1 |
282 |
jeremybenn |
;; GCC machine description for SSE instructions
|
2 |
|
|
;; Copyright (C) 2005, 2006, 2007, 2008, 2009, 2010
|
3 |
|
|
;; Free Software Foundation, Inc.
|
4 |
|
|
;;
|
5 |
|
|
;; This file is part of GCC.
|
6 |
|
|
;;
|
7 |
|
|
;; GCC is free software; you can redistribute it and/or modify
|
8 |
|
|
;; it under the terms of the GNU General Public License as published by
|
9 |
|
|
;; the Free Software Foundation; either version 3, or (at your option)
|
10 |
|
|
;; any later version.
|
11 |
|
|
;;
|
12 |
|
|
;; GCC is distributed in the hope that it will be useful,
|
13 |
|
|
;; but WITHOUT ANY WARRANTY; without even the implied warranty of
|
14 |
|
|
;; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
15 |
|
|
;; GNU General Public License for more details.
|
16 |
|
|
;;
|
17 |
|
|
;; You should have received a copy of the GNU General Public License
|
18 |
|
|
;; along with GCC; see the file COPYING3. If not see
|
19 |
|
|
;; .
|
20 |
|
|
|
21 |
|
|
|
22 |
|
|
;; 16 byte integral modes handled by SSE
|
23 |
|
|
(define_mode_iterator SSEMODEI [V16QI V8HI V4SI V2DI])
|
24 |
|
|
|
25 |
|
|
;; All 16-byte vector modes handled by SSE
|
26 |
|
|
(define_mode_iterator SSEMODE [V16QI V8HI V4SI V2DI V4SF V2DF])
|
27 |
|
|
(define_mode_iterator SSEMODE16 [V16QI V8HI V4SI V2DI V1TI V4SF V2DF])
|
28 |
|
|
|
29 |
|
|
;; 32 byte integral vector modes handled by AVX
|
30 |
|
|
(define_mode_iterator AVX256MODEI [V32QI V16HI V8SI V4DI])
|
31 |
|
|
|
32 |
|
|
;; All 32-byte vector modes handled by AVX
|
33 |
|
|
(define_mode_iterator AVX256MODE [V32QI V16HI V8SI V4DI V8SF V4DF])
|
34 |
|
|
|
35 |
|
|
;; All QI vector modes handled by AVX
|
36 |
|
|
(define_mode_iterator AVXMODEQI [V32QI V16QI])
|
37 |
|
|
|
38 |
|
|
;; All DI vector modes handled by AVX
|
39 |
|
|
(define_mode_iterator AVXMODEDI [V4DI V2DI])
|
40 |
|
|
|
41 |
|
|
;; All vector modes handled by AVX
|
42 |
|
|
(define_mode_iterator AVXMODE
|
43 |
|
|
[V16QI V8HI V4SI V2DI V4SF V2DF V32QI V16HI V8SI V4DI V8SF V4DF])
|
44 |
|
|
(define_mode_iterator AVXMODE16
|
45 |
|
|
[V16QI V8HI V4SI V2DI V1TI V4SF V2DF V32QI V16HI V8SI V4DI V8SF V4DF])
|
46 |
|
|
|
47 |
|
|
;; Mix-n-match
|
48 |
|
|
(define_mode_iterator SSEMODE12 [V16QI V8HI])
|
49 |
|
|
(define_mode_iterator SSEMODE24 [V8HI V4SI])
|
50 |
|
|
(define_mode_iterator SSEMODE14 [V16QI V4SI])
|
51 |
|
|
(define_mode_iterator SSEMODE124 [V16QI V8HI V4SI])
|
52 |
|
|
(define_mode_iterator SSEMODE248 [V8HI V4SI V2DI])
|
53 |
|
|
(define_mode_iterator SSEMODE1248 [V16QI V8HI V4SI V2DI])
|
54 |
|
|
(define_mode_iterator SSEMODEF4 [SF DF V4SF V2DF])
|
55 |
|
|
(define_mode_iterator FMA4MODEF4 [V8SF V4DF])
|
56 |
|
|
(define_mode_iterator SSEMODEF2P [V4SF V2DF])
|
57 |
|
|
|
58 |
|
|
(define_mode_iterator AVX256MODEF2P [V8SF V4DF])
|
59 |
|
|
(define_mode_iterator AVX256MODE2P [V8SI V8SF V4DF])
|
60 |
|
|
(define_mode_iterator AVX256MODE24P [V8SI V8SF V4DI V4DF])
|
61 |
|
|
(define_mode_iterator AVX256MODE4P [V4DI V4DF])
|
62 |
|
|
(define_mode_iterator AVX256MODE8P [V8SI V8SF])
|
63 |
|
|
(define_mode_iterator AVXMODEF2P [V4SF V2DF V8SF V4DF])
|
64 |
|
|
(define_mode_iterator AVXMODEF4P [V4SF V4DF])
|
65 |
|
|
(define_mode_iterator AVXMODEFDP [V2DF V4DF])
|
66 |
|
|
(define_mode_iterator AVXMODEFSP [V4SF V8SF])
|
67 |
|
|
(define_mode_iterator AVXMODEDCVTDQ2PS [V4SF V8SF])
|
68 |
|
|
(define_mode_iterator AVXMODEDCVTPS2DQ [V4SI V8SI])
|
69 |
|
|
|
70 |
|
|
;; Int-float size matches
|
71 |
|
|
(define_mode_iterator SSEMODE4S [V4SF V4SI])
|
72 |
|
|
(define_mode_iterator SSEMODE2D [V2DF V2DI])
|
73 |
|
|
|
74 |
|
|
;; Modes handled by integer vcond pattern
|
75 |
|
|
(define_mode_iterator SSEMODE124C8 [V16QI V8HI V4SI
|
76 |
|
|
(V2DI "TARGET_SSE4_2")])
|
77 |
|
|
|
78 |
|
|
;; Modes handled by vec_extract_even/odd pattern.
|
79 |
|
|
(define_mode_iterator SSEMODE_EO
|
80 |
|
|
[(V4SF "TARGET_SSE")
|
81 |
|
|
(V2DF "TARGET_SSE2")
|
82 |
|
|
(V2DI "TARGET_SSE2") (V4SI "TARGET_SSE2")
|
83 |
|
|
(V8HI "TARGET_SSE2") (V16QI "TARGET_SSE2")
|
84 |
|
|
(V4DF "TARGET_AVX") (V8SF "TARGET_AVX")])
|
85 |
|
|
|
86 |
|
|
;; Mapping from float mode to required SSE level
|
87 |
|
|
(define_mode_attr sse [(SF "sse") (DF "sse2") (V4SF "sse") (V2DF "sse2")])
|
88 |
|
|
|
89 |
|
|
;; Mapping from integer vector mode to mnemonic suffix
|
90 |
|
|
(define_mode_attr ssevecsize [(V16QI "b") (V8HI "w") (V4SI "d") (V2DI "q")])
|
91 |
|
|
|
92 |
|
|
;; Mapping of the fma4 suffix
|
93 |
|
|
(define_mode_attr fma4modesuffixf4 [(V8SF "ps") (V4DF "pd")])
|
94 |
|
|
(define_mode_attr ssemodesuffixf2s [(SF "ss") (DF "sd")
|
95 |
|
|
(V4SF "ss") (V2DF "sd")])
|
96 |
|
|
|
97 |
|
|
;; Mapping of the avx suffix
|
98 |
|
|
(define_mode_attr ssemodesuffixf4 [(SF "ss") (DF "sd")
|
99 |
|
|
(V4SF "ps") (V2DF "pd")])
|
100 |
|
|
|
101 |
|
|
(define_mode_attr ssemodesuffixf2c [(V4SF "s") (V2DF "d")])
|
102 |
|
|
|
103 |
|
|
(define_mode_attr ssescalarmodesuffix2s [(V4SF "ss") (V4SI "d")])
|
104 |
|
|
|
105 |
|
|
;; Mapping of the max integer size for xop rotate immediate constraint
|
106 |
|
|
(define_mode_attr sserotatemax [(V16QI "7") (V8HI "15") (V4SI "31") (V2DI "63")])
|
107 |
|
|
|
108 |
|
|
;; Mapping of vector modes back to the scalar modes
|
109 |
|
|
(define_mode_attr ssescalarmode [(V4SF "SF") (V2DF "DF")
|
110 |
|
|
(V16QI "QI") (V8HI "HI")
|
111 |
|
|
(V4SI "SI") (V2DI "DI")])
|
112 |
|
|
|
113 |
|
|
;; Mapping of vector modes to a vector mode of double size
|
114 |
|
|
(define_mode_attr ssedoublesizemode
|
115 |
|
|
[(V2DF "V4DF") (V2DI "V4DI") (V4SF "V8SF") (V4SI "V8SI")
|
116 |
|
|
(V8HI "V16HI") (V16QI "V32QI")
|
117 |
|
|
(V4DF "V8DF") (V8SF "V16SF")
|
118 |
|
|
(V4DI "V8DI") (V8SI "V16SI") (V16HI "V32HI") (V32QI "V64QI")])
|
119 |
|
|
|
120 |
|
|
;; Number of scalar elements in each vector type
|
121 |
|
|
(define_mode_attr ssescalarnum
|
122 |
|
|
[(V4SF "4") (V2DF "2") (V16QI "16") (V8HI "8") (V4SI "4") (V2DI "2")
|
123 |
|
|
(V8SF "8") (V4DF "4") (V32QI "32") (V16HI "16") (V8SI "8") (V4DI "4")])
|
124 |
|
|
|
125 |
|
|
;; Mapping for AVX
|
126 |
|
|
(define_mode_attr avxvecmode
|
127 |
|
|
[(V16QI "TI") (V8HI "TI") (V4SI "TI") (V2DI "TI") (V1TI "TI")
|
128 |
|
|
(V4SF "V4SF") (V8SF "V8SF") (V2DF "V2DF") (V4DF "V4DF")
|
129 |
|
|
(V32QI "OI") (V16HI "OI") (V8SI "OI") (V4DI "OI")])
|
130 |
|
|
(define_mode_attr avxvecpsmode
|
131 |
|
|
[(V16QI "V4SF") (V8HI "V4SF") (V4SI "V4SF") (V2DI "V4SF")
|
132 |
|
|
(V32QI "V8SF") (V16HI "V8SF") (V8SI "V8SF") (V4DI "V8SF")])
|
133 |
|
|
(define_mode_attr avxhalfvecmode
|
134 |
|
|
[(V32QI "V16QI") (V16HI "V8HI") (V8SI "V4SI") (V4DI "V2DI")
|
135 |
|
|
(V8SF "V4SF") (V4DF "V2DF")
|
136 |
|
|
(V16QI "V8QI") (V8HI "V4HI") (V4SI "V2SI") (V4SF "V2SF")])
|
137 |
|
|
(define_mode_attr avxscalarmode
|
138 |
|
|
[(V16QI "QI") (V8HI "HI") (V4SI "SI") (V2DI "DI") (V4SF "SF") (V2DF "DF")
|
139 |
|
|
(V32QI "QI") (V16HI "HI") (V8SI "SI") (V4DI "DI") (V8SF "SF") (V4DF "DF")])
|
140 |
|
|
(define_mode_attr avxcvtvecmode
|
141 |
|
|
[(V4SF "V4SI") (V8SF "V8SI") (V4SI "V4SF") (V8SI "V8SF")])
|
142 |
|
|
(define_mode_attr avxpermvecmode
|
143 |
|
|
[(V2DF "V2DI") (V4SF "V4SI") (V4DF "V4DI") (V8SF "V8SI")])
|
144 |
|
|
(define_mode_attr avxmodesuffixf2c
|
145 |
|
|
[(V4SF "s") (V2DF "d") (V8SI "s") (V8SF "s") (V4DI "d") (V4DF "d")])
|
146 |
|
|
(define_mode_attr avxmodesuffixp
|
147 |
|
|
[(V2DF "pd") (V4SI "si") (V4SF "ps") (V8SF "ps") (V8SI "si")
|
148 |
|
|
(V4DF "pd")])
|
149 |
|
|
(define_mode_attr avxmodesuffix
|
150 |
|
|
[(V16QI "") (V32QI "256") (V4SI "") (V4SF "") (V2DF "")
|
151 |
|
|
(V8SI "256") (V8SF "256") (V4DF "256")])
|
152 |
|
|
|
153 |
|
|
;; Mapping of immediate bits for blend instructions
|
154 |
|
|
(define_mode_attr blendbits
|
155 |
|
|
[(V8SF "255") (V4SF "15") (V4DF "15") (V2DF "3")])
|
156 |
|
|
|
157 |
|
|
;; Mapping of immediate bits for pinsr instructions
|
158 |
|
|
(define_mode_attr pinsrbits [(V16QI "32768") (V8HI "128") (V4SI "8")])
|
159 |
|
|
|
160 |
|
|
;; Patterns whose name begins with "sse{,2,3}_" are invoked by intrinsics.
|
161 |
|
|
|
162 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
163 |
|
|
;;
|
164 |
|
|
;; Move patterns
|
165 |
|
|
;;
|
166 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
167 |
|
|
|
168 |
|
|
(define_expand "mov"
|
169 |
|
|
[(set (match_operand:AVX256MODE 0 "nonimmediate_operand" "")
|
170 |
|
|
(match_operand:AVX256MODE 1 "nonimmediate_operand" ""))]
|
171 |
|
|
"TARGET_AVX"
|
172 |
|
|
{
|
173 |
|
|
ix86_expand_vector_move (mode, operands);
|
174 |
|
|
DONE;
|
175 |
|
|
})
|
176 |
|
|
|
177 |
|
|
(define_insn "*avx_mov_internal"
|
178 |
|
|
[(set (match_operand:AVXMODE16 0 "nonimmediate_operand" "=x,x ,m")
|
179 |
|
|
(match_operand:AVXMODE16 1 "nonimmediate_or_sse_const_operand" "C ,xm,x"))]
|
180 |
|
|
"TARGET_AVX
|
181 |
|
|
&& (register_operand (operands[0], mode)
|
182 |
|
|
|| register_operand (operands[1], mode))"
|
183 |
|
|
{
|
184 |
|
|
switch (which_alternative)
|
185 |
|
|
{
|
186 |
|
|
case 0:
|
187 |
|
|
return standard_sse_constant_opcode (insn, operands[1]);
|
188 |
|
|
case 1:
|
189 |
|
|
case 2:
|
190 |
|
|
switch (get_attr_mode (insn))
|
191 |
|
|
{
|
192 |
|
|
case MODE_V8SF:
|
193 |
|
|
case MODE_V4SF:
|
194 |
|
|
return "vmovaps\t{%1, %0|%0, %1}";
|
195 |
|
|
case MODE_V4DF:
|
196 |
|
|
case MODE_V2DF:
|
197 |
|
|
return "vmovapd\t{%1, %0|%0, %1}";
|
198 |
|
|
default:
|
199 |
|
|
return "vmovdqa\t{%1, %0|%0, %1}";
|
200 |
|
|
}
|
201 |
|
|
default:
|
202 |
|
|
gcc_unreachable ();
|
203 |
|
|
}
|
204 |
|
|
}
|
205 |
|
|
[(set_attr "type" "sselog1,ssemov,ssemov")
|
206 |
|
|
(set_attr "prefix" "vex")
|
207 |
|
|
(set_attr "mode" "")])
|
208 |
|
|
|
209 |
|
|
;; All of these patterns are enabled for SSE1 as well as SSE2.
|
210 |
|
|
;; This is essential for maintaining stable calling conventions.
|
211 |
|
|
|
212 |
|
|
(define_expand "mov"
|
213 |
|
|
[(set (match_operand:SSEMODE16 0 "nonimmediate_operand" "")
|
214 |
|
|
(match_operand:SSEMODE16 1 "nonimmediate_operand" ""))]
|
215 |
|
|
"TARGET_SSE"
|
216 |
|
|
{
|
217 |
|
|
ix86_expand_vector_move (mode, operands);
|
218 |
|
|
DONE;
|
219 |
|
|
})
|
220 |
|
|
|
221 |
|
|
(define_insn "*mov_internal"
|
222 |
|
|
[(set (match_operand:SSEMODE16 0 "nonimmediate_operand" "=x,x ,m")
|
223 |
|
|
(match_operand:SSEMODE16 1 "nonimmediate_or_sse_const_operand" "C ,xm,x"))]
|
224 |
|
|
"TARGET_SSE
|
225 |
|
|
&& (register_operand (operands[0], mode)
|
226 |
|
|
|| register_operand (operands[1], mode))"
|
227 |
|
|
{
|
228 |
|
|
switch (which_alternative)
|
229 |
|
|
{
|
230 |
|
|
case 0:
|
231 |
|
|
return standard_sse_constant_opcode (insn, operands[1]);
|
232 |
|
|
case 1:
|
233 |
|
|
case 2:
|
234 |
|
|
switch (get_attr_mode (insn))
|
235 |
|
|
{
|
236 |
|
|
case MODE_V4SF:
|
237 |
|
|
return "movaps\t{%1, %0|%0, %1}";
|
238 |
|
|
case MODE_V2DF:
|
239 |
|
|
return "movapd\t{%1, %0|%0, %1}";
|
240 |
|
|
default:
|
241 |
|
|
return "movdqa\t{%1, %0|%0, %1}";
|
242 |
|
|
}
|
243 |
|
|
default:
|
244 |
|
|
gcc_unreachable ();
|
245 |
|
|
}
|
246 |
|
|
}
|
247 |
|
|
[(set_attr "type" "sselog1,ssemov,ssemov")
|
248 |
|
|
(set (attr "mode")
|
249 |
|
|
(cond [(ior (ior (ne (symbol_ref "optimize_function_for_size_p (cfun)") (const_int 0))
|
250 |
|
|
(eq (symbol_ref "TARGET_SSE2") (const_int 0)))
|
251 |
|
|
(and (eq_attr "alternative" "2")
|
252 |
|
|
(ne (symbol_ref "TARGET_SSE_TYPELESS_STORES")
|
253 |
|
|
(const_int 0))))
|
254 |
|
|
(const_string "V4SF")
|
255 |
|
|
(eq (const_string "mode") (const_string "V4SFmode"))
|
256 |
|
|
(const_string "V4SF")
|
257 |
|
|
(eq (const_string "mode") (const_string "V2DFmode"))
|
258 |
|
|
(const_string "V2DF")
|
259 |
|
|
]
|
260 |
|
|
(const_string "TI")))])
|
261 |
|
|
|
262 |
|
|
;; Move a DI from a 32-bit register pair (e.g. %edx:%eax) to an xmm.
|
263 |
|
|
;; We'd rather avoid this entirely; if the 32-bit reg pair was loaded
|
264 |
|
|
;; from memory, we'd prefer to load the memory directly into the %xmm
|
265 |
|
|
;; register. To facilitate this happy circumstance, this pattern won't
|
266 |
|
|
;; split until after register allocation. If the 64-bit value didn't
|
267 |
|
|
;; come from memory, this is the best we can do. This is much better
|
268 |
|
|
;; than storing %edx:%eax into a stack temporary and loading an %xmm
|
269 |
|
|
;; from there.
|
270 |
|
|
|
271 |
|
|
(define_insn_and_split "movdi_to_sse"
|
272 |
|
|
[(parallel
|
273 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=?x,x")
|
274 |
|
|
(subreg:V4SI (match_operand:DI 1 "nonimmediate_operand" "r,m") 0))
|
275 |
|
|
(clobber (match_scratch:V4SI 2 "=&x,X"))])]
|
276 |
|
|
"!TARGET_64BIT && TARGET_SSE2 && TARGET_INTER_UNIT_MOVES"
|
277 |
|
|
"#"
|
278 |
|
|
"&& reload_completed"
|
279 |
|
|
[(const_int 0)]
|
280 |
|
|
{
|
281 |
|
|
if (register_operand (operands[1], DImode))
|
282 |
|
|
{
|
283 |
|
|
/* The DImode arrived in a pair of integral registers (e.g. %edx:%eax).
|
284 |
|
|
Assemble the 64-bit DImode value in an xmm register. */
|
285 |
|
|
emit_insn (gen_sse2_loadld (operands[0], CONST0_RTX (V4SImode),
|
286 |
|
|
gen_rtx_SUBREG (SImode, operands[1], 0)));
|
287 |
|
|
emit_insn (gen_sse2_loadld (operands[2], CONST0_RTX (V4SImode),
|
288 |
|
|
gen_rtx_SUBREG (SImode, operands[1], 4)));
|
289 |
|
|
emit_insn (gen_vec_interleave_lowv4si (operands[0], operands[0],
|
290 |
|
|
operands[2]));
|
291 |
|
|
}
|
292 |
|
|
else if (memory_operand (operands[1], DImode))
|
293 |
|
|
emit_insn (gen_vec_concatv2di (gen_lowpart (V2DImode, operands[0]),
|
294 |
|
|
operands[1], const0_rtx));
|
295 |
|
|
else
|
296 |
|
|
gcc_unreachable ();
|
297 |
|
|
})
|
298 |
|
|
|
299 |
|
|
(define_split
|
300 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "")
|
301 |
|
|
(match_operand:V4SF 1 "zero_extended_scalar_load_operand" ""))]
|
302 |
|
|
"TARGET_SSE && reload_completed"
|
303 |
|
|
[(set (match_dup 0)
|
304 |
|
|
(vec_merge:V4SF
|
305 |
|
|
(vec_duplicate:V4SF (match_dup 1))
|
306 |
|
|
(match_dup 2)
|
307 |
|
|
(const_int 1)))]
|
308 |
|
|
{
|
309 |
|
|
operands[1] = simplify_gen_subreg (SFmode, operands[1], V4SFmode, 0);
|
310 |
|
|
operands[2] = CONST0_RTX (V4SFmode);
|
311 |
|
|
})
|
312 |
|
|
|
313 |
|
|
(define_split
|
314 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "")
|
315 |
|
|
(match_operand:V2DF 1 "zero_extended_scalar_load_operand" ""))]
|
316 |
|
|
"TARGET_SSE2 && reload_completed"
|
317 |
|
|
[(set (match_dup 0) (vec_concat:V2DF (match_dup 1) (match_dup 2)))]
|
318 |
|
|
{
|
319 |
|
|
operands[1] = simplify_gen_subreg (DFmode, operands[1], V2DFmode, 0);
|
320 |
|
|
operands[2] = CONST0_RTX (DFmode);
|
321 |
|
|
})
|
322 |
|
|
|
323 |
|
|
(define_expand "push1"
|
324 |
|
|
[(match_operand:AVX256MODE 0 "register_operand" "")]
|
325 |
|
|
"TARGET_AVX"
|
326 |
|
|
{
|
327 |
|
|
ix86_expand_push (mode, operands[0]);
|
328 |
|
|
DONE;
|
329 |
|
|
})
|
330 |
|
|
|
331 |
|
|
(define_expand "push1"
|
332 |
|
|
[(match_operand:SSEMODE16 0 "register_operand" "")]
|
333 |
|
|
"TARGET_SSE"
|
334 |
|
|
{
|
335 |
|
|
ix86_expand_push (mode, operands[0]);
|
336 |
|
|
DONE;
|
337 |
|
|
})
|
338 |
|
|
|
339 |
|
|
(define_expand "movmisalign"
|
340 |
|
|
[(set (match_operand:AVX256MODE 0 "nonimmediate_operand" "")
|
341 |
|
|
(match_operand:AVX256MODE 1 "nonimmediate_operand" ""))]
|
342 |
|
|
"TARGET_AVX"
|
343 |
|
|
{
|
344 |
|
|
ix86_expand_vector_move_misalign (mode, operands);
|
345 |
|
|
DONE;
|
346 |
|
|
})
|
347 |
|
|
|
348 |
|
|
(define_expand "movmisalign"
|
349 |
|
|
[(set (match_operand:SSEMODE16 0 "nonimmediate_operand" "")
|
350 |
|
|
(match_operand:SSEMODE16 1 "nonimmediate_operand" ""))]
|
351 |
|
|
"TARGET_SSE"
|
352 |
|
|
{
|
353 |
|
|
ix86_expand_vector_move_misalign (mode, operands);
|
354 |
|
|
DONE;
|
355 |
|
|
})
|
356 |
|
|
|
357 |
|
|
(define_insn "avx_movup"
|
358 |
|
|
[(set (match_operand:AVXMODEF2P 0 "nonimmediate_operand" "=x,m")
|
359 |
|
|
(unspec:AVXMODEF2P
|
360 |
|
|
[(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "xm,x")]
|
361 |
|
|
UNSPEC_MOVU))]
|
362 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)
|
363 |
|
|
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
364 |
|
|
"vmovup\t{%1, %0|%0, %1}"
|
365 |
|
|
[(set_attr "type" "ssemov")
|
366 |
|
|
(set_attr "movu" "1")
|
367 |
|
|
(set_attr "prefix" "vex")
|
368 |
|
|
(set_attr "mode" "")])
|
369 |
|
|
|
370 |
|
|
(define_insn "sse2_movq128"
|
371 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
372 |
|
|
(vec_concat:V2DI
|
373 |
|
|
(vec_select:DI
|
374 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "xm")
|
375 |
|
|
(parallel [(const_int 0)]))
|
376 |
|
|
(const_int 0)))]
|
377 |
|
|
"TARGET_SSE2"
|
378 |
|
|
"%vmovq\t{%1, %0|%0, %1}"
|
379 |
|
|
[(set_attr "type" "ssemov")
|
380 |
|
|
(set_attr "prefix" "maybe_vex")
|
381 |
|
|
(set_attr "mode" "TI")])
|
382 |
|
|
|
383 |
|
|
(define_insn "_movup"
|
384 |
|
|
[(set (match_operand:SSEMODEF2P 0 "nonimmediate_operand" "=x,m")
|
385 |
|
|
(unspec:SSEMODEF2P
|
386 |
|
|
[(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "xm,x")]
|
387 |
|
|
UNSPEC_MOVU))]
|
388 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)
|
389 |
|
|
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
390 |
|
|
"movup\t{%1, %0|%0, %1}"
|
391 |
|
|
[(set_attr "type" "ssemov")
|
392 |
|
|
(set_attr "movu" "1")
|
393 |
|
|
(set_attr "mode" "")])
|
394 |
|
|
|
395 |
|
|
(define_insn "avx_movdqu"
|
396 |
|
|
[(set (match_operand:AVXMODEQI 0 "nonimmediate_operand" "=x,m")
|
397 |
|
|
(unspec:AVXMODEQI
|
398 |
|
|
[(match_operand:AVXMODEQI 1 "nonimmediate_operand" "xm,x")]
|
399 |
|
|
UNSPEC_MOVU))]
|
400 |
|
|
"TARGET_AVX && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
401 |
|
|
"vmovdqu\t{%1, %0|%0, %1}"
|
402 |
|
|
[(set_attr "type" "ssemov")
|
403 |
|
|
(set_attr "movu" "1")
|
404 |
|
|
(set_attr "prefix" "vex")
|
405 |
|
|
(set_attr "mode" "")])
|
406 |
|
|
|
407 |
|
|
(define_insn "sse2_movdqu"
|
408 |
|
|
[(set (match_operand:V16QI 0 "nonimmediate_operand" "=x,m")
|
409 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "nonimmediate_operand" "xm,x")]
|
410 |
|
|
UNSPEC_MOVU))]
|
411 |
|
|
"TARGET_SSE2 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
412 |
|
|
"movdqu\t{%1, %0|%0, %1}"
|
413 |
|
|
[(set_attr "type" "ssemov")
|
414 |
|
|
(set_attr "movu" "1")
|
415 |
|
|
(set_attr "prefix_data16" "1")
|
416 |
|
|
(set_attr "mode" "TI")])
|
417 |
|
|
|
418 |
|
|
(define_insn "avx_movnt"
|
419 |
|
|
[(set (match_operand:AVXMODEF2P 0 "memory_operand" "=m")
|
420 |
|
|
(unspec:AVXMODEF2P
|
421 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")]
|
422 |
|
|
UNSPEC_MOVNT))]
|
423 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
424 |
|
|
"vmovntp\t{%1, %0|%0, %1}"
|
425 |
|
|
[(set_attr "type" "ssemov")
|
426 |
|
|
(set_attr "prefix" "vex")
|
427 |
|
|
(set_attr "mode" "")])
|
428 |
|
|
|
429 |
|
|
(define_insn "_movnt"
|
430 |
|
|
[(set (match_operand:SSEMODEF2P 0 "memory_operand" "=m")
|
431 |
|
|
(unspec:SSEMODEF2P
|
432 |
|
|
[(match_operand:SSEMODEF2P 1 "register_operand" "x")]
|
433 |
|
|
UNSPEC_MOVNT))]
|
434 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
435 |
|
|
"movntp\t{%1, %0|%0, %1}"
|
436 |
|
|
[(set_attr "type" "ssemov")
|
437 |
|
|
(set_attr "mode" "")])
|
438 |
|
|
|
439 |
|
|
(define_insn "avx_movnt"
|
440 |
|
|
[(set (match_operand:AVXMODEDI 0 "memory_operand" "=m")
|
441 |
|
|
(unspec:AVXMODEDI
|
442 |
|
|
[(match_operand:AVXMODEDI 1 "register_operand" "x")]
|
443 |
|
|
UNSPEC_MOVNT))]
|
444 |
|
|
"TARGET_AVX"
|
445 |
|
|
"vmovntdq\t{%1, %0|%0, %1}"
|
446 |
|
|
[(set_attr "type" "ssecvt")
|
447 |
|
|
(set_attr "prefix" "vex")
|
448 |
|
|
(set_attr "mode" "")])
|
449 |
|
|
|
450 |
|
|
(define_insn "sse2_movntv2di"
|
451 |
|
|
[(set (match_operand:V2DI 0 "memory_operand" "=m")
|
452 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "x")]
|
453 |
|
|
UNSPEC_MOVNT))]
|
454 |
|
|
"TARGET_SSE2"
|
455 |
|
|
"movntdq\t{%1, %0|%0, %1}"
|
456 |
|
|
[(set_attr "type" "ssemov")
|
457 |
|
|
(set_attr "prefix_data16" "1")
|
458 |
|
|
(set_attr "mode" "TI")])
|
459 |
|
|
|
460 |
|
|
(define_insn "sse2_movntsi"
|
461 |
|
|
[(set (match_operand:SI 0 "memory_operand" "=m")
|
462 |
|
|
(unspec:SI [(match_operand:SI 1 "register_operand" "r")]
|
463 |
|
|
UNSPEC_MOVNT))]
|
464 |
|
|
"TARGET_SSE2"
|
465 |
|
|
"movnti\t{%1, %0|%0, %1}"
|
466 |
|
|
[(set_attr "type" "ssemov")
|
467 |
|
|
(set_attr "prefix_data16" "0")
|
468 |
|
|
(set_attr "mode" "V2DF")])
|
469 |
|
|
|
470 |
|
|
(define_insn "avx_lddqu"
|
471 |
|
|
[(set (match_operand:AVXMODEQI 0 "register_operand" "=x")
|
472 |
|
|
(unspec:AVXMODEQI
|
473 |
|
|
[(match_operand:AVXMODEQI 1 "memory_operand" "m")]
|
474 |
|
|
UNSPEC_LDDQU))]
|
475 |
|
|
"TARGET_AVX"
|
476 |
|
|
"vlddqu\t{%1, %0|%0, %1}"
|
477 |
|
|
[(set_attr "type" "ssecvt")
|
478 |
|
|
(set_attr "movu" "1")
|
479 |
|
|
(set_attr "prefix" "vex")
|
480 |
|
|
(set_attr "mode" "")])
|
481 |
|
|
|
482 |
|
|
(define_insn "sse3_lddqu"
|
483 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
484 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "memory_operand" "m")]
|
485 |
|
|
UNSPEC_LDDQU))]
|
486 |
|
|
"TARGET_SSE3"
|
487 |
|
|
"lddqu\t{%1, %0|%0, %1}"
|
488 |
|
|
[(set_attr "type" "ssemov")
|
489 |
|
|
(set_attr "movu" "1")
|
490 |
|
|
(set_attr "prefix_data16" "0")
|
491 |
|
|
(set_attr "prefix_rep" "1")
|
492 |
|
|
(set_attr "mode" "TI")])
|
493 |
|
|
|
494 |
|
|
; Expand patterns for non-temporal stores. At the moment, only those
|
495 |
|
|
; that directly map to insns are defined; it would be possible to
|
496 |
|
|
; define patterns for other modes that would expand to several insns.
|
497 |
|
|
|
498 |
|
|
(define_expand "storent"
|
499 |
|
|
[(set (match_operand:SSEMODEF2P 0 "memory_operand" "")
|
500 |
|
|
(unspec:SSEMODEF2P
|
501 |
|
|
[(match_operand:SSEMODEF2P 1 "register_operand" "")]
|
502 |
|
|
UNSPEC_MOVNT))]
|
503 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
504 |
|
|
"")
|
505 |
|
|
|
506 |
|
|
(define_expand "storent"
|
507 |
|
|
[(set (match_operand:MODEF 0 "memory_operand" "")
|
508 |
|
|
(unspec:MODEF
|
509 |
|
|
[(match_operand:MODEF 1 "register_operand" "")]
|
510 |
|
|
UNSPEC_MOVNT))]
|
511 |
|
|
"TARGET_SSE4A"
|
512 |
|
|
"")
|
513 |
|
|
|
514 |
|
|
(define_expand "storentv2di"
|
515 |
|
|
[(set (match_operand:V2DI 0 "memory_operand" "")
|
516 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "")]
|
517 |
|
|
UNSPEC_MOVNT))]
|
518 |
|
|
"TARGET_SSE2"
|
519 |
|
|
"")
|
520 |
|
|
|
521 |
|
|
(define_expand "storentsi"
|
522 |
|
|
[(set (match_operand:SI 0 "memory_operand" "")
|
523 |
|
|
(unspec:SI [(match_operand:SI 1 "register_operand" "")]
|
524 |
|
|
UNSPEC_MOVNT))]
|
525 |
|
|
"TARGET_SSE2"
|
526 |
|
|
"")
|
527 |
|
|
|
528 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
529 |
|
|
;;
|
530 |
|
|
;; Parallel floating point arithmetic
|
531 |
|
|
;;
|
532 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
533 |
|
|
|
534 |
|
|
(define_expand "2"
|
535 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "")
|
536 |
|
|
(absneg:SSEMODEF2P
|
537 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "")))]
|
538 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
539 |
|
|
"ix86_expand_fp_absneg_operator (, mode, operands); DONE;")
|
540 |
|
|
|
541 |
|
|
(define_expand "3"
|
542 |
|
|
[(set (match_operand:AVX256MODEF2P 0 "register_operand" "")
|
543 |
|
|
(plusminus:AVX256MODEF2P
|
544 |
|
|
(match_operand:AVX256MODEF2P 1 "nonimmediate_operand" "")
|
545 |
|
|
(match_operand:AVX256MODEF2P 2 "nonimmediate_operand" "")))]
|
546 |
|
|
"AVX256_VEC_FLOAT_MODE_P (mode)"
|
547 |
|
|
"ix86_fixup_binary_operands_no_copy (, mode, operands);")
|
548 |
|
|
|
549 |
|
|
(define_insn "*avx_3"
|
550 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
551 |
|
|
(plusminus:AVXMODEF2P
|
552 |
|
|
(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "x")
|
553 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")))]
|
554 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)
|
555 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
556 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
557 |
|
|
[(set_attr "type" "sseadd")
|
558 |
|
|
(set_attr "prefix" "vex")
|
559 |
|
|
(set_attr "mode" "")])
|
560 |
|
|
|
561 |
|
|
(define_expand "3"
|
562 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "")
|
563 |
|
|
(plusminus:SSEMODEF2P
|
564 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "")
|
565 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "")))]
|
566 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
567 |
|
|
"ix86_fixup_binary_operands_no_copy (, mode, operands);")
|
568 |
|
|
|
569 |
|
|
(define_insn "*3"
|
570 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
571 |
|
|
(plusminus:SSEMODEF2P
|
572 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "0")
|
573 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")))]
|
574 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)
|
575 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
576 |
|
|
"p\t{%2, %0|%0, %2}"
|
577 |
|
|
[(set_attr "type" "sseadd")
|
578 |
|
|
(set_attr "mode" "")])
|
579 |
|
|
|
580 |
|
|
(define_insn "*avx_vm3"
|
581 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
582 |
|
|
(vec_merge:SSEMODEF2P
|
583 |
|
|
(plusminus:SSEMODEF2P
|
584 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
585 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm"))
|
586 |
|
|
(match_dup 1)
|
587 |
|
|
(const_int 1)))]
|
588 |
|
|
"AVX128_VEC_FLOAT_MODE_P (mode)"
|
589 |
|
|
"vs\t{%2, %1, %0|%0, %1, %2}"
|
590 |
|
|
[(set_attr "type" "sseadd")
|
591 |
|
|
(set_attr "prefix" "vex")
|
592 |
|
|
(set_attr "mode" "")])
|
593 |
|
|
|
594 |
|
|
(define_insn "_vm3"
|
595 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
596 |
|
|
(vec_merge:SSEMODEF2P
|
597 |
|
|
(plusminus:SSEMODEF2P
|
598 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
599 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm"))
|
600 |
|
|
(match_dup 1)
|
601 |
|
|
(const_int 1)))]
|
602 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
603 |
|
|
"s\t{%2, %0|%0, %2}"
|
604 |
|
|
[(set_attr "type" "sseadd")
|
605 |
|
|
(set_attr "mode" "")])
|
606 |
|
|
|
607 |
|
|
(define_expand "mul3"
|
608 |
|
|
[(set (match_operand:AVX256MODEF2P 0 "register_operand" "")
|
609 |
|
|
(mult:AVX256MODEF2P
|
610 |
|
|
(match_operand:AVX256MODEF2P 1 "nonimmediate_operand" "")
|
611 |
|
|
(match_operand:AVX256MODEF2P 2 "nonimmediate_operand" "")))]
|
612 |
|
|
"AVX256_VEC_FLOAT_MODE_P (mode)"
|
613 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, mode, operands);")
|
614 |
|
|
|
615 |
|
|
(define_insn "*avx_mul3"
|
616 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
617 |
|
|
(mult:AVXMODEF2P
|
618 |
|
|
(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "%x")
|
619 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")))]
|
620 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)
|
621 |
|
|
&& ix86_binary_operator_ok (MULT, mode, operands)"
|
622 |
|
|
"vmulp\t{%2, %1, %0|%0, %1, %2}"
|
623 |
|
|
[(set_attr "type" "ssemul")
|
624 |
|
|
(set_attr "prefix" "vex")
|
625 |
|
|
(set_attr "mode" "")])
|
626 |
|
|
|
627 |
|
|
(define_expand "mul3"
|
628 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "")
|
629 |
|
|
(mult:SSEMODEF2P
|
630 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "")
|
631 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "")))]
|
632 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
633 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, mode, operands);")
|
634 |
|
|
|
635 |
|
|
(define_insn "*mul3"
|
636 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
637 |
|
|
(mult:SSEMODEF2P
|
638 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%0")
|
639 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")))]
|
640 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)
|
641 |
|
|
&& ix86_binary_operator_ok (MULT, mode, operands)"
|
642 |
|
|
"mulp\t{%2, %0|%0, %2}"
|
643 |
|
|
[(set_attr "type" "ssemul")
|
644 |
|
|
(set_attr "mode" "")])
|
645 |
|
|
|
646 |
|
|
(define_insn "*avx_vmmul3"
|
647 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
648 |
|
|
(vec_merge:SSEMODEF2P
|
649 |
|
|
(mult:SSEMODEF2P
|
650 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
651 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm"))
|
652 |
|
|
(match_dup 1)
|
653 |
|
|
(const_int 1)))]
|
654 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
655 |
|
|
"vmuls\t{%2, %1, %0|%0, %1, %2}"
|
656 |
|
|
[(set_attr "type" "ssemul")
|
657 |
|
|
(set_attr "prefix" "vex")
|
658 |
|
|
(set_attr "mode" "")])
|
659 |
|
|
|
660 |
|
|
(define_insn "_vmmul3"
|
661 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
662 |
|
|
(vec_merge:SSEMODEF2P
|
663 |
|
|
(mult:SSEMODEF2P
|
664 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
665 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm"))
|
666 |
|
|
(match_dup 1)
|
667 |
|
|
(const_int 1)))]
|
668 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
669 |
|
|
"muls\t{%2, %0|%0, %2}"
|
670 |
|
|
[(set_attr "type" "ssemul")
|
671 |
|
|
(set_attr "mode" "")])
|
672 |
|
|
|
673 |
|
|
(define_expand "divv8sf3"
|
674 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "")
|
675 |
|
|
(div:V8SF (match_operand:V8SF 1 "register_operand" "")
|
676 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "")))]
|
677 |
|
|
"TARGET_AVX"
|
678 |
|
|
{
|
679 |
|
|
ix86_fixup_binary_operands_no_copy (DIV, V8SFmode, operands);
|
680 |
|
|
|
681 |
|
|
if (TARGET_SSE_MATH && TARGET_RECIP && !optimize_insn_for_size_p ()
|
682 |
|
|
&& flag_finite_math_only && !flag_trapping_math
|
683 |
|
|
&& flag_unsafe_math_optimizations)
|
684 |
|
|
{
|
685 |
|
|
ix86_emit_swdivsf (operands[0], operands[1],
|
686 |
|
|
operands[2], V8SFmode);
|
687 |
|
|
DONE;
|
688 |
|
|
}
|
689 |
|
|
})
|
690 |
|
|
|
691 |
|
|
(define_expand "divv4df3"
|
692 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "")
|
693 |
|
|
(div:V4DF (match_operand:V4DF 1 "register_operand" "")
|
694 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "")))]
|
695 |
|
|
"TARGET_AVX"
|
696 |
|
|
"ix86_fixup_binary_operands_no_copy (DIV, V4DFmode, operands);")
|
697 |
|
|
|
698 |
|
|
(define_insn "avx_div3"
|
699 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
700 |
|
|
(div:AVXMODEF2P
|
701 |
|
|
(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
702 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")))]
|
703 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
704 |
|
|
"vdivp\t{%2, %1, %0|%0, %1, %2}"
|
705 |
|
|
[(set_attr "type" "ssediv")
|
706 |
|
|
(set_attr "prefix" "vex")
|
707 |
|
|
(set_attr "mode" "")])
|
708 |
|
|
|
709 |
|
|
(define_expand "divv4sf3"
|
710 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "")
|
711 |
|
|
(div:V4SF (match_operand:V4SF 1 "register_operand" "")
|
712 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "")))]
|
713 |
|
|
"TARGET_SSE"
|
714 |
|
|
{
|
715 |
|
|
if (TARGET_SSE_MATH && TARGET_RECIP && optimize_insn_for_speed_p ()
|
716 |
|
|
&& flag_finite_math_only && !flag_trapping_math
|
717 |
|
|
&& flag_unsafe_math_optimizations)
|
718 |
|
|
{
|
719 |
|
|
ix86_emit_swdivsf (operands[0], operands[1],
|
720 |
|
|
operands[2], V4SFmode);
|
721 |
|
|
DONE;
|
722 |
|
|
}
|
723 |
|
|
})
|
724 |
|
|
|
725 |
|
|
(define_expand "divv2df3"
|
726 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "")
|
727 |
|
|
(div:V2DF (match_operand:V2DF 1 "register_operand" "")
|
728 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "")))]
|
729 |
|
|
"TARGET_SSE2"
|
730 |
|
|
"")
|
731 |
|
|
|
732 |
|
|
(define_insn "*avx_div3"
|
733 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
734 |
|
|
(div:SSEMODEF2P
|
735 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
736 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")))]
|
737 |
|
|
"AVX128_VEC_FLOAT_MODE_P (mode)"
|
738 |
|
|
"vdivp\t{%2, %1, %0|%0, %1, %2}"
|
739 |
|
|
[(set_attr "type" "ssediv")
|
740 |
|
|
(set_attr "prefix" "vex")
|
741 |
|
|
(set_attr "mode" "")])
|
742 |
|
|
|
743 |
|
|
(define_insn "_div3"
|
744 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
745 |
|
|
(div:SSEMODEF2P
|
746 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
747 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")))]
|
748 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
749 |
|
|
"divp\t{%2, %0|%0, %2}"
|
750 |
|
|
[(set_attr "type" "ssediv")
|
751 |
|
|
(set_attr "mode" "")])
|
752 |
|
|
|
753 |
|
|
(define_insn "*avx_vmdiv3"
|
754 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
755 |
|
|
(vec_merge:SSEMODEF2P
|
756 |
|
|
(div:SSEMODEF2P
|
757 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
758 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm"))
|
759 |
|
|
(match_dup 1)
|
760 |
|
|
(const_int 1)))]
|
761 |
|
|
"AVX128_VEC_FLOAT_MODE_P (mode)"
|
762 |
|
|
"vdivs\t{%2, %1, %0|%0, %1, %2}"
|
763 |
|
|
[(set_attr "type" "ssediv")
|
764 |
|
|
(set_attr "prefix" "vex")
|
765 |
|
|
(set_attr "mode" "")])
|
766 |
|
|
|
767 |
|
|
(define_insn "_vmdiv3"
|
768 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
769 |
|
|
(vec_merge:SSEMODEF2P
|
770 |
|
|
(div:SSEMODEF2P
|
771 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
772 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm"))
|
773 |
|
|
(match_dup 1)
|
774 |
|
|
(const_int 1)))]
|
775 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
776 |
|
|
"divs\t{%2, %0|%0, %2}"
|
777 |
|
|
[(set_attr "type" "ssediv")
|
778 |
|
|
(set_attr "mode" "")])
|
779 |
|
|
|
780 |
|
|
(define_insn "avx_rcpv8sf2"
|
781 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
782 |
|
|
(unspec:V8SF
|
783 |
|
|
[(match_operand:V8SF 1 "nonimmediate_operand" "xm")] UNSPEC_RCP))]
|
784 |
|
|
"TARGET_AVX"
|
785 |
|
|
"vrcpps\t{%1, %0|%0, %1}"
|
786 |
|
|
[(set_attr "type" "sse")
|
787 |
|
|
(set_attr "prefix" "vex")
|
788 |
|
|
(set_attr "mode" "V8SF")])
|
789 |
|
|
|
790 |
|
|
(define_insn "sse_rcpv4sf2"
|
791 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
792 |
|
|
(unspec:V4SF
|
793 |
|
|
[(match_operand:V4SF 1 "nonimmediate_operand" "xm")] UNSPEC_RCP))]
|
794 |
|
|
"TARGET_SSE"
|
795 |
|
|
"%vrcpps\t{%1, %0|%0, %1}"
|
796 |
|
|
[(set_attr "type" "sse")
|
797 |
|
|
(set_attr "atom_sse_attr" "rcp")
|
798 |
|
|
(set_attr "prefix" "maybe_vex")
|
799 |
|
|
(set_attr "mode" "V4SF")])
|
800 |
|
|
|
801 |
|
|
(define_insn "*avx_vmrcpv4sf2"
|
802 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
803 |
|
|
(vec_merge:V4SF
|
804 |
|
|
(unspec:V4SF [(match_operand:V4SF 1 "nonimmediate_operand" "xm")]
|
805 |
|
|
UNSPEC_RCP)
|
806 |
|
|
(match_operand:V4SF 2 "register_operand" "x")
|
807 |
|
|
(const_int 1)))]
|
808 |
|
|
"TARGET_AVX"
|
809 |
|
|
"vrcpss\t{%1, %2, %0|%0, %2, %1}"
|
810 |
|
|
[(set_attr "type" "sse")
|
811 |
|
|
(set_attr "prefix" "vex")
|
812 |
|
|
(set_attr "mode" "SF")])
|
813 |
|
|
|
814 |
|
|
(define_insn "sse_vmrcpv4sf2"
|
815 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
816 |
|
|
(vec_merge:V4SF
|
817 |
|
|
(unspec:V4SF [(match_operand:V4SF 1 "nonimmediate_operand" "xm")]
|
818 |
|
|
UNSPEC_RCP)
|
819 |
|
|
(match_operand:V4SF 2 "register_operand" "0")
|
820 |
|
|
(const_int 1)))]
|
821 |
|
|
"TARGET_SSE"
|
822 |
|
|
"rcpss\t{%1, %0|%0, %1}"
|
823 |
|
|
[(set_attr "type" "sse")
|
824 |
|
|
(set_attr "atom_sse_attr" "rcp")
|
825 |
|
|
(set_attr "mode" "SF")])
|
826 |
|
|
|
827 |
|
|
(define_expand "sqrtv8sf2"
|
828 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "")
|
829 |
|
|
(sqrt:V8SF (match_operand:V8SF 1 "nonimmediate_operand" "")))]
|
830 |
|
|
"TARGET_AVX"
|
831 |
|
|
{
|
832 |
|
|
if (TARGET_SSE_MATH && TARGET_RECIP && !optimize_insn_for_size_p ()
|
833 |
|
|
&& flag_finite_math_only && !flag_trapping_math
|
834 |
|
|
&& flag_unsafe_math_optimizations)
|
835 |
|
|
{
|
836 |
|
|
ix86_emit_swsqrtsf (operands[0], operands[1], V8SFmode, 0);
|
837 |
|
|
DONE;
|
838 |
|
|
}
|
839 |
|
|
})
|
840 |
|
|
|
841 |
|
|
(define_insn "avx_sqrtv8sf2"
|
842 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
843 |
|
|
(sqrt:V8SF (match_operand:V8SF 1 "nonimmediate_operand" "xm")))]
|
844 |
|
|
"TARGET_AVX"
|
845 |
|
|
"vsqrtps\t{%1, %0|%0, %1}"
|
846 |
|
|
[(set_attr "type" "sse")
|
847 |
|
|
(set_attr "prefix" "vex")
|
848 |
|
|
(set_attr "mode" "V8SF")])
|
849 |
|
|
|
850 |
|
|
(define_expand "sqrtv4sf2"
|
851 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "")
|
852 |
|
|
(sqrt:V4SF (match_operand:V4SF 1 "nonimmediate_operand" "")))]
|
853 |
|
|
"TARGET_SSE"
|
854 |
|
|
{
|
855 |
|
|
if (TARGET_SSE_MATH && TARGET_RECIP && optimize_insn_for_speed_p ()
|
856 |
|
|
&& flag_finite_math_only && !flag_trapping_math
|
857 |
|
|
&& flag_unsafe_math_optimizations)
|
858 |
|
|
{
|
859 |
|
|
ix86_emit_swsqrtsf (operands[0], operands[1], V4SFmode, 0);
|
860 |
|
|
DONE;
|
861 |
|
|
}
|
862 |
|
|
})
|
863 |
|
|
|
864 |
|
|
(define_insn "sse_sqrtv4sf2"
|
865 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
866 |
|
|
(sqrt:V4SF (match_operand:V4SF 1 "nonimmediate_operand" "xm")))]
|
867 |
|
|
"TARGET_SSE"
|
868 |
|
|
"%vsqrtps\t{%1, %0|%0, %1}"
|
869 |
|
|
[(set_attr "type" "sse")
|
870 |
|
|
(set_attr "atom_sse_attr" "sqrt")
|
871 |
|
|
(set_attr "prefix" "maybe_vex")
|
872 |
|
|
(set_attr "mode" "V4SF")])
|
873 |
|
|
|
874 |
|
|
(define_insn "sqrtv4df2"
|
875 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x")
|
876 |
|
|
(sqrt:V4DF (match_operand:V4DF 1 "nonimmediate_operand" "xm")))]
|
877 |
|
|
"TARGET_AVX"
|
878 |
|
|
"vsqrtpd\t{%1, %0|%0, %1}"
|
879 |
|
|
[(set_attr "type" "sse")
|
880 |
|
|
(set_attr "prefix" "vex")
|
881 |
|
|
(set_attr "mode" "V4DF")])
|
882 |
|
|
|
883 |
|
|
(define_insn "sqrtv2df2"
|
884 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
885 |
|
|
(sqrt:V2DF (match_operand:V2DF 1 "nonimmediate_operand" "xm")))]
|
886 |
|
|
"TARGET_SSE2"
|
887 |
|
|
"%vsqrtpd\t{%1, %0|%0, %1}"
|
888 |
|
|
[(set_attr "type" "sse")
|
889 |
|
|
(set_attr "prefix" "maybe_vex")
|
890 |
|
|
(set_attr "mode" "V2DF")])
|
891 |
|
|
|
892 |
|
|
(define_insn "*avx_vmsqrt2"
|
893 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
894 |
|
|
(vec_merge:SSEMODEF2P
|
895 |
|
|
(sqrt:SSEMODEF2P
|
896 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "xm"))
|
897 |
|
|
(match_operand:SSEMODEF2P 2 "register_operand" "x")
|
898 |
|
|
(const_int 1)))]
|
899 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
900 |
|
|
"vsqrts\t{%1, %2, %0|%0, %2, %1}"
|
901 |
|
|
[(set_attr "type" "sse")
|
902 |
|
|
(set_attr "prefix" "vex")
|
903 |
|
|
(set_attr "mode" "")])
|
904 |
|
|
|
905 |
|
|
(define_insn "_vmsqrt2"
|
906 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
907 |
|
|
(vec_merge:SSEMODEF2P
|
908 |
|
|
(sqrt:SSEMODEF2P
|
909 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "xm"))
|
910 |
|
|
(match_operand:SSEMODEF2P 2 "register_operand" "0")
|
911 |
|
|
(const_int 1)))]
|
912 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
913 |
|
|
"sqrts\t{%1, %0|%0, %1}"
|
914 |
|
|
[(set_attr "type" "sse")
|
915 |
|
|
(set_attr "atom_sse_attr" "sqrt")
|
916 |
|
|
(set_attr "mode" "")])
|
917 |
|
|
|
918 |
|
|
(define_expand "rsqrtv8sf2"
|
919 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "")
|
920 |
|
|
(unspec:V8SF
|
921 |
|
|
[(match_operand:V8SF 1 "nonimmediate_operand" "")] UNSPEC_RSQRT))]
|
922 |
|
|
"TARGET_AVX && TARGET_SSE_MATH"
|
923 |
|
|
{
|
924 |
|
|
ix86_emit_swsqrtsf (operands[0], operands[1], V8SFmode, 1);
|
925 |
|
|
DONE;
|
926 |
|
|
})
|
927 |
|
|
|
928 |
|
|
(define_insn "avx_rsqrtv8sf2"
|
929 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
930 |
|
|
(unspec:V8SF
|
931 |
|
|
[(match_operand:V8SF 1 "nonimmediate_operand" "xm")] UNSPEC_RSQRT))]
|
932 |
|
|
"TARGET_AVX"
|
933 |
|
|
"vrsqrtps\t{%1, %0|%0, %1}"
|
934 |
|
|
[(set_attr "type" "sse")
|
935 |
|
|
(set_attr "prefix" "vex")
|
936 |
|
|
(set_attr "mode" "V8SF")])
|
937 |
|
|
|
938 |
|
|
(define_expand "rsqrtv4sf2"
|
939 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "")
|
940 |
|
|
(unspec:V4SF
|
941 |
|
|
[(match_operand:V4SF 1 "nonimmediate_operand" "")] UNSPEC_RSQRT))]
|
942 |
|
|
"TARGET_SSE_MATH"
|
943 |
|
|
{
|
944 |
|
|
ix86_emit_swsqrtsf (operands[0], operands[1], V4SFmode, 1);
|
945 |
|
|
DONE;
|
946 |
|
|
})
|
947 |
|
|
|
948 |
|
|
(define_insn "sse_rsqrtv4sf2"
|
949 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
950 |
|
|
(unspec:V4SF
|
951 |
|
|
[(match_operand:V4SF 1 "nonimmediate_operand" "xm")] UNSPEC_RSQRT))]
|
952 |
|
|
"TARGET_SSE"
|
953 |
|
|
"%vrsqrtps\t{%1, %0|%0, %1}"
|
954 |
|
|
[(set_attr "type" "sse")
|
955 |
|
|
(set_attr "prefix" "maybe_vex")
|
956 |
|
|
(set_attr "mode" "V4SF")])
|
957 |
|
|
|
958 |
|
|
(define_insn "*avx_vmrsqrtv4sf2"
|
959 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
960 |
|
|
(vec_merge:V4SF
|
961 |
|
|
(unspec:V4SF [(match_operand:V4SF 1 "nonimmediate_operand" "xm")]
|
962 |
|
|
UNSPEC_RSQRT)
|
963 |
|
|
(match_operand:V4SF 2 "register_operand" "x")
|
964 |
|
|
(const_int 1)))]
|
965 |
|
|
"TARGET_AVX"
|
966 |
|
|
"vrsqrtss\t{%1, %2, %0|%0, %2, %1}"
|
967 |
|
|
[(set_attr "type" "sse")
|
968 |
|
|
(set_attr "prefix" "vex")
|
969 |
|
|
(set_attr "mode" "SF")])
|
970 |
|
|
|
971 |
|
|
(define_insn "sse_vmrsqrtv4sf2"
|
972 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
973 |
|
|
(vec_merge:V4SF
|
974 |
|
|
(unspec:V4SF [(match_operand:V4SF 1 "nonimmediate_operand" "xm")]
|
975 |
|
|
UNSPEC_RSQRT)
|
976 |
|
|
(match_operand:V4SF 2 "register_operand" "0")
|
977 |
|
|
(const_int 1)))]
|
978 |
|
|
"TARGET_SSE"
|
979 |
|
|
"rsqrtss\t{%1, %0|%0, %1}"
|
980 |
|
|
[(set_attr "type" "sse")
|
981 |
|
|
(set_attr "mode" "SF")])
|
982 |
|
|
|
983 |
|
|
;; ??? For !flag_finite_math_only, the representation with SMIN/SMAX
|
984 |
|
|
;; isn't really correct, as those rtl operators aren't defined when
|
985 |
|
|
;; applied to NaNs. Hopefully the optimizers won't get too smart on us.
|
986 |
|
|
|
987 |
|
|
(define_expand "3"
|
988 |
|
|
[(set (match_operand:AVX256MODEF2P 0 "register_operand" "")
|
989 |
|
|
(smaxmin:AVX256MODEF2P
|
990 |
|
|
(match_operand:AVX256MODEF2P 1 "nonimmediate_operand" "")
|
991 |
|
|
(match_operand:AVX256MODEF2P 2 "nonimmediate_operand" "")))]
|
992 |
|
|
"AVX256_VEC_FLOAT_MODE_P (mode)"
|
993 |
|
|
{
|
994 |
|
|
if (!flag_finite_math_only)
|
995 |
|
|
operands[1] = force_reg (mode, operands[1]);
|
996 |
|
|
ix86_fixup_binary_operands_no_copy (, mode, operands);
|
997 |
|
|
})
|
998 |
|
|
|
999 |
|
|
(define_expand "3"
|
1000 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "")
|
1001 |
|
|
(smaxmin:SSEMODEF2P
|
1002 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "")
|
1003 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "")))]
|
1004 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1005 |
|
|
{
|
1006 |
|
|
if (!flag_finite_math_only)
|
1007 |
|
|
operands[1] = force_reg (mode, operands[1]);
|
1008 |
|
|
ix86_fixup_binary_operands_no_copy (, mode, operands);
|
1009 |
|
|
})
|
1010 |
|
|
|
1011 |
|
|
(define_insn "*avx_3_finite"
|
1012 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
1013 |
|
|
(smaxmin:AVXMODEF2P
|
1014 |
|
|
(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "%x")
|
1015 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")))]
|
1016 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode) && flag_finite_math_only
|
1017 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
1018 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
1019 |
|
|
[(set_attr "type" "sseadd")
|
1020 |
|
|
(set_attr "prefix" "vex")
|
1021 |
|
|
(set_attr "mode" "")])
|
1022 |
|
|
|
1023 |
|
|
(define_insn "*3_finite"
|
1024 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1025 |
|
|
(smaxmin:SSEMODEF2P
|
1026 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%0")
|
1027 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")))]
|
1028 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode) && flag_finite_math_only
|
1029 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
1030 |
|
|
"p\t{%2, %0|%0, %2}"
|
1031 |
|
|
[(set_attr "type" "sseadd")
|
1032 |
|
|
(set_attr "mode" "")])
|
1033 |
|
|
|
1034 |
|
|
(define_insn "*avx_3"
|
1035 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
1036 |
|
|
(smaxmin:AVXMODEF2P
|
1037 |
|
|
(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "%x")
|
1038 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")))]
|
1039 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
1040 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
1041 |
|
|
[(set_attr "type" "sseadd")
|
1042 |
|
|
(set_attr "prefix" "vex")
|
1043 |
|
|
(set_attr "mode" "")])
|
1044 |
|
|
|
1045 |
|
|
(define_insn "*3"
|
1046 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1047 |
|
|
(smaxmin:SSEMODEF2P
|
1048 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
1049 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")))]
|
1050 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1051 |
|
|
"p\t{%2, %0|%0, %2}"
|
1052 |
|
|
[(set_attr "type" "sseadd")
|
1053 |
|
|
(set_attr "mode" "")])
|
1054 |
|
|
|
1055 |
|
|
(define_insn "*avx_vm3"
|
1056 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1057 |
|
|
(vec_merge:SSEMODEF2P
|
1058 |
|
|
(smaxmin:SSEMODEF2P
|
1059 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
1060 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm"))
|
1061 |
|
|
(match_dup 1)
|
1062 |
|
|
(const_int 1)))]
|
1063 |
|
|
"AVX128_VEC_FLOAT_MODE_P (mode)"
|
1064 |
|
|
"vs\t{%2, %1, %0|%0, %1, %2}"
|
1065 |
|
|
[(set_attr "type" "sse")
|
1066 |
|
|
(set_attr "prefix" "vex")
|
1067 |
|
|
(set_attr "mode" "")])
|
1068 |
|
|
|
1069 |
|
|
(define_insn "_vm3"
|
1070 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1071 |
|
|
(vec_merge:SSEMODEF2P
|
1072 |
|
|
(smaxmin:SSEMODEF2P
|
1073 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
1074 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm"))
|
1075 |
|
|
(match_dup 1)
|
1076 |
|
|
(const_int 1)))]
|
1077 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1078 |
|
|
"s\t{%2, %0|%0, %2}"
|
1079 |
|
|
[(set_attr "type" "sseadd")
|
1080 |
|
|
(set_attr "mode" "")])
|
1081 |
|
|
|
1082 |
|
|
;; These versions of the min/max patterns implement exactly the operations
|
1083 |
|
|
;; min = (op1 < op2 ? op1 : op2)
|
1084 |
|
|
;; max = (!(op1 < op2) ? op1 : op2)
|
1085 |
|
|
;; Their operands are not commutative, and thus they may be used in the
|
1086 |
|
|
;; presence of -0.0 and NaN.
|
1087 |
|
|
|
1088 |
|
|
(define_insn "*avx_ieee_smin3"
|
1089 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
1090 |
|
|
(unspec:AVXMODEF2P
|
1091 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
1092 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")]
|
1093 |
|
|
UNSPEC_IEEE_MIN))]
|
1094 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
1095 |
|
|
"vminp\t{%2, %1, %0|%0, %1, %2}"
|
1096 |
|
|
[(set_attr "type" "sseadd")
|
1097 |
|
|
(set_attr "prefix" "vex")
|
1098 |
|
|
(set_attr "mode" "")])
|
1099 |
|
|
|
1100 |
|
|
(define_insn "*avx_ieee_smax3"
|
1101 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
1102 |
|
|
(unspec:AVXMODEF2P
|
1103 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
1104 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")]
|
1105 |
|
|
UNSPEC_IEEE_MAX))]
|
1106 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
1107 |
|
|
"vmaxp\t{%2, %1, %0|%0, %1, %2}"
|
1108 |
|
|
[(set_attr "type" "sseadd")
|
1109 |
|
|
(set_attr "prefix" "vex")
|
1110 |
|
|
(set_attr "mode" "")])
|
1111 |
|
|
|
1112 |
|
|
(define_insn "*ieee_smin3"
|
1113 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1114 |
|
|
(unspec:SSEMODEF2P
|
1115 |
|
|
[(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
1116 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")]
|
1117 |
|
|
UNSPEC_IEEE_MIN))]
|
1118 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1119 |
|
|
"minp\t{%2, %0|%0, %2}"
|
1120 |
|
|
[(set_attr "type" "sseadd")
|
1121 |
|
|
(set_attr "mode" "")])
|
1122 |
|
|
|
1123 |
|
|
(define_insn "*ieee_smax3"
|
1124 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1125 |
|
|
(unspec:SSEMODEF2P
|
1126 |
|
|
[(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
1127 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")]
|
1128 |
|
|
UNSPEC_IEEE_MAX))]
|
1129 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1130 |
|
|
"maxp\t{%2, %0|%0, %2}"
|
1131 |
|
|
[(set_attr "type" "sseadd")
|
1132 |
|
|
(set_attr "mode" "")])
|
1133 |
|
|
|
1134 |
|
|
(define_insn "avx_addsubv8sf3"
|
1135 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
1136 |
|
|
(vec_merge:V8SF
|
1137 |
|
|
(plus:V8SF
|
1138 |
|
|
(match_operand:V8SF 1 "register_operand" "x")
|
1139 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "xm"))
|
1140 |
|
|
(minus:V8SF (match_dup 1) (match_dup 2))
|
1141 |
|
|
(const_int 170)))]
|
1142 |
|
|
"TARGET_AVX"
|
1143 |
|
|
"vaddsubps\t{%2, %1, %0|%0, %1, %2}"
|
1144 |
|
|
[(set_attr "type" "sseadd")
|
1145 |
|
|
(set_attr "prefix" "vex")
|
1146 |
|
|
(set_attr "mode" "V8SF")])
|
1147 |
|
|
|
1148 |
|
|
(define_insn "avx_addsubv4df3"
|
1149 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x")
|
1150 |
|
|
(vec_merge:V4DF
|
1151 |
|
|
(plus:V4DF
|
1152 |
|
|
(match_operand:V4DF 1 "register_operand" "x")
|
1153 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "xm"))
|
1154 |
|
|
(minus:V4DF (match_dup 1) (match_dup 2))
|
1155 |
|
|
(const_int 10)))]
|
1156 |
|
|
"TARGET_AVX"
|
1157 |
|
|
"vaddsubpd\t{%2, %1, %0|%0, %1, %2}"
|
1158 |
|
|
[(set_attr "type" "sseadd")
|
1159 |
|
|
(set_attr "prefix" "vex")
|
1160 |
|
|
(set_attr "mode" "V4DF")])
|
1161 |
|
|
|
1162 |
|
|
(define_insn "*avx_addsubv4sf3"
|
1163 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
1164 |
|
|
(vec_merge:V4SF
|
1165 |
|
|
(plus:V4SF
|
1166 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
1167 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm"))
|
1168 |
|
|
(minus:V4SF (match_dup 1) (match_dup 2))
|
1169 |
|
|
(const_int 10)))]
|
1170 |
|
|
"TARGET_AVX"
|
1171 |
|
|
"vaddsubps\t{%2, %1, %0|%0, %1, %2}"
|
1172 |
|
|
[(set_attr "type" "sseadd")
|
1173 |
|
|
(set_attr "prefix" "vex")
|
1174 |
|
|
(set_attr "mode" "V4SF")])
|
1175 |
|
|
|
1176 |
|
|
(define_insn "sse3_addsubv4sf3"
|
1177 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
1178 |
|
|
(vec_merge:V4SF
|
1179 |
|
|
(plus:V4SF
|
1180 |
|
|
(match_operand:V4SF 1 "register_operand" "0")
|
1181 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm"))
|
1182 |
|
|
(minus:V4SF (match_dup 1) (match_dup 2))
|
1183 |
|
|
(const_int 10)))]
|
1184 |
|
|
"TARGET_SSE3"
|
1185 |
|
|
"addsubps\t{%2, %0|%0, %2}"
|
1186 |
|
|
[(set_attr "type" "sseadd")
|
1187 |
|
|
(set_attr "prefix_rep" "1")
|
1188 |
|
|
(set_attr "mode" "V4SF")])
|
1189 |
|
|
|
1190 |
|
|
(define_insn "*avx_addsubv2df3"
|
1191 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
1192 |
|
|
(vec_merge:V2DF
|
1193 |
|
|
(plus:V2DF
|
1194 |
|
|
(match_operand:V2DF 1 "register_operand" "x")
|
1195 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "xm"))
|
1196 |
|
|
(minus:V2DF (match_dup 1) (match_dup 2))
|
1197 |
|
|
(const_int 2)))]
|
1198 |
|
|
"TARGET_AVX"
|
1199 |
|
|
"vaddsubpd\t{%2, %1, %0|%0, %1, %2}"
|
1200 |
|
|
[(set_attr "type" "sseadd")
|
1201 |
|
|
(set_attr "prefix" "vex")
|
1202 |
|
|
(set_attr "mode" "V2DF")])
|
1203 |
|
|
|
1204 |
|
|
(define_insn "sse3_addsubv2df3"
|
1205 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
1206 |
|
|
(vec_merge:V2DF
|
1207 |
|
|
(plus:V2DF
|
1208 |
|
|
(match_operand:V2DF 1 "register_operand" "0")
|
1209 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "xm"))
|
1210 |
|
|
(minus:V2DF (match_dup 1) (match_dup 2))
|
1211 |
|
|
(const_int 2)))]
|
1212 |
|
|
"TARGET_SSE3"
|
1213 |
|
|
"addsubpd\t{%2, %0|%0, %2}"
|
1214 |
|
|
[(set_attr "type" "sseadd")
|
1215 |
|
|
(set_attr "atom_unit" "complex")
|
1216 |
|
|
(set_attr "mode" "V2DF")])
|
1217 |
|
|
|
1218 |
|
|
(define_insn "avx_hv4df3"
|
1219 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x")
|
1220 |
|
|
(vec_concat:V4DF
|
1221 |
|
|
(vec_concat:V2DF
|
1222 |
|
|
(plusminus:DF
|
1223 |
|
|
(vec_select:DF
|
1224 |
|
|
(match_operand:V4DF 1 "register_operand" "x")
|
1225 |
|
|
(parallel [(const_int 0)]))
|
1226 |
|
|
(vec_select:DF (match_dup 1) (parallel [(const_int 1)])))
|
1227 |
|
|
(plusminus:DF
|
1228 |
|
|
(vec_select:DF (match_dup 1) (parallel [(const_int 2)]))
|
1229 |
|
|
(vec_select:DF (match_dup 1) (parallel [(const_int 3)]))))
|
1230 |
|
|
(vec_concat:V2DF
|
1231 |
|
|
(plusminus:DF
|
1232 |
|
|
(vec_select:DF
|
1233 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "xm")
|
1234 |
|
|
(parallel [(const_int 0)]))
|
1235 |
|
|
(vec_select:DF (match_dup 2) (parallel [(const_int 1)])))
|
1236 |
|
|
(plusminus:DF
|
1237 |
|
|
(vec_select:DF (match_dup 2) (parallel [(const_int 2)]))
|
1238 |
|
|
(vec_select:DF (match_dup 2) (parallel [(const_int 3)]))))))]
|
1239 |
|
|
"TARGET_AVX"
|
1240 |
|
|
"vhpd\t{%2, %1, %0|%0, %1, %2}"
|
1241 |
|
|
[(set_attr "type" "sseadd")
|
1242 |
|
|
(set_attr "prefix" "vex")
|
1243 |
|
|
(set_attr "mode" "V4DF")])
|
1244 |
|
|
|
1245 |
|
|
(define_insn "avx_hv8sf3"
|
1246 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
1247 |
|
|
(vec_concat:V8SF
|
1248 |
|
|
(vec_concat:V4SF
|
1249 |
|
|
(vec_concat:V2SF
|
1250 |
|
|
(plusminus:SF
|
1251 |
|
|
(vec_select:SF
|
1252 |
|
|
(match_operand:V8SF 1 "register_operand" "x")
|
1253 |
|
|
(parallel [(const_int 0)]))
|
1254 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 1)])))
|
1255 |
|
|
(plusminus:SF
|
1256 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 2)]))
|
1257 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 3)]))))
|
1258 |
|
|
(vec_concat:V2SF
|
1259 |
|
|
(plusminus:SF
|
1260 |
|
|
(vec_select:SF
|
1261 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "xm")
|
1262 |
|
|
(parallel [(const_int 0)]))
|
1263 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 1)])))
|
1264 |
|
|
(plusminus:SF
|
1265 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 2)]))
|
1266 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 3)])))))
|
1267 |
|
|
(vec_concat:V4SF
|
1268 |
|
|
(vec_concat:V2SF
|
1269 |
|
|
(plusminus:SF
|
1270 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 4)]))
|
1271 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 5)])))
|
1272 |
|
|
(plusminus:SF
|
1273 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 6)]))
|
1274 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 7)]))))
|
1275 |
|
|
(vec_concat:V2SF
|
1276 |
|
|
(plusminus:SF
|
1277 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 4)]))
|
1278 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 5)])))
|
1279 |
|
|
(plusminus:SF
|
1280 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 6)]))
|
1281 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 7)])))))))]
|
1282 |
|
|
"TARGET_AVX"
|
1283 |
|
|
"vhps\t{%2, %1, %0|%0, %1, %2}"
|
1284 |
|
|
[(set_attr "type" "sseadd")
|
1285 |
|
|
(set_attr "prefix" "vex")
|
1286 |
|
|
(set_attr "mode" "V8SF")])
|
1287 |
|
|
|
1288 |
|
|
(define_insn "*avx_hv4sf3"
|
1289 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
1290 |
|
|
(vec_concat:V4SF
|
1291 |
|
|
(vec_concat:V2SF
|
1292 |
|
|
(plusminus:SF
|
1293 |
|
|
(vec_select:SF
|
1294 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
1295 |
|
|
(parallel [(const_int 0)]))
|
1296 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 1)])))
|
1297 |
|
|
(plusminus:SF
|
1298 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 2)]))
|
1299 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 3)]))))
|
1300 |
|
|
(vec_concat:V2SF
|
1301 |
|
|
(plusminus:SF
|
1302 |
|
|
(vec_select:SF
|
1303 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm")
|
1304 |
|
|
(parallel [(const_int 0)]))
|
1305 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 1)])))
|
1306 |
|
|
(plusminus:SF
|
1307 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 2)]))
|
1308 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 3)]))))))]
|
1309 |
|
|
"TARGET_AVX"
|
1310 |
|
|
"vhps\t{%2, %1, %0|%0, %1, %2}"
|
1311 |
|
|
[(set_attr "type" "sseadd")
|
1312 |
|
|
(set_attr "prefix" "vex")
|
1313 |
|
|
(set_attr "mode" "V4SF")])
|
1314 |
|
|
|
1315 |
|
|
(define_insn "sse3_hv4sf3"
|
1316 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
1317 |
|
|
(vec_concat:V4SF
|
1318 |
|
|
(vec_concat:V2SF
|
1319 |
|
|
(plusminus:SF
|
1320 |
|
|
(vec_select:SF
|
1321 |
|
|
(match_operand:V4SF 1 "register_operand" "0")
|
1322 |
|
|
(parallel [(const_int 0)]))
|
1323 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 1)])))
|
1324 |
|
|
(plusminus:SF
|
1325 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 2)]))
|
1326 |
|
|
(vec_select:SF (match_dup 1) (parallel [(const_int 3)]))))
|
1327 |
|
|
(vec_concat:V2SF
|
1328 |
|
|
(plusminus:SF
|
1329 |
|
|
(vec_select:SF
|
1330 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm")
|
1331 |
|
|
(parallel [(const_int 0)]))
|
1332 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 1)])))
|
1333 |
|
|
(plusminus:SF
|
1334 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 2)]))
|
1335 |
|
|
(vec_select:SF (match_dup 2) (parallel [(const_int 3)]))))))]
|
1336 |
|
|
"TARGET_SSE3"
|
1337 |
|
|
"hps\t{%2, %0|%0, %2}"
|
1338 |
|
|
[(set_attr "type" "sseadd")
|
1339 |
|
|
(set_attr "atom_unit" "complex")
|
1340 |
|
|
(set_attr "prefix_rep" "1")
|
1341 |
|
|
(set_attr "mode" "V4SF")])
|
1342 |
|
|
|
1343 |
|
|
(define_insn "*avx_hv2df3"
|
1344 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
1345 |
|
|
(vec_concat:V2DF
|
1346 |
|
|
(plusminus:DF
|
1347 |
|
|
(vec_select:DF
|
1348 |
|
|
(match_operand:V2DF 1 "register_operand" "x")
|
1349 |
|
|
(parallel [(const_int 0)]))
|
1350 |
|
|
(vec_select:DF (match_dup 1) (parallel [(const_int 1)])))
|
1351 |
|
|
(plusminus:DF
|
1352 |
|
|
(vec_select:DF
|
1353 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "xm")
|
1354 |
|
|
(parallel [(const_int 0)]))
|
1355 |
|
|
(vec_select:DF (match_dup 2) (parallel [(const_int 1)])))))]
|
1356 |
|
|
"TARGET_AVX"
|
1357 |
|
|
"vhpd\t{%2, %1, %0|%0, %1, %2}"
|
1358 |
|
|
[(set_attr "type" "sseadd")
|
1359 |
|
|
(set_attr "prefix" "vex")
|
1360 |
|
|
(set_attr "mode" "V2DF")])
|
1361 |
|
|
|
1362 |
|
|
(define_insn "sse3_hv2df3"
|
1363 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
1364 |
|
|
(vec_concat:V2DF
|
1365 |
|
|
(plusminus:DF
|
1366 |
|
|
(vec_select:DF
|
1367 |
|
|
(match_operand:V2DF 1 "register_operand" "0")
|
1368 |
|
|
(parallel [(const_int 0)]))
|
1369 |
|
|
(vec_select:DF (match_dup 1) (parallel [(const_int 1)])))
|
1370 |
|
|
(plusminus:DF
|
1371 |
|
|
(vec_select:DF
|
1372 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "xm")
|
1373 |
|
|
(parallel [(const_int 0)]))
|
1374 |
|
|
(vec_select:DF (match_dup 2) (parallel [(const_int 1)])))))]
|
1375 |
|
|
"TARGET_SSE3"
|
1376 |
|
|
"hpd\t{%2, %0|%0, %2}"
|
1377 |
|
|
[(set_attr "type" "sseadd")
|
1378 |
|
|
(set_attr "mode" "V2DF")])
|
1379 |
|
|
|
1380 |
|
|
(define_expand "reduc_splus_v4sf"
|
1381 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
1382 |
|
|
(match_operand:V4SF 1 "register_operand" "")]
|
1383 |
|
|
"TARGET_SSE"
|
1384 |
|
|
{
|
1385 |
|
|
if (TARGET_SSE3)
|
1386 |
|
|
{
|
1387 |
|
|
rtx tmp = gen_reg_rtx (V4SFmode);
|
1388 |
|
|
emit_insn (gen_sse3_haddv4sf3 (tmp, operands[1], operands[1]));
|
1389 |
|
|
emit_insn (gen_sse3_haddv4sf3 (operands[0], tmp, tmp));
|
1390 |
|
|
}
|
1391 |
|
|
else
|
1392 |
|
|
ix86_expand_reduc_v4sf (gen_addv4sf3, operands[0], operands[1]);
|
1393 |
|
|
DONE;
|
1394 |
|
|
})
|
1395 |
|
|
|
1396 |
|
|
(define_expand "reduc_splus_v2df"
|
1397 |
|
|
[(match_operand:V2DF 0 "register_operand" "")
|
1398 |
|
|
(match_operand:V2DF 1 "register_operand" "")]
|
1399 |
|
|
"TARGET_SSE3"
|
1400 |
|
|
{
|
1401 |
|
|
emit_insn (gen_sse3_haddv2df3 (operands[0], operands[1], operands[1]));
|
1402 |
|
|
DONE;
|
1403 |
|
|
})
|
1404 |
|
|
|
1405 |
|
|
(define_expand "reduc_smax_v4sf"
|
1406 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
1407 |
|
|
(match_operand:V4SF 1 "register_operand" "")]
|
1408 |
|
|
"TARGET_SSE"
|
1409 |
|
|
{
|
1410 |
|
|
ix86_expand_reduc_v4sf (gen_smaxv4sf3, operands[0], operands[1]);
|
1411 |
|
|
DONE;
|
1412 |
|
|
})
|
1413 |
|
|
|
1414 |
|
|
(define_expand "reduc_smin_v4sf"
|
1415 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
1416 |
|
|
(match_operand:V4SF 1 "register_operand" "")]
|
1417 |
|
|
"TARGET_SSE"
|
1418 |
|
|
{
|
1419 |
|
|
ix86_expand_reduc_v4sf (gen_sminv4sf3, operands[0], operands[1]);
|
1420 |
|
|
DONE;
|
1421 |
|
|
})
|
1422 |
|
|
|
1423 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
1424 |
|
|
;;
|
1425 |
|
|
;; Parallel floating point comparisons
|
1426 |
|
|
;;
|
1427 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
1428 |
|
|
|
1429 |
|
|
(define_insn "avx_cmpp3"
|
1430 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
1431 |
|
|
(unspec:AVXMODEF2P
|
1432 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
1433 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")
|
1434 |
|
|
(match_operand:SI 3 "const_0_to_31_operand" "n")]
|
1435 |
|
|
UNSPEC_PCMP))]
|
1436 |
|
|
"TARGET_AVX"
|
1437 |
|
|
"vcmpp\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1438 |
|
|
[(set_attr "type" "ssecmp")
|
1439 |
|
|
(set_attr "length_immediate" "1")
|
1440 |
|
|
(set_attr "prefix" "vex")
|
1441 |
|
|
(set_attr "mode" "")])
|
1442 |
|
|
|
1443 |
|
|
(define_insn "avx_cmps3"
|
1444 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "")
|
1445 |
|
|
(vec_merge:SSEMODEF2P
|
1446 |
|
|
(unspec:SSEMODEF2P
|
1447 |
|
|
[(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
1448 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")
|
1449 |
|
|
(match_operand:SI 3 "const_0_to_31_operand" "n")]
|
1450 |
|
|
UNSPEC_PCMP)
|
1451 |
|
|
(match_dup 1)
|
1452 |
|
|
(const_int 1)))]
|
1453 |
|
|
"TARGET_AVX"
|
1454 |
|
|
"vcmps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1455 |
|
|
[(set_attr "type" "ssecmp")
|
1456 |
|
|
(set_attr "length_immediate" "1")
|
1457 |
|
|
(set_attr "prefix" "vex")
|
1458 |
|
|
(set_attr "mode" "")])
|
1459 |
|
|
|
1460 |
|
|
;; We don't promote 128bit vector compare intrinsics. But vectorizer
|
1461 |
|
|
;; may generate 256bit vector compare instructions.
|
1462 |
|
|
(define_insn "*avx_maskcmp3"
|
1463 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
1464 |
|
|
(match_operator:AVXMODEF2P 3 "avx_comparison_float_operator"
|
1465 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
1466 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")]))]
|
1467 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
1468 |
|
|
"vcmp%D3p\t{%2, %1, %0|%0, %1, %2}"
|
1469 |
|
|
[(set_attr "type" "ssecmp")
|
1470 |
|
|
(set_attr "prefix" "vex")
|
1471 |
|
|
(set_attr "length_immediate" "1")
|
1472 |
|
|
(set_attr "mode" "")])
|
1473 |
|
|
|
1474 |
|
|
(define_insn "_maskcmp3"
|
1475 |
|
|
[(set (match_operand:SSEMODEF4 0 "register_operand" "=x")
|
1476 |
|
|
(match_operator:SSEMODEF4 3 "sse_comparison_operator"
|
1477 |
|
|
[(match_operand:SSEMODEF4 1 "register_operand" "0")
|
1478 |
|
|
(match_operand:SSEMODEF4 2 "nonimmediate_operand" "xm")]))]
|
1479 |
|
|
"!TARGET_XOP
|
1480 |
|
|
&& (SSE_FLOAT_MODE_P (mode) || SSE_VEC_FLOAT_MODE_P (mode))"
|
1481 |
|
|
"cmp%D3\t{%2, %0|%0, %2}"
|
1482 |
|
|
[(set_attr "type" "ssecmp")
|
1483 |
|
|
(set_attr "length_immediate" "1")
|
1484 |
|
|
(set_attr "mode" "")])
|
1485 |
|
|
|
1486 |
|
|
(define_insn "*avx_vmmaskcmp3"
|
1487 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1488 |
|
|
(vec_merge:SSEMODEF2P
|
1489 |
|
|
(match_operator:SSEMODEF2P 3 "sse_comparison_operator"
|
1490 |
|
|
[(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
1491 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")])
|
1492 |
|
|
(match_dup 1)
|
1493 |
|
|
(const_int 1)))]
|
1494 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
1495 |
|
|
"vcmp%D3s\t{%2, %1, %0|%0, %1, %2}"
|
1496 |
|
|
[(set_attr "type" "ssecmp")
|
1497 |
|
|
(set_attr "prefix" "vex")
|
1498 |
|
|
(set_attr "mode" "")])
|
1499 |
|
|
|
1500 |
|
|
(define_insn "_vmmaskcmp3"
|
1501 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1502 |
|
|
(vec_merge:SSEMODEF2P
|
1503 |
|
|
(match_operator:SSEMODEF2P 3 "sse_comparison_operator"
|
1504 |
|
|
[(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
1505 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")])
|
1506 |
|
|
(match_dup 1)
|
1507 |
|
|
(const_int 1)))]
|
1508 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1509 |
|
|
"cmp%D3s\t{%2, %0|%0, %2}"
|
1510 |
|
|
[(set_attr "type" "ssecmp")
|
1511 |
|
|
(set_attr "length_immediate" "1")
|
1512 |
|
|
(set_attr "mode" "")])
|
1513 |
|
|
|
1514 |
|
|
(define_insn "_comi"
|
1515 |
|
|
[(set (reg:CCFP FLAGS_REG)
|
1516 |
|
|
(compare:CCFP
|
1517 |
|
|
(vec_select:MODEF
|
1518 |
|
|
(match_operand: 0 "register_operand" "x")
|
1519 |
|
|
(parallel [(const_int 0)]))
|
1520 |
|
|
(vec_select:MODEF
|
1521 |
|
|
(match_operand: 1 "nonimmediate_operand" "xm")
|
1522 |
|
|
(parallel [(const_int 0)]))))]
|
1523 |
|
|
"SSE_FLOAT_MODE_P (mode)"
|
1524 |
|
|
"%vcomis\t{%1, %0|%0, %1}"
|
1525 |
|
|
[(set_attr "type" "ssecomi")
|
1526 |
|
|
(set_attr "prefix" "maybe_vex")
|
1527 |
|
|
(set_attr "prefix_rep" "0")
|
1528 |
|
|
(set (attr "prefix_data16")
|
1529 |
|
|
(if_then_else (eq_attr "mode" "DF")
|
1530 |
|
|
(const_string "1")
|
1531 |
|
|
(const_string "0")))
|
1532 |
|
|
(set_attr "mode" "")])
|
1533 |
|
|
|
1534 |
|
|
(define_insn "_ucomi"
|
1535 |
|
|
[(set (reg:CCFPU FLAGS_REG)
|
1536 |
|
|
(compare:CCFPU
|
1537 |
|
|
(vec_select:MODEF
|
1538 |
|
|
(match_operand: 0 "register_operand" "x")
|
1539 |
|
|
(parallel [(const_int 0)]))
|
1540 |
|
|
(vec_select:MODEF
|
1541 |
|
|
(match_operand: 1 "nonimmediate_operand" "xm")
|
1542 |
|
|
(parallel [(const_int 0)]))))]
|
1543 |
|
|
"SSE_FLOAT_MODE_P (mode)"
|
1544 |
|
|
"%vucomis\t{%1, %0|%0, %1}"
|
1545 |
|
|
[(set_attr "type" "ssecomi")
|
1546 |
|
|
(set_attr "prefix" "maybe_vex")
|
1547 |
|
|
(set_attr "prefix_rep" "0")
|
1548 |
|
|
(set (attr "prefix_data16")
|
1549 |
|
|
(if_then_else (eq_attr "mode" "DF")
|
1550 |
|
|
(const_string "1")
|
1551 |
|
|
(const_string "0")))
|
1552 |
|
|
(set_attr "mode" "")])
|
1553 |
|
|
|
1554 |
|
|
(define_expand "vcond"
|
1555 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "")
|
1556 |
|
|
(if_then_else:SSEMODEF2P
|
1557 |
|
|
(match_operator 3 ""
|
1558 |
|
|
[(match_operand:SSEMODEF2P 4 "nonimmediate_operand" "")
|
1559 |
|
|
(match_operand:SSEMODEF2P 5 "nonimmediate_operand" "")])
|
1560 |
|
|
(match_operand:SSEMODEF2P 1 "general_operand" "")
|
1561 |
|
|
(match_operand:SSEMODEF2P 2 "general_operand" "")))]
|
1562 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1563 |
|
|
{
|
1564 |
|
|
bool ok = ix86_expand_fp_vcond (operands);
|
1565 |
|
|
gcc_assert (ok);
|
1566 |
|
|
DONE;
|
1567 |
|
|
})
|
1568 |
|
|
|
1569 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
1570 |
|
|
;;
|
1571 |
|
|
;; Parallel floating point logical operations
|
1572 |
|
|
;;
|
1573 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
1574 |
|
|
|
1575 |
|
|
(define_insn "avx_andnot3"
|
1576 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
1577 |
|
|
(and:AVXMODEF2P
|
1578 |
|
|
(not:AVXMODEF2P
|
1579 |
|
|
(match_operand:AVXMODEF2P 1 "register_operand" "x"))
|
1580 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")))]
|
1581 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)"
|
1582 |
|
|
"vandnp\t{%2, %1, %0|%0, %1, %2}"
|
1583 |
|
|
[(set_attr "type" "sselog")
|
1584 |
|
|
(set_attr "prefix" "vex")
|
1585 |
|
|
(set_attr "mode" "")])
|
1586 |
|
|
|
1587 |
|
|
(define_insn "_andnot3"
|
1588 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1589 |
|
|
(and:SSEMODEF2P
|
1590 |
|
|
(not:SSEMODEF2P
|
1591 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0"))
|
1592 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")))]
|
1593 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1594 |
|
|
"andnp\t{%2, %0|%0, %2}"
|
1595 |
|
|
[(set_attr "type" "sselog")
|
1596 |
|
|
(set_attr "mode" "")])
|
1597 |
|
|
|
1598 |
|
|
(define_expand "3"
|
1599 |
|
|
[(set (match_operand:AVX256MODEF2P 0 "register_operand" "")
|
1600 |
|
|
(any_logic:AVX256MODEF2P
|
1601 |
|
|
(match_operand:AVX256MODEF2P 1 "nonimmediate_operand" "")
|
1602 |
|
|
(match_operand:AVX256MODEF2P 2 "nonimmediate_operand" "")))]
|
1603 |
|
|
"AVX256_VEC_FLOAT_MODE_P (mode)"
|
1604 |
|
|
"ix86_fixup_binary_operands_no_copy (, mode, operands);")
|
1605 |
|
|
|
1606 |
|
|
(define_insn "*avx_3"
|
1607 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
1608 |
|
|
(any_logic:AVXMODEF2P
|
1609 |
|
|
(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "%x")
|
1610 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")))]
|
1611 |
|
|
"AVX_VEC_FLOAT_MODE_P (mode)
|
1612 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
1613 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
1614 |
|
|
[(set_attr "type" "sselog")
|
1615 |
|
|
(set_attr "prefix" "vex")
|
1616 |
|
|
(set_attr "mode" "")])
|
1617 |
|
|
|
1618 |
|
|
(define_expand "3"
|
1619 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "")
|
1620 |
|
|
(any_logic:SSEMODEF2P
|
1621 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "")
|
1622 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "")))]
|
1623 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1624 |
|
|
"ix86_fixup_binary_operands_no_copy (, mode, operands);")
|
1625 |
|
|
|
1626 |
|
|
(define_insn "*3"
|
1627 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
1628 |
|
|
(any_logic:SSEMODEF2P
|
1629 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%0")
|
1630 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")))]
|
1631 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)
|
1632 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
1633 |
|
|
"p\t{%2, %0|%0, %2}"
|
1634 |
|
|
[(set_attr "type" "sselog")
|
1635 |
|
|
(set_attr "mode" "")])
|
1636 |
|
|
|
1637 |
|
|
(define_expand "copysign3"
|
1638 |
|
|
[(set (match_dup 4)
|
1639 |
|
|
(and:SSEMODEF2P
|
1640 |
|
|
(not:SSEMODEF2P (match_dup 3))
|
1641 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "")))
|
1642 |
|
|
(set (match_dup 5)
|
1643 |
|
|
(and:SSEMODEF2P (match_dup 3)
|
1644 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "")))
|
1645 |
|
|
(set (match_operand:SSEMODEF2P 0 "register_operand" "")
|
1646 |
|
|
(ior:SSEMODEF2P (match_dup 4) (match_dup 5)))]
|
1647 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
1648 |
|
|
{
|
1649 |
|
|
operands[3] = ix86_build_signbit_mask (mode, 1, 0);
|
1650 |
|
|
|
1651 |
|
|
operands[4] = gen_reg_rtx (mode);
|
1652 |
|
|
operands[5] = gen_reg_rtx (mode);
|
1653 |
|
|
})
|
1654 |
|
|
|
1655 |
|
|
;; Also define scalar versions. These are used for abs, neg, and
|
1656 |
|
|
;; conditional move. Using subregs into vector modes causes register
|
1657 |
|
|
;; allocation lossage. These patterns do not allow memory operands
|
1658 |
|
|
;; because the native instructions read the full 128-bits.
|
1659 |
|
|
|
1660 |
|
|
(define_insn "*avx_andnot3"
|
1661 |
|
|
[(set (match_operand:MODEF 0 "register_operand" "=x")
|
1662 |
|
|
(and:MODEF
|
1663 |
|
|
(not:MODEF
|
1664 |
|
|
(match_operand:MODEF 1 "register_operand" "x"))
|
1665 |
|
|
(match_operand:MODEF 2 "register_operand" "x")))]
|
1666 |
|
|
"AVX_FLOAT_MODE_P (mode)"
|
1667 |
|
|
"vandnp\t{%2, %1, %0|%0, %1, %2}"
|
1668 |
|
|
[(set_attr "type" "sselog")
|
1669 |
|
|
(set_attr "prefix" "vex")
|
1670 |
|
|
(set_attr "mode" "")])
|
1671 |
|
|
|
1672 |
|
|
(define_insn "*andnot3"
|
1673 |
|
|
[(set (match_operand:MODEF 0 "register_operand" "=x")
|
1674 |
|
|
(and:MODEF
|
1675 |
|
|
(not:MODEF
|
1676 |
|
|
(match_operand:MODEF 1 "register_operand" "0"))
|
1677 |
|
|
(match_operand:MODEF 2 "register_operand" "x")))]
|
1678 |
|
|
"SSE_FLOAT_MODE_P (mode)"
|
1679 |
|
|
"andnp\t{%2, %0|%0, %2}"
|
1680 |
|
|
[(set_attr "type" "sselog")
|
1681 |
|
|
(set_attr "mode" "")])
|
1682 |
|
|
|
1683 |
|
|
(define_insn "*avx_3"
|
1684 |
|
|
[(set (match_operand:MODEF 0 "register_operand" "=x")
|
1685 |
|
|
(any_logic:MODEF
|
1686 |
|
|
(match_operand:MODEF 1 "register_operand" "x")
|
1687 |
|
|
(match_operand:MODEF 2 "register_operand" "x")))]
|
1688 |
|
|
"AVX_FLOAT_MODE_P (mode)"
|
1689 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
1690 |
|
|
[(set_attr "type" "sselog")
|
1691 |
|
|
(set_attr "prefix" "vex")
|
1692 |
|
|
(set_attr "mode" "")])
|
1693 |
|
|
|
1694 |
|
|
(define_insn "*3"
|
1695 |
|
|
[(set (match_operand:MODEF 0 "register_operand" "=x")
|
1696 |
|
|
(any_logic:MODEF
|
1697 |
|
|
(match_operand:MODEF 1 "register_operand" "0")
|
1698 |
|
|
(match_operand:MODEF 2 "register_operand" "x")))]
|
1699 |
|
|
"SSE_FLOAT_MODE_P (mode)"
|
1700 |
|
|
"p\t{%2, %0|%0, %2}"
|
1701 |
|
|
[(set_attr "type" "sselog")
|
1702 |
|
|
(set_attr "mode" "")])
|
1703 |
|
|
|
1704 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
1705 |
|
|
;;
|
1706 |
|
|
;; FMA4 floating point multiply/accumulate instructions. This
|
1707 |
|
|
;; includes the scalar version of the instructions as well as the
|
1708 |
|
|
;; vector.
|
1709 |
|
|
;;
|
1710 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
1711 |
|
|
|
1712 |
|
|
;; In order to match (*a * *b) + *c, particularly when vectorizing, allow
|
1713 |
|
|
;; combine to generate a multiply/add with two memory references. We then
|
1714 |
|
|
;; split this insn, into loading up the destination register with one of the
|
1715 |
|
|
;; memory operations. If we don't manage to split the insn, reload will
|
1716 |
|
|
;; generate the appropriate moves. The reason this is needed, is that combine
|
1717 |
|
|
;; has already folded one of the memory references into both the multiply and
|
1718 |
|
|
;; add insns, and it can't generate a new pseudo. I.e.:
|
1719 |
|
|
;; (set (reg1) (mem (addr1)))
|
1720 |
|
|
;; (set (reg2) (mult (reg1) (mem (addr2))))
|
1721 |
|
|
;; (set (reg3) (plus (reg2) (mem (addr3))))
|
1722 |
|
|
|
1723 |
|
|
(define_insn "fma4_fmadd4256"
|
1724 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x,x")
|
1725 |
|
|
(plus:FMA4MODEF4
|
1726 |
|
|
(mult:FMA4MODEF4
|
1727 |
|
|
(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "%x,x")
|
1728 |
|
|
(match_operand:FMA4MODEF4 2 "nonimmediate_operand" "x,m"))
|
1729 |
|
|
(match_operand:FMA4MODEF4 3 "nonimmediate_operand" "xm,x")))]
|
1730 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1731 |
|
|
"vfmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1732 |
|
|
[(set_attr "type" "ssemuladd")
|
1733 |
|
|
(set_attr "mode" "")])
|
1734 |
|
|
|
1735 |
|
|
;; Floating multiply and subtract.
|
1736 |
|
|
(define_insn "fma4_fmsub4256"
|
1737 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x,x")
|
1738 |
|
|
(minus:FMA4MODEF4
|
1739 |
|
|
(mult:FMA4MODEF4
|
1740 |
|
|
(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "%x,x")
|
1741 |
|
|
(match_operand:FMA4MODEF4 2 "nonimmediate_operand" "x,m"))
|
1742 |
|
|
(match_operand:FMA4MODEF4 3 "nonimmediate_operand" "xm,x")))]
|
1743 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1744 |
|
|
"vfmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1745 |
|
|
[(set_attr "type" "ssemuladd")
|
1746 |
|
|
(set_attr "mode" "")])
|
1747 |
|
|
|
1748 |
|
|
;; Floating point negative multiply and add.
|
1749 |
|
|
;; Rewrite (- (a * b) + c) into the canonical form: c - (a * b).
|
1750 |
|
|
(define_insn "fma4_fnmadd4256"
|
1751 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x,x")
|
1752 |
|
|
(minus:FMA4MODEF4
|
1753 |
|
|
(match_operand:FMA4MODEF4 3 "nonimmediate_operand" "xm,x")
|
1754 |
|
|
(mult:FMA4MODEF4
|
1755 |
|
|
(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "%x,x")
|
1756 |
|
|
(match_operand:FMA4MODEF4 2 "nonimmediate_operand" "x,m"))))]
|
1757 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1758 |
|
|
"vfnmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1759 |
|
|
[(set_attr "type" "ssemuladd")
|
1760 |
|
|
(set_attr "mode" "")])
|
1761 |
|
|
|
1762 |
|
|
;; Floating point negative multiply and subtract.
|
1763 |
|
|
(define_insn "fma4_fnmsub4256"
|
1764 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x,x")
|
1765 |
|
|
(minus:FMA4MODEF4
|
1766 |
|
|
(mult:FMA4MODEF4
|
1767 |
|
|
(neg:FMA4MODEF4
|
1768 |
|
|
(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "%x,x"))
|
1769 |
|
|
(match_operand:FMA4MODEF4 2 "nonimmediate_operand" "x,m"))
|
1770 |
|
|
(match_operand:FMA4MODEF4 3 "nonimmediate_operand" "xm,x")))]
|
1771 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1772 |
|
|
"vfnmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1773 |
|
|
[(set_attr "type" "ssemuladd")
|
1774 |
|
|
(set_attr "mode" "")])
|
1775 |
|
|
|
1776 |
|
|
(define_insn "fma4_fmadd4"
|
1777 |
|
|
[(set (match_operand:SSEMODEF4 0 "register_operand" "=x,x")
|
1778 |
|
|
(plus:SSEMODEF4
|
1779 |
|
|
(mult:SSEMODEF4
|
1780 |
|
|
(match_operand:SSEMODEF4 1 "nonimmediate_operand" "%x,x")
|
1781 |
|
|
(match_operand:SSEMODEF4 2 "nonimmediate_operand" "x,m"))
|
1782 |
|
|
(match_operand:SSEMODEF4 3 "nonimmediate_operand" "xm,x")))]
|
1783 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1784 |
|
|
"vfmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1785 |
|
|
[(set_attr "type" "ssemuladd")
|
1786 |
|
|
(set_attr "mode" "")])
|
1787 |
|
|
|
1788 |
|
|
;; For the scalar operations, use operand1 for the upper words that aren't
|
1789 |
|
|
;; modified, so restrict the forms that are generated.
|
1790 |
|
|
;; Scalar version of fmadd.
|
1791 |
|
|
(define_insn "fma4_vmfmadd4"
|
1792 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
1793 |
|
|
(vec_merge:SSEMODEF2P
|
1794 |
|
|
(plus:SSEMODEF2P
|
1795 |
|
|
(mult:SSEMODEF2P
|
1796 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
1797 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
1798 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))
|
1799 |
|
|
(match_dup 0)
|
1800 |
|
|
(const_int 1)))]
|
1801 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1802 |
|
|
"vfmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1803 |
|
|
[(set_attr "type" "ssemuladd")
|
1804 |
|
|
(set_attr "mode" "")])
|
1805 |
|
|
|
1806 |
|
|
;; Floating multiply and subtract.
|
1807 |
|
|
;; Allow two memory operands the same as fmadd.
|
1808 |
|
|
(define_insn "fma4_fmsub4"
|
1809 |
|
|
[(set (match_operand:SSEMODEF4 0 "register_operand" "=x,x")
|
1810 |
|
|
(minus:SSEMODEF4
|
1811 |
|
|
(mult:SSEMODEF4
|
1812 |
|
|
(match_operand:SSEMODEF4 1 "nonimmediate_operand" "%x,x")
|
1813 |
|
|
(match_operand:SSEMODEF4 2 "nonimmediate_operand" "x,m"))
|
1814 |
|
|
(match_operand:SSEMODEF4 3 "nonimmediate_operand" "xm,x")))]
|
1815 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1816 |
|
|
"vfmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1817 |
|
|
[(set_attr "type" "ssemuladd")
|
1818 |
|
|
(set_attr "mode" "")])
|
1819 |
|
|
|
1820 |
|
|
;; For the scalar operations, use operand1 for the upper words that aren't
|
1821 |
|
|
;; modified, so restrict the forms that are generated.
|
1822 |
|
|
;; Scalar version of fmsub.
|
1823 |
|
|
(define_insn "fma4_vmfmsub4"
|
1824 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
1825 |
|
|
(vec_merge:SSEMODEF2P
|
1826 |
|
|
(minus:SSEMODEF2P
|
1827 |
|
|
(mult:SSEMODEF2P
|
1828 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
1829 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
1830 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))
|
1831 |
|
|
(match_dup 0)
|
1832 |
|
|
(const_int 1)))]
|
1833 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1834 |
|
|
"vfmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1835 |
|
|
[(set_attr "type" "ssemuladd")
|
1836 |
|
|
(set_attr "mode" "")])
|
1837 |
|
|
|
1838 |
|
|
;; Floating point negative multiply and add.
|
1839 |
|
|
;; Rewrite (- (a * b) + c) into the canonical form: c - (a * b).
|
1840 |
|
|
(define_insn "fma4_fnmadd4"
|
1841 |
|
|
[(set (match_operand:SSEMODEF4 0 "register_operand" "=x,x")
|
1842 |
|
|
(minus:SSEMODEF4
|
1843 |
|
|
(match_operand:SSEMODEF4 3 "nonimmediate_operand" "xm,x")
|
1844 |
|
|
(mult:SSEMODEF4
|
1845 |
|
|
(match_operand:SSEMODEF4 1 "nonimmediate_operand" "%x,x")
|
1846 |
|
|
(match_operand:SSEMODEF4 2 "nonimmediate_operand" "x,m"))))]
|
1847 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1848 |
|
|
"vfnmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1849 |
|
|
[(set_attr "type" "ssemuladd")
|
1850 |
|
|
(set_attr "mode" "")])
|
1851 |
|
|
|
1852 |
|
|
;; For the scalar operations, use operand1 for the upper words that aren't
|
1853 |
|
|
;; modified, so restrict the forms that are generated.
|
1854 |
|
|
;; Scalar version of fnmadd.
|
1855 |
|
|
(define_insn "fma4_vmfnmadd4"
|
1856 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
1857 |
|
|
(vec_merge:SSEMODEF2P
|
1858 |
|
|
(minus:SSEMODEF2P
|
1859 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x")
|
1860 |
|
|
(mult:SSEMODEF2P
|
1861 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
1862 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m")))
|
1863 |
|
|
(match_dup 0)
|
1864 |
|
|
(const_int 1)))]
|
1865 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1866 |
|
|
"vfnmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1867 |
|
|
[(set_attr "type" "ssemuladd")
|
1868 |
|
|
(set_attr "mode" "")])
|
1869 |
|
|
|
1870 |
|
|
;; Floating point negative multiply and subtract.
|
1871 |
|
|
;; Rewrite (- (a * b) - c) into the canonical form: ((-a) * b) - c.
|
1872 |
|
|
(define_insn "fma4_fnmsub4"
|
1873 |
|
|
[(set (match_operand:SSEMODEF4 0 "register_operand" "=x,x")
|
1874 |
|
|
(minus:SSEMODEF4
|
1875 |
|
|
(mult:SSEMODEF4
|
1876 |
|
|
(neg:SSEMODEF4
|
1877 |
|
|
(match_operand:SSEMODEF4 1 "nonimmediate_operand" "%x,x"))
|
1878 |
|
|
(match_operand:SSEMODEF4 2 "nonimmediate_operand" "x,m"))
|
1879 |
|
|
(match_operand:SSEMODEF4 3 "nonimmediate_operand" "xm,x")))]
|
1880 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1881 |
|
|
"vfnmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1882 |
|
|
[(set_attr "type" "ssemuladd")
|
1883 |
|
|
(set_attr "mode" "")])
|
1884 |
|
|
|
1885 |
|
|
;; For the scalar operations, use operand1 for the upper words that aren't
|
1886 |
|
|
;; modified, so restrict the forms that are generated.
|
1887 |
|
|
;; Scalar version of fnmsub.
|
1888 |
|
|
(define_insn "fma4_vmfnmsub4"
|
1889 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
1890 |
|
|
(vec_merge:SSEMODEF2P
|
1891 |
|
|
(minus:SSEMODEF2P
|
1892 |
|
|
(mult:SSEMODEF2P
|
1893 |
|
|
(neg:SSEMODEF2P
|
1894 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x"))
|
1895 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
1896 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))
|
1897 |
|
|
(match_dup 0)
|
1898 |
|
|
(const_int 1)))]
|
1899 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
1900 |
|
|
"vfnmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1901 |
|
|
[(set_attr "type" "ssemuladd")
|
1902 |
|
|
(set_attr "mode" "")])
|
1903 |
|
|
|
1904 |
|
|
(define_insn "fma4i_fmadd4256"
|
1905 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x,x")
|
1906 |
|
|
(unspec:FMA4MODEF4
|
1907 |
|
|
[(plus:FMA4MODEF4
|
1908 |
|
|
(mult:FMA4MODEF4
|
1909 |
|
|
(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "%x,x")
|
1910 |
|
|
(match_operand:FMA4MODEF4 2 "nonimmediate_operand" "x,m"))
|
1911 |
|
|
(match_operand:FMA4MODEF4 3 "nonimmediate_operand" "xm,x"))]
|
1912 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
1913 |
|
|
"TARGET_FMA4"
|
1914 |
|
|
"vfmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1915 |
|
|
[(set_attr "type" "ssemuladd")
|
1916 |
|
|
(set_attr "mode" "")])
|
1917 |
|
|
|
1918 |
|
|
(define_insn "fma4i_fmsub4256"
|
1919 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x,x")
|
1920 |
|
|
(unspec:FMA4MODEF4
|
1921 |
|
|
[(minus:FMA4MODEF4
|
1922 |
|
|
(mult:FMA4MODEF4
|
1923 |
|
|
(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "%x,x")
|
1924 |
|
|
(match_operand:FMA4MODEF4 2 "nonimmediate_operand" "x,m"))
|
1925 |
|
|
(match_operand:FMA4MODEF4 3 "nonimmediate_operand" "xm,x"))]
|
1926 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
1927 |
|
|
"TARGET_FMA4"
|
1928 |
|
|
"vfmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1929 |
|
|
[(set_attr "type" "ssemuladd")
|
1930 |
|
|
(set_attr "mode" "")])
|
1931 |
|
|
|
1932 |
|
|
(define_insn "fma4i_fnmadd4256"
|
1933 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x,x")
|
1934 |
|
|
(unspec:FMA4MODEF4
|
1935 |
|
|
[(minus:FMA4MODEF4
|
1936 |
|
|
(match_operand:FMA4MODEF4 3 "nonimmediate_operand" "xm,x")
|
1937 |
|
|
(mult:FMA4MODEF4
|
1938 |
|
|
(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "%x,x")
|
1939 |
|
|
(match_operand:FMA4MODEF4 2 "nonimmediate_operand" "x,m")))]
|
1940 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
1941 |
|
|
"TARGET_FMA4"
|
1942 |
|
|
"vfnmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1943 |
|
|
[(set_attr "type" "ssemuladd")
|
1944 |
|
|
(set_attr "mode" "")])
|
1945 |
|
|
|
1946 |
|
|
(define_insn "fma4i_fnmsub4256"
|
1947 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x,x")
|
1948 |
|
|
(unspec:FMA4MODEF4
|
1949 |
|
|
[(minus:FMA4MODEF4
|
1950 |
|
|
(mult:FMA4MODEF4
|
1951 |
|
|
(neg:FMA4MODEF4
|
1952 |
|
|
(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "%x,x"))
|
1953 |
|
|
(match_operand:FMA4MODEF4 2 "nonimmediate_operand" "x,m"))
|
1954 |
|
|
(match_operand:FMA4MODEF4 3 "nonimmediate_operand" "xm,x"))]
|
1955 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
1956 |
|
|
"TARGET_FMA4"
|
1957 |
|
|
"vfnmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1958 |
|
|
[(set_attr "type" "ssemuladd")
|
1959 |
|
|
(set_attr "mode" "")])
|
1960 |
|
|
|
1961 |
|
|
(define_insn "fma4i_fmadd4"
|
1962 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
1963 |
|
|
(unspec:SSEMODEF2P
|
1964 |
|
|
[(plus:SSEMODEF2P
|
1965 |
|
|
(mult:SSEMODEF2P
|
1966 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
1967 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
1968 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))]
|
1969 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
1970 |
|
|
"TARGET_FMA4"
|
1971 |
|
|
"vfmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1972 |
|
|
[(set_attr "type" "ssemuladd")
|
1973 |
|
|
(set_attr "mode" "")])
|
1974 |
|
|
|
1975 |
|
|
(define_insn "fma4i_fmsub4"
|
1976 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
1977 |
|
|
(unspec:SSEMODEF2P
|
1978 |
|
|
[(minus:SSEMODEF2P
|
1979 |
|
|
(mult:SSEMODEF2P
|
1980 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
1981 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
1982 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))]
|
1983 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
1984 |
|
|
"TARGET_FMA4"
|
1985 |
|
|
"vfmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
1986 |
|
|
[(set_attr "type" "ssemuladd")
|
1987 |
|
|
(set_attr "mode" "")])
|
1988 |
|
|
|
1989 |
|
|
(define_insn "fma4i_fnmadd4"
|
1990 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
1991 |
|
|
(unspec:SSEMODEF2P
|
1992 |
|
|
[(minus:SSEMODEF2P
|
1993 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x")
|
1994 |
|
|
(mult:SSEMODEF2P
|
1995 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
1996 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m")))]
|
1997 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
1998 |
|
|
"TARGET_FMA4"
|
1999 |
|
|
"vfnmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2000 |
|
|
[(set_attr "type" "ssemuladd")
|
2001 |
|
|
(set_attr "mode" "")])
|
2002 |
|
|
|
2003 |
|
|
(define_insn "fma4i_fnmsub4"
|
2004 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
2005 |
|
|
(unspec:SSEMODEF2P
|
2006 |
|
|
[(minus:SSEMODEF2P
|
2007 |
|
|
(mult:SSEMODEF2P
|
2008 |
|
|
(neg:SSEMODEF2P
|
2009 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x"))
|
2010 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
2011 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))]
|
2012 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2013 |
|
|
"TARGET_FMA4"
|
2014 |
|
|
"vfnmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2015 |
|
|
[(set_attr "type" "ssemuladd")
|
2016 |
|
|
(set_attr "mode" "")])
|
2017 |
|
|
|
2018 |
|
|
;; For the scalar operations, use operand1 for the upper words that aren't
|
2019 |
|
|
;; modified, so restrict the forms that are accepted.
|
2020 |
|
|
(define_insn "fma4i_vmfmadd4"
|
2021 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
2022 |
|
|
(unspec:SSEMODEF2P
|
2023 |
|
|
[(vec_merge:SSEMODEF2P
|
2024 |
|
|
(plus:SSEMODEF2P
|
2025 |
|
|
(mult:SSEMODEF2P
|
2026 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
2027 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
2028 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))
|
2029 |
|
|
(match_dup 0)
|
2030 |
|
|
(const_int 1))]
|
2031 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2032 |
|
|
"TARGET_FMA4"
|
2033 |
|
|
"vfmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2034 |
|
|
[(set_attr "type" "ssemuladd")
|
2035 |
|
|
(set_attr "mode" "")])
|
2036 |
|
|
|
2037 |
|
|
(define_insn "fma4i_vmfmsub4"
|
2038 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
2039 |
|
|
(unspec:SSEMODEF2P
|
2040 |
|
|
[(vec_merge:SSEMODEF2P
|
2041 |
|
|
(minus:SSEMODEF2P
|
2042 |
|
|
(mult:SSEMODEF2P
|
2043 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
2044 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
2045 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))
|
2046 |
|
|
(match_dup 0)
|
2047 |
|
|
(const_int 1))]
|
2048 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2049 |
|
|
"TARGET_FMA4"
|
2050 |
|
|
"vfmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2051 |
|
|
[(set_attr "type" "ssemuladd")
|
2052 |
|
|
(set_attr "mode" "")])
|
2053 |
|
|
|
2054 |
|
|
(define_insn "fma4i_vmfnmadd4"
|
2055 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
2056 |
|
|
(unspec:SSEMODEF2P
|
2057 |
|
|
[(vec_merge:SSEMODEF2P
|
2058 |
|
|
(minus:SSEMODEF2P
|
2059 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x")
|
2060 |
|
|
(mult:SSEMODEF2P
|
2061 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x")
|
2062 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m")))
|
2063 |
|
|
(match_dup 0)
|
2064 |
|
|
(const_int 1))]
|
2065 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2066 |
|
|
"TARGET_FMA4"
|
2067 |
|
|
"vfnmadd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2068 |
|
|
[(set_attr "type" "ssemuladd")
|
2069 |
|
|
(set_attr "mode" "")])
|
2070 |
|
|
|
2071 |
|
|
(define_insn "fma4i_vmfnmsub4"
|
2072 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x,x")
|
2073 |
|
|
(unspec:SSEMODEF2P
|
2074 |
|
|
[(vec_merge:SSEMODEF2P
|
2075 |
|
|
(minus:SSEMODEF2P
|
2076 |
|
|
(mult:SSEMODEF2P
|
2077 |
|
|
(neg:SSEMODEF2P
|
2078 |
|
|
(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%x,x"))
|
2079 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "x,m"))
|
2080 |
|
|
(match_operand:SSEMODEF2P 3 "nonimmediate_operand" "xm,x"))
|
2081 |
|
|
(match_dup 0)
|
2082 |
|
|
(const_int 1))]
|
2083 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2084 |
|
|
"TARGET_FMA4"
|
2085 |
|
|
"vfnmsub\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2086 |
|
|
[(set_attr "type" "ssemuladd")
|
2087 |
|
|
(set_attr "mode" "")])
|
2088 |
|
|
|
2089 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
2090 |
|
|
;;
|
2091 |
|
|
;; FMA4 Parallel floating point multiply addsub and subadd operations.
|
2092 |
|
|
;;
|
2093 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
2094 |
|
|
|
2095 |
|
|
(define_insn "fma4_fmaddsubv8sf4"
|
2096 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x,x")
|
2097 |
|
|
(vec_merge:V8SF
|
2098 |
|
|
(plus:V8SF
|
2099 |
|
|
(mult:V8SF
|
2100 |
|
|
(match_operand:V8SF 1 "nonimmediate_operand" "%x,x")
|
2101 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "x,m"))
|
2102 |
|
|
(match_operand:V8SF 3 "nonimmediate_operand" "xm,x"))
|
2103 |
|
|
(minus:V8SF
|
2104 |
|
|
(mult:V8SF
|
2105 |
|
|
(match_dup 1)
|
2106 |
|
|
(match_dup 2))
|
2107 |
|
|
(match_dup 3))
|
2108 |
|
|
(const_int 170)))]
|
2109 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
2110 |
|
|
"vfmaddsubps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2111 |
|
|
[(set_attr "type" "ssemuladd")
|
2112 |
|
|
(set_attr "mode" "V8SF")])
|
2113 |
|
|
|
2114 |
|
|
(define_insn "fma4_fmaddsubv4df4"
|
2115 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x,x")
|
2116 |
|
|
(vec_merge:V4DF
|
2117 |
|
|
(plus:V4DF
|
2118 |
|
|
(mult:V4DF
|
2119 |
|
|
(match_operand:V4DF 1 "nonimmediate_operand" "%x,x")
|
2120 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "x,m"))
|
2121 |
|
|
(match_operand:V4DF 3 "nonimmediate_operand" "xm,x"))
|
2122 |
|
|
(minus:V4DF
|
2123 |
|
|
(mult:V4DF
|
2124 |
|
|
(match_dup 1)
|
2125 |
|
|
(match_dup 2))
|
2126 |
|
|
(match_dup 3))
|
2127 |
|
|
(const_int 10)))]
|
2128 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
2129 |
|
|
"vfmaddsubpd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2130 |
|
|
[(set_attr "type" "ssemuladd")
|
2131 |
|
|
(set_attr "mode" "V4DF")])
|
2132 |
|
|
|
2133 |
|
|
(define_insn "fma4_fmaddsubv4sf4"
|
2134 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
2135 |
|
|
(vec_merge:V4SF
|
2136 |
|
|
(plus:V4SF
|
2137 |
|
|
(mult:V4SF
|
2138 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "%x,x")
|
2139 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "x,m"))
|
2140 |
|
|
(match_operand:V4SF 3 "nonimmediate_operand" "xm,x"))
|
2141 |
|
|
(minus:V4SF
|
2142 |
|
|
(mult:V4SF
|
2143 |
|
|
(match_dup 1)
|
2144 |
|
|
(match_dup 2))
|
2145 |
|
|
(match_dup 3))
|
2146 |
|
|
(const_int 10)))]
|
2147 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
2148 |
|
|
"vfmaddsubps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2149 |
|
|
[(set_attr "type" "ssemuladd")
|
2150 |
|
|
(set_attr "mode" "V4SF")])
|
2151 |
|
|
|
2152 |
|
|
(define_insn "fma4_fmaddsubv2df4"
|
2153 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x")
|
2154 |
|
|
(vec_merge:V2DF
|
2155 |
|
|
(plus:V2DF
|
2156 |
|
|
(mult:V2DF
|
2157 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "%x,x")
|
2158 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "x,m"))
|
2159 |
|
|
(match_operand:V2DF 3 "nonimmediate_operand" "xm,x"))
|
2160 |
|
|
(minus:V2DF
|
2161 |
|
|
(mult:V2DF
|
2162 |
|
|
(match_dup 1)
|
2163 |
|
|
(match_dup 2))
|
2164 |
|
|
(match_dup 3))
|
2165 |
|
|
(const_int 2)))]
|
2166 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
2167 |
|
|
"vfmaddsubpd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2168 |
|
|
[(set_attr "type" "ssemuladd")
|
2169 |
|
|
(set_attr "mode" "V2DF")])
|
2170 |
|
|
|
2171 |
|
|
(define_insn "fma4_fmsubaddv8sf4"
|
2172 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x,x")
|
2173 |
|
|
(vec_merge:V8SF
|
2174 |
|
|
(plus:V8SF
|
2175 |
|
|
(mult:V8SF
|
2176 |
|
|
(match_operand:V8SF 1 "nonimmediate_operand" "%x,x")
|
2177 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "x,m"))
|
2178 |
|
|
(match_operand:V8SF 3 "nonimmediate_operand" "xm,x"))
|
2179 |
|
|
(minus:V8SF
|
2180 |
|
|
(mult:V8SF
|
2181 |
|
|
(match_dup 1)
|
2182 |
|
|
(match_dup 2))
|
2183 |
|
|
(match_dup 3))
|
2184 |
|
|
(const_int 85)))]
|
2185 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
2186 |
|
|
"vfmsubaddps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2187 |
|
|
[(set_attr "type" "ssemuladd")
|
2188 |
|
|
(set_attr "mode" "V8SF")])
|
2189 |
|
|
|
2190 |
|
|
(define_insn "fma4_fmsubaddv4df4"
|
2191 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x,x")
|
2192 |
|
|
(vec_merge:V4DF
|
2193 |
|
|
(plus:V4DF
|
2194 |
|
|
(mult:V4DF
|
2195 |
|
|
(match_operand:V4DF 1 "nonimmediate_operand" "%x,x")
|
2196 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "x,m"))
|
2197 |
|
|
(match_operand:V4DF 3 "nonimmediate_operand" "xm,x"))
|
2198 |
|
|
(minus:V4DF
|
2199 |
|
|
(mult:V4DF
|
2200 |
|
|
(match_dup 1)
|
2201 |
|
|
(match_dup 2))
|
2202 |
|
|
(match_dup 3))
|
2203 |
|
|
(const_int 5)))]
|
2204 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
2205 |
|
|
"vfmsubaddpd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2206 |
|
|
[(set_attr "type" "ssemuladd")
|
2207 |
|
|
(set_attr "mode" "V4DF")])
|
2208 |
|
|
|
2209 |
|
|
(define_insn "fma4_fmsubaddv4sf4"
|
2210 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
2211 |
|
|
(vec_merge:V4SF
|
2212 |
|
|
(plus:V4SF
|
2213 |
|
|
(mult:V4SF
|
2214 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "%x,x")
|
2215 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "x,m"))
|
2216 |
|
|
(match_operand:V4SF 3 "nonimmediate_operand" "xm,x"))
|
2217 |
|
|
(minus:V4SF
|
2218 |
|
|
(mult:V4SF
|
2219 |
|
|
(match_dup 1)
|
2220 |
|
|
(match_dup 2))
|
2221 |
|
|
(match_dup 3))
|
2222 |
|
|
(const_int 5)))]
|
2223 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
2224 |
|
|
"vfmsubaddps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2225 |
|
|
[(set_attr "type" "ssemuladd")
|
2226 |
|
|
(set_attr "mode" "V4SF")])
|
2227 |
|
|
|
2228 |
|
|
(define_insn "fma4_fmsubaddv2df4"
|
2229 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x")
|
2230 |
|
|
(vec_merge:V2DF
|
2231 |
|
|
(plus:V2DF
|
2232 |
|
|
(mult:V2DF
|
2233 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "%x,x")
|
2234 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "x,m"))
|
2235 |
|
|
(match_operand:V2DF 3 "nonimmediate_operand" "xm,x"))
|
2236 |
|
|
(minus:V2DF
|
2237 |
|
|
(mult:V2DF
|
2238 |
|
|
(match_dup 1)
|
2239 |
|
|
(match_dup 2))
|
2240 |
|
|
(match_dup 3))
|
2241 |
|
|
(const_int 1)))]
|
2242 |
|
|
"TARGET_FMA4 && TARGET_FUSED_MADD"
|
2243 |
|
|
"vfmsubaddpd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2244 |
|
|
[(set_attr "type" "ssemuladd")
|
2245 |
|
|
(set_attr "mode" "V2DF")])
|
2246 |
|
|
|
2247 |
|
|
(define_insn "fma4i_fmaddsubv8sf4"
|
2248 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x,x")
|
2249 |
|
|
(unspec:V8SF
|
2250 |
|
|
[(vec_merge:V8SF
|
2251 |
|
|
(plus:V8SF
|
2252 |
|
|
(mult:V8SF
|
2253 |
|
|
(match_operand:V8SF 1 "nonimmediate_operand" "%x,x")
|
2254 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "x,m"))
|
2255 |
|
|
(match_operand:V8SF 3 "nonimmediate_operand" "xm,x"))
|
2256 |
|
|
(minus:V8SF
|
2257 |
|
|
(mult:V8SF
|
2258 |
|
|
(match_dup 1)
|
2259 |
|
|
(match_dup 2))
|
2260 |
|
|
(match_dup 3))
|
2261 |
|
|
(const_int 170))]
|
2262 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2263 |
|
|
"TARGET_FMA4"
|
2264 |
|
|
"vfmaddsubps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2265 |
|
|
[(set_attr "type" "ssemuladd")
|
2266 |
|
|
(set_attr "mode" "V8SF")])
|
2267 |
|
|
|
2268 |
|
|
(define_insn "fma4i_fmaddsubv4df4"
|
2269 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x,x")
|
2270 |
|
|
(unspec:V4DF
|
2271 |
|
|
[(vec_merge:V4DF
|
2272 |
|
|
(plus:V4DF
|
2273 |
|
|
(mult:V4DF
|
2274 |
|
|
(match_operand:V4DF 1 "nonimmediate_operand" "%x,x")
|
2275 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "x,m"))
|
2276 |
|
|
(match_operand:V4DF 3 "nonimmediate_operand" "xm,x"))
|
2277 |
|
|
(minus:V4DF
|
2278 |
|
|
(mult:V4DF
|
2279 |
|
|
(match_dup 1)
|
2280 |
|
|
(match_dup 2))
|
2281 |
|
|
(match_dup 3))
|
2282 |
|
|
(const_int 10))]
|
2283 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2284 |
|
|
"TARGET_FMA4"
|
2285 |
|
|
"vfmaddsubpd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2286 |
|
|
[(set_attr "type" "ssemuladd")
|
2287 |
|
|
(set_attr "mode" "V4DF")])
|
2288 |
|
|
|
2289 |
|
|
(define_insn "fma4i_fmaddsubv4sf4"
|
2290 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
2291 |
|
|
(unspec:V4SF
|
2292 |
|
|
[(vec_merge:V4SF
|
2293 |
|
|
(plus:V4SF
|
2294 |
|
|
(mult:V4SF
|
2295 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "%x,x")
|
2296 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "x,m"))
|
2297 |
|
|
(match_operand:V4SF 3 "nonimmediate_operand" "xm,x"))
|
2298 |
|
|
(minus:V4SF
|
2299 |
|
|
(mult:V4SF
|
2300 |
|
|
(match_dup 1)
|
2301 |
|
|
(match_dup 2))
|
2302 |
|
|
(match_dup 3))
|
2303 |
|
|
(const_int 10))]
|
2304 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2305 |
|
|
"TARGET_FMA4"
|
2306 |
|
|
"vfmaddsubps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2307 |
|
|
[(set_attr "type" "ssemuladd")
|
2308 |
|
|
(set_attr "mode" "V4SF")])
|
2309 |
|
|
|
2310 |
|
|
(define_insn "fma4i_fmaddsubv2df4"
|
2311 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x")
|
2312 |
|
|
(unspec:V2DF
|
2313 |
|
|
[(vec_merge:V2DF
|
2314 |
|
|
(plus:V2DF
|
2315 |
|
|
(mult:V2DF
|
2316 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "%x,x")
|
2317 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "x,m"))
|
2318 |
|
|
(match_operand:V2DF 3 "nonimmediate_operand" "xm,x"))
|
2319 |
|
|
(minus:V2DF
|
2320 |
|
|
(mult:V2DF
|
2321 |
|
|
(match_dup 1)
|
2322 |
|
|
(match_dup 2))
|
2323 |
|
|
(match_dup 3))
|
2324 |
|
|
(const_int 2))]
|
2325 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2326 |
|
|
"TARGET_FMA4"
|
2327 |
|
|
"vfmaddsubpd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2328 |
|
|
[(set_attr "type" "ssemuladd")
|
2329 |
|
|
(set_attr "mode" "V2DF")])
|
2330 |
|
|
|
2331 |
|
|
(define_insn "fma4i_fmsubaddv8sf4"
|
2332 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x,x")
|
2333 |
|
|
(unspec:V8SF
|
2334 |
|
|
[(vec_merge:V8SF
|
2335 |
|
|
(plus:V8SF
|
2336 |
|
|
(mult:V8SF
|
2337 |
|
|
(match_operand:V8SF 1 "nonimmediate_operand" "%x,x")
|
2338 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "x,m"))
|
2339 |
|
|
(match_operand:V8SF 3 "nonimmediate_operand" "xm,x"))
|
2340 |
|
|
(minus:V8SF
|
2341 |
|
|
(mult:V8SF
|
2342 |
|
|
(match_dup 1)
|
2343 |
|
|
(match_dup 2))
|
2344 |
|
|
(match_dup 3))
|
2345 |
|
|
(const_int 85))]
|
2346 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2347 |
|
|
"TARGET_FMA4"
|
2348 |
|
|
"vfmsubaddps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2349 |
|
|
[(set_attr "type" "ssemuladd")
|
2350 |
|
|
(set_attr "mode" "V8SF")])
|
2351 |
|
|
|
2352 |
|
|
(define_insn "fma4i_fmsubaddv4df4"
|
2353 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x,x")
|
2354 |
|
|
(unspec:V4DF
|
2355 |
|
|
[(vec_merge:V4DF
|
2356 |
|
|
(plus:V4DF
|
2357 |
|
|
(mult:V4DF
|
2358 |
|
|
(match_operand:V4DF 1 "nonimmediate_operand" "%x,x")
|
2359 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "x,m"))
|
2360 |
|
|
(match_operand:V4DF 3 "nonimmediate_operand" "xm,x"))
|
2361 |
|
|
(minus:V4DF
|
2362 |
|
|
(mult:V4DF
|
2363 |
|
|
(match_dup 1)
|
2364 |
|
|
(match_dup 2))
|
2365 |
|
|
(match_dup 3))
|
2366 |
|
|
(const_int 5))]
|
2367 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2368 |
|
|
"TARGET_FMA4"
|
2369 |
|
|
"vfmsubaddpd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2370 |
|
|
[(set_attr "type" "ssemuladd")
|
2371 |
|
|
(set_attr "mode" "V4DF")])
|
2372 |
|
|
|
2373 |
|
|
(define_insn "fma4i_fmsubaddv4sf4"
|
2374 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
2375 |
|
|
(unspec:V4SF
|
2376 |
|
|
[(vec_merge:V4SF
|
2377 |
|
|
(plus:V4SF
|
2378 |
|
|
(mult:V4SF
|
2379 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "%x,x")
|
2380 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "x,m"))
|
2381 |
|
|
(match_operand:V4SF 3 "nonimmediate_operand" "xm,x"))
|
2382 |
|
|
(minus:V4SF
|
2383 |
|
|
(mult:V4SF
|
2384 |
|
|
(match_dup 1)
|
2385 |
|
|
(match_dup 2))
|
2386 |
|
|
(match_dup 3))
|
2387 |
|
|
(const_int 5))]
|
2388 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2389 |
|
|
"TARGET_FMA4"
|
2390 |
|
|
"vfmsubaddps\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2391 |
|
|
[(set_attr "type" "ssemuladd")
|
2392 |
|
|
(set_attr "mode" "V4SF")])
|
2393 |
|
|
|
2394 |
|
|
(define_insn "fma4i_fmsubaddv2df4"
|
2395 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x")
|
2396 |
|
|
(unspec:V2DF
|
2397 |
|
|
[(vec_merge:V2DF
|
2398 |
|
|
(plus:V2DF
|
2399 |
|
|
(mult:V2DF
|
2400 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "%x,x")
|
2401 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "x,m"))
|
2402 |
|
|
(match_operand:V2DF 3 "nonimmediate_operand" "xm,x"))
|
2403 |
|
|
(minus:V2DF
|
2404 |
|
|
(mult:V2DF
|
2405 |
|
|
(match_dup 1)
|
2406 |
|
|
(match_dup 2))
|
2407 |
|
|
(match_dup 3))
|
2408 |
|
|
(const_int 1))]
|
2409 |
|
|
UNSPEC_FMA4_INTRINSIC))]
|
2410 |
|
|
"TARGET_FMA4"
|
2411 |
|
|
"vfmsubaddpd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
2412 |
|
|
[(set_attr "type" "ssemuladd")
|
2413 |
|
|
(set_attr "mode" "V2DF")])
|
2414 |
|
|
|
2415 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
2416 |
|
|
;;
|
2417 |
|
|
;; Parallel single-precision floating point conversion operations
|
2418 |
|
|
;;
|
2419 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
2420 |
|
|
|
2421 |
|
|
(define_insn "sse_cvtpi2ps"
|
2422 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
2423 |
|
|
(vec_merge:V4SF
|
2424 |
|
|
(vec_duplicate:V4SF
|
2425 |
|
|
(float:V2SF (match_operand:V2SI 2 "nonimmediate_operand" "ym")))
|
2426 |
|
|
(match_operand:V4SF 1 "register_operand" "0")
|
2427 |
|
|
(const_int 3)))]
|
2428 |
|
|
"TARGET_SSE"
|
2429 |
|
|
"cvtpi2ps\t{%2, %0|%0, %2}"
|
2430 |
|
|
[(set_attr "type" "ssecvt")
|
2431 |
|
|
(set_attr "mode" "V4SF")])
|
2432 |
|
|
|
2433 |
|
|
(define_insn "sse_cvtps2pi"
|
2434 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=y")
|
2435 |
|
|
(vec_select:V2SI
|
2436 |
|
|
(unspec:V4SI [(match_operand:V4SF 1 "nonimmediate_operand" "xm")]
|
2437 |
|
|
UNSPEC_FIX_NOTRUNC)
|
2438 |
|
|
(parallel [(const_int 0) (const_int 1)])))]
|
2439 |
|
|
"TARGET_SSE"
|
2440 |
|
|
"cvtps2pi\t{%1, %0|%0, %1}"
|
2441 |
|
|
[(set_attr "type" "ssecvt")
|
2442 |
|
|
(set_attr "unit" "mmx")
|
2443 |
|
|
(set_attr "mode" "DI")])
|
2444 |
|
|
|
2445 |
|
|
(define_insn "sse_cvttps2pi"
|
2446 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=y")
|
2447 |
|
|
(vec_select:V2SI
|
2448 |
|
|
(fix:V4SI (match_operand:V4SF 1 "nonimmediate_operand" "xm"))
|
2449 |
|
|
(parallel [(const_int 0) (const_int 1)])))]
|
2450 |
|
|
"TARGET_SSE"
|
2451 |
|
|
"cvttps2pi\t{%1, %0|%0, %1}"
|
2452 |
|
|
[(set_attr "type" "ssecvt")
|
2453 |
|
|
(set_attr "unit" "mmx")
|
2454 |
|
|
(set_attr "prefix_rep" "0")
|
2455 |
|
|
(set_attr "mode" "SF")])
|
2456 |
|
|
|
2457 |
|
|
(define_insn "*avx_cvtsi2ss"
|
2458 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
2459 |
|
|
(vec_merge:V4SF
|
2460 |
|
|
(vec_duplicate:V4SF
|
2461 |
|
|
(float:SF (match_operand:SI 2 "nonimmediate_operand" "rm")))
|
2462 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
2463 |
|
|
(const_int 1)))]
|
2464 |
|
|
"TARGET_AVX"
|
2465 |
|
|
"vcvtsi2ss\t{%2, %1, %0|%0, %1, %2}"
|
2466 |
|
|
[(set_attr "type" "sseicvt")
|
2467 |
|
|
(set_attr "prefix" "vex")
|
2468 |
|
|
(set_attr "mode" "SF")])
|
2469 |
|
|
|
2470 |
|
|
(define_insn "sse_cvtsi2ss"
|
2471 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
2472 |
|
|
(vec_merge:V4SF
|
2473 |
|
|
(vec_duplicate:V4SF
|
2474 |
|
|
(float:SF (match_operand:SI 2 "nonimmediate_operand" "r,m")))
|
2475 |
|
|
(match_operand:V4SF 1 "register_operand" "0,0")
|
2476 |
|
|
(const_int 1)))]
|
2477 |
|
|
"TARGET_SSE"
|
2478 |
|
|
"cvtsi2ss\t{%2, %0|%0, %2}"
|
2479 |
|
|
[(set_attr "type" "sseicvt")
|
2480 |
|
|
(set_attr "athlon_decode" "vector,double")
|
2481 |
|
|
(set_attr "amdfam10_decode" "vector,double")
|
2482 |
|
|
(set_attr "mode" "SF")])
|
2483 |
|
|
|
2484 |
|
|
(define_insn "*avx_cvtsi2ssq"
|
2485 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
2486 |
|
|
(vec_merge:V4SF
|
2487 |
|
|
(vec_duplicate:V4SF
|
2488 |
|
|
(float:SF (match_operand:DI 2 "nonimmediate_operand" "rm")))
|
2489 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
2490 |
|
|
(const_int 1)))]
|
2491 |
|
|
"TARGET_AVX && TARGET_64BIT"
|
2492 |
|
|
"vcvtsi2ssq\t{%2, %1, %0|%0, %1, %2}"
|
2493 |
|
|
[(set_attr "type" "sseicvt")
|
2494 |
|
|
(set_attr "length_vex" "4")
|
2495 |
|
|
(set_attr "prefix" "vex")
|
2496 |
|
|
(set_attr "mode" "SF")])
|
2497 |
|
|
|
2498 |
|
|
(define_insn "sse_cvtsi2ssq"
|
2499 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
2500 |
|
|
(vec_merge:V4SF
|
2501 |
|
|
(vec_duplicate:V4SF
|
2502 |
|
|
(float:SF (match_operand:DI 2 "nonimmediate_operand" "r,rm")))
|
2503 |
|
|
(match_operand:V4SF 1 "register_operand" "0,0")
|
2504 |
|
|
(const_int 1)))]
|
2505 |
|
|
"TARGET_SSE && TARGET_64BIT"
|
2506 |
|
|
"cvtsi2ssq\t{%2, %0|%0, %2}"
|
2507 |
|
|
[(set_attr "type" "sseicvt")
|
2508 |
|
|
(set_attr "prefix_rex" "1")
|
2509 |
|
|
(set_attr "athlon_decode" "vector,double")
|
2510 |
|
|
(set_attr "amdfam10_decode" "vector,double")
|
2511 |
|
|
(set_attr "mode" "SF")])
|
2512 |
|
|
|
2513 |
|
|
(define_insn "sse_cvtss2si"
|
2514 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r,r")
|
2515 |
|
|
(unspec:SI
|
2516 |
|
|
[(vec_select:SF
|
2517 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,m")
|
2518 |
|
|
(parallel [(const_int 0)]))]
|
2519 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2520 |
|
|
"TARGET_SSE"
|
2521 |
|
|
"%vcvtss2si\t{%1, %0|%0, %1}"
|
2522 |
|
|
[(set_attr "type" "sseicvt")
|
2523 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2524 |
|
|
(set_attr "prefix_rep" "1")
|
2525 |
|
|
(set_attr "prefix" "maybe_vex")
|
2526 |
|
|
(set_attr "mode" "SI")])
|
2527 |
|
|
|
2528 |
|
|
(define_insn "sse_cvtss2si_2"
|
2529 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r,r")
|
2530 |
|
|
(unspec:SI [(match_operand:SF 1 "nonimmediate_operand" "x,m")]
|
2531 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2532 |
|
|
"TARGET_SSE"
|
2533 |
|
|
"%vcvtss2si\t{%1, %0|%0, %1}"
|
2534 |
|
|
[(set_attr "type" "sseicvt")
|
2535 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2536 |
|
|
(set_attr "amdfam10_decode" "double,double")
|
2537 |
|
|
(set_attr "prefix_rep" "1")
|
2538 |
|
|
(set_attr "prefix" "maybe_vex")
|
2539 |
|
|
(set_attr "mode" "SI")])
|
2540 |
|
|
|
2541 |
|
|
(define_insn "sse_cvtss2siq"
|
2542 |
|
|
[(set (match_operand:DI 0 "register_operand" "=r,r")
|
2543 |
|
|
(unspec:DI
|
2544 |
|
|
[(vec_select:SF
|
2545 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,m")
|
2546 |
|
|
(parallel [(const_int 0)]))]
|
2547 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2548 |
|
|
"TARGET_SSE && TARGET_64BIT"
|
2549 |
|
|
"%vcvtss2si{q}\t{%1, %0|%0, %1}"
|
2550 |
|
|
[(set_attr "type" "sseicvt")
|
2551 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2552 |
|
|
(set_attr "prefix_rep" "1")
|
2553 |
|
|
(set_attr "prefix" "maybe_vex")
|
2554 |
|
|
(set_attr "mode" "DI")])
|
2555 |
|
|
|
2556 |
|
|
(define_insn "sse_cvtss2siq_2"
|
2557 |
|
|
[(set (match_operand:DI 0 "register_operand" "=r,r")
|
2558 |
|
|
(unspec:DI [(match_operand:SF 1 "nonimmediate_operand" "x,m")]
|
2559 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2560 |
|
|
"TARGET_SSE && TARGET_64BIT"
|
2561 |
|
|
"%vcvtss2si{q}\t{%1, %0|%0, %1}"
|
2562 |
|
|
[(set_attr "type" "sseicvt")
|
2563 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2564 |
|
|
(set_attr "amdfam10_decode" "double,double")
|
2565 |
|
|
(set_attr "prefix_rep" "1")
|
2566 |
|
|
(set_attr "prefix" "maybe_vex")
|
2567 |
|
|
(set_attr "mode" "DI")])
|
2568 |
|
|
|
2569 |
|
|
(define_insn "sse_cvttss2si"
|
2570 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r,r")
|
2571 |
|
|
(fix:SI
|
2572 |
|
|
(vec_select:SF
|
2573 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,m")
|
2574 |
|
|
(parallel [(const_int 0)]))))]
|
2575 |
|
|
"TARGET_SSE"
|
2576 |
|
|
"%vcvttss2si\t{%1, %0|%0, %1}"
|
2577 |
|
|
[(set_attr "type" "sseicvt")
|
2578 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2579 |
|
|
(set_attr "amdfam10_decode" "double,double")
|
2580 |
|
|
(set_attr "prefix_rep" "1")
|
2581 |
|
|
(set_attr "prefix" "maybe_vex")
|
2582 |
|
|
(set_attr "mode" "SI")])
|
2583 |
|
|
|
2584 |
|
|
(define_insn "sse_cvttss2siq"
|
2585 |
|
|
[(set (match_operand:DI 0 "register_operand" "=r,r")
|
2586 |
|
|
(fix:DI
|
2587 |
|
|
(vec_select:SF
|
2588 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,m")
|
2589 |
|
|
(parallel [(const_int 0)]))))]
|
2590 |
|
|
"TARGET_SSE && TARGET_64BIT"
|
2591 |
|
|
"%vcvttss2si{q}\t{%1, %0|%0, %1}"
|
2592 |
|
|
[(set_attr "type" "sseicvt")
|
2593 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2594 |
|
|
(set_attr "amdfam10_decode" "double,double")
|
2595 |
|
|
(set_attr "prefix_rep" "1")
|
2596 |
|
|
(set_attr "prefix" "maybe_vex")
|
2597 |
|
|
(set_attr "mode" "DI")])
|
2598 |
|
|
|
2599 |
|
|
(define_insn "avx_cvtdq2ps"
|
2600 |
|
|
[(set (match_operand:AVXMODEDCVTDQ2PS 0 "register_operand" "=x")
|
2601 |
|
|
(float:AVXMODEDCVTDQ2PS
|
2602 |
|
|
(match_operand: 1 "nonimmediate_operand" "xm")))]
|
2603 |
|
|
"TARGET_AVX"
|
2604 |
|
|
"vcvtdq2ps\t{%1, %0|%0, %1}"
|
2605 |
|
|
[(set_attr "type" "ssecvt")
|
2606 |
|
|
(set_attr "prefix" "vex")
|
2607 |
|
|
(set_attr "mode" "")])
|
2608 |
|
|
|
2609 |
|
|
(define_insn "sse2_cvtdq2ps"
|
2610 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
2611 |
|
|
(float:V4SF (match_operand:V4SI 1 "nonimmediate_operand" "xm")))]
|
2612 |
|
|
"TARGET_SSE2"
|
2613 |
|
|
"cvtdq2ps\t{%1, %0|%0, %1}"
|
2614 |
|
|
[(set_attr "type" "ssecvt")
|
2615 |
|
|
(set_attr "mode" "V4SF")])
|
2616 |
|
|
|
2617 |
|
|
(define_expand "sse2_cvtudq2ps"
|
2618 |
|
|
[(set (match_dup 5)
|
2619 |
|
|
(float:V4SF (match_operand:V4SI 1 "nonimmediate_operand" "")))
|
2620 |
|
|
(set (match_dup 6)
|
2621 |
|
|
(lt:V4SF (match_dup 5) (match_dup 3)))
|
2622 |
|
|
(set (match_dup 7)
|
2623 |
|
|
(and:V4SF (match_dup 6) (match_dup 4)))
|
2624 |
|
|
(set (match_operand:V4SF 0 "register_operand" "")
|
2625 |
|
|
(plus:V4SF (match_dup 5) (match_dup 7)))]
|
2626 |
|
|
"TARGET_SSE2"
|
2627 |
|
|
{
|
2628 |
|
|
REAL_VALUE_TYPE TWO32r;
|
2629 |
|
|
rtx x;
|
2630 |
|
|
int i;
|
2631 |
|
|
|
2632 |
|
|
real_ldexp (&TWO32r, &dconst1, 32);
|
2633 |
|
|
x = const_double_from_real_value (TWO32r, SFmode);
|
2634 |
|
|
|
2635 |
|
|
operands[3] = force_reg (V4SFmode, CONST0_RTX (V4SFmode));
|
2636 |
|
|
operands[4] = force_reg (V4SFmode, ix86_build_const_vector (SFmode, 1, x));
|
2637 |
|
|
|
2638 |
|
|
for (i = 5; i < 8; i++)
|
2639 |
|
|
operands[i] = gen_reg_rtx (V4SFmode);
|
2640 |
|
|
})
|
2641 |
|
|
|
2642 |
|
|
(define_insn "avx_cvtps2dq"
|
2643 |
|
|
[(set (match_operand:AVXMODEDCVTPS2DQ 0 "register_operand" "=x")
|
2644 |
|
|
(unspec:AVXMODEDCVTPS2DQ
|
2645 |
|
|
[(match_operand: 1 "nonimmediate_operand" "xm")]
|
2646 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2647 |
|
|
"TARGET_AVX"
|
2648 |
|
|
"vcvtps2dq\t{%1, %0|%0, %1}"
|
2649 |
|
|
[(set_attr "type" "ssecvt")
|
2650 |
|
|
(set_attr "prefix" "vex")
|
2651 |
|
|
(set_attr "mode" "")])
|
2652 |
|
|
|
2653 |
|
|
(define_insn "sse2_cvtps2dq"
|
2654 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
2655 |
|
|
(unspec:V4SI [(match_operand:V4SF 1 "nonimmediate_operand" "xm")]
|
2656 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2657 |
|
|
"TARGET_SSE2"
|
2658 |
|
|
"cvtps2dq\t{%1, %0|%0, %1}"
|
2659 |
|
|
[(set_attr "type" "ssecvt")
|
2660 |
|
|
(set_attr "prefix_data16" "1")
|
2661 |
|
|
(set_attr "mode" "TI")])
|
2662 |
|
|
|
2663 |
|
|
(define_insn "avx_cvttps2dq"
|
2664 |
|
|
[(set (match_operand:AVXMODEDCVTPS2DQ 0 "register_operand" "=x")
|
2665 |
|
|
(fix:AVXMODEDCVTPS2DQ
|
2666 |
|
|
(match_operand: 1 "nonimmediate_operand" "xm")))]
|
2667 |
|
|
"TARGET_AVX"
|
2668 |
|
|
"vcvttps2dq\t{%1, %0|%0, %1}"
|
2669 |
|
|
[(set_attr "type" "ssecvt")
|
2670 |
|
|
(set_attr "prefix" "vex")
|
2671 |
|
|
(set_attr "mode" "")])
|
2672 |
|
|
|
2673 |
|
|
(define_insn "sse2_cvttps2dq"
|
2674 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
2675 |
|
|
(fix:V4SI (match_operand:V4SF 1 "nonimmediate_operand" "xm")))]
|
2676 |
|
|
"TARGET_SSE2"
|
2677 |
|
|
"cvttps2dq\t{%1, %0|%0, %1}"
|
2678 |
|
|
[(set_attr "type" "ssecvt")
|
2679 |
|
|
(set_attr "prefix_rep" "1")
|
2680 |
|
|
(set_attr "prefix_data16" "0")
|
2681 |
|
|
(set_attr "mode" "TI")])
|
2682 |
|
|
|
2683 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
2684 |
|
|
;;
|
2685 |
|
|
;; Parallel double-precision floating point conversion operations
|
2686 |
|
|
;;
|
2687 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
2688 |
|
|
|
2689 |
|
|
(define_insn "sse2_cvtpi2pd"
|
2690 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x")
|
2691 |
|
|
(float:V2DF (match_operand:V2SI 1 "nonimmediate_operand" "y,m")))]
|
2692 |
|
|
"TARGET_SSE2"
|
2693 |
|
|
"cvtpi2pd\t{%1, %0|%0, %1}"
|
2694 |
|
|
[(set_attr "type" "ssecvt")
|
2695 |
|
|
(set_attr "unit" "mmx,*")
|
2696 |
|
|
(set_attr "prefix_data16" "1,*")
|
2697 |
|
|
(set_attr "mode" "V2DF")])
|
2698 |
|
|
|
2699 |
|
|
(define_insn "sse2_cvtpd2pi"
|
2700 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=y")
|
2701 |
|
|
(unspec:V2SI [(match_operand:V2DF 1 "nonimmediate_operand" "xm")]
|
2702 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2703 |
|
|
"TARGET_SSE2"
|
2704 |
|
|
"cvtpd2pi\t{%1, %0|%0, %1}"
|
2705 |
|
|
[(set_attr "type" "ssecvt")
|
2706 |
|
|
(set_attr "unit" "mmx")
|
2707 |
|
|
(set_attr "prefix_data16" "1")
|
2708 |
|
|
(set_attr "mode" "DI")])
|
2709 |
|
|
|
2710 |
|
|
(define_insn "sse2_cvttpd2pi"
|
2711 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=y")
|
2712 |
|
|
(fix:V2SI (match_operand:V2DF 1 "nonimmediate_operand" "xm")))]
|
2713 |
|
|
"TARGET_SSE2"
|
2714 |
|
|
"cvttpd2pi\t{%1, %0|%0, %1}"
|
2715 |
|
|
[(set_attr "type" "ssecvt")
|
2716 |
|
|
(set_attr "unit" "mmx")
|
2717 |
|
|
(set_attr "prefix_data16" "1")
|
2718 |
|
|
(set_attr "mode" "TI")])
|
2719 |
|
|
|
2720 |
|
|
(define_insn "*avx_cvtsi2sd"
|
2721 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
2722 |
|
|
(vec_merge:V2DF
|
2723 |
|
|
(vec_duplicate:V2DF
|
2724 |
|
|
(float:DF (match_operand:SI 2 "nonimmediate_operand" "rm")))
|
2725 |
|
|
(match_operand:V2DF 1 "register_operand" "x")
|
2726 |
|
|
(const_int 1)))]
|
2727 |
|
|
"TARGET_AVX"
|
2728 |
|
|
"vcvtsi2sd\t{%2, %1, %0|%0, %1, %2}"
|
2729 |
|
|
[(set_attr "type" "sseicvt")
|
2730 |
|
|
(set_attr "prefix" "vex")
|
2731 |
|
|
(set_attr "mode" "DF")])
|
2732 |
|
|
|
2733 |
|
|
(define_insn "sse2_cvtsi2sd"
|
2734 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x")
|
2735 |
|
|
(vec_merge:V2DF
|
2736 |
|
|
(vec_duplicate:V2DF
|
2737 |
|
|
(float:DF (match_operand:SI 2 "nonimmediate_operand" "r,m")))
|
2738 |
|
|
(match_operand:V2DF 1 "register_operand" "0,0")
|
2739 |
|
|
(const_int 1)))]
|
2740 |
|
|
"TARGET_SSE2"
|
2741 |
|
|
"cvtsi2sd\t{%2, %0|%0, %2}"
|
2742 |
|
|
[(set_attr "type" "sseicvt")
|
2743 |
|
|
(set_attr "mode" "DF")
|
2744 |
|
|
(set_attr "athlon_decode" "double,direct")
|
2745 |
|
|
(set_attr "amdfam10_decode" "vector,double")])
|
2746 |
|
|
|
2747 |
|
|
(define_insn "*avx_cvtsi2sdq"
|
2748 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
2749 |
|
|
(vec_merge:V2DF
|
2750 |
|
|
(vec_duplicate:V2DF
|
2751 |
|
|
(float:DF (match_operand:DI 2 "nonimmediate_operand" "rm")))
|
2752 |
|
|
(match_operand:V2DF 1 "register_operand" "x")
|
2753 |
|
|
(const_int 1)))]
|
2754 |
|
|
"TARGET_AVX && TARGET_64BIT"
|
2755 |
|
|
"vcvtsi2sdq\t{%2, %1, %0|%0, %1, %2}"
|
2756 |
|
|
[(set_attr "type" "sseicvt")
|
2757 |
|
|
(set_attr "length_vex" "4")
|
2758 |
|
|
(set_attr "prefix" "vex")
|
2759 |
|
|
(set_attr "mode" "DF")])
|
2760 |
|
|
|
2761 |
|
|
(define_insn "sse2_cvtsi2sdq"
|
2762 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x")
|
2763 |
|
|
(vec_merge:V2DF
|
2764 |
|
|
(vec_duplicate:V2DF
|
2765 |
|
|
(float:DF (match_operand:DI 2 "nonimmediate_operand" "r,m")))
|
2766 |
|
|
(match_operand:V2DF 1 "register_operand" "0,0")
|
2767 |
|
|
(const_int 1)))]
|
2768 |
|
|
"TARGET_SSE2 && TARGET_64BIT"
|
2769 |
|
|
"cvtsi2sdq\t{%2, %0|%0, %2}"
|
2770 |
|
|
[(set_attr "type" "sseicvt")
|
2771 |
|
|
(set_attr "prefix_rex" "1")
|
2772 |
|
|
(set_attr "mode" "DF")
|
2773 |
|
|
(set_attr "athlon_decode" "double,direct")
|
2774 |
|
|
(set_attr "amdfam10_decode" "vector,double")])
|
2775 |
|
|
|
2776 |
|
|
(define_insn "sse2_cvtsd2si"
|
2777 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r,r")
|
2778 |
|
|
(unspec:SI
|
2779 |
|
|
[(vec_select:DF
|
2780 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "x,m")
|
2781 |
|
|
(parallel [(const_int 0)]))]
|
2782 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2783 |
|
|
"TARGET_SSE2"
|
2784 |
|
|
"%vcvtsd2si\t{%1, %0|%0, %1}"
|
2785 |
|
|
[(set_attr "type" "sseicvt")
|
2786 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2787 |
|
|
(set_attr "prefix_rep" "1")
|
2788 |
|
|
(set_attr "prefix" "maybe_vex")
|
2789 |
|
|
(set_attr "mode" "SI")])
|
2790 |
|
|
|
2791 |
|
|
(define_insn "sse2_cvtsd2si_2"
|
2792 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r,r")
|
2793 |
|
|
(unspec:SI [(match_operand:DF 1 "nonimmediate_operand" "x,m")]
|
2794 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2795 |
|
|
"TARGET_SSE2"
|
2796 |
|
|
"%vcvtsd2si\t{%1, %0|%0, %1}"
|
2797 |
|
|
[(set_attr "type" "sseicvt")
|
2798 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2799 |
|
|
(set_attr "amdfam10_decode" "double,double")
|
2800 |
|
|
(set_attr "prefix_rep" "1")
|
2801 |
|
|
(set_attr "prefix" "maybe_vex")
|
2802 |
|
|
(set_attr "mode" "SI")])
|
2803 |
|
|
|
2804 |
|
|
(define_insn "sse2_cvtsd2siq"
|
2805 |
|
|
[(set (match_operand:DI 0 "register_operand" "=r,r")
|
2806 |
|
|
(unspec:DI
|
2807 |
|
|
[(vec_select:DF
|
2808 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "x,m")
|
2809 |
|
|
(parallel [(const_int 0)]))]
|
2810 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2811 |
|
|
"TARGET_SSE2 && TARGET_64BIT"
|
2812 |
|
|
"%vcvtsd2siq\t{%1, %0|%0, %1}"
|
2813 |
|
|
[(set_attr "type" "sseicvt")
|
2814 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2815 |
|
|
(set_attr "prefix_rep" "1")
|
2816 |
|
|
(set_attr "prefix" "maybe_vex")
|
2817 |
|
|
(set_attr "mode" "DI")])
|
2818 |
|
|
|
2819 |
|
|
(define_insn "sse2_cvtsd2siq_2"
|
2820 |
|
|
[(set (match_operand:DI 0 "register_operand" "=r,r")
|
2821 |
|
|
(unspec:DI [(match_operand:DF 1 "nonimmediate_operand" "x,m")]
|
2822 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2823 |
|
|
"TARGET_SSE2 && TARGET_64BIT"
|
2824 |
|
|
"%vcvtsd2siq\t{%1, %0|%0, %1}"
|
2825 |
|
|
[(set_attr "type" "sseicvt")
|
2826 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2827 |
|
|
(set_attr "amdfam10_decode" "double,double")
|
2828 |
|
|
(set_attr "prefix_rep" "1")
|
2829 |
|
|
(set_attr "prefix" "maybe_vex")
|
2830 |
|
|
(set_attr "mode" "DI")])
|
2831 |
|
|
|
2832 |
|
|
(define_insn "sse2_cvttsd2si"
|
2833 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r,r")
|
2834 |
|
|
(fix:SI
|
2835 |
|
|
(vec_select:DF
|
2836 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "x,m")
|
2837 |
|
|
(parallel [(const_int 0)]))))]
|
2838 |
|
|
"TARGET_SSE2"
|
2839 |
|
|
"%vcvttsd2si\t{%1, %0|%0, %1}"
|
2840 |
|
|
[(set_attr "type" "sseicvt")
|
2841 |
|
|
(set_attr "prefix_rep" "1")
|
2842 |
|
|
(set_attr "prefix" "maybe_vex")
|
2843 |
|
|
(set_attr "mode" "SI")
|
2844 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2845 |
|
|
(set_attr "amdfam10_decode" "double,double")])
|
2846 |
|
|
|
2847 |
|
|
(define_insn "sse2_cvttsd2siq"
|
2848 |
|
|
[(set (match_operand:DI 0 "register_operand" "=r,r")
|
2849 |
|
|
(fix:DI
|
2850 |
|
|
(vec_select:DF
|
2851 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "x,m")
|
2852 |
|
|
(parallel [(const_int 0)]))))]
|
2853 |
|
|
"TARGET_SSE2 && TARGET_64BIT"
|
2854 |
|
|
"%vcvttsd2siq\t{%1, %0|%0, %1}"
|
2855 |
|
|
[(set_attr "type" "sseicvt")
|
2856 |
|
|
(set_attr "prefix_rep" "1")
|
2857 |
|
|
(set_attr "prefix" "maybe_vex")
|
2858 |
|
|
(set_attr "mode" "DI")
|
2859 |
|
|
(set_attr "athlon_decode" "double,vector")
|
2860 |
|
|
(set_attr "amdfam10_decode" "double,double")])
|
2861 |
|
|
|
2862 |
|
|
(define_insn "avx_cvtdq2pd256"
|
2863 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x")
|
2864 |
|
|
(float:V4DF (match_operand:V4SI 1 "nonimmediate_operand" "xm")))]
|
2865 |
|
|
"TARGET_AVX"
|
2866 |
|
|
"vcvtdq2pd\t{%1, %0|%0, %1}"
|
2867 |
|
|
[(set_attr "type" "ssecvt")
|
2868 |
|
|
(set_attr "prefix" "vex")
|
2869 |
|
|
(set_attr "mode" "V4DF")])
|
2870 |
|
|
|
2871 |
|
|
(define_insn "sse2_cvtdq2pd"
|
2872 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
2873 |
|
|
(float:V2DF
|
2874 |
|
|
(vec_select:V2SI
|
2875 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "xm")
|
2876 |
|
|
(parallel [(const_int 0) (const_int 1)]))))]
|
2877 |
|
|
"TARGET_SSE2"
|
2878 |
|
|
"%vcvtdq2pd\t{%1, %0|%0, %1}"
|
2879 |
|
|
[(set_attr "type" "ssecvt")
|
2880 |
|
|
(set_attr "prefix" "maybe_vex")
|
2881 |
|
|
(set_attr "mode" "V2DF")])
|
2882 |
|
|
|
2883 |
|
|
(define_insn "avx_cvtpd2dq256"
|
2884 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
2885 |
|
|
(unspec:V4SI [(match_operand:V4DF 1 "nonimmediate_operand" "xm")]
|
2886 |
|
|
UNSPEC_FIX_NOTRUNC))]
|
2887 |
|
|
"TARGET_AVX"
|
2888 |
|
|
"vcvtpd2dq{y}\t{%1, %0|%0, %1}"
|
2889 |
|
|
[(set_attr "type" "ssecvt")
|
2890 |
|
|
(set_attr "prefix" "vex")
|
2891 |
|
|
(set_attr "mode" "OI")])
|
2892 |
|
|
|
2893 |
|
|
(define_expand "sse2_cvtpd2dq"
|
2894 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "")
|
2895 |
|
|
(vec_concat:V4SI
|
2896 |
|
|
(unspec:V2SI [(match_operand:V2DF 1 "nonimmediate_operand" "")]
|
2897 |
|
|
UNSPEC_FIX_NOTRUNC)
|
2898 |
|
|
(match_dup 2)))]
|
2899 |
|
|
"TARGET_SSE2"
|
2900 |
|
|
"operands[2] = CONST0_RTX (V2SImode);")
|
2901 |
|
|
|
2902 |
|
|
(define_insn "*sse2_cvtpd2dq"
|
2903 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
2904 |
|
|
(vec_concat:V4SI
|
2905 |
|
|
(unspec:V2SI [(match_operand:V2DF 1 "nonimmediate_operand" "xm")]
|
2906 |
|
|
UNSPEC_FIX_NOTRUNC)
|
2907 |
|
|
(match_operand:V2SI 2 "const0_operand" "")))]
|
2908 |
|
|
"TARGET_SSE2"
|
2909 |
|
|
"* return TARGET_AVX ? \"vcvtpd2dq{x}\t{%1, %0|%0, %1}\"
|
2910 |
|
|
: \"cvtpd2dq\t{%1, %0|%0, %1}\";"
|
2911 |
|
|
[(set_attr "type" "ssecvt")
|
2912 |
|
|
(set_attr "prefix_rep" "1")
|
2913 |
|
|
(set_attr "prefix_data16" "0")
|
2914 |
|
|
(set_attr "prefix" "maybe_vex")
|
2915 |
|
|
(set_attr "mode" "TI")
|
2916 |
|
|
(set_attr "amdfam10_decode" "double")])
|
2917 |
|
|
|
2918 |
|
|
(define_insn "avx_cvttpd2dq256"
|
2919 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
2920 |
|
|
(fix:V4SI (match_operand:V4DF 1 "nonimmediate_operand" "xm")))]
|
2921 |
|
|
"TARGET_AVX"
|
2922 |
|
|
"vcvttpd2dq{y}\t{%1, %0|%0, %1}"
|
2923 |
|
|
[(set_attr "type" "ssecvt")
|
2924 |
|
|
(set_attr "prefix" "vex")
|
2925 |
|
|
(set_attr "mode" "OI")])
|
2926 |
|
|
|
2927 |
|
|
(define_expand "sse2_cvttpd2dq"
|
2928 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "")
|
2929 |
|
|
(vec_concat:V4SI
|
2930 |
|
|
(fix:V2SI (match_operand:V2DF 1 "nonimmediate_operand" ""))
|
2931 |
|
|
(match_dup 2)))]
|
2932 |
|
|
"TARGET_SSE2"
|
2933 |
|
|
"operands[2] = CONST0_RTX (V2SImode);")
|
2934 |
|
|
|
2935 |
|
|
(define_insn "*sse2_cvttpd2dq"
|
2936 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
2937 |
|
|
(vec_concat:V4SI
|
2938 |
|
|
(fix:V2SI (match_operand:V2DF 1 "nonimmediate_operand" "xm"))
|
2939 |
|
|
(match_operand:V2SI 2 "const0_operand" "")))]
|
2940 |
|
|
"TARGET_SSE2"
|
2941 |
|
|
"* return TARGET_AVX ? \"vcvttpd2dq{x}\t{%1, %0|%0, %1}\"
|
2942 |
|
|
: \"cvttpd2dq\t{%1, %0|%0, %1}\";"
|
2943 |
|
|
[(set_attr "type" "ssecvt")
|
2944 |
|
|
(set_attr "prefix" "maybe_vex")
|
2945 |
|
|
(set_attr "mode" "TI")
|
2946 |
|
|
(set_attr "amdfam10_decode" "double")])
|
2947 |
|
|
|
2948 |
|
|
(define_insn "*avx_cvtsd2ss"
|
2949 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
2950 |
|
|
(vec_merge:V4SF
|
2951 |
|
|
(vec_duplicate:V4SF
|
2952 |
|
|
(float_truncate:V2SF
|
2953 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "xm")))
|
2954 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
2955 |
|
|
(const_int 1)))]
|
2956 |
|
|
"TARGET_AVX"
|
2957 |
|
|
"vcvtsd2ss\t{%2, %1, %0|%0, %1, %2}"
|
2958 |
|
|
[(set_attr "type" "ssecvt")
|
2959 |
|
|
(set_attr "prefix" "vex")
|
2960 |
|
|
(set_attr "mode" "SF")])
|
2961 |
|
|
|
2962 |
|
|
(define_insn "sse2_cvtsd2ss"
|
2963 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
2964 |
|
|
(vec_merge:V4SF
|
2965 |
|
|
(vec_duplicate:V4SF
|
2966 |
|
|
(float_truncate:V2SF
|
2967 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "x,m")))
|
2968 |
|
|
(match_operand:V4SF 1 "register_operand" "0,0")
|
2969 |
|
|
(const_int 1)))]
|
2970 |
|
|
"TARGET_SSE2"
|
2971 |
|
|
"cvtsd2ss\t{%2, %0|%0, %2}"
|
2972 |
|
|
[(set_attr "type" "ssecvt")
|
2973 |
|
|
(set_attr "athlon_decode" "vector,double")
|
2974 |
|
|
(set_attr "amdfam10_decode" "vector,double")
|
2975 |
|
|
(set_attr "mode" "SF")])
|
2976 |
|
|
|
2977 |
|
|
(define_insn "*avx_cvtss2sd"
|
2978 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
2979 |
|
|
(vec_merge:V2DF
|
2980 |
|
|
(float_extend:V2DF
|
2981 |
|
|
(vec_select:V2SF
|
2982 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm")
|
2983 |
|
|
(parallel [(const_int 0) (const_int 1)])))
|
2984 |
|
|
(match_operand:V2DF 1 "register_operand" "x")
|
2985 |
|
|
(const_int 1)))]
|
2986 |
|
|
"TARGET_AVX"
|
2987 |
|
|
"vcvtss2sd\t{%2, %1, %0|%0, %1, %2}"
|
2988 |
|
|
[(set_attr "type" "ssecvt")
|
2989 |
|
|
(set_attr "prefix" "vex")
|
2990 |
|
|
(set_attr "mode" "DF")])
|
2991 |
|
|
|
2992 |
|
|
(define_insn "sse2_cvtss2sd"
|
2993 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x")
|
2994 |
|
|
(vec_merge:V2DF
|
2995 |
|
|
(float_extend:V2DF
|
2996 |
|
|
(vec_select:V2SF
|
2997 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "x,m")
|
2998 |
|
|
(parallel [(const_int 0) (const_int 1)])))
|
2999 |
|
|
(match_operand:V2DF 1 "register_operand" "0,0")
|
3000 |
|
|
(const_int 1)))]
|
3001 |
|
|
"TARGET_SSE2"
|
3002 |
|
|
"cvtss2sd\t{%2, %0|%0, %2}"
|
3003 |
|
|
[(set_attr "type" "ssecvt")
|
3004 |
|
|
(set_attr "amdfam10_decode" "vector,double")
|
3005 |
|
|
(set_attr "mode" "DF")])
|
3006 |
|
|
|
3007 |
|
|
(define_insn "avx_cvtpd2ps256"
|
3008 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3009 |
|
|
(float_truncate:V4SF
|
3010 |
|
|
(match_operand:V4DF 1 "nonimmediate_operand" "xm")))]
|
3011 |
|
|
"TARGET_AVX"
|
3012 |
|
|
"vcvtpd2ps{y}\t{%1, %0|%0, %1}"
|
3013 |
|
|
[(set_attr "type" "ssecvt")
|
3014 |
|
|
(set_attr "prefix" "vex")
|
3015 |
|
|
(set_attr "mode" "V4SF")])
|
3016 |
|
|
|
3017 |
|
|
(define_expand "sse2_cvtpd2ps"
|
3018 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "")
|
3019 |
|
|
(vec_concat:V4SF
|
3020 |
|
|
(float_truncate:V2SF
|
3021 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" ""))
|
3022 |
|
|
(match_dup 2)))]
|
3023 |
|
|
"TARGET_SSE2"
|
3024 |
|
|
"operands[2] = CONST0_RTX (V2SFmode);")
|
3025 |
|
|
|
3026 |
|
|
(define_insn "*sse2_cvtpd2ps"
|
3027 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3028 |
|
|
(vec_concat:V4SF
|
3029 |
|
|
(float_truncate:V2SF
|
3030 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "xm"))
|
3031 |
|
|
(match_operand:V2SF 2 "const0_operand" "")))]
|
3032 |
|
|
"TARGET_SSE2"
|
3033 |
|
|
"* return TARGET_AVX ? \"vcvtpd2ps{x}\t{%1, %0|%0, %1}\"
|
3034 |
|
|
: \"cvtpd2ps\t{%1, %0|%0, %1}\";"
|
3035 |
|
|
[(set_attr "type" "ssecvt")
|
3036 |
|
|
(set_attr "prefix_data16" "1")
|
3037 |
|
|
(set_attr "prefix" "maybe_vex")
|
3038 |
|
|
(set_attr "mode" "V4SF")
|
3039 |
|
|
(set_attr "amdfam10_decode" "double")])
|
3040 |
|
|
|
3041 |
|
|
(define_insn "avx_cvtps2pd256"
|
3042 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x")
|
3043 |
|
|
(float_extend:V4DF
|
3044 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "xm")))]
|
3045 |
|
|
"TARGET_AVX"
|
3046 |
|
|
"vcvtps2pd\t{%1, %0|%0, %1}"
|
3047 |
|
|
[(set_attr "type" "ssecvt")
|
3048 |
|
|
(set_attr "prefix" "vex")
|
3049 |
|
|
(set_attr "mode" "V4DF")])
|
3050 |
|
|
|
3051 |
|
|
(define_insn "sse2_cvtps2pd"
|
3052 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
3053 |
|
|
(float_extend:V2DF
|
3054 |
|
|
(vec_select:V2SF
|
3055 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "xm")
|
3056 |
|
|
(parallel [(const_int 0) (const_int 1)]))))]
|
3057 |
|
|
"TARGET_SSE2"
|
3058 |
|
|
"%vcvtps2pd\t{%1, %0|%0, %1}"
|
3059 |
|
|
[(set_attr "type" "ssecvt")
|
3060 |
|
|
(set_attr "prefix" "maybe_vex")
|
3061 |
|
|
(set_attr "mode" "V2DF")
|
3062 |
|
|
(set_attr "prefix_data16" "0")
|
3063 |
|
|
(set_attr "amdfam10_decode" "direct")])
|
3064 |
|
|
|
3065 |
|
|
(define_expand "vec_unpacks_hi_v4sf"
|
3066 |
|
|
[(set (match_dup 2)
|
3067 |
|
|
(vec_select:V4SF
|
3068 |
|
|
(vec_concat:V8SF
|
3069 |
|
|
(match_dup 2)
|
3070 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" ""))
|
3071 |
|
|
(parallel [(const_int 6)
|
3072 |
|
|
(const_int 7)
|
3073 |
|
|
(const_int 2)
|
3074 |
|
|
(const_int 3)])))
|
3075 |
|
|
(set (match_operand:V2DF 0 "register_operand" "")
|
3076 |
|
|
(float_extend:V2DF
|
3077 |
|
|
(vec_select:V2SF
|
3078 |
|
|
(match_dup 2)
|
3079 |
|
|
(parallel [(const_int 0) (const_int 1)]))))]
|
3080 |
|
|
"TARGET_SSE2"
|
3081 |
|
|
{
|
3082 |
|
|
operands[2] = gen_reg_rtx (V4SFmode);
|
3083 |
|
|
})
|
3084 |
|
|
|
3085 |
|
|
(define_expand "vec_unpacks_lo_v4sf"
|
3086 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "")
|
3087 |
|
|
(float_extend:V2DF
|
3088 |
|
|
(vec_select:V2SF
|
3089 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "")
|
3090 |
|
|
(parallel [(const_int 0) (const_int 1)]))))]
|
3091 |
|
|
"TARGET_SSE2")
|
3092 |
|
|
|
3093 |
|
|
(define_expand "vec_unpacks_float_hi_v8hi"
|
3094 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
3095 |
|
|
(match_operand:V8HI 1 "register_operand" "")]
|
3096 |
|
|
"TARGET_SSE2"
|
3097 |
|
|
{
|
3098 |
|
|
rtx tmp = gen_reg_rtx (V4SImode);
|
3099 |
|
|
|
3100 |
|
|
emit_insn (gen_vec_unpacks_hi_v8hi (tmp, operands[1]));
|
3101 |
|
|
emit_insn (gen_sse2_cvtdq2ps (operands[0], tmp));
|
3102 |
|
|
DONE;
|
3103 |
|
|
})
|
3104 |
|
|
|
3105 |
|
|
(define_expand "vec_unpacks_float_lo_v8hi"
|
3106 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
3107 |
|
|
(match_operand:V8HI 1 "register_operand" "")]
|
3108 |
|
|
"TARGET_SSE2"
|
3109 |
|
|
{
|
3110 |
|
|
rtx tmp = gen_reg_rtx (V4SImode);
|
3111 |
|
|
|
3112 |
|
|
emit_insn (gen_vec_unpacks_lo_v8hi (tmp, operands[1]));
|
3113 |
|
|
emit_insn (gen_sse2_cvtdq2ps (operands[0], tmp));
|
3114 |
|
|
DONE;
|
3115 |
|
|
})
|
3116 |
|
|
|
3117 |
|
|
(define_expand "vec_unpacku_float_hi_v8hi"
|
3118 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
3119 |
|
|
(match_operand:V8HI 1 "register_operand" "")]
|
3120 |
|
|
"TARGET_SSE2"
|
3121 |
|
|
{
|
3122 |
|
|
rtx tmp = gen_reg_rtx (V4SImode);
|
3123 |
|
|
|
3124 |
|
|
emit_insn (gen_vec_unpacku_hi_v8hi (tmp, operands[1]));
|
3125 |
|
|
emit_insn (gen_sse2_cvtdq2ps (operands[0], tmp));
|
3126 |
|
|
DONE;
|
3127 |
|
|
})
|
3128 |
|
|
|
3129 |
|
|
(define_expand "vec_unpacku_float_lo_v8hi"
|
3130 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
3131 |
|
|
(match_operand:V8HI 1 "register_operand" "")]
|
3132 |
|
|
"TARGET_SSE2"
|
3133 |
|
|
{
|
3134 |
|
|
rtx tmp = gen_reg_rtx (V4SImode);
|
3135 |
|
|
|
3136 |
|
|
emit_insn (gen_vec_unpacku_lo_v8hi (tmp, operands[1]));
|
3137 |
|
|
emit_insn (gen_sse2_cvtdq2ps (operands[0], tmp));
|
3138 |
|
|
DONE;
|
3139 |
|
|
})
|
3140 |
|
|
|
3141 |
|
|
(define_expand "vec_unpacks_float_hi_v4si"
|
3142 |
|
|
[(set (match_dup 2)
|
3143 |
|
|
(vec_select:V4SI
|
3144 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "")
|
3145 |
|
|
(parallel [(const_int 2)
|
3146 |
|
|
(const_int 3)
|
3147 |
|
|
(const_int 2)
|
3148 |
|
|
(const_int 3)])))
|
3149 |
|
|
(set (match_operand:V2DF 0 "register_operand" "")
|
3150 |
|
|
(float:V2DF
|
3151 |
|
|
(vec_select:V2SI
|
3152 |
|
|
(match_dup 2)
|
3153 |
|
|
(parallel [(const_int 0) (const_int 1)]))))]
|
3154 |
|
|
"TARGET_SSE2"
|
3155 |
|
|
"operands[2] = gen_reg_rtx (V4SImode);")
|
3156 |
|
|
|
3157 |
|
|
(define_expand "vec_unpacks_float_lo_v4si"
|
3158 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "")
|
3159 |
|
|
(float:V2DF
|
3160 |
|
|
(vec_select:V2SI
|
3161 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "")
|
3162 |
|
|
(parallel [(const_int 0) (const_int 1)]))))]
|
3163 |
|
|
"TARGET_SSE2")
|
3164 |
|
|
|
3165 |
|
|
(define_expand "vec_unpacku_float_hi_v4si"
|
3166 |
|
|
[(set (match_dup 5)
|
3167 |
|
|
(vec_select:V4SI
|
3168 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "")
|
3169 |
|
|
(parallel [(const_int 2)
|
3170 |
|
|
(const_int 3)
|
3171 |
|
|
(const_int 2)
|
3172 |
|
|
(const_int 3)])))
|
3173 |
|
|
(set (match_dup 6)
|
3174 |
|
|
(float:V2DF
|
3175 |
|
|
(vec_select:V2SI
|
3176 |
|
|
(match_dup 5)
|
3177 |
|
|
(parallel [(const_int 0) (const_int 1)]))))
|
3178 |
|
|
(set (match_dup 7)
|
3179 |
|
|
(lt:V2DF (match_dup 6) (match_dup 3)))
|
3180 |
|
|
(set (match_dup 8)
|
3181 |
|
|
(and:V2DF (match_dup 7) (match_dup 4)))
|
3182 |
|
|
(set (match_operand:V2DF 0 "register_operand" "")
|
3183 |
|
|
(plus:V2DF (match_dup 6) (match_dup 8)))]
|
3184 |
|
|
"TARGET_SSE2"
|
3185 |
|
|
{
|
3186 |
|
|
REAL_VALUE_TYPE TWO32r;
|
3187 |
|
|
rtx x;
|
3188 |
|
|
int i;
|
3189 |
|
|
|
3190 |
|
|
real_ldexp (&TWO32r, &dconst1, 32);
|
3191 |
|
|
x = const_double_from_real_value (TWO32r, DFmode);
|
3192 |
|
|
|
3193 |
|
|
operands[3] = force_reg (V2DFmode, CONST0_RTX (V2DFmode));
|
3194 |
|
|
operands[4] = force_reg (V2DFmode, ix86_build_const_vector (DFmode, 1, x));
|
3195 |
|
|
|
3196 |
|
|
operands[5] = gen_reg_rtx (V4SImode);
|
3197 |
|
|
|
3198 |
|
|
for (i = 6; i < 9; i++)
|
3199 |
|
|
operands[i] = gen_reg_rtx (V2DFmode);
|
3200 |
|
|
})
|
3201 |
|
|
|
3202 |
|
|
(define_expand "vec_unpacku_float_lo_v4si"
|
3203 |
|
|
[(set (match_dup 5)
|
3204 |
|
|
(float:V2DF
|
3205 |
|
|
(vec_select:V2SI
|
3206 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "")
|
3207 |
|
|
(parallel [(const_int 0) (const_int 1)]))))
|
3208 |
|
|
(set (match_dup 6)
|
3209 |
|
|
(lt:V2DF (match_dup 5) (match_dup 3)))
|
3210 |
|
|
(set (match_dup 7)
|
3211 |
|
|
(and:V2DF (match_dup 6) (match_dup 4)))
|
3212 |
|
|
(set (match_operand:V2DF 0 "register_operand" "")
|
3213 |
|
|
(plus:V2DF (match_dup 5) (match_dup 7)))]
|
3214 |
|
|
"TARGET_SSE2"
|
3215 |
|
|
{
|
3216 |
|
|
REAL_VALUE_TYPE TWO32r;
|
3217 |
|
|
rtx x;
|
3218 |
|
|
int i;
|
3219 |
|
|
|
3220 |
|
|
real_ldexp (&TWO32r, &dconst1, 32);
|
3221 |
|
|
x = const_double_from_real_value (TWO32r, DFmode);
|
3222 |
|
|
|
3223 |
|
|
operands[3] = force_reg (V2DFmode, CONST0_RTX (V2DFmode));
|
3224 |
|
|
operands[4] = force_reg (V2DFmode, ix86_build_const_vector (DFmode, 1, x));
|
3225 |
|
|
|
3226 |
|
|
for (i = 5; i < 8; i++)
|
3227 |
|
|
operands[i] = gen_reg_rtx (V2DFmode);
|
3228 |
|
|
})
|
3229 |
|
|
|
3230 |
|
|
(define_expand "vec_pack_trunc_v2df"
|
3231 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
3232 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "")
|
3233 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "")]
|
3234 |
|
|
"TARGET_SSE2"
|
3235 |
|
|
{
|
3236 |
|
|
rtx r1, r2;
|
3237 |
|
|
|
3238 |
|
|
r1 = gen_reg_rtx (V4SFmode);
|
3239 |
|
|
r2 = gen_reg_rtx (V4SFmode);
|
3240 |
|
|
|
3241 |
|
|
emit_insn (gen_sse2_cvtpd2ps (r1, operands[1]));
|
3242 |
|
|
emit_insn (gen_sse2_cvtpd2ps (r2, operands[2]));
|
3243 |
|
|
emit_insn (gen_sse_movlhps (operands[0], r1, r2));
|
3244 |
|
|
DONE;
|
3245 |
|
|
})
|
3246 |
|
|
|
3247 |
|
|
(define_expand "vec_pack_sfix_trunc_v2df"
|
3248 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
3249 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "")
|
3250 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "")]
|
3251 |
|
|
"TARGET_SSE2"
|
3252 |
|
|
{
|
3253 |
|
|
rtx r1, r2;
|
3254 |
|
|
|
3255 |
|
|
r1 = gen_reg_rtx (V4SImode);
|
3256 |
|
|
r2 = gen_reg_rtx (V4SImode);
|
3257 |
|
|
|
3258 |
|
|
emit_insn (gen_sse2_cvttpd2dq (r1, operands[1]));
|
3259 |
|
|
emit_insn (gen_sse2_cvttpd2dq (r2, operands[2]));
|
3260 |
|
|
emit_insn (gen_vec_interleave_lowv2di (gen_lowpart (V2DImode, operands[0]),
|
3261 |
|
|
gen_lowpart (V2DImode, r1),
|
3262 |
|
|
gen_lowpart (V2DImode, r2)));
|
3263 |
|
|
DONE;
|
3264 |
|
|
})
|
3265 |
|
|
|
3266 |
|
|
(define_expand "vec_pack_sfix_v2df"
|
3267 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
3268 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "")
|
3269 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "")]
|
3270 |
|
|
"TARGET_SSE2"
|
3271 |
|
|
{
|
3272 |
|
|
rtx r1, r2;
|
3273 |
|
|
|
3274 |
|
|
r1 = gen_reg_rtx (V4SImode);
|
3275 |
|
|
r2 = gen_reg_rtx (V4SImode);
|
3276 |
|
|
|
3277 |
|
|
emit_insn (gen_sse2_cvtpd2dq (r1, operands[1]));
|
3278 |
|
|
emit_insn (gen_sse2_cvtpd2dq (r2, operands[2]));
|
3279 |
|
|
emit_insn (gen_vec_interleave_lowv2di (gen_lowpart (V2DImode, operands[0]),
|
3280 |
|
|
gen_lowpart (V2DImode, r1),
|
3281 |
|
|
gen_lowpart (V2DImode, r2)));
|
3282 |
|
|
DONE;
|
3283 |
|
|
})
|
3284 |
|
|
|
3285 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
3286 |
|
|
;;
|
3287 |
|
|
;; Parallel single-precision floating point element swizzling
|
3288 |
|
|
;;
|
3289 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
3290 |
|
|
|
3291 |
|
|
(define_expand "sse_movhlps_exp"
|
3292 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "")
|
3293 |
|
|
(vec_select:V4SF
|
3294 |
|
|
(vec_concat:V8SF
|
3295 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "")
|
3296 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" ""))
|
3297 |
|
|
(parallel [(const_int 6)
|
3298 |
|
|
(const_int 7)
|
3299 |
|
|
(const_int 2)
|
3300 |
|
|
(const_int 3)])))]
|
3301 |
|
|
"TARGET_SSE"
|
3302 |
|
|
"ix86_fixup_binary_operands (UNKNOWN, V4SFmode, operands);")
|
3303 |
|
|
|
3304 |
|
|
(define_insn "*avx_movhlps"
|
3305 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,x,m")
|
3306 |
|
|
(vec_select:V4SF
|
3307 |
|
|
(vec_concat:V8SF
|
3308 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" " x,x,0")
|
3309 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" " x,o,x"))
|
3310 |
|
|
(parallel [(const_int 6)
|
3311 |
|
|
(const_int 7)
|
3312 |
|
|
(const_int 2)
|
3313 |
|
|
(const_int 3)])))]
|
3314 |
|
|
"TARGET_AVX && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
3315 |
|
|
"@
|
3316 |
|
|
vmovhlps\t{%2, %1, %0|%0, %1, %2}
|
3317 |
|
|
vmovlps\t{%H2, %1, %0|%0, %1, %H2}
|
3318 |
|
|
vmovhps\t{%2, %0|%0, %2}"
|
3319 |
|
|
[(set_attr "type" "ssemov")
|
3320 |
|
|
(set_attr "prefix" "vex")
|
3321 |
|
|
(set_attr "mode" "V4SF,V2SF,V2SF")])
|
3322 |
|
|
|
3323 |
|
|
(define_insn "sse_movhlps"
|
3324 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,x,m")
|
3325 |
|
|
(vec_select:V4SF
|
3326 |
|
|
(vec_concat:V8SF
|
3327 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" " 0,0,0")
|
3328 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" " x,o,x"))
|
3329 |
|
|
(parallel [(const_int 6)
|
3330 |
|
|
(const_int 7)
|
3331 |
|
|
(const_int 2)
|
3332 |
|
|
(const_int 3)])))]
|
3333 |
|
|
"TARGET_SSE && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
3334 |
|
|
"@
|
3335 |
|
|
movhlps\t{%2, %0|%0, %2}
|
3336 |
|
|
movlps\t{%H2, %0|%0, %H2}
|
3337 |
|
|
movhps\t{%2, %0|%0, %2}"
|
3338 |
|
|
[(set_attr "type" "ssemov")
|
3339 |
|
|
(set_attr "mode" "V4SF,V2SF,V2SF")])
|
3340 |
|
|
|
3341 |
|
|
(define_expand "sse_movlhps_exp"
|
3342 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "")
|
3343 |
|
|
(vec_select:V4SF
|
3344 |
|
|
(vec_concat:V8SF
|
3345 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "")
|
3346 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" ""))
|
3347 |
|
|
(parallel [(const_int 0)
|
3348 |
|
|
(const_int 1)
|
3349 |
|
|
(const_int 4)
|
3350 |
|
|
(const_int 5)])))]
|
3351 |
|
|
"TARGET_SSE"
|
3352 |
|
|
"ix86_fixup_binary_operands (UNKNOWN, V4SFmode, operands);")
|
3353 |
|
|
|
3354 |
|
|
(define_insn "*avx_movlhps"
|
3355 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,x,o")
|
3356 |
|
|
(vec_select:V4SF
|
3357 |
|
|
(vec_concat:V8SF
|
3358 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" " x,x,0")
|
3359 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" " x,m,x"))
|
3360 |
|
|
(parallel [(const_int 0)
|
3361 |
|
|
(const_int 1)
|
3362 |
|
|
(const_int 4)
|
3363 |
|
|
(const_int 5)])))]
|
3364 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (UNKNOWN, V4SFmode, operands)"
|
3365 |
|
|
"@
|
3366 |
|
|
vmovlhps\t{%2, %1, %0|%0, %1, %2}
|
3367 |
|
|
vmovhps\t{%2, %1, %0|%0, %1, %2}
|
3368 |
|
|
vmovlps\t{%2, %H0|%H0, %2}"
|
3369 |
|
|
[(set_attr "type" "ssemov")
|
3370 |
|
|
(set_attr "prefix" "vex")
|
3371 |
|
|
(set_attr "mode" "V4SF,V2SF,V2SF")])
|
3372 |
|
|
|
3373 |
|
|
(define_insn "sse_movlhps"
|
3374 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,x,o")
|
3375 |
|
|
(vec_select:V4SF
|
3376 |
|
|
(vec_concat:V8SF
|
3377 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" " 0,0,0")
|
3378 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" " x,m,x"))
|
3379 |
|
|
(parallel [(const_int 0)
|
3380 |
|
|
(const_int 1)
|
3381 |
|
|
(const_int 4)
|
3382 |
|
|
(const_int 5)])))]
|
3383 |
|
|
"TARGET_SSE && ix86_binary_operator_ok (UNKNOWN, V4SFmode, operands)"
|
3384 |
|
|
"@
|
3385 |
|
|
movlhps\t{%2, %0|%0, %2}
|
3386 |
|
|
movhps\t{%2, %0|%0, %2}
|
3387 |
|
|
movlps\t{%2, %H0|%H0, %2}"
|
3388 |
|
|
[(set_attr "type" "ssemov")
|
3389 |
|
|
(set_attr "mode" "V4SF,V2SF,V2SF")])
|
3390 |
|
|
|
3391 |
|
|
;; Recall that the 256-bit unpck insns only shuffle within their lanes.
|
3392 |
|
|
(define_insn "avx_unpckhps256"
|
3393 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
3394 |
|
|
(vec_select:V8SF
|
3395 |
|
|
(vec_concat:V16SF
|
3396 |
|
|
(match_operand:V8SF 1 "register_operand" "x")
|
3397 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "xm"))
|
3398 |
|
|
(parallel [(const_int 2) (const_int 10)
|
3399 |
|
|
(const_int 3) (const_int 11)
|
3400 |
|
|
(const_int 6) (const_int 14)
|
3401 |
|
|
(const_int 7) (const_int 15)])))]
|
3402 |
|
|
"TARGET_AVX"
|
3403 |
|
|
"vunpckhps\t{%2, %1, %0|%0, %1, %2}"
|
3404 |
|
|
[(set_attr "type" "sselog")
|
3405 |
|
|
(set_attr "prefix" "vex")
|
3406 |
|
|
(set_attr "mode" "V8SF")])
|
3407 |
|
|
|
3408 |
|
|
(define_insn "*avx_interleave_highv4sf"
|
3409 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3410 |
|
|
(vec_select:V4SF
|
3411 |
|
|
(vec_concat:V8SF
|
3412 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
3413 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm"))
|
3414 |
|
|
(parallel [(const_int 2) (const_int 6)
|
3415 |
|
|
(const_int 3) (const_int 7)])))]
|
3416 |
|
|
"TARGET_AVX"
|
3417 |
|
|
"vunpckhps\t{%2, %1, %0|%0, %1, %2}"
|
3418 |
|
|
[(set_attr "type" "sselog")
|
3419 |
|
|
(set_attr "prefix" "vex")
|
3420 |
|
|
(set_attr "mode" "V4SF")])
|
3421 |
|
|
|
3422 |
|
|
(define_insn "vec_interleave_highv4sf"
|
3423 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3424 |
|
|
(vec_select:V4SF
|
3425 |
|
|
(vec_concat:V8SF
|
3426 |
|
|
(match_operand:V4SF 1 "register_operand" "0")
|
3427 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm"))
|
3428 |
|
|
(parallel [(const_int 2) (const_int 6)
|
3429 |
|
|
(const_int 3) (const_int 7)])))]
|
3430 |
|
|
"TARGET_SSE"
|
3431 |
|
|
"unpckhps\t{%2, %0|%0, %2}"
|
3432 |
|
|
[(set_attr "type" "sselog")
|
3433 |
|
|
(set_attr "mode" "V4SF")])
|
3434 |
|
|
|
3435 |
|
|
;; Recall that the 256-bit unpck insns only shuffle within their lanes.
|
3436 |
|
|
(define_insn "avx_unpcklps256"
|
3437 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
3438 |
|
|
(vec_select:V8SF
|
3439 |
|
|
(vec_concat:V16SF
|
3440 |
|
|
(match_operand:V8SF 1 "register_operand" "x")
|
3441 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "xm"))
|
3442 |
|
|
(parallel [(const_int 0) (const_int 8)
|
3443 |
|
|
(const_int 1) (const_int 9)
|
3444 |
|
|
(const_int 4) (const_int 12)
|
3445 |
|
|
(const_int 5) (const_int 13)])))]
|
3446 |
|
|
"TARGET_AVX"
|
3447 |
|
|
"vunpcklps\t{%2, %1, %0|%0, %1, %2}"
|
3448 |
|
|
[(set_attr "type" "sselog")
|
3449 |
|
|
(set_attr "prefix" "vex")
|
3450 |
|
|
(set_attr "mode" "V8SF")])
|
3451 |
|
|
|
3452 |
|
|
(define_insn "*avx_interleave_lowv4sf"
|
3453 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3454 |
|
|
(vec_select:V4SF
|
3455 |
|
|
(vec_concat:V8SF
|
3456 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
3457 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm"))
|
3458 |
|
|
(parallel [(const_int 0) (const_int 4)
|
3459 |
|
|
(const_int 1) (const_int 5)])))]
|
3460 |
|
|
"TARGET_AVX"
|
3461 |
|
|
"vunpcklps\t{%2, %1, %0|%0, %1, %2}"
|
3462 |
|
|
[(set_attr "type" "sselog")
|
3463 |
|
|
(set_attr "prefix" "vex")
|
3464 |
|
|
(set_attr "mode" "V4SF")])
|
3465 |
|
|
|
3466 |
|
|
(define_insn "vec_interleave_lowv4sf"
|
3467 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3468 |
|
|
(vec_select:V4SF
|
3469 |
|
|
(vec_concat:V8SF
|
3470 |
|
|
(match_operand:V4SF 1 "register_operand" "0")
|
3471 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "xm"))
|
3472 |
|
|
(parallel [(const_int 0) (const_int 4)
|
3473 |
|
|
(const_int 1) (const_int 5)])))]
|
3474 |
|
|
"TARGET_SSE"
|
3475 |
|
|
"unpcklps\t{%2, %0|%0, %2}"
|
3476 |
|
|
[(set_attr "type" "sselog")
|
3477 |
|
|
(set_attr "mode" "V4SF")])
|
3478 |
|
|
|
3479 |
|
|
;; These are modeled with the same vec_concat as the others so that we
|
3480 |
|
|
;; capture users of shufps that can use the new instructions
|
3481 |
|
|
(define_insn "avx_movshdup256"
|
3482 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
3483 |
|
|
(vec_select:V8SF
|
3484 |
|
|
(vec_concat:V16SF
|
3485 |
|
|
(match_operand:V8SF 1 "nonimmediate_operand" "xm")
|
3486 |
|
|
(match_dup 1))
|
3487 |
|
|
(parallel [(const_int 1) (const_int 1)
|
3488 |
|
|
(const_int 3) (const_int 3)
|
3489 |
|
|
(const_int 5) (const_int 5)
|
3490 |
|
|
(const_int 7) (const_int 7)])))]
|
3491 |
|
|
"TARGET_AVX"
|
3492 |
|
|
"vmovshdup\t{%1, %0|%0, %1}"
|
3493 |
|
|
[(set_attr "type" "sse")
|
3494 |
|
|
(set_attr "prefix" "vex")
|
3495 |
|
|
(set_attr "mode" "V8SF")])
|
3496 |
|
|
|
3497 |
|
|
(define_insn "sse3_movshdup"
|
3498 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3499 |
|
|
(vec_select:V4SF
|
3500 |
|
|
(vec_concat:V8SF
|
3501 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "xm")
|
3502 |
|
|
(match_dup 1))
|
3503 |
|
|
(parallel [(const_int 1)
|
3504 |
|
|
(const_int 1)
|
3505 |
|
|
(const_int 7)
|
3506 |
|
|
(const_int 7)])))]
|
3507 |
|
|
"TARGET_SSE3"
|
3508 |
|
|
"%vmovshdup\t{%1, %0|%0, %1}"
|
3509 |
|
|
[(set_attr "type" "sse")
|
3510 |
|
|
(set_attr "prefix_rep" "1")
|
3511 |
|
|
(set_attr "prefix" "maybe_vex")
|
3512 |
|
|
(set_attr "mode" "V4SF")])
|
3513 |
|
|
|
3514 |
|
|
(define_insn "avx_movsldup256"
|
3515 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
3516 |
|
|
(vec_select:V8SF
|
3517 |
|
|
(vec_concat:V16SF
|
3518 |
|
|
(match_operand:V8SF 1 "nonimmediate_operand" "xm")
|
3519 |
|
|
(match_dup 1))
|
3520 |
|
|
(parallel [(const_int 0) (const_int 0)
|
3521 |
|
|
(const_int 2) (const_int 2)
|
3522 |
|
|
(const_int 4) (const_int 4)
|
3523 |
|
|
(const_int 6) (const_int 6)])))]
|
3524 |
|
|
"TARGET_AVX"
|
3525 |
|
|
"vmovsldup\t{%1, %0|%0, %1}"
|
3526 |
|
|
[(set_attr "type" "sse")
|
3527 |
|
|
(set_attr "prefix" "vex")
|
3528 |
|
|
(set_attr "mode" "V8SF")])
|
3529 |
|
|
|
3530 |
|
|
(define_insn "sse3_movsldup"
|
3531 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3532 |
|
|
(vec_select:V4SF
|
3533 |
|
|
(vec_concat:V8SF
|
3534 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "xm")
|
3535 |
|
|
(match_dup 1))
|
3536 |
|
|
(parallel [(const_int 0)
|
3537 |
|
|
(const_int 0)
|
3538 |
|
|
(const_int 6)
|
3539 |
|
|
(const_int 6)])))]
|
3540 |
|
|
"TARGET_SSE3"
|
3541 |
|
|
"%vmovsldup\t{%1, %0|%0, %1}"
|
3542 |
|
|
[(set_attr "type" "sse")
|
3543 |
|
|
(set_attr "prefix_rep" "1")
|
3544 |
|
|
(set_attr "prefix" "maybe_vex")
|
3545 |
|
|
(set_attr "mode" "V4SF")])
|
3546 |
|
|
|
3547 |
|
|
(define_expand "avx_shufps256"
|
3548 |
|
|
[(match_operand:V8SF 0 "register_operand" "")
|
3549 |
|
|
(match_operand:V8SF 1 "register_operand" "")
|
3550 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "")
|
3551 |
|
|
(match_operand:SI 3 "const_int_operand" "")]
|
3552 |
|
|
"TARGET_AVX"
|
3553 |
|
|
{
|
3554 |
|
|
int mask = INTVAL (operands[3]);
|
3555 |
|
|
emit_insn (gen_avx_shufps256_1 (operands[0], operands[1], operands[2],
|
3556 |
|
|
GEN_INT ((mask >> 0) & 3),
|
3557 |
|
|
GEN_INT ((mask >> 2) & 3),
|
3558 |
|
|
GEN_INT (((mask >> 4) & 3) + 8),
|
3559 |
|
|
GEN_INT (((mask >> 6) & 3) + 8),
|
3560 |
|
|
GEN_INT (((mask >> 0) & 3) + 4),
|
3561 |
|
|
GEN_INT (((mask >> 2) & 3) + 4),
|
3562 |
|
|
GEN_INT (((mask >> 4) & 3) + 12),
|
3563 |
|
|
GEN_INT (((mask >> 6) & 3) + 12)));
|
3564 |
|
|
DONE;
|
3565 |
|
|
})
|
3566 |
|
|
|
3567 |
|
|
;; One bit in mask selects 2 elements.
|
3568 |
|
|
(define_insn "avx_shufps256_1"
|
3569 |
|
|
[(set (match_operand:V8SF 0 "register_operand" "=x")
|
3570 |
|
|
(vec_select:V8SF
|
3571 |
|
|
(vec_concat:V16SF
|
3572 |
|
|
(match_operand:V8SF 1 "register_operand" "x")
|
3573 |
|
|
(match_operand:V8SF 2 "nonimmediate_operand" "xm"))
|
3574 |
|
|
(parallel [(match_operand 3 "const_0_to_3_operand" "")
|
3575 |
|
|
(match_operand 4 "const_0_to_3_operand" "")
|
3576 |
|
|
(match_operand 5 "const_8_to_11_operand" "")
|
3577 |
|
|
(match_operand 6 "const_8_to_11_operand" "")
|
3578 |
|
|
(match_operand 7 "const_4_to_7_operand" "")
|
3579 |
|
|
(match_operand 8 "const_4_to_7_operand" "")
|
3580 |
|
|
(match_operand 9 "const_12_to_15_operand" "")
|
3581 |
|
|
(match_operand 10 "const_12_to_15_operand" "")])))]
|
3582 |
|
|
"TARGET_AVX
|
3583 |
|
|
&& (INTVAL (operands[3]) == (INTVAL (operands[7]) - 4)
|
3584 |
|
|
&& INTVAL (operands[4]) == (INTVAL (operands[8]) - 4)
|
3585 |
|
|
&& INTVAL (operands[5]) == (INTVAL (operands[9]) - 4)
|
3586 |
|
|
&& INTVAL (operands[6]) == (INTVAL (operands[10]) - 4))"
|
3587 |
|
|
{
|
3588 |
|
|
int mask;
|
3589 |
|
|
mask = INTVAL (operands[3]);
|
3590 |
|
|
mask |= INTVAL (operands[4]) << 2;
|
3591 |
|
|
mask |= (INTVAL (operands[5]) - 8) << 4;
|
3592 |
|
|
mask |= (INTVAL (operands[6]) - 8) << 6;
|
3593 |
|
|
operands[3] = GEN_INT (mask);
|
3594 |
|
|
|
3595 |
|
|
return "vshufps\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
3596 |
|
|
}
|
3597 |
|
|
[(set_attr "type" "sselog")
|
3598 |
|
|
(set_attr "length_immediate" "1")
|
3599 |
|
|
(set_attr "prefix" "vex")
|
3600 |
|
|
(set_attr "mode" "V8SF")])
|
3601 |
|
|
|
3602 |
|
|
(define_expand "sse_shufps"
|
3603 |
|
|
[(match_operand:V4SF 0 "register_operand" "")
|
3604 |
|
|
(match_operand:V4SF 1 "register_operand" "")
|
3605 |
|
|
(match_operand:V4SF 2 "nonimmediate_operand" "")
|
3606 |
|
|
(match_operand:SI 3 "const_int_operand" "")]
|
3607 |
|
|
"TARGET_SSE"
|
3608 |
|
|
{
|
3609 |
|
|
int mask = INTVAL (operands[3]);
|
3610 |
|
|
emit_insn (gen_sse_shufps_v4sf (operands[0], operands[1], operands[2],
|
3611 |
|
|
GEN_INT ((mask >> 0) & 3),
|
3612 |
|
|
GEN_INT ((mask >> 2) & 3),
|
3613 |
|
|
GEN_INT (((mask >> 4) & 3) + 4),
|
3614 |
|
|
GEN_INT (((mask >> 6) & 3) + 4)));
|
3615 |
|
|
DONE;
|
3616 |
|
|
})
|
3617 |
|
|
|
3618 |
|
|
(define_insn "*avx_shufps_"
|
3619 |
|
|
[(set (match_operand:SSEMODE4S 0 "register_operand" "=x")
|
3620 |
|
|
(vec_select:SSEMODE4S
|
3621 |
|
|
(vec_concat:
|
3622 |
|
|
(match_operand:SSEMODE4S 1 "register_operand" "x")
|
3623 |
|
|
(match_operand:SSEMODE4S 2 "nonimmediate_operand" "xm"))
|
3624 |
|
|
(parallel [(match_operand 3 "const_0_to_3_operand" "")
|
3625 |
|
|
(match_operand 4 "const_0_to_3_operand" "")
|
3626 |
|
|
(match_operand 5 "const_4_to_7_operand" "")
|
3627 |
|
|
(match_operand 6 "const_4_to_7_operand" "")])))]
|
3628 |
|
|
"TARGET_AVX"
|
3629 |
|
|
{
|
3630 |
|
|
int mask = 0;
|
3631 |
|
|
mask |= INTVAL (operands[3]) << 0;
|
3632 |
|
|
mask |= INTVAL (operands[4]) << 2;
|
3633 |
|
|
mask |= (INTVAL (operands[5]) - 4) << 4;
|
3634 |
|
|
mask |= (INTVAL (operands[6]) - 4) << 6;
|
3635 |
|
|
operands[3] = GEN_INT (mask);
|
3636 |
|
|
|
3637 |
|
|
return "vshufps\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
3638 |
|
|
}
|
3639 |
|
|
[(set_attr "type" "sselog")
|
3640 |
|
|
(set_attr "length_immediate" "1")
|
3641 |
|
|
(set_attr "prefix" "vex")
|
3642 |
|
|
(set_attr "mode" "V4SF")])
|
3643 |
|
|
|
3644 |
|
|
(define_insn "sse_shufps_"
|
3645 |
|
|
[(set (match_operand:SSEMODE4S 0 "register_operand" "=x")
|
3646 |
|
|
(vec_select:SSEMODE4S
|
3647 |
|
|
(vec_concat:
|
3648 |
|
|
(match_operand:SSEMODE4S 1 "register_operand" "0")
|
3649 |
|
|
(match_operand:SSEMODE4S 2 "nonimmediate_operand" "xm"))
|
3650 |
|
|
(parallel [(match_operand 3 "const_0_to_3_operand" "")
|
3651 |
|
|
(match_operand 4 "const_0_to_3_operand" "")
|
3652 |
|
|
(match_operand 5 "const_4_to_7_operand" "")
|
3653 |
|
|
(match_operand 6 "const_4_to_7_operand" "")])))]
|
3654 |
|
|
"TARGET_SSE"
|
3655 |
|
|
{
|
3656 |
|
|
int mask = 0;
|
3657 |
|
|
mask |= INTVAL (operands[3]) << 0;
|
3658 |
|
|
mask |= INTVAL (operands[4]) << 2;
|
3659 |
|
|
mask |= (INTVAL (operands[5]) - 4) << 4;
|
3660 |
|
|
mask |= (INTVAL (operands[6]) - 4) << 6;
|
3661 |
|
|
operands[3] = GEN_INT (mask);
|
3662 |
|
|
|
3663 |
|
|
return "shufps\t{%3, %2, %0|%0, %2, %3}";
|
3664 |
|
|
}
|
3665 |
|
|
[(set_attr "type" "sselog")
|
3666 |
|
|
(set_attr "length_immediate" "1")
|
3667 |
|
|
(set_attr "mode" "V4SF")])
|
3668 |
|
|
|
3669 |
|
|
(define_insn "sse_storehps"
|
3670 |
|
|
[(set (match_operand:V2SF 0 "nonimmediate_operand" "=m,x,x")
|
3671 |
|
|
(vec_select:V2SF
|
3672 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,x,o")
|
3673 |
|
|
(parallel [(const_int 2) (const_int 3)])))]
|
3674 |
|
|
"TARGET_SSE"
|
3675 |
|
|
"@
|
3676 |
|
|
%vmovhps\t{%1, %0|%0, %1}
|
3677 |
|
|
%vmovhlps\t{%1, %d0|%d0, %1}
|
3678 |
|
|
%vmovlps\t{%H1, %d0|%d0, %H1}"
|
3679 |
|
|
[(set_attr "type" "ssemov")
|
3680 |
|
|
(set_attr "prefix" "maybe_vex")
|
3681 |
|
|
(set_attr "mode" "V2SF,V4SF,V2SF")])
|
3682 |
|
|
|
3683 |
|
|
(define_expand "sse_loadhps_exp"
|
3684 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "")
|
3685 |
|
|
(vec_concat:V4SF
|
3686 |
|
|
(vec_select:V2SF
|
3687 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "")
|
3688 |
|
|
(parallel [(const_int 0) (const_int 1)]))
|
3689 |
|
|
(match_operand:V2SF 2 "nonimmediate_operand" "")))]
|
3690 |
|
|
"TARGET_SSE"
|
3691 |
|
|
"ix86_fixup_binary_operands (UNKNOWN, V4SFmode, operands);")
|
3692 |
|
|
|
3693 |
|
|
(define_insn "*avx_loadhps"
|
3694 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,x,o")
|
3695 |
|
|
(vec_concat:V4SF
|
3696 |
|
|
(vec_select:V2SF
|
3697 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,x,0")
|
3698 |
|
|
(parallel [(const_int 0) (const_int 1)]))
|
3699 |
|
|
(match_operand:V2SF 2 "nonimmediate_operand" "m,x,x")))]
|
3700 |
|
|
"TARGET_AVX"
|
3701 |
|
|
"@
|
3702 |
|
|
vmovhps\t{%2, %1, %0|%0, %1, %2}
|
3703 |
|
|
vmovlhps\t{%2, %1, %0|%0, %1, %2}
|
3704 |
|
|
vmovlps\t{%2, %H0|%H0, %2}"
|
3705 |
|
|
[(set_attr "type" "ssemov")
|
3706 |
|
|
(set_attr "prefix" "vex")
|
3707 |
|
|
(set_attr "mode" "V2SF,V4SF,V2SF")])
|
3708 |
|
|
|
3709 |
|
|
(define_insn "sse_loadhps"
|
3710 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,x,o")
|
3711 |
|
|
(vec_concat:V4SF
|
3712 |
|
|
(vec_select:V2SF
|
3713 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "0,0,0")
|
3714 |
|
|
(parallel [(const_int 0) (const_int 1)]))
|
3715 |
|
|
(match_operand:V2SF 2 "nonimmediate_operand" "m,x,x")))]
|
3716 |
|
|
"TARGET_SSE"
|
3717 |
|
|
"@
|
3718 |
|
|
movhps\t{%2, %0|%0, %2}
|
3719 |
|
|
movlhps\t{%2, %0|%0, %2}
|
3720 |
|
|
movlps\t{%2, %H0|%H0, %2}"
|
3721 |
|
|
[(set_attr "type" "ssemov")
|
3722 |
|
|
(set_attr "mode" "V2SF,V4SF,V2SF")])
|
3723 |
|
|
|
3724 |
|
|
(define_insn "*avx_storelps"
|
3725 |
|
|
[(set (match_operand:V2SF 0 "nonimmediate_operand" "=m,x,x")
|
3726 |
|
|
(vec_select:V2SF
|
3727 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,x,m")
|
3728 |
|
|
(parallel [(const_int 0) (const_int 1)])))]
|
3729 |
|
|
"TARGET_AVX"
|
3730 |
|
|
"@
|
3731 |
|
|
vmovlps\t{%1, %0|%0, %1}
|
3732 |
|
|
vmovaps\t{%1, %0|%0, %1}
|
3733 |
|
|
vmovlps\t{%1, %0, %0|%0, %0, %1}"
|
3734 |
|
|
[(set_attr "type" "ssemov")
|
3735 |
|
|
(set_attr "prefix" "vex")
|
3736 |
|
|
(set_attr "mode" "V2SF,V2DF,V2SF")])
|
3737 |
|
|
|
3738 |
|
|
(define_insn "sse_storelps"
|
3739 |
|
|
[(set (match_operand:V2SF 0 "nonimmediate_operand" "=m,x,x")
|
3740 |
|
|
(vec_select:V2SF
|
3741 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,x,m")
|
3742 |
|
|
(parallel [(const_int 0) (const_int 1)])))]
|
3743 |
|
|
"TARGET_SSE"
|
3744 |
|
|
"@
|
3745 |
|
|
movlps\t{%1, %0|%0, %1}
|
3746 |
|
|
movaps\t{%1, %0|%0, %1}
|
3747 |
|
|
movlps\t{%1, %0|%0, %1}"
|
3748 |
|
|
[(set_attr "type" "ssemov")
|
3749 |
|
|
(set_attr "mode" "V2SF,V4SF,V2SF")])
|
3750 |
|
|
|
3751 |
|
|
(define_expand "sse_loadlps_exp"
|
3752 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "")
|
3753 |
|
|
(vec_concat:V4SF
|
3754 |
|
|
(match_operand:V2SF 2 "nonimmediate_operand" "")
|
3755 |
|
|
(vec_select:V2SF
|
3756 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "")
|
3757 |
|
|
(parallel [(const_int 2) (const_int 3)]))))]
|
3758 |
|
|
"TARGET_SSE"
|
3759 |
|
|
"ix86_fixup_binary_operands (UNKNOWN, V4SFmode, operands);")
|
3760 |
|
|
|
3761 |
|
|
(define_insn "*avx_loadlps"
|
3762 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,x,m")
|
3763 |
|
|
(vec_concat:V4SF
|
3764 |
|
|
(match_operand:V2SF 2 "nonimmediate_operand" "x,m,x")
|
3765 |
|
|
(vec_select:V2SF
|
3766 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,x,0")
|
3767 |
|
|
(parallel [(const_int 2) (const_int 3)]))))]
|
3768 |
|
|
"TARGET_AVX"
|
3769 |
|
|
"@
|
3770 |
|
|
shufps\t{$0xe4, %1, %2, %0|%0, %2, %1, 0xe4}
|
3771 |
|
|
vmovlps\t{%2, %1, %0|%0, %1, %2}
|
3772 |
|
|
vmovlps\t{%2, %0|%0, %2}"
|
3773 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov")
|
3774 |
|
|
(set_attr "length_immediate" "1,*,*")
|
3775 |
|
|
(set_attr "prefix" "vex")
|
3776 |
|
|
(set_attr "mode" "V4SF,V2SF,V2SF")])
|
3777 |
|
|
|
3778 |
|
|
(define_insn "sse_loadlps"
|
3779 |
|
|
[(set (match_operand:V4SF 0 "nonimmediate_operand" "=x,x,m")
|
3780 |
|
|
(vec_concat:V4SF
|
3781 |
|
|
(match_operand:V2SF 2 "nonimmediate_operand" "0,m,x")
|
3782 |
|
|
(vec_select:V2SF
|
3783 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "x,0,0")
|
3784 |
|
|
(parallel [(const_int 2) (const_int 3)]))))]
|
3785 |
|
|
"TARGET_SSE"
|
3786 |
|
|
"@
|
3787 |
|
|
shufps\t{$0xe4, %1, %0|%0, %1, 0xe4}
|
3788 |
|
|
movlps\t{%2, %0|%0, %2}
|
3789 |
|
|
movlps\t{%2, %0|%0, %2}"
|
3790 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov")
|
3791 |
|
|
(set_attr "length_immediate" "1,*,*")
|
3792 |
|
|
(set_attr "mode" "V4SF,V2SF,V2SF")])
|
3793 |
|
|
|
3794 |
|
|
(define_insn "*avx_movss"
|
3795 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3796 |
|
|
(vec_merge:V4SF
|
3797 |
|
|
(match_operand:V4SF 2 "register_operand" "x")
|
3798 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
3799 |
|
|
(const_int 1)))]
|
3800 |
|
|
"TARGET_AVX"
|
3801 |
|
|
"vmovss\t{%2, %1, %0|%0, %1, %2}"
|
3802 |
|
|
[(set_attr "type" "ssemov")
|
3803 |
|
|
(set_attr "prefix" "vex")
|
3804 |
|
|
(set_attr "mode" "SF")])
|
3805 |
|
|
|
3806 |
|
|
(define_insn "sse_movss"
|
3807 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3808 |
|
|
(vec_merge:V4SF
|
3809 |
|
|
(match_operand:V4SF 2 "register_operand" "x")
|
3810 |
|
|
(match_operand:V4SF 1 "register_operand" "0")
|
3811 |
|
|
(const_int 1)))]
|
3812 |
|
|
"TARGET_SSE"
|
3813 |
|
|
"movss\t{%2, %0|%0, %2}"
|
3814 |
|
|
[(set_attr "type" "ssemov")
|
3815 |
|
|
(set_attr "mode" "SF")])
|
3816 |
|
|
|
3817 |
|
|
(define_expand "vec_dupv4sf"
|
3818 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "")
|
3819 |
|
|
(vec_duplicate:V4SF
|
3820 |
|
|
(match_operand:SF 1 "nonimmediate_operand" "")))]
|
3821 |
|
|
"TARGET_SSE"
|
3822 |
|
|
{
|
3823 |
|
|
if (!TARGET_AVX)
|
3824 |
|
|
operands[1] = force_reg (V4SFmode, operands[1]);
|
3825 |
|
|
})
|
3826 |
|
|
|
3827 |
|
|
(define_insn "*vec_dupv4sf_avx"
|
3828 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
3829 |
|
|
(vec_duplicate:V4SF
|
3830 |
|
|
(match_operand:SF 1 "nonimmediate_operand" "x,m")))]
|
3831 |
|
|
"TARGET_AVX"
|
3832 |
|
|
"@
|
3833 |
|
|
vshufps\t{$0, %1, %1, %0|%0, %1, %1, 0}
|
3834 |
|
|
vbroadcastss\t{%1, %0|%0, %1}"
|
3835 |
|
|
[(set_attr "type" "sselog1,ssemov")
|
3836 |
|
|
(set_attr "length_immediate" "1,0")
|
3837 |
|
|
(set_attr "prefix_extra" "0,1")
|
3838 |
|
|
(set_attr "prefix" "vex")
|
3839 |
|
|
(set_attr "mode" "V4SF")])
|
3840 |
|
|
|
3841 |
|
|
(define_insn "*vec_dupv4sf"
|
3842 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
3843 |
|
|
(vec_duplicate:V4SF
|
3844 |
|
|
(match_operand:SF 1 "register_operand" "0")))]
|
3845 |
|
|
"TARGET_SSE"
|
3846 |
|
|
"shufps\t{$0, %0, %0|%0, %0, 0}"
|
3847 |
|
|
[(set_attr "type" "sselog1")
|
3848 |
|
|
(set_attr "length_immediate" "1")
|
3849 |
|
|
(set_attr "mode" "V4SF")])
|
3850 |
|
|
|
3851 |
|
|
(define_insn "*vec_concatv2sf_avx"
|
3852 |
|
|
[(set (match_operand:V2SF 0 "register_operand" "=x,x,x,*y ,*y")
|
3853 |
|
|
(vec_concat:V2SF
|
3854 |
|
|
(match_operand:SF 1 "nonimmediate_operand" " x,x,m, x , m")
|
3855 |
|
|
(match_operand:SF 2 "vector_move_operand" " x,m,C,*ym, C")))]
|
3856 |
|
|
"TARGET_AVX"
|
3857 |
|
|
"@
|
3858 |
|
|
vunpcklps\t{%2, %1, %0|%0, %1, %2}
|
3859 |
|
|
vinsertps\t{$0x10, %2, %1, %0|%0, %1, %2, 0x10}
|
3860 |
|
|
vmovss\t{%1, %0|%0, %1}
|
3861 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
3862 |
|
|
movd\t{%1, %0|%0, %1}"
|
3863 |
|
|
[(set_attr "type" "sselog,sselog,ssemov,mmxcvt,mmxmov")
|
3864 |
|
|
(set_attr "length_immediate" "*,1,*,*,*")
|
3865 |
|
|
(set_attr "prefix_extra" "*,1,*,*,*")
|
3866 |
|
|
(set (attr "prefix")
|
3867 |
|
|
(if_then_else (eq_attr "alternative" "3,4")
|
3868 |
|
|
(const_string "orig")
|
3869 |
|
|
(const_string "vex")))
|
3870 |
|
|
(set_attr "mode" "V4SF,V4SF,SF,DI,DI")])
|
3871 |
|
|
|
3872 |
|
|
;; Although insertps takes register source, we prefer
|
3873 |
|
|
;; unpcklps with register source since it is shorter.
|
3874 |
|
|
(define_insn "*vec_concatv2sf_sse4_1"
|
3875 |
|
|
[(set (match_operand:V2SF 0 "register_operand" "=x,x,x,*y ,*y")
|
3876 |
|
|
(vec_concat:V2SF
|
3877 |
|
|
(match_operand:SF 1 "nonimmediate_operand" " 0,0,m, 0 , m")
|
3878 |
|
|
(match_operand:SF 2 "vector_move_operand" " x,m,C,*ym, C")))]
|
3879 |
|
|
"TARGET_SSE4_1"
|
3880 |
|
|
"@
|
3881 |
|
|
unpcklps\t{%2, %0|%0, %2}
|
3882 |
|
|
insertps\t{$0x10, %2, %0|%0, %2, 0x10}
|
3883 |
|
|
movss\t{%1, %0|%0, %1}
|
3884 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
3885 |
|
|
movd\t{%1, %0|%0, %1}"
|
3886 |
|
|
[(set_attr "type" "sselog,sselog,ssemov,mmxcvt,mmxmov")
|
3887 |
|
|
(set_attr "prefix_data16" "*,1,*,*,*")
|
3888 |
|
|
(set_attr "prefix_extra" "*,1,*,*,*")
|
3889 |
|
|
(set_attr "length_immediate" "*,1,*,*,*")
|
3890 |
|
|
(set_attr "mode" "V4SF,V4SF,SF,DI,DI")])
|
3891 |
|
|
|
3892 |
|
|
;; ??? In theory we can match memory for the MMX alternative, but allowing
|
3893 |
|
|
;; nonimmediate_operand for operand 2 and *not* allowing memory for the SSE
|
3894 |
|
|
;; alternatives pretty much forces the MMX alternative to be chosen.
|
3895 |
|
|
(define_insn "*vec_concatv2sf_sse"
|
3896 |
|
|
[(set (match_operand:V2SF 0 "register_operand" "=x,x,*y,*y")
|
3897 |
|
|
(vec_concat:V2SF
|
3898 |
|
|
(match_operand:SF 1 "nonimmediate_operand" " 0,m, 0, m")
|
3899 |
|
|
(match_operand:SF 2 "reg_or_0_operand" " x,C,*y, C")))]
|
3900 |
|
|
"TARGET_SSE"
|
3901 |
|
|
"@
|
3902 |
|
|
unpcklps\t{%2, %0|%0, %2}
|
3903 |
|
|
movss\t{%1, %0|%0, %1}
|
3904 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
3905 |
|
|
movd\t{%1, %0|%0, %1}"
|
3906 |
|
|
[(set_attr "type" "sselog,ssemov,mmxcvt,mmxmov")
|
3907 |
|
|
(set_attr "mode" "V4SF,SF,DI,DI")])
|
3908 |
|
|
|
3909 |
|
|
(define_insn "*vec_concatv4sf_avx"
|
3910 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
3911 |
|
|
(vec_concat:V4SF
|
3912 |
|
|
(match_operand:V2SF 1 "register_operand" " x,x")
|
3913 |
|
|
(match_operand:V2SF 2 "nonimmediate_operand" " x,m")))]
|
3914 |
|
|
"TARGET_AVX"
|
3915 |
|
|
"@
|
3916 |
|
|
vmovlhps\t{%2, %1, %0|%0, %1, %2}
|
3917 |
|
|
vmovhps\t{%2, %1, %0|%0, %1, %2}"
|
3918 |
|
|
[(set_attr "type" "ssemov")
|
3919 |
|
|
(set_attr "prefix" "vex")
|
3920 |
|
|
(set_attr "mode" "V4SF,V2SF")])
|
3921 |
|
|
|
3922 |
|
|
(define_insn "*vec_concatv4sf_sse"
|
3923 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x")
|
3924 |
|
|
(vec_concat:V4SF
|
3925 |
|
|
(match_operand:V2SF 1 "register_operand" " 0,0")
|
3926 |
|
|
(match_operand:V2SF 2 "nonimmediate_operand" " x,m")))]
|
3927 |
|
|
"TARGET_SSE"
|
3928 |
|
|
"@
|
3929 |
|
|
movlhps\t{%2, %0|%0, %2}
|
3930 |
|
|
movhps\t{%2, %0|%0, %2}"
|
3931 |
|
|
[(set_attr "type" "ssemov")
|
3932 |
|
|
(set_attr "mode" "V4SF,V2SF")])
|
3933 |
|
|
|
3934 |
|
|
(define_expand "vec_init"
|
3935 |
|
|
[(match_operand:SSEMODE 0 "register_operand" "")
|
3936 |
|
|
(match_operand 1 "" "")]
|
3937 |
|
|
"TARGET_SSE"
|
3938 |
|
|
{
|
3939 |
|
|
ix86_expand_vector_init (false, operands[0], operands[1]);
|
3940 |
|
|
DONE;
|
3941 |
|
|
})
|
3942 |
|
|
|
3943 |
|
|
(define_insn "*vec_set_0_avx"
|
3944 |
|
|
[(set (match_operand:SSEMODE4S 0 "nonimmediate_operand" "=x,x, x,x, x,m")
|
3945 |
|
|
(vec_merge:SSEMODE4S
|
3946 |
|
|
(vec_duplicate:SSEMODE4S
|
3947 |
|
|
(match_operand: 2
|
3948 |
|
|
"general_operand" " x,m,*r,x,*rm,x*rfF"))
|
3949 |
|
|
(match_operand:SSEMODE4S 1 "vector_move_operand" " C,C, C,x, x,0")
|
3950 |
|
|
(const_int 1)))]
|
3951 |
|
|
"TARGET_AVX"
|
3952 |
|
|
"@
|
3953 |
|
|
vinsertps\t{$0xe, %2, %2, %0|%0, %2, %2, 0xe}
|
3954 |
|
|
vmov\t{%2, %0|%0, %2}
|
3955 |
|
|
vmovd\t{%2, %0|%0, %2}
|
3956 |
|
|
vmovss\t{%2, %1, %0|%0, %1, %2}
|
3957 |
|
|
vpinsrd\t{$0, %2, %1, %0|%0, %1, %2, 0}
|
3958 |
|
|
#"
|
3959 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov,ssemov,sselog,*")
|
3960 |
|
|
(set_attr "prefix_extra" "*,*,*,*,1,*")
|
3961 |
|
|
(set_attr "length_immediate" "*,*,*,*,1,*")
|
3962 |
|
|
(set_attr "prefix" "vex")
|
3963 |
|
|
(set_attr "mode" "SF,,SI,SF,TI,*")])
|
3964 |
|
|
|
3965 |
|
|
(define_insn "*vec_set_0_sse4_1"
|
3966 |
|
|
[(set (match_operand:SSEMODE4S 0 "nonimmediate_operand" "=x,x, x,x, x,m")
|
3967 |
|
|
(vec_merge:SSEMODE4S
|
3968 |
|
|
(vec_duplicate:SSEMODE4S
|
3969 |
|
|
(match_operand: 2
|
3970 |
|
|
"general_operand" " x,m,*r,x,*rm,*rfF"))
|
3971 |
|
|
(match_operand:SSEMODE4S 1 "vector_move_operand" " C,C, C,0, 0,0")
|
3972 |
|
|
(const_int 1)))]
|
3973 |
|
|
"TARGET_SSE4_1"
|
3974 |
|
|
"@
|
3975 |
|
|
insertps\t{$0xe, %2, %0|%0, %2, 0xe}
|
3976 |
|
|
mov\t{%2, %0|%0, %2}
|
3977 |
|
|
movd\t{%2, %0|%0, %2}
|
3978 |
|
|
movss\t{%2, %0|%0, %2}
|
3979 |
|
|
pinsrd\t{$0, %2, %0|%0, %2, 0}
|
3980 |
|
|
#"
|
3981 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov,ssemov,sselog,*")
|
3982 |
|
|
(set_attr "prefix_extra" "*,*,*,*,1,*")
|
3983 |
|
|
(set_attr "length_immediate" "*,*,*,*,1,*")
|
3984 |
|
|
(set_attr "mode" "SF,,SI,SF,TI,*")])
|
3985 |
|
|
|
3986 |
|
|
(define_insn "*vec_set_0_sse2"
|
3987 |
|
|
[(set (match_operand:SSEMODE4S 0 "nonimmediate_operand" "=x, x,x,m")
|
3988 |
|
|
(vec_merge:SSEMODE4S
|
3989 |
|
|
(vec_duplicate:SSEMODE4S
|
3990 |
|
|
(match_operand: 2
|
3991 |
|
|
"general_operand" " m,*r,x,x*rfF"))
|
3992 |
|
|
(match_operand:SSEMODE4S 1 "vector_move_operand" " C, C,0,0")
|
3993 |
|
|
(const_int 1)))]
|
3994 |
|
|
"TARGET_SSE2"
|
3995 |
|
|
"@
|
3996 |
|
|
mov\t{%2, %0|%0, %2}
|
3997 |
|
|
movd\t{%2, %0|%0, %2}
|
3998 |
|
|
movss\t{%2, %0|%0, %2}
|
3999 |
|
|
#"
|
4000 |
|
|
[(set_attr "type" "ssemov")
|
4001 |
|
|
(set_attr "mode" ",SI,SF,*")])
|
4002 |
|
|
|
4003 |
|
|
(define_insn "vec_set_0"
|
4004 |
|
|
[(set (match_operand:SSEMODE4S 0 "nonimmediate_operand" "=x,x,m")
|
4005 |
|
|
(vec_merge:SSEMODE4S
|
4006 |
|
|
(vec_duplicate:SSEMODE4S
|
4007 |
|
|
(match_operand: 2
|
4008 |
|
|
"general_operand" " m,x,x*rfF"))
|
4009 |
|
|
(match_operand:SSEMODE4S 1 "vector_move_operand" " C,0,0")
|
4010 |
|
|
(const_int 1)))]
|
4011 |
|
|
"TARGET_SSE"
|
4012 |
|
|
"@
|
4013 |
|
|
movss\t{%2, %0|%0, %2}
|
4014 |
|
|
movss\t{%2, %0|%0, %2}
|
4015 |
|
|
#"
|
4016 |
|
|
[(set_attr "type" "ssemov")
|
4017 |
|
|
(set_attr "mode" "SF")])
|
4018 |
|
|
|
4019 |
|
|
;; A subset is vec_setv4sf.
|
4020 |
|
|
(define_insn "*vec_setv4sf_avx"
|
4021 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
4022 |
|
|
(vec_merge:V4SF
|
4023 |
|
|
(vec_duplicate:V4SF
|
4024 |
|
|
(match_operand:SF 2 "nonimmediate_operand" "xm"))
|
4025 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
4026 |
|
|
(match_operand:SI 3 "const_pow2_1_to_8_operand" "n")))]
|
4027 |
|
|
"TARGET_AVX"
|
4028 |
|
|
{
|
4029 |
|
|
operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])) << 4);
|
4030 |
|
|
return "vinsertps\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
4031 |
|
|
}
|
4032 |
|
|
[(set_attr "type" "sselog")
|
4033 |
|
|
(set_attr "prefix_extra" "1")
|
4034 |
|
|
(set_attr "length_immediate" "1")
|
4035 |
|
|
(set_attr "prefix" "vex")
|
4036 |
|
|
(set_attr "mode" "V4SF")])
|
4037 |
|
|
|
4038 |
|
|
(define_insn "*vec_setv4sf_sse4_1"
|
4039 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
4040 |
|
|
(vec_merge:V4SF
|
4041 |
|
|
(vec_duplicate:V4SF
|
4042 |
|
|
(match_operand:SF 2 "nonimmediate_operand" "xm"))
|
4043 |
|
|
(match_operand:V4SF 1 "register_operand" "0")
|
4044 |
|
|
(match_operand:SI 3 "const_pow2_1_to_8_operand" "n")))]
|
4045 |
|
|
"TARGET_SSE4_1"
|
4046 |
|
|
{
|
4047 |
|
|
operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])) << 4);
|
4048 |
|
|
return "insertps\t{%3, %2, %0|%0, %2, %3}";
|
4049 |
|
|
}
|
4050 |
|
|
[(set_attr "type" "sselog")
|
4051 |
|
|
(set_attr "prefix_data16" "1")
|
4052 |
|
|
(set_attr "prefix_extra" "1")
|
4053 |
|
|
(set_attr "length_immediate" "1")
|
4054 |
|
|
(set_attr "mode" "V4SF")])
|
4055 |
|
|
|
4056 |
|
|
(define_insn "*avx_insertps"
|
4057 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
4058 |
|
|
(unspec:V4SF [(match_operand:V4SF 2 "nonimmediate_operand" "xm")
|
4059 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
4060 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
4061 |
|
|
UNSPEC_INSERTPS))]
|
4062 |
|
|
"TARGET_AVX"
|
4063 |
|
|
"vinsertps\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
4064 |
|
|
[(set_attr "type" "sselog")
|
4065 |
|
|
(set_attr "prefix" "vex")
|
4066 |
|
|
(set_attr "prefix_extra" "1")
|
4067 |
|
|
(set_attr "length_immediate" "1")
|
4068 |
|
|
(set_attr "mode" "V4SF")])
|
4069 |
|
|
|
4070 |
|
|
(define_insn "sse4_1_insertps"
|
4071 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x")
|
4072 |
|
|
(unspec:V4SF [(match_operand:V4SF 2 "register_operand" "x")
|
4073 |
|
|
(match_operand:V4SF 1 "register_operand" "0")
|
4074 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
4075 |
|
|
UNSPEC_INSERTPS))]
|
4076 |
|
|
"TARGET_SSE4_1"
|
4077 |
|
|
"insertps\t{%3, %2, %0|%0, %2, %3}";
|
4078 |
|
|
[(set_attr "type" "sselog")
|
4079 |
|
|
(set_attr "prefix_data16" "1")
|
4080 |
|
|
(set_attr "prefix_extra" "1")
|
4081 |
|
|
(set_attr "length_immediate" "1")
|
4082 |
|
|
(set_attr "mode" "V4SF")])
|
4083 |
|
|
|
4084 |
|
|
(define_split
|
4085 |
|
|
[(set (match_operand:V4SF 0 "memory_operand" "")
|
4086 |
|
|
(vec_merge:V4SF
|
4087 |
|
|
(vec_duplicate:V4SF
|
4088 |
|
|
(match_operand:SF 1 "nonmemory_operand" ""))
|
4089 |
|
|
(match_dup 0)
|
4090 |
|
|
(const_int 1)))]
|
4091 |
|
|
"TARGET_SSE && reload_completed"
|
4092 |
|
|
[(const_int 0)]
|
4093 |
|
|
{
|
4094 |
|
|
emit_move_insn (adjust_address (operands[0], SFmode, 0), operands[1]);
|
4095 |
|
|
DONE;
|
4096 |
|
|
})
|
4097 |
|
|
|
4098 |
|
|
(define_expand "vec_set"
|
4099 |
|
|
[(match_operand:SSEMODE 0 "register_operand" "")
|
4100 |
|
|
(match_operand: 1 "register_operand" "")
|
4101 |
|
|
(match_operand 2 "const_int_operand" "")]
|
4102 |
|
|
"TARGET_SSE"
|
4103 |
|
|
{
|
4104 |
|
|
ix86_expand_vector_set (false, operands[0], operands[1],
|
4105 |
|
|
INTVAL (operands[2]));
|
4106 |
|
|
DONE;
|
4107 |
|
|
})
|
4108 |
|
|
|
4109 |
|
|
(define_insn_and_split "*vec_extractv4sf_0"
|
4110 |
|
|
[(set (match_operand:SF 0 "nonimmediate_operand" "=x,m,f,r")
|
4111 |
|
|
(vec_select:SF
|
4112 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "xm,x,m,m")
|
4113 |
|
|
(parallel [(const_int 0)])))]
|
4114 |
|
|
"TARGET_SSE && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
4115 |
|
|
"#"
|
4116 |
|
|
"&& reload_completed"
|
4117 |
|
|
[(const_int 0)]
|
4118 |
|
|
{
|
4119 |
|
|
rtx op1 = operands[1];
|
4120 |
|
|
if (REG_P (op1))
|
4121 |
|
|
op1 = gen_rtx_REG (SFmode, REGNO (op1));
|
4122 |
|
|
else
|
4123 |
|
|
op1 = gen_lowpart (SFmode, op1);
|
4124 |
|
|
emit_move_insn (operands[0], op1);
|
4125 |
|
|
DONE;
|
4126 |
|
|
})
|
4127 |
|
|
|
4128 |
|
|
(define_expand "avx_vextractf128"
|
4129 |
|
|
[(match_operand: 0 "nonimmediate_operand" "")
|
4130 |
|
|
(match_operand:AVX256MODE 1 "register_operand" "")
|
4131 |
|
|
(match_operand:SI 2 "const_0_to_1_operand" "")]
|
4132 |
|
|
"TARGET_AVX"
|
4133 |
|
|
{
|
4134 |
|
|
switch (INTVAL (operands[2]))
|
4135 |
|
|
{
|
4136 |
|
|
case 0:
|
4137 |
|
|
emit_insn (gen_vec_extract_lo_ (operands[0], operands[1]));
|
4138 |
|
|
break;
|
4139 |
|
|
case 1:
|
4140 |
|
|
emit_insn (gen_vec_extract_hi_ (operands[0], operands[1]));
|
4141 |
|
|
break;
|
4142 |
|
|
default:
|
4143 |
|
|
gcc_unreachable ();
|
4144 |
|
|
}
|
4145 |
|
|
DONE;
|
4146 |
|
|
})
|
4147 |
|
|
|
4148 |
|
|
(define_insn "vec_extract_lo_"
|
4149 |
|
|
[(set (match_operand: 0 "nonimmediate_operand" "=x,m")
|
4150 |
|
|
(vec_select:
|
4151 |
|
|
(match_operand:AVX256MODE4P 1 "register_operand" "x,x")
|
4152 |
|
|
(parallel [(const_int 0) (const_int 1)])))]
|
4153 |
|
|
"TARGET_AVX"
|
4154 |
|
|
"vextractf128\t{$0x0, %1, %0|%0, %1, 0x0}"
|
4155 |
|
|
[(set_attr "type" "sselog")
|
4156 |
|
|
(set_attr "prefix_extra" "1")
|
4157 |
|
|
(set_attr "length_immediate" "1")
|
4158 |
|
|
(set_attr "memory" "none,store")
|
4159 |
|
|
(set_attr "prefix" "vex")
|
4160 |
|
|
(set_attr "mode" "V8SF")])
|
4161 |
|
|
|
4162 |
|
|
(define_insn "vec_extract_hi_"
|
4163 |
|
|
[(set (match_operand: 0 "nonimmediate_operand" "=x,m")
|
4164 |
|
|
(vec_select:
|
4165 |
|
|
(match_operand:AVX256MODE4P 1 "register_operand" "x,x")
|
4166 |
|
|
(parallel [(const_int 2) (const_int 3)])))]
|
4167 |
|
|
"TARGET_AVX"
|
4168 |
|
|
"vextractf128\t{$0x1, %1, %0|%0, %1, 0x1}"
|
4169 |
|
|
[(set_attr "type" "sselog")
|
4170 |
|
|
(set_attr "prefix_extra" "1")
|
4171 |
|
|
(set_attr "length_immediate" "1")
|
4172 |
|
|
(set_attr "memory" "none,store")
|
4173 |
|
|
(set_attr "prefix" "vex")
|
4174 |
|
|
(set_attr "mode" "V8SF")])
|
4175 |
|
|
|
4176 |
|
|
(define_insn "vec_extract_lo_"
|
4177 |
|
|
[(set (match_operand: 0 "nonimmediate_operand" "=x,m")
|
4178 |
|
|
(vec_select:
|
4179 |
|
|
(match_operand:AVX256MODE8P 1 "register_operand" "x,x")
|
4180 |
|
|
(parallel [(const_int 0) (const_int 1)
|
4181 |
|
|
(const_int 2) (const_int 3)])))]
|
4182 |
|
|
"TARGET_AVX"
|
4183 |
|
|
"vextractf128\t{$0x0, %1, %0|%0, %1, 0x0}"
|
4184 |
|
|
[(set_attr "type" "sselog")
|
4185 |
|
|
(set_attr "prefix_extra" "1")
|
4186 |
|
|
(set_attr "length_immediate" "1")
|
4187 |
|
|
(set_attr "memory" "none,store")
|
4188 |
|
|
(set_attr "prefix" "vex")
|
4189 |
|
|
(set_attr "mode" "V8SF")])
|
4190 |
|
|
|
4191 |
|
|
(define_insn "vec_extract_hi_"
|
4192 |
|
|
[(set (match_operand: 0 "nonimmediate_operand" "=x,m")
|
4193 |
|
|
(vec_select:
|
4194 |
|
|
(match_operand:AVX256MODE8P 1 "register_operand" "x,x")
|
4195 |
|
|
(parallel [(const_int 4) (const_int 5)
|
4196 |
|
|
(const_int 6) (const_int 7)])))]
|
4197 |
|
|
"TARGET_AVX"
|
4198 |
|
|
"vextractf128\t{$0x1, %1, %0|%0, %1, 0x1}"
|
4199 |
|
|
[(set_attr "type" "sselog")
|
4200 |
|
|
(set_attr "prefix_extra" "1")
|
4201 |
|
|
(set_attr "length_immediate" "1")
|
4202 |
|
|
(set_attr "memory" "none,store")
|
4203 |
|
|
(set_attr "prefix" "vex")
|
4204 |
|
|
(set_attr "mode" "V8SF")])
|
4205 |
|
|
|
4206 |
|
|
(define_insn "vec_extract_lo_v16hi"
|
4207 |
|
|
[(set (match_operand:V8HI 0 "nonimmediate_operand" "=x,m")
|
4208 |
|
|
(vec_select:V8HI
|
4209 |
|
|
(match_operand:V16HI 1 "register_operand" "x,x")
|
4210 |
|
|
(parallel [(const_int 0) (const_int 1)
|
4211 |
|
|
(const_int 2) (const_int 3)
|
4212 |
|
|
(const_int 4) (const_int 5)
|
4213 |
|
|
(const_int 6) (const_int 7)])))]
|
4214 |
|
|
"TARGET_AVX"
|
4215 |
|
|
"vextractf128\t{$0x0, %1, %0|%0, %1, 0x0}"
|
4216 |
|
|
[(set_attr "type" "sselog")
|
4217 |
|
|
(set_attr "prefix_extra" "1")
|
4218 |
|
|
(set_attr "length_immediate" "1")
|
4219 |
|
|
(set_attr "memory" "none,store")
|
4220 |
|
|
(set_attr "prefix" "vex")
|
4221 |
|
|
(set_attr "mode" "V8SF")])
|
4222 |
|
|
|
4223 |
|
|
(define_insn "vec_extract_hi_v16hi"
|
4224 |
|
|
[(set (match_operand:V8HI 0 "nonimmediate_operand" "=x,m")
|
4225 |
|
|
(vec_select:V8HI
|
4226 |
|
|
(match_operand:V16HI 1 "register_operand" "x,x")
|
4227 |
|
|
(parallel [(const_int 8) (const_int 9)
|
4228 |
|
|
(const_int 10) (const_int 11)
|
4229 |
|
|
(const_int 12) (const_int 13)
|
4230 |
|
|
(const_int 14) (const_int 15)])))]
|
4231 |
|
|
"TARGET_AVX"
|
4232 |
|
|
"vextractf128\t{$0x1, %1, %0|%0, %1, 0x1}"
|
4233 |
|
|
[(set_attr "type" "sselog")
|
4234 |
|
|
(set_attr "prefix_extra" "1")
|
4235 |
|
|
(set_attr "length_immediate" "1")
|
4236 |
|
|
(set_attr "memory" "none,store")
|
4237 |
|
|
(set_attr "prefix" "vex")
|
4238 |
|
|
(set_attr "mode" "V8SF")])
|
4239 |
|
|
|
4240 |
|
|
(define_insn "vec_extract_lo_v32qi"
|
4241 |
|
|
[(set (match_operand:V16QI 0 "nonimmediate_operand" "=x,m")
|
4242 |
|
|
(vec_select:V16QI
|
4243 |
|
|
(match_operand:V32QI 1 "register_operand" "x,x")
|
4244 |
|
|
(parallel [(const_int 0) (const_int 1)
|
4245 |
|
|
(const_int 2) (const_int 3)
|
4246 |
|
|
(const_int 4) (const_int 5)
|
4247 |
|
|
(const_int 6) (const_int 7)
|
4248 |
|
|
(const_int 8) (const_int 9)
|
4249 |
|
|
(const_int 10) (const_int 11)
|
4250 |
|
|
(const_int 12) (const_int 13)
|
4251 |
|
|
(const_int 14) (const_int 15)])))]
|
4252 |
|
|
"TARGET_AVX"
|
4253 |
|
|
"vextractf128\t{$0x0, %1, %0|%0, %1, 0x0}"
|
4254 |
|
|
[(set_attr "type" "sselog")
|
4255 |
|
|
(set_attr "prefix_extra" "1")
|
4256 |
|
|
(set_attr "length_immediate" "1")
|
4257 |
|
|
(set_attr "memory" "none,store")
|
4258 |
|
|
(set_attr "prefix" "vex")
|
4259 |
|
|
(set_attr "mode" "V8SF")])
|
4260 |
|
|
|
4261 |
|
|
(define_insn "vec_extract_hi_v32qi"
|
4262 |
|
|
[(set (match_operand:V16QI 0 "nonimmediate_operand" "=x,m")
|
4263 |
|
|
(vec_select:V16QI
|
4264 |
|
|
(match_operand:V32QI 1 "register_operand" "x,x")
|
4265 |
|
|
(parallel [(const_int 16) (const_int 17)
|
4266 |
|
|
(const_int 18) (const_int 19)
|
4267 |
|
|
(const_int 20) (const_int 21)
|
4268 |
|
|
(const_int 22) (const_int 23)
|
4269 |
|
|
(const_int 24) (const_int 25)
|
4270 |
|
|
(const_int 26) (const_int 27)
|
4271 |
|
|
(const_int 28) (const_int 29)
|
4272 |
|
|
(const_int 30) (const_int 31)])))]
|
4273 |
|
|
"TARGET_AVX"
|
4274 |
|
|
"vextractf128\t{$0x1, %1, %0|%0, %1, 0x1}"
|
4275 |
|
|
[(set_attr "type" "sselog")
|
4276 |
|
|
(set_attr "prefix_extra" "1")
|
4277 |
|
|
(set_attr "length_immediate" "1")
|
4278 |
|
|
(set_attr "memory" "none,store")
|
4279 |
|
|
(set_attr "prefix" "vex")
|
4280 |
|
|
(set_attr "mode" "V8SF")])
|
4281 |
|
|
|
4282 |
|
|
(define_insn "*sse4_1_extractps"
|
4283 |
|
|
[(set (match_operand:SF 0 "nonimmediate_operand" "=rm")
|
4284 |
|
|
(vec_select:SF
|
4285 |
|
|
(match_operand:V4SF 1 "register_operand" "x")
|
4286 |
|
|
(parallel [(match_operand:SI 2 "const_0_to_3_operand" "n")])))]
|
4287 |
|
|
"TARGET_SSE4_1"
|
4288 |
|
|
"%vextractps\t{%2, %1, %0|%0, %1, %2}"
|
4289 |
|
|
[(set_attr "type" "sselog")
|
4290 |
|
|
(set_attr "prefix_data16" "1")
|
4291 |
|
|
(set_attr "prefix_extra" "1")
|
4292 |
|
|
(set_attr "length_immediate" "1")
|
4293 |
|
|
(set_attr "prefix" "maybe_vex")
|
4294 |
|
|
(set_attr "mode" "V4SF")])
|
4295 |
|
|
|
4296 |
|
|
(define_insn_and_split "*vec_extract_v4sf_mem"
|
4297 |
|
|
[(set (match_operand:SF 0 "register_operand" "=x*rf")
|
4298 |
|
|
(vec_select:SF
|
4299 |
|
|
(match_operand:V4SF 1 "memory_operand" "o")
|
4300 |
|
|
(parallel [(match_operand 2 "const_0_to_3_operand" "n")])))]
|
4301 |
|
|
""
|
4302 |
|
|
"#"
|
4303 |
|
|
"reload_completed"
|
4304 |
|
|
[(const_int 0)]
|
4305 |
|
|
{
|
4306 |
|
|
int i = INTVAL (operands[2]);
|
4307 |
|
|
|
4308 |
|
|
emit_move_insn (operands[0], adjust_address (operands[1], SFmode, i*4));
|
4309 |
|
|
DONE;
|
4310 |
|
|
})
|
4311 |
|
|
|
4312 |
|
|
(define_expand "vec_extract"
|
4313 |
|
|
[(match_operand: 0 "register_operand" "")
|
4314 |
|
|
(match_operand:SSEMODE 1 "register_operand" "")
|
4315 |
|
|
(match_operand 2 "const_int_operand" "")]
|
4316 |
|
|
"TARGET_SSE"
|
4317 |
|
|
{
|
4318 |
|
|
ix86_expand_vector_extract (false, operands[0], operands[1],
|
4319 |
|
|
INTVAL (operands[2]));
|
4320 |
|
|
DONE;
|
4321 |
|
|
})
|
4322 |
|
|
|
4323 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
4324 |
|
|
;;
|
4325 |
|
|
;; Parallel double-precision floating point element swizzling
|
4326 |
|
|
;;
|
4327 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
4328 |
|
|
|
4329 |
|
|
;; Recall that the 256-bit unpck insns only shuffle within their lanes.
|
4330 |
|
|
(define_insn "avx_unpckhpd256"
|
4331 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x")
|
4332 |
|
|
(vec_select:V4DF
|
4333 |
|
|
(vec_concat:V8DF
|
4334 |
|
|
(match_operand:V4DF 1 "register_operand" "x")
|
4335 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "xm"))
|
4336 |
|
|
(parallel [(const_int 1) (const_int 5)
|
4337 |
|
|
(const_int 3) (const_int 7)])))]
|
4338 |
|
|
"TARGET_AVX"
|
4339 |
|
|
"vunpckhpd\t{%2, %1, %0|%0, %1, %2}"
|
4340 |
|
|
[(set_attr "type" "sselog")
|
4341 |
|
|
(set_attr "prefix" "vex")
|
4342 |
|
|
(set_attr "mode" "V4DF")])
|
4343 |
|
|
|
4344 |
|
|
(define_expand "vec_interleave_highv2df"
|
4345 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "")
|
4346 |
|
|
(vec_select:V2DF
|
4347 |
|
|
(vec_concat:V4DF
|
4348 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "")
|
4349 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" ""))
|
4350 |
|
|
(parallel [(const_int 1)
|
4351 |
|
|
(const_int 3)])))]
|
4352 |
|
|
"TARGET_SSE2"
|
4353 |
|
|
{
|
4354 |
|
|
if (!ix86_vec_interleave_v2df_operator_ok (operands, 1))
|
4355 |
|
|
operands[2] = force_reg (V2DFmode, operands[2]);
|
4356 |
|
|
})
|
4357 |
|
|
|
4358 |
|
|
(define_insn "*avx_interleave_highv2df"
|
4359 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,x,m")
|
4360 |
|
|
(vec_select:V2DF
|
4361 |
|
|
(vec_concat:V4DF
|
4362 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " x,o,o,x")
|
4363 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" " x,1,x,0"))
|
4364 |
|
|
(parallel [(const_int 1)
|
4365 |
|
|
(const_int 3)])))]
|
4366 |
|
|
"TARGET_AVX && ix86_vec_interleave_v2df_operator_ok (operands, 1)"
|
4367 |
|
|
"@
|
4368 |
|
|
vunpckhpd\t{%2, %1, %0|%0, %1, %2}
|
4369 |
|
|
vmovddup\t{%H1, %0|%0, %H1}
|
4370 |
|
|
vmovlpd\t{%H1, %2, %0|%0, %2, %H1}
|
4371 |
|
|
vmovhpd\t{%1, %0|%0, %1}"
|
4372 |
|
|
[(set_attr "type" "sselog,sselog,ssemov,ssemov")
|
4373 |
|
|
(set_attr "prefix" "vex")
|
4374 |
|
|
(set_attr "mode" "V2DF,V2DF,V1DF,V1DF")])
|
4375 |
|
|
|
4376 |
|
|
(define_insn "*sse3_interleave_highv2df"
|
4377 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,x,m")
|
4378 |
|
|
(vec_select:V2DF
|
4379 |
|
|
(vec_concat:V4DF
|
4380 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " 0,o,o,x")
|
4381 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" " x,1,0,0"))
|
4382 |
|
|
(parallel [(const_int 1)
|
4383 |
|
|
(const_int 3)])))]
|
4384 |
|
|
"TARGET_SSE3 && ix86_vec_interleave_v2df_operator_ok (operands, 1)"
|
4385 |
|
|
"@
|
4386 |
|
|
unpckhpd\t{%2, %0|%0, %2}
|
4387 |
|
|
movddup\t{%H1, %0|%0, %H1}
|
4388 |
|
|
movlpd\t{%H1, %0|%0, %H1}
|
4389 |
|
|
movhpd\t{%1, %0|%0, %1}"
|
4390 |
|
|
[(set_attr "type" "sselog,sselog,ssemov,ssemov")
|
4391 |
|
|
(set_attr "prefix_data16" "*,*,1,1")
|
4392 |
|
|
(set_attr "mode" "V2DF,V2DF,V1DF,V1DF")])
|
4393 |
|
|
|
4394 |
|
|
(define_insn "*sse2_interleave_highv2df"
|
4395 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,m")
|
4396 |
|
|
(vec_select:V2DF
|
4397 |
|
|
(vec_concat:V4DF
|
4398 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " 0,o,x")
|
4399 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" " x,0,0"))
|
4400 |
|
|
(parallel [(const_int 1)
|
4401 |
|
|
(const_int 3)])))]
|
4402 |
|
|
"TARGET_SSE2 && ix86_vec_interleave_v2df_operator_ok (operands, 1)"
|
4403 |
|
|
"@
|
4404 |
|
|
unpckhpd\t{%2, %0|%0, %2}
|
4405 |
|
|
movlpd\t{%H1, %0|%0, %H1}
|
4406 |
|
|
movhpd\t{%1, %0|%0, %1}"
|
4407 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov")
|
4408 |
|
|
(set_attr "prefix_data16" "*,1,1")
|
4409 |
|
|
(set_attr "mode" "V2DF,V1DF,V1DF")])
|
4410 |
|
|
|
4411 |
|
|
;; Recall that the 256-bit unpck insns only shuffle within their lanes.
|
4412 |
|
|
(define_expand "avx_movddup256"
|
4413 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "")
|
4414 |
|
|
(vec_select:V4DF
|
4415 |
|
|
(vec_concat:V8DF
|
4416 |
|
|
(match_operand:V4DF 1 "nonimmediate_operand" "")
|
4417 |
|
|
(match_dup 1))
|
4418 |
|
|
(parallel [(const_int 0) (const_int 4)
|
4419 |
|
|
(const_int 2) (const_int 6)])))]
|
4420 |
|
|
"TARGET_AVX"
|
4421 |
|
|
"")
|
4422 |
|
|
|
4423 |
|
|
(define_expand "avx_unpcklpd256"
|
4424 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "")
|
4425 |
|
|
(vec_select:V4DF
|
4426 |
|
|
(vec_concat:V8DF
|
4427 |
|
|
(match_operand:V4DF 1 "register_operand" "")
|
4428 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" ""))
|
4429 |
|
|
(parallel [(const_int 0) (const_int 4)
|
4430 |
|
|
(const_int 2) (const_int 6)])))]
|
4431 |
|
|
"TARGET_AVX"
|
4432 |
|
|
"")
|
4433 |
|
|
|
4434 |
|
|
(define_insn "*avx_unpcklpd256"
|
4435 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x,x")
|
4436 |
|
|
(vec_select:V4DF
|
4437 |
|
|
(vec_concat:V8DF
|
4438 |
|
|
(match_operand:V4DF 1 "nonimmediate_operand" "xm,x")
|
4439 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" " 1,xm"))
|
4440 |
|
|
(parallel [(const_int 0) (const_int 4)
|
4441 |
|
|
(const_int 2) (const_int 6)])))]
|
4442 |
|
|
"TARGET_AVX
|
4443 |
|
|
&& (!MEM_P (operands[1]) || rtx_equal_p (operands[1], operands[2]))"
|
4444 |
|
|
"@
|
4445 |
|
|
vmovddup\t{%1, %0|%0, %1}
|
4446 |
|
|
vunpcklpd\t{%2, %1, %0|%0, %1, %2}"
|
4447 |
|
|
[(set_attr "type" "sselog")
|
4448 |
|
|
(set_attr "prefix" "vex")
|
4449 |
|
|
(set_attr "mode" "V4DF")])
|
4450 |
|
|
|
4451 |
|
|
(define_expand "vec_interleave_lowv2df"
|
4452 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "")
|
4453 |
|
|
(vec_select:V2DF
|
4454 |
|
|
(vec_concat:V4DF
|
4455 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "")
|
4456 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" ""))
|
4457 |
|
|
(parallel [(const_int 0)
|
4458 |
|
|
(const_int 2)])))]
|
4459 |
|
|
"TARGET_SSE2"
|
4460 |
|
|
{
|
4461 |
|
|
if (!ix86_vec_interleave_v2df_operator_ok (operands, 0))
|
4462 |
|
|
operands[1] = force_reg (V2DFmode, operands[1]);
|
4463 |
|
|
})
|
4464 |
|
|
|
4465 |
|
|
(define_insn "*avx_interleave_lowv2df"
|
4466 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,x,o")
|
4467 |
|
|
(vec_select:V2DF
|
4468 |
|
|
(vec_concat:V4DF
|
4469 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " x,m,x,0")
|
4470 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" " x,1,m,x"))
|
4471 |
|
|
(parallel [(const_int 0)
|
4472 |
|
|
(const_int 2)])))]
|
4473 |
|
|
"TARGET_AVX && ix86_vec_interleave_v2df_operator_ok (operands, 0)"
|
4474 |
|
|
"@
|
4475 |
|
|
vunpcklpd\t{%2, %1, %0|%0, %1, %2}
|
4476 |
|
|
vmovddup\t{%1, %0|%0, %1}
|
4477 |
|
|
vmovhpd\t{%2, %1, %0|%0, %1, %2}
|
4478 |
|
|
vmovlpd\t{%2, %H0|%H0, %2}"
|
4479 |
|
|
[(set_attr "type" "sselog,sselog,ssemov,ssemov")
|
4480 |
|
|
(set_attr "prefix" "vex")
|
4481 |
|
|
(set_attr "mode" "V2DF,V2DF,V1DF,V1DF")])
|
4482 |
|
|
|
4483 |
|
|
(define_insn "*sse3_interleave_lowv2df"
|
4484 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,x,o")
|
4485 |
|
|
(vec_select:V2DF
|
4486 |
|
|
(vec_concat:V4DF
|
4487 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " 0,m,0,0")
|
4488 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" " x,1,m,x"))
|
4489 |
|
|
(parallel [(const_int 0)
|
4490 |
|
|
(const_int 2)])))]
|
4491 |
|
|
"TARGET_SSE3 && ix86_vec_interleave_v2df_operator_ok (operands, 0)"
|
4492 |
|
|
"@
|
4493 |
|
|
unpcklpd\t{%2, %0|%0, %2}
|
4494 |
|
|
movddup\t{%1, %0|%0, %1}
|
4495 |
|
|
movhpd\t{%2, %0|%0, %2}
|
4496 |
|
|
movlpd\t{%2, %H0|%H0, %2}"
|
4497 |
|
|
[(set_attr "type" "sselog,sselog,ssemov,ssemov")
|
4498 |
|
|
(set_attr "prefix_data16" "*,*,1,1")
|
4499 |
|
|
(set_attr "mode" "V2DF,V2DF,V1DF,V1DF")])
|
4500 |
|
|
|
4501 |
|
|
(define_insn "*sse2_interleave_lowv2df"
|
4502 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,o")
|
4503 |
|
|
(vec_select:V2DF
|
4504 |
|
|
(vec_concat:V4DF
|
4505 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " 0,0,0")
|
4506 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" " x,m,x"))
|
4507 |
|
|
(parallel [(const_int 0)
|
4508 |
|
|
(const_int 2)])))]
|
4509 |
|
|
"TARGET_SSE2 && ix86_vec_interleave_v2df_operator_ok (operands, 0)"
|
4510 |
|
|
"@
|
4511 |
|
|
unpcklpd\t{%2, %0|%0, %2}
|
4512 |
|
|
movhpd\t{%2, %0|%0, %2}
|
4513 |
|
|
movlpd\t{%2, %H0|%H0, %2}"
|
4514 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov")
|
4515 |
|
|
(set_attr "prefix_data16" "*,1,1")
|
4516 |
|
|
(set_attr "mode" "V2DF,V1DF,V1DF")])
|
4517 |
|
|
|
4518 |
|
|
(define_split
|
4519 |
|
|
[(set (match_operand:V2DF 0 "memory_operand" "")
|
4520 |
|
|
(vec_select:V2DF
|
4521 |
|
|
(vec_concat:V4DF
|
4522 |
|
|
(match_operand:V2DF 1 "register_operand" "")
|
4523 |
|
|
(match_dup 1))
|
4524 |
|
|
(parallel [(const_int 0)
|
4525 |
|
|
(const_int 2)])))]
|
4526 |
|
|
"TARGET_SSE3 && reload_completed"
|
4527 |
|
|
[(const_int 0)]
|
4528 |
|
|
{
|
4529 |
|
|
rtx low = gen_rtx_REG (DFmode, REGNO (operands[1]));
|
4530 |
|
|
emit_move_insn (adjust_address (operands[0], DFmode, 0), low);
|
4531 |
|
|
emit_move_insn (adjust_address (operands[0], DFmode, 8), low);
|
4532 |
|
|
DONE;
|
4533 |
|
|
})
|
4534 |
|
|
|
4535 |
|
|
(define_split
|
4536 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "")
|
4537 |
|
|
(vec_select:V2DF
|
4538 |
|
|
(vec_concat:V4DF
|
4539 |
|
|
(match_operand:V2DF 1 "memory_operand" "")
|
4540 |
|
|
(match_dup 1))
|
4541 |
|
|
(parallel [(match_operand:SI 2 "const_0_to_1_operand" "")
|
4542 |
|
|
(match_operand:SI 3 "const_int_operand" "")])))]
|
4543 |
|
|
"TARGET_SSE3 && INTVAL (operands[2]) + 2 == INTVAL (operands[3])"
|
4544 |
|
|
[(set (match_dup 0) (vec_duplicate:V2DF (match_dup 1)))]
|
4545 |
|
|
{
|
4546 |
|
|
operands[1] = adjust_address (operands[1], DFmode, INTVAL (operands[2]) * 8);
|
4547 |
|
|
})
|
4548 |
|
|
|
4549 |
|
|
(define_expand "avx_shufpd256"
|
4550 |
|
|
[(match_operand:V4DF 0 "register_operand" "")
|
4551 |
|
|
(match_operand:V4DF 1 "register_operand" "")
|
4552 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "")
|
4553 |
|
|
(match_operand:SI 3 "const_int_operand" "")]
|
4554 |
|
|
"TARGET_AVX"
|
4555 |
|
|
{
|
4556 |
|
|
int mask = INTVAL (operands[3]);
|
4557 |
|
|
emit_insn (gen_avx_shufpd256_1 (operands[0], operands[1], operands[2],
|
4558 |
|
|
GEN_INT (mask & 1),
|
4559 |
|
|
GEN_INT (mask & 2 ? 5 : 4),
|
4560 |
|
|
GEN_INT (mask & 4 ? 3 : 2),
|
4561 |
|
|
GEN_INT (mask & 8 ? 7 : 6)));
|
4562 |
|
|
DONE;
|
4563 |
|
|
})
|
4564 |
|
|
|
4565 |
|
|
(define_insn "avx_shufpd256_1"
|
4566 |
|
|
[(set (match_operand:V4DF 0 "register_operand" "=x")
|
4567 |
|
|
(vec_select:V4DF
|
4568 |
|
|
(vec_concat:V8DF
|
4569 |
|
|
(match_operand:V4DF 1 "register_operand" "x")
|
4570 |
|
|
(match_operand:V4DF 2 "nonimmediate_operand" "xm"))
|
4571 |
|
|
(parallel [(match_operand 3 "const_0_to_1_operand" "")
|
4572 |
|
|
(match_operand 4 "const_4_to_5_operand" "")
|
4573 |
|
|
(match_operand 5 "const_2_to_3_operand" "")
|
4574 |
|
|
(match_operand 6 "const_6_to_7_operand" "")])))]
|
4575 |
|
|
"TARGET_AVX"
|
4576 |
|
|
{
|
4577 |
|
|
int mask;
|
4578 |
|
|
mask = INTVAL (operands[3]);
|
4579 |
|
|
mask |= (INTVAL (operands[4]) - 4) << 1;
|
4580 |
|
|
mask |= (INTVAL (operands[5]) - 2) << 2;
|
4581 |
|
|
mask |= (INTVAL (operands[6]) - 6) << 3;
|
4582 |
|
|
operands[3] = GEN_INT (mask);
|
4583 |
|
|
|
4584 |
|
|
return "vshufpd\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
4585 |
|
|
}
|
4586 |
|
|
[(set_attr "type" "sselog")
|
4587 |
|
|
(set_attr "length_immediate" "1")
|
4588 |
|
|
(set_attr "prefix" "vex")
|
4589 |
|
|
(set_attr "mode" "V4DF")])
|
4590 |
|
|
|
4591 |
|
|
(define_expand "sse2_shufpd"
|
4592 |
|
|
[(match_operand:V2DF 0 "register_operand" "")
|
4593 |
|
|
(match_operand:V2DF 1 "register_operand" "")
|
4594 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" "")
|
4595 |
|
|
(match_operand:SI 3 "const_int_operand" "")]
|
4596 |
|
|
"TARGET_SSE2"
|
4597 |
|
|
{
|
4598 |
|
|
int mask = INTVAL (operands[3]);
|
4599 |
|
|
emit_insn (gen_sse2_shufpd_v2df (operands[0], operands[1], operands[2],
|
4600 |
|
|
GEN_INT (mask & 1),
|
4601 |
|
|
GEN_INT (mask & 2 ? 3 : 2)));
|
4602 |
|
|
DONE;
|
4603 |
|
|
})
|
4604 |
|
|
|
4605 |
|
|
(define_expand "vec_extract_even"
|
4606 |
|
|
[(match_operand:SSEMODE_EO 0 "register_operand" "")
|
4607 |
|
|
(match_operand:SSEMODE_EO 1 "register_operand" "")
|
4608 |
|
|
(match_operand:SSEMODE_EO 2 "register_operand" "")]
|
4609 |
|
|
""
|
4610 |
|
|
{
|
4611 |
|
|
ix86_expand_vec_extract_even_odd (operands[0], operands[1], operands[2], 0);
|
4612 |
|
|
DONE;
|
4613 |
|
|
})
|
4614 |
|
|
|
4615 |
|
|
(define_expand "vec_extract_odd"
|
4616 |
|
|
[(match_operand:SSEMODE_EO 0 "register_operand" "")
|
4617 |
|
|
(match_operand:SSEMODE_EO 1 "register_operand" "")
|
4618 |
|
|
(match_operand:SSEMODE_EO 2 "register_operand" "")]
|
4619 |
|
|
""
|
4620 |
|
|
{
|
4621 |
|
|
ix86_expand_vec_extract_even_odd (operands[0], operands[1], operands[2], 1);
|
4622 |
|
|
DONE;
|
4623 |
|
|
})
|
4624 |
|
|
|
4625 |
|
|
;; punpcklqdq and punpckhqdq are shorter than shufpd.
|
4626 |
|
|
(define_insn "*avx_interleave_highv2di"
|
4627 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
4628 |
|
|
(vec_select:V2DI
|
4629 |
|
|
(vec_concat:V4DI
|
4630 |
|
|
(match_operand:V2DI 1 "register_operand" "x")
|
4631 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm"))
|
4632 |
|
|
(parallel [(const_int 1)
|
4633 |
|
|
(const_int 3)])))]
|
4634 |
|
|
"TARGET_AVX"
|
4635 |
|
|
"vpunpckhqdq\t{%2, %1, %0|%0, %1, %2}"
|
4636 |
|
|
[(set_attr "type" "sselog")
|
4637 |
|
|
(set_attr "prefix" "vex")
|
4638 |
|
|
(set_attr "mode" "TI")])
|
4639 |
|
|
|
4640 |
|
|
(define_insn "vec_interleave_highv2di"
|
4641 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
4642 |
|
|
(vec_select:V2DI
|
4643 |
|
|
(vec_concat:V4DI
|
4644 |
|
|
(match_operand:V2DI 1 "register_operand" "0")
|
4645 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm"))
|
4646 |
|
|
(parallel [(const_int 1)
|
4647 |
|
|
(const_int 3)])))]
|
4648 |
|
|
"TARGET_SSE2"
|
4649 |
|
|
"punpckhqdq\t{%2, %0|%0, %2}"
|
4650 |
|
|
[(set_attr "type" "sselog")
|
4651 |
|
|
(set_attr "prefix_data16" "1")
|
4652 |
|
|
(set_attr "mode" "TI")])
|
4653 |
|
|
|
4654 |
|
|
(define_insn "*avx_interleave_lowv2di"
|
4655 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
4656 |
|
|
(vec_select:V2DI
|
4657 |
|
|
(vec_concat:V4DI
|
4658 |
|
|
(match_operand:V2DI 1 "register_operand" "x")
|
4659 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm"))
|
4660 |
|
|
(parallel [(const_int 0)
|
4661 |
|
|
(const_int 2)])))]
|
4662 |
|
|
"TARGET_AVX"
|
4663 |
|
|
"vpunpcklqdq\t{%2, %1, %0|%0, %1, %2}"
|
4664 |
|
|
[(set_attr "type" "sselog")
|
4665 |
|
|
(set_attr "prefix" "vex")
|
4666 |
|
|
(set_attr "mode" "TI")])
|
4667 |
|
|
|
4668 |
|
|
(define_insn "vec_interleave_lowv2di"
|
4669 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
4670 |
|
|
(vec_select:V2DI
|
4671 |
|
|
(vec_concat:V4DI
|
4672 |
|
|
(match_operand:V2DI 1 "register_operand" "0")
|
4673 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm"))
|
4674 |
|
|
(parallel [(const_int 0)
|
4675 |
|
|
(const_int 2)])))]
|
4676 |
|
|
"TARGET_SSE2"
|
4677 |
|
|
"punpcklqdq\t{%2, %0|%0, %2}"
|
4678 |
|
|
[(set_attr "type" "sselog")
|
4679 |
|
|
(set_attr "prefix_data16" "1")
|
4680 |
|
|
(set_attr "mode" "TI")])
|
4681 |
|
|
|
4682 |
|
|
(define_insn "*avx_shufpd_"
|
4683 |
|
|
[(set (match_operand:SSEMODE2D 0 "register_operand" "=x")
|
4684 |
|
|
(vec_select:SSEMODE2D
|
4685 |
|
|
(vec_concat:
|
4686 |
|
|
(match_operand:SSEMODE2D 1 "register_operand" "x")
|
4687 |
|
|
(match_operand:SSEMODE2D 2 "nonimmediate_operand" "xm"))
|
4688 |
|
|
(parallel [(match_operand 3 "const_0_to_1_operand" "")
|
4689 |
|
|
(match_operand 4 "const_2_to_3_operand" "")])))]
|
4690 |
|
|
"TARGET_AVX"
|
4691 |
|
|
{
|
4692 |
|
|
int mask;
|
4693 |
|
|
mask = INTVAL (operands[3]);
|
4694 |
|
|
mask |= (INTVAL (operands[4]) - 2) << 1;
|
4695 |
|
|
operands[3] = GEN_INT (mask);
|
4696 |
|
|
|
4697 |
|
|
return "vshufpd\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
4698 |
|
|
}
|
4699 |
|
|
[(set_attr "type" "sselog")
|
4700 |
|
|
(set_attr "length_immediate" "1")
|
4701 |
|
|
(set_attr "prefix" "vex")
|
4702 |
|
|
(set_attr "mode" "V2DF")])
|
4703 |
|
|
|
4704 |
|
|
(define_insn "sse2_shufpd_"
|
4705 |
|
|
[(set (match_operand:SSEMODE2D 0 "register_operand" "=x")
|
4706 |
|
|
(vec_select:SSEMODE2D
|
4707 |
|
|
(vec_concat:
|
4708 |
|
|
(match_operand:SSEMODE2D 1 "register_operand" "0")
|
4709 |
|
|
(match_operand:SSEMODE2D 2 "nonimmediate_operand" "xm"))
|
4710 |
|
|
(parallel [(match_operand 3 "const_0_to_1_operand" "")
|
4711 |
|
|
(match_operand 4 "const_2_to_3_operand" "")])))]
|
4712 |
|
|
"TARGET_SSE2"
|
4713 |
|
|
{
|
4714 |
|
|
int mask;
|
4715 |
|
|
mask = INTVAL (operands[3]);
|
4716 |
|
|
mask |= (INTVAL (operands[4]) - 2) << 1;
|
4717 |
|
|
operands[3] = GEN_INT (mask);
|
4718 |
|
|
|
4719 |
|
|
return "shufpd\t{%3, %2, %0|%0, %2, %3}";
|
4720 |
|
|
}
|
4721 |
|
|
[(set_attr "type" "sselog")
|
4722 |
|
|
(set_attr "length_immediate" "1")
|
4723 |
|
|
(set_attr "mode" "V2DF")])
|
4724 |
|
|
|
4725 |
|
|
;; Avoid combining registers from different units in a single alternative,
|
4726 |
|
|
;; see comment above inline_secondary_memory_needed function in i386.c
|
4727 |
|
|
(define_insn "*avx_storehpd"
|
4728 |
|
|
[(set (match_operand:DF 0 "nonimmediate_operand" "=m,x,x,*f,r")
|
4729 |
|
|
(vec_select:DF
|
4730 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " x,x,o,o,o")
|
4731 |
|
|
(parallel [(const_int 1)])))]
|
4732 |
|
|
"TARGET_AVX && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
4733 |
|
|
"@
|
4734 |
|
|
vmovhpd\t{%1, %0|%0, %1}
|
4735 |
|
|
vunpckhpd\t{%1, %1, %0|%0, %1, %1}
|
4736 |
|
|
#
|
4737 |
|
|
#
|
4738 |
|
|
#"
|
4739 |
|
|
[(set_attr "type" "ssemov,sselog1,ssemov,fmov,imov")
|
4740 |
|
|
(set_attr "prefix" "vex")
|
4741 |
|
|
(set_attr "mode" "V1DF,V2DF,DF,DF,DF")])
|
4742 |
|
|
|
4743 |
|
|
(define_insn "sse2_storehpd"
|
4744 |
|
|
[(set (match_operand:DF 0 "nonimmediate_operand" "=m,x,x,*f,r")
|
4745 |
|
|
(vec_select:DF
|
4746 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " x,0,o,o,o")
|
4747 |
|
|
(parallel [(const_int 1)])))]
|
4748 |
|
|
"TARGET_SSE2 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
4749 |
|
|
"@
|
4750 |
|
|
movhpd\t{%1, %0|%0, %1}
|
4751 |
|
|
unpckhpd\t%0, %0
|
4752 |
|
|
#
|
4753 |
|
|
#
|
4754 |
|
|
#"
|
4755 |
|
|
[(set_attr "type" "ssemov,sselog1,ssemov,fmov,imov")
|
4756 |
|
|
(set_attr "prefix_data16" "1,*,*,*,*")
|
4757 |
|
|
(set_attr "mode" "V1DF,V2DF,DF,DF,DF")])
|
4758 |
|
|
|
4759 |
|
|
(define_split
|
4760 |
|
|
[(set (match_operand:DF 0 "register_operand" "")
|
4761 |
|
|
(vec_select:DF
|
4762 |
|
|
(match_operand:V2DF 1 "memory_operand" "")
|
4763 |
|
|
(parallel [(const_int 1)])))]
|
4764 |
|
|
"TARGET_SSE2 && reload_completed"
|
4765 |
|
|
[(set (match_dup 0) (match_dup 1))]
|
4766 |
|
|
{
|
4767 |
|
|
operands[1] = adjust_address (operands[1], DFmode, 8);
|
4768 |
|
|
})
|
4769 |
|
|
|
4770 |
|
|
;; Avoid combining registers from different units in a single alternative,
|
4771 |
|
|
;; see comment above inline_secondary_memory_needed function in i386.c
|
4772 |
|
|
(define_insn "sse2_storelpd"
|
4773 |
|
|
[(set (match_operand:DF 0 "nonimmediate_operand" "=m,x,x,*f,r")
|
4774 |
|
|
(vec_select:DF
|
4775 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " x,x,m,m,m")
|
4776 |
|
|
(parallel [(const_int 0)])))]
|
4777 |
|
|
"TARGET_SSE2 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
4778 |
|
|
"@
|
4779 |
|
|
%vmovlpd\t{%1, %0|%0, %1}
|
4780 |
|
|
#
|
4781 |
|
|
#
|
4782 |
|
|
#
|
4783 |
|
|
#"
|
4784 |
|
|
[(set_attr "type" "ssemov,ssemov,ssemov,fmov,imov")
|
4785 |
|
|
(set_attr "prefix_data16" "1,*,*,*,*")
|
4786 |
|
|
(set_attr "prefix" "maybe_vex")
|
4787 |
|
|
(set_attr "mode" "V1DF,DF,DF,DF,DF")])
|
4788 |
|
|
|
4789 |
|
|
(define_split
|
4790 |
|
|
[(set (match_operand:DF 0 "register_operand" "")
|
4791 |
|
|
(vec_select:DF
|
4792 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "")
|
4793 |
|
|
(parallel [(const_int 0)])))]
|
4794 |
|
|
"TARGET_SSE2 && reload_completed"
|
4795 |
|
|
[(const_int 0)]
|
4796 |
|
|
{
|
4797 |
|
|
rtx op1 = operands[1];
|
4798 |
|
|
if (REG_P (op1))
|
4799 |
|
|
op1 = gen_rtx_REG (DFmode, REGNO (op1));
|
4800 |
|
|
else
|
4801 |
|
|
op1 = gen_lowpart (DFmode, op1);
|
4802 |
|
|
emit_move_insn (operands[0], op1);
|
4803 |
|
|
DONE;
|
4804 |
|
|
})
|
4805 |
|
|
|
4806 |
|
|
(define_expand "sse2_loadhpd_exp"
|
4807 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "")
|
4808 |
|
|
(vec_concat:V2DF
|
4809 |
|
|
(vec_select:DF
|
4810 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "")
|
4811 |
|
|
(parallel [(const_int 0)]))
|
4812 |
|
|
(match_operand:DF 2 "nonimmediate_operand" "")))]
|
4813 |
|
|
"TARGET_SSE2"
|
4814 |
|
|
"ix86_fixup_binary_operands (UNKNOWN, V2DFmode, operands);")
|
4815 |
|
|
|
4816 |
|
|
;; Avoid combining registers from different units in a single alternative,
|
4817 |
|
|
;; see comment above inline_secondary_memory_needed function in i386.c
|
4818 |
|
|
(define_insn "*avx_loadhpd"
|
4819 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,o,o,o")
|
4820 |
|
|
(vec_concat:V2DF
|
4821 |
|
|
(vec_select:DF
|
4822 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " x,x,0,0,0")
|
4823 |
|
|
(parallel [(const_int 0)]))
|
4824 |
|
|
(match_operand:DF 2 "nonimmediate_operand" " m,x,x,*f,r")))]
|
4825 |
|
|
"TARGET_AVX && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
4826 |
|
|
"@
|
4827 |
|
|
vmovhpd\t{%2, %1, %0|%0, %1, %2}
|
4828 |
|
|
vunpcklpd\t{%2, %1, %0|%0, %1, %2}
|
4829 |
|
|
#
|
4830 |
|
|
#
|
4831 |
|
|
#"
|
4832 |
|
|
[(set_attr "type" "ssemov,sselog,ssemov,fmov,imov")
|
4833 |
|
|
(set_attr "prefix" "vex")
|
4834 |
|
|
(set_attr "mode" "V1DF,V2DF,DF,DF,DF")])
|
4835 |
|
|
|
4836 |
|
|
(define_insn "sse2_loadhpd"
|
4837 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,x,o,o,o")
|
4838 |
|
|
(vec_concat:V2DF
|
4839 |
|
|
(vec_select:DF
|
4840 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " 0,0,x,0,0,0")
|
4841 |
|
|
(parallel [(const_int 0)]))
|
4842 |
|
|
(match_operand:DF 2 "nonimmediate_operand" " m,x,0,x,*f,r")))]
|
4843 |
|
|
"TARGET_SSE2 && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
4844 |
|
|
"@
|
4845 |
|
|
movhpd\t{%2, %0|%0, %2}
|
4846 |
|
|
unpcklpd\t{%2, %0|%0, %2}
|
4847 |
|
|
shufpd\t{$1, %1, %0|%0, %1, 1}
|
4848 |
|
|
#
|
4849 |
|
|
#
|
4850 |
|
|
#"
|
4851 |
|
|
[(set_attr "type" "ssemov,sselog,sselog,ssemov,fmov,imov")
|
4852 |
|
|
(set_attr "prefix_data16" "1,*,*,*,*,*")
|
4853 |
|
|
(set_attr "length_immediate" "*,*,1,*,*,*")
|
4854 |
|
|
(set_attr "mode" "V1DF,V2DF,V2DF,DF,DF,DF")])
|
4855 |
|
|
|
4856 |
|
|
(define_split
|
4857 |
|
|
[(set (match_operand:V2DF 0 "memory_operand" "")
|
4858 |
|
|
(vec_concat:V2DF
|
4859 |
|
|
(vec_select:DF (match_dup 0) (parallel [(const_int 0)]))
|
4860 |
|
|
(match_operand:DF 1 "register_operand" "")))]
|
4861 |
|
|
"TARGET_SSE2 && reload_completed"
|
4862 |
|
|
[(set (match_dup 0) (match_dup 1))]
|
4863 |
|
|
{
|
4864 |
|
|
operands[0] = adjust_address (operands[0], DFmode, 8);
|
4865 |
|
|
})
|
4866 |
|
|
|
4867 |
|
|
(define_expand "sse2_loadlpd_exp"
|
4868 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "")
|
4869 |
|
|
(vec_concat:V2DF
|
4870 |
|
|
(match_operand:DF 2 "nonimmediate_operand" "")
|
4871 |
|
|
(vec_select:DF
|
4872 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "")
|
4873 |
|
|
(parallel [(const_int 1)]))))]
|
4874 |
|
|
"TARGET_SSE2"
|
4875 |
|
|
"ix86_fixup_binary_operands (UNKNOWN, V2DFmode, operands);")
|
4876 |
|
|
|
4877 |
|
|
;; Avoid combining registers from different units in a single alternative,
|
4878 |
|
|
;; see comment above inline_secondary_memory_needed function in i386.c
|
4879 |
|
|
(define_insn "*avx_loadlpd"
|
4880 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,x,x,m,m,m")
|
4881 |
|
|
(vec_concat:V2DF
|
4882 |
|
|
(match_operand:DF 2 "nonimmediate_operand" " m,m,x,x,x,*f,r")
|
4883 |
|
|
(vec_select:DF
|
4884 |
|
|
(match_operand:V2DF 1 "vector_move_operand" " C,x,x,o,0,0,0")
|
4885 |
|
|
(parallel [(const_int 1)]))))]
|
4886 |
|
|
"TARGET_AVX && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
4887 |
|
|
"@
|
4888 |
|
|
vmovsd\t{%2, %0|%0, %2}
|
4889 |
|
|
vmovlpd\t{%2, %1, %0|%0, %1, %2}
|
4890 |
|
|
vmovsd\t{%2, %1, %0|%0, %1, %2}
|
4891 |
|
|
vmovhpd\t{%H1, %2, %0|%0, %2, %H1}
|
4892 |
|
|
#
|
4893 |
|
|
#
|
4894 |
|
|
#"
|
4895 |
|
|
[(set_attr "type" "ssemov,ssemov,ssemov,ssemov,ssemov,fmov,imov")
|
4896 |
|
|
(set_attr "prefix" "vex")
|
4897 |
|
|
(set_attr "mode" "DF,V1DF,V1DF,V1DF,DF,DF,DF")])
|
4898 |
|
|
|
4899 |
|
|
(define_insn "sse2_loadlpd"
|
4900 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,x,x,x,m,m,m")
|
4901 |
|
|
(vec_concat:V2DF
|
4902 |
|
|
(match_operand:DF 2 "nonimmediate_operand" " m,m,x,0,0,x,*f,r")
|
4903 |
|
|
(vec_select:DF
|
4904 |
|
|
(match_operand:V2DF 1 "vector_move_operand" " C,0,0,x,o,0,0,0")
|
4905 |
|
|
(parallel [(const_int 1)]))))]
|
4906 |
|
|
"TARGET_SSE2 && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
4907 |
|
|
"@
|
4908 |
|
|
movsd\t{%2, %0|%0, %2}
|
4909 |
|
|
movlpd\t{%2, %0|%0, %2}
|
4910 |
|
|
movsd\t{%2, %0|%0, %2}
|
4911 |
|
|
shufpd\t{$2, %2, %0|%0, %2, 2}
|
4912 |
|
|
movhpd\t{%H1, %0|%0, %H1}
|
4913 |
|
|
#
|
4914 |
|
|
#
|
4915 |
|
|
#"
|
4916 |
|
|
[(set_attr "type" "ssemov,ssemov,ssemov,sselog,ssemov,ssemov,fmov,imov")
|
4917 |
|
|
(set_attr "prefix_data16" "*,1,*,*,1,*,*,*")
|
4918 |
|
|
(set_attr "length_immediate" "*,*,*,1,*,*,*,*")
|
4919 |
|
|
(set_attr "mode" "DF,V1DF,V1DF,V2DF,V1DF,DF,DF,DF")])
|
4920 |
|
|
|
4921 |
|
|
(define_split
|
4922 |
|
|
[(set (match_operand:V2DF 0 "memory_operand" "")
|
4923 |
|
|
(vec_concat:V2DF
|
4924 |
|
|
(match_operand:DF 1 "register_operand" "")
|
4925 |
|
|
(vec_select:DF (match_dup 0) (parallel [(const_int 1)]))))]
|
4926 |
|
|
"TARGET_SSE2 && reload_completed"
|
4927 |
|
|
[(set (match_dup 0) (match_dup 1))]
|
4928 |
|
|
{
|
4929 |
|
|
operands[0] = adjust_address (operands[0], DFmode, 8);
|
4930 |
|
|
})
|
4931 |
|
|
|
4932 |
|
|
;; Not sure these two are ever used, but it doesn't hurt to have
|
4933 |
|
|
;; them. -aoliva
|
4934 |
|
|
(define_insn "*vec_extractv2df_1_sse"
|
4935 |
|
|
[(set (match_operand:DF 0 "nonimmediate_operand" "=m,x,x")
|
4936 |
|
|
(vec_select:DF
|
4937 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "x,x,o")
|
4938 |
|
|
(parallel [(const_int 1)])))]
|
4939 |
|
|
"!TARGET_SSE2 && TARGET_SSE
|
4940 |
|
|
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
4941 |
|
|
"@
|
4942 |
|
|
movhps\t{%1, %0|%0, %1}
|
4943 |
|
|
movhlps\t{%1, %0|%0, %1}
|
4944 |
|
|
movlps\t{%H1, %0|%0, %H1}"
|
4945 |
|
|
[(set_attr "type" "ssemov")
|
4946 |
|
|
(set_attr "mode" "V2SF,V4SF,V2SF")])
|
4947 |
|
|
|
4948 |
|
|
(define_insn "*vec_extractv2df_0_sse"
|
4949 |
|
|
[(set (match_operand:DF 0 "nonimmediate_operand" "=m,x,x")
|
4950 |
|
|
(vec_select:DF
|
4951 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" "x,x,m")
|
4952 |
|
|
(parallel [(const_int 0)])))]
|
4953 |
|
|
"!TARGET_SSE2 && TARGET_SSE
|
4954 |
|
|
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
4955 |
|
|
"@
|
4956 |
|
|
movlps\t{%1, %0|%0, %1}
|
4957 |
|
|
movaps\t{%1, %0|%0, %1}
|
4958 |
|
|
movlps\t{%1, %0|%0, %1}"
|
4959 |
|
|
[(set_attr "type" "ssemov")
|
4960 |
|
|
(set_attr "mode" "V2SF,V4SF,V2SF")])
|
4961 |
|
|
|
4962 |
|
|
(define_insn "*avx_movsd"
|
4963 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,m,x,o")
|
4964 |
|
|
(vec_merge:V2DF
|
4965 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" " x,m,x,x,0")
|
4966 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " x,x,0,o,x")
|
4967 |
|
|
(const_int 1)))]
|
4968 |
|
|
"TARGET_AVX"
|
4969 |
|
|
"@
|
4970 |
|
|
vmovsd\t{%2, %1, %0|%0, %1, %2}
|
4971 |
|
|
vmovlpd\t{%2, %1, %0|%0, %1, %2}
|
4972 |
|
|
vmovlpd\t{%2, %0|%0, %2}
|
4973 |
|
|
vmovhps\t{%H1, %2, %0|%0, %2, %H1}
|
4974 |
|
|
vmovhps\t{%1, %H0|%H0, %1}"
|
4975 |
|
|
[(set_attr "type" "ssemov,ssemov,ssemov,ssemov,ssemov")
|
4976 |
|
|
(set_attr "prefix" "vex")
|
4977 |
|
|
(set_attr "mode" "DF,V1DF,V1DF,V1DF,V1DF")])
|
4978 |
|
|
|
4979 |
|
|
(define_insn "sse2_movsd"
|
4980 |
|
|
[(set (match_operand:V2DF 0 "nonimmediate_operand" "=x,x,m,x,x,o")
|
4981 |
|
|
(vec_merge:V2DF
|
4982 |
|
|
(match_operand:V2DF 2 "nonimmediate_operand" " x,m,x,0,0,0")
|
4983 |
|
|
(match_operand:V2DF 1 "nonimmediate_operand" " 0,0,0,x,o,x")
|
4984 |
|
|
(const_int 1)))]
|
4985 |
|
|
"TARGET_SSE2"
|
4986 |
|
|
"@
|
4987 |
|
|
movsd\t{%2, %0|%0, %2}
|
4988 |
|
|
movlpd\t{%2, %0|%0, %2}
|
4989 |
|
|
movlpd\t{%2, %0|%0, %2}
|
4990 |
|
|
shufpd\t{$2, %2, %0|%0, %2, 2}
|
4991 |
|
|
movhps\t{%H1, %0|%0, %H1}
|
4992 |
|
|
movhps\t{%1, %H0|%H0, %1}"
|
4993 |
|
|
[(set_attr "type" "ssemov,ssemov,ssemov,sselog,ssemov,ssemov")
|
4994 |
|
|
(set_attr "prefix_data16" "*,1,1,*,*,*")
|
4995 |
|
|
(set_attr "length_immediate" "*,*,*,1,*,*")
|
4996 |
|
|
(set_attr "mode" "DF,V1DF,V1DF,V2DF,V1DF,V1DF")])
|
4997 |
|
|
|
4998 |
|
|
(define_insn "*vec_dupv2df_sse3"
|
4999 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
5000 |
|
|
(vec_duplicate:V2DF
|
5001 |
|
|
(match_operand:DF 1 "nonimmediate_operand" "xm")))]
|
5002 |
|
|
"TARGET_SSE3"
|
5003 |
|
|
"%vmovddup\t{%1, %0|%0, %1}"
|
5004 |
|
|
[(set_attr "type" "sselog1")
|
5005 |
|
|
(set_attr "prefix" "maybe_vex")
|
5006 |
|
|
(set_attr "mode" "DF")])
|
5007 |
|
|
|
5008 |
|
|
(define_insn "vec_dupv2df"
|
5009 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
5010 |
|
|
(vec_duplicate:V2DF
|
5011 |
|
|
(match_operand:DF 1 "register_operand" "0")))]
|
5012 |
|
|
"TARGET_SSE2"
|
5013 |
|
|
"unpcklpd\t%0, %0"
|
5014 |
|
|
[(set_attr "type" "sselog1")
|
5015 |
|
|
(set_attr "mode" "V2DF")])
|
5016 |
|
|
|
5017 |
|
|
(define_insn "*vec_concatv2df_sse3"
|
5018 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x")
|
5019 |
|
|
(vec_concat:V2DF
|
5020 |
|
|
(match_operand:DF 1 "nonimmediate_operand" "xm")
|
5021 |
|
|
(match_dup 1)))]
|
5022 |
|
|
"TARGET_SSE3"
|
5023 |
|
|
"%vmovddup\t{%1, %0|%0, %1}"
|
5024 |
|
|
[(set_attr "type" "sselog1")
|
5025 |
|
|
(set_attr "prefix" "maybe_vex")
|
5026 |
|
|
(set_attr "mode" "DF")])
|
5027 |
|
|
|
5028 |
|
|
(define_insn "*vec_concatv2df_avx"
|
5029 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=x,x,x")
|
5030 |
|
|
(vec_concat:V2DF
|
5031 |
|
|
(match_operand:DF 1 "nonimmediate_operand" " x,x,m")
|
5032 |
|
|
(match_operand:DF 2 "vector_move_operand" " x,m,C")))]
|
5033 |
|
|
"TARGET_AVX"
|
5034 |
|
|
"@
|
5035 |
|
|
vunpcklpd\t{%2, %1, %0|%0, %1, %2}
|
5036 |
|
|
vmovhpd\t{%2, %1, %0|%0, %1, %2}
|
5037 |
|
|
vmovsd\t{%1, %0|%0, %1}"
|
5038 |
|
|
[(set_attr "type" "ssemov")
|
5039 |
|
|
(set_attr "prefix" "vex")
|
5040 |
|
|
(set_attr "mode" "DF,V1DF,DF")])
|
5041 |
|
|
|
5042 |
|
|
(define_insn "*vec_concatv2df"
|
5043 |
|
|
[(set (match_operand:V2DF 0 "register_operand" "=Y2,Y2,Y2,x,x")
|
5044 |
|
|
(vec_concat:V2DF
|
5045 |
|
|
(match_operand:DF 1 "nonimmediate_operand" " 0 ,0 ,m ,0,0")
|
5046 |
|
|
(match_operand:DF 2 "vector_move_operand" " Y2,m ,C ,x,m")))]
|
5047 |
|
|
"TARGET_SSE"
|
5048 |
|
|
"@
|
5049 |
|
|
unpcklpd\t{%2, %0|%0, %2}
|
5050 |
|
|
movhpd\t{%2, %0|%0, %2}
|
5051 |
|
|
movsd\t{%1, %0|%0, %1}
|
5052 |
|
|
movlhps\t{%2, %0|%0, %2}
|
5053 |
|
|
movhps\t{%2, %0|%0, %2}"
|
5054 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov,ssemov,ssemov")
|
5055 |
|
|
(set_attr "prefix_data16" "*,1,*,*,*")
|
5056 |
|
|
(set_attr "mode" "V2DF,V1DF,DF,V4SF,V2SF")])
|
5057 |
|
|
|
5058 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
5059 |
|
|
;;
|
5060 |
|
|
;; Parallel integral arithmetic
|
5061 |
|
|
;;
|
5062 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
5063 |
|
|
|
5064 |
|
|
(define_expand "neg2"
|
5065 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "")
|
5066 |
|
|
(minus:SSEMODEI
|
5067 |
|
|
(match_dup 2)
|
5068 |
|
|
(match_operand:SSEMODEI 1 "nonimmediate_operand" "")))]
|
5069 |
|
|
"TARGET_SSE2"
|
5070 |
|
|
"operands[2] = force_reg (mode, CONST0_RTX (mode));")
|
5071 |
|
|
|
5072 |
|
|
(define_expand "3"
|
5073 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "")
|
5074 |
|
|
(plusminus:SSEMODEI
|
5075 |
|
|
(match_operand:SSEMODEI 1 "nonimmediate_operand" "")
|
5076 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "")))]
|
5077 |
|
|
"TARGET_SSE2"
|
5078 |
|
|
"ix86_fixup_binary_operands_no_copy (, mode, operands);")
|
5079 |
|
|
|
5080 |
|
|
(define_insn "*avx_3"
|
5081 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "=x")
|
5082 |
|
|
(plusminus:SSEMODEI
|
5083 |
|
|
(match_operand:SSEMODEI 1 "nonimmediate_operand" "x")
|
5084 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "xm")))]
|
5085 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (, mode, operands)"
|
5086 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
5087 |
|
|
[(set_attr "type" "sseiadd")
|
5088 |
|
|
(set_attr "prefix" "vex")
|
5089 |
|
|
(set_attr "mode" "TI")])
|
5090 |
|
|
|
5091 |
|
|
(define_insn "*3"
|
5092 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "=x")
|
5093 |
|
|
(plusminus:SSEMODEI
|
5094 |
|
|
(match_operand:SSEMODEI 1 "nonimmediate_operand" "0")
|
5095 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "xm")))]
|
5096 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (, mode, operands)"
|
5097 |
|
|
"p\t{%2, %0|%0, %2}"
|
5098 |
|
|
[(set_attr "type" "sseiadd")
|
5099 |
|
|
(set_attr "prefix_data16" "1")
|
5100 |
|
|
(set_attr "mode" "TI")])
|
5101 |
|
|
|
5102 |
|
|
(define_expand "sse2_3"
|
5103 |
|
|
[(set (match_operand:SSEMODE12 0 "register_operand" "")
|
5104 |
|
|
(sat_plusminus:SSEMODE12
|
5105 |
|
|
(match_operand:SSEMODE12 1 "nonimmediate_operand" "")
|
5106 |
|
|
(match_operand:SSEMODE12 2 "nonimmediate_operand" "")))]
|
5107 |
|
|
"TARGET_SSE2"
|
5108 |
|
|
"ix86_fixup_binary_operands_no_copy (, mode, operands);")
|
5109 |
|
|
|
5110 |
|
|
(define_insn "*avx_3"
|
5111 |
|
|
[(set (match_operand:SSEMODE12 0 "register_operand" "=x")
|
5112 |
|
|
(sat_plusminus:SSEMODE12
|
5113 |
|
|
(match_operand:SSEMODE12 1 "nonimmediate_operand" "x")
|
5114 |
|
|
(match_operand:SSEMODE12 2 "nonimmediate_operand" "xm")))]
|
5115 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (, mode, operands)"
|
5116 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
5117 |
|
|
[(set_attr "type" "sseiadd")
|
5118 |
|
|
(set_attr "prefix" "vex")
|
5119 |
|
|
(set_attr "mode" "TI")])
|
5120 |
|
|
|
5121 |
|
|
(define_insn "*sse2_3"
|
5122 |
|
|
[(set (match_operand:SSEMODE12 0 "register_operand" "=x")
|
5123 |
|
|
(sat_plusminus:SSEMODE12
|
5124 |
|
|
(match_operand:SSEMODE12 1 "nonimmediate_operand" "0")
|
5125 |
|
|
(match_operand:SSEMODE12 2 "nonimmediate_operand" "xm")))]
|
5126 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (, mode, operands)"
|
5127 |
|
|
"p\t{%2, %0|%0, %2}"
|
5128 |
|
|
[(set_attr "type" "sseiadd")
|
5129 |
|
|
(set_attr "prefix_data16" "1")
|
5130 |
|
|
(set_attr "mode" "TI")])
|
5131 |
|
|
|
5132 |
|
|
(define_insn_and_split "mulv16qi3"
|
5133 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "")
|
5134 |
|
|
(mult:V16QI (match_operand:V16QI 1 "register_operand" "")
|
5135 |
|
|
(match_operand:V16QI 2 "register_operand" "")))]
|
5136 |
|
|
"TARGET_SSE2
|
5137 |
|
|
&& can_create_pseudo_p ()"
|
5138 |
|
|
"#"
|
5139 |
|
|
"&& 1"
|
5140 |
|
|
[(const_int 0)]
|
5141 |
|
|
{
|
5142 |
|
|
rtx t[6];
|
5143 |
|
|
int i;
|
5144 |
|
|
|
5145 |
|
|
for (i = 0; i < 6; ++i)
|
5146 |
|
|
t[i] = gen_reg_rtx (V16QImode);
|
5147 |
|
|
|
5148 |
|
|
/* Unpack data such that we've got a source byte in each low byte of
|
5149 |
|
|
each word. We don't care what goes into the high byte of each word.
|
5150 |
|
|
Rather than trying to get zero in there, most convenient is to let
|
5151 |
|
|
it be a copy of the low byte. */
|
5152 |
|
|
emit_insn (gen_vec_interleave_highv16qi (t[0], operands[1], operands[1]));
|
5153 |
|
|
emit_insn (gen_vec_interleave_highv16qi (t[1], operands[2], operands[2]));
|
5154 |
|
|
emit_insn (gen_vec_interleave_lowv16qi (t[2], operands[1], operands[1]));
|
5155 |
|
|
emit_insn (gen_vec_interleave_lowv16qi (t[3], operands[2], operands[2]));
|
5156 |
|
|
|
5157 |
|
|
/* Multiply words. The end-of-line annotations here give a picture of what
|
5158 |
|
|
the output of that instruction looks like. Dot means don't care; the
|
5159 |
|
|
letters are the bytes of the result with A being the most significant. */
|
5160 |
|
|
emit_insn (gen_mulv8hi3 (gen_lowpart (V8HImode, t[4]), /* .A.B.C.D.E.F.G.H */
|
5161 |
|
|
gen_lowpart (V8HImode, t[0]),
|
5162 |
|
|
gen_lowpart (V8HImode, t[1])));
|
5163 |
|
|
emit_insn (gen_mulv8hi3 (gen_lowpart (V8HImode, t[5]), /* .I.J.K.L.M.N.O.P */
|
5164 |
|
|
gen_lowpart (V8HImode, t[2]),
|
5165 |
|
|
gen_lowpart (V8HImode, t[3])));
|
5166 |
|
|
|
5167 |
|
|
/* Extract the even bytes and merge them back together. */
|
5168 |
|
|
ix86_expand_vec_extract_even_odd (operands[0], t[5], t[4], 0);
|
5169 |
|
|
DONE;
|
5170 |
|
|
})
|
5171 |
|
|
|
5172 |
|
|
(define_expand "mulv8hi3"
|
5173 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "")
|
5174 |
|
|
(mult:V8HI (match_operand:V8HI 1 "nonimmediate_operand" "")
|
5175 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "")))]
|
5176 |
|
|
"TARGET_SSE2"
|
5177 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, V8HImode, operands);")
|
5178 |
|
|
|
5179 |
|
|
(define_insn "*avx_mulv8hi3"
|
5180 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
5181 |
|
|
(mult:V8HI (match_operand:V8HI 1 "nonimmediate_operand" "%x")
|
5182 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))]
|
5183 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
5184 |
|
|
"vpmullw\t{%2, %1, %0|%0, %1, %2}"
|
5185 |
|
|
[(set_attr "type" "sseimul")
|
5186 |
|
|
(set_attr "prefix" "vex")
|
5187 |
|
|
(set_attr "mode" "TI")])
|
5188 |
|
|
|
5189 |
|
|
(define_insn "*mulv8hi3"
|
5190 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
5191 |
|
|
(mult:V8HI (match_operand:V8HI 1 "nonimmediate_operand" "%0")
|
5192 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))]
|
5193 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
5194 |
|
|
"pmullw\t{%2, %0|%0, %2}"
|
5195 |
|
|
[(set_attr "type" "sseimul")
|
5196 |
|
|
(set_attr "prefix_data16" "1")
|
5197 |
|
|
(set_attr "mode" "TI")])
|
5198 |
|
|
|
5199 |
|
|
(define_expand "smulv8hi3_highpart"
|
5200 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "")
|
5201 |
|
|
(truncate:V8HI
|
5202 |
|
|
(lshiftrt:V8SI
|
5203 |
|
|
(mult:V8SI
|
5204 |
|
|
(sign_extend:V8SI
|
5205 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" ""))
|
5206 |
|
|
(sign_extend:V8SI
|
5207 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "")))
|
5208 |
|
|
(const_int 16))))]
|
5209 |
|
|
"TARGET_SSE2"
|
5210 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, V8HImode, operands);")
|
5211 |
|
|
|
5212 |
|
|
(define_insn "*avxv8hi3_highpart"
|
5213 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
5214 |
|
|
(truncate:V8HI
|
5215 |
|
|
(lshiftrt:V8SI
|
5216 |
|
|
(mult:V8SI
|
5217 |
|
|
(sign_extend:V8SI
|
5218 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x"))
|
5219 |
|
|
(sign_extend:V8SI
|
5220 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))
|
5221 |
|
|
(const_int 16))))]
|
5222 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
5223 |
|
|
"vpmulhw\t{%2, %1, %0|%0, %1, %2}"
|
5224 |
|
|
[(set_attr "type" "sseimul")
|
5225 |
|
|
(set_attr "prefix" "vex")
|
5226 |
|
|
(set_attr "mode" "TI")])
|
5227 |
|
|
|
5228 |
|
|
(define_insn "*smulv8hi3_highpart"
|
5229 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
5230 |
|
|
(truncate:V8HI
|
5231 |
|
|
(lshiftrt:V8SI
|
5232 |
|
|
(mult:V8SI
|
5233 |
|
|
(sign_extend:V8SI
|
5234 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%0"))
|
5235 |
|
|
(sign_extend:V8SI
|
5236 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))
|
5237 |
|
|
(const_int 16))))]
|
5238 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
5239 |
|
|
"pmulhw\t{%2, %0|%0, %2}"
|
5240 |
|
|
[(set_attr "type" "sseimul")
|
5241 |
|
|
(set_attr "prefix_data16" "1")
|
5242 |
|
|
(set_attr "mode" "TI")])
|
5243 |
|
|
|
5244 |
|
|
(define_expand "umulv8hi3_highpart"
|
5245 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "")
|
5246 |
|
|
(truncate:V8HI
|
5247 |
|
|
(lshiftrt:V8SI
|
5248 |
|
|
(mult:V8SI
|
5249 |
|
|
(zero_extend:V8SI
|
5250 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" ""))
|
5251 |
|
|
(zero_extend:V8SI
|
5252 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "")))
|
5253 |
|
|
(const_int 16))))]
|
5254 |
|
|
"TARGET_SSE2"
|
5255 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, V8HImode, operands);")
|
5256 |
|
|
|
5257 |
|
|
(define_insn "*avx_umulv8hi3_highpart"
|
5258 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
5259 |
|
|
(truncate:V8HI
|
5260 |
|
|
(lshiftrt:V8SI
|
5261 |
|
|
(mult:V8SI
|
5262 |
|
|
(zero_extend:V8SI
|
5263 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x"))
|
5264 |
|
|
(zero_extend:V8SI
|
5265 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))
|
5266 |
|
|
(const_int 16))))]
|
5267 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
5268 |
|
|
"vpmulhuw\t{%2, %1, %0|%0, %1, %2}"
|
5269 |
|
|
[(set_attr "type" "sseimul")
|
5270 |
|
|
(set_attr "prefix" "vex")
|
5271 |
|
|
(set_attr "mode" "TI")])
|
5272 |
|
|
|
5273 |
|
|
(define_insn "*umulv8hi3_highpart"
|
5274 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
5275 |
|
|
(truncate:V8HI
|
5276 |
|
|
(lshiftrt:V8SI
|
5277 |
|
|
(mult:V8SI
|
5278 |
|
|
(zero_extend:V8SI
|
5279 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%0"))
|
5280 |
|
|
(zero_extend:V8SI
|
5281 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))
|
5282 |
|
|
(const_int 16))))]
|
5283 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
5284 |
|
|
"pmulhuw\t{%2, %0|%0, %2}"
|
5285 |
|
|
[(set_attr "type" "sseimul")
|
5286 |
|
|
(set_attr "prefix_data16" "1")
|
5287 |
|
|
(set_attr "mode" "TI")])
|
5288 |
|
|
|
5289 |
|
|
(define_expand "sse2_umulv2siv2di3"
|
5290 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "")
|
5291 |
|
|
(mult:V2DI
|
5292 |
|
|
(zero_extend:V2DI
|
5293 |
|
|
(vec_select:V2SI
|
5294 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "")
|
5295 |
|
|
(parallel [(const_int 0) (const_int 2)])))
|
5296 |
|
|
(zero_extend:V2DI
|
5297 |
|
|
(vec_select:V2SI
|
5298 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "")
|
5299 |
|
|
(parallel [(const_int 0) (const_int 2)])))))]
|
5300 |
|
|
"TARGET_SSE2"
|
5301 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, V4SImode, operands);")
|
5302 |
|
|
|
5303 |
|
|
(define_insn "*avx_umulv2siv2di3"
|
5304 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
5305 |
|
|
(mult:V2DI
|
5306 |
|
|
(zero_extend:V2DI
|
5307 |
|
|
(vec_select:V2SI
|
5308 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
5309 |
|
|
(parallel [(const_int 0) (const_int 2)])))
|
5310 |
|
|
(zero_extend:V2DI
|
5311 |
|
|
(vec_select:V2SI
|
5312 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
5313 |
|
|
(parallel [(const_int 0) (const_int 2)])))))]
|
5314 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (MULT, V4SImode, operands)"
|
5315 |
|
|
"vpmuludq\t{%2, %1, %0|%0, %1, %2}"
|
5316 |
|
|
[(set_attr "type" "sseimul")
|
5317 |
|
|
(set_attr "prefix" "vex")
|
5318 |
|
|
(set_attr "mode" "TI")])
|
5319 |
|
|
|
5320 |
|
|
(define_insn "*sse2_umulv2siv2di3"
|
5321 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
5322 |
|
|
(mult:V2DI
|
5323 |
|
|
(zero_extend:V2DI
|
5324 |
|
|
(vec_select:V2SI
|
5325 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%0")
|
5326 |
|
|
(parallel [(const_int 0) (const_int 2)])))
|
5327 |
|
|
(zero_extend:V2DI
|
5328 |
|
|
(vec_select:V2SI
|
5329 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
5330 |
|
|
(parallel [(const_int 0) (const_int 2)])))))]
|
5331 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (MULT, V4SImode, operands)"
|
5332 |
|
|
"pmuludq\t{%2, %0|%0, %2}"
|
5333 |
|
|
[(set_attr "type" "sseimul")
|
5334 |
|
|
(set_attr "prefix_data16" "1")
|
5335 |
|
|
(set_attr "mode" "TI")])
|
5336 |
|
|
|
5337 |
|
|
(define_expand "sse4_1_mulv2siv2di3"
|
5338 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "")
|
5339 |
|
|
(mult:V2DI
|
5340 |
|
|
(sign_extend:V2DI
|
5341 |
|
|
(vec_select:V2SI
|
5342 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "")
|
5343 |
|
|
(parallel [(const_int 0) (const_int 2)])))
|
5344 |
|
|
(sign_extend:V2DI
|
5345 |
|
|
(vec_select:V2SI
|
5346 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "")
|
5347 |
|
|
(parallel [(const_int 0) (const_int 2)])))))]
|
5348 |
|
|
"TARGET_SSE4_1"
|
5349 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, V4SImode, operands);")
|
5350 |
|
|
|
5351 |
|
|
(define_insn "*avx_mulv2siv2di3"
|
5352 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
5353 |
|
|
(mult:V2DI
|
5354 |
|
|
(sign_extend:V2DI
|
5355 |
|
|
(vec_select:V2SI
|
5356 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
5357 |
|
|
(parallel [(const_int 0) (const_int 2)])))
|
5358 |
|
|
(sign_extend:V2DI
|
5359 |
|
|
(vec_select:V2SI
|
5360 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
5361 |
|
|
(parallel [(const_int 0) (const_int 2)])))))]
|
5362 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (MULT, V4SImode, operands)"
|
5363 |
|
|
"vpmuldq\t{%2, %1, %0|%0, %1, %2}"
|
5364 |
|
|
[(set_attr "type" "sseimul")
|
5365 |
|
|
(set_attr "prefix_extra" "1")
|
5366 |
|
|
(set_attr "prefix" "vex")
|
5367 |
|
|
(set_attr "mode" "TI")])
|
5368 |
|
|
|
5369 |
|
|
(define_insn "*sse4_1_mulv2siv2di3"
|
5370 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
5371 |
|
|
(mult:V2DI
|
5372 |
|
|
(sign_extend:V2DI
|
5373 |
|
|
(vec_select:V2SI
|
5374 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%0")
|
5375 |
|
|
(parallel [(const_int 0) (const_int 2)])))
|
5376 |
|
|
(sign_extend:V2DI
|
5377 |
|
|
(vec_select:V2SI
|
5378 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
5379 |
|
|
(parallel [(const_int 0) (const_int 2)])))))]
|
5380 |
|
|
"TARGET_SSE4_1 && ix86_binary_operator_ok (MULT, V4SImode, operands)"
|
5381 |
|
|
"pmuldq\t{%2, %0|%0, %2}"
|
5382 |
|
|
[(set_attr "type" "sseimul")
|
5383 |
|
|
(set_attr "prefix_extra" "1")
|
5384 |
|
|
(set_attr "mode" "TI")])
|
5385 |
|
|
|
5386 |
|
|
(define_expand "sse2_pmaddwd"
|
5387 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "")
|
5388 |
|
|
(plus:V4SI
|
5389 |
|
|
(mult:V4SI
|
5390 |
|
|
(sign_extend:V4SI
|
5391 |
|
|
(vec_select:V4HI
|
5392 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "")
|
5393 |
|
|
(parallel [(const_int 0)
|
5394 |
|
|
(const_int 2)
|
5395 |
|
|
(const_int 4)
|
5396 |
|
|
(const_int 6)])))
|
5397 |
|
|
(sign_extend:V4SI
|
5398 |
|
|
(vec_select:V4HI
|
5399 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "")
|
5400 |
|
|
(parallel [(const_int 0)
|
5401 |
|
|
(const_int 2)
|
5402 |
|
|
(const_int 4)
|
5403 |
|
|
(const_int 6)]))))
|
5404 |
|
|
(mult:V4SI
|
5405 |
|
|
(sign_extend:V4SI
|
5406 |
|
|
(vec_select:V4HI (match_dup 1)
|
5407 |
|
|
(parallel [(const_int 1)
|
5408 |
|
|
(const_int 3)
|
5409 |
|
|
(const_int 5)
|
5410 |
|
|
(const_int 7)])))
|
5411 |
|
|
(sign_extend:V4SI
|
5412 |
|
|
(vec_select:V4HI (match_dup 2)
|
5413 |
|
|
(parallel [(const_int 1)
|
5414 |
|
|
(const_int 3)
|
5415 |
|
|
(const_int 5)
|
5416 |
|
|
(const_int 7)]))))))]
|
5417 |
|
|
"TARGET_SSE2"
|
5418 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, V8HImode, operands);")
|
5419 |
|
|
|
5420 |
|
|
(define_insn "*avx_pmaddwd"
|
5421 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
5422 |
|
|
(plus:V4SI
|
5423 |
|
|
(mult:V4SI
|
5424 |
|
|
(sign_extend:V4SI
|
5425 |
|
|
(vec_select:V4HI
|
5426 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x")
|
5427 |
|
|
(parallel [(const_int 0)
|
5428 |
|
|
(const_int 2)
|
5429 |
|
|
(const_int 4)
|
5430 |
|
|
(const_int 6)])))
|
5431 |
|
|
(sign_extend:V4SI
|
5432 |
|
|
(vec_select:V4HI
|
5433 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
5434 |
|
|
(parallel [(const_int 0)
|
5435 |
|
|
(const_int 2)
|
5436 |
|
|
(const_int 4)
|
5437 |
|
|
(const_int 6)]))))
|
5438 |
|
|
(mult:V4SI
|
5439 |
|
|
(sign_extend:V4SI
|
5440 |
|
|
(vec_select:V4HI (match_dup 1)
|
5441 |
|
|
(parallel [(const_int 1)
|
5442 |
|
|
(const_int 3)
|
5443 |
|
|
(const_int 5)
|
5444 |
|
|
(const_int 7)])))
|
5445 |
|
|
(sign_extend:V4SI
|
5446 |
|
|
(vec_select:V4HI (match_dup 2)
|
5447 |
|
|
(parallel [(const_int 1)
|
5448 |
|
|
(const_int 3)
|
5449 |
|
|
(const_int 5)
|
5450 |
|
|
(const_int 7)]))))))]
|
5451 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
5452 |
|
|
"vpmaddwd\t{%2, %1, %0|%0, %1, %2}"
|
5453 |
|
|
[(set_attr "type" "sseiadd")
|
5454 |
|
|
(set_attr "prefix" "vex")
|
5455 |
|
|
(set_attr "mode" "TI")])
|
5456 |
|
|
|
5457 |
|
|
(define_insn "*sse2_pmaddwd"
|
5458 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
5459 |
|
|
(plus:V4SI
|
5460 |
|
|
(mult:V4SI
|
5461 |
|
|
(sign_extend:V4SI
|
5462 |
|
|
(vec_select:V4HI
|
5463 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%0")
|
5464 |
|
|
(parallel [(const_int 0)
|
5465 |
|
|
(const_int 2)
|
5466 |
|
|
(const_int 4)
|
5467 |
|
|
(const_int 6)])))
|
5468 |
|
|
(sign_extend:V4SI
|
5469 |
|
|
(vec_select:V4HI
|
5470 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
5471 |
|
|
(parallel [(const_int 0)
|
5472 |
|
|
(const_int 2)
|
5473 |
|
|
(const_int 4)
|
5474 |
|
|
(const_int 6)]))))
|
5475 |
|
|
(mult:V4SI
|
5476 |
|
|
(sign_extend:V4SI
|
5477 |
|
|
(vec_select:V4HI (match_dup 1)
|
5478 |
|
|
(parallel [(const_int 1)
|
5479 |
|
|
(const_int 3)
|
5480 |
|
|
(const_int 5)
|
5481 |
|
|
(const_int 7)])))
|
5482 |
|
|
(sign_extend:V4SI
|
5483 |
|
|
(vec_select:V4HI (match_dup 2)
|
5484 |
|
|
(parallel [(const_int 1)
|
5485 |
|
|
(const_int 3)
|
5486 |
|
|
(const_int 5)
|
5487 |
|
|
(const_int 7)]))))))]
|
5488 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
5489 |
|
|
"pmaddwd\t{%2, %0|%0, %2}"
|
5490 |
|
|
[(set_attr "type" "sseiadd")
|
5491 |
|
|
(set_attr "atom_unit" "simul")
|
5492 |
|
|
(set_attr "prefix_data16" "1")
|
5493 |
|
|
(set_attr "mode" "TI")])
|
5494 |
|
|
|
5495 |
|
|
(define_expand "mulv4si3"
|
5496 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "")
|
5497 |
|
|
(mult:V4SI (match_operand:V4SI 1 "register_operand" "")
|
5498 |
|
|
(match_operand:V4SI 2 "register_operand" "")))]
|
5499 |
|
|
"TARGET_SSE2"
|
5500 |
|
|
{
|
5501 |
|
|
if (TARGET_SSE4_1 || TARGET_AVX)
|
5502 |
|
|
ix86_fixup_binary_operands_no_copy (MULT, V4SImode, operands);
|
5503 |
|
|
})
|
5504 |
|
|
|
5505 |
|
|
(define_insn "*avx_mulv4si3"
|
5506 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
5507 |
|
|
(mult:V4SI (match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
5508 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")))]
|
5509 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (MULT, V4SImode, operands)"
|
5510 |
|
|
"vpmulld\t{%2, %1, %0|%0, %1, %2}"
|
5511 |
|
|
[(set_attr "type" "sseimul")
|
5512 |
|
|
(set_attr "prefix_extra" "1")
|
5513 |
|
|
(set_attr "prefix" "vex")
|
5514 |
|
|
(set_attr "mode" "TI")])
|
5515 |
|
|
|
5516 |
|
|
(define_insn "*sse4_1_mulv4si3"
|
5517 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
5518 |
|
|
(mult:V4SI (match_operand:V4SI 1 "nonimmediate_operand" "%0")
|
5519 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")))]
|
5520 |
|
|
"TARGET_SSE4_1 && ix86_binary_operator_ok (MULT, V4SImode, operands)"
|
5521 |
|
|
"pmulld\t{%2, %0|%0, %2}"
|
5522 |
|
|
[(set_attr "type" "sseimul")
|
5523 |
|
|
(set_attr "prefix_extra" "1")
|
5524 |
|
|
(set_attr "mode" "TI")])
|
5525 |
|
|
|
5526 |
|
|
(define_insn_and_split "*sse2_mulv4si3"
|
5527 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "")
|
5528 |
|
|
(mult:V4SI (match_operand:V4SI 1 "register_operand" "")
|
5529 |
|
|
(match_operand:V4SI 2 "register_operand" "")))]
|
5530 |
|
|
"TARGET_SSE2 && !TARGET_SSE4_1 && !TARGET_AVX
|
5531 |
|
|
&& can_create_pseudo_p ()"
|
5532 |
|
|
"#"
|
5533 |
|
|
"&& 1"
|
5534 |
|
|
[(const_int 0)]
|
5535 |
|
|
{
|
5536 |
|
|
rtx t1, t2, t3, t4, t5, t6, thirtytwo;
|
5537 |
|
|
rtx op0, op1, op2;
|
5538 |
|
|
|
5539 |
|
|
op0 = operands[0];
|
5540 |
|
|
op1 = operands[1];
|
5541 |
|
|
op2 = operands[2];
|
5542 |
|
|
t1 = gen_reg_rtx (V4SImode);
|
5543 |
|
|
t2 = gen_reg_rtx (V4SImode);
|
5544 |
|
|
t3 = gen_reg_rtx (V4SImode);
|
5545 |
|
|
t4 = gen_reg_rtx (V4SImode);
|
5546 |
|
|
t5 = gen_reg_rtx (V4SImode);
|
5547 |
|
|
t6 = gen_reg_rtx (V4SImode);
|
5548 |
|
|
thirtytwo = GEN_INT (32);
|
5549 |
|
|
|
5550 |
|
|
/* Multiply elements 2 and 0. */
|
5551 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (gen_lowpart (V2DImode, t1),
|
5552 |
|
|
op1, op2));
|
5553 |
|
|
|
5554 |
|
|
/* Shift both input vectors down one element, so that elements 3
|
5555 |
|
|
and 1 are now in the slots for elements 2 and 0. For K8, at
|
5556 |
|
|
least, this is faster than using a shuffle. */
|
5557 |
|
|
emit_insn (gen_sse2_lshrv1ti3 (gen_lowpart (V1TImode, t2),
|
5558 |
|
|
gen_lowpart (V1TImode, op1),
|
5559 |
|
|
thirtytwo));
|
5560 |
|
|
emit_insn (gen_sse2_lshrv1ti3 (gen_lowpart (V1TImode, t3),
|
5561 |
|
|
gen_lowpart (V1TImode, op2),
|
5562 |
|
|
thirtytwo));
|
5563 |
|
|
/* Multiply elements 3 and 1. */
|
5564 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (gen_lowpart (V2DImode, t4),
|
5565 |
|
|
t2, t3));
|
5566 |
|
|
|
5567 |
|
|
/* Move the results in element 2 down to element 1; we don't care
|
5568 |
|
|
what goes in elements 2 and 3. */
|
5569 |
|
|
emit_insn (gen_sse2_pshufd_1 (t5, t1, const0_rtx, const2_rtx,
|
5570 |
|
|
const0_rtx, const0_rtx));
|
5571 |
|
|
emit_insn (gen_sse2_pshufd_1 (t6, t4, const0_rtx, const2_rtx,
|
5572 |
|
|
const0_rtx, const0_rtx));
|
5573 |
|
|
|
5574 |
|
|
/* Merge the parts back together. */
|
5575 |
|
|
emit_insn (gen_vec_interleave_lowv4si (op0, t5, t6));
|
5576 |
|
|
DONE;
|
5577 |
|
|
})
|
5578 |
|
|
|
5579 |
|
|
(define_insn_and_split "mulv2di3"
|
5580 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "")
|
5581 |
|
|
(mult:V2DI (match_operand:V2DI 1 "register_operand" "")
|
5582 |
|
|
(match_operand:V2DI 2 "register_operand" "")))]
|
5583 |
|
|
"TARGET_SSE2
|
5584 |
|
|
&& can_create_pseudo_p ()"
|
5585 |
|
|
"#"
|
5586 |
|
|
"&& 1"
|
5587 |
|
|
[(const_int 0)]
|
5588 |
|
|
{
|
5589 |
|
|
rtx t1, t2, t3, t4, t5, t6, thirtytwo;
|
5590 |
|
|
rtx op0, op1, op2;
|
5591 |
|
|
|
5592 |
|
|
op0 = operands[0];
|
5593 |
|
|
op1 = operands[1];
|
5594 |
|
|
op2 = operands[2];
|
5595 |
|
|
|
5596 |
|
|
if (TARGET_XOP)
|
5597 |
|
|
{
|
5598 |
|
|
/* op1: A,B,C,D, op2: E,F,G,H */
|
5599 |
|
|
op1 = gen_lowpart (V4SImode, op1);
|
5600 |
|
|
op2 = gen_lowpart (V4SImode, op2);
|
5601 |
|
|
|
5602 |
|
|
t1 = gen_reg_rtx (V4SImode);
|
5603 |
|
|
t2 = gen_reg_rtx (V4SImode);
|
5604 |
|
|
t3 = gen_reg_rtx (V2DImode);
|
5605 |
|
|
t4 = gen_reg_rtx (V2DImode);
|
5606 |
|
|
|
5607 |
|
|
/* t1: B,A,D,C */
|
5608 |
|
|
emit_insn (gen_sse2_pshufd_1 (t1, op1,
|
5609 |
|
|
GEN_INT (1),
|
5610 |
|
|
GEN_INT (0),
|
5611 |
|
|
GEN_INT (3),
|
5612 |
|
|
GEN_INT (2)));
|
5613 |
|
|
|
5614 |
|
|
/* t2: (B*E),(A*F),(D*G),(C*H) */
|
5615 |
|
|
emit_insn (gen_mulv4si3 (t2, t1, op2));
|
5616 |
|
|
|
5617 |
|
|
/* t4: (B*E)+(A*F), (D*G)+(C*H) */
|
5618 |
|
|
emit_insn (gen_xop_phadddq (t3, t2));
|
5619 |
|
|
|
5620 |
|
|
/* t5: ((B*E)+(A*F))<<32, ((D*G)+(C*H))<<32 */
|
5621 |
|
|
emit_insn (gen_ashlv2di3 (t4, t3, GEN_INT (32)));
|
5622 |
|
|
|
5623 |
|
|
/* op0: (((B*E)+(A*F))<<32)+(B*F), (((D*G)+(C*H))<<32)+(D*H) */
|
5624 |
|
|
emit_insn (gen_xop_pmacsdql (op0, op1, op2, t4));
|
5625 |
|
|
}
|
5626 |
|
|
else
|
5627 |
|
|
{
|
5628 |
|
|
t1 = gen_reg_rtx (V2DImode);
|
5629 |
|
|
t2 = gen_reg_rtx (V2DImode);
|
5630 |
|
|
t3 = gen_reg_rtx (V2DImode);
|
5631 |
|
|
t4 = gen_reg_rtx (V2DImode);
|
5632 |
|
|
t5 = gen_reg_rtx (V2DImode);
|
5633 |
|
|
t6 = gen_reg_rtx (V2DImode);
|
5634 |
|
|
thirtytwo = GEN_INT (32);
|
5635 |
|
|
|
5636 |
|
|
/* Multiply low parts. */
|
5637 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (t1, gen_lowpart (V4SImode, op1),
|
5638 |
|
|
gen_lowpart (V4SImode, op2)));
|
5639 |
|
|
|
5640 |
|
|
/* Shift input vectors left 32 bits so we can multiply high parts. */
|
5641 |
|
|
emit_insn (gen_lshrv2di3 (t2, op1, thirtytwo));
|
5642 |
|
|
emit_insn (gen_lshrv2di3 (t3, op2, thirtytwo));
|
5643 |
|
|
|
5644 |
|
|
/* Multiply high parts by low parts. */
|
5645 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (t4, gen_lowpart (V4SImode, op1),
|
5646 |
|
|
gen_lowpart (V4SImode, t3)));
|
5647 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (t5, gen_lowpart (V4SImode, op2),
|
5648 |
|
|
gen_lowpart (V4SImode, t2)));
|
5649 |
|
|
|
5650 |
|
|
/* Shift them back. */
|
5651 |
|
|
emit_insn (gen_ashlv2di3 (t4, t4, thirtytwo));
|
5652 |
|
|
emit_insn (gen_ashlv2di3 (t5, t5, thirtytwo));
|
5653 |
|
|
|
5654 |
|
|
/* Add the three parts together. */
|
5655 |
|
|
emit_insn (gen_addv2di3 (t6, t1, t4));
|
5656 |
|
|
emit_insn (gen_addv2di3 (op0, t6, t5));
|
5657 |
|
|
}
|
5658 |
|
|
DONE;
|
5659 |
|
|
})
|
5660 |
|
|
|
5661 |
|
|
(define_expand "vec_widen_smult_hi_v8hi"
|
5662 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
5663 |
|
|
(match_operand:V8HI 1 "register_operand" "")
|
5664 |
|
|
(match_operand:V8HI 2 "register_operand" "")]
|
5665 |
|
|
"TARGET_SSE2"
|
5666 |
|
|
{
|
5667 |
|
|
rtx op1, op2, t1, t2, dest;
|
5668 |
|
|
|
5669 |
|
|
op1 = operands[1];
|
5670 |
|
|
op2 = operands[2];
|
5671 |
|
|
t1 = gen_reg_rtx (V8HImode);
|
5672 |
|
|
t2 = gen_reg_rtx (V8HImode);
|
5673 |
|
|
dest = gen_lowpart (V8HImode, operands[0]);
|
5674 |
|
|
|
5675 |
|
|
emit_insn (gen_mulv8hi3 (t1, op1, op2));
|
5676 |
|
|
emit_insn (gen_smulv8hi3_highpart (t2, op1, op2));
|
5677 |
|
|
emit_insn (gen_vec_interleave_highv8hi (dest, t1, t2));
|
5678 |
|
|
DONE;
|
5679 |
|
|
})
|
5680 |
|
|
|
5681 |
|
|
(define_expand "vec_widen_smult_lo_v8hi"
|
5682 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
5683 |
|
|
(match_operand:V8HI 1 "register_operand" "")
|
5684 |
|
|
(match_operand:V8HI 2 "register_operand" "")]
|
5685 |
|
|
"TARGET_SSE2"
|
5686 |
|
|
{
|
5687 |
|
|
rtx op1, op2, t1, t2, dest;
|
5688 |
|
|
|
5689 |
|
|
op1 = operands[1];
|
5690 |
|
|
op2 = operands[2];
|
5691 |
|
|
t1 = gen_reg_rtx (V8HImode);
|
5692 |
|
|
t2 = gen_reg_rtx (V8HImode);
|
5693 |
|
|
dest = gen_lowpart (V8HImode, operands[0]);
|
5694 |
|
|
|
5695 |
|
|
emit_insn (gen_mulv8hi3 (t1, op1, op2));
|
5696 |
|
|
emit_insn (gen_smulv8hi3_highpart (t2, op1, op2));
|
5697 |
|
|
emit_insn (gen_vec_interleave_lowv8hi (dest, t1, t2));
|
5698 |
|
|
DONE;
|
5699 |
|
|
})
|
5700 |
|
|
|
5701 |
|
|
(define_expand "vec_widen_umult_hi_v8hi"
|
5702 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
5703 |
|
|
(match_operand:V8HI 1 "register_operand" "")
|
5704 |
|
|
(match_operand:V8HI 2 "register_operand" "")]
|
5705 |
|
|
"TARGET_SSE2"
|
5706 |
|
|
{
|
5707 |
|
|
rtx op1, op2, t1, t2, dest;
|
5708 |
|
|
|
5709 |
|
|
op1 = operands[1];
|
5710 |
|
|
op2 = operands[2];
|
5711 |
|
|
t1 = gen_reg_rtx (V8HImode);
|
5712 |
|
|
t2 = gen_reg_rtx (V8HImode);
|
5713 |
|
|
dest = gen_lowpart (V8HImode, operands[0]);
|
5714 |
|
|
|
5715 |
|
|
emit_insn (gen_mulv8hi3 (t1, op1, op2));
|
5716 |
|
|
emit_insn (gen_umulv8hi3_highpart (t2, op1, op2));
|
5717 |
|
|
emit_insn (gen_vec_interleave_highv8hi (dest, t1, t2));
|
5718 |
|
|
DONE;
|
5719 |
|
|
})
|
5720 |
|
|
|
5721 |
|
|
(define_expand "vec_widen_umult_lo_v8hi"
|
5722 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
5723 |
|
|
(match_operand:V8HI 1 "register_operand" "")
|
5724 |
|
|
(match_operand:V8HI 2 "register_operand" "")]
|
5725 |
|
|
"TARGET_SSE2"
|
5726 |
|
|
{
|
5727 |
|
|
rtx op1, op2, t1, t2, dest;
|
5728 |
|
|
|
5729 |
|
|
op1 = operands[1];
|
5730 |
|
|
op2 = operands[2];
|
5731 |
|
|
t1 = gen_reg_rtx (V8HImode);
|
5732 |
|
|
t2 = gen_reg_rtx (V8HImode);
|
5733 |
|
|
dest = gen_lowpart (V8HImode, operands[0]);
|
5734 |
|
|
|
5735 |
|
|
emit_insn (gen_mulv8hi3 (t1, op1, op2));
|
5736 |
|
|
emit_insn (gen_umulv8hi3_highpart (t2, op1, op2));
|
5737 |
|
|
emit_insn (gen_vec_interleave_lowv8hi (dest, t1, t2));
|
5738 |
|
|
DONE;
|
5739 |
|
|
})
|
5740 |
|
|
|
5741 |
|
|
(define_expand "vec_widen_smult_hi_v4si"
|
5742 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
5743 |
|
|
(match_operand:V4SI 1 "register_operand" "")
|
5744 |
|
|
(match_operand:V4SI 2 "register_operand" "")]
|
5745 |
|
|
"TARGET_XOP"
|
5746 |
|
|
{
|
5747 |
|
|
rtx t1, t2;
|
5748 |
|
|
|
5749 |
|
|
t1 = gen_reg_rtx (V4SImode);
|
5750 |
|
|
t2 = gen_reg_rtx (V4SImode);
|
5751 |
|
|
|
5752 |
|
|
emit_insn (gen_sse2_pshufd_1 (t1, operands[1],
|
5753 |
|
|
GEN_INT (0),
|
5754 |
|
|
GEN_INT (2),
|
5755 |
|
|
GEN_INT (1),
|
5756 |
|
|
GEN_INT (3)));
|
5757 |
|
|
emit_insn (gen_sse2_pshufd_1 (t2, operands[2],
|
5758 |
|
|
GEN_INT (0),
|
5759 |
|
|
GEN_INT (2),
|
5760 |
|
|
GEN_INT (1),
|
5761 |
|
|
GEN_INT (3)));
|
5762 |
|
|
emit_insn (gen_xop_mulv2div2di3_high (operands[0], t1, t2));
|
5763 |
|
|
DONE;
|
5764 |
|
|
})
|
5765 |
|
|
|
5766 |
|
|
(define_expand "vec_widen_smult_lo_v4si"
|
5767 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
5768 |
|
|
(match_operand:V4SI 1 "register_operand" "")
|
5769 |
|
|
(match_operand:V4SI 2 "register_operand" "")]
|
5770 |
|
|
"TARGET_XOP"
|
5771 |
|
|
{
|
5772 |
|
|
rtx t1, t2;
|
5773 |
|
|
|
5774 |
|
|
t1 = gen_reg_rtx (V4SImode);
|
5775 |
|
|
t2 = gen_reg_rtx (V4SImode);
|
5776 |
|
|
|
5777 |
|
|
emit_insn (gen_sse2_pshufd_1 (t1, operands[1],
|
5778 |
|
|
GEN_INT (0),
|
5779 |
|
|
GEN_INT (2),
|
5780 |
|
|
GEN_INT (1),
|
5781 |
|
|
GEN_INT (3)));
|
5782 |
|
|
emit_insn (gen_sse2_pshufd_1 (t2, operands[2],
|
5783 |
|
|
GEN_INT (0),
|
5784 |
|
|
GEN_INT (2),
|
5785 |
|
|
GEN_INT (1),
|
5786 |
|
|
GEN_INT (3)));
|
5787 |
|
|
emit_insn (gen_xop_mulv2div2di3_low (operands[0], t1, t2));
|
5788 |
|
|
DONE;
|
5789 |
|
|
})
|
5790 |
|
|
|
5791 |
|
|
(define_expand "vec_widen_umult_hi_v4si"
|
5792 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
5793 |
|
|
(match_operand:V4SI 1 "register_operand" "")
|
5794 |
|
|
(match_operand:V4SI 2 "register_operand" "")]
|
5795 |
|
|
"TARGET_SSE2"
|
5796 |
|
|
{
|
5797 |
|
|
rtx op1, op2, t1, t2;
|
5798 |
|
|
|
5799 |
|
|
op1 = operands[1];
|
5800 |
|
|
op2 = operands[2];
|
5801 |
|
|
t1 = gen_reg_rtx (V4SImode);
|
5802 |
|
|
t2 = gen_reg_rtx (V4SImode);
|
5803 |
|
|
|
5804 |
|
|
emit_insn (gen_vec_interleave_highv4si (t1, op1, op1));
|
5805 |
|
|
emit_insn (gen_vec_interleave_highv4si (t2, op2, op2));
|
5806 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (operands[0], t1, t2));
|
5807 |
|
|
DONE;
|
5808 |
|
|
})
|
5809 |
|
|
|
5810 |
|
|
(define_expand "vec_widen_umult_lo_v4si"
|
5811 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
5812 |
|
|
(match_operand:V4SI 1 "register_operand" "")
|
5813 |
|
|
(match_operand:V4SI 2 "register_operand" "")]
|
5814 |
|
|
"TARGET_SSE2"
|
5815 |
|
|
{
|
5816 |
|
|
rtx op1, op2, t1, t2;
|
5817 |
|
|
|
5818 |
|
|
op1 = operands[1];
|
5819 |
|
|
op2 = operands[2];
|
5820 |
|
|
t1 = gen_reg_rtx (V4SImode);
|
5821 |
|
|
t2 = gen_reg_rtx (V4SImode);
|
5822 |
|
|
|
5823 |
|
|
emit_insn (gen_vec_interleave_lowv4si (t1, op1, op1));
|
5824 |
|
|
emit_insn (gen_vec_interleave_lowv4si (t2, op2, op2));
|
5825 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (operands[0], t1, t2));
|
5826 |
|
|
DONE;
|
5827 |
|
|
})
|
5828 |
|
|
|
5829 |
|
|
(define_expand "sdot_prodv8hi"
|
5830 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
5831 |
|
|
(match_operand:V8HI 1 "register_operand" "")
|
5832 |
|
|
(match_operand:V8HI 2 "register_operand" "")
|
5833 |
|
|
(match_operand:V4SI 3 "register_operand" "")]
|
5834 |
|
|
"TARGET_SSE2"
|
5835 |
|
|
{
|
5836 |
|
|
rtx t = gen_reg_rtx (V4SImode);
|
5837 |
|
|
emit_insn (gen_sse2_pmaddwd (t, operands[1], operands[2]));
|
5838 |
|
|
emit_insn (gen_addv4si3 (operands[0], operands[3], t));
|
5839 |
|
|
DONE;
|
5840 |
|
|
})
|
5841 |
|
|
|
5842 |
|
|
(define_expand "udot_prodv4si"
|
5843 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
5844 |
|
|
(match_operand:V4SI 1 "register_operand" "")
|
5845 |
|
|
(match_operand:V4SI 2 "register_operand" "")
|
5846 |
|
|
(match_operand:V2DI 3 "register_operand" "")]
|
5847 |
|
|
"TARGET_SSE2"
|
5848 |
|
|
{
|
5849 |
|
|
rtx t1, t2, t3, t4;
|
5850 |
|
|
|
5851 |
|
|
t1 = gen_reg_rtx (V2DImode);
|
5852 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (t1, operands[1], operands[2]));
|
5853 |
|
|
emit_insn (gen_addv2di3 (t1, t1, operands[3]));
|
5854 |
|
|
|
5855 |
|
|
t2 = gen_reg_rtx (V4SImode);
|
5856 |
|
|
t3 = gen_reg_rtx (V4SImode);
|
5857 |
|
|
emit_insn (gen_sse2_lshrv1ti3 (gen_lowpart (V1TImode, t2),
|
5858 |
|
|
gen_lowpart (V1TImode, operands[1]),
|
5859 |
|
|
GEN_INT (32)));
|
5860 |
|
|
emit_insn (gen_sse2_lshrv1ti3 (gen_lowpart (V1TImode, t3),
|
5861 |
|
|
gen_lowpart (V1TImode, operands[2]),
|
5862 |
|
|
GEN_INT (32)));
|
5863 |
|
|
|
5864 |
|
|
t4 = gen_reg_rtx (V2DImode);
|
5865 |
|
|
emit_insn (gen_sse2_umulv2siv2di3 (t4, t2, t3));
|
5866 |
|
|
|
5867 |
|
|
emit_insn (gen_addv2di3 (operands[0], t1, t4));
|
5868 |
|
|
DONE;
|
5869 |
|
|
})
|
5870 |
|
|
|
5871 |
|
|
(define_insn "*avx_ashr3"
|
5872 |
|
|
[(set (match_operand:SSEMODE24 0 "register_operand" "=x")
|
5873 |
|
|
(ashiftrt:SSEMODE24
|
5874 |
|
|
(match_operand:SSEMODE24 1 "register_operand" "x")
|
5875 |
|
|
(match_operand:SI 2 "nonmemory_operand" "xN")))]
|
5876 |
|
|
"TARGET_AVX"
|
5877 |
|
|
"vpsra\t{%2, %1, %0|%0, %1, %2}"
|
5878 |
|
|
[(set_attr "type" "sseishft")
|
5879 |
|
|
(set_attr "prefix" "vex")
|
5880 |
|
|
(set (attr "length_immediate")
|
5881 |
|
|
(if_then_else (match_operand 2 "const_int_operand" "")
|
5882 |
|
|
(const_string "1")
|
5883 |
|
|
(const_string "0")))
|
5884 |
|
|
(set_attr "mode" "TI")])
|
5885 |
|
|
|
5886 |
|
|
(define_insn "ashr3"
|
5887 |
|
|
[(set (match_operand:SSEMODE24 0 "register_operand" "=x")
|
5888 |
|
|
(ashiftrt:SSEMODE24
|
5889 |
|
|
(match_operand:SSEMODE24 1 "register_operand" "0")
|
5890 |
|
|
(match_operand:SI 2 "nonmemory_operand" "xN")))]
|
5891 |
|
|
"TARGET_SSE2"
|
5892 |
|
|
"psra\t{%2, %0|%0, %2}"
|
5893 |
|
|
[(set_attr "type" "sseishft")
|
5894 |
|
|
(set_attr "prefix_data16" "1")
|
5895 |
|
|
(set (attr "length_immediate")
|
5896 |
|
|
(if_then_else (match_operand 2 "const_int_operand" "")
|
5897 |
|
|
(const_string "1")
|
5898 |
|
|
(const_string "0")))
|
5899 |
|
|
(set_attr "mode" "TI")])
|
5900 |
|
|
|
5901 |
|
|
(define_insn "*avx_lshrv1ti3"
|
5902 |
|
|
[(set (match_operand:V1TI 0 "register_operand" "=x")
|
5903 |
|
|
(lshiftrt:V1TI
|
5904 |
|
|
(match_operand:V1TI 1 "register_operand" "x")
|
5905 |
|
|
(match_operand:SI 2 "const_0_to_255_mul_8_operand" "n")))]
|
5906 |
|
|
"TARGET_AVX"
|
5907 |
|
|
{
|
5908 |
|
|
operands[2] = GEN_INT (INTVAL (operands[2]) / 8);
|
5909 |
|
|
return "vpsrldq\t{%2, %1, %0|%0, %1, %2}";
|
5910 |
|
|
}
|
5911 |
|
|
[(set_attr "type" "sseishft")
|
5912 |
|
|
(set_attr "prefix" "vex")
|
5913 |
|
|
(set_attr "length_immediate" "1")
|
5914 |
|
|
(set_attr "mode" "TI")])
|
5915 |
|
|
|
5916 |
|
|
(define_insn "*avx_lshr3"
|
5917 |
|
|
[(set (match_operand:SSEMODE248 0 "register_operand" "=x")
|
5918 |
|
|
(lshiftrt:SSEMODE248
|
5919 |
|
|
(match_operand:SSEMODE248 1 "register_operand" "x")
|
5920 |
|
|
(match_operand:SI 2 "nonmemory_operand" "xN")))]
|
5921 |
|
|
"TARGET_AVX"
|
5922 |
|
|
"vpsrl\t{%2, %1, %0|%0, %1, %2}"
|
5923 |
|
|
[(set_attr "type" "sseishft")
|
5924 |
|
|
(set_attr "prefix" "vex")
|
5925 |
|
|
(set (attr "length_immediate")
|
5926 |
|
|
(if_then_else (match_operand 2 "const_int_operand" "")
|
5927 |
|
|
(const_string "1")
|
5928 |
|
|
(const_string "0")))
|
5929 |
|
|
(set_attr "mode" "TI")])
|
5930 |
|
|
|
5931 |
|
|
(define_insn "sse2_lshrv1ti3"
|
5932 |
|
|
[(set (match_operand:V1TI 0 "register_operand" "=x")
|
5933 |
|
|
(lshiftrt:V1TI
|
5934 |
|
|
(match_operand:V1TI 1 "register_operand" "0")
|
5935 |
|
|
(match_operand:SI 2 "const_0_to_255_mul_8_operand" "n")))]
|
5936 |
|
|
"TARGET_SSE2"
|
5937 |
|
|
{
|
5938 |
|
|
operands[2] = GEN_INT (INTVAL (operands[2]) / 8);
|
5939 |
|
|
return "psrldq\t{%2, %0|%0, %2}";
|
5940 |
|
|
}
|
5941 |
|
|
[(set_attr "type" "sseishft")
|
5942 |
|
|
(set_attr "prefix_data16" "1")
|
5943 |
|
|
(set_attr "length_immediate" "1")
|
5944 |
|
|
(set_attr "atom_unit" "sishuf")
|
5945 |
|
|
(set_attr "mode" "TI")])
|
5946 |
|
|
|
5947 |
|
|
(define_insn "lshr3"
|
5948 |
|
|
[(set (match_operand:SSEMODE248 0 "register_operand" "=x")
|
5949 |
|
|
(lshiftrt:SSEMODE248
|
5950 |
|
|
(match_operand:SSEMODE248 1 "register_operand" "0")
|
5951 |
|
|
(match_operand:SI 2 "nonmemory_operand" "xN")))]
|
5952 |
|
|
"TARGET_SSE2"
|
5953 |
|
|
"psrl\t{%2, %0|%0, %2}"
|
5954 |
|
|
[(set_attr "type" "sseishft")
|
5955 |
|
|
(set_attr "prefix_data16" "1")
|
5956 |
|
|
(set (attr "length_immediate")
|
5957 |
|
|
(if_then_else (match_operand 2 "const_int_operand" "")
|
5958 |
|
|
(const_string "1")
|
5959 |
|
|
(const_string "0")))
|
5960 |
|
|
(set_attr "mode" "TI")])
|
5961 |
|
|
|
5962 |
|
|
(define_insn "*avx_ashlv1ti3"
|
5963 |
|
|
[(set (match_operand:V1TI 0 "register_operand" "=x")
|
5964 |
|
|
(ashift:V1TI (match_operand:V1TI 1 "register_operand" "x")
|
5965 |
|
|
(match_operand:SI 2 "const_0_to_255_mul_8_operand" "n")))]
|
5966 |
|
|
"TARGET_AVX"
|
5967 |
|
|
{
|
5968 |
|
|
operands[2] = GEN_INT (INTVAL (operands[2]) / 8);
|
5969 |
|
|
return "vpslldq\t{%2, %1, %0|%0, %1, %2}";
|
5970 |
|
|
}
|
5971 |
|
|
[(set_attr "type" "sseishft")
|
5972 |
|
|
(set_attr "prefix" "vex")
|
5973 |
|
|
(set_attr "length_immediate" "1")
|
5974 |
|
|
(set_attr "mode" "TI")])
|
5975 |
|
|
|
5976 |
|
|
(define_insn "*avx_ashl3"
|
5977 |
|
|
[(set (match_operand:SSEMODE248 0 "register_operand" "=x")
|
5978 |
|
|
(ashift:SSEMODE248
|
5979 |
|
|
(match_operand:SSEMODE248 1 "register_operand" "x")
|
5980 |
|
|
(match_operand:SI 2 "nonmemory_operand" "xN")))]
|
5981 |
|
|
"TARGET_AVX"
|
5982 |
|
|
"vpsll\t{%2, %1, %0|%0, %1, %2}"
|
5983 |
|
|
[(set_attr "type" "sseishft")
|
5984 |
|
|
(set_attr "prefix" "vex")
|
5985 |
|
|
(set (attr "length_immediate")
|
5986 |
|
|
(if_then_else (match_operand 2 "const_int_operand" "")
|
5987 |
|
|
(const_string "1")
|
5988 |
|
|
(const_string "0")))
|
5989 |
|
|
(set_attr "mode" "TI")])
|
5990 |
|
|
|
5991 |
|
|
(define_insn "sse2_ashlv1ti3"
|
5992 |
|
|
[(set (match_operand:V1TI 0 "register_operand" "=x")
|
5993 |
|
|
(ashift:V1TI (match_operand:V1TI 1 "register_operand" "0")
|
5994 |
|
|
(match_operand:SI 2 "const_0_to_255_mul_8_operand" "n")))]
|
5995 |
|
|
"TARGET_SSE2"
|
5996 |
|
|
{
|
5997 |
|
|
operands[2] = GEN_INT (INTVAL (operands[2]) / 8);
|
5998 |
|
|
return "pslldq\t{%2, %0|%0, %2}";
|
5999 |
|
|
}
|
6000 |
|
|
[(set_attr "type" "sseishft")
|
6001 |
|
|
(set_attr "prefix_data16" "1")
|
6002 |
|
|
(set_attr "length_immediate" "1")
|
6003 |
|
|
(set_attr "mode" "TI")])
|
6004 |
|
|
|
6005 |
|
|
(define_insn "ashl3"
|
6006 |
|
|
[(set (match_operand:SSEMODE248 0 "register_operand" "=x")
|
6007 |
|
|
(ashift:SSEMODE248
|
6008 |
|
|
(match_operand:SSEMODE248 1 "register_operand" "0")
|
6009 |
|
|
(match_operand:SI 2 "nonmemory_operand" "xN")))]
|
6010 |
|
|
"TARGET_SSE2"
|
6011 |
|
|
"psll\t{%2, %0|%0, %2}"
|
6012 |
|
|
[(set_attr "type" "sseishft")
|
6013 |
|
|
(set_attr "prefix_data16" "1")
|
6014 |
|
|
(set (attr "length_immediate")
|
6015 |
|
|
(if_then_else (match_operand 2 "const_int_operand" "")
|
6016 |
|
|
(const_string "1")
|
6017 |
|
|
(const_string "0")))
|
6018 |
|
|
(set_attr "mode" "TI")])
|
6019 |
|
|
|
6020 |
|
|
(define_expand "vec_shl_"
|
6021 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "")
|
6022 |
|
|
(ashift:V1TI
|
6023 |
|
|
(match_operand:SSEMODEI 1 "register_operand" "")
|
6024 |
|
|
(match_operand:SI 2 "const_0_to_255_mul_8_operand" "")))]
|
6025 |
|
|
"TARGET_SSE2"
|
6026 |
|
|
{
|
6027 |
|
|
operands[0] = gen_lowpart (V1TImode, operands[0]);
|
6028 |
|
|
operands[1] = gen_lowpart (V1TImode, operands[1]);
|
6029 |
|
|
})
|
6030 |
|
|
|
6031 |
|
|
(define_expand "vec_shr_"
|
6032 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "")
|
6033 |
|
|
(lshiftrt:V1TI
|
6034 |
|
|
(match_operand:SSEMODEI 1 "register_operand" "")
|
6035 |
|
|
(match_operand:SI 2 "const_0_to_255_mul_8_operand" "")))]
|
6036 |
|
|
"TARGET_SSE2"
|
6037 |
|
|
{
|
6038 |
|
|
operands[0] = gen_lowpart (V1TImode, operands[0]);
|
6039 |
|
|
operands[1] = gen_lowpart (V1TImode, operands[1]);
|
6040 |
|
|
})
|
6041 |
|
|
|
6042 |
|
|
(define_insn "*avx_3"
|
6043 |
|
|
[(set (match_operand:SSEMODE124 0 "register_operand" "=x")
|
6044 |
|
|
(maxmin:SSEMODE124
|
6045 |
|
|
(match_operand:SSEMODE124 1 "nonimmediate_operand" "%x")
|
6046 |
|
|
(match_operand:SSEMODE124 2 "nonimmediate_operand" "xm")))]
|
6047 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (, mode, operands)"
|
6048 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
6049 |
|
|
[(set_attr "type" "sseiadd")
|
6050 |
|
|
(set (attr "prefix_extra")
|
6051 |
|
|
(if_then_else
|
6052 |
|
|
(ne (symbol_ref "mode != (( == SMAX || == SMIN) ? V8HImode : V16QImode)")
|
6053 |
|
|
(const_int 0))
|
6054 |
|
|
(const_string "1")
|
6055 |
|
|
(const_string "0")))
|
6056 |
|
|
(set_attr "prefix" "vex")
|
6057 |
|
|
(set_attr "mode" "TI")])
|
6058 |
|
|
|
6059 |
|
|
(define_expand "v16qi3"
|
6060 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "")
|
6061 |
|
|
(umaxmin:V16QI
|
6062 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "")
|
6063 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "")))]
|
6064 |
|
|
"TARGET_SSE2"
|
6065 |
|
|
"ix86_fixup_binary_operands_no_copy (, V16QImode, operands);")
|
6066 |
|
|
|
6067 |
|
|
(define_insn "*v16qi3"
|
6068 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6069 |
|
|
(umaxmin:V16QI
|
6070 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "%0")
|
6071 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")))]
|
6072 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (, V16QImode, operands)"
|
6073 |
|
|
"pb\t{%2, %0|%0, %2}"
|
6074 |
|
|
[(set_attr "type" "sseiadd")
|
6075 |
|
|
(set_attr "prefix_data16" "1")
|
6076 |
|
|
(set_attr "mode" "TI")])
|
6077 |
|
|
|
6078 |
|
|
(define_expand "v8hi3"
|
6079 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "")
|
6080 |
|
|
(smaxmin:V8HI
|
6081 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "")
|
6082 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "")))]
|
6083 |
|
|
"TARGET_SSE2"
|
6084 |
|
|
"ix86_fixup_binary_operands_no_copy (, V8HImode, operands);")
|
6085 |
|
|
|
6086 |
|
|
(define_insn "*v8hi3"
|
6087 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
6088 |
|
|
(smaxmin:V8HI
|
6089 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%0")
|
6090 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))]
|
6091 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (, V8HImode, operands)"
|
6092 |
|
|
"pw\t{%2, %0|%0, %2}"
|
6093 |
|
|
[(set_attr "type" "sseiadd")
|
6094 |
|
|
(set_attr "prefix_data16" "1")
|
6095 |
|
|
(set_attr "mode" "TI")])
|
6096 |
|
|
|
6097 |
|
|
(define_expand "umaxv8hi3"
|
6098 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "")
|
6099 |
|
|
(umax:V8HI (match_operand:V8HI 1 "register_operand" "")
|
6100 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "")))]
|
6101 |
|
|
"TARGET_SSE2"
|
6102 |
|
|
{
|
6103 |
|
|
if (TARGET_SSE4_1)
|
6104 |
|
|
ix86_fixup_binary_operands_no_copy (UMAX, V8HImode, operands);
|
6105 |
|
|
else
|
6106 |
|
|
{
|
6107 |
|
|
rtx op0 = operands[0], op2 = operands[2], op3 = op0;
|
6108 |
|
|
if (rtx_equal_p (op3, op2))
|
6109 |
|
|
op3 = gen_reg_rtx (V8HImode);
|
6110 |
|
|
emit_insn (gen_sse2_ussubv8hi3 (op3, operands[1], op2));
|
6111 |
|
|
emit_insn (gen_addv8hi3 (op0, op3, op2));
|
6112 |
|
|
DONE;
|
6113 |
|
|
}
|
6114 |
|
|
})
|
6115 |
|
|
|
6116 |
|
|
(define_expand "smax3"
|
6117 |
|
|
[(set (match_operand:SSEMODE14 0 "register_operand" "")
|
6118 |
|
|
(smax:SSEMODE14 (match_operand:SSEMODE14 1 "register_operand" "")
|
6119 |
|
|
(match_operand:SSEMODE14 2 "register_operand" "")))]
|
6120 |
|
|
"TARGET_SSE2"
|
6121 |
|
|
{
|
6122 |
|
|
if (TARGET_SSE4_1)
|
6123 |
|
|
ix86_fixup_binary_operands_no_copy (SMAX, mode, operands);
|
6124 |
|
|
else
|
6125 |
|
|
{
|
6126 |
|
|
rtx xops[6];
|
6127 |
|
|
bool ok;
|
6128 |
|
|
|
6129 |
|
|
xops[0] = operands[0];
|
6130 |
|
|
xops[1] = operands[1];
|
6131 |
|
|
xops[2] = operands[2];
|
6132 |
|
|
xops[3] = gen_rtx_GT (VOIDmode, operands[1], operands[2]);
|
6133 |
|
|
xops[4] = operands[1];
|
6134 |
|
|
xops[5] = operands[2];
|
6135 |
|
|
ok = ix86_expand_int_vcond (xops);
|
6136 |
|
|
gcc_assert (ok);
|
6137 |
|
|
DONE;
|
6138 |
|
|
}
|
6139 |
|
|
})
|
6140 |
|
|
|
6141 |
|
|
(define_insn "*sse4_1_3"
|
6142 |
|
|
[(set (match_operand:SSEMODE14 0 "register_operand" "=x")
|
6143 |
|
|
(smaxmin:SSEMODE14
|
6144 |
|
|
(match_operand:SSEMODE14 1 "nonimmediate_operand" "%0")
|
6145 |
|
|
(match_operand:SSEMODE14 2 "nonimmediate_operand" "xm")))]
|
6146 |
|
|
"TARGET_SSE4_1 && ix86_binary_operator_ok (, mode, operands)"
|
6147 |
|
|
"p\t{%2, %0|%0, %2}"
|
6148 |
|
|
[(set_attr "type" "sseiadd")
|
6149 |
|
|
(set_attr "prefix_extra" "1")
|
6150 |
|
|
(set_attr "mode" "TI")])
|
6151 |
|
|
|
6152 |
|
|
(define_expand "smaxv2di3"
|
6153 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "")
|
6154 |
|
|
(smax:V2DI (match_operand:V2DI 1 "register_operand" "")
|
6155 |
|
|
(match_operand:V2DI 2 "register_operand" "")))]
|
6156 |
|
|
"TARGET_SSE4_2"
|
6157 |
|
|
{
|
6158 |
|
|
rtx xops[6];
|
6159 |
|
|
bool ok;
|
6160 |
|
|
|
6161 |
|
|
xops[0] = operands[0];
|
6162 |
|
|
xops[1] = operands[1];
|
6163 |
|
|
xops[2] = operands[2];
|
6164 |
|
|
xops[3] = gen_rtx_GT (VOIDmode, operands[1], operands[2]);
|
6165 |
|
|
xops[4] = operands[1];
|
6166 |
|
|
xops[5] = operands[2];
|
6167 |
|
|
ok = ix86_expand_int_vcond (xops);
|
6168 |
|
|
gcc_assert (ok);
|
6169 |
|
|
DONE;
|
6170 |
|
|
})
|
6171 |
|
|
|
6172 |
|
|
(define_expand "umaxv4si3"
|
6173 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "")
|
6174 |
|
|
(umax:V4SI (match_operand:V4SI 1 "register_operand" "")
|
6175 |
|
|
(match_operand:V4SI 2 "register_operand" "")))]
|
6176 |
|
|
"TARGET_SSE2"
|
6177 |
|
|
{
|
6178 |
|
|
if (TARGET_SSE4_1)
|
6179 |
|
|
ix86_fixup_binary_operands_no_copy (UMAX, V4SImode, operands);
|
6180 |
|
|
else
|
6181 |
|
|
{
|
6182 |
|
|
rtx xops[6];
|
6183 |
|
|
bool ok;
|
6184 |
|
|
|
6185 |
|
|
xops[0] = operands[0];
|
6186 |
|
|
xops[1] = operands[1];
|
6187 |
|
|
xops[2] = operands[2];
|
6188 |
|
|
xops[3] = gen_rtx_GTU (VOIDmode, operands[1], operands[2]);
|
6189 |
|
|
xops[4] = operands[1];
|
6190 |
|
|
xops[5] = operands[2];
|
6191 |
|
|
ok = ix86_expand_int_vcond (xops);
|
6192 |
|
|
gcc_assert (ok);
|
6193 |
|
|
DONE;
|
6194 |
|
|
}
|
6195 |
|
|
})
|
6196 |
|
|
|
6197 |
|
|
(define_insn "*sse4_1_3"
|
6198 |
|
|
[(set (match_operand:SSEMODE24 0 "register_operand" "=x")
|
6199 |
|
|
(umaxmin:SSEMODE24
|
6200 |
|
|
(match_operand:SSEMODE24 1 "nonimmediate_operand" "%0")
|
6201 |
|
|
(match_operand:SSEMODE24 2 "nonimmediate_operand" "xm")))]
|
6202 |
|
|
"TARGET_SSE4_1 && ix86_binary_operator_ok (, mode, operands)"
|
6203 |
|
|
"p\t{%2, %0|%0, %2}"
|
6204 |
|
|
[(set_attr "type" "sseiadd")
|
6205 |
|
|
(set_attr "prefix_extra" "1")
|
6206 |
|
|
(set_attr "mode" "TI")])
|
6207 |
|
|
|
6208 |
|
|
(define_expand "umaxv2di3"
|
6209 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "")
|
6210 |
|
|
(umax:V2DI (match_operand:V2DI 1 "register_operand" "")
|
6211 |
|
|
(match_operand:V2DI 2 "register_operand" "")))]
|
6212 |
|
|
"TARGET_SSE4_2"
|
6213 |
|
|
{
|
6214 |
|
|
rtx xops[6];
|
6215 |
|
|
bool ok;
|
6216 |
|
|
|
6217 |
|
|
xops[0] = operands[0];
|
6218 |
|
|
xops[1] = operands[1];
|
6219 |
|
|
xops[2] = operands[2];
|
6220 |
|
|
xops[3] = gen_rtx_GTU (VOIDmode, operands[1], operands[2]);
|
6221 |
|
|
xops[4] = operands[1];
|
6222 |
|
|
xops[5] = operands[2];
|
6223 |
|
|
ok = ix86_expand_int_vcond (xops);
|
6224 |
|
|
gcc_assert (ok);
|
6225 |
|
|
DONE;
|
6226 |
|
|
})
|
6227 |
|
|
|
6228 |
|
|
(define_expand "smin3"
|
6229 |
|
|
[(set (match_operand:SSEMODE14 0 "register_operand" "")
|
6230 |
|
|
(smin:SSEMODE14 (match_operand:SSEMODE14 1 "register_operand" "")
|
6231 |
|
|
(match_operand:SSEMODE14 2 "register_operand" "")))]
|
6232 |
|
|
"TARGET_SSE2"
|
6233 |
|
|
{
|
6234 |
|
|
if (TARGET_SSE4_1)
|
6235 |
|
|
ix86_fixup_binary_operands_no_copy (SMIN, mode, operands);
|
6236 |
|
|
else
|
6237 |
|
|
{
|
6238 |
|
|
rtx xops[6];
|
6239 |
|
|
bool ok;
|
6240 |
|
|
|
6241 |
|
|
xops[0] = operands[0];
|
6242 |
|
|
xops[1] = operands[2];
|
6243 |
|
|
xops[2] = operands[1];
|
6244 |
|
|
xops[3] = gen_rtx_GT (VOIDmode, operands[1], operands[2]);
|
6245 |
|
|
xops[4] = operands[1];
|
6246 |
|
|
xops[5] = operands[2];
|
6247 |
|
|
ok = ix86_expand_int_vcond (xops);
|
6248 |
|
|
gcc_assert (ok);
|
6249 |
|
|
DONE;
|
6250 |
|
|
}
|
6251 |
|
|
})
|
6252 |
|
|
|
6253 |
|
|
(define_expand "sminv2di3"
|
6254 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "")
|
6255 |
|
|
(smin:V2DI (match_operand:V2DI 1 "register_operand" "")
|
6256 |
|
|
(match_operand:V2DI 2 "register_operand" "")))]
|
6257 |
|
|
"TARGET_SSE4_2"
|
6258 |
|
|
{
|
6259 |
|
|
rtx xops[6];
|
6260 |
|
|
bool ok;
|
6261 |
|
|
|
6262 |
|
|
xops[0] = operands[0];
|
6263 |
|
|
xops[1] = operands[2];
|
6264 |
|
|
xops[2] = operands[1];
|
6265 |
|
|
xops[3] = gen_rtx_GT (VOIDmode, operands[1], operands[2]);
|
6266 |
|
|
xops[4] = operands[1];
|
6267 |
|
|
xops[5] = operands[2];
|
6268 |
|
|
ok = ix86_expand_int_vcond (xops);
|
6269 |
|
|
gcc_assert (ok);
|
6270 |
|
|
DONE;
|
6271 |
|
|
})
|
6272 |
|
|
|
6273 |
|
|
(define_expand "umin3"
|
6274 |
|
|
[(set (match_operand:SSEMODE24 0 "register_operand" "")
|
6275 |
|
|
(umin:SSEMODE24 (match_operand:SSEMODE24 1 "register_operand" "")
|
6276 |
|
|
(match_operand:SSEMODE24 2 "register_operand" "")))]
|
6277 |
|
|
"TARGET_SSE2"
|
6278 |
|
|
{
|
6279 |
|
|
if (TARGET_SSE4_1)
|
6280 |
|
|
ix86_fixup_binary_operands_no_copy (UMIN, mode, operands);
|
6281 |
|
|
else
|
6282 |
|
|
{
|
6283 |
|
|
rtx xops[6];
|
6284 |
|
|
bool ok;
|
6285 |
|
|
|
6286 |
|
|
xops[0] = operands[0];
|
6287 |
|
|
xops[1] = operands[2];
|
6288 |
|
|
xops[2] = operands[1];
|
6289 |
|
|
xops[3] = gen_rtx_GTU (VOIDmode, operands[1], operands[2]);
|
6290 |
|
|
xops[4] = operands[1];
|
6291 |
|
|
xops[5] = operands[2];
|
6292 |
|
|
ok = ix86_expand_int_vcond (xops);
|
6293 |
|
|
gcc_assert (ok);
|
6294 |
|
|
DONE;
|
6295 |
|
|
}
|
6296 |
|
|
})
|
6297 |
|
|
|
6298 |
|
|
(define_expand "uminv2di3"
|
6299 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "")
|
6300 |
|
|
(umin:V2DI (match_operand:V2DI 1 "register_operand" "")
|
6301 |
|
|
(match_operand:V2DI 2 "register_operand" "")))]
|
6302 |
|
|
"TARGET_SSE4_2"
|
6303 |
|
|
{
|
6304 |
|
|
rtx xops[6];
|
6305 |
|
|
bool ok;
|
6306 |
|
|
|
6307 |
|
|
xops[0] = operands[0];
|
6308 |
|
|
xops[1] = operands[2];
|
6309 |
|
|
xops[2] = operands[1];
|
6310 |
|
|
xops[3] = gen_rtx_GTU (VOIDmode, operands[1], operands[2]);
|
6311 |
|
|
xops[4] = operands[1];
|
6312 |
|
|
xops[5] = operands[2];
|
6313 |
|
|
ok = ix86_expand_int_vcond (xops);
|
6314 |
|
|
gcc_assert (ok);
|
6315 |
|
|
DONE;
|
6316 |
|
|
})
|
6317 |
|
|
|
6318 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
6319 |
|
|
;;
|
6320 |
|
|
;; Parallel integral comparisons
|
6321 |
|
|
;;
|
6322 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
6323 |
|
|
|
6324 |
|
|
(define_expand "sse2_eq3"
|
6325 |
|
|
[(set (match_operand:SSEMODE124 0 "register_operand" "")
|
6326 |
|
|
(eq:SSEMODE124
|
6327 |
|
|
(match_operand:SSEMODE124 1 "nonimmediate_operand" "")
|
6328 |
|
|
(match_operand:SSEMODE124 2 "nonimmediate_operand" "")))]
|
6329 |
|
|
"TARGET_SSE2 && !TARGET_XOP "
|
6330 |
|
|
"ix86_fixup_binary_operands_no_copy (EQ, mode, operands);")
|
6331 |
|
|
|
6332 |
|
|
(define_insn "*avx_eq3"
|
6333 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x")
|
6334 |
|
|
(eq:SSEMODE1248
|
6335 |
|
|
(match_operand:SSEMODE1248 1 "nonimmediate_operand" "%x")
|
6336 |
|
|
(match_operand:SSEMODE1248 2 "nonimmediate_operand" "xm")))]
|
6337 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (EQ, mode, operands)"
|
6338 |
|
|
"vpcmpeq\t{%2, %1, %0|%0, %1, %2}"
|
6339 |
|
|
[(set_attr "type" "ssecmp")
|
6340 |
|
|
(set (attr "prefix_extra")
|
6341 |
|
|
(if_then_else (match_operand:V2DI 0 "" "")
|
6342 |
|
|
(const_string "1")
|
6343 |
|
|
(const_string "*")))
|
6344 |
|
|
(set_attr "prefix" "vex")
|
6345 |
|
|
(set_attr "mode" "TI")])
|
6346 |
|
|
|
6347 |
|
|
(define_insn "*sse2_eq3"
|
6348 |
|
|
[(set (match_operand:SSEMODE124 0 "register_operand" "=x")
|
6349 |
|
|
(eq:SSEMODE124
|
6350 |
|
|
(match_operand:SSEMODE124 1 "nonimmediate_operand" "%0")
|
6351 |
|
|
(match_operand:SSEMODE124 2 "nonimmediate_operand" "xm")))]
|
6352 |
|
|
"TARGET_SSE2 && !TARGET_XOP
|
6353 |
|
|
&& ix86_binary_operator_ok (EQ, mode, operands)"
|
6354 |
|
|
"pcmpeq\t{%2, %0|%0, %2}"
|
6355 |
|
|
[(set_attr "type" "ssecmp")
|
6356 |
|
|
(set_attr "prefix_data16" "1")
|
6357 |
|
|
(set_attr "mode" "TI")])
|
6358 |
|
|
|
6359 |
|
|
(define_expand "sse4_1_eqv2di3"
|
6360 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "")
|
6361 |
|
|
(eq:V2DI
|
6362 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "")
|
6363 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "")))]
|
6364 |
|
|
"TARGET_SSE4_1"
|
6365 |
|
|
"ix86_fixup_binary_operands_no_copy (EQ, V2DImode, operands);")
|
6366 |
|
|
|
6367 |
|
|
(define_insn "*sse4_1_eqv2di3"
|
6368 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
6369 |
|
|
(eq:V2DI
|
6370 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "%0")
|
6371 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")))]
|
6372 |
|
|
"TARGET_SSE4_1 && ix86_binary_operator_ok (EQ, V2DImode, operands)"
|
6373 |
|
|
"pcmpeqq\t{%2, %0|%0, %2}"
|
6374 |
|
|
[(set_attr "type" "ssecmp")
|
6375 |
|
|
(set_attr "prefix_extra" "1")
|
6376 |
|
|
(set_attr "mode" "TI")])
|
6377 |
|
|
|
6378 |
|
|
(define_insn "*avx_gt3"
|
6379 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x")
|
6380 |
|
|
(gt:SSEMODE1248
|
6381 |
|
|
(match_operand:SSEMODE1248 1 "register_operand" "x")
|
6382 |
|
|
(match_operand:SSEMODE1248 2 "nonimmediate_operand" "xm")))]
|
6383 |
|
|
"TARGET_AVX"
|
6384 |
|
|
"vpcmpgt\t{%2, %1, %0|%0, %1, %2}"
|
6385 |
|
|
[(set_attr "type" "ssecmp")
|
6386 |
|
|
(set (attr "prefix_extra")
|
6387 |
|
|
(if_then_else (match_operand:V2DI 0 "" "")
|
6388 |
|
|
(const_string "1")
|
6389 |
|
|
(const_string "*")))
|
6390 |
|
|
(set_attr "prefix" "vex")
|
6391 |
|
|
(set_attr "mode" "TI")])
|
6392 |
|
|
|
6393 |
|
|
(define_insn "sse2_gt3"
|
6394 |
|
|
[(set (match_operand:SSEMODE124 0 "register_operand" "=x")
|
6395 |
|
|
(gt:SSEMODE124
|
6396 |
|
|
(match_operand:SSEMODE124 1 "register_operand" "0")
|
6397 |
|
|
(match_operand:SSEMODE124 2 "nonimmediate_operand" "xm")))]
|
6398 |
|
|
"TARGET_SSE2 && !TARGET_XOP"
|
6399 |
|
|
"pcmpgt\t{%2, %0|%0, %2}"
|
6400 |
|
|
[(set_attr "type" "ssecmp")
|
6401 |
|
|
(set_attr "prefix_data16" "1")
|
6402 |
|
|
(set_attr "mode" "TI")])
|
6403 |
|
|
|
6404 |
|
|
(define_insn "sse4_2_gtv2di3"
|
6405 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
6406 |
|
|
(gt:V2DI
|
6407 |
|
|
(match_operand:V2DI 1 "register_operand" "0")
|
6408 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")))]
|
6409 |
|
|
"TARGET_SSE4_2"
|
6410 |
|
|
"pcmpgtq\t{%2, %0|%0, %2}"
|
6411 |
|
|
[(set_attr "type" "ssecmp")
|
6412 |
|
|
(set_attr "prefix_extra" "1")
|
6413 |
|
|
(set_attr "mode" "TI")])
|
6414 |
|
|
|
6415 |
|
|
(define_expand "vcond"
|
6416 |
|
|
[(set (match_operand:SSEMODE124C8 0 "register_operand" "")
|
6417 |
|
|
(if_then_else:SSEMODE124C8
|
6418 |
|
|
(match_operator 3 ""
|
6419 |
|
|
[(match_operand:SSEMODE124C8 4 "nonimmediate_operand" "")
|
6420 |
|
|
(match_operand:SSEMODE124C8 5 "nonimmediate_operand" "")])
|
6421 |
|
|
(match_operand:SSEMODE124C8 1 "general_operand" "")
|
6422 |
|
|
(match_operand:SSEMODE124C8 2 "general_operand" "")))]
|
6423 |
|
|
"TARGET_SSE2"
|
6424 |
|
|
{
|
6425 |
|
|
bool ok = ix86_expand_int_vcond (operands);
|
6426 |
|
|
gcc_assert (ok);
|
6427 |
|
|
DONE;
|
6428 |
|
|
})
|
6429 |
|
|
|
6430 |
|
|
(define_expand "vcondu"
|
6431 |
|
|
[(set (match_operand:SSEMODE124C8 0 "register_operand" "")
|
6432 |
|
|
(if_then_else:SSEMODE124C8
|
6433 |
|
|
(match_operator 3 ""
|
6434 |
|
|
[(match_operand:SSEMODE124C8 4 "nonimmediate_operand" "")
|
6435 |
|
|
(match_operand:SSEMODE124C8 5 "nonimmediate_operand" "")])
|
6436 |
|
|
(match_operand:SSEMODE124C8 1 "general_operand" "")
|
6437 |
|
|
(match_operand:SSEMODE124C8 2 "general_operand" "")))]
|
6438 |
|
|
"TARGET_SSE2"
|
6439 |
|
|
{
|
6440 |
|
|
bool ok = ix86_expand_int_vcond (operands);
|
6441 |
|
|
gcc_assert (ok);
|
6442 |
|
|
DONE;
|
6443 |
|
|
})
|
6444 |
|
|
|
6445 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
6446 |
|
|
;;
|
6447 |
|
|
;; Parallel bitwise logical operations
|
6448 |
|
|
;;
|
6449 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
6450 |
|
|
|
6451 |
|
|
(define_expand "one_cmpl2"
|
6452 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "")
|
6453 |
|
|
(xor:SSEMODEI (match_operand:SSEMODEI 1 "nonimmediate_operand" "")
|
6454 |
|
|
(match_dup 2)))]
|
6455 |
|
|
"TARGET_SSE2"
|
6456 |
|
|
{
|
6457 |
|
|
int i, n = GET_MODE_NUNITS (mode);
|
6458 |
|
|
rtvec v = rtvec_alloc (n);
|
6459 |
|
|
|
6460 |
|
|
for (i = 0; i < n; ++i)
|
6461 |
|
|
RTVEC_ELT (v, i) = constm1_rtx;
|
6462 |
|
|
|
6463 |
|
|
operands[2] = force_reg (mode, gen_rtx_CONST_VECTOR (mode, v));
|
6464 |
|
|
})
|
6465 |
|
|
|
6466 |
|
|
(define_insn "*avx_andnot3"
|
6467 |
|
|
[(set (match_operand:AVX256MODEI 0 "register_operand" "=x")
|
6468 |
|
|
(and:AVX256MODEI
|
6469 |
|
|
(not:AVX256MODEI (match_operand:AVX256MODEI 1 "register_operand" "x"))
|
6470 |
|
|
(match_operand:AVX256MODEI 2 "nonimmediate_operand" "xm")))]
|
6471 |
|
|
"TARGET_AVX"
|
6472 |
|
|
"vandnps\t{%2, %1, %0|%0, %1, %2}"
|
6473 |
|
|
[(set_attr "type" "sselog")
|
6474 |
|
|
(set_attr "prefix" "vex")
|
6475 |
|
|
(set_attr "mode" "")])
|
6476 |
|
|
|
6477 |
|
|
(define_insn "*sse_andnot3"
|
6478 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "=x")
|
6479 |
|
|
(and:SSEMODEI
|
6480 |
|
|
(not:SSEMODEI (match_operand:SSEMODEI 1 "register_operand" "0"))
|
6481 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "xm")))]
|
6482 |
|
|
"(TARGET_SSE && !TARGET_SSE2)"
|
6483 |
|
|
"andnps\t{%2, %0|%0, %2}"
|
6484 |
|
|
[(set_attr "type" "sselog")
|
6485 |
|
|
(set_attr "mode" "V4SF")])
|
6486 |
|
|
|
6487 |
|
|
(define_insn "*avx_andnot3"
|
6488 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "=x")
|
6489 |
|
|
(and:SSEMODEI
|
6490 |
|
|
(not:SSEMODEI (match_operand:SSEMODEI 1 "register_operand" "x"))
|
6491 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "xm")))]
|
6492 |
|
|
"TARGET_AVX"
|
6493 |
|
|
"vpandn\t{%2, %1, %0|%0, %1, %2}"
|
6494 |
|
|
[(set_attr "type" "sselog")
|
6495 |
|
|
(set_attr "prefix" "vex")
|
6496 |
|
|
(set_attr "mode" "TI")])
|
6497 |
|
|
|
6498 |
|
|
(define_insn "sse2_andnot3"
|
6499 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "=x")
|
6500 |
|
|
(and:SSEMODEI
|
6501 |
|
|
(not:SSEMODEI (match_operand:SSEMODEI 1 "register_operand" "0"))
|
6502 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "xm")))]
|
6503 |
|
|
"TARGET_SSE2"
|
6504 |
|
|
"pandn\t{%2, %0|%0, %2}"
|
6505 |
|
|
[(set_attr "type" "sselog")
|
6506 |
|
|
(set_attr "prefix_data16" "1")
|
6507 |
|
|
(set_attr "mode" "TI")])
|
6508 |
|
|
|
6509 |
|
|
(define_insn "*andnottf3"
|
6510 |
|
|
[(set (match_operand:TF 0 "register_operand" "=x")
|
6511 |
|
|
(and:TF
|
6512 |
|
|
(not:TF (match_operand:TF 1 "register_operand" "0"))
|
6513 |
|
|
(match_operand:TF 2 "nonimmediate_operand" "xm")))]
|
6514 |
|
|
"TARGET_SSE2"
|
6515 |
|
|
"pandn\t{%2, %0|%0, %2}"
|
6516 |
|
|
[(set_attr "type" "sselog")
|
6517 |
|
|
(set_attr "prefix_data16" "1")
|
6518 |
|
|
(set_attr "mode" "TI")])
|
6519 |
|
|
|
6520 |
|
|
(define_expand "3"
|
6521 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "")
|
6522 |
|
|
(any_logic:SSEMODEI
|
6523 |
|
|
(match_operand:SSEMODEI 1 "nonimmediate_operand" "")
|
6524 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "")))]
|
6525 |
|
|
"TARGET_SSE"
|
6526 |
|
|
"ix86_fixup_binary_operands_no_copy (, mode, operands);")
|
6527 |
|
|
|
6528 |
|
|
(define_insn "*avx_3"
|
6529 |
|
|
[(set (match_operand:AVX256MODEI 0 "register_operand" "=x")
|
6530 |
|
|
(any_logic:AVX256MODEI
|
6531 |
|
|
(match_operand:AVX256MODEI 1 "nonimmediate_operand" "%x")
|
6532 |
|
|
(match_operand:AVX256MODEI 2 "nonimmediate_operand" "xm")))]
|
6533 |
|
|
"TARGET_AVX
|
6534 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
6535 |
|
|
"vps\t{%2, %1, %0|%0, %1, %2}"
|
6536 |
|
|
[(set_attr "type" "sselog")
|
6537 |
|
|
(set_attr "prefix" "vex")
|
6538 |
|
|
(set_attr "mode" "")])
|
6539 |
|
|
|
6540 |
|
|
(define_insn "*sse_3"
|
6541 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "=x")
|
6542 |
|
|
(any_logic:SSEMODEI
|
6543 |
|
|
(match_operand:SSEMODEI 1 "nonimmediate_operand" "%0")
|
6544 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "xm")))]
|
6545 |
|
|
"(TARGET_SSE && !TARGET_SSE2)
|
6546 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
6547 |
|
|
"ps\t{%2, %0|%0, %2}"
|
6548 |
|
|
[(set_attr "type" "sselog")
|
6549 |
|
|
(set_attr "mode" "V4SF")])
|
6550 |
|
|
|
6551 |
|
|
(define_insn "*avx_3"
|
6552 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "=x")
|
6553 |
|
|
(any_logic:SSEMODEI
|
6554 |
|
|
(match_operand:SSEMODEI 1 "nonimmediate_operand" "%x")
|
6555 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "xm")))]
|
6556 |
|
|
"TARGET_AVX
|
6557 |
|
|
&& ix86_binary_operator_ok (, mode, operands)"
|
6558 |
|
|
"vp\t{%2, %1, %0|%0, %1, %2}"
|
6559 |
|
|
[(set_attr "type" "sselog")
|
6560 |
|
|
(set_attr "prefix" "vex")
|
6561 |
|
|
(set_attr "mode" "TI")])
|
6562 |
|
|
|
6563 |
|
|
(define_insn "*sse2_3"
|
6564 |
|
|
[(set (match_operand:SSEMODEI 0 "register_operand" "=x")
|
6565 |
|
|
(any_logic:SSEMODEI
|
6566 |
|
|
(match_operand:SSEMODEI 1 "nonimmediate_operand" "%0")
|
6567 |
|
|
(match_operand:SSEMODEI 2 "nonimmediate_operand" "xm")))]
|
6568 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (, mode, operands)"
|
6569 |
|
|
"p\t{%2, %0|%0, %2}"
|
6570 |
|
|
[(set_attr "type" "sselog")
|
6571 |
|
|
(set_attr "prefix_data16" "1")
|
6572 |
|
|
(set_attr "mode" "TI")])
|
6573 |
|
|
|
6574 |
|
|
(define_expand "tf3"
|
6575 |
|
|
[(set (match_operand:TF 0 "register_operand" "")
|
6576 |
|
|
(any_logic:TF
|
6577 |
|
|
(match_operand:TF 1 "nonimmediate_operand" "")
|
6578 |
|
|
(match_operand:TF 2 "nonimmediate_operand" "")))]
|
6579 |
|
|
"TARGET_SSE2"
|
6580 |
|
|
"ix86_fixup_binary_operands_no_copy (, TFmode, operands);")
|
6581 |
|
|
|
6582 |
|
|
(define_insn "*tf3"
|
6583 |
|
|
[(set (match_operand:TF 0 "register_operand" "=x")
|
6584 |
|
|
(any_logic:TF
|
6585 |
|
|
(match_operand:TF 1 "nonimmediate_operand" "%0")
|
6586 |
|
|
(match_operand:TF 2 "nonimmediate_operand" "xm")))]
|
6587 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (, TFmode, operands)"
|
6588 |
|
|
"p\t{%2, %0|%0, %2}"
|
6589 |
|
|
[(set_attr "type" "sselog")
|
6590 |
|
|
(set_attr "prefix_data16" "1")
|
6591 |
|
|
(set_attr "mode" "TI")])
|
6592 |
|
|
|
6593 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
6594 |
|
|
;;
|
6595 |
|
|
;; Parallel integral element swizzling
|
6596 |
|
|
;;
|
6597 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
6598 |
|
|
|
6599 |
|
|
(define_expand "vec_pack_trunc_v8hi"
|
6600 |
|
|
[(match_operand:V16QI 0 "register_operand" "")
|
6601 |
|
|
(match_operand:V8HI 1 "register_operand" "")
|
6602 |
|
|
(match_operand:V8HI 2 "register_operand" "")]
|
6603 |
|
|
"TARGET_SSE2"
|
6604 |
|
|
{
|
6605 |
|
|
rtx op1 = gen_lowpart (V16QImode, operands[1]);
|
6606 |
|
|
rtx op2 = gen_lowpart (V16QImode, operands[2]);
|
6607 |
|
|
ix86_expand_vec_extract_even_odd (operands[0], op1, op2, 0);
|
6608 |
|
|
DONE;
|
6609 |
|
|
})
|
6610 |
|
|
|
6611 |
|
|
(define_expand "vec_pack_trunc_v4si"
|
6612 |
|
|
[(match_operand:V8HI 0 "register_operand" "")
|
6613 |
|
|
(match_operand:V4SI 1 "register_operand" "")
|
6614 |
|
|
(match_operand:V4SI 2 "register_operand" "")]
|
6615 |
|
|
"TARGET_SSE2"
|
6616 |
|
|
{
|
6617 |
|
|
rtx op1 = gen_lowpart (V8HImode, operands[1]);
|
6618 |
|
|
rtx op2 = gen_lowpart (V8HImode, operands[2]);
|
6619 |
|
|
ix86_expand_vec_extract_even_odd (operands[0], op1, op2, 0);
|
6620 |
|
|
DONE;
|
6621 |
|
|
})
|
6622 |
|
|
|
6623 |
|
|
(define_expand "vec_pack_trunc_v2di"
|
6624 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
6625 |
|
|
(match_operand:V2DI 1 "register_operand" "")
|
6626 |
|
|
(match_operand:V2DI 2 "register_operand" "")]
|
6627 |
|
|
"TARGET_SSE2"
|
6628 |
|
|
{
|
6629 |
|
|
rtx op1 = gen_lowpart (V4SImode, operands[1]);
|
6630 |
|
|
rtx op2 = gen_lowpart (V4SImode, operands[2]);
|
6631 |
|
|
ix86_expand_vec_extract_even_odd (operands[0], op1, op2, 0);
|
6632 |
|
|
DONE;
|
6633 |
|
|
})
|
6634 |
|
|
|
6635 |
|
|
(define_insn "*avx_packsswb"
|
6636 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6637 |
|
|
(vec_concat:V16QI
|
6638 |
|
|
(ss_truncate:V8QI
|
6639 |
|
|
(match_operand:V8HI 1 "register_operand" "x"))
|
6640 |
|
|
(ss_truncate:V8QI
|
6641 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))))]
|
6642 |
|
|
"TARGET_AVX"
|
6643 |
|
|
"vpacksswb\t{%2, %1, %0|%0, %1, %2}"
|
6644 |
|
|
[(set_attr "type" "sselog")
|
6645 |
|
|
(set_attr "prefix" "vex")
|
6646 |
|
|
(set_attr "mode" "TI")])
|
6647 |
|
|
|
6648 |
|
|
(define_insn "sse2_packsswb"
|
6649 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6650 |
|
|
(vec_concat:V16QI
|
6651 |
|
|
(ss_truncate:V8QI
|
6652 |
|
|
(match_operand:V8HI 1 "register_operand" "0"))
|
6653 |
|
|
(ss_truncate:V8QI
|
6654 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))))]
|
6655 |
|
|
"TARGET_SSE2"
|
6656 |
|
|
"packsswb\t{%2, %0|%0, %2}"
|
6657 |
|
|
[(set_attr "type" "sselog")
|
6658 |
|
|
(set_attr "prefix_data16" "1")
|
6659 |
|
|
(set_attr "mode" "TI")])
|
6660 |
|
|
|
6661 |
|
|
(define_insn "*avx_packssdw"
|
6662 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
6663 |
|
|
(vec_concat:V8HI
|
6664 |
|
|
(ss_truncate:V4HI
|
6665 |
|
|
(match_operand:V4SI 1 "register_operand" "x"))
|
6666 |
|
|
(ss_truncate:V4HI
|
6667 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))))]
|
6668 |
|
|
"TARGET_AVX"
|
6669 |
|
|
"vpackssdw\t{%2, %1, %0|%0, %1, %2}"
|
6670 |
|
|
[(set_attr "type" "sselog")
|
6671 |
|
|
(set_attr "prefix" "vex")
|
6672 |
|
|
(set_attr "mode" "TI")])
|
6673 |
|
|
|
6674 |
|
|
(define_insn "sse2_packssdw"
|
6675 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
6676 |
|
|
(vec_concat:V8HI
|
6677 |
|
|
(ss_truncate:V4HI
|
6678 |
|
|
(match_operand:V4SI 1 "register_operand" "0"))
|
6679 |
|
|
(ss_truncate:V4HI
|
6680 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))))]
|
6681 |
|
|
"TARGET_SSE2"
|
6682 |
|
|
"packssdw\t{%2, %0|%0, %2}"
|
6683 |
|
|
[(set_attr "type" "sselog")
|
6684 |
|
|
(set_attr "prefix_data16" "1")
|
6685 |
|
|
(set_attr "mode" "TI")])
|
6686 |
|
|
|
6687 |
|
|
(define_insn "*avx_packuswb"
|
6688 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6689 |
|
|
(vec_concat:V16QI
|
6690 |
|
|
(us_truncate:V8QI
|
6691 |
|
|
(match_operand:V8HI 1 "register_operand" "x"))
|
6692 |
|
|
(us_truncate:V8QI
|
6693 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))))]
|
6694 |
|
|
"TARGET_AVX"
|
6695 |
|
|
"vpackuswb\t{%2, %1, %0|%0, %1, %2}"
|
6696 |
|
|
[(set_attr "type" "sselog")
|
6697 |
|
|
(set_attr "prefix" "vex")
|
6698 |
|
|
(set_attr "mode" "TI")])
|
6699 |
|
|
|
6700 |
|
|
(define_insn "sse2_packuswb"
|
6701 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6702 |
|
|
(vec_concat:V16QI
|
6703 |
|
|
(us_truncate:V8QI
|
6704 |
|
|
(match_operand:V8HI 1 "register_operand" "0"))
|
6705 |
|
|
(us_truncate:V8QI
|
6706 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))))]
|
6707 |
|
|
"TARGET_SSE2"
|
6708 |
|
|
"packuswb\t{%2, %0|%0, %2}"
|
6709 |
|
|
[(set_attr "type" "sselog")
|
6710 |
|
|
(set_attr "prefix_data16" "1")
|
6711 |
|
|
(set_attr "mode" "TI")])
|
6712 |
|
|
|
6713 |
|
|
(define_insn "*avx_interleave_highv16qi"
|
6714 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6715 |
|
|
(vec_select:V16QI
|
6716 |
|
|
(vec_concat:V32QI
|
6717 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
6718 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm"))
|
6719 |
|
|
(parallel [(const_int 8) (const_int 24)
|
6720 |
|
|
(const_int 9) (const_int 25)
|
6721 |
|
|
(const_int 10) (const_int 26)
|
6722 |
|
|
(const_int 11) (const_int 27)
|
6723 |
|
|
(const_int 12) (const_int 28)
|
6724 |
|
|
(const_int 13) (const_int 29)
|
6725 |
|
|
(const_int 14) (const_int 30)
|
6726 |
|
|
(const_int 15) (const_int 31)])))]
|
6727 |
|
|
"TARGET_AVX"
|
6728 |
|
|
"vpunpckhbw\t{%2, %1, %0|%0, %1, %2}"
|
6729 |
|
|
[(set_attr "type" "sselog")
|
6730 |
|
|
(set_attr "prefix" "vex")
|
6731 |
|
|
(set_attr "mode" "TI")])
|
6732 |
|
|
|
6733 |
|
|
(define_insn "vec_interleave_highv16qi"
|
6734 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6735 |
|
|
(vec_select:V16QI
|
6736 |
|
|
(vec_concat:V32QI
|
6737 |
|
|
(match_operand:V16QI 1 "register_operand" "0")
|
6738 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm"))
|
6739 |
|
|
(parallel [(const_int 8) (const_int 24)
|
6740 |
|
|
(const_int 9) (const_int 25)
|
6741 |
|
|
(const_int 10) (const_int 26)
|
6742 |
|
|
(const_int 11) (const_int 27)
|
6743 |
|
|
(const_int 12) (const_int 28)
|
6744 |
|
|
(const_int 13) (const_int 29)
|
6745 |
|
|
(const_int 14) (const_int 30)
|
6746 |
|
|
(const_int 15) (const_int 31)])))]
|
6747 |
|
|
"TARGET_SSE2"
|
6748 |
|
|
"punpckhbw\t{%2, %0|%0, %2}"
|
6749 |
|
|
[(set_attr "type" "sselog")
|
6750 |
|
|
(set_attr "prefix_data16" "1")
|
6751 |
|
|
(set_attr "mode" "TI")])
|
6752 |
|
|
|
6753 |
|
|
(define_insn "*avx_interleave_lowv16qi"
|
6754 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6755 |
|
|
(vec_select:V16QI
|
6756 |
|
|
(vec_concat:V32QI
|
6757 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
6758 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm"))
|
6759 |
|
|
(parallel [(const_int 0) (const_int 16)
|
6760 |
|
|
(const_int 1) (const_int 17)
|
6761 |
|
|
(const_int 2) (const_int 18)
|
6762 |
|
|
(const_int 3) (const_int 19)
|
6763 |
|
|
(const_int 4) (const_int 20)
|
6764 |
|
|
(const_int 5) (const_int 21)
|
6765 |
|
|
(const_int 6) (const_int 22)
|
6766 |
|
|
(const_int 7) (const_int 23)])))]
|
6767 |
|
|
"TARGET_AVX"
|
6768 |
|
|
"vpunpcklbw\t{%2, %1, %0|%0, %1, %2}"
|
6769 |
|
|
[(set_attr "type" "sselog")
|
6770 |
|
|
(set_attr "prefix" "vex")
|
6771 |
|
|
(set_attr "mode" "TI")])
|
6772 |
|
|
|
6773 |
|
|
(define_insn "vec_interleave_lowv16qi"
|
6774 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6775 |
|
|
(vec_select:V16QI
|
6776 |
|
|
(vec_concat:V32QI
|
6777 |
|
|
(match_operand:V16QI 1 "register_operand" "0")
|
6778 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm"))
|
6779 |
|
|
(parallel [(const_int 0) (const_int 16)
|
6780 |
|
|
(const_int 1) (const_int 17)
|
6781 |
|
|
(const_int 2) (const_int 18)
|
6782 |
|
|
(const_int 3) (const_int 19)
|
6783 |
|
|
(const_int 4) (const_int 20)
|
6784 |
|
|
(const_int 5) (const_int 21)
|
6785 |
|
|
(const_int 6) (const_int 22)
|
6786 |
|
|
(const_int 7) (const_int 23)])))]
|
6787 |
|
|
"TARGET_SSE2"
|
6788 |
|
|
"punpcklbw\t{%2, %0|%0, %2}"
|
6789 |
|
|
[(set_attr "type" "sselog")
|
6790 |
|
|
(set_attr "prefix_data16" "1")
|
6791 |
|
|
(set_attr "mode" "TI")])
|
6792 |
|
|
|
6793 |
|
|
(define_insn "*avx_interleave_highv8hi"
|
6794 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
6795 |
|
|
(vec_select:V8HI
|
6796 |
|
|
(vec_concat:V16HI
|
6797 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
6798 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))
|
6799 |
|
|
(parallel [(const_int 4) (const_int 12)
|
6800 |
|
|
(const_int 5) (const_int 13)
|
6801 |
|
|
(const_int 6) (const_int 14)
|
6802 |
|
|
(const_int 7) (const_int 15)])))]
|
6803 |
|
|
"TARGET_AVX"
|
6804 |
|
|
"vpunpckhwd\t{%2, %1, %0|%0, %1, %2}"
|
6805 |
|
|
[(set_attr "type" "sselog")
|
6806 |
|
|
(set_attr "prefix" "vex")
|
6807 |
|
|
(set_attr "mode" "TI")])
|
6808 |
|
|
|
6809 |
|
|
(define_insn "vec_interleave_highv8hi"
|
6810 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
6811 |
|
|
(vec_select:V8HI
|
6812 |
|
|
(vec_concat:V16HI
|
6813 |
|
|
(match_operand:V8HI 1 "register_operand" "0")
|
6814 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))
|
6815 |
|
|
(parallel [(const_int 4) (const_int 12)
|
6816 |
|
|
(const_int 5) (const_int 13)
|
6817 |
|
|
(const_int 6) (const_int 14)
|
6818 |
|
|
(const_int 7) (const_int 15)])))]
|
6819 |
|
|
"TARGET_SSE2"
|
6820 |
|
|
"punpckhwd\t{%2, %0|%0, %2}"
|
6821 |
|
|
[(set_attr "type" "sselog")
|
6822 |
|
|
(set_attr "prefix_data16" "1")
|
6823 |
|
|
(set_attr "mode" "TI")])
|
6824 |
|
|
|
6825 |
|
|
(define_insn "*avx_interleave_lowv8hi"
|
6826 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
6827 |
|
|
(vec_select:V8HI
|
6828 |
|
|
(vec_concat:V16HI
|
6829 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
6830 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))
|
6831 |
|
|
(parallel [(const_int 0) (const_int 8)
|
6832 |
|
|
(const_int 1) (const_int 9)
|
6833 |
|
|
(const_int 2) (const_int 10)
|
6834 |
|
|
(const_int 3) (const_int 11)])))]
|
6835 |
|
|
"TARGET_AVX"
|
6836 |
|
|
"vpunpcklwd\t{%2, %1, %0|%0, %1, %2}"
|
6837 |
|
|
[(set_attr "type" "sselog")
|
6838 |
|
|
(set_attr "prefix" "vex")
|
6839 |
|
|
(set_attr "mode" "TI")])
|
6840 |
|
|
|
6841 |
|
|
(define_insn "vec_interleave_lowv8hi"
|
6842 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
6843 |
|
|
(vec_select:V8HI
|
6844 |
|
|
(vec_concat:V16HI
|
6845 |
|
|
(match_operand:V8HI 1 "register_operand" "0")
|
6846 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))
|
6847 |
|
|
(parallel [(const_int 0) (const_int 8)
|
6848 |
|
|
(const_int 1) (const_int 9)
|
6849 |
|
|
(const_int 2) (const_int 10)
|
6850 |
|
|
(const_int 3) (const_int 11)])))]
|
6851 |
|
|
"TARGET_SSE2"
|
6852 |
|
|
"punpcklwd\t{%2, %0|%0, %2}"
|
6853 |
|
|
[(set_attr "type" "sselog")
|
6854 |
|
|
(set_attr "prefix_data16" "1")
|
6855 |
|
|
(set_attr "mode" "TI")])
|
6856 |
|
|
|
6857 |
|
|
(define_insn "*avx_interleave_highv4si"
|
6858 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
6859 |
|
|
(vec_select:V4SI
|
6860 |
|
|
(vec_concat:V8SI
|
6861 |
|
|
(match_operand:V4SI 1 "register_operand" "x")
|
6862 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))
|
6863 |
|
|
(parallel [(const_int 2) (const_int 6)
|
6864 |
|
|
(const_int 3) (const_int 7)])))]
|
6865 |
|
|
"TARGET_AVX"
|
6866 |
|
|
"vpunpckhdq\t{%2, %1, %0|%0, %1, %2}"
|
6867 |
|
|
[(set_attr "type" "sselog")
|
6868 |
|
|
(set_attr "prefix" "vex")
|
6869 |
|
|
(set_attr "mode" "TI")])
|
6870 |
|
|
|
6871 |
|
|
(define_insn "vec_interleave_highv4si"
|
6872 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
6873 |
|
|
(vec_select:V4SI
|
6874 |
|
|
(vec_concat:V8SI
|
6875 |
|
|
(match_operand:V4SI 1 "register_operand" "0")
|
6876 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))
|
6877 |
|
|
(parallel [(const_int 2) (const_int 6)
|
6878 |
|
|
(const_int 3) (const_int 7)])))]
|
6879 |
|
|
"TARGET_SSE2"
|
6880 |
|
|
"punpckhdq\t{%2, %0|%0, %2}"
|
6881 |
|
|
[(set_attr "type" "sselog")
|
6882 |
|
|
(set_attr "prefix_data16" "1")
|
6883 |
|
|
(set_attr "mode" "TI")])
|
6884 |
|
|
|
6885 |
|
|
(define_insn "*avx_interleave_lowv4si"
|
6886 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
6887 |
|
|
(vec_select:V4SI
|
6888 |
|
|
(vec_concat:V8SI
|
6889 |
|
|
(match_operand:V4SI 1 "register_operand" "x")
|
6890 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))
|
6891 |
|
|
(parallel [(const_int 0) (const_int 4)
|
6892 |
|
|
(const_int 1) (const_int 5)])))]
|
6893 |
|
|
"TARGET_AVX"
|
6894 |
|
|
"vpunpckldq\t{%2, %1, %0|%0, %1, %2}"
|
6895 |
|
|
[(set_attr "type" "sselog")
|
6896 |
|
|
(set_attr "prefix" "vex")
|
6897 |
|
|
(set_attr "mode" "TI")])
|
6898 |
|
|
|
6899 |
|
|
(define_insn "vec_interleave_lowv4si"
|
6900 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
6901 |
|
|
(vec_select:V4SI
|
6902 |
|
|
(vec_concat:V8SI
|
6903 |
|
|
(match_operand:V4SI 1 "register_operand" "0")
|
6904 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))
|
6905 |
|
|
(parallel [(const_int 0) (const_int 4)
|
6906 |
|
|
(const_int 1) (const_int 5)])))]
|
6907 |
|
|
"TARGET_SSE2"
|
6908 |
|
|
"punpckldq\t{%2, %0|%0, %2}"
|
6909 |
|
|
[(set_attr "type" "sselog")
|
6910 |
|
|
(set_attr "prefix_data16" "1")
|
6911 |
|
|
(set_attr "mode" "TI")])
|
6912 |
|
|
|
6913 |
|
|
(define_insn "*avx_pinsr"
|
6914 |
|
|
[(set (match_operand:SSEMODE124 0 "register_operand" "=x")
|
6915 |
|
|
(vec_merge:SSEMODE124
|
6916 |
|
|
(vec_duplicate:SSEMODE124
|
6917 |
|
|
(match_operand: 2 "nonimmediate_operand" "rm"))
|
6918 |
|
|
(match_operand:SSEMODE124 1 "register_operand" "x")
|
6919 |
|
|
(match_operand:SI 3 "const_pow2_1_to__operand" "n")))]
|
6920 |
|
|
"TARGET_AVX"
|
6921 |
|
|
{
|
6922 |
|
|
operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])));
|
6923 |
|
|
if (MEM_P (operands[2]))
|
6924 |
|
|
return "vpinsr\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
6925 |
|
|
else
|
6926 |
|
|
return "vpinsr\t{%3, %k2, %1, %0|%0, %1, %k2, %3}";
|
6927 |
|
|
}
|
6928 |
|
|
[(set_attr "type" "sselog")
|
6929 |
|
|
(set (attr "prefix_extra")
|
6930 |
|
|
(if_then_else (match_operand:V8HI 0 "register_operand" "")
|
6931 |
|
|
(const_string "0")
|
6932 |
|
|
(const_string "1")))
|
6933 |
|
|
(set_attr "length_immediate" "1")
|
6934 |
|
|
(set_attr "prefix" "vex")
|
6935 |
|
|
(set_attr "mode" "TI")])
|
6936 |
|
|
|
6937 |
|
|
(define_insn "*sse4_1_pinsrb"
|
6938 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
6939 |
|
|
(vec_merge:V16QI
|
6940 |
|
|
(vec_duplicate:V16QI
|
6941 |
|
|
(match_operand:QI 2 "nonimmediate_operand" "rm"))
|
6942 |
|
|
(match_operand:V16QI 1 "register_operand" "0")
|
6943 |
|
|
(match_operand:SI 3 "const_pow2_1_to_32768_operand" "n")))]
|
6944 |
|
|
"TARGET_SSE4_1"
|
6945 |
|
|
{
|
6946 |
|
|
operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])));
|
6947 |
|
|
if (MEM_P (operands[2]))
|
6948 |
|
|
return "pinsrb\t{%3, %2, %0|%0, %2, %3}";
|
6949 |
|
|
else
|
6950 |
|
|
return "pinsrb\t{%3, %k2, %0|%0, %k2, %3}";
|
6951 |
|
|
}
|
6952 |
|
|
[(set_attr "type" "sselog")
|
6953 |
|
|
(set_attr "prefix_extra" "1")
|
6954 |
|
|
(set_attr "length_immediate" "1")
|
6955 |
|
|
(set_attr "mode" "TI")])
|
6956 |
|
|
|
6957 |
|
|
(define_insn "*sse2_pinsrw"
|
6958 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
6959 |
|
|
(vec_merge:V8HI
|
6960 |
|
|
(vec_duplicate:V8HI
|
6961 |
|
|
(match_operand:HI 2 "nonimmediate_operand" "rm"))
|
6962 |
|
|
(match_operand:V8HI 1 "register_operand" "0")
|
6963 |
|
|
(match_operand:SI 3 "const_pow2_1_to_128_operand" "n")))]
|
6964 |
|
|
"TARGET_SSE2"
|
6965 |
|
|
{
|
6966 |
|
|
operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])));
|
6967 |
|
|
if (MEM_P (operands[2]))
|
6968 |
|
|
return "pinsrw\t{%3, %2, %0|%0, %2, %3}";
|
6969 |
|
|
else
|
6970 |
|
|
return "pinsrw\t{%3, %k2, %0|%0, %k2, %3}";
|
6971 |
|
|
}
|
6972 |
|
|
[(set_attr "type" "sselog")
|
6973 |
|
|
(set_attr "prefix_data16" "1")
|
6974 |
|
|
(set_attr "length_immediate" "1")
|
6975 |
|
|
(set_attr "mode" "TI")])
|
6976 |
|
|
|
6977 |
|
|
;; It must come before sse2_loadld since it is preferred.
|
6978 |
|
|
(define_insn "*sse4_1_pinsrd"
|
6979 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
6980 |
|
|
(vec_merge:V4SI
|
6981 |
|
|
(vec_duplicate:V4SI
|
6982 |
|
|
(match_operand:SI 2 "nonimmediate_operand" "rm"))
|
6983 |
|
|
(match_operand:V4SI 1 "register_operand" "0")
|
6984 |
|
|
(match_operand:SI 3 "const_pow2_1_to_8_operand" "n")))]
|
6985 |
|
|
"TARGET_SSE4_1"
|
6986 |
|
|
{
|
6987 |
|
|
operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])));
|
6988 |
|
|
return "pinsrd\t{%3, %2, %0|%0, %2, %3}";
|
6989 |
|
|
}
|
6990 |
|
|
[(set_attr "type" "sselog")
|
6991 |
|
|
(set_attr "prefix_extra" "1")
|
6992 |
|
|
(set_attr "length_immediate" "1")
|
6993 |
|
|
(set_attr "mode" "TI")])
|
6994 |
|
|
|
6995 |
|
|
(define_insn "*avx_pinsrq"
|
6996 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
6997 |
|
|
(vec_merge:V2DI
|
6998 |
|
|
(vec_duplicate:V2DI
|
6999 |
|
|
(match_operand:DI 2 "nonimmediate_operand" "rm"))
|
7000 |
|
|
(match_operand:V2DI 1 "register_operand" "x")
|
7001 |
|
|
(match_operand:SI 3 "const_pow2_1_to_2_operand" "n")))]
|
7002 |
|
|
"TARGET_AVX && TARGET_64BIT"
|
7003 |
|
|
{
|
7004 |
|
|
operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])));
|
7005 |
|
|
return "vpinsrq\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
7006 |
|
|
}
|
7007 |
|
|
[(set_attr "type" "sselog")
|
7008 |
|
|
(set_attr "prefix_extra" "1")
|
7009 |
|
|
(set_attr "length_immediate" "1")
|
7010 |
|
|
(set_attr "prefix" "vex")
|
7011 |
|
|
(set_attr "mode" "TI")])
|
7012 |
|
|
|
7013 |
|
|
(define_insn "*sse4_1_pinsrq"
|
7014 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
7015 |
|
|
(vec_merge:V2DI
|
7016 |
|
|
(vec_duplicate:V2DI
|
7017 |
|
|
(match_operand:DI 2 "nonimmediate_operand" "rm"))
|
7018 |
|
|
(match_operand:V2DI 1 "register_operand" "0")
|
7019 |
|
|
(match_operand:SI 3 "const_pow2_1_to_2_operand" "n")))]
|
7020 |
|
|
"TARGET_SSE4_1 && TARGET_64BIT"
|
7021 |
|
|
{
|
7022 |
|
|
operands[3] = GEN_INT (exact_log2 (INTVAL (operands[3])));
|
7023 |
|
|
return "pinsrq\t{%3, %2, %0|%0, %2, %3}";
|
7024 |
|
|
}
|
7025 |
|
|
[(set_attr "type" "sselog")
|
7026 |
|
|
(set_attr "prefix_rex" "1")
|
7027 |
|
|
(set_attr "prefix_extra" "1")
|
7028 |
|
|
(set_attr "length_immediate" "1")
|
7029 |
|
|
(set_attr "mode" "TI")])
|
7030 |
|
|
|
7031 |
|
|
(define_insn "*sse4_1_pextrb"
|
7032 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r")
|
7033 |
|
|
(zero_extend:SI
|
7034 |
|
|
(vec_select:QI
|
7035 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
7036 |
|
|
(parallel [(match_operand:SI 2 "const_0_to_15_operand" "n")]))))]
|
7037 |
|
|
"TARGET_SSE4_1"
|
7038 |
|
|
"%vpextrb\t{%2, %1, %0|%0, %1, %2}"
|
7039 |
|
|
[(set_attr "type" "sselog")
|
7040 |
|
|
(set_attr "prefix_extra" "1")
|
7041 |
|
|
(set_attr "length_immediate" "1")
|
7042 |
|
|
(set_attr "prefix" "maybe_vex")
|
7043 |
|
|
(set_attr "mode" "TI")])
|
7044 |
|
|
|
7045 |
|
|
(define_insn "*sse4_1_pextrb_memory"
|
7046 |
|
|
[(set (match_operand:QI 0 "memory_operand" "=m")
|
7047 |
|
|
(vec_select:QI
|
7048 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
7049 |
|
|
(parallel [(match_operand:SI 2 "const_0_to_15_operand" "n")])))]
|
7050 |
|
|
"TARGET_SSE4_1"
|
7051 |
|
|
"%vpextrb\t{%2, %1, %0|%0, %1, %2}"
|
7052 |
|
|
[(set_attr "type" "sselog")
|
7053 |
|
|
(set_attr "prefix_extra" "1")
|
7054 |
|
|
(set_attr "length_immediate" "1")
|
7055 |
|
|
(set_attr "prefix" "maybe_vex")
|
7056 |
|
|
(set_attr "mode" "TI")])
|
7057 |
|
|
|
7058 |
|
|
(define_insn "*sse2_pextrw"
|
7059 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r")
|
7060 |
|
|
(zero_extend:SI
|
7061 |
|
|
(vec_select:HI
|
7062 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
7063 |
|
|
(parallel [(match_operand:SI 2 "const_0_to_7_operand" "n")]))))]
|
7064 |
|
|
"TARGET_SSE2"
|
7065 |
|
|
"%vpextrw\t{%2, %1, %0|%0, %1, %2}"
|
7066 |
|
|
[(set_attr "type" "sselog")
|
7067 |
|
|
(set_attr "prefix_data16" "1")
|
7068 |
|
|
(set_attr "length_immediate" "1")
|
7069 |
|
|
(set_attr "prefix" "maybe_vex")
|
7070 |
|
|
(set_attr "mode" "TI")])
|
7071 |
|
|
|
7072 |
|
|
(define_insn "*sse4_1_pextrw_memory"
|
7073 |
|
|
[(set (match_operand:HI 0 "memory_operand" "=m")
|
7074 |
|
|
(vec_select:HI
|
7075 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
7076 |
|
|
(parallel [(match_operand:SI 2 "const_0_to_7_operand" "n")])))]
|
7077 |
|
|
"TARGET_SSE4_1"
|
7078 |
|
|
"%vpextrw\t{%2, %1, %0|%0, %1, %2}"
|
7079 |
|
|
[(set_attr "type" "sselog")
|
7080 |
|
|
(set_attr "prefix_extra" "1")
|
7081 |
|
|
(set_attr "length_immediate" "1")
|
7082 |
|
|
(set_attr "prefix" "maybe_vex")
|
7083 |
|
|
(set_attr "mode" "TI")])
|
7084 |
|
|
|
7085 |
|
|
(define_insn "*sse4_1_pextrd"
|
7086 |
|
|
[(set (match_operand:SI 0 "nonimmediate_operand" "=rm")
|
7087 |
|
|
(vec_select:SI
|
7088 |
|
|
(match_operand:V4SI 1 "register_operand" "x")
|
7089 |
|
|
(parallel [(match_operand:SI 2 "const_0_to_3_operand" "n")])))]
|
7090 |
|
|
"TARGET_SSE4_1"
|
7091 |
|
|
"%vpextrd\t{%2, %1, %0|%0, %1, %2}"
|
7092 |
|
|
[(set_attr "type" "sselog")
|
7093 |
|
|
(set_attr "prefix_extra" "1")
|
7094 |
|
|
(set_attr "length_immediate" "1")
|
7095 |
|
|
(set_attr "prefix" "maybe_vex")
|
7096 |
|
|
(set_attr "mode" "TI")])
|
7097 |
|
|
|
7098 |
|
|
;; It must come before *vec_extractv2di_1_sse since it is preferred.
|
7099 |
|
|
(define_insn "*sse4_1_pextrq"
|
7100 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "=rm")
|
7101 |
|
|
(vec_select:DI
|
7102 |
|
|
(match_operand:V2DI 1 "register_operand" "x")
|
7103 |
|
|
(parallel [(match_operand:SI 2 "const_0_to_1_operand" "n")])))]
|
7104 |
|
|
"TARGET_SSE4_1 && TARGET_64BIT"
|
7105 |
|
|
"%vpextrq\t{%2, %1, %0|%0, %1, %2}"
|
7106 |
|
|
[(set_attr "type" "sselog")
|
7107 |
|
|
(set_attr "prefix_rex" "1")
|
7108 |
|
|
(set_attr "prefix_extra" "1")
|
7109 |
|
|
(set_attr "length_immediate" "1")
|
7110 |
|
|
(set_attr "prefix" "maybe_vex")
|
7111 |
|
|
(set_attr "mode" "TI")])
|
7112 |
|
|
|
7113 |
|
|
(define_expand "sse2_pshufd"
|
7114 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
7115 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "")
|
7116 |
|
|
(match_operand:SI 2 "const_int_operand" "")]
|
7117 |
|
|
"TARGET_SSE2"
|
7118 |
|
|
{
|
7119 |
|
|
int mask = INTVAL (operands[2]);
|
7120 |
|
|
emit_insn (gen_sse2_pshufd_1 (operands[0], operands[1],
|
7121 |
|
|
GEN_INT ((mask >> 0) & 3),
|
7122 |
|
|
GEN_INT ((mask >> 2) & 3),
|
7123 |
|
|
GEN_INT ((mask >> 4) & 3),
|
7124 |
|
|
GEN_INT ((mask >> 6) & 3)));
|
7125 |
|
|
DONE;
|
7126 |
|
|
})
|
7127 |
|
|
|
7128 |
|
|
(define_insn "sse2_pshufd_1"
|
7129 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
7130 |
|
|
(vec_select:V4SI
|
7131 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "xm")
|
7132 |
|
|
(parallel [(match_operand 2 "const_0_to_3_operand" "")
|
7133 |
|
|
(match_operand 3 "const_0_to_3_operand" "")
|
7134 |
|
|
(match_operand 4 "const_0_to_3_operand" "")
|
7135 |
|
|
(match_operand 5 "const_0_to_3_operand" "")])))]
|
7136 |
|
|
"TARGET_SSE2"
|
7137 |
|
|
{
|
7138 |
|
|
int mask = 0;
|
7139 |
|
|
mask |= INTVAL (operands[2]) << 0;
|
7140 |
|
|
mask |= INTVAL (operands[3]) << 2;
|
7141 |
|
|
mask |= INTVAL (operands[4]) << 4;
|
7142 |
|
|
mask |= INTVAL (operands[5]) << 6;
|
7143 |
|
|
operands[2] = GEN_INT (mask);
|
7144 |
|
|
|
7145 |
|
|
return "%vpshufd\t{%2, %1, %0|%0, %1, %2}";
|
7146 |
|
|
}
|
7147 |
|
|
[(set_attr "type" "sselog1")
|
7148 |
|
|
(set_attr "prefix_data16" "1")
|
7149 |
|
|
(set_attr "prefix" "maybe_vex")
|
7150 |
|
|
(set_attr "length_immediate" "1")
|
7151 |
|
|
(set_attr "mode" "TI")])
|
7152 |
|
|
|
7153 |
|
|
(define_expand "sse2_pshuflw"
|
7154 |
|
|
[(match_operand:V8HI 0 "register_operand" "")
|
7155 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "")
|
7156 |
|
|
(match_operand:SI 2 "const_int_operand" "")]
|
7157 |
|
|
"TARGET_SSE2"
|
7158 |
|
|
{
|
7159 |
|
|
int mask = INTVAL (operands[2]);
|
7160 |
|
|
emit_insn (gen_sse2_pshuflw_1 (operands[0], operands[1],
|
7161 |
|
|
GEN_INT ((mask >> 0) & 3),
|
7162 |
|
|
GEN_INT ((mask >> 2) & 3),
|
7163 |
|
|
GEN_INT ((mask >> 4) & 3),
|
7164 |
|
|
GEN_INT ((mask >> 6) & 3)));
|
7165 |
|
|
DONE;
|
7166 |
|
|
})
|
7167 |
|
|
|
7168 |
|
|
(define_insn "sse2_pshuflw_1"
|
7169 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
7170 |
|
|
(vec_select:V8HI
|
7171 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "xm")
|
7172 |
|
|
(parallel [(match_operand 2 "const_0_to_3_operand" "")
|
7173 |
|
|
(match_operand 3 "const_0_to_3_operand" "")
|
7174 |
|
|
(match_operand 4 "const_0_to_3_operand" "")
|
7175 |
|
|
(match_operand 5 "const_0_to_3_operand" "")
|
7176 |
|
|
(const_int 4)
|
7177 |
|
|
(const_int 5)
|
7178 |
|
|
(const_int 6)
|
7179 |
|
|
(const_int 7)])))]
|
7180 |
|
|
"TARGET_SSE2"
|
7181 |
|
|
{
|
7182 |
|
|
int mask = 0;
|
7183 |
|
|
mask |= INTVAL (operands[2]) << 0;
|
7184 |
|
|
mask |= INTVAL (operands[3]) << 2;
|
7185 |
|
|
mask |= INTVAL (operands[4]) << 4;
|
7186 |
|
|
mask |= INTVAL (operands[5]) << 6;
|
7187 |
|
|
operands[2] = GEN_INT (mask);
|
7188 |
|
|
|
7189 |
|
|
return "%vpshuflw\t{%2, %1, %0|%0, %1, %2}";
|
7190 |
|
|
}
|
7191 |
|
|
[(set_attr "type" "sselog")
|
7192 |
|
|
(set_attr "prefix_data16" "0")
|
7193 |
|
|
(set_attr "prefix_rep" "1")
|
7194 |
|
|
(set_attr "prefix" "maybe_vex")
|
7195 |
|
|
(set_attr "length_immediate" "1")
|
7196 |
|
|
(set_attr "mode" "TI")])
|
7197 |
|
|
|
7198 |
|
|
(define_expand "sse2_pshufhw"
|
7199 |
|
|
[(match_operand:V8HI 0 "register_operand" "")
|
7200 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "")
|
7201 |
|
|
(match_operand:SI 2 "const_int_operand" "")]
|
7202 |
|
|
"TARGET_SSE2"
|
7203 |
|
|
{
|
7204 |
|
|
int mask = INTVAL (operands[2]);
|
7205 |
|
|
emit_insn (gen_sse2_pshufhw_1 (operands[0], operands[1],
|
7206 |
|
|
GEN_INT (((mask >> 0) & 3) + 4),
|
7207 |
|
|
GEN_INT (((mask >> 2) & 3) + 4),
|
7208 |
|
|
GEN_INT (((mask >> 4) & 3) + 4),
|
7209 |
|
|
GEN_INT (((mask >> 6) & 3) + 4)));
|
7210 |
|
|
DONE;
|
7211 |
|
|
})
|
7212 |
|
|
|
7213 |
|
|
(define_insn "sse2_pshufhw_1"
|
7214 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
7215 |
|
|
(vec_select:V8HI
|
7216 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "xm")
|
7217 |
|
|
(parallel [(const_int 0)
|
7218 |
|
|
(const_int 1)
|
7219 |
|
|
(const_int 2)
|
7220 |
|
|
(const_int 3)
|
7221 |
|
|
(match_operand 2 "const_4_to_7_operand" "")
|
7222 |
|
|
(match_operand 3 "const_4_to_7_operand" "")
|
7223 |
|
|
(match_operand 4 "const_4_to_7_operand" "")
|
7224 |
|
|
(match_operand 5 "const_4_to_7_operand" "")])))]
|
7225 |
|
|
"TARGET_SSE2"
|
7226 |
|
|
{
|
7227 |
|
|
int mask = 0;
|
7228 |
|
|
mask |= (INTVAL (operands[2]) - 4) << 0;
|
7229 |
|
|
mask |= (INTVAL (operands[3]) - 4) << 2;
|
7230 |
|
|
mask |= (INTVAL (operands[4]) - 4) << 4;
|
7231 |
|
|
mask |= (INTVAL (operands[5]) - 4) << 6;
|
7232 |
|
|
operands[2] = GEN_INT (mask);
|
7233 |
|
|
|
7234 |
|
|
return "%vpshufhw\t{%2, %1, %0|%0, %1, %2}";
|
7235 |
|
|
}
|
7236 |
|
|
[(set_attr "type" "sselog")
|
7237 |
|
|
(set_attr "prefix_rep" "1")
|
7238 |
|
|
(set_attr "prefix_data16" "0")
|
7239 |
|
|
(set_attr "prefix" "maybe_vex")
|
7240 |
|
|
(set_attr "length_immediate" "1")
|
7241 |
|
|
(set_attr "mode" "TI")])
|
7242 |
|
|
|
7243 |
|
|
(define_expand "sse2_loadd"
|
7244 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "")
|
7245 |
|
|
(vec_merge:V4SI
|
7246 |
|
|
(vec_duplicate:V4SI
|
7247 |
|
|
(match_operand:SI 1 "nonimmediate_operand" ""))
|
7248 |
|
|
(match_dup 2)
|
7249 |
|
|
(const_int 1)))]
|
7250 |
|
|
"TARGET_SSE"
|
7251 |
|
|
"operands[2] = CONST0_RTX (V4SImode);")
|
7252 |
|
|
|
7253 |
|
|
(define_insn "*avx_loadld"
|
7254 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x,Yi,x")
|
7255 |
|
|
(vec_merge:V4SI
|
7256 |
|
|
(vec_duplicate:V4SI
|
7257 |
|
|
(match_operand:SI 2 "nonimmediate_operand" "m ,r ,x"))
|
7258 |
|
|
(match_operand:V4SI 1 "reg_or_0_operand" "C ,C ,x")
|
7259 |
|
|
(const_int 1)))]
|
7260 |
|
|
"TARGET_AVX"
|
7261 |
|
|
"@
|
7262 |
|
|
vmovd\t{%2, %0|%0, %2}
|
7263 |
|
|
vmovd\t{%2, %0|%0, %2}
|
7264 |
|
|
vmovss\t{%2, %1, %0|%0, %1, %2}"
|
7265 |
|
|
[(set_attr "type" "ssemov")
|
7266 |
|
|
(set_attr "prefix" "vex")
|
7267 |
|
|
(set_attr "mode" "TI,TI,V4SF")])
|
7268 |
|
|
|
7269 |
|
|
(define_insn "sse2_loadld"
|
7270 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=Y2,Yi,x,x")
|
7271 |
|
|
(vec_merge:V4SI
|
7272 |
|
|
(vec_duplicate:V4SI
|
7273 |
|
|
(match_operand:SI 2 "nonimmediate_operand" "m ,r ,m,x"))
|
7274 |
|
|
(match_operand:V4SI 1 "reg_or_0_operand" "C ,C ,C,0")
|
7275 |
|
|
(const_int 1)))]
|
7276 |
|
|
"TARGET_SSE"
|
7277 |
|
|
"@
|
7278 |
|
|
movd\t{%2, %0|%0, %2}
|
7279 |
|
|
movd\t{%2, %0|%0, %2}
|
7280 |
|
|
movss\t{%2, %0|%0, %2}
|
7281 |
|
|
movss\t{%2, %0|%0, %2}"
|
7282 |
|
|
[(set_attr "type" "ssemov")
|
7283 |
|
|
(set_attr "mode" "TI,TI,V4SF,SF")])
|
7284 |
|
|
|
7285 |
|
|
(define_insn_and_split "sse2_stored"
|
7286 |
|
|
[(set (match_operand:SI 0 "nonimmediate_operand" "=mx,r")
|
7287 |
|
|
(vec_select:SI
|
7288 |
|
|
(match_operand:V4SI 1 "register_operand" "x,Yi")
|
7289 |
|
|
(parallel [(const_int 0)])))]
|
7290 |
|
|
"TARGET_SSE"
|
7291 |
|
|
"#"
|
7292 |
|
|
"&& reload_completed
|
7293 |
|
|
&& (TARGET_INTER_UNIT_MOVES
|
7294 |
|
|
|| MEM_P (operands [0])
|
7295 |
|
|
|| !GENERAL_REGNO_P (true_regnum (operands [0])))"
|
7296 |
|
|
[(set (match_dup 0) (match_dup 1))]
|
7297 |
|
|
{
|
7298 |
|
|
operands[1] = gen_rtx_REG (SImode, REGNO (operands[1]));
|
7299 |
|
|
})
|
7300 |
|
|
|
7301 |
|
|
(define_insn_and_split "*vec_ext_v4si_mem"
|
7302 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r")
|
7303 |
|
|
(vec_select:SI
|
7304 |
|
|
(match_operand:V4SI 1 "memory_operand" "o")
|
7305 |
|
|
(parallel [(match_operand 2 "const_0_to_3_operand" "")])))]
|
7306 |
|
|
""
|
7307 |
|
|
"#"
|
7308 |
|
|
"reload_completed"
|
7309 |
|
|
[(const_int 0)]
|
7310 |
|
|
{
|
7311 |
|
|
int i = INTVAL (operands[2]);
|
7312 |
|
|
|
7313 |
|
|
emit_move_insn (operands[0], adjust_address (operands[1], SImode, i*4));
|
7314 |
|
|
DONE;
|
7315 |
|
|
})
|
7316 |
|
|
|
7317 |
|
|
(define_expand "sse_storeq"
|
7318 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "")
|
7319 |
|
|
(vec_select:DI
|
7320 |
|
|
(match_operand:V2DI 1 "register_operand" "")
|
7321 |
|
|
(parallel [(const_int 0)])))]
|
7322 |
|
|
"TARGET_SSE"
|
7323 |
|
|
"")
|
7324 |
|
|
|
7325 |
|
|
(define_insn "*sse2_storeq_rex64"
|
7326 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "=mx,*r,r")
|
7327 |
|
|
(vec_select:DI
|
7328 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "x,Yi,o")
|
7329 |
|
|
(parallel [(const_int 0)])))]
|
7330 |
|
|
"TARGET_64BIT && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
7331 |
|
|
"@
|
7332 |
|
|
#
|
7333 |
|
|
#
|
7334 |
|
|
%vmov{q}\t{%1, %0|%0, %1}"
|
7335 |
|
|
[(set_attr "type" "*,*,imov")
|
7336 |
|
|
(set_attr "prefix" "*,*,maybe_vex")
|
7337 |
|
|
(set_attr "mode" "*,*,DI")])
|
7338 |
|
|
|
7339 |
|
|
(define_insn "*sse2_storeq"
|
7340 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "=mx")
|
7341 |
|
|
(vec_select:DI
|
7342 |
|
|
(match_operand:V2DI 1 "register_operand" "x")
|
7343 |
|
|
(parallel [(const_int 0)])))]
|
7344 |
|
|
"TARGET_SSE"
|
7345 |
|
|
"#")
|
7346 |
|
|
|
7347 |
|
|
(define_split
|
7348 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "")
|
7349 |
|
|
(vec_select:DI
|
7350 |
|
|
(match_operand:V2DI 1 "register_operand" "")
|
7351 |
|
|
(parallel [(const_int 0)])))]
|
7352 |
|
|
"TARGET_SSE
|
7353 |
|
|
&& reload_completed
|
7354 |
|
|
&& (TARGET_INTER_UNIT_MOVES
|
7355 |
|
|
|| MEM_P (operands [0])
|
7356 |
|
|
|| !GENERAL_REGNO_P (true_regnum (operands [0])))"
|
7357 |
|
|
[(set (match_dup 0) (match_dup 1))]
|
7358 |
|
|
{
|
7359 |
|
|
operands[1] = gen_rtx_REG (DImode, REGNO (operands[1]));
|
7360 |
|
|
})
|
7361 |
|
|
|
7362 |
|
|
(define_insn "*vec_extractv2di_1_rex64_avx"
|
7363 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "=m,x,x,r")
|
7364 |
|
|
(vec_select:DI
|
7365 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "x,x,o,o")
|
7366 |
|
|
(parallel [(const_int 1)])))]
|
7367 |
|
|
"TARGET_64BIT
|
7368 |
|
|
&& TARGET_AVX
|
7369 |
|
|
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
7370 |
|
|
"@
|
7371 |
|
|
vmovhps\t{%1, %0|%0, %1}
|
7372 |
|
|
vpsrldq\t{$8, %1, %0|%0, %1, 8}
|
7373 |
|
|
vmovq\t{%H1, %0|%0, %H1}
|
7374 |
|
|
vmov{q}\t{%H1, %0|%0, %H1}"
|
7375 |
|
|
[(set_attr "type" "ssemov,sseishft1,ssemov,imov")
|
7376 |
|
|
(set_attr "length_immediate" "*,1,*,*")
|
7377 |
|
|
(set_attr "memory" "*,none,*,*")
|
7378 |
|
|
(set_attr "prefix" "vex")
|
7379 |
|
|
(set_attr "mode" "V2SF,TI,TI,DI")])
|
7380 |
|
|
|
7381 |
|
|
(define_insn "*vec_extractv2di_1_rex64"
|
7382 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "=m,x,x,r")
|
7383 |
|
|
(vec_select:DI
|
7384 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "x,0,o,o")
|
7385 |
|
|
(parallel [(const_int 1)])))]
|
7386 |
|
|
"TARGET_64BIT && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
7387 |
|
|
"@
|
7388 |
|
|
movhps\t{%1, %0|%0, %1}
|
7389 |
|
|
psrldq\t{$8, %0|%0, 8}
|
7390 |
|
|
movq\t{%H1, %0|%0, %H1}
|
7391 |
|
|
mov{q}\t{%H1, %0|%0, %H1}"
|
7392 |
|
|
[(set_attr "type" "ssemov,sseishft1,ssemov,imov")
|
7393 |
|
|
(set_attr "length_immediate" "*,1,*,*")
|
7394 |
|
|
(set_attr "memory" "*,none,*,*")
|
7395 |
|
|
(set_attr "mode" "V2SF,TI,TI,DI")])
|
7396 |
|
|
|
7397 |
|
|
(define_insn "*vec_extractv2di_1_avx"
|
7398 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "=m,x,x")
|
7399 |
|
|
(vec_select:DI
|
7400 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "x,x,o")
|
7401 |
|
|
(parallel [(const_int 1)])))]
|
7402 |
|
|
"!TARGET_64BIT
|
7403 |
|
|
&& TARGET_AVX
|
7404 |
|
|
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
7405 |
|
|
"@
|
7406 |
|
|
vmovhps\t{%1, %0|%0, %1}
|
7407 |
|
|
vpsrldq\t{$8, %1, %0|%0, %1, 8}
|
7408 |
|
|
vmovq\t{%H1, %0|%0, %H1}"
|
7409 |
|
|
[(set_attr "type" "ssemov,sseishft1,ssemov")
|
7410 |
|
|
(set_attr "length_immediate" "*,1,*")
|
7411 |
|
|
(set_attr "memory" "*,none,*")
|
7412 |
|
|
(set_attr "prefix" "vex")
|
7413 |
|
|
(set_attr "mode" "V2SF,TI,TI")])
|
7414 |
|
|
|
7415 |
|
|
(define_insn "*vec_extractv2di_1_sse2"
|
7416 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "=m,x,x")
|
7417 |
|
|
(vec_select:DI
|
7418 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "x,0,o")
|
7419 |
|
|
(parallel [(const_int 1)])))]
|
7420 |
|
|
"!TARGET_64BIT
|
7421 |
|
|
&& TARGET_SSE2 && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
7422 |
|
|
"@
|
7423 |
|
|
movhps\t{%1, %0|%0, %1}
|
7424 |
|
|
psrldq\t{$8, %0|%0, 8}
|
7425 |
|
|
movq\t{%H1, %0|%0, %H1}"
|
7426 |
|
|
[(set_attr "type" "ssemov,sseishft1,ssemov")
|
7427 |
|
|
(set_attr "length_immediate" "*,1,*")
|
7428 |
|
|
(set_attr "memory" "*,none,*")
|
7429 |
|
|
(set_attr "mode" "V2SF,TI,TI")])
|
7430 |
|
|
|
7431 |
|
|
;; Not sure this is ever used, but it doesn't hurt to have it. -aoliva
|
7432 |
|
|
(define_insn "*vec_extractv2di_1_sse"
|
7433 |
|
|
[(set (match_operand:DI 0 "nonimmediate_operand" "=m,x,x")
|
7434 |
|
|
(vec_select:DI
|
7435 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "x,x,o")
|
7436 |
|
|
(parallel [(const_int 1)])))]
|
7437 |
|
|
"!TARGET_SSE2 && TARGET_SSE
|
7438 |
|
|
&& !(MEM_P (operands[0]) && MEM_P (operands[1]))"
|
7439 |
|
|
"@
|
7440 |
|
|
movhps\t{%1, %0|%0, %1}
|
7441 |
|
|
movhlps\t{%1, %0|%0, %1}
|
7442 |
|
|
movlps\t{%H1, %0|%0, %H1}"
|
7443 |
|
|
[(set_attr "type" "ssemov")
|
7444 |
|
|
(set_attr "mode" "V2SF,V4SF,V2SF")])
|
7445 |
|
|
|
7446 |
|
|
(define_insn "*vec_dupv4si_avx"
|
7447 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x,x")
|
7448 |
|
|
(vec_duplicate:V4SI
|
7449 |
|
|
(match_operand:SI 1 "register_operand" "x,m")))]
|
7450 |
|
|
"TARGET_AVX"
|
7451 |
|
|
"@
|
7452 |
|
|
vpshufd\t{$0, %1, %0|%0, %1, 0}
|
7453 |
|
|
vbroadcastss\t{%1, %0|%0, %1}"
|
7454 |
|
|
[(set_attr "type" "sselog1,ssemov")
|
7455 |
|
|
(set_attr "length_immediate" "1,0")
|
7456 |
|
|
(set_attr "prefix_extra" "0,1")
|
7457 |
|
|
(set_attr "prefix" "vex")
|
7458 |
|
|
(set_attr "mode" "TI,V4SF")])
|
7459 |
|
|
|
7460 |
|
|
(define_insn "*vec_dupv4si"
|
7461 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=Y2,x")
|
7462 |
|
|
(vec_duplicate:V4SI
|
7463 |
|
|
(match_operand:SI 1 "register_operand" " Y2,0")))]
|
7464 |
|
|
"TARGET_SSE"
|
7465 |
|
|
"@
|
7466 |
|
|
%vpshufd\t{$0, %1, %0|%0, %1, 0}
|
7467 |
|
|
shufps\t{$0, %0, %0|%0, %0, 0}"
|
7468 |
|
|
[(set_attr "type" "sselog1")
|
7469 |
|
|
(set_attr "length_immediate" "1")
|
7470 |
|
|
(set_attr "mode" "TI,V4SF")])
|
7471 |
|
|
|
7472 |
|
|
(define_insn "*vec_dupv2di_avx"
|
7473 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x,x")
|
7474 |
|
|
(vec_duplicate:V2DI
|
7475 |
|
|
(match_operand:DI 1 "nonimmediate_operand" " x,m")))]
|
7476 |
|
|
"TARGET_AVX"
|
7477 |
|
|
"@
|
7478 |
|
|
vpunpcklqdq\t{%1, %1, %0|%0, %1, %1}
|
7479 |
|
|
vmovddup\t{%1, %0|%0, %1}"
|
7480 |
|
|
[(set_attr "type" "sselog1")
|
7481 |
|
|
(set_attr "prefix" "vex")
|
7482 |
|
|
(set_attr "mode" "TI,DF")])
|
7483 |
|
|
|
7484 |
|
|
(define_insn "*vec_dupv2di_sse3"
|
7485 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x,x")
|
7486 |
|
|
(vec_duplicate:V2DI
|
7487 |
|
|
(match_operand:DI 1 "nonimmediate_operand" " 0,m")))]
|
7488 |
|
|
"TARGET_SSE3"
|
7489 |
|
|
"@
|
7490 |
|
|
punpcklqdq\t%0, %0
|
7491 |
|
|
movddup\t{%1, %0|%0, %1}"
|
7492 |
|
|
[(set_attr "type" "sselog1")
|
7493 |
|
|
(set_attr "mode" "TI,DF")])
|
7494 |
|
|
|
7495 |
|
|
(define_insn "*vec_dupv2di"
|
7496 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=Y2,x")
|
7497 |
|
|
(vec_duplicate:V2DI
|
7498 |
|
|
(match_operand:DI 1 "register_operand" " 0 ,0")))]
|
7499 |
|
|
"TARGET_SSE"
|
7500 |
|
|
"@
|
7501 |
|
|
punpcklqdq\t%0, %0
|
7502 |
|
|
movlhps\t%0, %0"
|
7503 |
|
|
[(set_attr "type" "sselog1,ssemov")
|
7504 |
|
|
(set_attr "mode" "TI,V4SF")])
|
7505 |
|
|
|
7506 |
|
|
(define_insn "*vec_concatv2si_avx"
|
7507 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=x,x,x ,*y ,*y")
|
7508 |
|
|
(vec_concat:V2SI
|
7509 |
|
|
(match_operand:SI 1 "nonimmediate_operand" "x ,x,rm, 0 ,rm")
|
7510 |
|
|
(match_operand:SI 2 "vector_move_operand" "rm,x,C ,*ym,C")))]
|
7511 |
|
|
"TARGET_AVX"
|
7512 |
|
|
"@
|
7513 |
|
|
vpinsrd\t{$0x1, %2, %1, %0|%0, %1, %2, 0x1}
|
7514 |
|
|
vpunpckldq\t{%2, %1, %0|%0, %1, %2}
|
7515 |
|
|
vmovd\t{%1, %0|%0, %1}
|
7516 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
7517 |
|
|
movd\t{%1, %0|%0, %1}"
|
7518 |
|
|
[(set_attr "type" "sselog,sselog,ssemov,mmxcvt,mmxmov")
|
7519 |
|
|
(set_attr "prefix_extra" "1,*,*,*,*")
|
7520 |
|
|
(set_attr "length_immediate" "1,*,*,*,*")
|
7521 |
|
|
(set (attr "prefix")
|
7522 |
|
|
(if_then_else (eq_attr "alternative" "3,4")
|
7523 |
|
|
(const_string "orig")
|
7524 |
|
|
(const_string "vex")))
|
7525 |
|
|
(set_attr "mode" "TI,TI,TI,DI,DI")])
|
7526 |
|
|
|
7527 |
|
|
(define_insn "*vec_concatv2si_sse4_1"
|
7528 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=x,x,x ,*y ,*y")
|
7529 |
|
|
(vec_concat:V2SI
|
7530 |
|
|
(match_operand:SI 1 "nonimmediate_operand" "0 ,0,rm, 0 ,rm")
|
7531 |
|
|
(match_operand:SI 2 "vector_move_operand" "rm,x,C ,*ym,C")))]
|
7532 |
|
|
"TARGET_SSE4_1"
|
7533 |
|
|
"@
|
7534 |
|
|
pinsrd\t{$0x1, %2, %0|%0, %2, 0x1}
|
7535 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
7536 |
|
|
movd\t{%1, %0|%0, %1}
|
7537 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
7538 |
|
|
movd\t{%1, %0|%0, %1}"
|
7539 |
|
|
[(set_attr "type" "sselog,sselog,ssemov,mmxcvt,mmxmov")
|
7540 |
|
|
(set_attr "prefix_extra" "1,*,*,*,*")
|
7541 |
|
|
(set_attr "length_immediate" "1,*,*,*,*")
|
7542 |
|
|
(set_attr "mode" "TI,TI,TI,DI,DI")])
|
7543 |
|
|
|
7544 |
|
|
;; ??? In theory we can match memory for the MMX alternative, but allowing
|
7545 |
|
|
;; nonimmediate_operand for operand 2 and *not* allowing memory for the SSE
|
7546 |
|
|
;; alternatives pretty much forces the MMX alternative to be chosen.
|
7547 |
|
|
(define_insn "*vec_concatv2si_sse2"
|
7548 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=x,x ,*y,*y")
|
7549 |
|
|
(vec_concat:V2SI
|
7550 |
|
|
(match_operand:SI 1 "nonimmediate_operand" " 0,rm, 0,rm")
|
7551 |
|
|
(match_operand:SI 2 "reg_or_0_operand" " x,C ,*y, C")))]
|
7552 |
|
|
"TARGET_SSE2"
|
7553 |
|
|
"@
|
7554 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
7555 |
|
|
movd\t{%1, %0|%0, %1}
|
7556 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
7557 |
|
|
movd\t{%1, %0|%0, %1}"
|
7558 |
|
|
[(set_attr "type" "sselog,ssemov,mmxcvt,mmxmov")
|
7559 |
|
|
(set_attr "mode" "TI,TI,DI,DI")])
|
7560 |
|
|
|
7561 |
|
|
(define_insn "*vec_concatv2si_sse"
|
7562 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=x,x,*y,*y")
|
7563 |
|
|
(vec_concat:V2SI
|
7564 |
|
|
(match_operand:SI 1 "nonimmediate_operand" " 0,m, 0,*rm")
|
7565 |
|
|
(match_operand:SI 2 "reg_or_0_operand" " x,C,*y,C")))]
|
7566 |
|
|
"TARGET_SSE"
|
7567 |
|
|
"@
|
7568 |
|
|
unpcklps\t{%2, %0|%0, %2}
|
7569 |
|
|
movss\t{%1, %0|%0, %1}
|
7570 |
|
|
punpckldq\t{%2, %0|%0, %2}
|
7571 |
|
|
movd\t{%1, %0|%0, %1}"
|
7572 |
|
|
[(set_attr "type" "sselog,ssemov,mmxcvt,mmxmov")
|
7573 |
|
|
(set_attr "mode" "V4SF,V4SF,DI,DI")])
|
7574 |
|
|
|
7575 |
|
|
(define_insn "*vec_concatv4si_1_avx"
|
7576 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x,x")
|
7577 |
|
|
(vec_concat:V4SI
|
7578 |
|
|
(match_operand:V2SI 1 "register_operand" " x,x")
|
7579 |
|
|
(match_operand:V2SI 2 "nonimmediate_operand" " x,m")))]
|
7580 |
|
|
"TARGET_AVX"
|
7581 |
|
|
"@
|
7582 |
|
|
vpunpcklqdq\t{%2, %1, %0|%0, %1, %2}
|
7583 |
|
|
vmovhps\t{%2, %1, %0|%0, %1, %2}"
|
7584 |
|
|
[(set_attr "type" "sselog,ssemov")
|
7585 |
|
|
(set_attr "prefix" "vex")
|
7586 |
|
|
(set_attr "mode" "TI,V2SF")])
|
7587 |
|
|
|
7588 |
|
|
(define_insn "*vec_concatv4si_1"
|
7589 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=Y2,x,x")
|
7590 |
|
|
(vec_concat:V4SI
|
7591 |
|
|
(match_operand:V2SI 1 "register_operand" " 0 ,0,0")
|
7592 |
|
|
(match_operand:V2SI 2 "nonimmediate_operand" " Y2,x,m")))]
|
7593 |
|
|
"TARGET_SSE"
|
7594 |
|
|
"@
|
7595 |
|
|
punpcklqdq\t{%2, %0|%0, %2}
|
7596 |
|
|
movlhps\t{%2, %0|%0, %2}
|
7597 |
|
|
movhps\t{%2, %0|%0, %2}"
|
7598 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov")
|
7599 |
|
|
(set_attr "mode" "TI,V4SF,V2SF")])
|
7600 |
|
|
|
7601 |
|
|
(define_insn "*vec_concatv2di_avx"
|
7602 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x,?x,x,x")
|
7603 |
|
|
(vec_concat:V2DI
|
7604 |
|
|
(match_operand:DI 1 "nonimmediate_operand" " m,*y,x,x")
|
7605 |
|
|
(match_operand:DI 2 "vector_move_operand" " C, C,x,m")))]
|
7606 |
|
|
"!TARGET_64BIT && TARGET_AVX"
|
7607 |
|
|
"@
|
7608 |
|
|
vmovq\t{%1, %0|%0, %1}
|
7609 |
|
|
movq2dq\t{%1, %0|%0, %1}
|
7610 |
|
|
vpunpcklqdq\t{%2, %1, %0|%0, %1, %2}
|
7611 |
|
|
vmovhps\t{%2, %1, %0|%0, %1, %2}"
|
7612 |
|
|
[(set_attr "type" "ssemov,ssemov,sselog,ssemov")
|
7613 |
|
|
(set (attr "prefix")
|
7614 |
|
|
(if_then_else (eq_attr "alternative" "1")
|
7615 |
|
|
(const_string "orig")
|
7616 |
|
|
(const_string "vex")))
|
7617 |
|
|
(set_attr "mode" "TI,TI,TI,V2SF")])
|
7618 |
|
|
|
7619 |
|
|
(define_insn "vec_concatv2di"
|
7620 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=Y2 ,?Y2,Y2,x,x")
|
7621 |
|
|
(vec_concat:V2DI
|
7622 |
|
|
(match_operand:DI 1 "nonimmediate_operand" " mY2,*y ,0 ,0,0")
|
7623 |
|
|
(match_operand:DI 2 "vector_move_operand" " C , C,Y2,x,m")))]
|
7624 |
|
|
"!TARGET_64BIT && TARGET_SSE"
|
7625 |
|
|
"@
|
7626 |
|
|
movq\t{%1, %0|%0, %1}
|
7627 |
|
|
movq2dq\t{%1, %0|%0, %1}
|
7628 |
|
|
punpcklqdq\t{%2, %0|%0, %2}
|
7629 |
|
|
movlhps\t{%2, %0|%0, %2}
|
7630 |
|
|
movhps\t{%2, %0|%0, %2}"
|
7631 |
|
|
[(set_attr "type" "ssemov,ssemov,sselog,ssemov,ssemov")
|
7632 |
|
|
(set_attr "mode" "TI,TI,TI,V4SF,V2SF")])
|
7633 |
|
|
|
7634 |
|
|
(define_insn "*vec_concatv2di_rex64_avx"
|
7635 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x,x,Yi,!x,x,x")
|
7636 |
|
|
(vec_concat:V2DI
|
7637 |
|
|
(match_operand:DI 1 "nonimmediate_operand" " x,m,r ,*y,x,x")
|
7638 |
|
|
(match_operand:DI 2 "vector_move_operand" "rm,C,C ,C ,x,m")))]
|
7639 |
|
|
"TARGET_64BIT && TARGET_AVX"
|
7640 |
|
|
"@
|
7641 |
|
|
vpinsrq\t{$0x1, %2, %1, %0|%0, %1, %2, 0x1}
|
7642 |
|
|
vmovq\t{%1, %0|%0, %1}
|
7643 |
|
|
vmovq\t{%1, %0|%0, %1}
|
7644 |
|
|
movq2dq\t{%1, %0|%0, %1}
|
7645 |
|
|
vpunpcklqdq\t{%2, %1, %0|%0, %1, %2}
|
7646 |
|
|
vmovhps\t{%2, %1, %0|%0, %1, %2}"
|
7647 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov,ssemov,sselog,ssemov")
|
7648 |
|
|
(set_attr "prefix_extra" "1,*,*,*,*,*")
|
7649 |
|
|
(set_attr "length_immediate" "1,*,*,*,*,*")
|
7650 |
|
|
(set (attr "prefix")
|
7651 |
|
|
(if_then_else (eq_attr "alternative" "3")
|
7652 |
|
|
(const_string "orig")
|
7653 |
|
|
(const_string "vex")))
|
7654 |
|
|
(set_attr "mode" "TI,TI,TI,TI,TI,V2SF")])
|
7655 |
|
|
|
7656 |
|
|
(define_insn "*vec_concatv2di_rex64_sse4_1"
|
7657 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x ,x ,Yi,!x,x,x,x")
|
7658 |
|
|
(vec_concat:V2DI
|
7659 |
|
|
(match_operand:DI 1 "nonimmediate_operand" " 0 ,mx,r ,*y,0,0,0")
|
7660 |
|
|
(match_operand:DI 2 "vector_move_operand" " rm,C ,C ,C ,x,x,m")))]
|
7661 |
|
|
"TARGET_64BIT && TARGET_SSE4_1"
|
7662 |
|
|
"@
|
7663 |
|
|
pinsrq\t{$0x1, %2, %0|%0, %2, 0x1}
|
7664 |
|
|
movq\t{%1, %0|%0, %1}
|
7665 |
|
|
movq\t{%1, %0|%0, %1}
|
7666 |
|
|
movq2dq\t{%1, %0|%0, %1}
|
7667 |
|
|
punpcklqdq\t{%2, %0|%0, %2}
|
7668 |
|
|
movlhps\t{%2, %0|%0, %2}
|
7669 |
|
|
movhps\t{%2, %0|%0, %2}"
|
7670 |
|
|
[(set_attr "type" "sselog,ssemov,ssemov,ssemov,sselog,ssemov,ssemov")
|
7671 |
|
|
(set_attr "prefix_rex" "1,*,1,*,*,*,*")
|
7672 |
|
|
(set_attr "prefix_extra" "1,*,*,*,*,*,*")
|
7673 |
|
|
(set_attr "length_immediate" "1,*,*,*,*,*,*")
|
7674 |
|
|
(set_attr "mode" "TI,TI,TI,TI,TI,V4SF,V2SF")])
|
7675 |
|
|
|
7676 |
|
|
(define_insn "*vec_concatv2di_rex64_sse"
|
7677 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=Y2 ,Yi,!Y2,Y2,x,x")
|
7678 |
|
|
(vec_concat:V2DI
|
7679 |
|
|
(match_operand:DI 1 "nonimmediate_operand" " mY2,r ,*y ,0 ,0,0")
|
7680 |
|
|
(match_operand:DI 2 "vector_move_operand" " C ,C ,C ,Y2,x,m")))]
|
7681 |
|
|
"TARGET_64BIT && TARGET_SSE"
|
7682 |
|
|
"@
|
7683 |
|
|
movq\t{%1, %0|%0, %1}
|
7684 |
|
|
movq\t{%1, %0|%0, %1}
|
7685 |
|
|
movq2dq\t{%1, %0|%0, %1}
|
7686 |
|
|
punpcklqdq\t{%2, %0|%0, %2}
|
7687 |
|
|
movlhps\t{%2, %0|%0, %2}
|
7688 |
|
|
movhps\t{%2, %0|%0, %2}"
|
7689 |
|
|
[(set_attr "type" "ssemov,ssemov,ssemov,sselog,ssemov,ssemov")
|
7690 |
|
|
(set_attr "prefix_rex" "*,1,*,*,*,*")
|
7691 |
|
|
(set_attr "mode" "TI,TI,TI,TI,V4SF,V2SF")])
|
7692 |
|
|
|
7693 |
|
|
(define_expand "vec_unpacku_hi_v16qi"
|
7694 |
|
|
[(match_operand:V8HI 0 "register_operand" "")
|
7695 |
|
|
(match_operand:V16QI 1 "register_operand" "")]
|
7696 |
|
|
"TARGET_SSE2"
|
7697 |
|
|
{
|
7698 |
|
|
if (TARGET_SSE4_1)
|
7699 |
|
|
ix86_expand_sse4_unpack (operands, true, true);
|
7700 |
|
|
else
|
7701 |
|
|
ix86_expand_sse_unpack (operands, true, true);
|
7702 |
|
|
DONE;
|
7703 |
|
|
})
|
7704 |
|
|
|
7705 |
|
|
(define_expand "vec_unpacks_hi_v16qi"
|
7706 |
|
|
[(match_operand:V8HI 0 "register_operand" "")
|
7707 |
|
|
(match_operand:V16QI 1 "register_operand" "")]
|
7708 |
|
|
"TARGET_SSE2"
|
7709 |
|
|
{
|
7710 |
|
|
if (TARGET_SSE4_1)
|
7711 |
|
|
ix86_expand_sse4_unpack (operands, false, true);
|
7712 |
|
|
else
|
7713 |
|
|
ix86_expand_sse_unpack (operands, false, true);
|
7714 |
|
|
DONE;
|
7715 |
|
|
})
|
7716 |
|
|
|
7717 |
|
|
(define_expand "vec_unpacku_lo_v16qi"
|
7718 |
|
|
[(match_operand:V8HI 0 "register_operand" "")
|
7719 |
|
|
(match_operand:V16QI 1 "register_operand" "")]
|
7720 |
|
|
"TARGET_SSE2"
|
7721 |
|
|
{
|
7722 |
|
|
if (TARGET_SSE4_1)
|
7723 |
|
|
ix86_expand_sse4_unpack (operands, true, false);
|
7724 |
|
|
else
|
7725 |
|
|
ix86_expand_sse_unpack (operands, true, false);
|
7726 |
|
|
DONE;
|
7727 |
|
|
})
|
7728 |
|
|
|
7729 |
|
|
(define_expand "vec_unpacks_lo_v16qi"
|
7730 |
|
|
[(match_operand:V8HI 0 "register_operand" "")
|
7731 |
|
|
(match_operand:V16QI 1 "register_operand" "")]
|
7732 |
|
|
"TARGET_SSE2"
|
7733 |
|
|
{
|
7734 |
|
|
if (TARGET_SSE4_1)
|
7735 |
|
|
ix86_expand_sse4_unpack (operands, false, false);
|
7736 |
|
|
else
|
7737 |
|
|
ix86_expand_sse_unpack (operands, false, false);
|
7738 |
|
|
DONE;
|
7739 |
|
|
})
|
7740 |
|
|
|
7741 |
|
|
(define_expand "vec_unpacku_hi_v8hi"
|
7742 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
7743 |
|
|
(match_operand:V8HI 1 "register_operand" "")]
|
7744 |
|
|
"TARGET_SSE2"
|
7745 |
|
|
{
|
7746 |
|
|
if (TARGET_SSE4_1)
|
7747 |
|
|
ix86_expand_sse4_unpack (operands, true, true);
|
7748 |
|
|
else
|
7749 |
|
|
ix86_expand_sse_unpack (operands, true, true);
|
7750 |
|
|
DONE;
|
7751 |
|
|
})
|
7752 |
|
|
|
7753 |
|
|
(define_expand "vec_unpacks_hi_v8hi"
|
7754 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
7755 |
|
|
(match_operand:V8HI 1 "register_operand" "")]
|
7756 |
|
|
"TARGET_SSE2"
|
7757 |
|
|
{
|
7758 |
|
|
if (TARGET_SSE4_1)
|
7759 |
|
|
ix86_expand_sse4_unpack (operands, false, true);
|
7760 |
|
|
else
|
7761 |
|
|
ix86_expand_sse_unpack (operands, false, true);
|
7762 |
|
|
DONE;
|
7763 |
|
|
})
|
7764 |
|
|
|
7765 |
|
|
(define_expand "vec_unpacku_lo_v8hi"
|
7766 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
7767 |
|
|
(match_operand:V8HI 1 "register_operand" "")]
|
7768 |
|
|
"TARGET_SSE2"
|
7769 |
|
|
{
|
7770 |
|
|
if (TARGET_SSE4_1)
|
7771 |
|
|
ix86_expand_sse4_unpack (operands, true, false);
|
7772 |
|
|
else
|
7773 |
|
|
ix86_expand_sse_unpack (operands, true, false);
|
7774 |
|
|
DONE;
|
7775 |
|
|
})
|
7776 |
|
|
|
7777 |
|
|
(define_expand "vec_unpacks_lo_v8hi"
|
7778 |
|
|
[(match_operand:V4SI 0 "register_operand" "")
|
7779 |
|
|
(match_operand:V8HI 1 "register_operand" "")]
|
7780 |
|
|
"TARGET_SSE2"
|
7781 |
|
|
{
|
7782 |
|
|
if (TARGET_SSE4_1)
|
7783 |
|
|
ix86_expand_sse4_unpack (operands, false, false);
|
7784 |
|
|
else
|
7785 |
|
|
ix86_expand_sse_unpack (operands, false, false);
|
7786 |
|
|
DONE;
|
7787 |
|
|
})
|
7788 |
|
|
|
7789 |
|
|
(define_expand "vec_unpacku_hi_v4si"
|
7790 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
7791 |
|
|
(match_operand:V4SI 1 "register_operand" "")]
|
7792 |
|
|
"TARGET_SSE2"
|
7793 |
|
|
{
|
7794 |
|
|
if (TARGET_SSE4_1)
|
7795 |
|
|
ix86_expand_sse4_unpack (operands, true, true);
|
7796 |
|
|
else
|
7797 |
|
|
ix86_expand_sse_unpack (operands, true, true);
|
7798 |
|
|
DONE;
|
7799 |
|
|
})
|
7800 |
|
|
|
7801 |
|
|
(define_expand "vec_unpacks_hi_v4si"
|
7802 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
7803 |
|
|
(match_operand:V4SI 1 "register_operand" "")]
|
7804 |
|
|
"TARGET_SSE2"
|
7805 |
|
|
{
|
7806 |
|
|
if (TARGET_SSE4_1)
|
7807 |
|
|
ix86_expand_sse4_unpack (operands, false, true);
|
7808 |
|
|
else
|
7809 |
|
|
ix86_expand_sse_unpack (operands, false, true);
|
7810 |
|
|
DONE;
|
7811 |
|
|
})
|
7812 |
|
|
|
7813 |
|
|
(define_expand "vec_unpacku_lo_v4si"
|
7814 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
7815 |
|
|
(match_operand:V4SI 1 "register_operand" "")]
|
7816 |
|
|
"TARGET_SSE2"
|
7817 |
|
|
{
|
7818 |
|
|
if (TARGET_SSE4_1)
|
7819 |
|
|
ix86_expand_sse4_unpack (operands, true, false);
|
7820 |
|
|
else
|
7821 |
|
|
ix86_expand_sse_unpack (operands, true, false);
|
7822 |
|
|
DONE;
|
7823 |
|
|
})
|
7824 |
|
|
|
7825 |
|
|
(define_expand "vec_unpacks_lo_v4si"
|
7826 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
7827 |
|
|
(match_operand:V4SI 1 "register_operand" "")]
|
7828 |
|
|
"TARGET_SSE2"
|
7829 |
|
|
{
|
7830 |
|
|
if (TARGET_SSE4_1)
|
7831 |
|
|
ix86_expand_sse4_unpack (operands, false, false);
|
7832 |
|
|
else
|
7833 |
|
|
ix86_expand_sse_unpack (operands, false, false);
|
7834 |
|
|
DONE;
|
7835 |
|
|
})
|
7836 |
|
|
|
7837 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
7838 |
|
|
;;
|
7839 |
|
|
;; Miscellaneous
|
7840 |
|
|
;;
|
7841 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
7842 |
|
|
|
7843 |
|
|
(define_expand "sse2_uavgv16qi3"
|
7844 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "")
|
7845 |
|
|
(truncate:V16QI
|
7846 |
|
|
(lshiftrt:V16HI
|
7847 |
|
|
(plus:V16HI
|
7848 |
|
|
(plus:V16HI
|
7849 |
|
|
(zero_extend:V16HI
|
7850 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" ""))
|
7851 |
|
|
(zero_extend:V16HI
|
7852 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "")))
|
7853 |
|
|
(const_vector:V16QI [(const_int 1) (const_int 1)
|
7854 |
|
|
(const_int 1) (const_int 1)
|
7855 |
|
|
(const_int 1) (const_int 1)
|
7856 |
|
|
(const_int 1) (const_int 1)
|
7857 |
|
|
(const_int 1) (const_int 1)
|
7858 |
|
|
(const_int 1) (const_int 1)
|
7859 |
|
|
(const_int 1) (const_int 1)
|
7860 |
|
|
(const_int 1) (const_int 1)]))
|
7861 |
|
|
(const_int 1))))]
|
7862 |
|
|
"TARGET_SSE2"
|
7863 |
|
|
"ix86_fixup_binary_operands_no_copy (PLUS, V16QImode, operands);")
|
7864 |
|
|
|
7865 |
|
|
(define_insn "*avx_uavgv16qi3"
|
7866 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
7867 |
|
|
(truncate:V16QI
|
7868 |
|
|
(lshiftrt:V16HI
|
7869 |
|
|
(plus:V16HI
|
7870 |
|
|
(plus:V16HI
|
7871 |
|
|
(zero_extend:V16HI
|
7872 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "%x"))
|
7873 |
|
|
(zero_extend:V16HI
|
7874 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")))
|
7875 |
|
|
(const_vector:V16QI [(const_int 1) (const_int 1)
|
7876 |
|
|
(const_int 1) (const_int 1)
|
7877 |
|
|
(const_int 1) (const_int 1)
|
7878 |
|
|
(const_int 1) (const_int 1)
|
7879 |
|
|
(const_int 1) (const_int 1)
|
7880 |
|
|
(const_int 1) (const_int 1)
|
7881 |
|
|
(const_int 1) (const_int 1)
|
7882 |
|
|
(const_int 1) (const_int 1)]))
|
7883 |
|
|
(const_int 1))))]
|
7884 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (PLUS, V16QImode, operands)"
|
7885 |
|
|
"vpavgb\t{%2, %1, %0|%0, %1, %2}"
|
7886 |
|
|
[(set_attr "type" "sseiadd")
|
7887 |
|
|
(set_attr "prefix" "vex")
|
7888 |
|
|
(set_attr "mode" "TI")])
|
7889 |
|
|
|
7890 |
|
|
(define_insn "*sse2_uavgv16qi3"
|
7891 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
7892 |
|
|
(truncate:V16QI
|
7893 |
|
|
(lshiftrt:V16HI
|
7894 |
|
|
(plus:V16HI
|
7895 |
|
|
(plus:V16HI
|
7896 |
|
|
(zero_extend:V16HI
|
7897 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "%0"))
|
7898 |
|
|
(zero_extend:V16HI
|
7899 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")))
|
7900 |
|
|
(const_vector:V16QI [(const_int 1) (const_int 1)
|
7901 |
|
|
(const_int 1) (const_int 1)
|
7902 |
|
|
(const_int 1) (const_int 1)
|
7903 |
|
|
(const_int 1) (const_int 1)
|
7904 |
|
|
(const_int 1) (const_int 1)
|
7905 |
|
|
(const_int 1) (const_int 1)
|
7906 |
|
|
(const_int 1) (const_int 1)
|
7907 |
|
|
(const_int 1) (const_int 1)]))
|
7908 |
|
|
(const_int 1))))]
|
7909 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (PLUS, V16QImode, operands)"
|
7910 |
|
|
"pavgb\t{%2, %0|%0, %2}"
|
7911 |
|
|
[(set_attr "type" "sseiadd")
|
7912 |
|
|
(set_attr "prefix_data16" "1")
|
7913 |
|
|
(set_attr "mode" "TI")])
|
7914 |
|
|
|
7915 |
|
|
(define_expand "sse2_uavgv8hi3"
|
7916 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "")
|
7917 |
|
|
(truncate:V8HI
|
7918 |
|
|
(lshiftrt:V8SI
|
7919 |
|
|
(plus:V8SI
|
7920 |
|
|
(plus:V8SI
|
7921 |
|
|
(zero_extend:V8SI
|
7922 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" ""))
|
7923 |
|
|
(zero_extend:V8SI
|
7924 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "")))
|
7925 |
|
|
(const_vector:V8HI [(const_int 1) (const_int 1)
|
7926 |
|
|
(const_int 1) (const_int 1)
|
7927 |
|
|
(const_int 1) (const_int 1)
|
7928 |
|
|
(const_int 1) (const_int 1)]))
|
7929 |
|
|
(const_int 1))))]
|
7930 |
|
|
"TARGET_SSE2"
|
7931 |
|
|
"ix86_fixup_binary_operands_no_copy (PLUS, V8HImode, operands);")
|
7932 |
|
|
|
7933 |
|
|
(define_insn "*avx_uavgv8hi3"
|
7934 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
7935 |
|
|
(truncate:V8HI
|
7936 |
|
|
(lshiftrt:V8SI
|
7937 |
|
|
(plus:V8SI
|
7938 |
|
|
(plus:V8SI
|
7939 |
|
|
(zero_extend:V8SI
|
7940 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x"))
|
7941 |
|
|
(zero_extend:V8SI
|
7942 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))
|
7943 |
|
|
(const_vector:V8HI [(const_int 1) (const_int 1)
|
7944 |
|
|
(const_int 1) (const_int 1)
|
7945 |
|
|
(const_int 1) (const_int 1)
|
7946 |
|
|
(const_int 1) (const_int 1)]))
|
7947 |
|
|
(const_int 1))))]
|
7948 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (PLUS, V8HImode, operands)"
|
7949 |
|
|
"vpavgw\t{%2, %1, %0|%0, %1, %2}"
|
7950 |
|
|
[(set_attr "type" "sseiadd")
|
7951 |
|
|
(set_attr "prefix" "vex")
|
7952 |
|
|
(set_attr "mode" "TI")])
|
7953 |
|
|
|
7954 |
|
|
(define_insn "*sse2_uavgv8hi3"
|
7955 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
7956 |
|
|
(truncate:V8HI
|
7957 |
|
|
(lshiftrt:V8SI
|
7958 |
|
|
(plus:V8SI
|
7959 |
|
|
(plus:V8SI
|
7960 |
|
|
(zero_extend:V8SI
|
7961 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%0"))
|
7962 |
|
|
(zero_extend:V8SI
|
7963 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))
|
7964 |
|
|
(const_vector:V8HI [(const_int 1) (const_int 1)
|
7965 |
|
|
(const_int 1) (const_int 1)
|
7966 |
|
|
(const_int 1) (const_int 1)
|
7967 |
|
|
(const_int 1) (const_int 1)]))
|
7968 |
|
|
(const_int 1))))]
|
7969 |
|
|
"TARGET_SSE2 && ix86_binary_operator_ok (PLUS, V8HImode, operands)"
|
7970 |
|
|
"pavgw\t{%2, %0|%0, %2}"
|
7971 |
|
|
[(set_attr "type" "sseiadd")
|
7972 |
|
|
(set_attr "prefix_data16" "1")
|
7973 |
|
|
(set_attr "mode" "TI")])
|
7974 |
|
|
|
7975 |
|
|
;; The correct representation for this is absolutely enormous, and
|
7976 |
|
|
;; surely not generally useful.
|
7977 |
|
|
(define_insn "*avx_psadbw"
|
7978 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
7979 |
|
|
(unspec:V2DI [(match_operand:V16QI 1 "register_operand" "x")
|
7980 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")]
|
7981 |
|
|
UNSPEC_PSADBW))]
|
7982 |
|
|
"TARGET_AVX"
|
7983 |
|
|
"vpsadbw\t{%2, %1, %0|%0, %1, %2}"
|
7984 |
|
|
[(set_attr "type" "sseiadd")
|
7985 |
|
|
(set_attr "prefix" "vex")
|
7986 |
|
|
(set_attr "mode" "TI")])
|
7987 |
|
|
|
7988 |
|
|
(define_insn "sse2_psadbw"
|
7989 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
7990 |
|
|
(unspec:V2DI [(match_operand:V16QI 1 "register_operand" "0")
|
7991 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")]
|
7992 |
|
|
UNSPEC_PSADBW))]
|
7993 |
|
|
"TARGET_SSE2"
|
7994 |
|
|
"psadbw\t{%2, %0|%0, %2}"
|
7995 |
|
|
[(set_attr "type" "sseiadd")
|
7996 |
|
|
(set_attr "atom_unit" "simul")
|
7997 |
|
|
(set_attr "prefix_data16" "1")
|
7998 |
|
|
(set_attr "mode" "TI")])
|
7999 |
|
|
|
8000 |
|
|
(define_insn "avx_movmskp256"
|
8001 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r")
|
8002 |
|
|
(unspec:SI
|
8003 |
|
|
[(match_operand:AVX256MODEF2P 1 "register_operand" "x")]
|
8004 |
|
|
UNSPEC_MOVMSK))]
|
8005 |
|
|
"AVX256_VEC_FLOAT_MODE_P (mode)"
|
8006 |
|
|
"vmovmskp\t{%1, %0|%0, %1}"
|
8007 |
|
|
[(set_attr "type" "ssecvt")
|
8008 |
|
|
(set_attr "prefix" "vex")
|
8009 |
|
|
(set_attr "mode" "")])
|
8010 |
|
|
|
8011 |
|
|
(define_insn "_movmskp"
|
8012 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r")
|
8013 |
|
|
(unspec:SI
|
8014 |
|
|
[(match_operand:SSEMODEF2P 1 "register_operand" "x")]
|
8015 |
|
|
UNSPEC_MOVMSK))]
|
8016 |
|
|
"SSE_VEC_FLOAT_MODE_P (mode)"
|
8017 |
|
|
"%vmovmskp\t{%1, %0|%0, %1}"
|
8018 |
|
|
[(set_attr "type" "ssemov")
|
8019 |
|
|
(set_attr "prefix" "maybe_vex")
|
8020 |
|
|
(set_attr "mode" "")])
|
8021 |
|
|
|
8022 |
|
|
(define_insn "sse2_pmovmskb"
|
8023 |
|
|
[(set (match_operand:SI 0 "register_operand" "=r")
|
8024 |
|
|
(unspec:SI [(match_operand:V16QI 1 "register_operand" "x")]
|
8025 |
|
|
UNSPEC_MOVMSK))]
|
8026 |
|
|
"TARGET_SSE2"
|
8027 |
|
|
"%vpmovmskb\t{%1, %0|%0, %1}"
|
8028 |
|
|
[(set_attr "type" "ssemov")
|
8029 |
|
|
(set_attr "prefix_data16" "1")
|
8030 |
|
|
(set_attr "prefix" "maybe_vex")
|
8031 |
|
|
(set_attr "mode" "SI")])
|
8032 |
|
|
|
8033 |
|
|
(define_expand "sse2_maskmovdqu"
|
8034 |
|
|
[(set (match_operand:V16QI 0 "memory_operand" "")
|
8035 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "")
|
8036 |
|
|
(match_operand:V16QI 2 "register_operand" "")
|
8037 |
|
|
(match_dup 0)]
|
8038 |
|
|
UNSPEC_MASKMOV))]
|
8039 |
|
|
"TARGET_SSE2"
|
8040 |
|
|
"")
|
8041 |
|
|
|
8042 |
|
|
(define_insn "*sse2_maskmovdqu"
|
8043 |
|
|
[(set (mem:V16QI (match_operand:SI 0 "register_operand" "D"))
|
8044 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "x")
|
8045 |
|
|
(match_operand:V16QI 2 "register_operand" "x")
|
8046 |
|
|
(mem:V16QI (match_dup 0))]
|
8047 |
|
|
UNSPEC_MASKMOV))]
|
8048 |
|
|
"TARGET_SSE2 && !TARGET_64BIT"
|
8049 |
|
|
;; @@@ check ordering of operands in intel/nonintel syntax
|
8050 |
|
|
"%vmaskmovdqu\t{%2, %1|%1, %2}"
|
8051 |
|
|
[(set_attr "type" "ssemov")
|
8052 |
|
|
(set_attr "prefix_data16" "1")
|
8053 |
|
|
;; The implicit %rdi operand confuses default length_vex computation.
|
8054 |
|
|
(set_attr "length_vex" "3")
|
8055 |
|
|
(set_attr "prefix" "maybe_vex")
|
8056 |
|
|
(set_attr "mode" "TI")])
|
8057 |
|
|
|
8058 |
|
|
(define_insn "*sse2_maskmovdqu_rex64"
|
8059 |
|
|
[(set (mem:V16QI (match_operand:DI 0 "register_operand" "D"))
|
8060 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "x")
|
8061 |
|
|
(match_operand:V16QI 2 "register_operand" "x")
|
8062 |
|
|
(mem:V16QI (match_dup 0))]
|
8063 |
|
|
UNSPEC_MASKMOV))]
|
8064 |
|
|
"TARGET_SSE2 && TARGET_64BIT"
|
8065 |
|
|
;; @@@ check ordering of operands in intel/nonintel syntax
|
8066 |
|
|
"%vmaskmovdqu\t{%2, %1|%1, %2}"
|
8067 |
|
|
[(set_attr "type" "ssemov")
|
8068 |
|
|
(set_attr "prefix_data16" "1")
|
8069 |
|
|
;; The implicit %rdi operand confuses default length_vex computation.
|
8070 |
|
|
(set (attr "length_vex")
|
8071 |
|
|
(symbol_ref ("REGNO (operands[2]) >= FIRST_REX_SSE_REG ? 3 + 1 : 2 + 1")))
|
8072 |
|
|
(set_attr "prefix" "maybe_vex")
|
8073 |
|
|
(set_attr "mode" "TI")])
|
8074 |
|
|
|
8075 |
|
|
(define_insn "sse_ldmxcsr"
|
8076 |
|
|
[(unspec_volatile [(match_operand:SI 0 "memory_operand" "m")]
|
8077 |
|
|
UNSPECV_LDMXCSR)]
|
8078 |
|
|
"TARGET_SSE"
|
8079 |
|
|
"%vldmxcsr\t%0"
|
8080 |
|
|
[(set_attr "type" "sse")
|
8081 |
|
|
(set_attr "atom_sse_attr" "mxcsr")
|
8082 |
|
|
(set_attr "prefix" "maybe_vex")
|
8083 |
|
|
(set_attr "memory" "load")])
|
8084 |
|
|
|
8085 |
|
|
(define_insn "sse_stmxcsr"
|
8086 |
|
|
[(set (match_operand:SI 0 "memory_operand" "=m")
|
8087 |
|
|
(unspec_volatile:SI [(const_int 0)] UNSPECV_STMXCSR))]
|
8088 |
|
|
"TARGET_SSE"
|
8089 |
|
|
"%vstmxcsr\t%0"
|
8090 |
|
|
[(set_attr "type" "sse")
|
8091 |
|
|
(set_attr "atom_sse_attr" "mxcsr")
|
8092 |
|
|
(set_attr "prefix" "maybe_vex")
|
8093 |
|
|
(set_attr "memory" "store")])
|
8094 |
|
|
|
8095 |
|
|
(define_expand "sse_sfence"
|
8096 |
|
|
[(set (match_dup 0)
|
8097 |
|
|
(unspec:BLK [(match_dup 0)] UNSPEC_SFENCE))]
|
8098 |
|
|
"TARGET_SSE || TARGET_3DNOW_A"
|
8099 |
|
|
{
|
8100 |
|
|
operands[0] = gen_rtx_MEM (BLKmode, gen_rtx_SCRATCH (Pmode));
|
8101 |
|
|
MEM_VOLATILE_P (operands[0]) = 1;
|
8102 |
|
|
})
|
8103 |
|
|
|
8104 |
|
|
(define_insn "*sse_sfence"
|
8105 |
|
|
[(set (match_operand:BLK 0 "" "")
|
8106 |
|
|
(unspec:BLK [(match_dup 0)] UNSPEC_SFENCE))]
|
8107 |
|
|
"TARGET_SSE || TARGET_3DNOW_A"
|
8108 |
|
|
"sfence"
|
8109 |
|
|
[(set_attr "type" "sse")
|
8110 |
|
|
(set_attr "length_address" "0")
|
8111 |
|
|
(set_attr "atom_sse_attr" "fence")
|
8112 |
|
|
(set_attr "memory" "unknown")])
|
8113 |
|
|
|
8114 |
|
|
(define_insn "sse2_clflush"
|
8115 |
|
|
[(unspec_volatile [(match_operand 0 "address_operand" "p")]
|
8116 |
|
|
UNSPECV_CLFLUSH)]
|
8117 |
|
|
"TARGET_SSE2"
|
8118 |
|
|
"clflush\t%a0"
|
8119 |
|
|
[(set_attr "type" "sse")
|
8120 |
|
|
(set_attr "atom_sse_attr" "fence")
|
8121 |
|
|
(set_attr "memory" "unknown")])
|
8122 |
|
|
|
8123 |
|
|
(define_expand "sse2_mfence"
|
8124 |
|
|
[(set (match_dup 0)
|
8125 |
|
|
(unspec:BLK [(match_dup 0)] UNSPEC_MFENCE))]
|
8126 |
|
|
"TARGET_SSE2"
|
8127 |
|
|
{
|
8128 |
|
|
operands[0] = gen_rtx_MEM (BLKmode, gen_rtx_SCRATCH (Pmode));
|
8129 |
|
|
MEM_VOLATILE_P (operands[0]) = 1;
|
8130 |
|
|
})
|
8131 |
|
|
|
8132 |
|
|
(define_insn "*sse2_mfence"
|
8133 |
|
|
[(set (match_operand:BLK 0 "" "")
|
8134 |
|
|
(unspec:BLK [(match_dup 0)] UNSPEC_MFENCE))]
|
8135 |
|
|
"TARGET_64BIT || TARGET_SSE2"
|
8136 |
|
|
"mfence"
|
8137 |
|
|
[(set_attr "type" "sse")
|
8138 |
|
|
(set_attr "length_address" "0")
|
8139 |
|
|
(set_attr "atom_sse_attr" "fence")
|
8140 |
|
|
(set_attr "memory" "unknown")])
|
8141 |
|
|
|
8142 |
|
|
(define_expand "sse2_lfence"
|
8143 |
|
|
[(set (match_dup 0)
|
8144 |
|
|
(unspec:BLK [(match_dup 0)] UNSPEC_LFENCE))]
|
8145 |
|
|
"TARGET_SSE2"
|
8146 |
|
|
{
|
8147 |
|
|
operands[0] = gen_rtx_MEM (BLKmode, gen_rtx_SCRATCH (Pmode));
|
8148 |
|
|
MEM_VOLATILE_P (operands[0]) = 1;
|
8149 |
|
|
})
|
8150 |
|
|
|
8151 |
|
|
(define_insn "*sse2_lfence"
|
8152 |
|
|
[(set (match_operand:BLK 0 "" "")
|
8153 |
|
|
(unspec:BLK [(match_dup 0)] UNSPEC_LFENCE))]
|
8154 |
|
|
"TARGET_SSE2"
|
8155 |
|
|
"lfence"
|
8156 |
|
|
[(set_attr "type" "sse")
|
8157 |
|
|
(set_attr "length_address" "0")
|
8158 |
|
|
(set_attr "atom_sse_attr" "lfence")
|
8159 |
|
|
(set_attr "memory" "unknown")])
|
8160 |
|
|
|
8161 |
|
|
(define_insn "sse3_mwait"
|
8162 |
|
|
[(unspec_volatile [(match_operand:SI 0 "register_operand" "a")
|
8163 |
|
|
(match_operand:SI 1 "register_operand" "c")]
|
8164 |
|
|
UNSPECV_MWAIT)]
|
8165 |
|
|
"TARGET_SSE3"
|
8166 |
|
|
;; 64bit version is "mwait %rax,%rcx". But only lower 32bits are used.
|
8167 |
|
|
;; Since 32bit register operands are implicitly zero extended to 64bit,
|
8168 |
|
|
;; we only need to set up 32bit registers.
|
8169 |
|
|
"mwait"
|
8170 |
|
|
[(set_attr "length" "3")])
|
8171 |
|
|
|
8172 |
|
|
(define_insn "sse3_monitor"
|
8173 |
|
|
[(unspec_volatile [(match_operand:SI 0 "register_operand" "a")
|
8174 |
|
|
(match_operand:SI 1 "register_operand" "c")
|
8175 |
|
|
(match_operand:SI 2 "register_operand" "d")]
|
8176 |
|
|
UNSPECV_MONITOR)]
|
8177 |
|
|
"TARGET_SSE3 && !TARGET_64BIT"
|
8178 |
|
|
"monitor\t%0, %1, %2"
|
8179 |
|
|
[(set_attr "length" "3")])
|
8180 |
|
|
|
8181 |
|
|
(define_insn "sse3_monitor64"
|
8182 |
|
|
[(unspec_volatile [(match_operand:DI 0 "register_operand" "a")
|
8183 |
|
|
(match_operand:SI 1 "register_operand" "c")
|
8184 |
|
|
(match_operand:SI 2 "register_operand" "d")]
|
8185 |
|
|
UNSPECV_MONITOR)]
|
8186 |
|
|
"TARGET_SSE3 && TARGET_64BIT"
|
8187 |
|
|
;; 64bit version is "monitor %rax,%rcx,%rdx". But only lower 32bits in
|
8188 |
|
|
;; RCX and RDX are used. Since 32bit register operands are implicitly
|
8189 |
|
|
;; zero extended to 64bit, we only need to set up 32bit registers.
|
8190 |
|
|
"monitor"
|
8191 |
|
|
[(set_attr "length" "3")])
|
8192 |
|
|
|
8193 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
8194 |
|
|
;;
|
8195 |
|
|
;; SSSE3 instructions
|
8196 |
|
|
;;
|
8197 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
8198 |
|
|
|
8199 |
|
|
(define_insn "*avx_phaddwv8hi3"
|
8200 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8201 |
|
|
(vec_concat:V8HI
|
8202 |
|
|
(vec_concat:V4HI
|
8203 |
|
|
(vec_concat:V2HI
|
8204 |
|
|
(plus:HI
|
8205 |
|
|
(vec_select:HI
|
8206 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
8207 |
|
|
(parallel [(const_int 0)]))
|
8208 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8209 |
|
|
(plus:HI
|
8210 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8211 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8212 |
|
|
(vec_concat:V2HI
|
8213 |
|
|
(plus:HI
|
8214 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 4)]))
|
8215 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 5)])))
|
8216 |
|
|
(plus:HI
|
8217 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 6)]))
|
8218 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 7)])))))
|
8219 |
|
|
(vec_concat:V4HI
|
8220 |
|
|
(vec_concat:V2HI
|
8221 |
|
|
(plus:HI
|
8222 |
|
|
(vec_select:HI
|
8223 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
8224 |
|
|
(parallel [(const_int 0)]))
|
8225 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8226 |
|
|
(plus:HI
|
8227 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8228 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))
|
8229 |
|
|
(vec_concat:V2HI
|
8230 |
|
|
(plus:HI
|
8231 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 4)]))
|
8232 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 5)])))
|
8233 |
|
|
(plus:HI
|
8234 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 6)]))
|
8235 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 7)])))))))]
|
8236 |
|
|
"TARGET_AVX"
|
8237 |
|
|
"vphaddw\t{%2, %1, %0|%0, %1, %2}"
|
8238 |
|
|
[(set_attr "type" "sseiadd")
|
8239 |
|
|
(set_attr "prefix_extra" "1")
|
8240 |
|
|
(set_attr "prefix" "vex")
|
8241 |
|
|
(set_attr "mode" "TI")])
|
8242 |
|
|
|
8243 |
|
|
(define_insn "ssse3_phaddwv8hi3"
|
8244 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8245 |
|
|
(vec_concat:V8HI
|
8246 |
|
|
(vec_concat:V4HI
|
8247 |
|
|
(vec_concat:V2HI
|
8248 |
|
|
(plus:HI
|
8249 |
|
|
(vec_select:HI
|
8250 |
|
|
(match_operand:V8HI 1 "register_operand" "0")
|
8251 |
|
|
(parallel [(const_int 0)]))
|
8252 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8253 |
|
|
(plus:HI
|
8254 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8255 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8256 |
|
|
(vec_concat:V2HI
|
8257 |
|
|
(plus:HI
|
8258 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 4)]))
|
8259 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 5)])))
|
8260 |
|
|
(plus:HI
|
8261 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 6)]))
|
8262 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 7)])))))
|
8263 |
|
|
(vec_concat:V4HI
|
8264 |
|
|
(vec_concat:V2HI
|
8265 |
|
|
(plus:HI
|
8266 |
|
|
(vec_select:HI
|
8267 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
8268 |
|
|
(parallel [(const_int 0)]))
|
8269 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8270 |
|
|
(plus:HI
|
8271 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8272 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))
|
8273 |
|
|
(vec_concat:V2HI
|
8274 |
|
|
(plus:HI
|
8275 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 4)]))
|
8276 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 5)])))
|
8277 |
|
|
(plus:HI
|
8278 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 6)]))
|
8279 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 7)])))))))]
|
8280 |
|
|
"TARGET_SSSE3"
|
8281 |
|
|
"phaddw\t{%2, %0|%0, %2}"
|
8282 |
|
|
[(set_attr "type" "sseiadd")
|
8283 |
|
|
(set_attr "atom_unit" "complex")
|
8284 |
|
|
(set_attr "prefix_data16" "1")
|
8285 |
|
|
(set_attr "prefix_extra" "1")
|
8286 |
|
|
(set_attr "mode" "TI")])
|
8287 |
|
|
|
8288 |
|
|
(define_insn "ssse3_phaddwv4hi3"
|
8289 |
|
|
[(set (match_operand:V4HI 0 "register_operand" "=y")
|
8290 |
|
|
(vec_concat:V4HI
|
8291 |
|
|
(vec_concat:V2HI
|
8292 |
|
|
(plus:HI
|
8293 |
|
|
(vec_select:HI
|
8294 |
|
|
(match_operand:V4HI 1 "register_operand" "0")
|
8295 |
|
|
(parallel [(const_int 0)]))
|
8296 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8297 |
|
|
(plus:HI
|
8298 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8299 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8300 |
|
|
(vec_concat:V2HI
|
8301 |
|
|
(plus:HI
|
8302 |
|
|
(vec_select:HI
|
8303 |
|
|
(match_operand:V4HI 2 "nonimmediate_operand" "ym")
|
8304 |
|
|
(parallel [(const_int 0)]))
|
8305 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8306 |
|
|
(plus:HI
|
8307 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8308 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))))]
|
8309 |
|
|
"TARGET_SSSE3"
|
8310 |
|
|
"phaddw\t{%2, %0|%0, %2}"
|
8311 |
|
|
[(set_attr "type" "sseiadd")
|
8312 |
|
|
(set_attr "atom_unit" "complex")
|
8313 |
|
|
(set_attr "prefix_extra" "1")
|
8314 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
8315 |
|
|
(set_attr "mode" "DI")])
|
8316 |
|
|
|
8317 |
|
|
(define_insn "*avx_phadddv4si3"
|
8318 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
8319 |
|
|
(vec_concat:V4SI
|
8320 |
|
|
(vec_concat:V2SI
|
8321 |
|
|
(plus:SI
|
8322 |
|
|
(vec_select:SI
|
8323 |
|
|
(match_operand:V4SI 1 "register_operand" "x")
|
8324 |
|
|
(parallel [(const_int 0)]))
|
8325 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 1)])))
|
8326 |
|
|
(plus:SI
|
8327 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 2)]))
|
8328 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 3)]))))
|
8329 |
|
|
(vec_concat:V2SI
|
8330 |
|
|
(plus:SI
|
8331 |
|
|
(vec_select:SI
|
8332 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
8333 |
|
|
(parallel [(const_int 0)]))
|
8334 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 1)])))
|
8335 |
|
|
(plus:SI
|
8336 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 2)]))
|
8337 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 3)]))))))]
|
8338 |
|
|
"TARGET_AVX"
|
8339 |
|
|
"vphaddd\t{%2, %1, %0|%0, %1, %2}"
|
8340 |
|
|
[(set_attr "type" "sseiadd")
|
8341 |
|
|
(set_attr "prefix_extra" "1")
|
8342 |
|
|
(set_attr "prefix" "vex")
|
8343 |
|
|
(set_attr "mode" "TI")])
|
8344 |
|
|
|
8345 |
|
|
(define_insn "ssse3_phadddv4si3"
|
8346 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
8347 |
|
|
(vec_concat:V4SI
|
8348 |
|
|
(vec_concat:V2SI
|
8349 |
|
|
(plus:SI
|
8350 |
|
|
(vec_select:SI
|
8351 |
|
|
(match_operand:V4SI 1 "register_operand" "0")
|
8352 |
|
|
(parallel [(const_int 0)]))
|
8353 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 1)])))
|
8354 |
|
|
(plus:SI
|
8355 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 2)]))
|
8356 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 3)]))))
|
8357 |
|
|
(vec_concat:V2SI
|
8358 |
|
|
(plus:SI
|
8359 |
|
|
(vec_select:SI
|
8360 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
8361 |
|
|
(parallel [(const_int 0)]))
|
8362 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 1)])))
|
8363 |
|
|
(plus:SI
|
8364 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 2)]))
|
8365 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 3)]))))))]
|
8366 |
|
|
"TARGET_SSSE3"
|
8367 |
|
|
"phaddd\t{%2, %0|%0, %2}"
|
8368 |
|
|
[(set_attr "type" "sseiadd")
|
8369 |
|
|
(set_attr "atom_unit" "complex")
|
8370 |
|
|
(set_attr "prefix_data16" "1")
|
8371 |
|
|
(set_attr "prefix_extra" "1")
|
8372 |
|
|
(set_attr "mode" "TI")])
|
8373 |
|
|
|
8374 |
|
|
(define_insn "ssse3_phadddv2si3"
|
8375 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=y")
|
8376 |
|
|
(vec_concat:V2SI
|
8377 |
|
|
(plus:SI
|
8378 |
|
|
(vec_select:SI
|
8379 |
|
|
(match_operand:V2SI 1 "register_operand" "0")
|
8380 |
|
|
(parallel [(const_int 0)]))
|
8381 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 1)])))
|
8382 |
|
|
(plus:SI
|
8383 |
|
|
(vec_select:SI
|
8384 |
|
|
(match_operand:V2SI 2 "nonimmediate_operand" "ym")
|
8385 |
|
|
(parallel [(const_int 0)]))
|
8386 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 1)])))))]
|
8387 |
|
|
"TARGET_SSSE3"
|
8388 |
|
|
"phaddd\t{%2, %0|%0, %2}"
|
8389 |
|
|
[(set_attr "type" "sseiadd")
|
8390 |
|
|
(set_attr "atom_unit" "complex")
|
8391 |
|
|
(set_attr "prefix_extra" "1")
|
8392 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
8393 |
|
|
(set_attr "mode" "DI")])
|
8394 |
|
|
|
8395 |
|
|
(define_insn "*avx_phaddswv8hi3"
|
8396 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8397 |
|
|
(vec_concat:V8HI
|
8398 |
|
|
(vec_concat:V4HI
|
8399 |
|
|
(vec_concat:V2HI
|
8400 |
|
|
(ss_plus:HI
|
8401 |
|
|
(vec_select:HI
|
8402 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
8403 |
|
|
(parallel [(const_int 0)]))
|
8404 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8405 |
|
|
(ss_plus:HI
|
8406 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8407 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8408 |
|
|
(vec_concat:V2HI
|
8409 |
|
|
(ss_plus:HI
|
8410 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 4)]))
|
8411 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 5)])))
|
8412 |
|
|
(ss_plus:HI
|
8413 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 6)]))
|
8414 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 7)])))))
|
8415 |
|
|
(vec_concat:V4HI
|
8416 |
|
|
(vec_concat:V2HI
|
8417 |
|
|
(ss_plus:HI
|
8418 |
|
|
(vec_select:HI
|
8419 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
8420 |
|
|
(parallel [(const_int 0)]))
|
8421 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8422 |
|
|
(ss_plus:HI
|
8423 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8424 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))
|
8425 |
|
|
(vec_concat:V2HI
|
8426 |
|
|
(ss_plus:HI
|
8427 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 4)]))
|
8428 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 5)])))
|
8429 |
|
|
(ss_plus:HI
|
8430 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 6)]))
|
8431 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 7)])))))))]
|
8432 |
|
|
"TARGET_AVX"
|
8433 |
|
|
"vphaddsw\t{%2, %1, %0|%0, %1, %2}"
|
8434 |
|
|
[(set_attr "type" "sseiadd")
|
8435 |
|
|
(set_attr "prefix_extra" "1")
|
8436 |
|
|
(set_attr "prefix" "vex")
|
8437 |
|
|
(set_attr "mode" "TI")])
|
8438 |
|
|
|
8439 |
|
|
(define_insn "ssse3_phaddswv8hi3"
|
8440 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8441 |
|
|
(vec_concat:V8HI
|
8442 |
|
|
(vec_concat:V4HI
|
8443 |
|
|
(vec_concat:V2HI
|
8444 |
|
|
(ss_plus:HI
|
8445 |
|
|
(vec_select:HI
|
8446 |
|
|
(match_operand:V8HI 1 "register_operand" "0")
|
8447 |
|
|
(parallel [(const_int 0)]))
|
8448 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8449 |
|
|
(ss_plus:HI
|
8450 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8451 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8452 |
|
|
(vec_concat:V2HI
|
8453 |
|
|
(ss_plus:HI
|
8454 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 4)]))
|
8455 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 5)])))
|
8456 |
|
|
(ss_plus:HI
|
8457 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 6)]))
|
8458 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 7)])))))
|
8459 |
|
|
(vec_concat:V4HI
|
8460 |
|
|
(vec_concat:V2HI
|
8461 |
|
|
(ss_plus:HI
|
8462 |
|
|
(vec_select:HI
|
8463 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
8464 |
|
|
(parallel [(const_int 0)]))
|
8465 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8466 |
|
|
(ss_plus:HI
|
8467 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8468 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))
|
8469 |
|
|
(vec_concat:V2HI
|
8470 |
|
|
(ss_plus:HI
|
8471 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 4)]))
|
8472 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 5)])))
|
8473 |
|
|
(ss_plus:HI
|
8474 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 6)]))
|
8475 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 7)])))))))]
|
8476 |
|
|
"TARGET_SSSE3"
|
8477 |
|
|
"phaddsw\t{%2, %0|%0, %2}"
|
8478 |
|
|
[(set_attr "type" "sseiadd")
|
8479 |
|
|
(set_attr "atom_unit" "complex")
|
8480 |
|
|
(set_attr "prefix_data16" "1")
|
8481 |
|
|
(set_attr "prefix_extra" "1")
|
8482 |
|
|
(set_attr "mode" "TI")])
|
8483 |
|
|
|
8484 |
|
|
(define_insn "ssse3_phaddswv4hi3"
|
8485 |
|
|
[(set (match_operand:V4HI 0 "register_operand" "=y")
|
8486 |
|
|
(vec_concat:V4HI
|
8487 |
|
|
(vec_concat:V2HI
|
8488 |
|
|
(ss_plus:HI
|
8489 |
|
|
(vec_select:HI
|
8490 |
|
|
(match_operand:V4HI 1 "register_operand" "0")
|
8491 |
|
|
(parallel [(const_int 0)]))
|
8492 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8493 |
|
|
(ss_plus:HI
|
8494 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8495 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8496 |
|
|
(vec_concat:V2HI
|
8497 |
|
|
(ss_plus:HI
|
8498 |
|
|
(vec_select:HI
|
8499 |
|
|
(match_operand:V4HI 2 "nonimmediate_operand" "ym")
|
8500 |
|
|
(parallel [(const_int 0)]))
|
8501 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8502 |
|
|
(ss_plus:HI
|
8503 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8504 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))))]
|
8505 |
|
|
"TARGET_SSSE3"
|
8506 |
|
|
"phaddsw\t{%2, %0|%0, %2}"
|
8507 |
|
|
[(set_attr "type" "sseiadd")
|
8508 |
|
|
(set_attr "atom_unit" "complex")
|
8509 |
|
|
(set_attr "prefix_extra" "1")
|
8510 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
8511 |
|
|
(set_attr "mode" "DI")])
|
8512 |
|
|
|
8513 |
|
|
(define_insn "*avx_phsubwv8hi3"
|
8514 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8515 |
|
|
(vec_concat:V8HI
|
8516 |
|
|
(vec_concat:V4HI
|
8517 |
|
|
(vec_concat:V2HI
|
8518 |
|
|
(minus:HI
|
8519 |
|
|
(vec_select:HI
|
8520 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
8521 |
|
|
(parallel [(const_int 0)]))
|
8522 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8523 |
|
|
(minus:HI
|
8524 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8525 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8526 |
|
|
(vec_concat:V2HI
|
8527 |
|
|
(minus:HI
|
8528 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 4)]))
|
8529 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 5)])))
|
8530 |
|
|
(minus:HI
|
8531 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 6)]))
|
8532 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 7)])))))
|
8533 |
|
|
(vec_concat:V4HI
|
8534 |
|
|
(vec_concat:V2HI
|
8535 |
|
|
(minus:HI
|
8536 |
|
|
(vec_select:HI
|
8537 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
8538 |
|
|
(parallel [(const_int 0)]))
|
8539 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8540 |
|
|
(minus:HI
|
8541 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8542 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))
|
8543 |
|
|
(vec_concat:V2HI
|
8544 |
|
|
(minus:HI
|
8545 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 4)]))
|
8546 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 5)])))
|
8547 |
|
|
(minus:HI
|
8548 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 6)]))
|
8549 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 7)])))))))]
|
8550 |
|
|
"TARGET_AVX"
|
8551 |
|
|
"vphsubw\t{%2, %1, %0|%0, %1, %2}"
|
8552 |
|
|
[(set_attr "type" "sseiadd")
|
8553 |
|
|
(set_attr "prefix_extra" "1")
|
8554 |
|
|
(set_attr "prefix" "vex")
|
8555 |
|
|
(set_attr "mode" "TI")])
|
8556 |
|
|
|
8557 |
|
|
(define_insn "ssse3_phsubwv8hi3"
|
8558 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8559 |
|
|
(vec_concat:V8HI
|
8560 |
|
|
(vec_concat:V4HI
|
8561 |
|
|
(vec_concat:V2HI
|
8562 |
|
|
(minus:HI
|
8563 |
|
|
(vec_select:HI
|
8564 |
|
|
(match_operand:V8HI 1 "register_operand" "0")
|
8565 |
|
|
(parallel [(const_int 0)]))
|
8566 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8567 |
|
|
(minus:HI
|
8568 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8569 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8570 |
|
|
(vec_concat:V2HI
|
8571 |
|
|
(minus:HI
|
8572 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 4)]))
|
8573 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 5)])))
|
8574 |
|
|
(minus:HI
|
8575 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 6)]))
|
8576 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 7)])))))
|
8577 |
|
|
(vec_concat:V4HI
|
8578 |
|
|
(vec_concat:V2HI
|
8579 |
|
|
(minus:HI
|
8580 |
|
|
(vec_select:HI
|
8581 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
8582 |
|
|
(parallel [(const_int 0)]))
|
8583 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8584 |
|
|
(minus:HI
|
8585 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8586 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))
|
8587 |
|
|
(vec_concat:V2HI
|
8588 |
|
|
(minus:HI
|
8589 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 4)]))
|
8590 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 5)])))
|
8591 |
|
|
(minus:HI
|
8592 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 6)]))
|
8593 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 7)])))))))]
|
8594 |
|
|
"TARGET_SSSE3"
|
8595 |
|
|
"phsubw\t{%2, %0|%0, %2}"
|
8596 |
|
|
[(set_attr "type" "sseiadd")
|
8597 |
|
|
(set_attr "atom_unit" "complex")
|
8598 |
|
|
(set_attr "prefix_data16" "1")
|
8599 |
|
|
(set_attr "prefix_extra" "1")
|
8600 |
|
|
(set_attr "mode" "TI")])
|
8601 |
|
|
|
8602 |
|
|
(define_insn "ssse3_phsubwv4hi3"
|
8603 |
|
|
[(set (match_operand:V4HI 0 "register_operand" "=y")
|
8604 |
|
|
(vec_concat:V4HI
|
8605 |
|
|
(vec_concat:V2HI
|
8606 |
|
|
(minus:HI
|
8607 |
|
|
(vec_select:HI
|
8608 |
|
|
(match_operand:V4HI 1 "register_operand" "0")
|
8609 |
|
|
(parallel [(const_int 0)]))
|
8610 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8611 |
|
|
(minus:HI
|
8612 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8613 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8614 |
|
|
(vec_concat:V2HI
|
8615 |
|
|
(minus:HI
|
8616 |
|
|
(vec_select:HI
|
8617 |
|
|
(match_operand:V4HI 2 "nonimmediate_operand" "ym")
|
8618 |
|
|
(parallel [(const_int 0)]))
|
8619 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8620 |
|
|
(minus:HI
|
8621 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8622 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))))]
|
8623 |
|
|
"TARGET_SSSE3"
|
8624 |
|
|
"phsubw\t{%2, %0|%0, %2}"
|
8625 |
|
|
[(set_attr "type" "sseiadd")
|
8626 |
|
|
(set_attr "atom_unit" "complex")
|
8627 |
|
|
(set_attr "prefix_extra" "1")
|
8628 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
8629 |
|
|
(set_attr "mode" "DI")])
|
8630 |
|
|
|
8631 |
|
|
(define_insn "*avx_phsubdv4si3"
|
8632 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
8633 |
|
|
(vec_concat:V4SI
|
8634 |
|
|
(vec_concat:V2SI
|
8635 |
|
|
(minus:SI
|
8636 |
|
|
(vec_select:SI
|
8637 |
|
|
(match_operand:V4SI 1 "register_operand" "x")
|
8638 |
|
|
(parallel [(const_int 0)]))
|
8639 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 1)])))
|
8640 |
|
|
(minus:SI
|
8641 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 2)]))
|
8642 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 3)]))))
|
8643 |
|
|
(vec_concat:V2SI
|
8644 |
|
|
(minus:SI
|
8645 |
|
|
(vec_select:SI
|
8646 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
8647 |
|
|
(parallel [(const_int 0)]))
|
8648 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 1)])))
|
8649 |
|
|
(minus:SI
|
8650 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 2)]))
|
8651 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 3)]))))))]
|
8652 |
|
|
"TARGET_AVX"
|
8653 |
|
|
"vphsubd\t{%2, %1, %0|%0, %1, %2}"
|
8654 |
|
|
[(set_attr "type" "sseiadd")
|
8655 |
|
|
(set_attr "prefix_extra" "1")
|
8656 |
|
|
(set_attr "prefix" "vex")
|
8657 |
|
|
(set_attr "mode" "TI")])
|
8658 |
|
|
|
8659 |
|
|
(define_insn "ssse3_phsubdv4si3"
|
8660 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
8661 |
|
|
(vec_concat:V4SI
|
8662 |
|
|
(vec_concat:V2SI
|
8663 |
|
|
(minus:SI
|
8664 |
|
|
(vec_select:SI
|
8665 |
|
|
(match_operand:V4SI 1 "register_operand" "0")
|
8666 |
|
|
(parallel [(const_int 0)]))
|
8667 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 1)])))
|
8668 |
|
|
(minus:SI
|
8669 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 2)]))
|
8670 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 3)]))))
|
8671 |
|
|
(vec_concat:V2SI
|
8672 |
|
|
(minus:SI
|
8673 |
|
|
(vec_select:SI
|
8674 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
8675 |
|
|
(parallel [(const_int 0)]))
|
8676 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 1)])))
|
8677 |
|
|
(minus:SI
|
8678 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 2)]))
|
8679 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 3)]))))))]
|
8680 |
|
|
"TARGET_SSSE3"
|
8681 |
|
|
"phsubd\t{%2, %0|%0, %2}"
|
8682 |
|
|
[(set_attr "type" "sseiadd")
|
8683 |
|
|
(set_attr "atom_unit" "complex")
|
8684 |
|
|
(set_attr "prefix_data16" "1")
|
8685 |
|
|
(set_attr "prefix_extra" "1")
|
8686 |
|
|
(set_attr "mode" "TI")])
|
8687 |
|
|
|
8688 |
|
|
(define_insn "ssse3_phsubdv2si3"
|
8689 |
|
|
[(set (match_operand:V2SI 0 "register_operand" "=y")
|
8690 |
|
|
(vec_concat:V2SI
|
8691 |
|
|
(minus:SI
|
8692 |
|
|
(vec_select:SI
|
8693 |
|
|
(match_operand:V2SI 1 "register_operand" "0")
|
8694 |
|
|
(parallel [(const_int 0)]))
|
8695 |
|
|
(vec_select:SI (match_dup 1) (parallel [(const_int 1)])))
|
8696 |
|
|
(minus:SI
|
8697 |
|
|
(vec_select:SI
|
8698 |
|
|
(match_operand:V2SI 2 "nonimmediate_operand" "ym")
|
8699 |
|
|
(parallel [(const_int 0)]))
|
8700 |
|
|
(vec_select:SI (match_dup 2) (parallel [(const_int 1)])))))]
|
8701 |
|
|
"TARGET_SSSE3"
|
8702 |
|
|
"phsubd\t{%2, %0|%0, %2}"
|
8703 |
|
|
[(set_attr "type" "sseiadd")
|
8704 |
|
|
(set_attr "atom_unit" "complex")
|
8705 |
|
|
(set_attr "prefix_extra" "1")
|
8706 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
8707 |
|
|
(set_attr "mode" "DI")])
|
8708 |
|
|
|
8709 |
|
|
(define_insn "*avx_phsubswv8hi3"
|
8710 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8711 |
|
|
(vec_concat:V8HI
|
8712 |
|
|
(vec_concat:V4HI
|
8713 |
|
|
(vec_concat:V2HI
|
8714 |
|
|
(ss_minus:HI
|
8715 |
|
|
(vec_select:HI
|
8716 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
8717 |
|
|
(parallel [(const_int 0)]))
|
8718 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8719 |
|
|
(ss_minus:HI
|
8720 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8721 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8722 |
|
|
(vec_concat:V2HI
|
8723 |
|
|
(ss_minus:HI
|
8724 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 4)]))
|
8725 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 5)])))
|
8726 |
|
|
(ss_minus:HI
|
8727 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 6)]))
|
8728 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 7)])))))
|
8729 |
|
|
(vec_concat:V4HI
|
8730 |
|
|
(vec_concat:V2HI
|
8731 |
|
|
(ss_minus:HI
|
8732 |
|
|
(vec_select:HI
|
8733 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
8734 |
|
|
(parallel [(const_int 0)]))
|
8735 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8736 |
|
|
(ss_minus:HI
|
8737 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8738 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))
|
8739 |
|
|
(vec_concat:V2HI
|
8740 |
|
|
(ss_minus:HI
|
8741 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 4)]))
|
8742 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 5)])))
|
8743 |
|
|
(ss_minus:HI
|
8744 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 6)]))
|
8745 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 7)])))))))]
|
8746 |
|
|
"TARGET_AVX"
|
8747 |
|
|
"vphsubsw\t{%2, %1, %0|%0, %1, %2}"
|
8748 |
|
|
[(set_attr "type" "sseiadd")
|
8749 |
|
|
(set_attr "prefix_extra" "1")
|
8750 |
|
|
(set_attr "prefix" "vex")
|
8751 |
|
|
(set_attr "mode" "TI")])
|
8752 |
|
|
|
8753 |
|
|
(define_insn "ssse3_phsubswv8hi3"
|
8754 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8755 |
|
|
(vec_concat:V8HI
|
8756 |
|
|
(vec_concat:V4HI
|
8757 |
|
|
(vec_concat:V2HI
|
8758 |
|
|
(ss_minus:HI
|
8759 |
|
|
(vec_select:HI
|
8760 |
|
|
(match_operand:V8HI 1 "register_operand" "0")
|
8761 |
|
|
(parallel [(const_int 0)]))
|
8762 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8763 |
|
|
(ss_minus:HI
|
8764 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8765 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8766 |
|
|
(vec_concat:V2HI
|
8767 |
|
|
(ss_minus:HI
|
8768 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 4)]))
|
8769 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 5)])))
|
8770 |
|
|
(ss_minus:HI
|
8771 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 6)]))
|
8772 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 7)])))))
|
8773 |
|
|
(vec_concat:V4HI
|
8774 |
|
|
(vec_concat:V2HI
|
8775 |
|
|
(ss_minus:HI
|
8776 |
|
|
(vec_select:HI
|
8777 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
8778 |
|
|
(parallel [(const_int 0)]))
|
8779 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8780 |
|
|
(ss_minus:HI
|
8781 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8782 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))
|
8783 |
|
|
(vec_concat:V2HI
|
8784 |
|
|
(ss_minus:HI
|
8785 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 4)]))
|
8786 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 5)])))
|
8787 |
|
|
(ss_minus:HI
|
8788 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 6)]))
|
8789 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 7)])))))))]
|
8790 |
|
|
"TARGET_SSSE3"
|
8791 |
|
|
"phsubsw\t{%2, %0|%0, %2}"
|
8792 |
|
|
[(set_attr "type" "sseiadd")
|
8793 |
|
|
(set_attr "atom_unit" "complex")
|
8794 |
|
|
(set_attr "prefix_data16" "1")
|
8795 |
|
|
(set_attr "prefix_extra" "1")
|
8796 |
|
|
(set_attr "mode" "TI")])
|
8797 |
|
|
|
8798 |
|
|
(define_insn "ssse3_phsubswv4hi3"
|
8799 |
|
|
[(set (match_operand:V4HI 0 "register_operand" "=y")
|
8800 |
|
|
(vec_concat:V4HI
|
8801 |
|
|
(vec_concat:V2HI
|
8802 |
|
|
(ss_minus:HI
|
8803 |
|
|
(vec_select:HI
|
8804 |
|
|
(match_operand:V4HI 1 "register_operand" "0")
|
8805 |
|
|
(parallel [(const_int 0)]))
|
8806 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 1)])))
|
8807 |
|
|
(ss_minus:HI
|
8808 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 2)]))
|
8809 |
|
|
(vec_select:HI (match_dup 1) (parallel [(const_int 3)]))))
|
8810 |
|
|
(vec_concat:V2HI
|
8811 |
|
|
(ss_minus:HI
|
8812 |
|
|
(vec_select:HI
|
8813 |
|
|
(match_operand:V4HI 2 "nonimmediate_operand" "ym")
|
8814 |
|
|
(parallel [(const_int 0)]))
|
8815 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 1)])))
|
8816 |
|
|
(ss_minus:HI
|
8817 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 2)]))
|
8818 |
|
|
(vec_select:HI (match_dup 2) (parallel [(const_int 3)]))))))]
|
8819 |
|
|
"TARGET_SSSE3"
|
8820 |
|
|
"phsubsw\t{%2, %0|%0, %2}"
|
8821 |
|
|
[(set_attr "type" "sseiadd")
|
8822 |
|
|
(set_attr "atom_unit" "complex")
|
8823 |
|
|
(set_attr "prefix_extra" "1")
|
8824 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
8825 |
|
|
(set_attr "mode" "DI")])
|
8826 |
|
|
|
8827 |
|
|
(define_insn "*avx_pmaddubsw128"
|
8828 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8829 |
|
|
(ss_plus:V8HI
|
8830 |
|
|
(mult:V8HI
|
8831 |
|
|
(zero_extend:V8HI
|
8832 |
|
|
(vec_select:V4QI
|
8833 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
8834 |
|
|
(parallel [(const_int 0)
|
8835 |
|
|
(const_int 2)
|
8836 |
|
|
(const_int 4)
|
8837 |
|
|
(const_int 6)
|
8838 |
|
|
(const_int 8)
|
8839 |
|
|
(const_int 10)
|
8840 |
|
|
(const_int 12)
|
8841 |
|
|
(const_int 14)])))
|
8842 |
|
|
(sign_extend:V8HI
|
8843 |
|
|
(vec_select:V8QI
|
8844 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")
|
8845 |
|
|
(parallel [(const_int 0)
|
8846 |
|
|
(const_int 2)
|
8847 |
|
|
(const_int 4)
|
8848 |
|
|
(const_int 6)
|
8849 |
|
|
(const_int 8)
|
8850 |
|
|
(const_int 10)
|
8851 |
|
|
(const_int 12)
|
8852 |
|
|
(const_int 14)]))))
|
8853 |
|
|
(mult:V8HI
|
8854 |
|
|
(zero_extend:V8HI
|
8855 |
|
|
(vec_select:V16QI (match_dup 1)
|
8856 |
|
|
(parallel [(const_int 1)
|
8857 |
|
|
(const_int 3)
|
8858 |
|
|
(const_int 5)
|
8859 |
|
|
(const_int 7)
|
8860 |
|
|
(const_int 9)
|
8861 |
|
|
(const_int 11)
|
8862 |
|
|
(const_int 13)
|
8863 |
|
|
(const_int 15)])))
|
8864 |
|
|
(sign_extend:V8HI
|
8865 |
|
|
(vec_select:V16QI (match_dup 2)
|
8866 |
|
|
(parallel [(const_int 1)
|
8867 |
|
|
(const_int 3)
|
8868 |
|
|
(const_int 5)
|
8869 |
|
|
(const_int 7)
|
8870 |
|
|
(const_int 9)
|
8871 |
|
|
(const_int 11)
|
8872 |
|
|
(const_int 13)
|
8873 |
|
|
(const_int 15)]))))))]
|
8874 |
|
|
"TARGET_AVX"
|
8875 |
|
|
"vpmaddubsw\t{%2, %1, %0|%0, %1, %2}"
|
8876 |
|
|
[(set_attr "type" "sseiadd")
|
8877 |
|
|
(set_attr "prefix_extra" "1")
|
8878 |
|
|
(set_attr "prefix" "vex")
|
8879 |
|
|
(set_attr "mode" "TI")])
|
8880 |
|
|
|
8881 |
|
|
(define_insn "ssse3_pmaddubsw128"
|
8882 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8883 |
|
|
(ss_plus:V8HI
|
8884 |
|
|
(mult:V8HI
|
8885 |
|
|
(zero_extend:V8HI
|
8886 |
|
|
(vec_select:V4QI
|
8887 |
|
|
(match_operand:V16QI 1 "register_operand" "0")
|
8888 |
|
|
(parallel [(const_int 0)
|
8889 |
|
|
(const_int 2)
|
8890 |
|
|
(const_int 4)
|
8891 |
|
|
(const_int 6)
|
8892 |
|
|
(const_int 8)
|
8893 |
|
|
(const_int 10)
|
8894 |
|
|
(const_int 12)
|
8895 |
|
|
(const_int 14)])))
|
8896 |
|
|
(sign_extend:V8HI
|
8897 |
|
|
(vec_select:V8QI
|
8898 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")
|
8899 |
|
|
(parallel [(const_int 0)
|
8900 |
|
|
(const_int 2)
|
8901 |
|
|
(const_int 4)
|
8902 |
|
|
(const_int 6)
|
8903 |
|
|
(const_int 8)
|
8904 |
|
|
(const_int 10)
|
8905 |
|
|
(const_int 12)
|
8906 |
|
|
(const_int 14)]))))
|
8907 |
|
|
(mult:V8HI
|
8908 |
|
|
(zero_extend:V8HI
|
8909 |
|
|
(vec_select:V16QI (match_dup 1)
|
8910 |
|
|
(parallel [(const_int 1)
|
8911 |
|
|
(const_int 3)
|
8912 |
|
|
(const_int 5)
|
8913 |
|
|
(const_int 7)
|
8914 |
|
|
(const_int 9)
|
8915 |
|
|
(const_int 11)
|
8916 |
|
|
(const_int 13)
|
8917 |
|
|
(const_int 15)])))
|
8918 |
|
|
(sign_extend:V8HI
|
8919 |
|
|
(vec_select:V16QI (match_dup 2)
|
8920 |
|
|
(parallel [(const_int 1)
|
8921 |
|
|
(const_int 3)
|
8922 |
|
|
(const_int 5)
|
8923 |
|
|
(const_int 7)
|
8924 |
|
|
(const_int 9)
|
8925 |
|
|
(const_int 11)
|
8926 |
|
|
(const_int 13)
|
8927 |
|
|
(const_int 15)]))))))]
|
8928 |
|
|
"TARGET_SSSE3"
|
8929 |
|
|
"pmaddubsw\t{%2, %0|%0, %2}"
|
8930 |
|
|
[(set_attr "type" "sseiadd")
|
8931 |
|
|
(set_attr "atom_unit" "simul")
|
8932 |
|
|
(set_attr "prefix_data16" "1")
|
8933 |
|
|
(set_attr "prefix_extra" "1")
|
8934 |
|
|
(set_attr "mode" "TI")])
|
8935 |
|
|
|
8936 |
|
|
(define_insn "ssse3_pmaddubsw"
|
8937 |
|
|
[(set (match_operand:V4HI 0 "register_operand" "=y")
|
8938 |
|
|
(ss_plus:V4HI
|
8939 |
|
|
(mult:V4HI
|
8940 |
|
|
(zero_extend:V4HI
|
8941 |
|
|
(vec_select:V4QI
|
8942 |
|
|
(match_operand:V8QI 1 "register_operand" "0")
|
8943 |
|
|
(parallel [(const_int 0)
|
8944 |
|
|
(const_int 2)
|
8945 |
|
|
(const_int 4)
|
8946 |
|
|
(const_int 6)])))
|
8947 |
|
|
(sign_extend:V4HI
|
8948 |
|
|
(vec_select:V4QI
|
8949 |
|
|
(match_operand:V8QI 2 "nonimmediate_operand" "ym")
|
8950 |
|
|
(parallel [(const_int 0)
|
8951 |
|
|
(const_int 2)
|
8952 |
|
|
(const_int 4)
|
8953 |
|
|
(const_int 6)]))))
|
8954 |
|
|
(mult:V4HI
|
8955 |
|
|
(zero_extend:V4HI
|
8956 |
|
|
(vec_select:V8QI (match_dup 1)
|
8957 |
|
|
(parallel [(const_int 1)
|
8958 |
|
|
(const_int 3)
|
8959 |
|
|
(const_int 5)
|
8960 |
|
|
(const_int 7)])))
|
8961 |
|
|
(sign_extend:V4HI
|
8962 |
|
|
(vec_select:V8QI (match_dup 2)
|
8963 |
|
|
(parallel [(const_int 1)
|
8964 |
|
|
(const_int 3)
|
8965 |
|
|
(const_int 5)
|
8966 |
|
|
(const_int 7)]))))))]
|
8967 |
|
|
"TARGET_SSSE3"
|
8968 |
|
|
"pmaddubsw\t{%2, %0|%0, %2}"
|
8969 |
|
|
[(set_attr "type" "sseiadd")
|
8970 |
|
|
(set_attr "atom_unit" "simul")
|
8971 |
|
|
(set_attr "prefix_extra" "1")
|
8972 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
8973 |
|
|
(set_attr "mode" "DI")])
|
8974 |
|
|
|
8975 |
|
|
(define_expand "ssse3_pmulhrswv8hi3"
|
8976 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "")
|
8977 |
|
|
(truncate:V8HI
|
8978 |
|
|
(lshiftrt:V8SI
|
8979 |
|
|
(plus:V8SI
|
8980 |
|
|
(lshiftrt:V8SI
|
8981 |
|
|
(mult:V8SI
|
8982 |
|
|
(sign_extend:V8SI
|
8983 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" ""))
|
8984 |
|
|
(sign_extend:V8SI
|
8985 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "")))
|
8986 |
|
|
(const_int 14))
|
8987 |
|
|
(const_vector:V8HI [(const_int 1) (const_int 1)
|
8988 |
|
|
(const_int 1) (const_int 1)
|
8989 |
|
|
(const_int 1) (const_int 1)
|
8990 |
|
|
(const_int 1) (const_int 1)]))
|
8991 |
|
|
(const_int 1))))]
|
8992 |
|
|
"TARGET_SSSE3"
|
8993 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, V8HImode, operands);")
|
8994 |
|
|
|
8995 |
|
|
(define_insn "*avx_pmulhrswv8hi3"
|
8996 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
8997 |
|
|
(truncate:V8HI
|
8998 |
|
|
(lshiftrt:V8SI
|
8999 |
|
|
(plus:V8SI
|
9000 |
|
|
(lshiftrt:V8SI
|
9001 |
|
|
(mult:V8SI
|
9002 |
|
|
(sign_extend:V8SI
|
9003 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x"))
|
9004 |
|
|
(sign_extend:V8SI
|
9005 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))
|
9006 |
|
|
(const_int 14))
|
9007 |
|
|
(const_vector:V8HI [(const_int 1) (const_int 1)
|
9008 |
|
|
(const_int 1) (const_int 1)
|
9009 |
|
|
(const_int 1) (const_int 1)
|
9010 |
|
|
(const_int 1) (const_int 1)]))
|
9011 |
|
|
(const_int 1))))]
|
9012 |
|
|
"TARGET_AVX && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
9013 |
|
|
"vpmulhrsw\t{%2, %1, %0|%0, %1, %2}"
|
9014 |
|
|
[(set_attr "type" "sseimul")
|
9015 |
|
|
(set_attr "prefix_extra" "1")
|
9016 |
|
|
(set_attr "prefix" "vex")
|
9017 |
|
|
(set_attr "mode" "TI")])
|
9018 |
|
|
|
9019 |
|
|
(define_insn "*ssse3_pmulhrswv8hi3"
|
9020 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9021 |
|
|
(truncate:V8HI
|
9022 |
|
|
(lshiftrt:V8SI
|
9023 |
|
|
(plus:V8SI
|
9024 |
|
|
(lshiftrt:V8SI
|
9025 |
|
|
(mult:V8SI
|
9026 |
|
|
(sign_extend:V8SI
|
9027 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%0"))
|
9028 |
|
|
(sign_extend:V8SI
|
9029 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))
|
9030 |
|
|
(const_int 14))
|
9031 |
|
|
(const_vector:V8HI [(const_int 1) (const_int 1)
|
9032 |
|
|
(const_int 1) (const_int 1)
|
9033 |
|
|
(const_int 1) (const_int 1)
|
9034 |
|
|
(const_int 1) (const_int 1)]))
|
9035 |
|
|
(const_int 1))))]
|
9036 |
|
|
"TARGET_SSSE3 && ix86_binary_operator_ok (MULT, V8HImode, operands)"
|
9037 |
|
|
"pmulhrsw\t{%2, %0|%0, %2}"
|
9038 |
|
|
[(set_attr "type" "sseimul")
|
9039 |
|
|
(set_attr "prefix_data16" "1")
|
9040 |
|
|
(set_attr "prefix_extra" "1")
|
9041 |
|
|
(set_attr "mode" "TI")])
|
9042 |
|
|
|
9043 |
|
|
(define_expand "ssse3_pmulhrswv4hi3"
|
9044 |
|
|
[(set (match_operand:V4HI 0 "register_operand" "")
|
9045 |
|
|
(truncate:V4HI
|
9046 |
|
|
(lshiftrt:V4SI
|
9047 |
|
|
(plus:V4SI
|
9048 |
|
|
(lshiftrt:V4SI
|
9049 |
|
|
(mult:V4SI
|
9050 |
|
|
(sign_extend:V4SI
|
9051 |
|
|
(match_operand:V4HI 1 "nonimmediate_operand" ""))
|
9052 |
|
|
(sign_extend:V4SI
|
9053 |
|
|
(match_operand:V4HI 2 "nonimmediate_operand" "")))
|
9054 |
|
|
(const_int 14))
|
9055 |
|
|
(const_vector:V4HI [(const_int 1) (const_int 1)
|
9056 |
|
|
(const_int 1) (const_int 1)]))
|
9057 |
|
|
(const_int 1))))]
|
9058 |
|
|
"TARGET_SSSE3"
|
9059 |
|
|
"ix86_fixup_binary_operands_no_copy (MULT, V4HImode, operands);")
|
9060 |
|
|
|
9061 |
|
|
(define_insn "*ssse3_pmulhrswv4hi3"
|
9062 |
|
|
[(set (match_operand:V4HI 0 "register_operand" "=y")
|
9063 |
|
|
(truncate:V4HI
|
9064 |
|
|
(lshiftrt:V4SI
|
9065 |
|
|
(plus:V4SI
|
9066 |
|
|
(lshiftrt:V4SI
|
9067 |
|
|
(mult:V4SI
|
9068 |
|
|
(sign_extend:V4SI
|
9069 |
|
|
(match_operand:V4HI 1 "nonimmediate_operand" "%0"))
|
9070 |
|
|
(sign_extend:V4SI
|
9071 |
|
|
(match_operand:V4HI 2 "nonimmediate_operand" "ym")))
|
9072 |
|
|
(const_int 14))
|
9073 |
|
|
(const_vector:V4HI [(const_int 1) (const_int 1)
|
9074 |
|
|
(const_int 1) (const_int 1)]))
|
9075 |
|
|
(const_int 1))))]
|
9076 |
|
|
"TARGET_SSSE3 && ix86_binary_operator_ok (MULT, V4HImode, operands)"
|
9077 |
|
|
"pmulhrsw\t{%2, %0|%0, %2}"
|
9078 |
|
|
[(set_attr "type" "sseimul")
|
9079 |
|
|
(set_attr "prefix_extra" "1")
|
9080 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
9081 |
|
|
(set_attr "mode" "DI")])
|
9082 |
|
|
|
9083 |
|
|
(define_insn "*avx_pshufbv16qi3"
|
9084 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
9085 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "x")
|
9086 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")]
|
9087 |
|
|
UNSPEC_PSHUFB))]
|
9088 |
|
|
"TARGET_AVX"
|
9089 |
|
|
"vpshufb\t{%2, %1, %0|%0, %1, %2}";
|
9090 |
|
|
[(set_attr "type" "sselog1")
|
9091 |
|
|
(set_attr "prefix_extra" "1")
|
9092 |
|
|
(set_attr "prefix" "vex")
|
9093 |
|
|
(set_attr "mode" "TI")])
|
9094 |
|
|
|
9095 |
|
|
(define_insn "ssse3_pshufbv16qi3"
|
9096 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
9097 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "0")
|
9098 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")]
|
9099 |
|
|
UNSPEC_PSHUFB))]
|
9100 |
|
|
"TARGET_SSSE3"
|
9101 |
|
|
"pshufb\t{%2, %0|%0, %2}";
|
9102 |
|
|
[(set_attr "type" "sselog1")
|
9103 |
|
|
(set_attr "prefix_data16" "1")
|
9104 |
|
|
(set_attr "prefix_extra" "1")
|
9105 |
|
|
(set_attr "mode" "TI")])
|
9106 |
|
|
|
9107 |
|
|
(define_insn "ssse3_pshufbv8qi3"
|
9108 |
|
|
[(set (match_operand:V8QI 0 "register_operand" "=y")
|
9109 |
|
|
(unspec:V8QI [(match_operand:V8QI 1 "register_operand" "0")
|
9110 |
|
|
(match_operand:V8QI 2 "nonimmediate_operand" "ym")]
|
9111 |
|
|
UNSPEC_PSHUFB))]
|
9112 |
|
|
"TARGET_SSSE3"
|
9113 |
|
|
"pshufb\t{%2, %0|%0, %2}";
|
9114 |
|
|
[(set_attr "type" "sselog1")
|
9115 |
|
|
(set_attr "prefix_extra" "1")
|
9116 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
9117 |
|
|
(set_attr "mode" "DI")])
|
9118 |
|
|
|
9119 |
|
|
(define_insn "*avx_psign3"
|
9120 |
|
|
[(set (match_operand:SSEMODE124 0 "register_operand" "=x")
|
9121 |
|
|
(unspec:SSEMODE124
|
9122 |
|
|
[(match_operand:SSEMODE124 1 "register_operand" "x")
|
9123 |
|
|
(match_operand:SSEMODE124 2 "nonimmediate_operand" "xm")]
|
9124 |
|
|
UNSPEC_PSIGN))]
|
9125 |
|
|
"TARGET_AVX"
|
9126 |
|
|
"vpsign\t{%2, %1, %0|%0, %1, %2}";
|
9127 |
|
|
[(set_attr "type" "sselog1")
|
9128 |
|
|
(set_attr "prefix_extra" "1")
|
9129 |
|
|
(set_attr "prefix" "vex")
|
9130 |
|
|
(set_attr "mode" "TI")])
|
9131 |
|
|
|
9132 |
|
|
(define_insn "ssse3_psign3"
|
9133 |
|
|
[(set (match_operand:SSEMODE124 0 "register_operand" "=x")
|
9134 |
|
|
(unspec:SSEMODE124
|
9135 |
|
|
[(match_operand:SSEMODE124 1 "register_operand" "0")
|
9136 |
|
|
(match_operand:SSEMODE124 2 "nonimmediate_operand" "xm")]
|
9137 |
|
|
UNSPEC_PSIGN))]
|
9138 |
|
|
"TARGET_SSSE3"
|
9139 |
|
|
"psign\t{%2, %0|%0, %2}";
|
9140 |
|
|
[(set_attr "type" "sselog1")
|
9141 |
|
|
(set_attr "prefix_data16" "1")
|
9142 |
|
|
(set_attr "prefix_extra" "1")
|
9143 |
|
|
(set_attr "mode" "TI")])
|
9144 |
|
|
|
9145 |
|
|
(define_insn "ssse3_psign3"
|
9146 |
|
|
[(set (match_operand:MMXMODEI 0 "register_operand" "=y")
|
9147 |
|
|
(unspec:MMXMODEI
|
9148 |
|
|
[(match_operand:MMXMODEI 1 "register_operand" "0")
|
9149 |
|
|
(match_operand:MMXMODEI 2 "nonimmediate_operand" "ym")]
|
9150 |
|
|
UNSPEC_PSIGN))]
|
9151 |
|
|
"TARGET_SSSE3"
|
9152 |
|
|
"psign\t{%2, %0|%0, %2}";
|
9153 |
|
|
[(set_attr "type" "sselog1")
|
9154 |
|
|
(set_attr "prefix_extra" "1")
|
9155 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
9156 |
|
|
(set_attr "mode" "DI")])
|
9157 |
|
|
|
9158 |
|
|
(define_insn "*avx_palignrti"
|
9159 |
|
|
[(set (match_operand:TI 0 "register_operand" "=x")
|
9160 |
|
|
(unspec:TI [(match_operand:TI 1 "register_operand" "x")
|
9161 |
|
|
(match_operand:TI 2 "nonimmediate_operand" "xm")
|
9162 |
|
|
(match_operand:SI 3 "const_0_to_255_mul_8_operand" "n")]
|
9163 |
|
|
UNSPEC_PALIGNR))]
|
9164 |
|
|
"TARGET_AVX"
|
9165 |
|
|
{
|
9166 |
|
|
operands[3] = GEN_INT (INTVAL (operands[3]) / 8);
|
9167 |
|
|
return "vpalignr\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
9168 |
|
|
}
|
9169 |
|
|
[(set_attr "type" "sseishft")
|
9170 |
|
|
(set_attr "prefix_extra" "1")
|
9171 |
|
|
(set_attr "length_immediate" "1")
|
9172 |
|
|
(set_attr "prefix" "vex")
|
9173 |
|
|
(set_attr "mode" "TI")])
|
9174 |
|
|
|
9175 |
|
|
(define_insn "ssse3_palignrti"
|
9176 |
|
|
[(set (match_operand:TI 0 "register_operand" "=x")
|
9177 |
|
|
(unspec:TI [(match_operand:TI 1 "register_operand" "0")
|
9178 |
|
|
(match_operand:TI 2 "nonimmediate_operand" "xm")
|
9179 |
|
|
(match_operand:SI 3 "const_0_to_255_mul_8_operand" "n")]
|
9180 |
|
|
UNSPEC_PALIGNR))]
|
9181 |
|
|
"TARGET_SSSE3"
|
9182 |
|
|
{
|
9183 |
|
|
operands[3] = GEN_INT (INTVAL (operands[3]) / 8);
|
9184 |
|
|
return "palignr\t{%3, %2, %0|%0, %2, %3}";
|
9185 |
|
|
}
|
9186 |
|
|
[(set_attr "type" "sseishft")
|
9187 |
|
|
(set_attr "atom_unit" "sishuf")
|
9188 |
|
|
(set_attr "prefix_data16" "1")
|
9189 |
|
|
(set_attr "prefix_extra" "1")
|
9190 |
|
|
(set_attr "length_immediate" "1")
|
9191 |
|
|
(set_attr "mode" "TI")])
|
9192 |
|
|
|
9193 |
|
|
(define_insn "ssse3_palignrdi"
|
9194 |
|
|
[(set (match_operand:DI 0 "register_operand" "=y")
|
9195 |
|
|
(unspec:DI [(match_operand:DI 1 "register_operand" "0")
|
9196 |
|
|
(match_operand:DI 2 "nonimmediate_operand" "ym")
|
9197 |
|
|
(match_operand:SI 3 "const_0_to_255_mul_8_operand" "n")]
|
9198 |
|
|
UNSPEC_PALIGNR))]
|
9199 |
|
|
"TARGET_SSSE3"
|
9200 |
|
|
{
|
9201 |
|
|
operands[3] = GEN_INT (INTVAL (operands[3]) / 8);
|
9202 |
|
|
return "palignr\t{%3, %2, %0|%0, %2, %3}";
|
9203 |
|
|
}
|
9204 |
|
|
[(set_attr "type" "sseishft")
|
9205 |
|
|
(set_attr "atom_unit" "sishuf")
|
9206 |
|
|
(set_attr "prefix_extra" "1")
|
9207 |
|
|
(set_attr "length_immediate" "1")
|
9208 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
9209 |
|
|
(set_attr "mode" "DI")])
|
9210 |
|
|
|
9211 |
|
|
(define_insn "abs2"
|
9212 |
|
|
[(set (match_operand:SSEMODE124 0 "register_operand" "=x")
|
9213 |
|
|
(abs:SSEMODE124 (match_operand:SSEMODE124 1 "nonimmediate_operand" "xm")))]
|
9214 |
|
|
"TARGET_SSSE3"
|
9215 |
|
|
"%vpabs\t{%1, %0|%0, %1}"
|
9216 |
|
|
[(set_attr "type" "sselog1")
|
9217 |
|
|
(set_attr "prefix_data16" "1")
|
9218 |
|
|
(set_attr "prefix_extra" "1")
|
9219 |
|
|
(set_attr "prefix" "maybe_vex")
|
9220 |
|
|
(set_attr "mode" "TI")])
|
9221 |
|
|
|
9222 |
|
|
(define_insn "abs2"
|
9223 |
|
|
[(set (match_operand:MMXMODEI 0 "register_operand" "=y")
|
9224 |
|
|
(abs:MMXMODEI (match_operand:MMXMODEI 1 "nonimmediate_operand" "ym")))]
|
9225 |
|
|
"TARGET_SSSE3"
|
9226 |
|
|
"pabs\t{%1, %0|%0, %1}";
|
9227 |
|
|
[(set_attr "type" "sselog1")
|
9228 |
|
|
(set_attr "prefix_rep" "0")
|
9229 |
|
|
(set_attr "prefix_extra" "1")
|
9230 |
|
|
(set (attr "prefix_rex") (symbol_ref "x86_extended_reg_mentioned_p (insn)"))
|
9231 |
|
|
(set_attr "mode" "DI")])
|
9232 |
|
|
|
9233 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
9234 |
|
|
;;
|
9235 |
|
|
;; AMD SSE4A instructions
|
9236 |
|
|
;;
|
9237 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
9238 |
|
|
|
9239 |
|
|
(define_insn "sse4a_movnt"
|
9240 |
|
|
[(set (match_operand:MODEF 0 "memory_operand" "=m")
|
9241 |
|
|
(unspec:MODEF
|
9242 |
|
|
[(match_operand:MODEF 1 "register_operand" "x")]
|
9243 |
|
|
UNSPEC_MOVNT))]
|
9244 |
|
|
"TARGET_SSE4A"
|
9245 |
|
|
"movnts\t{%1, %0|%0, %1}"
|
9246 |
|
|
[(set_attr "type" "ssemov")
|
9247 |
|
|
(set_attr "mode" "")])
|
9248 |
|
|
|
9249 |
|
|
(define_insn "sse4a_vmmovnt"
|
9250 |
|
|
[(set (match_operand: 0 "memory_operand" "=m")
|
9251 |
|
|
(unspec:
|
9252 |
|
|
[(vec_select:
|
9253 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
9254 |
|
|
(parallel [(const_int 0)]))]
|
9255 |
|
|
UNSPEC_MOVNT))]
|
9256 |
|
|
"TARGET_SSE4A"
|
9257 |
|
|
"movnts\t{%1, %0|%0, %1}"
|
9258 |
|
|
[(set_attr "type" "ssemov")
|
9259 |
|
|
(set_attr "mode" "")])
|
9260 |
|
|
|
9261 |
|
|
(define_insn "sse4a_extrqi"
|
9262 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9263 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
9264 |
|
|
(match_operand 2 "const_int_operand" "")
|
9265 |
|
|
(match_operand 3 "const_int_operand" "")]
|
9266 |
|
|
UNSPEC_EXTRQI))]
|
9267 |
|
|
"TARGET_SSE4A"
|
9268 |
|
|
"extrq\t{%3, %2, %0|%0, %2, %3}"
|
9269 |
|
|
[(set_attr "type" "sse")
|
9270 |
|
|
(set_attr "prefix_data16" "1")
|
9271 |
|
|
(set_attr "length_immediate" "2")
|
9272 |
|
|
(set_attr "mode" "TI")])
|
9273 |
|
|
|
9274 |
|
|
(define_insn "sse4a_extrq"
|
9275 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9276 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
9277 |
|
|
(match_operand:V16QI 2 "register_operand" "x")]
|
9278 |
|
|
UNSPEC_EXTRQ))]
|
9279 |
|
|
"TARGET_SSE4A"
|
9280 |
|
|
"extrq\t{%2, %0|%0, %2}"
|
9281 |
|
|
[(set_attr "type" "sse")
|
9282 |
|
|
(set_attr "prefix_data16" "1")
|
9283 |
|
|
(set_attr "mode" "TI")])
|
9284 |
|
|
|
9285 |
|
|
(define_insn "sse4a_insertqi"
|
9286 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9287 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
9288 |
|
|
(match_operand:V2DI 2 "register_operand" "x")
|
9289 |
|
|
(match_operand 3 "const_int_operand" "")
|
9290 |
|
|
(match_operand 4 "const_int_operand" "")]
|
9291 |
|
|
UNSPEC_INSERTQI))]
|
9292 |
|
|
"TARGET_SSE4A"
|
9293 |
|
|
"insertq\t{%4, %3, %2, %0|%0, %2, %3, %4}"
|
9294 |
|
|
[(set_attr "type" "sseins")
|
9295 |
|
|
(set_attr "prefix_data16" "0")
|
9296 |
|
|
(set_attr "prefix_rep" "1")
|
9297 |
|
|
(set_attr "length_immediate" "2")
|
9298 |
|
|
(set_attr "mode" "TI")])
|
9299 |
|
|
|
9300 |
|
|
(define_insn "sse4a_insertq"
|
9301 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9302 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
9303 |
|
|
(match_operand:V2DI 2 "register_operand" "x")]
|
9304 |
|
|
UNSPEC_INSERTQ))]
|
9305 |
|
|
"TARGET_SSE4A"
|
9306 |
|
|
"insertq\t{%2, %0|%0, %2}"
|
9307 |
|
|
[(set_attr "type" "sseins")
|
9308 |
|
|
(set_attr "prefix_data16" "0")
|
9309 |
|
|
(set_attr "prefix_rep" "1")
|
9310 |
|
|
(set_attr "mode" "TI")])
|
9311 |
|
|
|
9312 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
9313 |
|
|
;;
|
9314 |
|
|
;; Intel SSE4.1 instructions
|
9315 |
|
|
;;
|
9316 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
9317 |
|
|
|
9318 |
|
|
(define_insn "avx_blendp"
|
9319 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
9320 |
|
|
(vec_merge:AVXMODEF2P
|
9321 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")
|
9322 |
|
|
(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
9323 |
|
|
(match_operand:SI 3 "const_0_to__operand" "n")))]
|
9324 |
|
|
"TARGET_AVX"
|
9325 |
|
|
"vblendp\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
9326 |
|
|
[(set_attr "type" "ssemov")
|
9327 |
|
|
(set_attr "prefix_extra" "1")
|
9328 |
|
|
(set_attr "length_immediate" "1")
|
9329 |
|
|
(set_attr "prefix" "vex")
|
9330 |
|
|
(set_attr "mode" "")])
|
9331 |
|
|
|
9332 |
|
|
(define_insn "avx_blendvp"
|
9333 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
9334 |
|
|
(unspec:AVXMODEF2P
|
9335 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
9336 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")
|
9337 |
|
|
(match_operand:AVXMODEF2P 3 "register_operand" "x")]
|
9338 |
|
|
UNSPEC_BLENDV))]
|
9339 |
|
|
"TARGET_AVX"
|
9340 |
|
|
"vblendvp\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
9341 |
|
|
[(set_attr "type" "ssemov")
|
9342 |
|
|
(set_attr "prefix_extra" "1")
|
9343 |
|
|
(set_attr "length_immediate" "1")
|
9344 |
|
|
(set_attr "prefix" "vex")
|
9345 |
|
|
(set_attr "mode" "")])
|
9346 |
|
|
|
9347 |
|
|
(define_insn "sse4_1_blendp"
|
9348 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
9349 |
|
|
(vec_merge:SSEMODEF2P
|
9350 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")
|
9351 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
9352 |
|
|
(match_operand:SI 3 "const_0_to__operand" "n")))]
|
9353 |
|
|
"TARGET_SSE4_1"
|
9354 |
|
|
"blendp\t{%3, %2, %0|%0, %2, %3}"
|
9355 |
|
|
[(set_attr "type" "ssemov")
|
9356 |
|
|
(set_attr "prefix_data16" "1")
|
9357 |
|
|
(set_attr "prefix_extra" "1")
|
9358 |
|
|
(set_attr "length_immediate" "1")
|
9359 |
|
|
(set_attr "mode" "")])
|
9360 |
|
|
|
9361 |
|
|
(define_insn "sse4_1_blendvp"
|
9362 |
|
|
[(set (match_operand:SSEMODEF2P 0 "reg_not_xmm0_operand" "=x")
|
9363 |
|
|
(unspec:SSEMODEF2P
|
9364 |
|
|
[(match_operand:SSEMODEF2P 1 "reg_not_xmm0_operand" "0")
|
9365 |
|
|
(match_operand:SSEMODEF2P 2 "nonimm_not_xmm0_operand" "xm")
|
9366 |
|
|
(match_operand:SSEMODEF2P 3 "register_operand" "Yz")]
|
9367 |
|
|
UNSPEC_BLENDV))]
|
9368 |
|
|
"TARGET_SSE4_1"
|
9369 |
|
|
"blendvp\t{%3, %2, %0|%0, %2, %3}"
|
9370 |
|
|
[(set_attr "type" "ssemov")
|
9371 |
|
|
(set_attr "prefix_data16" "1")
|
9372 |
|
|
(set_attr "prefix_extra" "1")
|
9373 |
|
|
(set_attr "mode" "")])
|
9374 |
|
|
|
9375 |
|
|
(define_insn "avx_dpp"
|
9376 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
9377 |
|
|
(unspec:AVXMODEF2P
|
9378 |
|
|
[(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "%x")
|
9379 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "xm")
|
9380 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
9381 |
|
|
UNSPEC_DP))]
|
9382 |
|
|
"TARGET_AVX"
|
9383 |
|
|
"vdpp\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
9384 |
|
|
[(set_attr "type" "ssemul")
|
9385 |
|
|
(set_attr "prefix" "vex")
|
9386 |
|
|
(set_attr "prefix_extra" "1")
|
9387 |
|
|
(set_attr "length_immediate" "1")
|
9388 |
|
|
(set_attr "mode" "")])
|
9389 |
|
|
|
9390 |
|
|
(define_insn "sse4_1_dpp"
|
9391 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
9392 |
|
|
(unspec:SSEMODEF2P
|
9393 |
|
|
[(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "%0")
|
9394 |
|
|
(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")
|
9395 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
9396 |
|
|
UNSPEC_DP))]
|
9397 |
|
|
"TARGET_SSE4_1"
|
9398 |
|
|
"dpp\t{%3, %2, %0|%0, %2, %3}"
|
9399 |
|
|
[(set_attr "type" "ssemul")
|
9400 |
|
|
(set_attr "prefix_data16" "1")
|
9401 |
|
|
(set_attr "prefix_extra" "1")
|
9402 |
|
|
(set_attr "length_immediate" "1")
|
9403 |
|
|
(set_attr "mode" "")])
|
9404 |
|
|
|
9405 |
|
|
(define_insn "sse4_1_movntdqa"
|
9406 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9407 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "memory_operand" "m")]
|
9408 |
|
|
UNSPEC_MOVNTDQA))]
|
9409 |
|
|
"TARGET_SSE4_1"
|
9410 |
|
|
"%vmovntdqa\t{%1, %0|%0, %1}"
|
9411 |
|
|
[(set_attr "type" "ssemov")
|
9412 |
|
|
(set_attr "prefix_extra" "1")
|
9413 |
|
|
(set_attr "prefix" "maybe_vex")
|
9414 |
|
|
(set_attr "mode" "TI")])
|
9415 |
|
|
|
9416 |
|
|
(define_insn "*avx_mpsadbw"
|
9417 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
9418 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "x")
|
9419 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")
|
9420 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
9421 |
|
|
UNSPEC_MPSADBW))]
|
9422 |
|
|
"TARGET_AVX"
|
9423 |
|
|
"vmpsadbw\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
9424 |
|
|
[(set_attr "type" "sselog1")
|
9425 |
|
|
(set_attr "prefix" "vex")
|
9426 |
|
|
(set_attr "prefix_extra" "1")
|
9427 |
|
|
(set_attr "length_immediate" "1")
|
9428 |
|
|
(set_attr "mode" "TI")])
|
9429 |
|
|
|
9430 |
|
|
(define_insn "sse4_1_mpsadbw"
|
9431 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
9432 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "0")
|
9433 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")
|
9434 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
9435 |
|
|
UNSPEC_MPSADBW))]
|
9436 |
|
|
"TARGET_SSE4_1"
|
9437 |
|
|
"mpsadbw\t{%3, %2, %0|%0, %2, %3}"
|
9438 |
|
|
[(set_attr "type" "sselog1")
|
9439 |
|
|
(set_attr "prefix_extra" "1")
|
9440 |
|
|
(set_attr "length_immediate" "1")
|
9441 |
|
|
(set_attr "mode" "TI")])
|
9442 |
|
|
|
9443 |
|
|
(define_insn "*avx_packusdw"
|
9444 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9445 |
|
|
(vec_concat:V8HI
|
9446 |
|
|
(us_truncate:V4HI
|
9447 |
|
|
(match_operand:V4SI 1 "register_operand" "x"))
|
9448 |
|
|
(us_truncate:V4HI
|
9449 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))))]
|
9450 |
|
|
"TARGET_AVX"
|
9451 |
|
|
"vpackusdw\t{%2, %1, %0|%0, %1, %2}"
|
9452 |
|
|
[(set_attr "type" "sselog")
|
9453 |
|
|
(set_attr "prefix_extra" "1")
|
9454 |
|
|
(set_attr "prefix" "vex")
|
9455 |
|
|
(set_attr "mode" "TI")])
|
9456 |
|
|
|
9457 |
|
|
(define_insn "sse4_1_packusdw"
|
9458 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9459 |
|
|
(vec_concat:V8HI
|
9460 |
|
|
(us_truncate:V4HI
|
9461 |
|
|
(match_operand:V4SI 1 "register_operand" "0"))
|
9462 |
|
|
(us_truncate:V4HI
|
9463 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))))]
|
9464 |
|
|
"TARGET_SSE4_1"
|
9465 |
|
|
"packusdw\t{%2, %0|%0, %2}"
|
9466 |
|
|
[(set_attr "type" "sselog")
|
9467 |
|
|
(set_attr "prefix_extra" "1")
|
9468 |
|
|
(set_attr "mode" "TI")])
|
9469 |
|
|
|
9470 |
|
|
(define_insn "*avx_pblendvb"
|
9471 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x")
|
9472 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "register_operand" "x")
|
9473 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")
|
9474 |
|
|
(match_operand:V16QI 3 "register_operand" "x")]
|
9475 |
|
|
UNSPEC_BLENDV))]
|
9476 |
|
|
"TARGET_AVX"
|
9477 |
|
|
"vpblendvb\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
9478 |
|
|
[(set_attr "type" "ssemov")
|
9479 |
|
|
(set_attr "prefix_extra" "1")
|
9480 |
|
|
(set_attr "length_immediate" "1")
|
9481 |
|
|
(set_attr "prefix" "vex")
|
9482 |
|
|
(set_attr "mode" "TI")])
|
9483 |
|
|
|
9484 |
|
|
(define_insn "sse4_1_pblendvb"
|
9485 |
|
|
[(set (match_operand:V16QI 0 "reg_not_xmm0_operand" "=x")
|
9486 |
|
|
(unspec:V16QI [(match_operand:V16QI 1 "reg_not_xmm0_operand" "0")
|
9487 |
|
|
(match_operand:V16QI 2 "nonimm_not_xmm0_operand" "xm")
|
9488 |
|
|
(match_operand:V16QI 3 "register_operand" "Yz")]
|
9489 |
|
|
UNSPEC_BLENDV))]
|
9490 |
|
|
"TARGET_SSE4_1"
|
9491 |
|
|
"pblendvb\t{%3, %2, %0|%0, %2, %3}"
|
9492 |
|
|
[(set_attr "type" "ssemov")
|
9493 |
|
|
(set_attr "prefix_extra" "1")
|
9494 |
|
|
(set_attr "mode" "TI")])
|
9495 |
|
|
|
9496 |
|
|
(define_insn "*avx_pblendw"
|
9497 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9498 |
|
|
(vec_merge:V8HI
|
9499 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
9500 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
9501 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")))]
|
9502 |
|
|
"TARGET_AVX"
|
9503 |
|
|
"vpblendw\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
9504 |
|
|
[(set_attr "type" "ssemov")
|
9505 |
|
|
(set_attr "prefix" "vex")
|
9506 |
|
|
(set_attr "prefix_extra" "1")
|
9507 |
|
|
(set_attr "length_immediate" "1")
|
9508 |
|
|
(set_attr "mode" "TI")])
|
9509 |
|
|
|
9510 |
|
|
(define_insn "sse4_1_pblendw"
|
9511 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9512 |
|
|
(vec_merge:V8HI
|
9513 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
9514 |
|
|
(match_operand:V8HI 1 "register_operand" "0")
|
9515 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")))]
|
9516 |
|
|
"TARGET_SSE4_1"
|
9517 |
|
|
"pblendw\t{%3, %2, %0|%0, %2, %3}"
|
9518 |
|
|
[(set_attr "type" "ssemov")
|
9519 |
|
|
(set_attr "prefix_extra" "1")
|
9520 |
|
|
(set_attr "length_immediate" "1")
|
9521 |
|
|
(set_attr "mode" "TI")])
|
9522 |
|
|
|
9523 |
|
|
(define_insn "sse4_1_phminposuw"
|
9524 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9525 |
|
|
(unspec:V8HI [(match_operand:V8HI 1 "nonimmediate_operand" "xm")]
|
9526 |
|
|
UNSPEC_PHMINPOSUW))]
|
9527 |
|
|
"TARGET_SSE4_1"
|
9528 |
|
|
"%vphminposuw\t{%1, %0|%0, %1}"
|
9529 |
|
|
[(set_attr "type" "sselog1")
|
9530 |
|
|
(set_attr "prefix_extra" "1")
|
9531 |
|
|
(set_attr "prefix" "maybe_vex")
|
9532 |
|
|
(set_attr "mode" "TI")])
|
9533 |
|
|
|
9534 |
|
|
(define_insn "sse4_1_extendv8qiv8hi2"
|
9535 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9536 |
|
|
(sign_extend:V8HI
|
9537 |
|
|
(vec_select:V8QI
|
9538 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
9539 |
|
|
(parallel [(const_int 0)
|
9540 |
|
|
(const_int 1)
|
9541 |
|
|
(const_int 2)
|
9542 |
|
|
(const_int 3)
|
9543 |
|
|
(const_int 4)
|
9544 |
|
|
(const_int 5)
|
9545 |
|
|
(const_int 6)
|
9546 |
|
|
(const_int 7)]))))]
|
9547 |
|
|
"TARGET_SSE4_1"
|
9548 |
|
|
"%vpmovsxbw\t{%1, %0|%0, %1}"
|
9549 |
|
|
[(set_attr "type" "ssemov")
|
9550 |
|
|
(set_attr "prefix_extra" "1")
|
9551 |
|
|
(set_attr "prefix" "maybe_vex")
|
9552 |
|
|
(set_attr "mode" "TI")])
|
9553 |
|
|
|
9554 |
|
|
(define_insn "*sse4_1_extendv8qiv8hi2"
|
9555 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9556 |
|
|
(sign_extend:V8HI
|
9557 |
|
|
(vec_select:V8QI
|
9558 |
|
|
(vec_duplicate:V16QI
|
9559 |
|
|
(match_operand:V8QI 1 "nonimmediate_operand" "xm"))
|
9560 |
|
|
(parallel [(const_int 0)
|
9561 |
|
|
(const_int 1)
|
9562 |
|
|
(const_int 2)
|
9563 |
|
|
(const_int 3)
|
9564 |
|
|
(const_int 4)
|
9565 |
|
|
(const_int 5)
|
9566 |
|
|
(const_int 6)
|
9567 |
|
|
(const_int 7)]))))]
|
9568 |
|
|
"TARGET_SSE4_1"
|
9569 |
|
|
"%vpmovsxbw\t{%1, %0|%0, %1}"
|
9570 |
|
|
[(set_attr "type" "ssemov")
|
9571 |
|
|
(set_attr "prefix_extra" "1")
|
9572 |
|
|
(set_attr "prefix" "maybe_vex")
|
9573 |
|
|
(set_attr "mode" "TI")])
|
9574 |
|
|
|
9575 |
|
|
(define_insn "sse4_1_extendv4qiv4si2"
|
9576 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
9577 |
|
|
(sign_extend:V4SI
|
9578 |
|
|
(vec_select:V4QI
|
9579 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
9580 |
|
|
(parallel [(const_int 0)
|
9581 |
|
|
(const_int 1)
|
9582 |
|
|
(const_int 2)
|
9583 |
|
|
(const_int 3)]))))]
|
9584 |
|
|
"TARGET_SSE4_1"
|
9585 |
|
|
"%vpmovsxbd\t{%1, %0|%0, %1}"
|
9586 |
|
|
[(set_attr "type" "ssemov")
|
9587 |
|
|
(set_attr "prefix_extra" "1")
|
9588 |
|
|
(set_attr "prefix" "maybe_vex")
|
9589 |
|
|
(set_attr "mode" "TI")])
|
9590 |
|
|
|
9591 |
|
|
(define_insn "*sse4_1_extendv4qiv4si2"
|
9592 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
9593 |
|
|
(sign_extend:V4SI
|
9594 |
|
|
(vec_select:V4QI
|
9595 |
|
|
(vec_duplicate:V16QI
|
9596 |
|
|
(match_operand:V4QI 1 "nonimmediate_operand" "xm"))
|
9597 |
|
|
(parallel [(const_int 0)
|
9598 |
|
|
(const_int 1)
|
9599 |
|
|
(const_int 2)
|
9600 |
|
|
(const_int 3)]))))]
|
9601 |
|
|
"TARGET_SSE4_1"
|
9602 |
|
|
"%vpmovsxbd\t{%1, %0|%0, %1}"
|
9603 |
|
|
[(set_attr "type" "ssemov")
|
9604 |
|
|
(set_attr "prefix_extra" "1")
|
9605 |
|
|
(set_attr "prefix" "maybe_vex")
|
9606 |
|
|
(set_attr "mode" "TI")])
|
9607 |
|
|
|
9608 |
|
|
(define_insn "sse4_1_extendv2qiv2di2"
|
9609 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9610 |
|
|
(sign_extend:V2DI
|
9611 |
|
|
(vec_select:V2QI
|
9612 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
9613 |
|
|
(parallel [(const_int 0)
|
9614 |
|
|
(const_int 1)]))))]
|
9615 |
|
|
"TARGET_SSE4_1"
|
9616 |
|
|
"%vpmovsxbq\t{%1, %0|%0, %1}"
|
9617 |
|
|
[(set_attr "type" "ssemov")
|
9618 |
|
|
(set_attr "prefix_extra" "1")
|
9619 |
|
|
(set_attr "prefix" "maybe_vex")
|
9620 |
|
|
(set_attr "mode" "TI")])
|
9621 |
|
|
|
9622 |
|
|
(define_insn "*sse4_1_extendv2qiv2di2"
|
9623 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9624 |
|
|
(sign_extend:V2DI
|
9625 |
|
|
(vec_select:V2QI
|
9626 |
|
|
(vec_duplicate:V16QI
|
9627 |
|
|
(match_operand:V2QI 1 "nonimmediate_operand" "xm"))
|
9628 |
|
|
(parallel [(const_int 0)
|
9629 |
|
|
(const_int 1)]))))]
|
9630 |
|
|
"TARGET_SSE4_1"
|
9631 |
|
|
"%vpmovsxbq\t{%1, %0|%0, %1}"
|
9632 |
|
|
[(set_attr "type" "ssemov")
|
9633 |
|
|
(set_attr "prefix_extra" "1")
|
9634 |
|
|
(set_attr "prefix" "maybe_vex")
|
9635 |
|
|
(set_attr "mode" "TI")])
|
9636 |
|
|
|
9637 |
|
|
(define_insn "sse4_1_extendv4hiv4si2"
|
9638 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
9639 |
|
|
(sign_extend:V4SI
|
9640 |
|
|
(vec_select:V4HI
|
9641 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
9642 |
|
|
(parallel [(const_int 0)
|
9643 |
|
|
(const_int 1)
|
9644 |
|
|
(const_int 2)
|
9645 |
|
|
(const_int 3)]))))]
|
9646 |
|
|
"TARGET_SSE4_1"
|
9647 |
|
|
"%vpmovsxwd\t{%1, %0|%0, %1}"
|
9648 |
|
|
[(set_attr "type" "ssemov")
|
9649 |
|
|
(set_attr "prefix_extra" "1")
|
9650 |
|
|
(set_attr "prefix" "maybe_vex")
|
9651 |
|
|
(set_attr "mode" "TI")])
|
9652 |
|
|
|
9653 |
|
|
(define_insn "*sse4_1_extendv4hiv4si2"
|
9654 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
9655 |
|
|
(sign_extend:V4SI
|
9656 |
|
|
(vec_select:V4HI
|
9657 |
|
|
(vec_duplicate:V8HI
|
9658 |
|
|
(match_operand:V2HI 1 "nonimmediate_operand" "xm"))
|
9659 |
|
|
(parallel [(const_int 0)
|
9660 |
|
|
(const_int 1)
|
9661 |
|
|
(const_int 2)
|
9662 |
|
|
(const_int 3)]))))]
|
9663 |
|
|
"TARGET_SSE4_1"
|
9664 |
|
|
"%vpmovsxwd\t{%1, %0|%0, %1}"
|
9665 |
|
|
[(set_attr "type" "ssemov")
|
9666 |
|
|
(set_attr "prefix_extra" "1")
|
9667 |
|
|
(set_attr "prefix" "maybe_vex")
|
9668 |
|
|
(set_attr "mode" "TI")])
|
9669 |
|
|
|
9670 |
|
|
(define_insn "sse4_1_extendv2hiv2di2"
|
9671 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9672 |
|
|
(sign_extend:V2DI
|
9673 |
|
|
(vec_select:V2HI
|
9674 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
9675 |
|
|
(parallel [(const_int 0)
|
9676 |
|
|
(const_int 1)]))))]
|
9677 |
|
|
"TARGET_SSE4_1"
|
9678 |
|
|
"%vpmovsxwq\t{%1, %0|%0, %1}"
|
9679 |
|
|
[(set_attr "type" "ssemov")
|
9680 |
|
|
(set_attr "prefix_extra" "1")
|
9681 |
|
|
(set_attr "prefix" "maybe_vex")
|
9682 |
|
|
(set_attr "mode" "TI")])
|
9683 |
|
|
|
9684 |
|
|
(define_insn "*sse4_1_extendv2hiv2di2"
|
9685 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9686 |
|
|
(sign_extend:V2DI
|
9687 |
|
|
(vec_select:V2HI
|
9688 |
|
|
(vec_duplicate:V8HI
|
9689 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "xm"))
|
9690 |
|
|
(parallel [(const_int 0)
|
9691 |
|
|
(const_int 1)]))))]
|
9692 |
|
|
"TARGET_SSE4_1"
|
9693 |
|
|
"%vpmovsxwq\t{%1, %0|%0, %1}"
|
9694 |
|
|
[(set_attr "type" "ssemov")
|
9695 |
|
|
(set_attr "prefix_extra" "1")
|
9696 |
|
|
(set_attr "prefix" "maybe_vex")
|
9697 |
|
|
(set_attr "mode" "TI")])
|
9698 |
|
|
|
9699 |
|
|
(define_insn "sse4_1_extendv2siv2di2"
|
9700 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9701 |
|
|
(sign_extend:V2DI
|
9702 |
|
|
(vec_select:V2SI
|
9703 |
|
|
(match_operand:V4SI 1 "register_operand" "x")
|
9704 |
|
|
(parallel [(const_int 0)
|
9705 |
|
|
(const_int 1)]))))]
|
9706 |
|
|
"TARGET_SSE4_1"
|
9707 |
|
|
"%vpmovsxdq\t{%1, %0|%0, %1}"
|
9708 |
|
|
[(set_attr "type" "ssemov")
|
9709 |
|
|
(set_attr "prefix_extra" "1")
|
9710 |
|
|
(set_attr "prefix" "maybe_vex")
|
9711 |
|
|
(set_attr "mode" "TI")])
|
9712 |
|
|
|
9713 |
|
|
(define_insn "*sse4_1_extendv2siv2di2"
|
9714 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9715 |
|
|
(sign_extend:V2DI
|
9716 |
|
|
(vec_select:V2SI
|
9717 |
|
|
(vec_duplicate:V4SI
|
9718 |
|
|
(match_operand:V2SI 1 "nonimmediate_operand" "xm"))
|
9719 |
|
|
(parallel [(const_int 0)
|
9720 |
|
|
(const_int 1)]))))]
|
9721 |
|
|
"TARGET_SSE4_1"
|
9722 |
|
|
"%vpmovsxdq\t{%1, %0|%0, %1}"
|
9723 |
|
|
[(set_attr "type" "ssemov")
|
9724 |
|
|
(set_attr "prefix_extra" "1")
|
9725 |
|
|
(set_attr "prefix" "maybe_vex")
|
9726 |
|
|
(set_attr "mode" "TI")])
|
9727 |
|
|
|
9728 |
|
|
(define_insn "sse4_1_zero_extendv8qiv8hi2"
|
9729 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9730 |
|
|
(zero_extend:V8HI
|
9731 |
|
|
(vec_select:V8QI
|
9732 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
9733 |
|
|
(parallel [(const_int 0)
|
9734 |
|
|
(const_int 1)
|
9735 |
|
|
(const_int 2)
|
9736 |
|
|
(const_int 3)
|
9737 |
|
|
(const_int 4)
|
9738 |
|
|
(const_int 5)
|
9739 |
|
|
(const_int 6)
|
9740 |
|
|
(const_int 7)]))))]
|
9741 |
|
|
"TARGET_SSE4_1"
|
9742 |
|
|
"%vpmovzxbw\t{%1, %0|%0, %1}"
|
9743 |
|
|
[(set_attr "type" "ssemov")
|
9744 |
|
|
(set_attr "prefix_extra" "1")
|
9745 |
|
|
(set_attr "prefix" "maybe_vex")
|
9746 |
|
|
(set_attr "mode" "TI")])
|
9747 |
|
|
|
9748 |
|
|
(define_insn "*sse4_1_zero_extendv8qiv8hi2"
|
9749 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
9750 |
|
|
(zero_extend:V8HI
|
9751 |
|
|
(vec_select:V8QI
|
9752 |
|
|
(vec_duplicate:V16QI
|
9753 |
|
|
(match_operand:V8QI 1 "nonimmediate_operand" "xm"))
|
9754 |
|
|
(parallel [(const_int 0)
|
9755 |
|
|
(const_int 1)
|
9756 |
|
|
(const_int 2)
|
9757 |
|
|
(const_int 3)
|
9758 |
|
|
(const_int 4)
|
9759 |
|
|
(const_int 5)
|
9760 |
|
|
(const_int 6)
|
9761 |
|
|
(const_int 7)]))))]
|
9762 |
|
|
"TARGET_SSE4_1"
|
9763 |
|
|
"%vpmovzxbw\t{%1, %0|%0, %1}"
|
9764 |
|
|
[(set_attr "type" "ssemov")
|
9765 |
|
|
(set_attr "prefix_extra" "1")
|
9766 |
|
|
(set_attr "prefix" "maybe_vex")
|
9767 |
|
|
(set_attr "mode" "TI")])
|
9768 |
|
|
|
9769 |
|
|
(define_insn "sse4_1_zero_extendv4qiv4si2"
|
9770 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
9771 |
|
|
(zero_extend:V4SI
|
9772 |
|
|
(vec_select:V4QI
|
9773 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
9774 |
|
|
(parallel [(const_int 0)
|
9775 |
|
|
(const_int 1)
|
9776 |
|
|
(const_int 2)
|
9777 |
|
|
(const_int 3)]))))]
|
9778 |
|
|
"TARGET_SSE4_1"
|
9779 |
|
|
"%vpmovzxbd\t{%1, %0|%0, %1}"
|
9780 |
|
|
[(set_attr "type" "ssemov")
|
9781 |
|
|
(set_attr "prefix_extra" "1")
|
9782 |
|
|
(set_attr "prefix" "maybe_vex")
|
9783 |
|
|
(set_attr "mode" "TI")])
|
9784 |
|
|
|
9785 |
|
|
(define_insn "*sse4_1_zero_extendv4qiv4si2"
|
9786 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
9787 |
|
|
(zero_extend:V4SI
|
9788 |
|
|
(vec_select:V4QI
|
9789 |
|
|
(vec_duplicate:V16QI
|
9790 |
|
|
(match_operand:V4QI 1 "nonimmediate_operand" "xm"))
|
9791 |
|
|
(parallel [(const_int 0)
|
9792 |
|
|
(const_int 1)
|
9793 |
|
|
(const_int 2)
|
9794 |
|
|
(const_int 3)]))))]
|
9795 |
|
|
"TARGET_SSE4_1"
|
9796 |
|
|
"%vpmovzxbd\t{%1, %0|%0, %1}"
|
9797 |
|
|
[(set_attr "type" "ssemov")
|
9798 |
|
|
(set_attr "prefix_extra" "1")
|
9799 |
|
|
(set_attr "prefix" "maybe_vex")
|
9800 |
|
|
(set_attr "mode" "TI")])
|
9801 |
|
|
|
9802 |
|
|
(define_insn "sse4_1_zero_extendv2qiv2di2"
|
9803 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9804 |
|
|
(zero_extend:V2DI
|
9805 |
|
|
(vec_select:V2QI
|
9806 |
|
|
(match_operand:V16QI 1 "register_operand" "x")
|
9807 |
|
|
(parallel [(const_int 0)
|
9808 |
|
|
(const_int 1)]))))]
|
9809 |
|
|
"TARGET_SSE4_1"
|
9810 |
|
|
"%vpmovzxbq\t{%1, %0|%0, %1}"
|
9811 |
|
|
[(set_attr "type" "ssemov")
|
9812 |
|
|
(set_attr "prefix_extra" "1")
|
9813 |
|
|
(set_attr "prefix" "maybe_vex")
|
9814 |
|
|
(set_attr "mode" "TI")])
|
9815 |
|
|
|
9816 |
|
|
(define_insn "*sse4_1_zero_extendv2qiv2di2"
|
9817 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9818 |
|
|
(zero_extend:V2DI
|
9819 |
|
|
(vec_select:V2QI
|
9820 |
|
|
(vec_duplicate:V16QI
|
9821 |
|
|
(match_operand:V2QI 1 "nonimmediate_operand" "xm"))
|
9822 |
|
|
(parallel [(const_int 0)
|
9823 |
|
|
(const_int 1)]))))]
|
9824 |
|
|
"TARGET_SSE4_1"
|
9825 |
|
|
"%vpmovzxbq\t{%1, %0|%0, %1}"
|
9826 |
|
|
[(set_attr "type" "ssemov")
|
9827 |
|
|
(set_attr "prefix_extra" "1")
|
9828 |
|
|
(set_attr "prefix" "maybe_vex")
|
9829 |
|
|
(set_attr "mode" "TI")])
|
9830 |
|
|
|
9831 |
|
|
(define_insn "sse4_1_zero_extendv4hiv4si2"
|
9832 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
9833 |
|
|
(zero_extend:V4SI
|
9834 |
|
|
(vec_select:V4HI
|
9835 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
9836 |
|
|
(parallel [(const_int 0)
|
9837 |
|
|
(const_int 1)
|
9838 |
|
|
(const_int 2)
|
9839 |
|
|
(const_int 3)]))))]
|
9840 |
|
|
"TARGET_SSE4_1"
|
9841 |
|
|
"%vpmovzxwd\t{%1, %0|%0, %1}"
|
9842 |
|
|
[(set_attr "type" "ssemov")
|
9843 |
|
|
(set_attr "prefix_extra" "1")
|
9844 |
|
|
(set_attr "prefix" "maybe_vex")
|
9845 |
|
|
(set_attr "mode" "TI")])
|
9846 |
|
|
|
9847 |
|
|
(define_insn "*sse4_1_zero_extendv4hiv4si2"
|
9848 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
9849 |
|
|
(zero_extend:V4SI
|
9850 |
|
|
(vec_select:V4HI
|
9851 |
|
|
(vec_duplicate:V8HI
|
9852 |
|
|
(match_operand:V4HI 1 "nonimmediate_operand" "xm"))
|
9853 |
|
|
(parallel [(const_int 0)
|
9854 |
|
|
(const_int 1)
|
9855 |
|
|
(const_int 2)
|
9856 |
|
|
(const_int 3)]))))]
|
9857 |
|
|
"TARGET_SSE4_1"
|
9858 |
|
|
"%vpmovzxwd\t{%1, %0|%0, %1}"
|
9859 |
|
|
[(set_attr "type" "ssemov")
|
9860 |
|
|
(set_attr "prefix_extra" "1")
|
9861 |
|
|
(set_attr "prefix" "maybe_vex")
|
9862 |
|
|
(set_attr "mode" "TI")])
|
9863 |
|
|
|
9864 |
|
|
(define_insn "sse4_1_zero_extendv2hiv2di2"
|
9865 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9866 |
|
|
(zero_extend:V2DI
|
9867 |
|
|
(vec_select:V2HI
|
9868 |
|
|
(match_operand:V8HI 1 "register_operand" "x")
|
9869 |
|
|
(parallel [(const_int 0)
|
9870 |
|
|
(const_int 1)]))))]
|
9871 |
|
|
"TARGET_SSE4_1"
|
9872 |
|
|
"%vpmovzxwq\t{%1, %0|%0, %1}"
|
9873 |
|
|
[(set_attr "type" "ssemov")
|
9874 |
|
|
(set_attr "prefix_extra" "1")
|
9875 |
|
|
(set_attr "prefix" "maybe_vex")
|
9876 |
|
|
(set_attr "mode" "TI")])
|
9877 |
|
|
|
9878 |
|
|
(define_insn "*sse4_1_zero_extendv2hiv2di2"
|
9879 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9880 |
|
|
(zero_extend:V2DI
|
9881 |
|
|
(vec_select:V2HI
|
9882 |
|
|
(vec_duplicate:V8HI
|
9883 |
|
|
(match_operand:V2HI 1 "nonimmediate_operand" "xm"))
|
9884 |
|
|
(parallel [(const_int 0)
|
9885 |
|
|
(const_int 1)]))))]
|
9886 |
|
|
"TARGET_SSE4_1"
|
9887 |
|
|
"%vpmovzxwq\t{%1, %0|%0, %1}"
|
9888 |
|
|
[(set_attr "type" "ssemov")
|
9889 |
|
|
(set_attr "prefix_extra" "1")
|
9890 |
|
|
(set_attr "prefix" "maybe_vex")
|
9891 |
|
|
(set_attr "mode" "TI")])
|
9892 |
|
|
|
9893 |
|
|
(define_insn "sse4_1_zero_extendv2siv2di2"
|
9894 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9895 |
|
|
(zero_extend:V2DI
|
9896 |
|
|
(vec_select:V2SI
|
9897 |
|
|
(match_operand:V4SI 1 "register_operand" "x")
|
9898 |
|
|
(parallel [(const_int 0)
|
9899 |
|
|
(const_int 1)]))))]
|
9900 |
|
|
"TARGET_SSE4_1"
|
9901 |
|
|
"%vpmovzxdq\t{%1, %0|%0, %1}"
|
9902 |
|
|
[(set_attr "type" "ssemov")
|
9903 |
|
|
(set_attr "prefix_extra" "1")
|
9904 |
|
|
(set_attr "prefix" "maybe_vex")
|
9905 |
|
|
(set_attr "mode" "TI")])
|
9906 |
|
|
|
9907 |
|
|
(define_insn "*sse4_1_zero_extendv2siv2di2"
|
9908 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
9909 |
|
|
(zero_extend:V2DI
|
9910 |
|
|
(vec_select:V2SI
|
9911 |
|
|
(vec_duplicate:V4SI
|
9912 |
|
|
(match_operand:V2SI 1 "nonimmediate_operand" "xm"))
|
9913 |
|
|
(parallel [(const_int 0)
|
9914 |
|
|
(const_int 1)]))))]
|
9915 |
|
|
"TARGET_SSE4_1"
|
9916 |
|
|
"%vpmovzxdq\t{%1, %0|%0, %1}"
|
9917 |
|
|
[(set_attr "type" "ssemov")
|
9918 |
|
|
(set_attr "prefix_extra" "1")
|
9919 |
|
|
(set_attr "prefix" "maybe_vex")
|
9920 |
|
|
(set_attr "mode" "TI")])
|
9921 |
|
|
|
9922 |
|
|
;; ptestps/ptestpd are very similar to comiss and ucomiss when
|
9923 |
|
|
;; setting FLAGS_REG. But it is not a really compare instruction.
|
9924 |
|
|
(define_insn "avx_vtestp"
|
9925 |
|
|
[(set (reg:CC FLAGS_REG)
|
9926 |
|
|
(unspec:CC [(match_operand:AVXMODEF2P 0 "register_operand" "x")
|
9927 |
|
|
(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "xm")]
|
9928 |
|
|
UNSPEC_VTESTP))]
|
9929 |
|
|
"TARGET_AVX"
|
9930 |
|
|
"vtestp\t{%1, %0|%0, %1}"
|
9931 |
|
|
[(set_attr "type" "ssecomi")
|
9932 |
|
|
(set_attr "prefix_extra" "1")
|
9933 |
|
|
(set_attr "prefix" "vex")
|
9934 |
|
|
(set_attr "mode" "")])
|
9935 |
|
|
|
9936 |
|
|
;; ptest is very similar to comiss and ucomiss when setting FLAGS_REG.
|
9937 |
|
|
;; But it is not a really compare instruction.
|
9938 |
|
|
(define_insn "avx_ptest256"
|
9939 |
|
|
[(set (reg:CC FLAGS_REG)
|
9940 |
|
|
(unspec:CC [(match_operand:V4DI 0 "register_operand" "x")
|
9941 |
|
|
(match_operand:V4DI 1 "nonimmediate_operand" "xm")]
|
9942 |
|
|
UNSPEC_PTEST))]
|
9943 |
|
|
"TARGET_AVX"
|
9944 |
|
|
"vptest\t{%1, %0|%0, %1}"
|
9945 |
|
|
[(set_attr "type" "ssecomi")
|
9946 |
|
|
(set_attr "prefix_extra" "1")
|
9947 |
|
|
(set_attr "prefix" "vex")
|
9948 |
|
|
(set_attr "mode" "OI")])
|
9949 |
|
|
|
9950 |
|
|
(define_insn "sse4_1_ptest"
|
9951 |
|
|
[(set (reg:CC FLAGS_REG)
|
9952 |
|
|
(unspec:CC [(match_operand:V2DI 0 "register_operand" "x")
|
9953 |
|
|
(match_operand:V2DI 1 "nonimmediate_operand" "xm")]
|
9954 |
|
|
UNSPEC_PTEST))]
|
9955 |
|
|
"TARGET_SSE4_1"
|
9956 |
|
|
"%vptest\t{%1, %0|%0, %1}"
|
9957 |
|
|
[(set_attr "type" "ssecomi")
|
9958 |
|
|
(set_attr "prefix_extra" "1")
|
9959 |
|
|
(set_attr "prefix" "maybe_vex")
|
9960 |
|
|
(set_attr "mode" "TI")])
|
9961 |
|
|
|
9962 |
|
|
(define_insn "avx_roundp256"
|
9963 |
|
|
[(set (match_operand:AVX256MODEF2P 0 "register_operand" "=x")
|
9964 |
|
|
(unspec:AVX256MODEF2P
|
9965 |
|
|
[(match_operand:AVX256MODEF2P 1 "nonimmediate_operand" "xm")
|
9966 |
|
|
(match_operand:SI 2 "const_0_to_15_operand" "n")]
|
9967 |
|
|
UNSPEC_ROUND))]
|
9968 |
|
|
"TARGET_AVX"
|
9969 |
|
|
"vroundp\t{%2, %1, %0|%0, %1, %2}"
|
9970 |
|
|
[(set_attr "type" "ssecvt")
|
9971 |
|
|
(set_attr "prefix_extra" "1")
|
9972 |
|
|
(set_attr "length_immediate" "1")
|
9973 |
|
|
(set_attr "prefix" "vex")
|
9974 |
|
|
(set_attr "mode" "")])
|
9975 |
|
|
|
9976 |
|
|
(define_insn "sse4_1_roundp"
|
9977 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
9978 |
|
|
(unspec:SSEMODEF2P
|
9979 |
|
|
[(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "xm")
|
9980 |
|
|
(match_operand:SI 2 "const_0_to_15_operand" "n")]
|
9981 |
|
|
UNSPEC_ROUND))]
|
9982 |
|
|
"TARGET_ROUND"
|
9983 |
|
|
"%vroundp\t{%2, %1, %0|%0, %1, %2}"
|
9984 |
|
|
[(set_attr "type" "ssecvt")
|
9985 |
|
|
(set_attr "prefix_data16" "1")
|
9986 |
|
|
(set_attr "prefix_extra" "1")
|
9987 |
|
|
(set_attr "length_immediate" "1")
|
9988 |
|
|
(set_attr "prefix" "maybe_vex")
|
9989 |
|
|
(set_attr "mode" "")])
|
9990 |
|
|
|
9991 |
|
|
(define_insn "*avx_rounds"
|
9992 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
9993 |
|
|
(vec_merge:SSEMODEF2P
|
9994 |
|
|
(unspec:SSEMODEF2P
|
9995 |
|
|
[(match_operand:SSEMODEF2P 2 "register_operand" "x")
|
9996 |
|
|
(match_operand:SI 3 "const_0_to_15_operand" "n")]
|
9997 |
|
|
UNSPEC_ROUND)
|
9998 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "x")
|
9999 |
|
|
(const_int 1)))]
|
10000 |
|
|
"TARGET_AVX"
|
10001 |
|
|
"vrounds\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10002 |
|
|
[(set_attr "type" "ssecvt")
|
10003 |
|
|
(set_attr "prefix_extra" "1")
|
10004 |
|
|
(set_attr "length_immediate" "1")
|
10005 |
|
|
(set_attr "prefix" "vex")
|
10006 |
|
|
(set_attr "mode" "")])
|
10007 |
|
|
|
10008 |
|
|
(define_insn "sse4_1_rounds"
|
10009 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
10010 |
|
|
(vec_merge:SSEMODEF2P
|
10011 |
|
|
(unspec:SSEMODEF2P
|
10012 |
|
|
[(match_operand:SSEMODEF2P 2 "register_operand" "x")
|
10013 |
|
|
(match_operand:SI 3 "const_0_to_15_operand" "n")]
|
10014 |
|
|
UNSPEC_ROUND)
|
10015 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
10016 |
|
|
(const_int 1)))]
|
10017 |
|
|
"TARGET_ROUND"
|
10018 |
|
|
"rounds\t{%3, %2, %0|%0, %2, %3}"
|
10019 |
|
|
[(set_attr "type" "ssecvt")
|
10020 |
|
|
(set_attr "prefix_data16" "1")
|
10021 |
|
|
(set_attr "prefix_extra" "1")
|
10022 |
|
|
(set_attr "length_immediate" "1")
|
10023 |
|
|
(set_attr "mode" "")])
|
10024 |
|
|
|
10025 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
10026 |
|
|
;;
|
10027 |
|
|
;; Intel SSE4.2 string/text processing instructions
|
10028 |
|
|
;;
|
10029 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
10030 |
|
|
|
10031 |
|
|
(define_insn_and_split "sse4_2_pcmpestr"
|
10032 |
|
|
[(set (match_operand:SI 0 "register_operand" "=c,c")
|
10033 |
|
|
(unspec:SI
|
10034 |
|
|
[(match_operand:V16QI 2 "reg_not_xmm0_operand" "x,x")
|
10035 |
|
|
(match_operand:SI 3 "register_operand" "a,a")
|
10036 |
|
|
(match_operand:V16QI 4 "nonimm_not_xmm0_operand" "x,m")
|
10037 |
|
|
(match_operand:SI 5 "register_operand" "d,d")
|
10038 |
|
|
(match_operand:SI 6 "const_0_to_255_operand" "n,n")]
|
10039 |
|
|
UNSPEC_PCMPESTR))
|
10040 |
|
|
(set (match_operand:V16QI 1 "register_operand" "=Yz,Yz")
|
10041 |
|
|
(unspec:V16QI
|
10042 |
|
|
[(match_dup 2)
|
10043 |
|
|
(match_dup 3)
|
10044 |
|
|
(match_dup 4)
|
10045 |
|
|
(match_dup 5)
|
10046 |
|
|
(match_dup 6)]
|
10047 |
|
|
UNSPEC_PCMPESTR))
|
10048 |
|
|
(set (reg:CC FLAGS_REG)
|
10049 |
|
|
(unspec:CC
|
10050 |
|
|
[(match_dup 2)
|
10051 |
|
|
(match_dup 3)
|
10052 |
|
|
(match_dup 4)
|
10053 |
|
|
(match_dup 5)
|
10054 |
|
|
(match_dup 6)]
|
10055 |
|
|
UNSPEC_PCMPESTR))]
|
10056 |
|
|
"TARGET_SSE4_2
|
10057 |
|
|
&& can_create_pseudo_p ()"
|
10058 |
|
|
"#"
|
10059 |
|
|
"&& 1"
|
10060 |
|
|
[(const_int 0)]
|
10061 |
|
|
{
|
10062 |
|
|
int ecx = !find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[0]));
|
10063 |
|
|
int xmm0 = !find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[1]));
|
10064 |
|
|
int flags = !find_regno_note (curr_insn, REG_UNUSED, FLAGS_REG);
|
10065 |
|
|
|
10066 |
|
|
if (ecx)
|
10067 |
|
|
emit_insn (gen_sse4_2_pcmpestri (operands[0], operands[2],
|
10068 |
|
|
operands[3], operands[4],
|
10069 |
|
|
operands[5], operands[6]));
|
10070 |
|
|
if (xmm0)
|
10071 |
|
|
emit_insn (gen_sse4_2_pcmpestrm (operands[1], operands[2],
|
10072 |
|
|
operands[3], operands[4],
|
10073 |
|
|
operands[5], operands[6]));
|
10074 |
|
|
if (flags && !(ecx || xmm0))
|
10075 |
|
|
emit_insn (gen_sse4_2_pcmpestr_cconly (NULL, NULL,
|
10076 |
|
|
operands[2], operands[3],
|
10077 |
|
|
operands[4], operands[5],
|
10078 |
|
|
operands[6]));
|
10079 |
|
|
DONE;
|
10080 |
|
|
}
|
10081 |
|
|
[(set_attr "type" "sselog")
|
10082 |
|
|
(set_attr "prefix_data16" "1")
|
10083 |
|
|
(set_attr "prefix_extra" "1")
|
10084 |
|
|
(set_attr "length_immediate" "1")
|
10085 |
|
|
(set_attr "memory" "none,load")
|
10086 |
|
|
(set_attr "mode" "TI")])
|
10087 |
|
|
|
10088 |
|
|
(define_insn "sse4_2_pcmpestri"
|
10089 |
|
|
[(set (match_operand:SI 0 "register_operand" "=c,c")
|
10090 |
|
|
(unspec:SI
|
10091 |
|
|
[(match_operand:V16QI 1 "register_operand" "x,x")
|
10092 |
|
|
(match_operand:SI 2 "register_operand" "a,a")
|
10093 |
|
|
(match_operand:V16QI 3 "nonimmediate_operand" "x,m")
|
10094 |
|
|
(match_operand:SI 4 "register_operand" "d,d")
|
10095 |
|
|
(match_operand:SI 5 "const_0_to_255_operand" "n,n")]
|
10096 |
|
|
UNSPEC_PCMPESTR))
|
10097 |
|
|
(set (reg:CC FLAGS_REG)
|
10098 |
|
|
(unspec:CC
|
10099 |
|
|
[(match_dup 1)
|
10100 |
|
|
(match_dup 2)
|
10101 |
|
|
(match_dup 3)
|
10102 |
|
|
(match_dup 4)
|
10103 |
|
|
(match_dup 5)]
|
10104 |
|
|
UNSPEC_PCMPESTR))]
|
10105 |
|
|
"TARGET_SSE4_2"
|
10106 |
|
|
"%vpcmpestri\t{%5, %3, %1|%1, %3, %5}"
|
10107 |
|
|
[(set_attr "type" "sselog")
|
10108 |
|
|
(set_attr "prefix_data16" "1")
|
10109 |
|
|
(set_attr "prefix_extra" "1")
|
10110 |
|
|
(set_attr "prefix" "maybe_vex")
|
10111 |
|
|
(set_attr "length_immediate" "1")
|
10112 |
|
|
(set_attr "memory" "none,load")
|
10113 |
|
|
(set_attr "mode" "TI")])
|
10114 |
|
|
|
10115 |
|
|
(define_insn "sse4_2_pcmpestrm"
|
10116 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=Yz,Yz")
|
10117 |
|
|
(unspec:V16QI
|
10118 |
|
|
[(match_operand:V16QI 1 "register_operand" "x,x")
|
10119 |
|
|
(match_operand:SI 2 "register_operand" "a,a")
|
10120 |
|
|
(match_operand:V16QI 3 "nonimmediate_operand" "x,m")
|
10121 |
|
|
(match_operand:SI 4 "register_operand" "d,d")
|
10122 |
|
|
(match_operand:SI 5 "const_0_to_255_operand" "n,n")]
|
10123 |
|
|
UNSPEC_PCMPESTR))
|
10124 |
|
|
(set (reg:CC FLAGS_REG)
|
10125 |
|
|
(unspec:CC
|
10126 |
|
|
[(match_dup 1)
|
10127 |
|
|
(match_dup 2)
|
10128 |
|
|
(match_dup 3)
|
10129 |
|
|
(match_dup 4)
|
10130 |
|
|
(match_dup 5)]
|
10131 |
|
|
UNSPEC_PCMPESTR))]
|
10132 |
|
|
"TARGET_SSE4_2"
|
10133 |
|
|
"%vpcmpestrm\t{%5, %3, %1|%1, %3, %5}"
|
10134 |
|
|
[(set_attr "type" "sselog")
|
10135 |
|
|
(set_attr "prefix_data16" "1")
|
10136 |
|
|
(set_attr "prefix_extra" "1")
|
10137 |
|
|
(set_attr "length_immediate" "1")
|
10138 |
|
|
(set_attr "prefix" "maybe_vex")
|
10139 |
|
|
(set_attr "memory" "none,load")
|
10140 |
|
|
(set_attr "mode" "TI")])
|
10141 |
|
|
|
10142 |
|
|
(define_insn "sse4_2_pcmpestr_cconly"
|
10143 |
|
|
[(set (reg:CC FLAGS_REG)
|
10144 |
|
|
(unspec:CC
|
10145 |
|
|
[(match_operand:V16QI 2 "register_operand" "x,x,x,x")
|
10146 |
|
|
(match_operand:SI 3 "register_operand" "a,a,a,a")
|
10147 |
|
|
(match_operand:V16QI 4 "nonimmediate_operand" "x,m,x,m")
|
10148 |
|
|
(match_operand:SI 5 "register_operand" "d,d,d,d")
|
10149 |
|
|
(match_operand:SI 6 "const_0_to_255_operand" "n,n,n,n")]
|
10150 |
|
|
UNSPEC_PCMPESTR))
|
10151 |
|
|
(clobber (match_scratch:V16QI 0 "=Yz,Yz,X,X"))
|
10152 |
|
|
(clobber (match_scratch:SI 1 "= X, X,c,c"))]
|
10153 |
|
|
"TARGET_SSE4_2"
|
10154 |
|
|
"@
|
10155 |
|
|
%vpcmpestrm\t{%6, %4, %2|%2, %4, %6}
|
10156 |
|
|
%vpcmpestrm\t{%6, %4, %2|%2, %4, %6}
|
10157 |
|
|
%vpcmpestri\t{%6, %4, %2|%2, %4, %6}
|
10158 |
|
|
%vpcmpestri\t{%6, %4, %2|%2, %4, %6}"
|
10159 |
|
|
[(set_attr "type" "sselog")
|
10160 |
|
|
(set_attr "prefix_data16" "1")
|
10161 |
|
|
(set_attr "prefix_extra" "1")
|
10162 |
|
|
(set_attr "length_immediate" "1")
|
10163 |
|
|
(set_attr "memory" "none,load,none,load")
|
10164 |
|
|
(set_attr "prefix" "maybe_vex")
|
10165 |
|
|
(set_attr "mode" "TI")])
|
10166 |
|
|
|
10167 |
|
|
(define_insn_and_split "sse4_2_pcmpistr"
|
10168 |
|
|
[(set (match_operand:SI 0 "register_operand" "=c,c")
|
10169 |
|
|
(unspec:SI
|
10170 |
|
|
[(match_operand:V16QI 2 "reg_not_xmm0_operand" "x,x")
|
10171 |
|
|
(match_operand:V16QI 3 "nonimm_not_xmm0_operand" "x,m")
|
10172 |
|
|
(match_operand:SI 4 "const_0_to_255_operand" "n,n")]
|
10173 |
|
|
UNSPEC_PCMPISTR))
|
10174 |
|
|
(set (match_operand:V16QI 1 "register_operand" "=Yz,Yz")
|
10175 |
|
|
(unspec:V16QI
|
10176 |
|
|
[(match_dup 2)
|
10177 |
|
|
(match_dup 3)
|
10178 |
|
|
(match_dup 4)]
|
10179 |
|
|
UNSPEC_PCMPISTR))
|
10180 |
|
|
(set (reg:CC FLAGS_REG)
|
10181 |
|
|
(unspec:CC
|
10182 |
|
|
[(match_dup 2)
|
10183 |
|
|
(match_dup 3)
|
10184 |
|
|
(match_dup 4)]
|
10185 |
|
|
UNSPEC_PCMPISTR))]
|
10186 |
|
|
"TARGET_SSE4_2
|
10187 |
|
|
&& can_create_pseudo_p ()"
|
10188 |
|
|
"#"
|
10189 |
|
|
"&& 1"
|
10190 |
|
|
[(const_int 0)]
|
10191 |
|
|
{
|
10192 |
|
|
int ecx = !find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[0]));
|
10193 |
|
|
int xmm0 = !find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[1]));
|
10194 |
|
|
int flags = !find_regno_note (curr_insn, REG_UNUSED, FLAGS_REG);
|
10195 |
|
|
|
10196 |
|
|
if (ecx)
|
10197 |
|
|
emit_insn (gen_sse4_2_pcmpistri (operands[0], operands[2],
|
10198 |
|
|
operands[3], operands[4]));
|
10199 |
|
|
if (xmm0)
|
10200 |
|
|
emit_insn (gen_sse4_2_pcmpistrm (operands[1], operands[2],
|
10201 |
|
|
operands[3], operands[4]));
|
10202 |
|
|
if (flags && !(ecx || xmm0))
|
10203 |
|
|
emit_insn (gen_sse4_2_pcmpistr_cconly (NULL, NULL,
|
10204 |
|
|
operands[2], operands[3],
|
10205 |
|
|
operands[4]));
|
10206 |
|
|
DONE;
|
10207 |
|
|
}
|
10208 |
|
|
[(set_attr "type" "sselog")
|
10209 |
|
|
(set_attr "prefix_data16" "1")
|
10210 |
|
|
(set_attr "prefix_extra" "1")
|
10211 |
|
|
(set_attr "length_immediate" "1")
|
10212 |
|
|
(set_attr "memory" "none,load")
|
10213 |
|
|
(set_attr "mode" "TI")])
|
10214 |
|
|
|
10215 |
|
|
(define_insn "sse4_2_pcmpistri"
|
10216 |
|
|
[(set (match_operand:SI 0 "register_operand" "=c,c")
|
10217 |
|
|
(unspec:SI
|
10218 |
|
|
[(match_operand:V16QI 1 "register_operand" "x,x")
|
10219 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "x,m")
|
10220 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n,n")]
|
10221 |
|
|
UNSPEC_PCMPISTR))
|
10222 |
|
|
(set (reg:CC FLAGS_REG)
|
10223 |
|
|
(unspec:CC
|
10224 |
|
|
[(match_dup 1)
|
10225 |
|
|
(match_dup 2)
|
10226 |
|
|
(match_dup 3)]
|
10227 |
|
|
UNSPEC_PCMPISTR))]
|
10228 |
|
|
"TARGET_SSE4_2"
|
10229 |
|
|
"%vpcmpistri\t{%3, %2, %1|%1, %2, %3}"
|
10230 |
|
|
[(set_attr "type" "sselog")
|
10231 |
|
|
(set_attr "prefix_data16" "1")
|
10232 |
|
|
(set_attr "prefix_extra" "1")
|
10233 |
|
|
(set_attr "length_immediate" "1")
|
10234 |
|
|
(set_attr "prefix" "maybe_vex")
|
10235 |
|
|
(set_attr "memory" "none,load")
|
10236 |
|
|
(set_attr "mode" "TI")])
|
10237 |
|
|
|
10238 |
|
|
(define_insn "sse4_2_pcmpistrm"
|
10239 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=Yz,Yz")
|
10240 |
|
|
(unspec:V16QI
|
10241 |
|
|
[(match_operand:V16QI 1 "register_operand" "x,x")
|
10242 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "x,m")
|
10243 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n,n")]
|
10244 |
|
|
UNSPEC_PCMPISTR))
|
10245 |
|
|
(set (reg:CC FLAGS_REG)
|
10246 |
|
|
(unspec:CC
|
10247 |
|
|
[(match_dup 1)
|
10248 |
|
|
(match_dup 2)
|
10249 |
|
|
(match_dup 3)]
|
10250 |
|
|
UNSPEC_PCMPISTR))]
|
10251 |
|
|
"TARGET_SSE4_2"
|
10252 |
|
|
"%vpcmpistrm\t{%3, %2, %1|%1, %2, %3}"
|
10253 |
|
|
[(set_attr "type" "sselog")
|
10254 |
|
|
(set_attr "prefix_data16" "1")
|
10255 |
|
|
(set_attr "prefix_extra" "1")
|
10256 |
|
|
(set_attr "length_immediate" "1")
|
10257 |
|
|
(set_attr "prefix" "maybe_vex")
|
10258 |
|
|
(set_attr "memory" "none,load")
|
10259 |
|
|
(set_attr "mode" "TI")])
|
10260 |
|
|
|
10261 |
|
|
(define_insn "sse4_2_pcmpistr_cconly"
|
10262 |
|
|
[(set (reg:CC FLAGS_REG)
|
10263 |
|
|
(unspec:CC
|
10264 |
|
|
[(match_operand:V16QI 2 "register_operand" "x,x,x,x")
|
10265 |
|
|
(match_operand:V16QI 3 "nonimmediate_operand" "x,m,x,m")
|
10266 |
|
|
(match_operand:SI 4 "const_0_to_255_operand" "n,n,n,n")]
|
10267 |
|
|
UNSPEC_PCMPISTR))
|
10268 |
|
|
(clobber (match_scratch:V16QI 0 "=Yz,Yz,X,X"))
|
10269 |
|
|
(clobber (match_scratch:SI 1 "= X, X,c,c"))]
|
10270 |
|
|
"TARGET_SSE4_2"
|
10271 |
|
|
"@
|
10272 |
|
|
%vpcmpistrm\t{%4, %3, %2|%2, %3, %4}
|
10273 |
|
|
%vpcmpistrm\t{%4, %3, %2|%2, %3, %4}
|
10274 |
|
|
%vpcmpistri\t{%4, %3, %2|%2, %3, %4}
|
10275 |
|
|
%vpcmpistri\t{%4, %3, %2|%2, %3, %4}"
|
10276 |
|
|
[(set_attr "type" "sselog")
|
10277 |
|
|
(set_attr "prefix_data16" "1")
|
10278 |
|
|
(set_attr "prefix_extra" "1")
|
10279 |
|
|
(set_attr "length_immediate" "1")
|
10280 |
|
|
(set_attr "memory" "none,load,none,load")
|
10281 |
|
|
(set_attr "prefix" "maybe_vex")
|
10282 |
|
|
(set_attr "mode" "TI")])
|
10283 |
|
|
|
10284 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
10285 |
|
|
;;
|
10286 |
|
|
;; XOP instructions
|
10287 |
|
|
;;
|
10288 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
10289 |
|
|
|
10290 |
|
|
;; XOP parallel integer multiply/add instructions.
|
10291 |
|
|
;; Note the XOP multiply/add instructions
|
10292 |
|
|
;; a[i] = b[i] * c[i] + d[i];
|
10293 |
|
|
;; do not allow the value being added to be a memory operation.
|
10294 |
|
|
(define_insn "xop_pmacsww"
|
10295 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
10296 |
|
|
(plus:V8HI
|
10297 |
|
|
(mult:V8HI
|
10298 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x")
|
10299 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))
|
10300 |
|
|
(match_operand:V8HI 3 "nonimmediate_operand" "x")))]
|
10301 |
|
|
"TARGET_XOP"
|
10302 |
|
|
"vpmacsww\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10303 |
|
|
[(set_attr "type" "ssemuladd")
|
10304 |
|
|
(set_attr "mode" "TI")])
|
10305 |
|
|
|
10306 |
|
|
(define_insn "xop_pmacssww"
|
10307 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
10308 |
|
|
(ss_plus:V8HI
|
10309 |
|
|
(mult:V8HI (match_operand:V8HI 1 "nonimmediate_operand" "%x")
|
10310 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm"))
|
10311 |
|
|
(match_operand:V8HI 3 "nonimmediate_operand" "x")))]
|
10312 |
|
|
"TARGET_XOP"
|
10313 |
|
|
"vpmacssww\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10314 |
|
|
[(set_attr "type" "ssemuladd")
|
10315 |
|
|
(set_attr "mode" "TI")])
|
10316 |
|
|
|
10317 |
|
|
(define_insn "xop_pmacsdd"
|
10318 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10319 |
|
|
(plus:V4SI
|
10320 |
|
|
(mult:V4SI
|
10321 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
10322 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))
|
10323 |
|
|
(match_operand:V4SI 3 "nonimmediate_operand" "x")))]
|
10324 |
|
|
"TARGET_XOP"
|
10325 |
|
|
"vpmacsdd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10326 |
|
|
[(set_attr "type" "ssemuladd")
|
10327 |
|
|
(set_attr "mode" "TI")])
|
10328 |
|
|
|
10329 |
|
|
(define_insn "xop_pmacssdd"
|
10330 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10331 |
|
|
(ss_plus:V4SI
|
10332 |
|
|
(mult:V4SI (match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
10333 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm"))
|
10334 |
|
|
(match_operand:V4SI 3 "nonimmediate_operand" "x")))]
|
10335 |
|
|
"TARGET_XOP"
|
10336 |
|
|
"vpmacssdd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10337 |
|
|
[(set_attr "type" "ssemuladd")
|
10338 |
|
|
(set_attr "mode" "TI")])
|
10339 |
|
|
|
10340 |
|
|
(define_insn "xop_pmacssdql"
|
10341 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10342 |
|
|
(ss_plus:V2DI
|
10343 |
|
|
(mult:V2DI
|
10344 |
|
|
(sign_extend:V2DI
|
10345 |
|
|
(vec_select:V2SI
|
10346 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
10347 |
|
|
(parallel [(const_int 1)
|
10348 |
|
|
(const_int 3)])))
|
10349 |
|
|
(vec_select:V2SI
|
10350 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
10351 |
|
|
(parallel [(const_int 1)
|
10352 |
|
|
(const_int 3)])))
|
10353 |
|
|
(match_operand:V2DI 3 "nonimmediate_operand" "x")))]
|
10354 |
|
|
"TARGET_XOP"
|
10355 |
|
|
"vpmacssdql\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10356 |
|
|
[(set_attr "type" "ssemuladd")
|
10357 |
|
|
(set_attr "mode" "TI")])
|
10358 |
|
|
|
10359 |
|
|
(define_insn "xop_pmacssdqh"
|
10360 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10361 |
|
|
(ss_plus:V2DI
|
10362 |
|
|
(mult:V2DI
|
10363 |
|
|
(sign_extend:V2DI
|
10364 |
|
|
(vec_select:V2SI
|
10365 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
10366 |
|
|
(parallel [(const_int 0)
|
10367 |
|
|
(const_int 2)])))
|
10368 |
|
|
(sign_extend:V2DI
|
10369 |
|
|
(vec_select:V2SI
|
10370 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
10371 |
|
|
(parallel [(const_int 0)
|
10372 |
|
|
(const_int 2)]))))
|
10373 |
|
|
(match_operand:V2DI 3 "nonimmediate_operand" "x")))]
|
10374 |
|
|
"TARGET_XOP"
|
10375 |
|
|
"vpmacssdqh\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10376 |
|
|
[(set_attr "type" "ssemuladd")
|
10377 |
|
|
(set_attr "mode" "TI")])
|
10378 |
|
|
|
10379 |
|
|
(define_insn "xop_pmacsdql"
|
10380 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10381 |
|
|
(plus:V2DI
|
10382 |
|
|
(mult:V2DI
|
10383 |
|
|
(sign_extend:V2DI
|
10384 |
|
|
(vec_select:V2SI
|
10385 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
10386 |
|
|
(parallel [(const_int 1)
|
10387 |
|
|
(const_int 3)])))
|
10388 |
|
|
(sign_extend:V2DI
|
10389 |
|
|
(vec_select:V2SI
|
10390 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
10391 |
|
|
(parallel [(const_int 1)
|
10392 |
|
|
(const_int 3)]))))
|
10393 |
|
|
(match_operand:V2DI 3 "nonimmediate_operand" "x")))]
|
10394 |
|
|
"TARGET_XOP"
|
10395 |
|
|
"vpmacsdql\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10396 |
|
|
[(set_attr "type" "ssemuladd")
|
10397 |
|
|
(set_attr "mode" "TI")])
|
10398 |
|
|
|
10399 |
|
|
;; We don't have a straight 32-bit parallel multiply and extend on XOP, so
|
10400 |
|
|
;; fake it with a multiply/add. In general, we expect the define_split to
|
10401 |
|
|
;; occur before register allocation, so we have to handle the corner case where
|
10402 |
|
|
;; the target is the same as operands 1/2
|
10403 |
|
|
(define_insn_and_split "xop_mulv2div2di3_low"
|
10404 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=&x")
|
10405 |
|
|
(mult:V2DI
|
10406 |
|
|
(sign_extend:V2DI
|
10407 |
|
|
(vec_select:V2SI
|
10408 |
|
|
(match_operand:V4SI 1 "register_operand" "%x")
|
10409 |
|
|
(parallel [(const_int 1)
|
10410 |
|
|
(const_int 3)])))
|
10411 |
|
|
(sign_extend:V2DI
|
10412 |
|
|
(vec_select:V2SI
|
10413 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
10414 |
|
|
(parallel [(const_int 1)
|
10415 |
|
|
(const_int 3)])))))]
|
10416 |
|
|
"TARGET_XOP"
|
10417 |
|
|
"#"
|
10418 |
|
|
"&& reload_completed"
|
10419 |
|
|
[(set (match_dup 0)
|
10420 |
|
|
(match_dup 3))
|
10421 |
|
|
(set (match_dup 0)
|
10422 |
|
|
(plus:V2DI
|
10423 |
|
|
(mult:V2DI
|
10424 |
|
|
(sign_extend:V2DI
|
10425 |
|
|
(vec_select:V2SI
|
10426 |
|
|
(match_dup 1)
|
10427 |
|
|
(parallel [(const_int 1)
|
10428 |
|
|
(const_int 3)])))
|
10429 |
|
|
(sign_extend:V2DI
|
10430 |
|
|
(vec_select:V2SI
|
10431 |
|
|
(match_dup 2)
|
10432 |
|
|
(parallel [(const_int 1)
|
10433 |
|
|
(const_int 3)]))))
|
10434 |
|
|
(match_dup 0)))]
|
10435 |
|
|
{
|
10436 |
|
|
operands[3] = CONST0_RTX (V2DImode);
|
10437 |
|
|
}
|
10438 |
|
|
[(set_attr "type" "ssemul")
|
10439 |
|
|
(set_attr "mode" "TI")])
|
10440 |
|
|
|
10441 |
|
|
(define_insn "xop_pmacsdqh"
|
10442 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10443 |
|
|
(plus:V2DI
|
10444 |
|
|
(mult:V2DI
|
10445 |
|
|
(sign_extend:V2DI
|
10446 |
|
|
(vec_select:V2SI
|
10447 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "%x")
|
10448 |
|
|
(parallel [(const_int 0)
|
10449 |
|
|
(const_int 2)])))
|
10450 |
|
|
(sign_extend:V2DI
|
10451 |
|
|
(vec_select:V2SI
|
10452 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
10453 |
|
|
(parallel [(const_int 0)
|
10454 |
|
|
(const_int 2)]))))
|
10455 |
|
|
(match_operand:V2DI 3 "nonimmediate_operand" "x")))]
|
10456 |
|
|
"TARGET_XOP"
|
10457 |
|
|
"vpmacsdqh\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10458 |
|
|
[(set_attr "type" "ssemuladd")
|
10459 |
|
|
(set_attr "mode" "TI")])
|
10460 |
|
|
|
10461 |
|
|
;; We don't have a straight 32-bit parallel multiply and extend on XOP, so
|
10462 |
|
|
;; fake it with a multiply/add. In general, we expect the define_split to
|
10463 |
|
|
;; occur before register allocation, so we have to handle the corner case where
|
10464 |
|
|
;; the target is the same as either operands[1] or operands[2]
|
10465 |
|
|
(define_insn_and_split "xop_mulv2div2di3_high"
|
10466 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=&x")
|
10467 |
|
|
(mult:V2DI
|
10468 |
|
|
(sign_extend:V2DI
|
10469 |
|
|
(vec_select:V2SI
|
10470 |
|
|
(match_operand:V4SI 1 "register_operand" "%x")
|
10471 |
|
|
(parallel [(const_int 0)
|
10472 |
|
|
(const_int 2)])))
|
10473 |
|
|
(sign_extend:V2DI
|
10474 |
|
|
(vec_select:V2SI
|
10475 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "xm")
|
10476 |
|
|
(parallel [(const_int 0)
|
10477 |
|
|
(const_int 2)])))))]
|
10478 |
|
|
"TARGET_XOP"
|
10479 |
|
|
"#"
|
10480 |
|
|
"&& reload_completed"
|
10481 |
|
|
[(set (match_dup 0)
|
10482 |
|
|
(match_dup 3))
|
10483 |
|
|
(set (match_dup 0)
|
10484 |
|
|
(plus:V2DI
|
10485 |
|
|
(mult:V2DI
|
10486 |
|
|
(sign_extend:V2DI
|
10487 |
|
|
(vec_select:V2SI
|
10488 |
|
|
(match_dup 1)
|
10489 |
|
|
(parallel [(const_int 0)
|
10490 |
|
|
(const_int 2)])))
|
10491 |
|
|
(sign_extend:V2DI
|
10492 |
|
|
(vec_select:V2SI
|
10493 |
|
|
(match_dup 2)
|
10494 |
|
|
(parallel [(const_int 0)
|
10495 |
|
|
(const_int 2)]))))
|
10496 |
|
|
(match_dup 0)))]
|
10497 |
|
|
{
|
10498 |
|
|
operands[3] = CONST0_RTX (V2DImode);
|
10499 |
|
|
}
|
10500 |
|
|
[(set_attr "type" "ssemul")
|
10501 |
|
|
(set_attr "mode" "TI")])
|
10502 |
|
|
|
10503 |
|
|
;; XOP parallel integer multiply/add instructions for the intrinisics
|
10504 |
|
|
(define_insn "xop_pmacsswd"
|
10505 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10506 |
|
|
(ss_plus:V4SI
|
10507 |
|
|
(mult:V4SI
|
10508 |
|
|
(sign_extend:V4SI
|
10509 |
|
|
(vec_select:V4HI
|
10510 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x")
|
10511 |
|
|
(parallel [(const_int 1)
|
10512 |
|
|
(const_int 3)
|
10513 |
|
|
(const_int 5)
|
10514 |
|
|
(const_int 7)])))
|
10515 |
|
|
(sign_extend:V4SI
|
10516 |
|
|
(vec_select:V4HI
|
10517 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
10518 |
|
|
(parallel [(const_int 1)
|
10519 |
|
|
(const_int 3)
|
10520 |
|
|
(const_int 5)
|
10521 |
|
|
(const_int 7)]))))
|
10522 |
|
|
(match_operand:V4SI 3 "nonimmediate_operand" "x")))]
|
10523 |
|
|
"TARGET_XOP"
|
10524 |
|
|
"vpmacsswd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10525 |
|
|
[(set_attr "type" "ssemuladd")
|
10526 |
|
|
(set_attr "mode" "TI")])
|
10527 |
|
|
|
10528 |
|
|
(define_insn "xop_pmacswd"
|
10529 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10530 |
|
|
(plus:V4SI
|
10531 |
|
|
(mult:V4SI
|
10532 |
|
|
(sign_extend:V4SI
|
10533 |
|
|
(vec_select:V4HI
|
10534 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x")
|
10535 |
|
|
(parallel [(const_int 1)
|
10536 |
|
|
(const_int 3)
|
10537 |
|
|
(const_int 5)
|
10538 |
|
|
(const_int 7)])))
|
10539 |
|
|
(sign_extend:V4SI
|
10540 |
|
|
(vec_select:V4HI
|
10541 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
10542 |
|
|
(parallel [(const_int 1)
|
10543 |
|
|
(const_int 3)
|
10544 |
|
|
(const_int 5)
|
10545 |
|
|
(const_int 7)]))))
|
10546 |
|
|
(match_operand:V4SI 3 "nonimmediate_operand" "x")))]
|
10547 |
|
|
"TARGET_XOP"
|
10548 |
|
|
"vpmacswd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10549 |
|
|
[(set_attr "type" "ssemuladd")
|
10550 |
|
|
(set_attr "mode" "TI")])
|
10551 |
|
|
|
10552 |
|
|
(define_insn "xop_pmadcsswd"
|
10553 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10554 |
|
|
(ss_plus:V4SI
|
10555 |
|
|
(plus:V4SI
|
10556 |
|
|
(mult:V4SI
|
10557 |
|
|
(sign_extend:V4SI
|
10558 |
|
|
(vec_select:V4HI
|
10559 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x")
|
10560 |
|
|
(parallel [(const_int 0)
|
10561 |
|
|
(const_int 2)
|
10562 |
|
|
(const_int 4)
|
10563 |
|
|
(const_int 6)])))
|
10564 |
|
|
(sign_extend:V4SI
|
10565 |
|
|
(vec_select:V4HI
|
10566 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
10567 |
|
|
(parallel [(const_int 0)
|
10568 |
|
|
(const_int 2)
|
10569 |
|
|
(const_int 4)
|
10570 |
|
|
(const_int 6)]))))
|
10571 |
|
|
(mult:V4SI
|
10572 |
|
|
(sign_extend:V4SI
|
10573 |
|
|
(vec_select:V4HI
|
10574 |
|
|
(match_dup 1)
|
10575 |
|
|
(parallel [(const_int 1)
|
10576 |
|
|
(const_int 3)
|
10577 |
|
|
(const_int 5)
|
10578 |
|
|
(const_int 7)])))
|
10579 |
|
|
(sign_extend:V4SI
|
10580 |
|
|
(vec_select:V4HI
|
10581 |
|
|
(match_dup 2)
|
10582 |
|
|
(parallel [(const_int 1)
|
10583 |
|
|
(const_int 3)
|
10584 |
|
|
(const_int 5)
|
10585 |
|
|
(const_int 7)])))))
|
10586 |
|
|
(match_operand:V4SI 3 "nonimmediate_operand" "x")))]
|
10587 |
|
|
"TARGET_XOP"
|
10588 |
|
|
"vpmadcsswd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10589 |
|
|
[(set_attr "type" "ssemuladd")
|
10590 |
|
|
(set_attr "mode" "TI")])
|
10591 |
|
|
|
10592 |
|
|
(define_insn "xop_pmadcswd"
|
10593 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10594 |
|
|
(plus:V4SI
|
10595 |
|
|
(plus:V4SI
|
10596 |
|
|
(mult:V4SI
|
10597 |
|
|
(sign_extend:V4SI
|
10598 |
|
|
(vec_select:V4HI
|
10599 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "%x")
|
10600 |
|
|
(parallel [(const_int 0)
|
10601 |
|
|
(const_int 2)
|
10602 |
|
|
(const_int 4)
|
10603 |
|
|
(const_int 6)])))
|
10604 |
|
|
(sign_extend:V4SI
|
10605 |
|
|
(vec_select:V4HI
|
10606 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
10607 |
|
|
(parallel [(const_int 0)
|
10608 |
|
|
(const_int 2)
|
10609 |
|
|
(const_int 4)
|
10610 |
|
|
(const_int 6)]))))
|
10611 |
|
|
(mult:V4SI
|
10612 |
|
|
(sign_extend:V4SI
|
10613 |
|
|
(vec_select:V4HI
|
10614 |
|
|
(match_dup 1)
|
10615 |
|
|
(parallel [(const_int 1)
|
10616 |
|
|
(const_int 3)
|
10617 |
|
|
(const_int 5)
|
10618 |
|
|
(const_int 7)])))
|
10619 |
|
|
(sign_extend:V4SI
|
10620 |
|
|
(vec_select:V4HI
|
10621 |
|
|
(match_dup 2)
|
10622 |
|
|
(parallel [(const_int 1)
|
10623 |
|
|
(const_int 3)
|
10624 |
|
|
(const_int 5)
|
10625 |
|
|
(const_int 7)])))))
|
10626 |
|
|
(match_operand:V4SI 3 "nonimmediate_operand" "x")))]
|
10627 |
|
|
"TARGET_XOP"
|
10628 |
|
|
"vpmadcswd\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10629 |
|
|
[(set_attr "type" "ssemuladd")
|
10630 |
|
|
(set_attr "mode" "TI")])
|
10631 |
|
|
|
10632 |
|
|
;; XOP parallel XMM conditional moves
|
10633 |
|
|
(define_insn "xop_pcmov_"
|
10634 |
|
|
[(set (match_operand:SSEMODE 0 "register_operand" "=x,x")
|
10635 |
|
|
(if_then_else:SSEMODE
|
10636 |
|
|
(match_operand:SSEMODE 3 "nonimmediate_operand" "x,m")
|
10637 |
|
|
(match_operand:SSEMODE 1 "vector_move_operand" "x,x")
|
10638 |
|
|
(match_operand:SSEMODE 2 "vector_move_operand" "xm,x")))]
|
10639 |
|
|
"TARGET_XOP"
|
10640 |
|
|
"vpcmov\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10641 |
|
|
[(set_attr "type" "sse4arg")])
|
10642 |
|
|
|
10643 |
|
|
(define_insn "xop_pcmov_256"
|
10644 |
|
|
[(set (match_operand:AVX256MODE 0 "register_operand" "=x,x")
|
10645 |
|
|
(if_then_else:AVX256MODE
|
10646 |
|
|
(match_operand:AVX256MODE 3 "nonimmediate_operand" "x,m")
|
10647 |
|
|
(match_operand:AVX256MODE 1 "vector_move_operand" "x,x")
|
10648 |
|
|
(match_operand:AVX256MODE 2 "vector_move_operand" "xm,x")))]
|
10649 |
|
|
"TARGET_XOP"
|
10650 |
|
|
"vpcmov\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
10651 |
|
|
[(set_attr "type" "sse4arg")])
|
10652 |
|
|
|
10653 |
|
|
;; XOP horizontal add/subtract instructions
|
10654 |
|
|
(define_insn "xop_phaddbw"
|
10655 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
10656 |
|
|
(plus:V8HI
|
10657 |
|
|
(sign_extend:V8HI
|
10658 |
|
|
(vec_select:V8QI
|
10659 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "xm")
|
10660 |
|
|
(parallel [(const_int 0)
|
10661 |
|
|
(const_int 2)
|
10662 |
|
|
(const_int 4)
|
10663 |
|
|
(const_int 6)
|
10664 |
|
|
(const_int 8)
|
10665 |
|
|
(const_int 10)
|
10666 |
|
|
(const_int 12)
|
10667 |
|
|
(const_int 14)])))
|
10668 |
|
|
(sign_extend:V8HI
|
10669 |
|
|
(vec_select:V8QI
|
10670 |
|
|
(match_dup 1)
|
10671 |
|
|
(parallel [(const_int 1)
|
10672 |
|
|
(const_int 3)
|
10673 |
|
|
(const_int 5)
|
10674 |
|
|
(const_int 7)
|
10675 |
|
|
(const_int 9)
|
10676 |
|
|
(const_int 11)
|
10677 |
|
|
(const_int 13)
|
10678 |
|
|
(const_int 15)])))))]
|
10679 |
|
|
"TARGET_XOP"
|
10680 |
|
|
"vphaddbw\t{%1, %0|%0, %1}"
|
10681 |
|
|
[(set_attr "type" "sseiadd1")])
|
10682 |
|
|
|
10683 |
|
|
(define_insn "xop_phaddbd"
|
10684 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10685 |
|
|
(plus:V4SI
|
10686 |
|
|
(plus:V4SI
|
10687 |
|
|
(sign_extend:V4SI
|
10688 |
|
|
(vec_select:V4QI
|
10689 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "xm")
|
10690 |
|
|
(parallel [(const_int 0)
|
10691 |
|
|
(const_int 4)
|
10692 |
|
|
(const_int 8)
|
10693 |
|
|
(const_int 12)])))
|
10694 |
|
|
(sign_extend:V4SI
|
10695 |
|
|
(vec_select:V4QI
|
10696 |
|
|
(match_dup 1)
|
10697 |
|
|
(parallel [(const_int 1)
|
10698 |
|
|
(const_int 5)
|
10699 |
|
|
(const_int 9)
|
10700 |
|
|
(const_int 13)]))))
|
10701 |
|
|
(plus:V4SI
|
10702 |
|
|
(sign_extend:V4SI
|
10703 |
|
|
(vec_select:V4QI
|
10704 |
|
|
(match_dup 1)
|
10705 |
|
|
(parallel [(const_int 2)
|
10706 |
|
|
(const_int 6)
|
10707 |
|
|
(const_int 10)
|
10708 |
|
|
(const_int 14)])))
|
10709 |
|
|
(sign_extend:V4SI
|
10710 |
|
|
(vec_select:V4QI
|
10711 |
|
|
(match_dup 1)
|
10712 |
|
|
(parallel [(const_int 3)
|
10713 |
|
|
(const_int 7)
|
10714 |
|
|
(const_int 11)
|
10715 |
|
|
(const_int 15)]))))))]
|
10716 |
|
|
"TARGET_XOP"
|
10717 |
|
|
"vphaddbd\t{%1, %0|%0, %1}"
|
10718 |
|
|
[(set_attr "type" "sseiadd1")])
|
10719 |
|
|
|
10720 |
|
|
(define_insn "xop_phaddbq"
|
10721 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10722 |
|
|
(plus:V2DI
|
10723 |
|
|
(plus:V2DI
|
10724 |
|
|
(plus:V2DI
|
10725 |
|
|
(sign_extend:V2DI
|
10726 |
|
|
(vec_select:V2QI
|
10727 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "xm")
|
10728 |
|
|
(parallel [(const_int 0)
|
10729 |
|
|
(const_int 4)])))
|
10730 |
|
|
(sign_extend:V2DI
|
10731 |
|
|
(vec_select:V2QI
|
10732 |
|
|
(match_dup 1)
|
10733 |
|
|
(parallel [(const_int 1)
|
10734 |
|
|
(const_int 5)]))))
|
10735 |
|
|
(plus:V2DI
|
10736 |
|
|
(sign_extend:V2DI
|
10737 |
|
|
(vec_select:V2QI
|
10738 |
|
|
(match_dup 1)
|
10739 |
|
|
(parallel [(const_int 2)
|
10740 |
|
|
(const_int 6)])))
|
10741 |
|
|
(sign_extend:V2DI
|
10742 |
|
|
(vec_select:V2QI
|
10743 |
|
|
(match_dup 1)
|
10744 |
|
|
(parallel [(const_int 3)
|
10745 |
|
|
(const_int 7)])))))
|
10746 |
|
|
(plus:V2DI
|
10747 |
|
|
(plus:V2DI
|
10748 |
|
|
(sign_extend:V2DI
|
10749 |
|
|
(vec_select:V2QI
|
10750 |
|
|
(match_dup 1)
|
10751 |
|
|
(parallel [(const_int 8)
|
10752 |
|
|
(const_int 12)])))
|
10753 |
|
|
(sign_extend:V2DI
|
10754 |
|
|
(vec_select:V2QI
|
10755 |
|
|
(match_dup 1)
|
10756 |
|
|
(parallel [(const_int 9)
|
10757 |
|
|
(const_int 13)]))))
|
10758 |
|
|
(plus:V2DI
|
10759 |
|
|
(sign_extend:V2DI
|
10760 |
|
|
(vec_select:V2QI
|
10761 |
|
|
(match_dup 1)
|
10762 |
|
|
(parallel [(const_int 10)
|
10763 |
|
|
(const_int 14)])))
|
10764 |
|
|
(sign_extend:V2DI
|
10765 |
|
|
(vec_select:V2QI
|
10766 |
|
|
(match_dup 1)
|
10767 |
|
|
(parallel [(const_int 11)
|
10768 |
|
|
(const_int 15)])))))))]
|
10769 |
|
|
"TARGET_XOP"
|
10770 |
|
|
"vphaddbq\t{%1, %0|%0, %1}"
|
10771 |
|
|
[(set_attr "type" "sseiadd1")])
|
10772 |
|
|
|
10773 |
|
|
(define_insn "xop_phaddwd"
|
10774 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10775 |
|
|
(plus:V4SI
|
10776 |
|
|
(sign_extend:V4SI
|
10777 |
|
|
(vec_select:V4HI
|
10778 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "xm")
|
10779 |
|
|
(parallel [(const_int 0)
|
10780 |
|
|
(const_int 2)
|
10781 |
|
|
(const_int 4)
|
10782 |
|
|
(const_int 6)])))
|
10783 |
|
|
(sign_extend:V4SI
|
10784 |
|
|
(vec_select:V4HI
|
10785 |
|
|
(match_dup 1)
|
10786 |
|
|
(parallel [(const_int 1)
|
10787 |
|
|
(const_int 3)
|
10788 |
|
|
(const_int 5)
|
10789 |
|
|
(const_int 7)])))))]
|
10790 |
|
|
"TARGET_XOP"
|
10791 |
|
|
"vphaddwd\t{%1, %0|%0, %1}"
|
10792 |
|
|
[(set_attr "type" "sseiadd1")])
|
10793 |
|
|
|
10794 |
|
|
(define_insn "xop_phaddwq"
|
10795 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10796 |
|
|
(plus:V2DI
|
10797 |
|
|
(plus:V2DI
|
10798 |
|
|
(sign_extend:V2DI
|
10799 |
|
|
(vec_select:V2HI
|
10800 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "xm")
|
10801 |
|
|
(parallel [(const_int 0)
|
10802 |
|
|
(const_int 4)])))
|
10803 |
|
|
(sign_extend:V2DI
|
10804 |
|
|
(vec_select:V2HI
|
10805 |
|
|
(match_dup 1)
|
10806 |
|
|
(parallel [(const_int 1)
|
10807 |
|
|
(const_int 5)]))))
|
10808 |
|
|
(plus:V2DI
|
10809 |
|
|
(sign_extend:V2DI
|
10810 |
|
|
(vec_select:V2HI
|
10811 |
|
|
(match_dup 1)
|
10812 |
|
|
(parallel [(const_int 2)
|
10813 |
|
|
(const_int 6)])))
|
10814 |
|
|
(sign_extend:V2DI
|
10815 |
|
|
(vec_select:V2HI
|
10816 |
|
|
(match_dup 1)
|
10817 |
|
|
(parallel [(const_int 3)
|
10818 |
|
|
(const_int 7)]))))))]
|
10819 |
|
|
"TARGET_XOP"
|
10820 |
|
|
"vphaddwq\t{%1, %0|%0, %1}"
|
10821 |
|
|
[(set_attr "type" "sseiadd1")])
|
10822 |
|
|
|
10823 |
|
|
(define_insn "xop_phadddq"
|
10824 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10825 |
|
|
(plus:V2DI
|
10826 |
|
|
(sign_extend:V2DI
|
10827 |
|
|
(vec_select:V2SI
|
10828 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "xm")
|
10829 |
|
|
(parallel [(const_int 0)
|
10830 |
|
|
(const_int 2)])))
|
10831 |
|
|
(sign_extend:V2DI
|
10832 |
|
|
(vec_select:V2SI
|
10833 |
|
|
(match_dup 1)
|
10834 |
|
|
(parallel [(const_int 1)
|
10835 |
|
|
(const_int 3)])))))]
|
10836 |
|
|
"TARGET_XOP"
|
10837 |
|
|
"vphadddq\t{%1, %0|%0, %1}"
|
10838 |
|
|
[(set_attr "type" "sseiadd1")])
|
10839 |
|
|
|
10840 |
|
|
(define_insn "xop_phaddubw"
|
10841 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
10842 |
|
|
(plus:V8HI
|
10843 |
|
|
(zero_extend:V8HI
|
10844 |
|
|
(vec_select:V8QI
|
10845 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "xm")
|
10846 |
|
|
(parallel [(const_int 0)
|
10847 |
|
|
(const_int 2)
|
10848 |
|
|
(const_int 4)
|
10849 |
|
|
(const_int 6)
|
10850 |
|
|
(const_int 8)
|
10851 |
|
|
(const_int 10)
|
10852 |
|
|
(const_int 12)
|
10853 |
|
|
(const_int 14)])))
|
10854 |
|
|
(zero_extend:V8HI
|
10855 |
|
|
(vec_select:V8QI
|
10856 |
|
|
(match_dup 1)
|
10857 |
|
|
(parallel [(const_int 1)
|
10858 |
|
|
(const_int 3)
|
10859 |
|
|
(const_int 5)
|
10860 |
|
|
(const_int 7)
|
10861 |
|
|
(const_int 9)
|
10862 |
|
|
(const_int 11)
|
10863 |
|
|
(const_int 13)
|
10864 |
|
|
(const_int 15)])))))]
|
10865 |
|
|
"TARGET_XOP"
|
10866 |
|
|
"vphaddubw\t{%1, %0|%0, %1}"
|
10867 |
|
|
[(set_attr "type" "sseiadd1")])
|
10868 |
|
|
|
10869 |
|
|
(define_insn "xop_phaddubd"
|
10870 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10871 |
|
|
(plus:V4SI
|
10872 |
|
|
(plus:V4SI
|
10873 |
|
|
(zero_extend:V4SI
|
10874 |
|
|
(vec_select:V4QI
|
10875 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "xm")
|
10876 |
|
|
(parallel [(const_int 0)
|
10877 |
|
|
(const_int 4)
|
10878 |
|
|
(const_int 8)
|
10879 |
|
|
(const_int 12)])))
|
10880 |
|
|
(zero_extend:V4SI
|
10881 |
|
|
(vec_select:V4QI
|
10882 |
|
|
(match_dup 1)
|
10883 |
|
|
(parallel [(const_int 1)
|
10884 |
|
|
(const_int 5)
|
10885 |
|
|
(const_int 9)
|
10886 |
|
|
(const_int 13)]))))
|
10887 |
|
|
(plus:V4SI
|
10888 |
|
|
(zero_extend:V4SI
|
10889 |
|
|
(vec_select:V4QI
|
10890 |
|
|
(match_dup 1)
|
10891 |
|
|
(parallel [(const_int 2)
|
10892 |
|
|
(const_int 6)
|
10893 |
|
|
(const_int 10)
|
10894 |
|
|
(const_int 14)])))
|
10895 |
|
|
(zero_extend:V4SI
|
10896 |
|
|
(vec_select:V4QI
|
10897 |
|
|
(match_dup 1)
|
10898 |
|
|
(parallel [(const_int 3)
|
10899 |
|
|
(const_int 7)
|
10900 |
|
|
(const_int 11)
|
10901 |
|
|
(const_int 15)]))))))]
|
10902 |
|
|
"TARGET_XOP"
|
10903 |
|
|
"vphaddubd\t{%1, %0|%0, %1}"
|
10904 |
|
|
[(set_attr "type" "sseiadd1")])
|
10905 |
|
|
|
10906 |
|
|
(define_insn "xop_phaddubq"
|
10907 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10908 |
|
|
(plus:V2DI
|
10909 |
|
|
(plus:V2DI
|
10910 |
|
|
(plus:V2DI
|
10911 |
|
|
(zero_extend:V2DI
|
10912 |
|
|
(vec_select:V2QI
|
10913 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "xm")
|
10914 |
|
|
(parallel [(const_int 0)
|
10915 |
|
|
(const_int 4)])))
|
10916 |
|
|
(sign_extend:V2DI
|
10917 |
|
|
(vec_select:V2QI
|
10918 |
|
|
(match_dup 1)
|
10919 |
|
|
(parallel [(const_int 1)
|
10920 |
|
|
(const_int 5)]))))
|
10921 |
|
|
(plus:V2DI
|
10922 |
|
|
(zero_extend:V2DI
|
10923 |
|
|
(vec_select:V2QI
|
10924 |
|
|
(match_dup 1)
|
10925 |
|
|
(parallel [(const_int 2)
|
10926 |
|
|
(const_int 6)])))
|
10927 |
|
|
(zero_extend:V2DI
|
10928 |
|
|
(vec_select:V2QI
|
10929 |
|
|
(match_dup 1)
|
10930 |
|
|
(parallel [(const_int 3)
|
10931 |
|
|
(const_int 7)])))))
|
10932 |
|
|
(plus:V2DI
|
10933 |
|
|
(plus:V2DI
|
10934 |
|
|
(zero_extend:V2DI
|
10935 |
|
|
(vec_select:V2QI
|
10936 |
|
|
(match_dup 1)
|
10937 |
|
|
(parallel [(const_int 8)
|
10938 |
|
|
(const_int 12)])))
|
10939 |
|
|
(sign_extend:V2DI
|
10940 |
|
|
(vec_select:V2QI
|
10941 |
|
|
(match_dup 1)
|
10942 |
|
|
(parallel [(const_int 9)
|
10943 |
|
|
(const_int 13)]))))
|
10944 |
|
|
(plus:V2DI
|
10945 |
|
|
(zero_extend:V2DI
|
10946 |
|
|
(vec_select:V2QI
|
10947 |
|
|
(match_dup 1)
|
10948 |
|
|
(parallel [(const_int 10)
|
10949 |
|
|
(const_int 14)])))
|
10950 |
|
|
(zero_extend:V2DI
|
10951 |
|
|
(vec_select:V2QI
|
10952 |
|
|
(match_dup 1)
|
10953 |
|
|
(parallel [(const_int 11)
|
10954 |
|
|
(const_int 15)])))))))]
|
10955 |
|
|
"TARGET_XOP"
|
10956 |
|
|
"vphaddubq\t{%1, %0|%0, %1}"
|
10957 |
|
|
[(set_attr "type" "sseiadd1")])
|
10958 |
|
|
|
10959 |
|
|
(define_insn "xop_phadduwd"
|
10960 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
10961 |
|
|
(plus:V4SI
|
10962 |
|
|
(zero_extend:V4SI
|
10963 |
|
|
(vec_select:V4HI
|
10964 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "xm")
|
10965 |
|
|
(parallel [(const_int 0)
|
10966 |
|
|
(const_int 2)
|
10967 |
|
|
(const_int 4)
|
10968 |
|
|
(const_int 6)])))
|
10969 |
|
|
(zero_extend:V4SI
|
10970 |
|
|
(vec_select:V4HI
|
10971 |
|
|
(match_dup 1)
|
10972 |
|
|
(parallel [(const_int 1)
|
10973 |
|
|
(const_int 3)
|
10974 |
|
|
(const_int 5)
|
10975 |
|
|
(const_int 7)])))))]
|
10976 |
|
|
"TARGET_XOP"
|
10977 |
|
|
"vphadduwd\t{%1, %0|%0, %1}"
|
10978 |
|
|
[(set_attr "type" "sseiadd1")])
|
10979 |
|
|
|
10980 |
|
|
(define_insn "xop_phadduwq"
|
10981 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
10982 |
|
|
(plus:V2DI
|
10983 |
|
|
(plus:V2DI
|
10984 |
|
|
(zero_extend:V2DI
|
10985 |
|
|
(vec_select:V2HI
|
10986 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "xm")
|
10987 |
|
|
(parallel [(const_int 0)
|
10988 |
|
|
(const_int 4)])))
|
10989 |
|
|
(zero_extend:V2DI
|
10990 |
|
|
(vec_select:V2HI
|
10991 |
|
|
(match_dup 1)
|
10992 |
|
|
(parallel [(const_int 1)
|
10993 |
|
|
(const_int 5)]))))
|
10994 |
|
|
(plus:V2DI
|
10995 |
|
|
(zero_extend:V2DI
|
10996 |
|
|
(vec_select:V2HI
|
10997 |
|
|
(match_dup 1)
|
10998 |
|
|
(parallel [(const_int 2)
|
10999 |
|
|
(const_int 6)])))
|
11000 |
|
|
(zero_extend:V2DI
|
11001 |
|
|
(vec_select:V2HI
|
11002 |
|
|
(match_dup 1)
|
11003 |
|
|
(parallel [(const_int 3)
|
11004 |
|
|
(const_int 7)]))))))]
|
11005 |
|
|
"TARGET_XOP"
|
11006 |
|
|
"vphadduwq\t{%1, %0|%0, %1}"
|
11007 |
|
|
[(set_attr "type" "sseiadd1")])
|
11008 |
|
|
|
11009 |
|
|
(define_insn "xop_phaddudq"
|
11010 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11011 |
|
|
(plus:V2DI
|
11012 |
|
|
(zero_extend:V2DI
|
11013 |
|
|
(vec_select:V2SI
|
11014 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "xm")
|
11015 |
|
|
(parallel [(const_int 0)
|
11016 |
|
|
(const_int 2)])))
|
11017 |
|
|
(zero_extend:V2DI
|
11018 |
|
|
(vec_select:V2SI
|
11019 |
|
|
(match_dup 1)
|
11020 |
|
|
(parallel [(const_int 1)
|
11021 |
|
|
(const_int 3)])))))]
|
11022 |
|
|
"TARGET_XOP"
|
11023 |
|
|
"vphaddudq\t{%1, %0|%0, %1}"
|
11024 |
|
|
[(set_attr "type" "sseiadd1")])
|
11025 |
|
|
|
11026 |
|
|
(define_insn "xop_phsubbw"
|
11027 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x")
|
11028 |
|
|
(minus:V8HI
|
11029 |
|
|
(sign_extend:V8HI
|
11030 |
|
|
(vec_select:V8QI
|
11031 |
|
|
(match_operand:V16QI 1 "nonimmediate_operand" "xm")
|
11032 |
|
|
(parallel [(const_int 0)
|
11033 |
|
|
(const_int 2)
|
11034 |
|
|
(const_int 4)
|
11035 |
|
|
(const_int 6)
|
11036 |
|
|
(const_int 8)
|
11037 |
|
|
(const_int 10)
|
11038 |
|
|
(const_int 12)
|
11039 |
|
|
(const_int 14)])))
|
11040 |
|
|
(sign_extend:V8HI
|
11041 |
|
|
(vec_select:V8QI
|
11042 |
|
|
(match_dup 1)
|
11043 |
|
|
(parallel [(const_int 1)
|
11044 |
|
|
(const_int 3)
|
11045 |
|
|
(const_int 5)
|
11046 |
|
|
(const_int 7)
|
11047 |
|
|
(const_int 9)
|
11048 |
|
|
(const_int 11)
|
11049 |
|
|
(const_int 13)
|
11050 |
|
|
(const_int 15)])))))]
|
11051 |
|
|
"TARGET_XOP"
|
11052 |
|
|
"vphsubbw\t{%1, %0|%0, %1}"
|
11053 |
|
|
[(set_attr "type" "sseiadd1")])
|
11054 |
|
|
|
11055 |
|
|
(define_insn "xop_phsubwd"
|
11056 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x")
|
11057 |
|
|
(minus:V4SI
|
11058 |
|
|
(sign_extend:V4SI
|
11059 |
|
|
(vec_select:V4HI
|
11060 |
|
|
(match_operand:V8HI 1 "nonimmediate_operand" "xm")
|
11061 |
|
|
(parallel [(const_int 0)
|
11062 |
|
|
(const_int 2)
|
11063 |
|
|
(const_int 4)
|
11064 |
|
|
(const_int 6)])))
|
11065 |
|
|
(sign_extend:V4SI
|
11066 |
|
|
(vec_select:V4HI
|
11067 |
|
|
(match_dup 1)
|
11068 |
|
|
(parallel [(const_int 1)
|
11069 |
|
|
(const_int 3)
|
11070 |
|
|
(const_int 5)
|
11071 |
|
|
(const_int 7)])))))]
|
11072 |
|
|
"TARGET_XOP"
|
11073 |
|
|
"vphsubwd\t{%1, %0|%0, %1}"
|
11074 |
|
|
[(set_attr "type" "sseiadd1")])
|
11075 |
|
|
|
11076 |
|
|
(define_insn "xop_phsubdq"
|
11077 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11078 |
|
|
(minus:V2DI
|
11079 |
|
|
(sign_extend:V2DI
|
11080 |
|
|
(vec_select:V2SI
|
11081 |
|
|
(match_operand:V4SI 1 "nonimmediate_operand" "xm")
|
11082 |
|
|
(parallel [(const_int 0)
|
11083 |
|
|
(const_int 2)])))
|
11084 |
|
|
(sign_extend:V2DI
|
11085 |
|
|
(vec_select:V2SI
|
11086 |
|
|
(match_dup 1)
|
11087 |
|
|
(parallel [(const_int 1)
|
11088 |
|
|
(const_int 3)])))))]
|
11089 |
|
|
"TARGET_XOP"
|
11090 |
|
|
"vphsubdq\t{%1, %0|%0, %1}"
|
11091 |
|
|
[(set_attr "type" "sseiadd1")])
|
11092 |
|
|
|
11093 |
|
|
;; XOP permute instructions
|
11094 |
|
|
(define_insn "xop_pperm"
|
11095 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x,x")
|
11096 |
|
|
(unspec:V16QI
|
11097 |
|
|
[(match_operand:V16QI 1 "register_operand" "x,x")
|
11098 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "x,m")
|
11099 |
|
|
(match_operand:V16QI 3 "nonimmediate_operand" "xm,x")]
|
11100 |
|
|
UNSPEC_XOP_PERMUTE))]
|
11101 |
|
|
"TARGET_XOP && !(MEM_P (operands[2]) && MEM_P (operands[3]))"
|
11102 |
|
|
"vpperm\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
11103 |
|
|
[(set_attr "type" "sse4arg")
|
11104 |
|
|
(set_attr "mode" "TI")])
|
11105 |
|
|
|
11106 |
|
|
;; XOP pack instructions that combine two vectors into a smaller vector
|
11107 |
|
|
(define_insn "xop_pperm_pack_v2di_v4si"
|
11108 |
|
|
[(set (match_operand:V4SI 0 "register_operand" "=x,x")
|
11109 |
|
|
(vec_concat:V4SI
|
11110 |
|
|
(truncate:V2SI
|
11111 |
|
|
(match_operand:V2DI 1 "register_operand" "x,x"))
|
11112 |
|
|
(truncate:V2SI
|
11113 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "x,m"))))
|
11114 |
|
|
(use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x"))]
|
11115 |
|
|
"TARGET_XOP && !(MEM_P (operands[2]) && MEM_P (operands[3]))"
|
11116 |
|
|
"vpperm\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
11117 |
|
|
[(set_attr "type" "sse4arg")
|
11118 |
|
|
(set_attr "mode" "TI")])
|
11119 |
|
|
|
11120 |
|
|
(define_insn "xop_pperm_pack_v4si_v8hi"
|
11121 |
|
|
[(set (match_operand:V8HI 0 "register_operand" "=x,x")
|
11122 |
|
|
(vec_concat:V8HI
|
11123 |
|
|
(truncate:V4HI
|
11124 |
|
|
(match_operand:V4SI 1 "register_operand" "x,x"))
|
11125 |
|
|
(truncate:V4HI
|
11126 |
|
|
(match_operand:V4SI 2 "nonimmediate_operand" "x,m"))))
|
11127 |
|
|
(use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x"))]
|
11128 |
|
|
"TARGET_XOP && !(MEM_P (operands[2]) && MEM_P (operands[3]))"
|
11129 |
|
|
"vpperm\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
11130 |
|
|
[(set_attr "type" "sse4arg")
|
11131 |
|
|
(set_attr "mode" "TI")])
|
11132 |
|
|
|
11133 |
|
|
(define_insn "xop_pperm_pack_v8hi_v16qi"
|
11134 |
|
|
[(set (match_operand:V16QI 0 "register_operand" "=x,x")
|
11135 |
|
|
(vec_concat:V16QI
|
11136 |
|
|
(truncate:V8QI
|
11137 |
|
|
(match_operand:V8HI 1 "register_operand" "x,x"))
|
11138 |
|
|
(truncate:V8QI
|
11139 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "x,m"))))
|
11140 |
|
|
(use (match_operand:V16QI 3 "nonimmediate_operand" "xm,x"))]
|
11141 |
|
|
"TARGET_XOP && !(MEM_P (operands[2]) && MEM_P (operands[3]))"
|
11142 |
|
|
"vpperm\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
11143 |
|
|
[(set_attr "type" "sse4arg")
|
11144 |
|
|
(set_attr "mode" "TI")])
|
11145 |
|
|
|
11146 |
|
|
;; XOP packed rotate instructions
|
11147 |
|
|
(define_expand "rotl3"
|
11148 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "")
|
11149 |
|
|
(rotate:SSEMODE1248
|
11150 |
|
|
(match_operand:SSEMODE1248 1 "nonimmediate_operand" "")
|
11151 |
|
|
(match_operand:SI 2 "general_operand")))]
|
11152 |
|
|
"TARGET_XOP"
|
11153 |
|
|
{
|
11154 |
|
|
/* If we were given a scalar, convert it to parallel */
|
11155 |
|
|
if (! const_0_to__operand (operands[2], SImode))
|
11156 |
|
|
{
|
11157 |
|
|
rtvec vs = rtvec_alloc ();
|
11158 |
|
|
rtx par = gen_rtx_PARALLEL (mode, vs);
|
11159 |
|
|
rtx reg = gen_reg_rtx (mode);
|
11160 |
|
|
rtx op2 = operands[2];
|
11161 |
|
|
int i;
|
11162 |
|
|
|
11163 |
|
|
if (GET_MODE (op2) != mode)
|
11164 |
|
|
{
|
11165 |
|
|
op2 = gen_reg_rtx (mode);
|
11166 |
|
|
convert_move (op2, operands[2], false);
|
11167 |
|
|
}
|
11168 |
|
|
|
11169 |
|
|
for (i = 0; i < ; i++)
|
11170 |
|
|
RTVEC_ELT (vs, i) = op2;
|
11171 |
|
|
|
11172 |
|
|
emit_insn (gen_vec_init (reg, par));
|
11173 |
|
|
emit_insn (gen_xop_vrotl3 (operands[0], operands[1], reg));
|
11174 |
|
|
DONE;
|
11175 |
|
|
}
|
11176 |
|
|
})
|
11177 |
|
|
|
11178 |
|
|
(define_expand "rotr3"
|
11179 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "")
|
11180 |
|
|
(rotatert:SSEMODE1248
|
11181 |
|
|
(match_operand:SSEMODE1248 1 "nonimmediate_operand" "")
|
11182 |
|
|
(match_operand:SI 2 "general_operand")))]
|
11183 |
|
|
"TARGET_XOP"
|
11184 |
|
|
{
|
11185 |
|
|
/* If we were given a scalar, convert it to parallel */
|
11186 |
|
|
if (! const_0_to__operand (operands[2], SImode))
|
11187 |
|
|
{
|
11188 |
|
|
rtvec vs = rtvec_alloc ();
|
11189 |
|
|
rtx par = gen_rtx_PARALLEL (mode, vs);
|
11190 |
|
|
rtx neg = gen_reg_rtx (mode);
|
11191 |
|
|
rtx reg = gen_reg_rtx (mode);
|
11192 |
|
|
rtx op2 = operands[2];
|
11193 |
|
|
int i;
|
11194 |
|
|
|
11195 |
|
|
if (GET_MODE (op2) != mode)
|
11196 |
|
|
{
|
11197 |
|
|
op2 = gen_reg_rtx (mode);
|
11198 |
|
|
convert_move (op2, operands[2], false);
|
11199 |
|
|
}
|
11200 |
|
|
|
11201 |
|
|
for (i = 0; i < ; i++)
|
11202 |
|
|
RTVEC_ELT (vs, i) = op2;
|
11203 |
|
|
|
11204 |
|
|
emit_insn (gen_vec_init (reg, par));
|
11205 |
|
|
emit_insn (gen_neg2 (neg, reg));
|
11206 |
|
|
emit_insn (gen_xop_vrotl3 (operands[0], operands[1], neg));
|
11207 |
|
|
DONE;
|
11208 |
|
|
}
|
11209 |
|
|
})
|
11210 |
|
|
|
11211 |
|
|
(define_insn "xop_rotl3"
|
11212 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x")
|
11213 |
|
|
(rotate:SSEMODE1248
|
11214 |
|
|
(match_operand:SSEMODE1248 1 "nonimmediate_operand" "xm")
|
11215 |
|
|
(match_operand:SI 2 "const_0_to__operand" "n")))]
|
11216 |
|
|
"TARGET_XOP"
|
11217 |
|
|
"vprot\t{%2, %1, %0|%0, %1, %2}"
|
11218 |
|
|
[(set_attr "type" "sseishft")
|
11219 |
|
|
(set_attr "length_immediate" "1")
|
11220 |
|
|
(set_attr "mode" "TI")])
|
11221 |
|
|
|
11222 |
|
|
(define_insn "xop_rotr3"
|
11223 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x")
|
11224 |
|
|
(rotatert:SSEMODE1248
|
11225 |
|
|
(match_operand:SSEMODE1248 1 "nonimmediate_operand" "xm")
|
11226 |
|
|
(match_operand:SI 2 "const_0_to__operand" "n")))]
|
11227 |
|
|
"TARGET_XOP"
|
11228 |
|
|
{
|
11229 |
|
|
operands[3] = GEN_INT (( * 8) - INTVAL (operands[2]));
|
11230 |
|
|
return \"vprot\t{%3, %1, %0|%0, %1, %3}\";
|
11231 |
|
|
}
|
11232 |
|
|
[(set_attr "type" "sseishft")
|
11233 |
|
|
(set_attr "length_immediate" "1")
|
11234 |
|
|
(set_attr "mode" "TI")])
|
11235 |
|
|
|
11236 |
|
|
(define_expand "vrotr3"
|
11237 |
|
|
[(match_operand:SSEMODE1248 0 "register_operand" "")
|
11238 |
|
|
(match_operand:SSEMODE1248 1 "register_operand" "")
|
11239 |
|
|
(match_operand:SSEMODE1248 2 "register_operand" "")]
|
11240 |
|
|
"TARGET_XOP"
|
11241 |
|
|
{
|
11242 |
|
|
rtx reg = gen_reg_rtx (mode);
|
11243 |
|
|
emit_insn (gen_neg2 (reg, operands[2]));
|
11244 |
|
|
emit_insn (gen_xop_vrotl3 (operands[0], operands[1], reg));
|
11245 |
|
|
DONE;
|
11246 |
|
|
})
|
11247 |
|
|
|
11248 |
|
|
(define_expand "vrotl3"
|
11249 |
|
|
[(match_operand:SSEMODE1248 0 "register_operand" "")
|
11250 |
|
|
(match_operand:SSEMODE1248 1 "register_operand" "")
|
11251 |
|
|
(match_operand:SSEMODE1248 2 "register_operand" "")]
|
11252 |
|
|
"TARGET_XOP"
|
11253 |
|
|
{
|
11254 |
|
|
emit_insn (gen_xop_vrotl3 (operands[0], operands[1], operands[2]));
|
11255 |
|
|
DONE;
|
11256 |
|
|
})
|
11257 |
|
|
|
11258 |
|
|
(define_insn "xop_vrotl3"
|
11259 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x,x")
|
11260 |
|
|
(if_then_else:SSEMODE1248
|
11261 |
|
|
(ge:SSEMODE1248
|
11262 |
|
|
(match_operand:SSEMODE1248 2 "nonimmediate_operand" "x,m")
|
11263 |
|
|
(const_int 0))
|
11264 |
|
|
(rotate:SSEMODE1248
|
11265 |
|
|
(match_operand:SSEMODE1248 1 "nonimmediate_operand" "xm,x")
|
11266 |
|
|
(match_dup 2))
|
11267 |
|
|
(rotatert:SSEMODE1248
|
11268 |
|
|
(match_dup 1)
|
11269 |
|
|
(neg:SSEMODE1248 (match_dup 2)))))]
|
11270 |
|
|
"TARGET_XOP && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
11271 |
|
|
"vprot\t{%2, %1, %0|%0, %1, %2}"
|
11272 |
|
|
[(set_attr "type" "sseishft")
|
11273 |
|
|
(set_attr "prefix_data16" "0")
|
11274 |
|
|
(set_attr "prefix_extra" "2")
|
11275 |
|
|
(set_attr "mode" "TI")])
|
11276 |
|
|
|
11277 |
|
|
;; XOP packed shift instructions.
|
11278 |
|
|
;; FIXME: add V2DI back in
|
11279 |
|
|
(define_expand "vlshr3"
|
11280 |
|
|
[(match_operand:SSEMODE124 0 "register_operand" "")
|
11281 |
|
|
(match_operand:SSEMODE124 1 "register_operand" "")
|
11282 |
|
|
(match_operand:SSEMODE124 2 "register_operand" "")]
|
11283 |
|
|
"TARGET_XOP"
|
11284 |
|
|
{
|
11285 |
|
|
rtx neg = gen_reg_rtx (mode);
|
11286 |
|
|
emit_insn (gen_neg2 (neg, operands[2]));
|
11287 |
|
|
emit_insn (gen_xop_lshl3 (operands[0], operands[1], neg));
|
11288 |
|
|
DONE;
|
11289 |
|
|
})
|
11290 |
|
|
|
11291 |
|
|
(define_expand "vashr3"
|
11292 |
|
|
[(match_operand:SSEMODE124 0 "register_operand" "")
|
11293 |
|
|
(match_operand:SSEMODE124 1 "register_operand" "")
|
11294 |
|
|
(match_operand:SSEMODE124 2 "register_operand" "")]
|
11295 |
|
|
"TARGET_XOP"
|
11296 |
|
|
{
|
11297 |
|
|
rtx neg = gen_reg_rtx (mode);
|
11298 |
|
|
emit_insn (gen_neg2 (neg, operands[2]));
|
11299 |
|
|
emit_insn (gen_xop_ashl3 (operands[0], operands[1], neg));
|
11300 |
|
|
DONE;
|
11301 |
|
|
})
|
11302 |
|
|
|
11303 |
|
|
(define_expand "vashl3"
|
11304 |
|
|
[(match_operand:SSEMODE124 0 "register_operand" "")
|
11305 |
|
|
(match_operand:SSEMODE124 1 "register_operand" "")
|
11306 |
|
|
(match_operand:SSEMODE124 2 "register_operand" "")]
|
11307 |
|
|
"TARGET_XOP"
|
11308 |
|
|
{
|
11309 |
|
|
emit_insn (gen_xop_ashl3 (operands[0], operands[1], operands[2]));
|
11310 |
|
|
DONE;
|
11311 |
|
|
})
|
11312 |
|
|
|
11313 |
|
|
(define_insn "xop_ashl3"
|
11314 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x,x")
|
11315 |
|
|
(if_then_else:SSEMODE1248
|
11316 |
|
|
(ge:SSEMODE1248
|
11317 |
|
|
(match_operand:SSEMODE1248 2 "nonimmediate_operand" "x,m")
|
11318 |
|
|
(const_int 0))
|
11319 |
|
|
(ashift:SSEMODE1248
|
11320 |
|
|
(match_operand:SSEMODE1248 1 "nonimmediate_operand" "xm,x")
|
11321 |
|
|
(match_dup 2))
|
11322 |
|
|
(ashiftrt:SSEMODE1248
|
11323 |
|
|
(match_dup 1)
|
11324 |
|
|
(neg:SSEMODE1248 (match_dup 2)))))]
|
11325 |
|
|
"TARGET_XOP && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
11326 |
|
|
"vpsha\t{%2, %1, %0|%0, %1, %2}"
|
11327 |
|
|
[(set_attr "type" "sseishft")
|
11328 |
|
|
(set_attr "prefix_data16" "0")
|
11329 |
|
|
(set_attr "prefix_extra" "2")
|
11330 |
|
|
(set_attr "mode" "TI")])
|
11331 |
|
|
|
11332 |
|
|
(define_insn "xop_lshl3"
|
11333 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x,x")
|
11334 |
|
|
(if_then_else:SSEMODE1248
|
11335 |
|
|
(ge:SSEMODE1248
|
11336 |
|
|
(match_operand:SSEMODE1248 2 "nonimmediate_operand" "x,m")
|
11337 |
|
|
(const_int 0))
|
11338 |
|
|
(ashift:SSEMODE1248
|
11339 |
|
|
(match_operand:SSEMODE1248 1 "nonimmediate_operand" "xm,x")
|
11340 |
|
|
(match_dup 2))
|
11341 |
|
|
(lshiftrt:SSEMODE1248
|
11342 |
|
|
(match_dup 1)
|
11343 |
|
|
(neg:SSEMODE1248 (match_dup 2)))))]
|
11344 |
|
|
"TARGET_XOP && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
|
11345 |
|
|
"vpshl\t{%2, %1, %0|%0, %1, %2}"
|
11346 |
|
|
[(set_attr "type" "sseishft")
|
11347 |
|
|
(set_attr "prefix_data16" "0")
|
11348 |
|
|
(set_attr "prefix_extra" "2")
|
11349 |
|
|
(set_attr "mode" "TI")])
|
11350 |
|
|
|
11351 |
|
|
;; SSE2 doesn't have some shift varients, so define versions for XOP
|
11352 |
|
|
(define_expand "ashlv16qi3"
|
11353 |
|
|
[(match_operand:V16QI 0 "register_operand" "")
|
11354 |
|
|
(match_operand:V16QI 1 "register_operand" "")
|
11355 |
|
|
(match_operand:SI 2 "nonmemory_operand" "")]
|
11356 |
|
|
"TARGET_XOP"
|
11357 |
|
|
{
|
11358 |
|
|
rtvec vs = rtvec_alloc (16);
|
11359 |
|
|
rtx par = gen_rtx_PARALLEL (V16QImode, vs);
|
11360 |
|
|
rtx reg = gen_reg_rtx (V16QImode);
|
11361 |
|
|
int i;
|
11362 |
|
|
for (i = 0; i < 16; i++)
|
11363 |
|
|
RTVEC_ELT (vs, i) = operands[2];
|
11364 |
|
|
|
11365 |
|
|
emit_insn (gen_vec_initv16qi (reg, par));
|
11366 |
|
|
emit_insn (gen_xop_ashlv16qi3 (operands[0], operands[1], reg));
|
11367 |
|
|
DONE;
|
11368 |
|
|
})
|
11369 |
|
|
|
11370 |
|
|
(define_expand "lshlv16qi3"
|
11371 |
|
|
[(match_operand:V16QI 0 "register_operand" "")
|
11372 |
|
|
(match_operand:V16QI 1 "register_operand" "")
|
11373 |
|
|
(match_operand:SI 2 "nonmemory_operand" "")]
|
11374 |
|
|
"TARGET_XOP"
|
11375 |
|
|
{
|
11376 |
|
|
rtvec vs = rtvec_alloc (16);
|
11377 |
|
|
rtx par = gen_rtx_PARALLEL (V16QImode, vs);
|
11378 |
|
|
rtx reg = gen_reg_rtx (V16QImode);
|
11379 |
|
|
int i;
|
11380 |
|
|
for (i = 0; i < 16; i++)
|
11381 |
|
|
RTVEC_ELT (vs, i) = operands[2];
|
11382 |
|
|
|
11383 |
|
|
emit_insn (gen_vec_initv16qi (reg, par));
|
11384 |
|
|
emit_insn (gen_xop_lshlv16qi3 (operands[0], operands[1], reg));
|
11385 |
|
|
DONE;
|
11386 |
|
|
})
|
11387 |
|
|
|
11388 |
|
|
(define_expand "ashrv16qi3"
|
11389 |
|
|
[(match_operand:V16QI 0 "register_operand" "")
|
11390 |
|
|
(match_operand:V16QI 1 "register_operand" "")
|
11391 |
|
|
(match_operand:SI 2 "nonmemory_operand" "")]
|
11392 |
|
|
"TARGET_XOP"
|
11393 |
|
|
{
|
11394 |
|
|
rtvec vs = rtvec_alloc (16);
|
11395 |
|
|
rtx par = gen_rtx_PARALLEL (V16QImode, vs);
|
11396 |
|
|
rtx reg = gen_reg_rtx (V16QImode);
|
11397 |
|
|
int i;
|
11398 |
|
|
rtx ele = ((CONST_INT_P (operands[2]))
|
11399 |
|
|
? GEN_INT (- INTVAL (operands[2]))
|
11400 |
|
|
: operands[2]);
|
11401 |
|
|
|
11402 |
|
|
for (i = 0; i < 16; i++)
|
11403 |
|
|
RTVEC_ELT (vs, i) = ele;
|
11404 |
|
|
|
11405 |
|
|
emit_insn (gen_vec_initv16qi (reg, par));
|
11406 |
|
|
|
11407 |
|
|
if (!CONST_INT_P (operands[2]))
|
11408 |
|
|
{
|
11409 |
|
|
rtx neg = gen_reg_rtx (V16QImode);
|
11410 |
|
|
emit_insn (gen_negv16qi2 (neg, reg));
|
11411 |
|
|
emit_insn (gen_xop_ashlv16qi3 (operands[0], operands[1], neg));
|
11412 |
|
|
}
|
11413 |
|
|
else
|
11414 |
|
|
emit_insn (gen_xop_ashlv16qi3 (operands[0], operands[1], reg));
|
11415 |
|
|
|
11416 |
|
|
DONE;
|
11417 |
|
|
})
|
11418 |
|
|
|
11419 |
|
|
(define_expand "ashrv2di3"
|
11420 |
|
|
[(match_operand:V2DI 0 "register_operand" "")
|
11421 |
|
|
(match_operand:V2DI 1 "register_operand" "")
|
11422 |
|
|
(match_operand:DI 2 "nonmemory_operand" "")]
|
11423 |
|
|
"TARGET_XOP"
|
11424 |
|
|
{
|
11425 |
|
|
rtvec vs = rtvec_alloc (2);
|
11426 |
|
|
rtx par = gen_rtx_PARALLEL (V2DImode, vs);
|
11427 |
|
|
rtx reg = gen_reg_rtx (V2DImode);
|
11428 |
|
|
rtx ele;
|
11429 |
|
|
|
11430 |
|
|
if (CONST_INT_P (operands[2]))
|
11431 |
|
|
ele = GEN_INT (- INTVAL (operands[2]));
|
11432 |
|
|
else if (GET_MODE (operands[2]) != DImode)
|
11433 |
|
|
{
|
11434 |
|
|
rtx move = gen_reg_rtx (DImode);
|
11435 |
|
|
ele = gen_reg_rtx (DImode);
|
11436 |
|
|
convert_move (move, operands[2], false);
|
11437 |
|
|
emit_insn (gen_negdi2 (ele, move));
|
11438 |
|
|
}
|
11439 |
|
|
else
|
11440 |
|
|
{
|
11441 |
|
|
ele = gen_reg_rtx (DImode);
|
11442 |
|
|
emit_insn (gen_negdi2 (ele, operands[2]));
|
11443 |
|
|
}
|
11444 |
|
|
|
11445 |
|
|
RTVEC_ELT (vs, 0) = ele;
|
11446 |
|
|
RTVEC_ELT (vs, 1) = ele;
|
11447 |
|
|
emit_insn (gen_vec_initv2di (reg, par));
|
11448 |
|
|
emit_insn (gen_xop_ashlv2di3 (operands[0], operands[1], reg));
|
11449 |
|
|
DONE;
|
11450 |
|
|
})
|
11451 |
|
|
|
11452 |
|
|
;; XOP FRCZ support
|
11453 |
|
|
;; parallel insns
|
11454 |
|
|
(define_insn "xop_frcz2"
|
11455 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
11456 |
|
|
(unspec:SSEMODEF2P
|
11457 |
|
|
[(match_operand:SSEMODEF2P 1 "nonimmediate_operand" "xm")]
|
11458 |
|
|
UNSPEC_FRCZ))]
|
11459 |
|
|
"TARGET_XOP"
|
11460 |
|
|
"vfrcz\t{%1, %0|%0, %1}"
|
11461 |
|
|
[(set_attr "type" "ssecvt1")
|
11462 |
|
|
(set_attr "mode" "")])
|
11463 |
|
|
|
11464 |
|
|
;; scalar insns
|
11465 |
|
|
(define_insn "xop_vmfrcz2"
|
11466 |
|
|
[(set (match_operand:SSEMODEF2P 0 "register_operand" "=x")
|
11467 |
|
|
(vec_merge:SSEMODEF2P
|
11468 |
|
|
(unspec:SSEMODEF2P
|
11469 |
|
|
[(match_operand:SSEMODEF2P 2 "nonimmediate_operand" "xm")]
|
11470 |
|
|
UNSPEC_FRCZ)
|
11471 |
|
|
(match_operand:SSEMODEF2P 1 "register_operand" "0")
|
11472 |
|
|
(const_int 1)))]
|
11473 |
|
|
"TARGET_XOP"
|
11474 |
|
|
"vfrcz\t{%2, %0|%0, %2}"
|
11475 |
|
|
[(set_attr "type" "ssecvt1")
|
11476 |
|
|
(set_attr "mode" "")])
|
11477 |
|
|
|
11478 |
|
|
(define_insn "xop_frcz2256"
|
11479 |
|
|
[(set (match_operand:FMA4MODEF4 0 "register_operand" "=x")
|
11480 |
|
|
(unspec:FMA4MODEF4
|
11481 |
|
|
[(match_operand:FMA4MODEF4 1 "nonimmediate_operand" "xm")]
|
11482 |
|
|
UNSPEC_FRCZ))]
|
11483 |
|
|
"TARGET_XOP"
|
11484 |
|
|
"vfrcz\t{%1, %0|%0, %1}"
|
11485 |
|
|
[(set_attr "type" "ssecvt1")
|
11486 |
|
|
(set_attr "mode" "")])
|
11487 |
|
|
|
11488 |
|
|
(define_insn "xop_maskcmp3"
|
11489 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x")
|
11490 |
|
|
(match_operator:SSEMODE1248 1 "ix86_comparison_int_operator"
|
11491 |
|
|
[(match_operand:SSEMODE1248 2 "register_operand" "x")
|
11492 |
|
|
(match_operand:SSEMODE1248 3 "nonimmediate_operand" "xm")]))]
|
11493 |
|
|
"TARGET_XOP"
|
11494 |
|
|
"vpcom%Y1\t{%3, %2, %0|%0, %2, %3}"
|
11495 |
|
|
[(set_attr "type" "sse4arg")
|
11496 |
|
|
(set_attr "prefix_data16" "0")
|
11497 |
|
|
(set_attr "prefix_rep" "0")
|
11498 |
|
|
(set_attr "prefix_extra" "2")
|
11499 |
|
|
(set_attr "length_immediate" "1")
|
11500 |
|
|
(set_attr "mode" "TI")])
|
11501 |
|
|
|
11502 |
|
|
(define_insn "xop_maskcmp_uns3"
|
11503 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x")
|
11504 |
|
|
(match_operator:SSEMODE1248 1 "ix86_comparison_uns_operator"
|
11505 |
|
|
[(match_operand:SSEMODE1248 2 "register_operand" "x")
|
11506 |
|
|
(match_operand:SSEMODE1248 3 "nonimmediate_operand" "xm")]))]
|
11507 |
|
|
"TARGET_XOP"
|
11508 |
|
|
"vpcom%Y1u\t{%3, %2, %0|%0, %2, %3}"
|
11509 |
|
|
[(set_attr "type" "ssecmp")
|
11510 |
|
|
(set_attr "prefix_data16" "0")
|
11511 |
|
|
(set_attr "prefix_rep" "0")
|
11512 |
|
|
(set_attr "prefix_extra" "2")
|
11513 |
|
|
(set_attr "length_immediate" "1")
|
11514 |
|
|
(set_attr "mode" "TI")])
|
11515 |
|
|
|
11516 |
|
|
;; Version of pcom*u* that is called from the intrinsics that allows pcomequ*
|
11517 |
|
|
;; and pcomneu* not to be converted to the signed ones in case somebody needs
|
11518 |
|
|
;; the exact instruction generated for the intrinsic.
|
11519 |
|
|
(define_insn "xop_maskcmp_uns23"
|
11520 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x")
|
11521 |
|
|
(unspec:SSEMODE1248
|
11522 |
|
|
[(match_operator:SSEMODE1248 1 "ix86_comparison_uns_operator"
|
11523 |
|
|
[(match_operand:SSEMODE1248 2 "register_operand" "x")
|
11524 |
|
|
(match_operand:SSEMODE1248 3 "nonimmediate_operand" "xm")])]
|
11525 |
|
|
UNSPEC_XOP_UNSIGNED_CMP))]
|
11526 |
|
|
"TARGET_XOP"
|
11527 |
|
|
"vpcom%Y1u\t{%3, %2, %0|%0, %2, %3}"
|
11528 |
|
|
[(set_attr "type" "ssecmp")
|
11529 |
|
|
(set_attr "prefix_data16" "0")
|
11530 |
|
|
(set_attr "prefix_extra" "2")
|
11531 |
|
|
(set_attr "length_immediate" "1")
|
11532 |
|
|
(set_attr "mode" "TI")])
|
11533 |
|
|
|
11534 |
|
|
;; Pcomtrue and pcomfalse support. These are useless instructions, but are
|
11535 |
|
|
;; being added here to be complete.
|
11536 |
|
|
(define_insn "xop_pcom_tf3"
|
11537 |
|
|
[(set (match_operand:SSEMODE1248 0 "register_operand" "=x")
|
11538 |
|
|
(unspec:SSEMODE1248
|
11539 |
|
|
[(match_operand:SSEMODE1248 1 "register_operand" "x")
|
11540 |
|
|
(match_operand:SSEMODE1248 2 "nonimmediate_operand" "xm")
|
11541 |
|
|
(match_operand:SI 3 "const_int_operand" "n")]
|
11542 |
|
|
UNSPEC_XOP_TRUEFALSE))]
|
11543 |
|
|
"TARGET_XOP"
|
11544 |
|
|
{
|
11545 |
|
|
return ((INTVAL (operands[3]) != 0)
|
11546 |
|
|
? "vpcomtrue\t{%2, %1, %0|%0, %1, %2}"
|
11547 |
|
|
: "vpcomfalse\t{%2, %1, %0|%0, %1, %2}");
|
11548 |
|
|
}
|
11549 |
|
|
[(set_attr "type" "ssecmp")
|
11550 |
|
|
(set_attr "prefix_data16" "0")
|
11551 |
|
|
(set_attr "prefix_extra" "2")
|
11552 |
|
|
(set_attr "length_immediate" "1")
|
11553 |
|
|
(set_attr "mode" "TI")])
|
11554 |
|
|
|
11555 |
|
|
(define_insn "xop_vpermil23"
|
11556 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
11557 |
|
|
(unspec:AVXMODEF2P
|
11558 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
11559 |
|
|
(match_operand:AVXMODEF2P 2 "nonimmediate_operand" "%x")
|
11560 |
|
|
(match_operand: 3 "nonimmediate_operand" "xm")
|
11561 |
|
|
(match_operand:SI 4 "const_0_to_3_operand" "n")]
|
11562 |
|
|
UNSPEC_VPERMIL2))]
|
11563 |
|
|
"TARGET_XOP"
|
11564 |
|
|
"vpermil2p\t{%4, %3, %2, %1, %0|%0, %1, %2, %3, %4}"
|
11565 |
|
|
[(set_attr "type" "sse4arg")
|
11566 |
|
|
(set_attr "length_immediate" "1")
|
11567 |
|
|
(set_attr "mode" "")])
|
11568 |
|
|
|
11569 |
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
11570 |
|
|
(define_insn "*avx_aesenc"
|
11571 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11572 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "x")
|
11573 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")]
|
11574 |
|
|
UNSPEC_AESENC))]
|
11575 |
|
|
"TARGET_AES && TARGET_AVX"
|
11576 |
|
|
"vaesenc\t{%2, %1, %0|%0, %1, %2}"
|
11577 |
|
|
[(set_attr "type" "sselog1")
|
11578 |
|
|
(set_attr "prefix_extra" "1")
|
11579 |
|
|
(set_attr "prefix" "vex")
|
11580 |
|
|
(set_attr "mode" "TI")])
|
11581 |
|
|
|
11582 |
|
|
(define_insn "aesenc"
|
11583 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11584 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
11585 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")]
|
11586 |
|
|
UNSPEC_AESENC))]
|
11587 |
|
|
"TARGET_AES"
|
11588 |
|
|
"aesenc\t{%2, %0|%0, %2}"
|
11589 |
|
|
[(set_attr "type" "sselog1")
|
11590 |
|
|
(set_attr "prefix_extra" "1")
|
11591 |
|
|
(set_attr "mode" "TI")])
|
11592 |
|
|
|
11593 |
|
|
(define_insn "*avx_aesenclast"
|
11594 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11595 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "x")
|
11596 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")]
|
11597 |
|
|
UNSPEC_AESENCLAST))]
|
11598 |
|
|
"TARGET_AES && TARGET_AVX"
|
11599 |
|
|
"vaesenclast\t{%2, %1, %0|%0, %1, %2}"
|
11600 |
|
|
[(set_attr "type" "sselog1")
|
11601 |
|
|
(set_attr "prefix_extra" "1")
|
11602 |
|
|
(set_attr "prefix" "vex")
|
11603 |
|
|
(set_attr "mode" "TI")])
|
11604 |
|
|
|
11605 |
|
|
(define_insn "aesenclast"
|
11606 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11607 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
11608 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")]
|
11609 |
|
|
UNSPEC_AESENCLAST))]
|
11610 |
|
|
"TARGET_AES"
|
11611 |
|
|
"aesenclast\t{%2, %0|%0, %2}"
|
11612 |
|
|
[(set_attr "type" "sselog1")
|
11613 |
|
|
(set_attr "prefix_extra" "1")
|
11614 |
|
|
(set_attr "mode" "TI")])
|
11615 |
|
|
|
11616 |
|
|
(define_insn "*avx_aesdec"
|
11617 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11618 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "x")
|
11619 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")]
|
11620 |
|
|
UNSPEC_AESDEC))]
|
11621 |
|
|
"TARGET_AES && TARGET_AVX"
|
11622 |
|
|
"vaesdec\t{%2, %1, %0|%0, %1, %2}"
|
11623 |
|
|
[(set_attr "type" "sselog1")
|
11624 |
|
|
(set_attr "prefix_extra" "1")
|
11625 |
|
|
(set_attr "prefix" "vex")
|
11626 |
|
|
(set_attr "mode" "TI")])
|
11627 |
|
|
|
11628 |
|
|
(define_insn "aesdec"
|
11629 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11630 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
11631 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")]
|
11632 |
|
|
UNSPEC_AESDEC))]
|
11633 |
|
|
"TARGET_AES"
|
11634 |
|
|
"aesdec\t{%2, %0|%0, %2}"
|
11635 |
|
|
[(set_attr "type" "sselog1")
|
11636 |
|
|
(set_attr "prefix_extra" "1")
|
11637 |
|
|
(set_attr "mode" "TI")])
|
11638 |
|
|
|
11639 |
|
|
(define_insn "*avx_aesdeclast"
|
11640 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11641 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "x")
|
11642 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")]
|
11643 |
|
|
UNSPEC_AESDECLAST))]
|
11644 |
|
|
"TARGET_AES && TARGET_AVX"
|
11645 |
|
|
"vaesdeclast\t{%2, %1, %0|%0, %1, %2}"
|
11646 |
|
|
[(set_attr "type" "sselog1")
|
11647 |
|
|
(set_attr "prefix_extra" "1")
|
11648 |
|
|
(set_attr "prefix" "vex")
|
11649 |
|
|
(set_attr "mode" "TI")])
|
11650 |
|
|
|
11651 |
|
|
(define_insn "aesdeclast"
|
11652 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11653 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
11654 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")]
|
11655 |
|
|
UNSPEC_AESDECLAST))]
|
11656 |
|
|
"TARGET_AES"
|
11657 |
|
|
"aesdeclast\t{%2, %0|%0, %2}"
|
11658 |
|
|
[(set_attr "type" "sselog1")
|
11659 |
|
|
(set_attr "prefix_extra" "1")
|
11660 |
|
|
(set_attr "mode" "TI")])
|
11661 |
|
|
|
11662 |
|
|
(define_insn "aesimc"
|
11663 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11664 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "nonimmediate_operand" "xm")]
|
11665 |
|
|
UNSPEC_AESIMC))]
|
11666 |
|
|
"TARGET_AES"
|
11667 |
|
|
"%vaesimc\t{%1, %0|%0, %1}"
|
11668 |
|
|
[(set_attr "type" "sselog1")
|
11669 |
|
|
(set_attr "prefix_extra" "1")
|
11670 |
|
|
(set_attr "prefix" "maybe_vex")
|
11671 |
|
|
(set_attr "mode" "TI")])
|
11672 |
|
|
|
11673 |
|
|
(define_insn "aeskeygenassist"
|
11674 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11675 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "nonimmediate_operand" "xm")
|
11676 |
|
|
(match_operand:SI 2 "const_0_to_255_operand" "n")]
|
11677 |
|
|
UNSPEC_AESKEYGENASSIST))]
|
11678 |
|
|
"TARGET_AES"
|
11679 |
|
|
"%vaeskeygenassist\t{%2, %1, %0|%0, %1, %2}"
|
11680 |
|
|
[(set_attr "type" "sselog1")
|
11681 |
|
|
(set_attr "prefix_extra" "1")
|
11682 |
|
|
(set_attr "length_immediate" "1")
|
11683 |
|
|
(set_attr "prefix" "maybe_vex")
|
11684 |
|
|
(set_attr "mode" "TI")])
|
11685 |
|
|
|
11686 |
|
|
(define_insn "*vpclmulqdq"
|
11687 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11688 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "x")
|
11689 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")
|
11690 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
11691 |
|
|
UNSPEC_PCLMUL))]
|
11692 |
|
|
"TARGET_PCLMUL && TARGET_AVX"
|
11693 |
|
|
"vpclmulqdq\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
11694 |
|
|
[(set_attr "type" "sselog1")
|
11695 |
|
|
(set_attr "prefix_extra" "1")
|
11696 |
|
|
(set_attr "length_immediate" "1")
|
11697 |
|
|
(set_attr "prefix" "vex")
|
11698 |
|
|
(set_attr "mode" "TI")])
|
11699 |
|
|
|
11700 |
|
|
(define_insn "pclmulqdq"
|
11701 |
|
|
[(set (match_operand:V2DI 0 "register_operand" "=x")
|
11702 |
|
|
(unspec:V2DI [(match_operand:V2DI 1 "register_operand" "0")
|
11703 |
|
|
(match_operand:V2DI 2 "nonimmediate_operand" "xm")
|
11704 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
11705 |
|
|
UNSPEC_PCLMUL))]
|
11706 |
|
|
"TARGET_PCLMUL"
|
11707 |
|
|
"pclmulqdq\t{%3, %2, %0|%0, %2, %3}"
|
11708 |
|
|
[(set_attr "type" "sselog1")
|
11709 |
|
|
(set_attr "prefix_extra" "1")
|
11710 |
|
|
(set_attr "length_immediate" "1")
|
11711 |
|
|
(set_attr "mode" "TI")])
|
11712 |
|
|
|
11713 |
|
|
(define_expand "avx_vzeroall"
|
11714 |
|
|
[(match_par_dup 0 [(const_int 0)])]
|
11715 |
|
|
"TARGET_AVX"
|
11716 |
|
|
{
|
11717 |
|
|
int nregs = TARGET_64BIT ? 16 : 8;
|
11718 |
|
|
int regno;
|
11719 |
|
|
|
11720 |
|
|
operands[0] = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (nregs + 1));
|
11721 |
|
|
|
11722 |
|
|
XVECEXP (operands[0], 0, 0)
|
11723 |
|
|
= gen_rtx_UNSPEC_VOLATILE (VOIDmode, gen_rtvec (1, const0_rtx),
|
11724 |
|
|
UNSPECV_VZEROALL);
|
11725 |
|
|
|
11726 |
|
|
for (regno = 0; regno < nregs; regno++)
|
11727 |
|
|
XVECEXP (operands[0], 0, regno + 1)
|
11728 |
|
|
= gen_rtx_SET (VOIDmode,
|
11729 |
|
|
gen_rtx_REG (V8SImode, SSE_REGNO (regno)),
|
11730 |
|
|
CONST0_RTX (V8SImode));
|
11731 |
|
|
})
|
11732 |
|
|
|
11733 |
|
|
(define_insn "*avx_vzeroall"
|
11734 |
|
|
[(match_parallel 0 "vzeroall_operation"
|
11735 |
|
|
[(unspec_volatile [(const_int 0)] UNSPECV_VZEROALL)])]
|
11736 |
|
|
"TARGET_AVX"
|
11737 |
|
|
"vzeroall"
|
11738 |
|
|
[(set_attr "type" "sse")
|
11739 |
|
|
(set_attr "modrm" "0")
|
11740 |
|
|
(set_attr "memory" "none")
|
11741 |
|
|
(set_attr "prefix" "vex")
|
11742 |
|
|
(set_attr "mode" "OI")])
|
11743 |
|
|
|
11744 |
|
|
;; vzeroupper clobbers the upper 128bits of AVX registers.
|
11745 |
|
|
(define_expand "avx_vzeroupper"
|
11746 |
|
|
[(match_par_dup 0 [(const_int 0)])]
|
11747 |
|
|
"TARGET_AVX"
|
11748 |
|
|
{
|
11749 |
|
|
int nregs = TARGET_64BIT ? 16 : 8;
|
11750 |
|
|
int regno;
|
11751 |
|
|
|
11752 |
|
|
operands[0] = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (nregs + 1));
|
11753 |
|
|
|
11754 |
|
|
XVECEXP (operands[0], 0, 0)
|
11755 |
|
|
= gen_rtx_UNSPEC_VOLATILE (VOIDmode, gen_rtvec (1, const0_rtx),
|
11756 |
|
|
UNSPECV_VZEROUPPER);
|
11757 |
|
|
|
11758 |
|
|
for (regno = 0; regno < nregs; regno++)
|
11759 |
|
|
XVECEXP (operands[0], 0, regno + 1)
|
11760 |
|
|
= gen_rtx_CLOBBER (VOIDmode,
|
11761 |
|
|
gen_rtx_REG (V8SImode, SSE_REGNO (regno)));
|
11762 |
|
|
})
|
11763 |
|
|
|
11764 |
|
|
(define_insn "*avx_vzeroupper"
|
11765 |
|
|
[(match_parallel 0 "vzeroupper_operation"
|
11766 |
|
|
[(unspec_volatile [(const_int 0)] UNSPECV_VZEROUPPER)])]
|
11767 |
|
|
"TARGET_AVX"
|
11768 |
|
|
"vzeroupper"
|
11769 |
|
|
[(set_attr "type" "sse")
|
11770 |
|
|
(set_attr "modrm" "0")
|
11771 |
|
|
(set_attr "memory" "none")
|
11772 |
|
|
(set_attr "prefix" "vex")
|
11773 |
|
|
(set_attr "mode" "OI")])
|
11774 |
|
|
|
11775 |
|
|
(define_insn_and_split "vec_dup"
|
11776 |
|
|
[(set (match_operand:AVX256MODE24P 0 "register_operand" "=x,x")
|
11777 |
|
|
(vec_duplicate:AVX256MODE24P
|
11778 |
|
|
(match_operand: 1 "nonimmediate_operand" "m,?x")))]
|
11779 |
|
|
"TARGET_AVX"
|
11780 |
|
|
"@
|
11781 |
|
|
vbroadcasts\t{%1, %0|%0, %1}
|
11782 |
|
|
#"
|
11783 |
|
|
"&& reload_completed && REG_P (operands[1])"
|
11784 |
|
|
[(set (match_dup 2) (vec_duplicate: (match_dup 1)))
|
11785 |
|
|
(set (match_dup 0) (vec_concat:AVX256MODE24P (match_dup 2) (match_dup 2)))]
|
11786 |
|
|
{
|
11787 |
|
|
operands[2] = gen_rtx_REG (mode, REGNO (operands[0]));
|
11788 |
|
|
}
|
11789 |
|
|
[(set_attr "type" "ssemov")
|
11790 |
|
|
(set_attr "prefix_extra" "1")
|
11791 |
|
|
(set_attr "prefix" "vex")
|
11792 |
|
|
(set_attr "mode" "V8SF")])
|
11793 |
|
|
|
11794 |
|
|
(define_insn "avx_vbroadcastf128_"
|
11795 |
|
|
[(set (match_operand:AVX256MODE 0 "register_operand" "=x,x,x")
|
11796 |
|
|
(vec_concat:AVX256MODE
|
11797 |
|
|
(match_operand: 1 "nonimmediate_operand" "m,0,?x")
|
11798 |
|
|
(match_dup 1)))]
|
11799 |
|
|
"TARGET_AVX"
|
11800 |
|
|
"@
|
11801 |
|
|
vbroadcastf128\t{%1, %0|%0, %1}
|
11802 |
|
|
vinsertf128\t{$1, %1, %0, %0|%0, %0, %1, 1}
|
11803 |
|
|
vperm2f128\t{$0, %t1, %t1, %0|%0, %t1, %t1, 0}"
|
11804 |
|
|
[(set_attr "type" "ssemov,sselog1,sselog1")
|
11805 |
|
|
(set_attr "prefix_extra" "1")
|
11806 |
|
|
(set_attr "length_immediate" "0,1,1")
|
11807 |
|
|
(set_attr "prefix" "vex")
|
11808 |
|
|
(set_attr "mode" "V4SF,V8SF,V8SF")])
|
11809 |
|
|
|
11810 |
|
|
;; Recognize broadcast as a vec_select as produced by builtin_vec_perm.
|
11811 |
|
|
;; If it so happens that the input is in memory, use vbroadcast.
|
11812 |
|
|
;; Otherwise use vpermilp (and in the case of 256-bit modes, vperm2f128).
|
11813 |
|
|
(define_insn "*avx_vperm_broadcast_v4sf"
|
11814 |
|
|
[(set (match_operand:V4SF 0 "register_operand" "=x,x,x")
|
11815 |
|
|
(vec_select:V4SF
|
11816 |
|
|
(match_operand:V4SF 1 "nonimmediate_operand" "m,o,x")
|
11817 |
|
|
(match_parallel 2 "avx_vbroadcast_operand"
|
11818 |
|
|
[(match_operand 3 "const_int_operand" "C,n,n")])))]
|
11819 |
|
|
"TARGET_AVX"
|
11820 |
|
|
{
|
11821 |
|
|
int elt = INTVAL (operands[3]);
|
11822 |
|
|
switch (which_alternative)
|
11823 |
|
|
{
|
11824 |
|
|
case 0:
|
11825 |
|
|
case 1:
|
11826 |
|
|
operands[1] = adjust_address_nv (operands[1], SFmode, elt * 4);
|
11827 |
|
|
return "vbroadcastss\t{%1, %0|%0, %1}";
|
11828 |
|
|
case 2:
|
11829 |
|
|
operands[2] = GEN_INT (elt * 0x55);
|
11830 |
|
|
return "vpermilps\t{%2, %1, %0|%0, %1, %2}";
|
11831 |
|
|
default:
|
11832 |
|
|
gcc_unreachable ();
|
11833 |
|
|
}
|
11834 |
|
|
}
|
11835 |
|
|
[(set_attr "type" "ssemov,ssemov,sselog1")
|
11836 |
|
|
(set_attr "prefix_extra" "1")
|
11837 |
|
|
(set_attr "length_immediate" "0,0,1")
|
11838 |
|
|
(set_attr "prefix" "vex")
|
11839 |
|
|
(set_attr "mode" "SF,SF,V4SF")])
|
11840 |
|
|
|
11841 |
|
|
(define_insn_and_split "*avx_vperm_broadcast_"
|
11842 |
|
|
[(set (match_operand:AVX256MODEF2P 0 "register_operand" "=x,x,x")
|
11843 |
|
|
(vec_select:AVX256MODEF2P
|
11844 |
|
|
(match_operand:AVX256MODEF2P 1 "nonimmediate_operand" "m,o,?x")
|
11845 |
|
|
(match_parallel 2 "avx_vbroadcast_operand"
|
11846 |
|
|
[(match_operand 3 "const_int_operand" "C,n,n")])))]
|
11847 |
|
|
"TARGET_AVX"
|
11848 |
|
|
"#"
|
11849 |
|
|
"&& reload_completed"
|
11850 |
|
|
[(set (match_dup 0) (vec_duplicate:AVX256MODEF2P (match_dup 1)))]
|
11851 |
|
|
{
|
11852 |
|
|
rtx op0 = operands[0], op1 = operands[1];
|
11853 |
|
|
int elt = INTVAL (operands[3]);
|
11854 |
|
|
|
11855 |
|
|
if (REG_P (op1))
|
11856 |
|
|
{
|
11857 |
|
|
int mask;
|
11858 |
|
|
|
11859 |
|
|
/* Shuffle element we care about into all elements of the 128-bit lane.
|
11860 |
|
|
The other lane gets shuffled too, but we don't care. */
|
11861 |
|
|
if (mode == V4DFmode)
|
11862 |
|
|
mask = (elt & 1 ? 15 : 0);
|
11863 |
|
|
else
|
11864 |
|
|
mask = (elt & 3) * 0x55;
|
11865 |
|
|
emit_insn (gen_avx_vpermil (op0, op1, GEN_INT (mask)));
|
11866 |
|
|
|
11867 |
|
|
/* Shuffle the lane we care about into both lanes of the dest. */
|
11868 |
|
|
mask = (elt / ( / 2)) * 0x11;
|
11869 |
|
|
emit_insn (gen_avx_vperm2f1283 (op0, op0, op0, GEN_INT (mask)));
|
11870 |
|
|
DONE;
|
11871 |
|
|
}
|
11872 |
|
|
|
11873 |
|
|
operands[1] = adjust_address_nv (op1, mode,
|
11874 |
|
|
elt * GET_MODE_SIZE (mode));
|
11875 |
|
|
})
|
11876 |
|
|
|
11877 |
|
|
(define_expand "avx_vpermil"
|
11878 |
|
|
[(set (match_operand:AVXMODEFDP 0 "register_operand" "")
|
11879 |
|
|
(vec_select:AVXMODEFDP
|
11880 |
|
|
(match_operand:AVXMODEFDP 1 "nonimmediate_operand" "")
|
11881 |
|
|
(match_operand:SI 2 "const_0_to_255_operand" "")))]
|
11882 |
|
|
"TARGET_AVX"
|
11883 |
|
|
{
|
11884 |
|
|
int mask = INTVAL (operands[2]);
|
11885 |
|
|
rtx perm[];
|
11886 |
|
|
|
11887 |
|
|
perm[0] = GEN_INT (mask & 1);
|
11888 |
|
|
perm[1] = GEN_INT ((mask >> 1) & 1);
|
11889 |
|
|
if (mode == V4DFmode)
|
11890 |
|
|
{
|
11891 |
|
|
perm[2] = GEN_INT (((mask >> 2) & 1) + 2);
|
11892 |
|
|
perm[3] = GEN_INT (((mask >> 3) & 1) + 2);
|
11893 |
|
|
}
|
11894 |
|
|
|
11895 |
|
|
operands[2]
|
11896 |
|
|
= gen_rtx_PARALLEL (VOIDmode, gen_rtvec_v (, perm));
|
11897 |
|
|
})
|
11898 |
|
|
|
11899 |
|
|
(define_expand "avx_vpermil"
|
11900 |
|
|
[(set (match_operand:AVXMODEFSP 0 "register_operand" "")
|
11901 |
|
|
(vec_select:AVXMODEFSP
|
11902 |
|
|
(match_operand:AVXMODEFSP 1 "nonimmediate_operand" "")
|
11903 |
|
|
(match_operand:SI 2 "const_0_to_255_operand" "")))]
|
11904 |
|
|
"TARGET_AVX"
|
11905 |
|
|
{
|
11906 |
|
|
int mask = INTVAL (operands[2]);
|
11907 |
|
|
rtx perm[];
|
11908 |
|
|
|
11909 |
|
|
perm[0] = GEN_INT (mask & 3);
|
11910 |
|
|
perm[1] = GEN_INT ((mask >> 2) & 3);
|
11911 |
|
|
perm[2] = GEN_INT ((mask >> 4) & 3);
|
11912 |
|
|
perm[3] = GEN_INT ((mask >> 6) & 3);
|
11913 |
|
|
if (mode == V8SFmode)
|
11914 |
|
|
{
|
11915 |
|
|
perm[4] = GEN_INT ((mask & 3) + 4);
|
11916 |
|
|
perm[5] = GEN_INT (((mask >> 2) & 3) + 4);
|
11917 |
|
|
perm[6] = GEN_INT (((mask >> 4) & 3) + 4);
|
11918 |
|
|
perm[7] = GEN_INT (((mask >> 6) & 3) + 4);
|
11919 |
|
|
}
|
11920 |
|
|
|
11921 |
|
|
operands[2]
|
11922 |
|
|
= gen_rtx_PARALLEL (VOIDmode, gen_rtvec_v (, perm));
|
11923 |
|
|
})
|
11924 |
|
|
|
11925 |
|
|
(define_insn "*avx_vpermilp"
|
11926 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
11927 |
|
|
(vec_select:AVXMODEF2P
|
11928 |
|
|
(match_operand:AVXMODEF2P 1 "nonimmediate_operand" "xm")
|
11929 |
|
|
(match_parallel 2 "avx_vpermilp__operand"
|
11930 |
|
|
[(match_operand 3 "const_int_operand" "")])))]
|
11931 |
|
|
"TARGET_AVX"
|
11932 |
|
|
{
|
11933 |
|
|
int mask = avx_vpermilp_parallel (operands[2], mode) - 1;
|
11934 |
|
|
operands[2] = GEN_INT (mask);
|
11935 |
|
|
return "vpermilp\t{%2, %1, %0|%0, %1, %2}";
|
11936 |
|
|
}
|
11937 |
|
|
[(set_attr "type" "sselog")
|
11938 |
|
|
(set_attr "prefix_extra" "1")
|
11939 |
|
|
(set_attr "length_immediate" "1")
|
11940 |
|
|
(set_attr "prefix" "vex")
|
11941 |
|
|
(set_attr "mode" "")])
|
11942 |
|
|
|
11943 |
|
|
(define_insn "avx_vpermilvar3"
|
11944 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
11945 |
|
|
(unspec:AVXMODEF2P
|
11946 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
11947 |
|
|
(match_operand: 2 "nonimmediate_operand" "xm")]
|
11948 |
|
|
UNSPEC_VPERMIL))]
|
11949 |
|
|
"TARGET_AVX"
|
11950 |
|
|
"vpermilp\t{%2, %1, %0|%0, %1, %2}"
|
11951 |
|
|
[(set_attr "type" "sselog")
|
11952 |
|
|
(set_attr "prefix_extra" "1")
|
11953 |
|
|
(set_attr "prefix" "vex")
|
11954 |
|
|
(set_attr "mode" "")])
|
11955 |
|
|
|
11956 |
|
|
(define_expand "avx_vperm2f1283"
|
11957 |
|
|
[(set (match_operand:AVX256MODE2P 0 "register_operand" "")
|
11958 |
|
|
(unspec:AVX256MODE2P
|
11959 |
|
|
[(match_operand:AVX256MODE2P 1 "register_operand" "")
|
11960 |
|
|
(match_operand:AVX256MODE2P 2 "nonimmediate_operand" "")
|
11961 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "")]
|
11962 |
|
|
UNSPEC_VPERMIL2F128))]
|
11963 |
|
|
"TARGET_AVX"
|
11964 |
|
|
{
|
11965 |
|
|
int mask = INTVAL (operands[3]);
|
11966 |
|
|
if ((mask & 0x88) == 0)
|
11967 |
|
|
{
|
11968 |
|
|
rtx perm[], t1, t2;
|
11969 |
|
|
int i, base, nelt = , nelt2 = nelt / 2;
|
11970 |
|
|
|
11971 |
|
|
base = (mask & 3) * nelt2;
|
11972 |
|
|
for (i = 0; i < nelt2; ++i)
|
11973 |
|
|
perm[i] = GEN_INT (base + i);
|
11974 |
|
|
|
11975 |
|
|
base = ((mask >> 4) & 3) * nelt2;
|
11976 |
|
|
for (i = 0; i < nelt2; ++i)
|
11977 |
|
|
perm[i + nelt2] = GEN_INT (base + i);
|
11978 |
|
|
|
11979 |
|
|
t2 = gen_rtx_VEC_CONCAT (mode,
|
11980 |
|
|
operands[1], operands[2]);
|
11981 |
|
|
t1 = gen_rtx_PARALLEL (VOIDmode, gen_rtvec_v (nelt, perm));
|
11982 |
|
|
t2 = gen_rtx_VEC_SELECT (mode, t2, t1);
|
11983 |
|
|
t2 = gen_rtx_SET (VOIDmode, operands[0], t2);
|
11984 |
|
|
emit_insn (t2);
|
11985 |
|
|
DONE;
|
11986 |
|
|
}
|
11987 |
|
|
})
|
11988 |
|
|
|
11989 |
|
|
;; Note that bits 7 and 3 of the imm8 allow lanes to be zeroed, which
|
11990 |
|
|
;; means that in order to represent this properly in rtl we'd have to
|
11991 |
|
|
;; nest *another* vec_concat with a zero operand and do the select from
|
11992 |
|
|
;; a 4x wide vector. That doesn't seem very nice.
|
11993 |
|
|
(define_insn "*avx_vperm2f128_full"
|
11994 |
|
|
[(set (match_operand:AVX256MODE2P 0 "register_operand" "=x")
|
11995 |
|
|
(unspec:AVX256MODE2P
|
11996 |
|
|
[(match_operand:AVX256MODE2P 1 "register_operand" "x")
|
11997 |
|
|
(match_operand:AVX256MODE2P 2 "nonimmediate_operand" "xm")
|
11998 |
|
|
(match_operand:SI 3 "const_0_to_255_operand" "n")]
|
11999 |
|
|
UNSPEC_VPERMIL2F128))]
|
12000 |
|
|
"TARGET_AVX"
|
12001 |
|
|
"vperm2f128\t{%3, %2, %1, %0|%0, %1, %2, %3}"
|
12002 |
|
|
[(set_attr "type" "sselog")
|
12003 |
|
|
(set_attr "prefix_extra" "1")
|
12004 |
|
|
(set_attr "length_immediate" "1")
|
12005 |
|
|
(set_attr "prefix" "vex")
|
12006 |
|
|
(set_attr "mode" "V8SF")])
|
12007 |
|
|
|
12008 |
|
|
(define_insn "*avx_vperm2f128_nozero"
|
12009 |
|
|
[(set (match_operand:AVX256MODE2P 0 "register_operand" "=x")
|
12010 |
|
|
(vec_select:AVX256MODE2P
|
12011 |
|
|
(vec_concat:
|
12012 |
|
|
(match_operand:AVX256MODE2P 1 "register_operand" "x")
|
12013 |
|
|
(match_operand:AVX256MODE2P 2 "nonimmediate_operand" "xm"))
|
12014 |
|
|
(match_parallel 3 "avx_vperm2f128__operand"
|
12015 |
|
|
[(match_operand 4 "const_int_operand" "")])))]
|
12016 |
|
|
"TARGET_AVX"
|
12017 |
|
|
{
|
12018 |
|
|
int mask = avx_vperm2f128_parallel (operands[3], mode) - 1;
|
12019 |
|
|
operands[3] = GEN_INT (mask);
|
12020 |
|
|
return "vperm2f128\t{%3, %2, %1, %0|%0, %1, %2, %3}";
|
12021 |
|
|
}
|
12022 |
|
|
[(set_attr "type" "sselog")
|
12023 |
|
|
(set_attr "prefix_extra" "1")
|
12024 |
|
|
(set_attr "length_immediate" "1")
|
12025 |
|
|
(set_attr "prefix" "vex")
|
12026 |
|
|
(set_attr "mode" "V8SF")])
|
12027 |
|
|
|
12028 |
|
|
(define_expand "avx_vinsertf128"
|
12029 |
|
|
[(match_operand:AVX256MODE 0 "register_operand" "")
|
12030 |
|
|
(match_operand:AVX256MODE 1 "register_operand" "")
|
12031 |
|
|
(match_operand: 2 "nonimmediate_operand" "")
|
12032 |
|
|
(match_operand:SI 3 "const_0_to_1_operand" "")]
|
12033 |
|
|
"TARGET_AVX"
|
12034 |
|
|
{
|
12035 |
|
|
switch (INTVAL (operands[3]))
|
12036 |
|
|
{
|
12037 |
|
|
case 0:
|
12038 |
|
|
emit_insn (gen_vec_set_lo_ (operands[0], operands[1],
|
12039 |
|
|
operands[2]));
|
12040 |
|
|
break;
|
12041 |
|
|
case 1:
|
12042 |
|
|
emit_insn (gen_vec_set_hi_ (operands[0], operands[1],
|
12043 |
|
|
operands[2]));
|
12044 |
|
|
break;
|
12045 |
|
|
default:
|
12046 |
|
|
gcc_unreachable ();
|
12047 |
|
|
}
|
12048 |
|
|
DONE;
|
12049 |
|
|
})
|
12050 |
|
|
|
12051 |
|
|
(define_insn "vec_set_lo_"
|
12052 |
|
|
[(set (match_operand:AVX256MODE4P 0 "register_operand" "=x")
|
12053 |
|
|
(vec_concat:AVX256MODE4P
|
12054 |
|
|
(match_operand: 2 "nonimmediate_operand" "xm")
|
12055 |
|
|
(vec_select:
|
12056 |
|
|
(match_operand:AVX256MODE4P 1 "register_operand" "x")
|
12057 |
|
|
(parallel [(const_int 2) (const_int 3)]))))]
|
12058 |
|
|
"TARGET_AVX"
|
12059 |
|
|
"vinsertf128\t{$0x0, %2, %1, %0|%0, %1, %2, 0x0}"
|
12060 |
|
|
[(set_attr "type" "sselog")
|
12061 |
|
|
(set_attr "prefix_extra" "1")
|
12062 |
|
|
(set_attr "length_immediate" "1")
|
12063 |
|
|
(set_attr "prefix" "vex")
|
12064 |
|
|
(set_attr "mode" "V8SF")])
|
12065 |
|
|
|
12066 |
|
|
(define_insn "vec_set_hi_"
|
12067 |
|
|
[(set (match_operand:AVX256MODE4P 0 "register_operand" "=x")
|
12068 |
|
|
(vec_concat:AVX256MODE4P
|
12069 |
|
|
(vec_select:
|
12070 |
|
|
(match_operand:AVX256MODE4P 1 "register_operand" "x")
|
12071 |
|
|
(parallel [(const_int 0) (const_int 1)]))
|
12072 |
|
|
(match_operand: 2 "nonimmediate_operand" "xm")))]
|
12073 |
|
|
"TARGET_AVX"
|
12074 |
|
|
"vinsertf128\t{$0x1, %2, %1, %0|%0, %1, %2, 0x1}"
|
12075 |
|
|
[(set_attr "type" "sselog")
|
12076 |
|
|
(set_attr "prefix_extra" "1")
|
12077 |
|
|
(set_attr "length_immediate" "1")
|
12078 |
|
|
(set_attr "prefix" "vex")
|
12079 |
|
|
(set_attr "mode" "V8SF")])
|
12080 |
|
|
|
12081 |
|
|
(define_insn "vec_set_lo_"
|
12082 |
|
|
[(set (match_operand:AVX256MODE8P 0 "register_operand" "=x")
|
12083 |
|
|
(vec_concat:AVX256MODE8P
|
12084 |
|
|
(match_operand: 2 "nonimmediate_operand" "xm")
|
12085 |
|
|
(vec_select:
|
12086 |
|
|
(match_operand:AVX256MODE8P 1 "register_operand" "x")
|
12087 |
|
|
(parallel [(const_int 4) (const_int 5)
|
12088 |
|
|
(const_int 6) (const_int 7)]))))]
|
12089 |
|
|
"TARGET_AVX"
|
12090 |
|
|
"vinsertf128\t{$0x0, %2, %1, %0|%0, %1, %2, 0x0}"
|
12091 |
|
|
[(set_attr "type" "sselog")
|
12092 |
|
|
(set_attr "prefix_extra" "1")
|
12093 |
|
|
(set_attr "length_immediate" "1")
|
12094 |
|
|
(set_attr "prefix" "vex")
|
12095 |
|
|
(set_attr "mode" "V8SF")])
|
12096 |
|
|
|
12097 |
|
|
(define_insn "vec_set_hi_"
|
12098 |
|
|
[(set (match_operand:AVX256MODE8P 0 "register_operand" "=x")
|
12099 |
|
|
(vec_concat:AVX256MODE8P
|
12100 |
|
|
(vec_select:
|
12101 |
|
|
(match_operand:AVX256MODE8P 1 "register_operand" "x")
|
12102 |
|
|
(parallel [(const_int 0) (const_int 1)
|
12103 |
|
|
(const_int 2) (const_int 3)]))
|
12104 |
|
|
(match_operand: 2 "nonimmediate_operand" "xm")))]
|
12105 |
|
|
"TARGET_AVX"
|
12106 |
|
|
"vinsertf128\t{$0x1, %2, %1, %0|%0, %1, %2, 0x1}"
|
12107 |
|
|
[(set_attr "type" "sselog")
|
12108 |
|
|
(set_attr "prefix_extra" "1")
|
12109 |
|
|
(set_attr "length_immediate" "1")
|
12110 |
|
|
(set_attr "prefix" "vex")
|
12111 |
|
|
(set_attr "mode" "V8SF")])
|
12112 |
|
|
|
12113 |
|
|
(define_insn "vec_set_lo_v16hi"
|
12114 |
|
|
[(set (match_operand:V16HI 0 "register_operand" "=x")
|
12115 |
|
|
(vec_concat:V16HI
|
12116 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")
|
12117 |
|
|
(vec_select:V8HI
|
12118 |
|
|
(match_operand:V16HI 1 "register_operand" "x")
|
12119 |
|
|
(parallel [(const_int 8) (const_int 9)
|
12120 |
|
|
(const_int 10) (const_int 11)
|
12121 |
|
|
(const_int 12) (const_int 13)
|
12122 |
|
|
(const_int 14) (const_int 15)]))))]
|
12123 |
|
|
"TARGET_AVX"
|
12124 |
|
|
"vinsertf128\t{$0x0, %2, %1, %0|%0, %1, %2, 0x0}"
|
12125 |
|
|
[(set_attr "type" "sselog")
|
12126 |
|
|
(set_attr "prefix_extra" "1")
|
12127 |
|
|
(set_attr "length_immediate" "1")
|
12128 |
|
|
(set_attr "prefix" "vex")
|
12129 |
|
|
(set_attr "mode" "V8SF")])
|
12130 |
|
|
|
12131 |
|
|
(define_insn "vec_set_hi_v16hi"
|
12132 |
|
|
[(set (match_operand:V16HI 0 "register_operand" "=x")
|
12133 |
|
|
(vec_concat:V16HI
|
12134 |
|
|
(vec_select:V8HI
|
12135 |
|
|
(match_operand:V16HI 1 "register_operand" "x")
|
12136 |
|
|
(parallel [(const_int 0) (const_int 1)
|
12137 |
|
|
(const_int 2) (const_int 3)
|
12138 |
|
|
(const_int 4) (const_int 5)
|
12139 |
|
|
(const_int 6) (const_int 7)]))
|
12140 |
|
|
(match_operand:V8HI 2 "nonimmediate_operand" "xm")))]
|
12141 |
|
|
"TARGET_AVX"
|
12142 |
|
|
"vinsertf128\t{$0x1, %2, %1, %0|%0, %1, %2, 0x1}"
|
12143 |
|
|
[(set_attr "type" "sselog")
|
12144 |
|
|
(set_attr "prefix_extra" "1")
|
12145 |
|
|
(set_attr "length_immediate" "1")
|
12146 |
|
|
(set_attr "prefix" "vex")
|
12147 |
|
|
(set_attr "mode" "V8SF")])
|
12148 |
|
|
|
12149 |
|
|
(define_insn "vec_set_lo_v32qi"
|
12150 |
|
|
[(set (match_operand:V32QI 0 "register_operand" "=x")
|
12151 |
|
|
(vec_concat:V32QI
|
12152 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")
|
12153 |
|
|
(vec_select:V16QI
|
12154 |
|
|
(match_operand:V32QI 1 "register_operand" "x")
|
12155 |
|
|
(parallel [(const_int 16) (const_int 17)
|
12156 |
|
|
(const_int 18) (const_int 19)
|
12157 |
|
|
(const_int 20) (const_int 21)
|
12158 |
|
|
(const_int 22) (const_int 23)
|
12159 |
|
|
(const_int 24) (const_int 25)
|
12160 |
|
|
(const_int 26) (const_int 27)
|
12161 |
|
|
(const_int 28) (const_int 29)
|
12162 |
|
|
(const_int 30) (const_int 31)]))))]
|
12163 |
|
|
"TARGET_AVX"
|
12164 |
|
|
"vinsertf128\t{$0x0, %2, %1, %0|%0, %1, %2, 0x0}"
|
12165 |
|
|
[(set_attr "type" "sselog")
|
12166 |
|
|
(set_attr "prefix_extra" "1")
|
12167 |
|
|
(set_attr "length_immediate" "1")
|
12168 |
|
|
(set_attr "prefix" "vex")
|
12169 |
|
|
(set_attr "mode" "V8SF")])
|
12170 |
|
|
|
12171 |
|
|
(define_insn "vec_set_hi_v32qi"
|
12172 |
|
|
[(set (match_operand:V32QI 0 "register_operand" "=x")
|
12173 |
|
|
(vec_concat:V32QI
|
12174 |
|
|
(vec_select:V16QI
|
12175 |
|
|
(match_operand:V32QI 1 "register_operand" "x")
|
12176 |
|
|
(parallel [(const_int 0) (const_int 1)
|
12177 |
|
|
(const_int 2) (const_int 3)
|
12178 |
|
|
(const_int 4) (const_int 5)
|
12179 |
|
|
(const_int 6) (const_int 7)
|
12180 |
|
|
(const_int 8) (const_int 9)
|
12181 |
|
|
(const_int 10) (const_int 11)
|
12182 |
|
|
(const_int 12) (const_int 13)
|
12183 |
|
|
(const_int 14) (const_int 15)]))
|
12184 |
|
|
(match_operand:V16QI 2 "nonimmediate_operand" "xm")))]
|
12185 |
|
|
"TARGET_AVX"
|
12186 |
|
|
"vinsertf128\t{$0x1, %2, %1, %0|%0, %1, %2, 0x1}"
|
12187 |
|
|
[(set_attr "type" "sselog")
|
12188 |
|
|
(set_attr "prefix_extra" "1")
|
12189 |
|
|
(set_attr "length_immediate" "1")
|
12190 |
|
|
(set_attr "prefix" "vex")
|
12191 |
|
|
(set_attr "mode" "V8SF")])
|
12192 |
|
|
|
12193 |
|
|
(define_insn "avx_maskloadp"
|
12194 |
|
|
[(set (match_operand:AVXMODEF2P 0 "register_operand" "=x")
|
12195 |
|
|
(unspec:AVXMODEF2P
|
12196 |
|
|
[(match_operand:AVXMODEF2P 1 "memory_operand" "m")
|
12197 |
|
|
(match_operand:AVXMODEF2P 2 "register_operand" "x")
|
12198 |
|
|
(match_dup 0)]
|
12199 |
|
|
UNSPEC_MASKLOAD))]
|
12200 |
|
|
"TARGET_AVX"
|
12201 |
|
|
"vmaskmovp\t{%1, %2, %0|%0, %2, %1}"
|
12202 |
|
|
[(set_attr "type" "sselog1")
|
12203 |
|
|
(set_attr "prefix_extra" "1")
|
12204 |
|
|
(set_attr "prefix" "vex")
|
12205 |
|
|
(set_attr "mode" "")])
|
12206 |
|
|
|
12207 |
|
|
(define_insn "avx_maskstorep"
|
12208 |
|
|
[(set (match_operand:AVXMODEF2P 0 "memory_operand" "=m")
|
12209 |
|
|
(unspec:AVXMODEF2P
|
12210 |
|
|
[(match_operand:AVXMODEF2P 1 "register_operand" "x")
|
12211 |
|
|
(match_operand:AVXMODEF2P 2 "register_operand" "x")
|
12212 |
|
|
(match_dup 0)]
|
12213 |
|
|
UNSPEC_MASKSTORE))]
|
12214 |
|
|
"TARGET_AVX"
|
12215 |
|
|
"vmaskmovp\t{%2, %1, %0|%0, %1, %2}"
|
12216 |
|
|
[(set_attr "type" "sselog1")
|
12217 |
|
|
(set_attr "prefix_extra" "1")
|
12218 |
|
|
(set_attr "prefix" "vex")
|
12219 |
|
|
(set_attr "mode" "")])
|
12220 |
|
|
|
12221 |
|
|
(define_insn "avx__"
|
12222 |
|
|
[(set (match_operand:AVX256MODE2P 0 "register_operand" "=x,x")
|
12223 |
|
|
(unspec:AVX256MODE2P
|
12224 |
|
|
[(match_operand: 1 "nonimmediate_operand" "0,xm")]
|
12225 |
|
|
UNSPEC_CAST))]
|
12226 |
|
|
"TARGET_AVX"
|
12227 |
|
|
{
|
12228 |
|
|
switch (which_alternative)
|
12229 |
|
|
{
|
12230 |
|
|
case 0:
|
12231 |
|
|
return "";
|
12232 |
|
|
case 1:
|
12233 |
|
|
switch (get_attr_mode (insn))
|
12234 |
|
|
{
|
12235 |
|
|
case MODE_V8SF:
|
12236 |
|
|
return "vmovaps\t{%1, %x0|%x0, %1}";
|
12237 |
|
|
case MODE_V4DF:
|
12238 |
|
|
return "vmovapd\t{%1, %x0|%x0, %1}";
|
12239 |
|
|
case MODE_OI:
|
12240 |
|
|
return "vmovdqa\t{%1, %x0|%x0, %1}";
|
12241 |
|
|
default:
|
12242 |
|
|
break;
|
12243 |
|
|
}
|
12244 |
|
|
default:
|
12245 |
|
|
break;
|
12246 |
|
|
}
|
12247 |
|
|
gcc_unreachable ();
|
12248 |
|
|
}
|
12249 |
|
|
[(set_attr "type" "ssemov")
|
12250 |
|
|
(set_attr "prefix" "vex")
|
12251 |
|
|
(set_attr "mode" "")
|
12252 |
|
|
(set (attr "length")
|
12253 |
|
|
(if_then_else (eq_attr "alternative" "0")
|
12254 |
|
|
(const_string "0")
|
12255 |
|
|
(const_string "*")))])
|
12256 |
|
|
|
12257 |
|
|
(define_insn "avx__"
|
12258 |
|
|
[(set (match_operand: 0 "register_operand" "=x,x")
|
12259 |
|
|
(unspec:
|
12260 |
|
|
[(match_operand:AVX256MODE2P 1 "nonimmediate_operand" "0,xm")]
|
12261 |
|
|
UNSPEC_CAST))]
|
12262 |
|
|
"TARGET_AVX"
|
12263 |
|
|
{
|
12264 |
|
|
switch (which_alternative)
|
12265 |
|
|
{
|
12266 |
|
|
case 0:
|
12267 |
|
|
return "";
|
12268 |
|
|
case 1:
|
12269 |
|
|
switch (get_attr_mode (insn))
|
12270 |
|
|
{
|
12271 |
|
|
case MODE_V8SF:
|
12272 |
|
|
return "vmovaps\t{%x1, %0|%0, %x1}";
|
12273 |
|
|
case MODE_V4DF:
|
12274 |
|
|
return "vmovapd\t{%x1, %0|%0, %x1}";
|
12275 |
|
|
case MODE_OI:
|
12276 |
|
|
return "vmovdqa\t{%x1, %0|%0, %x1}";
|
12277 |
|
|
default:
|
12278 |
|
|
break;
|
12279 |
|
|
}
|
12280 |
|
|
default:
|
12281 |
|
|
break;
|
12282 |
|
|
}
|
12283 |
|
|
gcc_unreachable ();
|
12284 |
|
|
}
|
12285 |
|
|
[(set_attr "type" "ssemov")
|
12286 |
|
|
(set_attr "prefix" "vex")
|
12287 |
|
|
(set_attr "mode" "")
|
12288 |
|
|
(set (attr "length")
|
12289 |
|
|
(if_then_else (eq_attr "alternative" "0")
|
12290 |
|
|
(const_string "0")
|
12291 |
|
|
(const_string "*")))])
|
12292 |
|
|
|
12293 |
|
|
(define_expand "vec_init"
|
12294 |
|
|
[(match_operand:AVX256MODE 0 "register_operand" "")
|
12295 |
|
|
(match_operand 1 "" "")]
|
12296 |
|
|
"TARGET_AVX"
|
12297 |
|
|
{
|
12298 |
|
|
ix86_expand_vector_init (false, operands[0], operands[1]);
|
12299 |
|
|
DONE;
|
12300 |
|
|
})
|
12301 |
|
|
|
12302 |
|
|
(define_insn "*vec_concat_avx"
|
12303 |
|
|
[(set (match_operand:AVX256MODE 0 "register_operand" "=x,x")
|
12304 |
|
|
(vec_concat:AVX256MODE
|
12305 |
|
|
(match_operand: 1 "register_operand" "x,x")
|
12306 |
|
|
(match_operand: 2 "vector_move_operand" "xm,C")))]
|
12307 |
|
|
"TARGET_AVX"
|
12308 |
|
|
{
|
12309 |
|
|
switch (which_alternative)
|
12310 |
|
|
{
|
12311 |
|
|
case 0:
|
12312 |
|
|
return "vinsertf128\t{$0x1, %2, %t1, %0|%0, %t1, %2, 0x1}";
|
12313 |
|
|
case 1:
|
12314 |
|
|
switch (get_attr_mode (insn))
|
12315 |
|
|
{
|
12316 |
|
|
case MODE_V8SF:
|
12317 |
|
|
return "vmovaps\t{%1, %x0|%x0, %1}";
|
12318 |
|
|
case MODE_V4DF:
|
12319 |
|
|
return "vmovapd\t{%1, %x0|%x0, %1}";
|
12320 |
|
|
default:
|
12321 |
|
|
return "vmovdqa\t{%1, %x0|%x0, %1}";
|
12322 |
|
|
}
|
12323 |
|
|
default:
|
12324 |
|
|
gcc_unreachable ();
|
12325 |
|
|
}
|
12326 |
|
|
}
|
12327 |
|
|
[(set_attr "type" "sselog,ssemov")
|
12328 |
|
|
(set_attr "prefix_extra" "1,*")
|
12329 |
|
|
(set_attr "length_immediate" "1,*")
|
12330 |
|
|
(set_attr "prefix" "vex")
|
12331 |
|
|
(set_attr "mode" "")])
|