1 |
282 |
jeremybenn |
;; VSX patterns.
|
2 |
|
|
;; Copyright (C) 2009
|
3 |
|
|
;; Free Software Foundation, Inc.
|
4 |
|
|
;; Contributed by Michael Meissner
|
5 |
|
|
|
6 |
|
|
;; This file is part of GCC.
|
7 |
|
|
|
8 |
|
|
;; GCC is free software; you can redistribute it and/or modify it
|
9 |
|
|
;; under the terms of the GNU General Public License as published
|
10 |
|
|
;; by the Free Software Foundation; either version 3, or (at your
|
11 |
|
|
;; option) any later version.
|
12 |
|
|
|
13 |
|
|
;; GCC is distributed in the hope that it will be useful, but WITHOUT
|
14 |
|
|
;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
15 |
|
|
;; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
|
16 |
|
|
;; License for more details.
|
17 |
|
|
|
18 |
|
|
;; You should have received a copy of the GNU General Public License
|
19 |
|
|
;; along with GCC; see the file COPYING3. If not see
|
20 |
|
|
;; .
|
21 |
|
|
|
22 |
|
|
;; Iterator for both scalar and vector floating point types supported by VSX
|
23 |
|
|
(define_mode_iterator VSX_B [DF V4SF V2DF])
|
24 |
|
|
|
25 |
|
|
;; Iterator for the 2 64-bit vector types
|
26 |
|
|
(define_mode_iterator VSX_D [V2DF V2DI])
|
27 |
|
|
|
28 |
|
|
;; Iterator for the 2 32-bit vector types
|
29 |
|
|
(define_mode_iterator VSX_W [V4SF V4SI])
|
30 |
|
|
|
31 |
|
|
;; Iterator for vector floating point types supported by VSX
|
32 |
|
|
(define_mode_iterator VSX_F [V4SF V2DF])
|
33 |
|
|
|
34 |
|
|
;; Iterator for logical types supported by VSX
|
35 |
|
|
(define_mode_iterator VSX_L [V16QI V8HI V4SI V2DI V4SF V2DF TI])
|
36 |
|
|
|
37 |
|
|
;; Iterator for memory move. Handle TImode specially to allow
|
38 |
|
|
;; it to use gprs as well as vsx registers.
|
39 |
|
|
(define_mode_iterator VSX_M [V16QI V8HI V4SI V2DI V4SF V2DF])
|
40 |
|
|
|
41 |
|
|
;; Map into the appropriate load/store name based on the type
|
42 |
|
|
(define_mode_attr VSm [(V16QI "vw4")
|
43 |
|
|
(V8HI "vw4")
|
44 |
|
|
(V4SI "vw4")
|
45 |
|
|
(V4SF "vw4")
|
46 |
|
|
(V2DF "vd2")
|
47 |
|
|
(V2DI "vd2")
|
48 |
|
|
(DF "d")
|
49 |
|
|
(TI "vw4")])
|
50 |
|
|
|
51 |
|
|
;; Map into the appropriate suffix based on the type
|
52 |
|
|
(define_mode_attr VSs [(V16QI "sp")
|
53 |
|
|
(V8HI "sp")
|
54 |
|
|
(V4SI "sp")
|
55 |
|
|
(V4SF "sp")
|
56 |
|
|
(V2DF "dp")
|
57 |
|
|
(V2DI "dp")
|
58 |
|
|
(DF "dp")
|
59 |
|
|
(SF "sp")
|
60 |
|
|
(TI "sp")])
|
61 |
|
|
|
62 |
|
|
;; Map the register class used
|
63 |
|
|
(define_mode_attr VSr [(V16QI "v")
|
64 |
|
|
(V8HI "v")
|
65 |
|
|
(V4SI "v")
|
66 |
|
|
(V4SF "wf")
|
67 |
|
|
(V2DI "wd")
|
68 |
|
|
(V2DF "wd")
|
69 |
|
|
(DF "ws")
|
70 |
|
|
(SF "d")
|
71 |
|
|
(TI "wd")])
|
72 |
|
|
|
73 |
|
|
;; Map the register class used for float<->int conversions
|
74 |
|
|
(define_mode_attr VSr2 [(V2DF "wd")
|
75 |
|
|
(V4SF "wf")
|
76 |
|
|
(DF "!f#r")])
|
77 |
|
|
|
78 |
|
|
(define_mode_attr VSr3 [(V2DF "wa")
|
79 |
|
|
(V4SF "wa")
|
80 |
|
|
(DF "!f#r")])
|
81 |
|
|
|
82 |
|
|
;; Map the register class for sp<->dp float conversions, destination
|
83 |
|
|
(define_mode_attr VSr4 [(SF "ws")
|
84 |
|
|
(DF "f")
|
85 |
|
|
(V2DF "wd")
|
86 |
|
|
(V4SF "v")])
|
87 |
|
|
|
88 |
|
|
;; Map the register class for sp<->dp float conversions, destination
|
89 |
|
|
(define_mode_attr VSr5 [(SF "ws")
|
90 |
|
|
(DF "f")
|
91 |
|
|
(V2DF "v")
|
92 |
|
|
(V4SF "wd")])
|
93 |
|
|
|
94 |
|
|
;; Same size integer type for floating point data
|
95 |
|
|
(define_mode_attr VSi [(V4SF "v4si")
|
96 |
|
|
(V2DF "v2di")
|
97 |
|
|
(DF "di")])
|
98 |
|
|
|
99 |
|
|
(define_mode_attr VSI [(V4SF "V4SI")
|
100 |
|
|
(V2DF "V2DI")
|
101 |
|
|
(DF "DI")])
|
102 |
|
|
|
103 |
|
|
;; Word size for same size conversion
|
104 |
|
|
(define_mode_attr VSc [(V4SF "w")
|
105 |
|
|
(V2DF "d")
|
106 |
|
|
(DF "d")])
|
107 |
|
|
|
108 |
|
|
;; Map into either s or v, depending on whether this is a scalar or vector
|
109 |
|
|
;; operation
|
110 |
|
|
(define_mode_attr VSv [(V16QI "v")
|
111 |
|
|
(V8HI "v")
|
112 |
|
|
(V4SI "v")
|
113 |
|
|
(V4SF "v")
|
114 |
|
|
(V2DI "v")
|
115 |
|
|
(V2DF "v")
|
116 |
|
|
(TI "v")
|
117 |
|
|
(DF "s")])
|
118 |
|
|
|
119 |
|
|
;; Appropriate type for add ops (and other simple FP ops)
|
120 |
|
|
(define_mode_attr VStype_simple [(V2DF "vecfloat")
|
121 |
|
|
(V4SF "vecfloat")
|
122 |
|
|
(DF "fp")])
|
123 |
|
|
|
124 |
|
|
(define_mode_attr VSfptype_simple [(V2DF "fp_addsub_d")
|
125 |
|
|
(V4SF "fp_addsub_s")
|
126 |
|
|
(DF "fp_addsub_d")])
|
127 |
|
|
|
128 |
|
|
;; Appropriate type for multiply ops
|
129 |
|
|
(define_mode_attr VStype_mul [(V2DF "vecfloat")
|
130 |
|
|
(V4SF "vecfloat")
|
131 |
|
|
(DF "dmul")])
|
132 |
|
|
|
133 |
|
|
(define_mode_attr VSfptype_mul [(V2DF "fp_mul_d")
|
134 |
|
|
(V4SF "fp_mul_s")
|
135 |
|
|
(DF "fp_mul_d")])
|
136 |
|
|
|
137 |
|
|
;; Appropriate type for divide ops. For now, just lump the vector divide with
|
138 |
|
|
;; the scalar divides
|
139 |
|
|
(define_mode_attr VStype_div [(V2DF "ddiv")
|
140 |
|
|
(V4SF "sdiv")
|
141 |
|
|
(DF "ddiv")])
|
142 |
|
|
|
143 |
|
|
(define_mode_attr VSfptype_div [(V2DF "fp_div_d")
|
144 |
|
|
(V4SF "fp_div_s")
|
145 |
|
|
(DF "fp_div_d")])
|
146 |
|
|
|
147 |
|
|
;; Appropriate type for sqrt ops. For now, just lump the vector sqrt with
|
148 |
|
|
;; the scalar sqrt
|
149 |
|
|
(define_mode_attr VStype_sqrt [(V2DF "dsqrt")
|
150 |
|
|
(V4SF "sdiv")
|
151 |
|
|
(DF "ddiv")])
|
152 |
|
|
|
153 |
|
|
(define_mode_attr VSfptype_sqrt [(V2DF "fp_sqrt_d")
|
154 |
|
|
(V4SF "fp_sqrt_s")
|
155 |
|
|
(DF "fp_sqrt_d")])
|
156 |
|
|
|
157 |
|
|
;; Iterator and modes for sp<->dp conversions
|
158 |
|
|
;; Because scalar SF values are represented internally as double, use the
|
159 |
|
|
;; V4SF type to represent this than SF.
|
160 |
|
|
(define_mode_iterator VSX_SPDP [DF V4SF V2DF])
|
161 |
|
|
|
162 |
|
|
(define_mode_attr VS_spdp_res [(DF "V4SF")
|
163 |
|
|
(V4SF "V2DF")
|
164 |
|
|
(V2DF "V4SF")])
|
165 |
|
|
|
166 |
|
|
(define_mode_attr VS_spdp_insn [(DF "xscvdpsp")
|
167 |
|
|
(V4SF "xvcvspdp")
|
168 |
|
|
(V2DF "xvcvdpsp")])
|
169 |
|
|
|
170 |
|
|
(define_mode_attr VS_spdp_type [(DF "fp")
|
171 |
|
|
(V4SF "vecfloat")
|
172 |
|
|
(V2DF "vecfloat")])
|
173 |
|
|
|
174 |
|
|
;; Map the scalar mode for a vector type
|
175 |
|
|
(define_mode_attr VS_scalar [(V2DF "DF")
|
176 |
|
|
(V2DI "DI")
|
177 |
|
|
(V4SF "SF")
|
178 |
|
|
(V4SI "SI")
|
179 |
|
|
(V8HI "HI")
|
180 |
|
|
(V16QI "QI")])
|
181 |
|
|
|
182 |
|
|
;; Constants for creating unspecs
|
183 |
|
|
(define_constants
|
184 |
|
|
[(UNSPEC_VSX_CONCAT 500)
|
185 |
|
|
(UNSPEC_VSX_CVDPSXWS 501)
|
186 |
|
|
(UNSPEC_VSX_CVDPUXWS 502)
|
187 |
|
|
(UNSPEC_VSX_CVSPDP 503)
|
188 |
|
|
(UNSPEC_VSX_CVSXWDP 504)
|
189 |
|
|
(UNSPEC_VSX_CVUXWDP 505)
|
190 |
|
|
(UNSPEC_VSX_CVSXDSP 506)
|
191 |
|
|
(UNSPEC_VSX_CVUXDSP 507)
|
192 |
|
|
(UNSPEC_VSX_CVSPSXDS 508)
|
193 |
|
|
(UNSPEC_VSX_CVSPUXDS 509)
|
194 |
|
|
(UNSPEC_VSX_MADD 510)
|
195 |
|
|
(UNSPEC_VSX_MSUB 511)
|
196 |
|
|
(UNSPEC_VSX_NMADD 512)
|
197 |
|
|
(UNSPEC_VSX_NMSUB 513)
|
198 |
|
|
(UNSPEC_VSX_RSQRTE 514)
|
199 |
|
|
(UNSPEC_VSX_TDIV 515)
|
200 |
|
|
(UNSPEC_VSX_TSQRT 516)
|
201 |
|
|
(UNSPEC_VSX_XXPERMDI 517)
|
202 |
|
|
(UNSPEC_VSX_SET 518)
|
203 |
|
|
(UNSPEC_VSX_ROUND_I 519)
|
204 |
|
|
(UNSPEC_VSX_ROUND_IC 520)
|
205 |
|
|
(UNSPEC_VSX_SLDWI 521)])
|
206 |
|
|
|
207 |
|
|
;; VSX moves
|
208 |
|
|
(define_insn "*vsx_mov"
|
209 |
|
|
[(set (match_operand:VSX_M 0 "nonimmediate_operand" "=Z,,,?Z,?wa,?wa,*o,*r,*r,,?wa,v,wZ,v")
|
210 |
|
|
(match_operand:VSX_M 1 "input_operand" ",Z,,wa,Z,wa,r,o,r,j,j,W,v,wZ"))]
|
211 |
|
|
"VECTOR_MEM_VSX_P (mode)
|
212 |
|
|
&& (register_operand (operands[0], mode)
|
213 |
|
|
|| register_operand (operands[1], mode))"
|
214 |
|
|
{
|
215 |
|
|
switch (which_alternative)
|
216 |
|
|
{
|
217 |
|
|
case 0:
|
218 |
|
|
case 3:
|
219 |
|
|
gcc_assert (MEM_P (operands[0])
|
220 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_INC
|
221 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_DEC
|
222 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_MODIFY);
|
223 |
|
|
return "stxx %x1,%y0";
|
224 |
|
|
|
225 |
|
|
case 1:
|
226 |
|
|
case 4:
|
227 |
|
|
gcc_assert (MEM_P (operands[1])
|
228 |
|
|
&& GET_CODE (XEXP (operands[1], 0)) != PRE_INC
|
229 |
|
|
&& GET_CODE (XEXP (operands[1], 0)) != PRE_DEC
|
230 |
|
|
&& GET_CODE (XEXP (operands[1], 0)) != PRE_MODIFY);
|
231 |
|
|
return "lxx %x0,%y1";
|
232 |
|
|
|
233 |
|
|
case 2:
|
234 |
|
|
case 5:
|
235 |
|
|
return "xxlor %x0,%x1,%x1";
|
236 |
|
|
|
237 |
|
|
case 6:
|
238 |
|
|
case 7:
|
239 |
|
|
case 8:
|
240 |
|
|
return "#";
|
241 |
|
|
|
242 |
|
|
case 9:
|
243 |
|
|
case 10:
|
244 |
|
|
return "xxlxor %x0,%x0,%x0";
|
245 |
|
|
|
246 |
|
|
case 11:
|
247 |
|
|
return output_vec_const_move (operands);
|
248 |
|
|
|
249 |
|
|
case 12:
|
250 |
|
|
gcc_assert (MEM_P (operands[0])
|
251 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_INC
|
252 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_DEC
|
253 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_MODIFY);
|
254 |
|
|
return "stvx %1,%y0";
|
255 |
|
|
|
256 |
|
|
case 13:
|
257 |
|
|
gcc_assert (MEM_P (operands[0])
|
258 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_INC
|
259 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_DEC
|
260 |
|
|
&& GET_CODE (XEXP (operands[0], 0)) != PRE_MODIFY);
|
261 |
|
|
return "lvx %0,%y1";
|
262 |
|
|
|
263 |
|
|
default:
|
264 |
|
|
gcc_unreachable ();
|
265 |
|
|
}
|
266 |
|
|
}
|
267 |
|
|
[(set_attr "type" "vecstore,vecload,vecsimple,vecstore,vecload,vecsimple,*,*,*,vecsimple,vecsimple,*,vecstore,vecload")])
|
268 |
|
|
|
269 |
|
|
;; Unlike other VSX moves, allow the GPRs, since a normal use of TImode is for
|
270 |
|
|
;; unions. However for plain data movement, slightly favor the vector loads
|
271 |
|
|
(define_insn "*vsx_movti"
|
272 |
|
|
[(set (match_operand:TI 0 "nonimmediate_operand" "=Z,wa,wa,?o,?r,?r,wa,v,v,wZ")
|
273 |
|
|
(match_operand:TI 1 "input_operand" "wa,Z,wa,r,o,r,j,W,wZ,v"))]
|
274 |
|
|
"VECTOR_MEM_VSX_P (TImode)
|
275 |
|
|
&& (register_operand (operands[0], TImode)
|
276 |
|
|
|| register_operand (operands[1], TImode))"
|
277 |
|
|
{
|
278 |
|
|
switch (which_alternative)
|
279 |
|
|
{
|
280 |
|
|
case 0:
|
281 |
|
|
return "stxvd2x %x1,%y0";
|
282 |
|
|
|
283 |
|
|
case 1:
|
284 |
|
|
return "lxvd2x %x0,%y1";
|
285 |
|
|
|
286 |
|
|
case 2:
|
287 |
|
|
return "xxlor %x0,%x1,%x1";
|
288 |
|
|
|
289 |
|
|
case 3:
|
290 |
|
|
case 4:
|
291 |
|
|
case 5:
|
292 |
|
|
return "#";
|
293 |
|
|
|
294 |
|
|
case 6:
|
295 |
|
|
return "xxlxor %x0,%x0,%x0";
|
296 |
|
|
|
297 |
|
|
case 7:
|
298 |
|
|
return output_vec_const_move (operands);
|
299 |
|
|
|
300 |
|
|
case 8:
|
301 |
|
|
return "stvx %1,%y0";
|
302 |
|
|
|
303 |
|
|
case 9:
|
304 |
|
|
return "lvx %0,%y1";
|
305 |
|
|
|
306 |
|
|
default:
|
307 |
|
|
gcc_unreachable ();
|
308 |
|
|
}
|
309 |
|
|
}
|
310 |
|
|
[(set_attr "type" "vecstore,vecload,vecsimple,*,*,*,vecsimple,*,vecstore,vecload")])
|
311 |
|
|
|
312 |
|
|
|
313 |
|
|
;; VSX scalar and vector floating point arithmetic instructions
|
314 |
|
|
(define_insn "*vsx_add3"
|
315 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
316 |
|
|
(plus:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")
|
317 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",wa")))]
|
318 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
319 |
|
|
"xadd %x0,%x1,%x2"
|
320 |
|
|
[(set_attr "type" "")
|
321 |
|
|
(set_attr "fp_type" "")])
|
322 |
|
|
|
323 |
|
|
(define_insn "*vsx_sub3"
|
324 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
325 |
|
|
(minus:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")
|
326 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",wa")))]
|
327 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
328 |
|
|
"xsub %x0,%x1,%x2"
|
329 |
|
|
[(set_attr "type" "")
|
330 |
|
|
(set_attr "fp_type" "")])
|
331 |
|
|
|
332 |
|
|
(define_insn "*vsx_mul3"
|
333 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
334 |
|
|
(mult:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")
|
335 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",wa")))]
|
336 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
337 |
|
|
"xmul %x0,%x1,%x2"
|
338 |
|
|
[(set_attr "type" "")
|
339 |
|
|
(set_attr "fp_type" "")])
|
340 |
|
|
|
341 |
|
|
(define_insn "*vsx_div3"
|
342 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
343 |
|
|
(div:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")
|
344 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",wa")))]
|
345 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
346 |
|
|
"xdiv %x0,%x1,%x2"
|
347 |
|
|
[(set_attr "type" "")
|
348 |
|
|
(set_attr "fp_type" "")])
|
349 |
|
|
|
350 |
|
|
;; *tdiv* instruction returning the FG flag
|
351 |
|
|
(define_expand "vsx_tdiv3_fg"
|
352 |
|
|
[(set (match_dup 3)
|
353 |
|
|
(unspec:CCFP [(match_operand:VSX_B 1 "vsx_register_operand" "")
|
354 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" "")]
|
355 |
|
|
UNSPEC_VSX_TDIV))
|
356 |
|
|
(set (match_operand:SI 0 "gpc_reg_operand" "")
|
357 |
|
|
(gt:SI (match_dup 3)
|
358 |
|
|
(const_int 0)))]
|
359 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
360 |
|
|
{
|
361 |
|
|
operands[3] = gen_reg_rtx (CCFPmode);
|
362 |
|
|
})
|
363 |
|
|
|
364 |
|
|
;; *tdiv* instruction returning the FE flag
|
365 |
|
|
(define_expand "vsx_tdiv3_fe"
|
366 |
|
|
[(set (match_dup 3)
|
367 |
|
|
(unspec:CCFP [(match_operand:VSX_B 1 "vsx_register_operand" "")
|
368 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" "")]
|
369 |
|
|
UNSPEC_VSX_TDIV))
|
370 |
|
|
(set (match_operand:SI 0 "gpc_reg_operand" "")
|
371 |
|
|
(eq:SI (match_dup 3)
|
372 |
|
|
(const_int 0)))]
|
373 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
374 |
|
|
{
|
375 |
|
|
operands[3] = gen_reg_rtx (CCFPmode);
|
376 |
|
|
})
|
377 |
|
|
|
378 |
|
|
(define_insn "*vsx_tdiv3_internal"
|
379 |
|
|
[(set (match_operand:CCFP 0 "cc_reg_operand" "=x,x")
|
380 |
|
|
(unspec:CCFP [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")
|
381 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",wa")]
|
382 |
|
|
UNSPEC_VSX_TDIV))]
|
383 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
384 |
|
|
"xtdiv %0,%x1,%x2"
|
385 |
|
|
[(set_attr "type" "")
|
386 |
|
|
(set_attr "fp_type" "")])
|
387 |
|
|
|
388 |
|
|
(define_insn "vsx_fre2"
|
389 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
390 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")]
|
391 |
|
|
UNSPEC_FRES))]
|
392 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
393 |
|
|
"xre %x0,%x1"
|
394 |
|
|
[(set_attr "type" "")
|
395 |
|
|
(set_attr "fp_type" "")])
|
396 |
|
|
|
397 |
|
|
(define_insn "*vsx_neg2"
|
398 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
399 |
|
|
(neg:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")))]
|
400 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
401 |
|
|
"xneg %x0,%x1"
|
402 |
|
|
[(set_attr "type" "")
|
403 |
|
|
(set_attr "fp_type" "")])
|
404 |
|
|
|
405 |
|
|
(define_insn "*vsx_abs2"
|
406 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
407 |
|
|
(abs:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")))]
|
408 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
409 |
|
|
"xabs %x0,%x1"
|
410 |
|
|
[(set_attr "type" "")
|
411 |
|
|
(set_attr "fp_type" "")])
|
412 |
|
|
|
413 |
|
|
(define_insn "vsx_nabs2"
|
414 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
415 |
|
|
(neg:VSX_B
|
416 |
|
|
(abs:VSX_B
|
417 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" ",wa"))))]
|
418 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
419 |
|
|
"xnabs %x0,%x1"
|
420 |
|
|
[(set_attr "type" "")
|
421 |
|
|
(set_attr "fp_type" "")])
|
422 |
|
|
|
423 |
|
|
(define_insn "vsx_smax3"
|
424 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
425 |
|
|
(smax:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")
|
426 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",wa")))]
|
427 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
428 |
|
|
"xmax %x0,%x1,%x2"
|
429 |
|
|
[(set_attr "type" "")
|
430 |
|
|
(set_attr "fp_type" "")])
|
431 |
|
|
|
432 |
|
|
(define_insn "*vsx_smin3"
|
433 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
434 |
|
|
(smin:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")
|
435 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",wa")))]
|
436 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
437 |
|
|
"xmin %x0,%x1,%x2"
|
438 |
|
|
[(set_attr "type" "")
|
439 |
|
|
(set_attr "fp_type" "")])
|
440 |
|
|
|
441 |
|
|
(define_insn "*vsx_sqrt2"
|
442 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
443 |
|
|
(sqrt:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")))]
|
444 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
445 |
|
|
"xsqrt %x0,%x1"
|
446 |
|
|
[(set_attr "type" "")
|
447 |
|
|
(set_attr "fp_type" "")])
|
448 |
|
|
|
449 |
|
|
(define_insn "vsx_rsqrte2"
|
450 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
451 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")]
|
452 |
|
|
UNSPEC_VSX_RSQRTE))]
|
453 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
454 |
|
|
"xrsqrte %x0,%x1"
|
455 |
|
|
[(set_attr "type" "")
|
456 |
|
|
(set_attr "fp_type" "")])
|
457 |
|
|
|
458 |
|
|
;; *tsqrt* returning the fg flag
|
459 |
|
|
(define_expand "vsx_tsqrt2_fg"
|
460 |
|
|
[(set (match_dup 3)
|
461 |
|
|
(unspec:CCFP [(match_operand:VSX_B 1 "vsx_register_operand" "")]
|
462 |
|
|
UNSPEC_VSX_TSQRT))
|
463 |
|
|
(set (match_operand:SI 0 "gpc_reg_operand" "")
|
464 |
|
|
(gt:SI (match_dup 3)
|
465 |
|
|
(const_int 0)))]
|
466 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
467 |
|
|
{
|
468 |
|
|
operands[3] = gen_reg_rtx (CCFPmode);
|
469 |
|
|
})
|
470 |
|
|
|
471 |
|
|
;; *tsqrt* returning the fe flag
|
472 |
|
|
(define_expand "vsx_tsqrt2_fe"
|
473 |
|
|
[(set (match_dup 3)
|
474 |
|
|
(unspec:CCFP [(match_operand:VSX_B 1 "vsx_register_operand" "")]
|
475 |
|
|
UNSPEC_VSX_TSQRT))
|
476 |
|
|
(set (match_operand:SI 0 "gpc_reg_operand" "")
|
477 |
|
|
(eq:SI (match_dup 3)
|
478 |
|
|
(const_int 0)))]
|
479 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
480 |
|
|
{
|
481 |
|
|
operands[3] = gen_reg_rtx (CCFPmode);
|
482 |
|
|
})
|
483 |
|
|
|
484 |
|
|
(define_insn "*vsx_tsqrt2_internal"
|
485 |
|
|
[(set (match_operand:CCFP 0 "cc_reg_operand" "=x,x")
|
486 |
|
|
(unspec:CCFP [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")]
|
487 |
|
|
UNSPEC_VSX_TSQRT))]
|
488 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
489 |
|
|
"xtsqrt %0,%x1"
|
490 |
|
|
[(set_attr "type" "")
|
491 |
|
|
(set_attr "fp_type" "")])
|
492 |
|
|
|
493 |
|
|
;; Fused vector multiply/add instructions
|
494 |
|
|
|
495 |
|
|
;; Note we have a pattern for the multiply/add operations that uses unspec and
|
496 |
|
|
;; does not check -mfused-madd to allow users to use these ops when they know
|
497 |
|
|
;; they want the fused multiply/add.
|
498 |
|
|
|
499 |
|
|
(define_expand "vsx_fmadd4"
|
500 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "")
|
501 |
|
|
(plus:VSX_B
|
502 |
|
|
(mult:VSX_B
|
503 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" "")
|
504 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ""))
|
505 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "")))]
|
506 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
507 |
|
|
{
|
508 |
|
|
if (!TARGET_FUSED_MADD)
|
509 |
|
|
{
|
510 |
|
|
emit_insn (gen_vsx_fmadd4_2 (operands[0], operands[1], operands[2],
|
511 |
|
|
operands[3]));
|
512 |
|
|
DONE;
|
513 |
|
|
}
|
514 |
|
|
})
|
515 |
|
|
|
516 |
|
|
(define_insn "*vsx_fmadd4_1"
|
517 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
518 |
|
|
(plus:VSX_B
|
519 |
|
|
(mult:VSX_B
|
520 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" "%,,wa,wa")
|
521 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0"))
|
522 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa")))]
|
523 |
|
|
"VECTOR_UNIT_VSX_P (mode) && TARGET_FUSED_MADD"
|
524 |
|
|
"@
|
525 |
|
|
xmadda %x0,%x1,%x2
|
526 |
|
|
xmaddm %x0,%x1,%x3
|
527 |
|
|
xmadda %x0,%x1,%x2
|
528 |
|
|
xmaddm %x0,%x1,%x3"
|
529 |
|
|
[(set_attr "type" "")
|
530 |
|
|
(set_attr "fp_type" "")])
|
531 |
|
|
|
532 |
|
|
(define_insn "vsx_fmadd4_2"
|
533 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
534 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" "%,,wa,wa")
|
535 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0")
|
536 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa")]
|
537 |
|
|
UNSPEC_VSX_MADD))]
|
538 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
539 |
|
|
"@
|
540 |
|
|
xmadda %x0,%x1,%x2
|
541 |
|
|
xmaddm %x0,%x1,%x3
|
542 |
|
|
xmadda %x0,%x1,%x2
|
543 |
|
|
xmaddm %x0,%x1,%x3"
|
544 |
|
|
[(set_attr "type" "")
|
545 |
|
|
(set_attr "fp_type" "")])
|
546 |
|
|
|
547 |
|
|
(define_expand "vsx_fmsub4"
|
548 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "")
|
549 |
|
|
(minus:VSX_B
|
550 |
|
|
(mult:VSX_B
|
551 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" "")
|
552 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ""))
|
553 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "")))]
|
554 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
555 |
|
|
{
|
556 |
|
|
if (!TARGET_FUSED_MADD)
|
557 |
|
|
{
|
558 |
|
|
emit_insn (gen_vsx_fmsub4_2 (operands[0], operands[1], operands[2],
|
559 |
|
|
operands[3]));
|
560 |
|
|
DONE;
|
561 |
|
|
}
|
562 |
|
|
})
|
563 |
|
|
|
564 |
|
|
(define_insn "*vsx_fmsub4_1"
|
565 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
566 |
|
|
(minus:VSX_B
|
567 |
|
|
(mult:VSX_B
|
568 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" "%,,wa,wa")
|
569 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0"))
|
570 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa")))]
|
571 |
|
|
"VECTOR_UNIT_VSX_P (mode) && TARGET_FUSED_MADD"
|
572 |
|
|
"@
|
573 |
|
|
xmsuba %x0,%x1,%x2
|
574 |
|
|
xmsubm %x0,%x1,%x3
|
575 |
|
|
xmsuba %x0,%x1,%x2
|
576 |
|
|
xmsubm %x0,%x1,%x3"
|
577 |
|
|
[(set_attr "type" "")
|
578 |
|
|
(set_attr "fp_type" "")])
|
579 |
|
|
|
580 |
|
|
(define_insn "vsx_fmsub4_2"
|
581 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
582 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" "%,,wa,wa")
|
583 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0")
|
584 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa")]
|
585 |
|
|
UNSPEC_VSX_MSUB))]
|
586 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
587 |
|
|
"@
|
588 |
|
|
xmsuba %x0,%x1,%x2
|
589 |
|
|
xmsubm %x0,%x1,%x3
|
590 |
|
|
xmsuba %x0,%x1,%x2
|
591 |
|
|
xmsubm %x0,%x1,%x3"
|
592 |
|
|
[(set_attr "type" "")
|
593 |
|
|
(set_attr "fp_type" "")])
|
594 |
|
|
|
595 |
|
|
(define_expand "vsx_fnmadd4"
|
596 |
|
|
[(match_operand:VSX_B 0 "vsx_register_operand" "")
|
597 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" "")
|
598 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" "")
|
599 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "")]
|
600 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
601 |
|
|
{
|
602 |
|
|
if (TARGET_FUSED_MADD && HONOR_SIGNED_ZEROS (DFmode))
|
603 |
|
|
{
|
604 |
|
|
emit_insn (gen_vsx_fnmadd4_1 (operands[0], operands[1],
|
605 |
|
|
operands[2], operands[3]));
|
606 |
|
|
DONE;
|
607 |
|
|
}
|
608 |
|
|
else if (TARGET_FUSED_MADD && !HONOR_SIGNED_ZEROS (DFmode))
|
609 |
|
|
{
|
610 |
|
|
emit_insn (gen_vsx_fnmadd4_2 (operands[0], operands[1],
|
611 |
|
|
operands[2], operands[3]));
|
612 |
|
|
DONE;
|
613 |
|
|
}
|
614 |
|
|
else
|
615 |
|
|
{
|
616 |
|
|
emit_insn (gen_vsx_fnmadd4_3 (operands[0], operands[1],
|
617 |
|
|
operands[2], operands[3]));
|
618 |
|
|
DONE;
|
619 |
|
|
}
|
620 |
|
|
})
|
621 |
|
|
|
622 |
|
|
(define_insn "vsx_fnmadd4_1"
|
623 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
624 |
|
|
(neg:VSX_B
|
625 |
|
|
(plus:VSX_B
|
626 |
|
|
(mult:VSX_B
|
627 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" ",,wa,wa")
|
628 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0"))
|
629 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa"))))]
|
630 |
|
|
"VECTOR_UNIT_VSX_P (mode) && TARGET_FUSED_MADD
|
631 |
|
|
&& HONOR_SIGNED_ZEROS (DFmode)"
|
632 |
|
|
"@
|
633 |
|
|
xnmadda %x0,%x1,%x2
|
634 |
|
|
xnmaddm %x0,%x1,%x3
|
635 |
|
|
xnmadda %x0,%x1,%x2
|
636 |
|
|
xnmaddm %x0,%x1,%x3"
|
637 |
|
|
[(set_attr "type" "")
|
638 |
|
|
(set_attr "fp_type" "")])
|
639 |
|
|
|
640 |
|
|
(define_insn "vsx_fnmadd4_2"
|
641 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
642 |
|
|
(minus:VSX_B
|
643 |
|
|
(mult:VSX_B
|
644 |
|
|
(neg:VSX_B
|
645 |
|
|
(match_operand:VSX_B 1 "gpc_reg_operand" ",,wa,wa"))
|
646 |
|
|
(match_operand:VSX_B 2 "gpc_reg_operand" ",0,wa,0"))
|
647 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa")))]
|
648 |
|
|
"VECTOR_UNIT_VSX_P (mode) && TARGET_FUSED_MADD
|
649 |
|
|
&& !HONOR_SIGNED_ZEROS (DFmode)"
|
650 |
|
|
"@
|
651 |
|
|
xnmadda %x0,%x1,%x2
|
652 |
|
|
xnmaddm %x0,%x1,%x3
|
653 |
|
|
xnmadda %x0,%x1,%x2
|
654 |
|
|
xnmaddm %x0,%x1,%x3"
|
655 |
|
|
[(set_attr "type" "")
|
656 |
|
|
(set_attr "fp_type" "")])
|
657 |
|
|
|
658 |
|
|
(define_insn "vsx_fnmadd4_3"
|
659 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
660 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" ",,wa,wa")
|
661 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0")
|
662 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa")]
|
663 |
|
|
UNSPEC_VSX_NMADD))]
|
664 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
665 |
|
|
"@
|
666 |
|
|
xnmadda %x0,%x1,%x2
|
667 |
|
|
xnmaddm %x0,%x1,%x3
|
668 |
|
|
xnmadda %x0,%x1,%x2
|
669 |
|
|
xnmaddm %x0,%x1,%x3"
|
670 |
|
|
[(set_attr "type" "")
|
671 |
|
|
(set_attr "fp_type" "")])
|
672 |
|
|
|
673 |
|
|
(define_expand "vsx_fnmsub4"
|
674 |
|
|
[(match_operand:VSX_B 0 "vsx_register_operand" "")
|
675 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" "")
|
676 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" "")
|
677 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "")]
|
678 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
679 |
|
|
{
|
680 |
|
|
if (TARGET_FUSED_MADD && HONOR_SIGNED_ZEROS (DFmode))
|
681 |
|
|
{
|
682 |
|
|
emit_insn (gen_vsx_fnmsub4_1 (operands[0], operands[1],
|
683 |
|
|
operands[2], operands[3]));
|
684 |
|
|
DONE;
|
685 |
|
|
}
|
686 |
|
|
else if (TARGET_FUSED_MADD && !HONOR_SIGNED_ZEROS (DFmode))
|
687 |
|
|
{
|
688 |
|
|
emit_insn (gen_vsx_fnmsub4_2 (operands[0], operands[1],
|
689 |
|
|
operands[2], operands[3]));
|
690 |
|
|
DONE;
|
691 |
|
|
}
|
692 |
|
|
else
|
693 |
|
|
{
|
694 |
|
|
emit_insn (gen_vsx_fnmsub4_3 (operands[0], operands[1],
|
695 |
|
|
operands[2], operands[3]));
|
696 |
|
|
DONE;
|
697 |
|
|
}
|
698 |
|
|
})
|
699 |
|
|
|
700 |
|
|
(define_insn "vsx_fnmsub4_1"
|
701 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
702 |
|
|
(neg:VSX_B
|
703 |
|
|
(minus:VSX_B
|
704 |
|
|
(mult:VSX_B
|
705 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" "%,,wa,wa")
|
706 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0"))
|
707 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa"))))]
|
708 |
|
|
"VECTOR_UNIT_VSX_P (mode) && TARGET_FUSED_MADD
|
709 |
|
|
&& HONOR_SIGNED_ZEROS (DFmode)"
|
710 |
|
|
"@
|
711 |
|
|
xnmsuba %x0,%x1,%x2
|
712 |
|
|
xnmsubm %x0,%x1,%x3
|
713 |
|
|
xnmsuba %x0,%x1,%x2
|
714 |
|
|
xnmsubm %x0,%x1,%x3"
|
715 |
|
|
[(set_attr "type" "")
|
716 |
|
|
(set_attr "fp_type" "")])
|
717 |
|
|
|
718 |
|
|
(define_insn "vsx_fnmsub4_2"
|
719 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
720 |
|
|
(minus:VSX_B
|
721 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa")
|
722 |
|
|
(mult:VSX_B
|
723 |
|
|
(match_operand:VSX_B 1 "vsx_register_operand" "%,,wa,wa")
|
724 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0"))))]
|
725 |
|
|
"VECTOR_UNIT_VSX_P (mode) && TARGET_FUSED_MADD
|
726 |
|
|
&& !HONOR_SIGNED_ZEROS (DFmode)"
|
727 |
|
|
"@
|
728 |
|
|
xnmsuba %x0,%x1,%x2
|
729 |
|
|
xnmsubm %x0,%x1,%x3
|
730 |
|
|
xnmsuba %x0,%x1,%x2
|
731 |
|
|
xnmsubm %x0,%x1,%x3"
|
732 |
|
|
[(set_attr "type" "")
|
733 |
|
|
(set_attr "fp_type" "")])
|
734 |
|
|
|
735 |
|
|
(define_insn "vsx_fnmsub4_3"
|
736 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,,?wa,?wa")
|
737 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" "%,,wa,wa")
|
738 |
|
|
(match_operand:VSX_B 2 "vsx_register_operand" ",0,wa,0")
|
739 |
|
|
(match_operand:VSX_B 3 "vsx_register_operand" "0,,0,wa")]
|
740 |
|
|
UNSPEC_VSX_NMSUB))]
|
741 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
742 |
|
|
"@
|
743 |
|
|
xnmsuba %x0,%x1,%x2
|
744 |
|
|
xnmsubm %x0,%x1,%x3
|
745 |
|
|
xnmsuba %x0,%x1,%x2
|
746 |
|
|
xnmsubm %x0,%x1,%x3"
|
747 |
|
|
[(set_attr "type" "")
|
748 |
|
|
(set_attr "fp_type" "")])
|
749 |
|
|
|
750 |
|
|
;; Vector conditional expressions (no scalar version for these instructions)
|
751 |
|
|
(define_insn "vsx_eq"
|
752 |
|
|
[(set (match_operand:VSX_F 0 "vsx_register_operand" "=,?wa")
|
753 |
|
|
(eq:VSX_F (match_operand:VSX_F 1 "vsx_register_operand" ",wa")
|
754 |
|
|
(match_operand:VSX_F 2 "vsx_register_operand" ",wa")))]
|
755 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
756 |
|
|
"xvcmpeq %x0,%x1,%x2"
|
757 |
|
|
[(set_attr "type" "")
|
758 |
|
|
(set_attr "fp_type" "")])
|
759 |
|
|
|
760 |
|
|
(define_insn "vsx_gt"
|
761 |
|
|
[(set (match_operand:VSX_F 0 "vsx_register_operand" "=,?wa")
|
762 |
|
|
(gt:VSX_F (match_operand:VSX_F 1 "vsx_register_operand" ",wa")
|
763 |
|
|
(match_operand:VSX_F 2 "vsx_register_operand" ",wa")))]
|
764 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
765 |
|
|
"xvcmpgt %x0,%x1,%x2"
|
766 |
|
|
[(set_attr "type" "")
|
767 |
|
|
(set_attr "fp_type" "")])
|
768 |
|
|
|
769 |
|
|
(define_insn "*vsx_ge"
|
770 |
|
|
[(set (match_operand:VSX_F 0 "vsx_register_operand" "=,?wa")
|
771 |
|
|
(ge:VSX_F (match_operand:VSX_F 1 "vsx_register_operand" ",wa")
|
772 |
|
|
(match_operand:VSX_F 2 "vsx_register_operand" ",wa")))]
|
773 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
774 |
|
|
"xvcmpge %x0,%x1,%x2"
|
775 |
|
|
[(set_attr "type" "")
|
776 |
|
|
(set_attr "fp_type" "")])
|
777 |
|
|
|
778 |
|
|
;; Floating point scalar compare
|
779 |
|
|
(define_insn "*vsx_cmpdf_internal1"
|
780 |
|
|
[(set (match_operand:CCFP 0 "cc_reg_operand" "=y,?y")
|
781 |
|
|
(compare:CCFP (match_operand:DF 1 "gpc_reg_operand" "ws,wa")
|
782 |
|
|
(match_operand:DF 2 "gpc_reg_operand" "ws,wa")))]
|
783 |
|
|
"TARGET_HARD_FLOAT && TARGET_FPRS && TARGET_DOUBLE_FLOAT
|
784 |
|
|
&& VECTOR_UNIT_VSX_P (DFmode)"
|
785 |
|
|
"xscmpudp %0,%x1,%x2"
|
786 |
|
|
[(set_attr "type" "fpcompare")])
|
787 |
|
|
|
788 |
|
|
;; Compare vectors producing a vector result and a predicate, setting CR6 to
|
789 |
|
|
;; indicate a combined status
|
790 |
|
|
(define_insn "*vsx_eq__p"
|
791 |
|
|
[(set (reg:CC 74)
|
792 |
|
|
(unspec:CC
|
793 |
|
|
[(eq:CC (match_operand:VSX_F 1 "vsx_register_operand" ",?wa")
|
794 |
|
|
(match_operand:VSX_F 2 "vsx_register_operand" ",?wa"))]
|
795 |
|
|
UNSPEC_PREDICATE))
|
796 |
|
|
(set (match_operand:VSX_F 0 "vsx_register_operand" "=,?wa")
|
797 |
|
|
(eq:VSX_F (match_dup 1)
|
798 |
|
|
(match_dup 2)))]
|
799 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
800 |
|
|
"xvcmpeq. %x0,%x1,%x2"
|
801 |
|
|
[(set_attr "type" "veccmp")])
|
802 |
|
|
|
803 |
|
|
(define_insn "*vsx_gt__p"
|
804 |
|
|
[(set (reg:CC 74)
|
805 |
|
|
(unspec:CC
|
806 |
|
|
[(gt:CC (match_operand:VSX_F 1 "vsx_register_operand" ",?wa")
|
807 |
|
|
(match_operand:VSX_F 2 "vsx_register_operand" ",?wa"))]
|
808 |
|
|
UNSPEC_PREDICATE))
|
809 |
|
|
(set (match_operand:VSX_F 0 "vsx_register_operand" "=,?wa")
|
810 |
|
|
(gt:VSX_F (match_dup 1)
|
811 |
|
|
(match_dup 2)))]
|
812 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
813 |
|
|
"xvcmpgt. %x0,%x1,%x2"
|
814 |
|
|
[(set_attr "type" "veccmp")])
|
815 |
|
|
|
816 |
|
|
(define_insn "*vsx_ge__p"
|
817 |
|
|
[(set (reg:CC 74)
|
818 |
|
|
(unspec:CC
|
819 |
|
|
[(ge:CC (match_operand:VSX_F 1 "vsx_register_operand" ",?wa")
|
820 |
|
|
(match_operand:VSX_F 2 "vsx_register_operand" ",?wa"))]
|
821 |
|
|
UNSPEC_PREDICATE))
|
822 |
|
|
(set (match_operand:VSX_F 0 "vsx_register_operand" "=,?wa")
|
823 |
|
|
(ge:VSX_F (match_dup 1)
|
824 |
|
|
(match_dup 2)))]
|
825 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
826 |
|
|
"xvcmpge. %x0,%x1,%x2"
|
827 |
|
|
[(set_attr "type" "veccmp")])
|
828 |
|
|
|
829 |
|
|
;; Vector select
|
830 |
|
|
(define_insn "*vsx_xxsel"
|
831 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=,?wa")
|
832 |
|
|
(if_then_else:VSX_L
|
833 |
|
|
(ne:CC (match_operand:VSX_L 1 "vsx_register_operand" ",wa")
|
834 |
|
|
(const_int 0))
|
835 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" ",wa")
|
836 |
|
|
(match_operand:VSX_L 3 "vsx_register_operand" ",wa")))]
|
837 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
838 |
|
|
"xxsel %x0,%x3,%x2,%x1"
|
839 |
|
|
[(set_attr "type" "vecperm")])
|
840 |
|
|
|
841 |
|
|
(define_insn "*vsx_xxsel_uns"
|
842 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=,?wa")
|
843 |
|
|
(if_then_else:VSX_L
|
844 |
|
|
(ne:CCUNS (match_operand:VSX_L 1 "vsx_register_operand" ",wa")
|
845 |
|
|
(const_int 0))
|
846 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" ",wa")
|
847 |
|
|
(match_operand:VSX_L 3 "vsx_register_operand" ",wa")))]
|
848 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
849 |
|
|
"xxsel %x0,%x3,%x2,%x1"
|
850 |
|
|
[(set_attr "type" "vecperm")])
|
851 |
|
|
|
852 |
|
|
;; Copy sign
|
853 |
|
|
(define_insn "vsx_copysign3"
|
854 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
855 |
|
|
(if_then_else:VSX_B
|
856 |
|
|
(ge:VSX_B (match_operand:VSX_B 2 "vsx_register_operand" ",wa")
|
857 |
|
|
(match_operand:VSX_B 3 "zero_constant" "j,j"))
|
858 |
|
|
(abs:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa"))
|
859 |
|
|
(neg:VSX_B (abs:VSX_B (match_dup 1)))))]
|
860 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
861 |
|
|
"xcpsgn %x0,%x2,%x1"
|
862 |
|
|
[(set_attr "type" "")
|
863 |
|
|
(set_attr "fp_type" "")])
|
864 |
|
|
|
865 |
|
|
;; For the conversions, limit the register class for the integer value to be
|
866 |
|
|
;; the fprs because we don't want to add the altivec registers to movdi/movsi.
|
867 |
|
|
;; For the unsigned tests, there isn't a generic double -> unsigned conversion
|
868 |
|
|
;; in rs6000.md so don't test VECTOR_UNIT_VSX_P, just test against VSX.
|
869 |
|
|
(define_insn "vsx_float2"
|
870 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
871 |
|
|
(float:VSX_B (match_operand: 1 "vsx_register_operand" ",")))]
|
872 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
873 |
|
|
"xcvsx %x0,%x1"
|
874 |
|
|
[(set_attr "type" "")
|
875 |
|
|
(set_attr "fp_type" "")])
|
876 |
|
|
|
877 |
|
|
(define_insn "vsx_floatuns2"
|
878 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
879 |
|
|
(unsigned_float:VSX_B (match_operand: 1 "vsx_register_operand" ",")))]
|
880 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
881 |
|
|
"xcvux %x0,%x1"
|
882 |
|
|
[(set_attr "type" "")
|
883 |
|
|
(set_attr "fp_type" "")])
|
884 |
|
|
|
885 |
|
|
(define_insn "vsx_fix_trunc2"
|
886 |
|
|
[(set (match_operand: 0 "vsx_register_operand" "=,?")
|
887 |
|
|
(fix: (match_operand:VSX_B 1 "vsx_register_operand" ",wa")))]
|
888 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
889 |
|
|
"xcvsxs %x0,%x1"
|
890 |
|
|
[(set_attr "type" "")
|
891 |
|
|
(set_attr "fp_type" "")])
|
892 |
|
|
|
893 |
|
|
(define_insn "vsx_fixuns_trunc2"
|
894 |
|
|
[(set (match_operand: 0 "vsx_register_operand" "=,?")
|
895 |
|
|
(unsigned_fix: (match_operand:VSX_B 1 "vsx_register_operand" ",wa")))]
|
896 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
897 |
|
|
"xcvuxs %x0,%x1"
|
898 |
|
|
[(set_attr "type" "")
|
899 |
|
|
(set_attr "fp_type" "")])
|
900 |
|
|
|
901 |
|
|
;; Math rounding functions
|
902 |
|
|
(define_insn "vsx_xri"
|
903 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
904 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")]
|
905 |
|
|
UNSPEC_VSX_ROUND_I))]
|
906 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
907 |
|
|
"xri %x0,%x1"
|
908 |
|
|
[(set_attr "type" "")
|
909 |
|
|
(set_attr "fp_type" "")])
|
910 |
|
|
|
911 |
|
|
(define_insn "vsx_xric"
|
912 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
913 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")]
|
914 |
|
|
UNSPEC_VSX_ROUND_IC))]
|
915 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
916 |
|
|
"xric %x0,%x1"
|
917 |
|
|
[(set_attr "type" "")
|
918 |
|
|
(set_attr "fp_type" "")])
|
919 |
|
|
|
920 |
|
|
(define_insn "vsx_btrunc2"
|
921 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
922 |
|
|
(fix:VSX_B (match_operand:VSX_B 1 "vsx_register_operand" ",wa")))]
|
923 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
924 |
|
|
"xriz %x0,%x1"
|
925 |
|
|
[(set_attr "type" "")
|
926 |
|
|
(set_attr "fp_type" "")])
|
927 |
|
|
|
928 |
|
|
(define_insn "*vsx_b2trunc2"
|
929 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
930 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")]
|
931 |
|
|
UNSPEC_FRIZ))]
|
932 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
933 |
|
|
"xriz %x0,%x1"
|
934 |
|
|
[(set_attr "type" "")
|
935 |
|
|
(set_attr "fp_type" "")])
|
936 |
|
|
|
937 |
|
|
(define_insn "vsx_floor2"
|
938 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
939 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")]
|
940 |
|
|
UNSPEC_FRIM))]
|
941 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
942 |
|
|
"xrim %x0,%x1"
|
943 |
|
|
[(set_attr "type" "")
|
944 |
|
|
(set_attr "fp_type" "")])
|
945 |
|
|
|
946 |
|
|
(define_insn "vsx_ceil2"
|
947 |
|
|
[(set (match_operand:VSX_B 0 "vsx_register_operand" "=,?wa")
|
948 |
|
|
(unspec:VSX_B [(match_operand:VSX_B 1 "vsx_register_operand" ",wa")]
|
949 |
|
|
UNSPEC_FRIP))]
|
950 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
951 |
|
|
"xrip %x0,%x1"
|
952 |
|
|
[(set_attr "type" "")
|
953 |
|
|
(set_attr "fp_type" "")])
|
954 |
|
|
|
955 |
|
|
|
956 |
|
|
;; VSX convert to/from double vector
|
957 |
|
|
|
958 |
|
|
;; Convert between single and double precision
|
959 |
|
|
;; Don't use xscvspdp and xscvdpsp for scalar conversions, since the normal
|
960 |
|
|
;; scalar single precision instructions internally use the double format.
|
961 |
|
|
;; Prefer the altivec registers, since we likely will need to do a vperm
|
962 |
|
|
(define_insn "vsx_"
|
963 |
|
|
[(set (match_operand: 0 "vsx_register_operand" "=,?wa")
|
964 |
|
|
(unspec: [(match_operand:VSX_SPDP 1 "vsx_register_operand" ",wa")]
|
965 |
|
|
UNSPEC_VSX_CVSPDP))]
|
966 |
|
|
"VECTOR_UNIT_VSX_P (mode)"
|
967 |
|
|
" %x0,%x1"
|
968 |
|
|
[(set_attr "type" "")])
|
969 |
|
|
|
970 |
|
|
;; xscvspdp, represent the scalar SF type as V4SF
|
971 |
|
|
(define_insn "vsx_xscvspdp"
|
972 |
|
|
[(set (match_operand:DF 0 "vsx_register_operand" "=ws,?wa")
|
973 |
|
|
(unspec:DF [(match_operand:V4SF 1 "vsx_register_operand" "wa,wa")]
|
974 |
|
|
UNSPEC_VSX_CVSPDP))]
|
975 |
|
|
"VECTOR_UNIT_VSX_P (DFmode)"
|
976 |
|
|
"xscvspdp %x0,%x1"
|
977 |
|
|
[(set_attr "type" "fp")])
|
978 |
|
|
|
979 |
|
|
;; xscvdpsp used for splat'ing a scalar to V4SF, knowing that the internal SF
|
980 |
|
|
;; format of scalars is actually DF.
|
981 |
|
|
(define_insn "vsx_xscvdpsp_scalar"
|
982 |
|
|
[(set (match_operand:V4SF 0 "vsx_register_operand" "=wa")
|
983 |
|
|
(unspec:V4SF [(match_operand:SF 1 "vsx_register_operand" "f")]
|
984 |
|
|
UNSPEC_VSX_CVSPDP))]
|
985 |
|
|
"VECTOR_UNIT_VSX_P (DFmode)"
|
986 |
|
|
"xscvdpsp %x0,%x1"
|
987 |
|
|
[(set_attr "type" "fp")])
|
988 |
|
|
|
989 |
|
|
;; Convert from 64-bit to 32-bit types
|
990 |
|
|
;; Note, favor the Altivec registers since the usual use of these instructions
|
991 |
|
|
;; is in vector converts and we need to use the Altivec vperm instruction.
|
992 |
|
|
|
993 |
|
|
(define_insn "vsx_xvcvdpsxws"
|
994 |
|
|
[(set (match_operand:V4SI 0 "vsx_register_operand" "=v,?wa")
|
995 |
|
|
(unspec:V4SI [(match_operand:V2DF 1 "vsx_register_operand" "wd,wa")]
|
996 |
|
|
UNSPEC_VSX_CVDPSXWS))]
|
997 |
|
|
"VECTOR_UNIT_VSX_P (V2DFmode)"
|
998 |
|
|
"xvcvdpsxws %x0,%x1"
|
999 |
|
|
[(set_attr "type" "vecfloat")])
|
1000 |
|
|
|
1001 |
|
|
(define_insn "vsx_xvcvdpuxws"
|
1002 |
|
|
[(set (match_operand:V4SI 0 "vsx_register_operand" "=v,?wa")
|
1003 |
|
|
(unspec:V4SI [(match_operand:V2DF 1 "vsx_register_operand" "wd,wa")]
|
1004 |
|
|
UNSPEC_VSX_CVDPUXWS))]
|
1005 |
|
|
"VECTOR_UNIT_VSX_P (V2DFmode)"
|
1006 |
|
|
"xvcvdpuxws %x0,%x1"
|
1007 |
|
|
[(set_attr "type" "vecfloat")])
|
1008 |
|
|
|
1009 |
|
|
(define_insn "vsx_xvcvsxdsp"
|
1010 |
|
|
[(set (match_operand:V4SI 0 "vsx_register_operand" "=wd,?wa")
|
1011 |
|
|
(unspec:V4SI [(match_operand:V2DF 1 "vsx_register_operand" "wf,wa")]
|
1012 |
|
|
UNSPEC_VSX_CVSXDSP))]
|
1013 |
|
|
"VECTOR_UNIT_VSX_P (V2DFmode)"
|
1014 |
|
|
"xvcvsxdsp %x0,%x1"
|
1015 |
|
|
[(set_attr "type" "vecfloat")])
|
1016 |
|
|
|
1017 |
|
|
(define_insn "vsx_xvcvuxdsp"
|
1018 |
|
|
[(set (match_operand:V4SI 0 "vsx_register_operand" "=wd,?wa")
|
1019 |
|
|
(unspec:V4SI [(match_operand:V2DF 1 "vsx_register_operand" "wf,wa")]
|
1020 |
|
|
UNSPEC_VSX_CVUXDSP))]
|
1021 |
|
|
"VECTOR_UNIT_VSX_P (V2DFmode)"
|
1022 |
|
|
"xvcvuxwdp %x0,%x1"
|
1023 |
|
|
[(set_attr "type" "vecfloat")])
|
1024 |
|
|
|
1025 |
|
|
;; Convert from 32-bit to 64-bit types
|
1026 |
|
|
(define_insn "vsx_xvcvsxwdp"
|
1027 |
|
|
[(set (match_operand:V2DF 0 "vsx_register_operand" "=wd,?wa")
|
1028 |
|
|
(unspec:V2DF [(match_operand:V4SI 1 "vsx_register_operand" "wf,wa")]
|
1029 |
|
|
UNSPEC_VSX_CVSXWDP))]
|
1030 |
|
|
"VECTOR_UNIT_VSX_P (V2DFmode)"
|
1031 |
|
|
"xvcvsxwdp %x0,%x1"
|
1032 |
|
|
[(set_attr "type" "vecfloat")])
|
1033 |
|
|
|
1034 |
|
|
(define_insn "vsx_xvcvuxwdp"
|
1035 |
|
|
[(set (match_operand:V2DF 0 "vsx_register_operand" "=wd,?wa")
|
1036 |
|
|
(unspec:V2DF [(match_operand:V4SI 1 "vsx_register_operand" "wf,wa")]
|
1037 |
|
|
UNSPEC_VSX_CVUXWDP))]
|
1038 |
|
|
"VECTOR_UNIT_VSX_P (V2DFmode)"
|
1039 |
|
|
"xvcvuxwdp %x0,%x1"
|
1040 |
|
|
[(set_attr "type" "vecfloat")])
|
1041 |
|
|
|
1042 |
|
|
(define_insn "vsx_xvcvspsxds"
|
1043 |
|
|
[(set (match_operand:V2DI 0 "vsx_register_operand" "=v,?wa")
|
1044 |
|
|
(unspec:V2DI [(match_operand:V4SF 1 "vsx_register_operand" "wd,wa")]
|
1045 |
|
|
UNSPEC_VSX_CVSPSXDS))]
|
1046 |
|
|
"VECTOR_UNIT_VSX_P (V2DFmode)"
|
1047 |
|
|
"xvcvspsxds %x0,%x1"
|
1048 |
|
|
[(set_attr "type" "vecfloat")])
|
1049 |
|
|
|
1050 |
|
|
(define_insn "vsx_xvcvspuxds"
|
1051 |
|
|
[(set (match_operand:V2DI 0 "vsx_register_operand" "=v,?wa")
|
1052 |
|
|
(unspec:V2DI [(match_operand:V4SF 1 "vsx_register_operand" "wd,wa")]
|
1053 |
|
|
UNSPEC_VSX_CVSPUXDS))]
|
1054 |
|
|
"VECTOR_UNIT_VSX_P (V2DFmode)"
|
1055 |
|
|
"xvcvspuxds %x0,%x1"
|
1056 |
|
|
[(set_attr "type" "vecfloat")])
|
1057 |
|
|
|
1058 |
|
|
;; Logical and permute operations
|
1059 |
|
|
(define_insn "*vsx_and3"
|
1060 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=,?wa")
|
1061 |
|
|
(and:VSX_L
|
1062 |
|
|
(match_operand:VSX_L 1 "vsx_register_operand" ",?wa")
|
1063 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" ",?wa")))]
|
1064 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1065 |
|
|
"xxland %x0,%x1,%x2"
|
1066 |
|
|
[(set_attr "type" "vecsimple")])
|
1067 |
|
|
|
1068 |
|
|
(define_insn "*vsx_ior3"
|
1069 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=,?wa")
|
1070 |
|
|
(ior:VSX_L (match_operand:VSX_L 1 "vsx_register_operand" ",?wa")
|
1071 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" ",?wa")))]
|
1072 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1073 |
|
|
"xxlor %x0,%x1,%x2"
|
1074 |
|
|
[(set_attr "type" "vecsimple")])
|
1075 |
|
|
|
1076 |
|
|
(define_insn "*vsx_xor3"
|
1077 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=,?wa")
|
1078 |
|
|
(xor:VSX_L
|
1079 |
|
|
(match_operand:VSX_L 1 "vsx_register_operand" ",?wa")
|
1080 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" ",?wa")))]
|
1081 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1082 |
|
|
"xxlxor %x0,%x1,%x2"
|
1083 |
|
|
[(set_attr "type" "vecsimple")])
|
1084 |
|
|
|
1085 |
|
|
(define_insn "*vsx_one_cmpl2"
|
1086 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=,?wa")
|
1087 |
|
|
(not:VSX_L
|
1088 |
|
|
(match_operand:VSX_L 1 "vsx_register_operand" ",?wa")))]
|
1089 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1090 |
|
|
"xxlnor %x0,%x1,%x1"
|
1091 |
|
|
[(set_attr "type" "vecsimple")])
|
1092 |
|
|
|
1093 |
|
|
(define_insn "*vsx_nor3"
|
1094 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=,?wa")
|
1095 |
|
|
(not:VSX_L
|
1096 |
|
|
(ior:VSX_L
|
1097 |
|
|
(match_operand:VSX_L 1 "vsx_register_operand" ",?wa")
|
1098 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" ",?wa"))))]
|
1099 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1100 |
|
|
"xxlnor %x0,%x1,%x2"
|
1101 |
|
|
[(set_attr "type" "vecsimple")])
|
1102 |
|
|
|
1103 |
|
|
(define_insn "*vsx_andc3"
|
1104 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=,?wa")
|
1105 |
|
|
(and:VSX_L
|
1106 |
|
|
(not:VSX_L
|
1107 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" ",?wa"))
|
1108 |
|
|
(match_operand:VSX_L 1 "vsx_register_operand" ",?wa")))]
|
1109 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1110 |
|
|
"xxlandc %x0,%x1,%x2"
|
1111 |
|
|
[(set_attr "type" "vecsimple")])
|
1112 |
|
|
|
1113 |
|
|
|
1114 |
|
|
;; Permute operations
|
1115 |
|
|
|
1116 |
|
|
;; Build a V2DF/V2DI vector from two scalars
|
1117 |
|
|
(define_insn "vsx_concat_"
|
1118 |
|
|
[(set (match_operand:VSX_D 0 "vsx_register_operand" "=wd,?wa")
|
1119 |
|
|
(unspec:VSX_D
|
1120 |
|
|
[(match_operand: 1 "vsx_register_operand" "ws,wa")
|
1121 |
|
|
(match_operand: 2 "vsx_register_operand" "ws,wa")]
|
1122 |
|
|
UNSPEC_VSX_CONCAT))]
|
1123 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1124 |
|
|
"xxpermdi %x0,%x1,%x2,0"
|
1125 |
|
|
[(set_attr "type" "vecperm")])
|
1126 |
|
|
|
1127 |
|
|
;; Special purpose concat using xxpermdi to glue two single precision values
|
1128 |
|
|
;; together, relying on the fact that internally scalar floats are represented
|
1129 |
|
|
;; as doubles. This is used to initialize a V4SF vector with 4 floats
|
1130 |
|
|
(define_insn "vsx_concat_v2sf"
|
1131 |
|
|
[(set (match_operand:V2DF 0 "vsx_register_operand" "=wd,?wa")
|
1132 |
|
|
(unspec:V2DF
|
1133 |
|
|
[(match_operand:SF 1 "vsx_register_operand" "f,f")
|
1134 |
|
|
(match_operand:SF 2 "vsx_register_operand" "f,f")]
|
1135 |
|
|
UNSPEC_VSX_CONCAT))]
|
1136 |
|
|
"VECTOR_MEM_VSX_P (V2DFmode)"
|
1137 |
|
|
"xxpermdi %x0,%x1,%x2,0"
|
1138 |
|
|
[(set_attr "type" "vecperm")])
|
1139 |
|
|
|
1140 |
|
|
;; Set the element of a V2DI/VD2F mode
|
1141 |
|
|
(define_insn "vsx_set_"
|
1142 |
|
|
[(set (match_operand:VSX_D 0 "vsx_register_operand" "=wd,?wa")
|
1143 |
|
|
(unspec:VSX_D [(match_operand:VSX_D 1 "vsx_register_operand" "wd,wa")
|
1144 |
|
|
(match_operand: 2 "vsx_register_operand" "ws,wa")
|
1145 |
|
|
(match_operand:QI 3 "u5bit_cint_operand" "i,i")]
|
1146 |
|
|
UNSPEC_VSX_SET))]
|
1147 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1148 |
|
|
{
|
1149 |
|
|
if (INTVAL (operands[3]) == 0)
|
1150 |
|
|
return \"xxpermdi %x0,%x1,%x2,1\";
|
1151 |
|
|
else if (INTVAL (operands[3]) == 1)
|
1152 |
|
|
return \"xxpermdi %x0,%x2,%x1,0\";
|
1153 |
|
|
else
|
1154 |
|
|
gcc_unreachable ();
|
1155 |
|
|
}
|
1156 |
|
|
[(set_attr "type" "vecperm")])
|
1157 |
|
|
|
1158 |
|
|
;; Extract a DF/DI element from V2DF/V2DI
|
1159 |
|
|
(define_insn "vsx_extract_"
|
1160 |
|
|
[(set (match_operand: 0 "vsx_register_operand" "=ws,d,?wa")
|
1161 |
|
|
(vec_select: (match_operand:VSX_D 1 "vsx_register_operand" "wd,wd,wa")
|
1162 |
|
|
(parallel
|
1163 |
|
|
[(match_operand:QI 2 "u5bit_cint_operand" "i,i,i")])))]
|
1164 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1165 |
|
|
{
|
1166 |
|
|
gcc_assert (UINTVAL (operands[2]) <= 1);
|
1167 |
|
|
operands[3] = GEN_INT (INTVAL (operands[2]) << 1);
|
1168 |
|
|
return \"xxpermdi %x0,%x1,%x1,%3\";
|
1169 |
|
|
}
|
1170 |
|
|
[(set_attr "type" "vecperm")])
|
1171 |
|
|
|
1172 |
|
|
;; Optimize extracting element 0 from memory
|
1173 |
|
|
(define_insn "*vsx_extract__zero"
|
1174 |
|
|
[(set (match_operand: 0 "vsx_register_operand" "=ws,d,?wa")
|
1175 |
|
|
(vec_select:
|
1176 |
|
|
(match_operand:VSX_D 1 "indexed_or_indirect_operand" "Z,Z,Z")
|
1177 |
|
|
(parallel [(const_int 0)])))]
|
1178 |
|
|
"VECTOR_MEM_VSX_P (mode) && WORDS_BIG_ENDIAN"
|
1179 |
|
|
"lxsd%U1x %x0,%y1"
|
1180 |
|
|
[(set_attr "type" "fpload")
|
1181 |
|
|
(set_attr "length" "4")])
|
1182 |
|
|
|
1183 |
|
|
;; General double word oriented permute, allow the other vector types for
|
1184 |
|
|
;; optimizing the permute instruction.
|
1185 |
|
|
(define_insn "vsx_xxpermdi_"
|
1186 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=wd,?wa")
|
1187 |
|
|
(unspec:VSX_L [(match_operand:VSX_L 1 "vsx_register_operand" "wd,wa")
|
1188 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" "wd,wa")
|
1189 |
|
|
(match_operand:QI 3 "u5bit_cint_operand" "i,i")]
|
1190 |
|
|
UNSPEC_VSX_XXPERMDI))]
|
1191 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1192 |
|
|
"xxpermdi %x0,%x1,%x2,%3"
|
1193 |
|
|
[(set_attr "type" "vecperm")])
|
1194 |
|
|
|
1195 |
|
|
;; Varient of xxpermdi that is emitted by the vec_interleave functions
|
1196 |
|
|
(define_insn "*vsx_xxpermdi2_"
|
1197 |
|
|
[(set (match_operand:VSX_D 0 "vsx_register_operand" "=wd")
|
1198 |
|
|
(vec_concat:VSX_D
|
1199 |
|
|
(vec_select:
|
1200 |
|
|
(match_operand:VSX_D 1 "vsx_register_operand" "wd")
|
1201 |
|
|
(parallel
|
1202 |
|
|
[(match_operand:QI 2 "u5bit_cint_operand" "i")]))
|
1203 |
|
|
(vec_select:
|
1204 |
|
|
(match_operand:VSX_D 3 "vsx_register_operand" "wd")
|
1205 |
|
|
(parallel
|
1206 |
|
|
[(match_operand:QI 4 "u5bit_cint_operand" "i")]))))]
|
1207 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1208 |
|
|
{
|
1209 |
|
|
gcc_assert ((UINTVAL (operands[2]) <= 1) && (UINTVAL (operands[4]) <= 1));
|
1210 |
|
|
operands[5] = GEN_INT (((INTVAL (operands[2]) & 1) << 1)
|
1211 |
|
|
| (INTVAL (operands[4]) & 1));
|
1212 |
|
|
return \"xxpermdi %x0,%x1,%x3,%5\";
|
1213 |
|
|
}
|
1214 |
|
|
[(set_attr "type" "vecperm")])
|
1215 |
|
|
|
1216 |
|
|
;; V2DF/V2DI splat
|
1217 |
|
|
(define_insn "vsx_splat_"
|
1218 |
|
|
[(set (match_operand:VSX_D 0 "vsx_register_operand" "=wd,wd,wd,?wa,?wa,?wa")
|
1219 |
|
|
(vec_duplicate:VSX_D
|
1220 |
|
|
(match_operand: 1 "input_operand" "ws,f,Z,wa,wa,Z")))]
|
1221 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1222 |
|
|
"@
|
1223 |
|
|
xxpermdi %x0,%x1,%x1,0
|
1224 |
|
|
xxpermdi %x0,%x1,%x1,0
|
1225 |
|
|
lxvdsx %x0,%y1
|
1226 |
|
|
xxpermdi %x0,%x1,%x1,0
|
1227 |
|
|
xxpermdi %x0,%x1,%x1,0
|
1228 |
|
|
lxvdsx %x0,%y1"
|
1229 |
|
|
[(set_attr "type" "vecperm,vecperm,vecload,vecperm,vecperm,vecload")])
|
1230 |
|
|
|
1231 |
|
|
;; V4SF/V4SI splat
|
1232 |
|
|
(define_insn "vsx_xxspltw_"
|
1233 |
|
|
[(set (match_operand:VSX_W 0 "vsx_register_operand" "=wf,?wa")
|
1234 |
|
|
(vec_duplicate:VSX_W
|
1235 |
|
|
(vec_select:
|
1236 |
|
|
(match_operand:VSX_W 1 "vsx_register_operand" "wf,wa")
|
1237 |
|
|
(parallel
|
1238 |
|
|
[(match_operand:QI 2 "u5bit_cint_operand" "i,i")]))))]
|
1239 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1240 |
|
|
"xxspltw %x0,%x1,%2"
|
1241 |
|
|
[(set_attr "type" "vecperm")])
|
1242 |
|
|
|
1243 |
|
|
;; V4SF/V4SI interleave
|
1244 |
|
|
(define_insn "vsx_xxmrghw_"
|
1245 |
|
|
[(set (match_operand:VSX_W 0 "vsx_register_operand" "=wf,?wa")
|
1246 |
|
|
(vec_merge:VSX_W
|
1247 |
|
|
(vec_select:VSX_W
|
1248 |
|
|
(match_operand:VSX_W 1 "vsx_register_operand" "wf,wa")
|
1249 |
|
|
(parallel [(const_int 0)
|
1250 |
|
|
(const_int 2)
|
1251 |
|
|
(const_int 1)
|
1252 |
|
|
(const_int 3)]))
|
1253 |
|
|
(vec_select:VSX_W
|
1254 |
|
|
(match_operand:VSX_W 2 "vsx_register_operand" "wf,wa")
|
1255 |
|
|
(parallel [(const_int 2)
|
1256 |
|
|
(const_int 0)
|
1257 |
|
|
(const_int 3)
|
1258 |
|
|
(const_int 1)]))
|
1259 |
|
|
(const_int 5)))]
|
1260 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1261 |
|
|
"xxmrghw %x0,%x1,%x2"
|
1262 |
|
|
[(set_attr "type" "vecperm")])
|
1263 |
|
|
|
1264 |
|
|
(define_insn "vsx_xxmrglw_"
|
1265 |
|
|
[(set (match_operand:VSX_W 0 "vsx_register_operand" "=wf,?wa")
|
1266 |
|
|
(vec_merge:VSX_W
|
1267 |
|
|
(vec_select:VSX_W
|
1268 |
|
|
(match_operand:VSX_W 1 "vsx_register_operand" "wf,wa")
|
1269 |
|
|
(parallel [(const_int 2)
|
1270 |
|
|
(const_int 0)
|
1271 |
|
|
(const_int 3)
|
1272 |
|
|
(const_int 1)]))
|
1273 |
|
|
(vec_select:VSX_W
|
1274 |
|
|
(match_operand:VSX_W 2 "vsx_register_operand" "wf,?wa")
|
1275 |
|
|
(parallel [(const_int 0)
|
1276 |
|
|
(const_int 2)
|
1277 |
|
|
(const_int 1)
|
1278 |
|
|
(const_int 3)]))
|
1279 |
|
|
(const_int 5)))]
|
1280 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1281 |
|
|
"xxmrglw %x0,%x1,%x2"
|
1282 |
|
|
[(set_attr "type" "vecperm")])
|
1283 |
|
|
|
1284 |
|
|
;; Shift left double by word immediate
|
1285 |
|
|
(define_insn "vsx_xxsldwi_"
|
1286 |
|
|
[(set (match_operand:VSX_L 0 "vsx_register_operand" "=wa")
|
1287 |
|
|
(unspec:VSX_L [(match_operand:VSX_L 1 "vsx_register_operand" "wa")
|
1288 |
|
|
(match_operand:VSX_L 2 "vsx_register_operand" "wa")
|
1289 |
|
|
(match_operand:QI 3 "u5bit_cint_operand" "i")]
|
1290 |
|
|
UNSPEC_VSX_SLDWI))]
|
1291 |
|
|
"VECTOR_MEM_VSX_P (mode)"
|
1292 |
|
|
"xxsldwi %x0,%x1,%x2,%3"
|
1293 |
|
|
[(set_attr "type" "vecperm")])
|