1 |
38 |
julius |
/* Perform simple optimizations to clean up the result of reload.
|
2 |
|
|
Copyright (C) 1987, 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
|
3 |
|
|
1999, 2000, 2001, 2002, 2003, 2004, 2005, 2007 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 it under
|
8 |
|
|
the terms of the GNU General Public License as published by the Free
|
9 |
|
|
Software Foundation; either version 3, or (at your option) any later
|
10 |
|
|
version.
|
11 |
|
|
|
12 |
|
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
13 |
|
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
14 |
|
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
15 |
|
|
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 |
|
|
<http://www.gnu.org/licenses/>. */
|
20 |
|
|
|
21 |
|
|
#include "config.h"
|
22 |
|
|
#include "system.h"
|
23 |
|
|
#include "coretypes.h"
|
24 |
|
|
#include "tm.h"
|
25 |
|
|
|
26 |
|
|
#include "machmode.h"
|
27 |
|
|
#include "hard-reg-set.h"
|
28 |
|
|
#include "rtl.h"
|
29 |
|
|
#include "tm_p.h"
|
30 |
|
|
#include "obstack.h"
|
31 |
|
|
#include "insn-config.h"
|
32 |
|
|
#include "flags.h"
|
33 |
|
|
#include "function.h"
|
34 |
|
|
#include "expr.h"
|
35 |
|
|
#include "optabs.h"
|
36 |
|
|
#include "regs.h"
|
37 |
|
|
#include "basic-block.h"
|
38 |
|
|
#include "reload.h"
|
39 |
|
|
#include "recog.h"
|
40 |
|
|
#include "output.h"
|
41 |
|
|
#include "cselib.h"
|
42 |
|
|
#include "real.h"
|
43 |
|
|
#include "toplev.h"
|
44 |
|
|
#include "except.h"
|
45 |
|
|
#include "tree.h"
|
46 |
|
|
#include "timevar.h"
|
47 |
|
|
#include "tree-pass.h"
|
48 |
|
|
|
49 |
|
|
static int reload_cse_noop_set_p (rtx);
|
50 |
|
|
static void reload_cse_simplify (rtx, rtx);
|
51 |
|
|
static void reload_cse_regs_1 (rtx);
|
52 |
|
|
static int reload_cse_simplify_set (rtx, rtx);
|
53 |
|
|
static int reload_cse_simplify_operands (rtx, rtx);
|
54 |
|
|
|
55 |
|
|
static void reload_combine (void);
|
56 |
|
|
static void reload_combine_note_use (rtx *, rtx);
|
57 |
|
|
static void reload_combine_note_store (rtx, rtx, void *);
|
58 |
|
|
|
59 |
|
|
static void reload_cse_move2add (rtx);
|
60 |
|
|
static void move2add_note_store (rtx, rtx, void *);
|
61 |
|
|
|
62 |
|
|
/* Call cse / combine like post-reload optimization phases.
|
63 |
|
|
FIRST is the first instruction. */
|
64 |
|
|
void
|
65 |
|
|
reload_cse_regs (rtx first ATTRIBUTE_UNUSED)
|
66 |
|
|
{
|
67 |
|
|
reload_cse_regs_1 (first);
|
68 |
|
|
reload_combine ();
|
69 |
|
|
reload_cse_move2add (first);
|
70 |
|
|
if (flag_expensive_optimizations)
|
71 |
|
|
reload_cse_regs_1 (first);
|
72 |
|
|
}
|
73 |
|
|
|
74 |
|
|
/* See whether a single set SET is a noop. */
|
75 |
|
|
static int
|
76 |
|
|
reload_cse_noop_set_p (rtx set)
|
77 |
|
|
{
|
78 |
|
|
if (cselib_reg_set_mode (SET_DEST (set)) != GET_MODE (SET_DEST (set)))
|
79 |
|
|
return 0;
|
80 |
|
|
|
81 |
|
|
return rtx_equal_for_cselib_p (SET_DEST (set), SET_SRC (set));
|
82 |
|
|
}
|
83 |
|
|
|
84 |
|
|
/* Try to simplify INSN. */
|
85 |
|
|
static void
|
86 |
|
|
reload_cse_simplify (rtx insn, rtx testreg)
|
87 |
|
|
{
|
88 |
|
|
rtx body = PATTERN (insn);
|
89 |
|
|
|
90 |
|
|
if (GET_CODE (body) == SET)
|
91 |
|
|
{
|
92 |
|
|
int count = 0;
|
93 |
|
|
|
94 |
|
|
/* Simplify even if we may think it is a no-op.
|
95 |
|
|
We may think a memory load of a value smaller than WORD_SIZE
|
96 |
|
|
is redundant because we haven't taken into account possible
|
97 |
|
|
implicit extension. reload_cse_simplify_set() will bring
|
98 |
|
|
this out, so it's safer to simplify before we delete. */
|
99 |
|
|
count += reload_cse_simplify_set (body, insn);
|
100 |
|
|
|
101 |
|
|
if (!count && reload_cse_noop_set_p (body))
|
102 |
|
|
{
|
103 |
|
|
rtx value = SET_DEST (body);
|
104 |
|
|
if (REG_P (value)
|
105 |
|
|
&& ! REG_FUNCTION_VALUE_P (value))
|
106 |
|
|
value = 0;
|
107 |
|
|
delete_insn_and_edges (insn);
|
108 |
|
|
return;
|
109 |
|
|
}
|
110 |
|
|
|
111 |
|
|
if (count > 0)
|
112 |
|
|
apply_change_group ();
|
113 |
|
|
else
|
114 |
|
|
reload_cse_simplify_operands (insn, testreg);
|
115 |
|
|
}
|
116 |
|
|
else if (GET_CODE (body) == PARALLEL)
|
117 |
|
|
{
|
118 |
|
|
int i;
|
119 |
|
|
int count = 0;
|
120 |
|
|
rtx value = NULL_RTX;
|
121 |
|
|
|
122 |
|
|
/* Registers mentioned in the clobber list for an asm cannot be reused
|
123 |
|
|
within the body of the asm. Invalidate those registers now so that
|
124 |
|
|
we don't try to substitute values for them. */
|
125 |
|
|
if (asm_noperands (body) >= 0)
|
126 |
|
|
{
|
127 |
|
|
for (i = XVECLEN (body, 0) - 1; i >= 0; --i)
|
128 |
|
|
{
|
129 |
|
|
rtx part = XVECEXP (body, 0, i);
|
130 |
|
|
if (GET_CODE (part) == CLOBBER && REG_P (XEXP (part, 0)))
|
131 |
|
|
cselib_invalidate_rtx (XEXP (part, 0));
|
132 |
|
|
}
|
133 |
|
|
}
|
134 |
|
|
|
135 |
|
|
/* If every action in a PARALLEL is a noop, we can delete
|
136 |
|
|
the entire PARALLEL. */
|
137 |
|
|
for (i = XVECLEN (body, 0) - 1; i >= 0; --i)
|
138 |
|
|
{
|
139 |
|
|
rtx part = XVECEXP (body, 0, i);
|
140 |
|
|
if (GET_CODE (part) == SET)
|
141 |
|
|
{
|
142 |
|
|
if (! reload_cse_noop_set_p (part))
|
143 |
|
|
break;
|
144 |
|
|
if (REG_P (SET_DEST (part))
|
145 |
|
|
&& REG_FUNCTION_VALUE_P (SET_DEST (part)))
|
146 |
|
|
{
|
147 |
|
|
if (value)
|
148 |
|
|
break;
|
149 |
|
|
value = SET_DEST (part);
|
150 |
|
|
}
|
151 |
|
|
}
|
152 |
|
|
else if (GET_CODE (part) != CLOBBER)
|
153 |
|
|
break;
|
154 |
|
|
}
|
155 |
|
|
|
156 |
|
|
if (i < 0)
|
157 |
|
|
{
|
158 |
|
|
delete_insn_and_edges (insn);
|
159 |
|
|
/* We're done with this insn. */
|
160 |
|
|
return;
|
161 |
|
|
}
|
162 |
|
|
|
163 |
|
|
/* It's not a no-op, but we can try to simplify it. */
|
164 |
|
|
for (i = XVECLEN (body, 0) - 1; i >= 0; --i)
|
165 |
|
|
if (GET_CODE (XVECEXP (body, 0, i)) == SET)
|
166 |
|
|
count += reload_cse_simplify_set (XVECEXP (body, 0, i), insn);
|
167 |
|
|
|
168 |
|
|
if (count > 0)
|
169 |
|
|
apply_change_group ();
|
170 |
|
|
else
|
171 |
|
|
reload_cse_simplify_operands (insn, testreg);
|
172 |
|
|
}
|
173 |
|
|
}
|
174 |
|
|
|
175 |
|
|
/* Do a very simple CSE pass over the hard registers.
|
176 |
|
|
|
177 |
|
|
This function detects no-op moves where we happened to assign two
|
178 |
|
|
different pseudo-registers to the same hard register, and then
|
179 |
|
|
copied one to the other. Reload will generate a useless
|
180 |
|
|
instruction copying a register to itself.
|
181 |
|
|
|
182 |
|
|
This function also detects cases where we load a value from memory
|
183 |
|
|
into two different registers, and (if memory is more expensive than
|
184 |
|
|
registers) changes it to simply copy the first register into the
|
185 |
|
|
second register.
|
186 |
|
|
|
187 |
|
|
Another optimization is performed that scans the operands of each
|
188 |
|
|
instruction to see whether the value is already available in a
|
189 |
|
|
hard register. It then replaces the operand with the hard register
|
190 |
|
|
if possible, much like an optional reload would. */
|
191 |
|
|
|
192 |
|
|
static void
|
193 |
|
|
reload_cse_regs_1 (rtx first)
|
194 |
|
|
{
|
195 |
|
|
rtx insn;
|
196 |
|
|
rtx testreg = gen_rtx_REG (VOIDmode, -1);
|
197 |
|
|
|
198 |
|
|
cselib_init (true);
|
199 |
|
|
init_alias_analysis ();
|
200 |
|
|
|
201 |
|
|
for (insn = first; insn; insn = NEXT_INSN (insn))
|
202 |
|
|
{
|
203 |
|
|
if (INSN_P (insn))
|
204 |
|
|
reload_cse_simplify (insn, testreg);
|
205 |
|
|
|
206 |
|
|
cselib_process_insn (insn);
|
207 |
|
|
}
|
208 |
|
|
|
209 |
|
|
/* Clean up. */
|
210 |
|
|
end_alias_analysis ();
|
211 |
|
|
cselib_finish ();
|
212 |
|
|
}
|
213 |
|
|
|
214 |
|
|
/* Try to simplify a single SET instruction. SET is the set pattern.
|
215 |
|
|
INSN is the instruction it came from.
|
216 |
|
|
This function only handles one case: if we set a register to a value
|
217 |
|
|
which is not a register, we try to find that value in some other register
|
218 |
|
|
and change the set into a register copy. */
|
219 |
|
|
|
220 |
|
|
static int
|
221 |
|
|
reload_cse_simplify_set (rtx set, rtx insn)
|
222 |
|
|
{
|
223 |
|
|
int did_change = 0;
|
224 |
|
|
int dreg;
|
225 |
|
|
rtx src;
|
226 |
|
|
enum reg_class dclass;
|
227 |
|
|
int old_cost;
|
228 |
|
|
cselib_val *val;
|
229 |
|
|
struct elt_loc_list *l;
|
230 |
|
|
#ifdef LOAD_EXTEND_OP
|
231 |
|
|
enum rtx_code extend_op = UNKNOWN;
|
232 |
|
|
#endif
|
233 |
|
|
|
234 |
|
|
dreg = true_regnum (SET_DEST (set));
|
235 |
|
|
if (dreg < 0)
|
236 |
|
|
return 0;
|
237 |
|
|
|
238 |
|
|
src = SET_SRC (set);
|
239 |
|
|
if (side_effects_p (src) || true_regnum (src) >= 0)
|
240 |
|
|
return 0;
|
241 |
|
|
|
242 |
|
|
dclass = REGNO_REG_CLASS (dreg);
|
243 |
|
|
|
244 |
|
|
#ifdef LOAD_EXTEND_OP
|
245 |
|
|
/* When replacing a memory with a register, we need to honor assumptions
|
246 |
|
|
that combine made wrt the contents of sign bits. We'll do this by
|
247 |
|
|
generating an extend instruction instead of a reg->reg copy. Thus
|
248 |
|
|
the destination must be a register that we can widen. */
|
249 |
|
|
if (MEM_P (src)
|
250 |
|
|
&& GET_MODE_BITSIZE (GET_MODE (src)) < BITS_PER_WORD
|
251 |
|
|
&& (extend_op = LOAD_EXTEND_OP (GET_MODE (src))) != UNKNOWN
|
252 |
|
|
&& !REG_P (SET_DEST (set)))
|
253 |
|
|
return 0;
|
254 |
|
|
#endif
|
255 |
|
|
|
256 |
|
|
val = cselib_lookup (src, GET_MODE (SET_DEST (set)), 0);
|
257 |
|
|
if (! val)
|
258 |
|
|
return 0;
|
259 |
|
|
|
260 |
|
|
/* If memory loads are cheaper than register copies, don't change them. */
|
261 |
|
|
if (MEM_P (src))
|
262 |
|
|
old_cost = MEMORY_MOVE_COST (GET_MODE (src), dclass, 1);
|
263 |
|
|
else if (REG_P (src))
|
264 |
|
|
old_cost = REGISTER_MOVE_COST (GET_MODE (src),
|
265 |
|
|
REGNO_REG_CLASS (REGNO (src)), dclass);
|
266 |
|
|
else
|
267 |
|
|
old_cost = rtx_cost (src, SET);
|
268 |
|
|
|
269 |
|
|
for (l = val->locs; l; l = l->next)
|
270 |
|
|
{
|
271 |
|
|
rtx this_rtx = l->loc;
|
272 |
|
|
int this_cost;
|
273 |
|
|
|
274 |
|
|
if (CONSTANT_P (this_rtx) && ! references_value_p (this_rtx, 0))
|
275 |
|
|
{
|
276 |
|
|
#ifdef LOAD_EXTEND_OP
|
277 |
|
|
if (extend_op != UNKNOWN)
|
278 |
|
|
{
|
279 |
|
|
HOST_WIDE_INT this_val;
|
280 |
|
|
|
281 |
|
|
/* ??? I'm lazy and don't wish to handle CONST_DOUBLE. Other
|
282 |
|
|
constants, such as SYMBOL_REF, cannot be extended. */
|
283 |
|
|
if (GET_CODE (this_rtx) != CONST_INT)
|
284 |
|
|
continue;
|
285 |
|
|
|
286 |
|
|
this_val = INTVAL (this_rtx);
|
287 |
|
|
switch (extend_op)
|
288 |
|
|
{
|
289 |
|
|
case ZERO_EXTEND:
|
290 |
|
|
this_val &= GET_MODE_MASK (GET_MODE (src));
|
291 |
|
|
break;
|
292 |
|
|
case SIGN_EXTEND:
|
293 |
|
|
/* ??? In theory we're already extended. */
|
294 |
|
|
if (this_val == trunc_int_for_mode (this_val, GET_MODE (src)))
|
295 |
|
|
break;
|
296 |
|
|
default:
|
297 |
|
|
gcc_unreachable ();
|
298 |
|
|
}
|
299 |
|
|
this_rtx = GEN_INT (this_val);
|
300 |
|
|
}
|
301 |
|
|
#endif
|
302 |
|
|
this_cost = rtx_cost (this_rtx, SET);
|
303 |
|
|
}
|
304 |
|
|
else if (REG_P (this_rtx))
|
305 |
|
|
{
|
306 |
|
|
#ifdef LOAD_EXTEND_OP
|
307 |
|
|
if (extend_op != UNKNOWN)
|
308 |
|
|
{
|
309 |
|
|
this_rtx = gen_rtx_fmt_e (extend_op, word_mode, this_rtx);
|
310 |
|
|
this_cost = rtx_cost (this_rtx, SET);
|
311 |
|
|
}
|
312 |
|
|
else
|
313 |
|
|
#endif
|
314 |
|
|
this_cost = REGISTER_MOVE_COST (GET_MODE (this_rtx),
|
315 |
|
|
REGNO_REG_CLASS (REGNO (this_rtx)),
|
316 |
|
|
dclass);
|
317 |
|
|
}
|
318 |
|
|
else
|
319 |
|
|
continue;
|
320 |
|
|
|
321 |
|
|
/* If equal costs, prefer registers over anything else. That
|
322 |
|
|
tends to lead to smaller instructions on some machines. */
|
323 |
|
|
if (this_cost < old_cost
|
324 |
|
|
|| (this_cost == old_cost
|
325 |
|
|
&& REG_P (this_rtx)
|
326 |
|
|
&& !REG_P (SET_SRC (set))))
|
327 |
|
|
{
|
328 |
|
|
#ifdef LOAD_EXTEND_OP
|
329 |
|
|
if (GET_MODE_BITSIZE (GET_MODE (SET_DEST (set))) < BITS_PER_WORD
|
330 |
|
|
&& extend_op != UNKNOWN
|
331 |
|
|
#ifdef CANNOT_CHANGE_MODE_CLASS
|
332 |
|
|
&& !CANNOT_CHANGE_MODE_CLASS (GET_MODE (SET_DEST (set)),
|
333 |
|
|
word_mode,
|
334 |
|
|
REGNO_REG_CLASS (REGNO (SET_DEST (set))))
|
335 |
|
|
#endif
|
336 |
|
|
)
|
337 |
|
|
{
|
338 |
|
|
rtx wide_dest = gen_rtx_REG (word_mode, REGNO (SET_DEST (set)));
|
339 |
|
|
ORIGINAL_REGNO (wide_dest) = ORIGINAL_REGNO (SET_DEST (set));
|
340 |
|
|
validate_change (insn, &SET_DEST (set), wide_dest, 1);
|
341 |
|
|
}
|
342 |
|
|
#endif
|
343 |
|
|
|
344 |
|
|
validate_change (insn, &SET_SRC (set), copy_rtx (this_rtx), 1);
|
345 |
|
|
old_cost = this_cost, did_change = 1;
|
346 |
|
|
}
|
347 |
|
|
}
|
348 |
|
|
|
349 |
|
|
return did_change;
|
350 |
|
|
}
|
351 |
|
|
|
352 |
|
|
/* Try to replace operands in INSN with equivalent values that are already
|
353 |
|
|
in registers. This can be viewed as optional reloading.
|
354 |
|
|
|
355 |
|
|
For each non-register operand in the insn, see if any hard regs are
|
356 |
|
|
known to be equivalent to that operand. Record the alternatives which
|
357 |
|
|
can accept these hard registers. Among all alternatives, select the
|
358 |
|
|
ones which are better or equal to the one currently matching, where
|
359 |
|
|
"better" is in terms of '?' and '!' constraints. Among the remaining
|
360 |
|
|
alternatives, select the one which replaces most operands with
|
361 |
|
|
hard registers. */
|
362 |
|
|
|
363 |
|
|
static int
|
364 |
|
|
reload_cse_simplify_operands (rtx insn, rtx testreg)
|
365 |
|
|
{
|
366 |
|
|
int i, j;
|
367 |
|
|
|
368 |
|
|
/* For each operand, all registers that are equivalent to it. */
|
369 |
|
|
HARD_REG_SET equiv_regs[MAX_RECOG_OPERANDS];
|
370 |
|
|
|
371 |
|
|
const char *constraints[MAX_RECOG_OPERANDS];
|
372 |
|
|
|
373 |
|
|
/* Vector recording how bad an alternative is. */
|
374 |
|
|
int *alternative_reject;
|
375 |
|
|
/* Vector recording how many registers can be introduced by choosing
|
376 |
|
|
this alternative. */
|
377 |
|
|
int *alternative_nregs;
|
378 |
|
|
/* Array of vectors recording, for each operand and each alternative,
|
379 |
|
|
which hard register to substitute, or -1 if the operand should be
|
380 |
|
|
left as it is. */
|
381 |
|
|
int *op_alt_regno[MAX_RECOG_OPERANDS];
|
382 |
|
|
/* Array of alternatives, sorted in order of decreasing desirability. */
|
383 |
|
|
int *alternative_order;
|
384 |
|
|
|
385 |
|
|
extract_insn (insn);
|
386 |
|
|
|
387 |
|
|
if (recog_data.n_alternatives == 0 || recog_data.n_operands == 0)
|
388 |
|
|
return 0;
|
389 |
|
|
|
390 |
|
|
/* Figure out which alternative currently matches. */
|
391 |
|
|
if (! constrain_operands (1))
|
392 |
|
|
fatal_insn_not_found (insn);
|
393 |
|
|
|
394 |
|
|
alternative_reject = alloca (recog_data.n_alternatives * sizeof (int));
|
395 |
|
|
alternative_nregs = alloca (recog_data.n_alternatives * sizeof (int));
|
396 |
|
|
alternative_order = alloca (recog_data.n_alternatives * sizeof (int));
|
397 |
|
|
memset (alternative_reject, 0, recog_data.n_alternatives * sizeof (int));
|
398 |
|
|
memset (alternative_nregs, 0, recog_data.n_alternatives * sizeof (int));
|
399 |
|
|
|
400 |
|
|
/* For each operand, find out which regs are equivalent. */
|
401 |
|
|
for (i = 0; i < recog_data.n_operands; i++)
|
402 |
|
|
{
|
403 |
|
|
cselib_val *v;
|
404 |
|
|
struct elt_loc_list *l;
|
405 |
|
|
rtx op;
|
406 |
|
|
enum machine_mode mode;
|
407 |
|
|
|
408 |
|
|
CLEAR_HARD_REG_SET (equiv_regs[i]);
|
409 |
|
|
|
410 |
|
|
/* cselib blows up on CODE_LABELs. Trying to fix that doesn't seem
|
411 |
|
|
right, so avoid the problem here. Likewise if we have a constant
|
412 |
|
|
and the insn pattern doesn't tell us the mode we need. */
|
413 |
|
|
if (LABEL_P (recog_data.operand[i])
|
414 |
|
|
|| (CONSTANT_P (recog_data.operand[i])
|
415 |
|
|
&& recog_data.operand_mode[i] == VOIDmode))
|
416 |
|
|
continue;
|
417 |
|
|
|
418 |
|
|
op = recog_data.operand[i];
|
419 |
|
|
mode = GET_MODE (op);
|
420 |
|
|
#ifdef LOAD_EXTEND_OP
|
421 |
|
|
if (MEM_P (op)
|
422 |
|
|
&& GET_MODE_BITSIZE (mode) < BITS_PER_WORD
|
423 |
|
|
&& LOAD_EXTEND_OP (mode) != UNKNOWN)
|
424 |
|
|
{
|
425 |
|
|
rtx set = single_set (insn);
|
426 |
|
|
|
427 |
|
|
/* We might have multiple sets, some of which do implicit
|
428 |
|
|
extension. Punt on this for now. */
|
429 |
|
|
if (! set)
|
430 |
|
|
continue;
|
431 |
|
|
/* If the destination is also a MEM or a STRICT_LOW_PART, no
|
432 |
|
|
extension applies.
|
433 |
|
|
Also, if there is an explicit extension, we don't have to
|
434 |
|
|
worry about an implicit one. */
|
435 |
|
|
else if (MEM_P (SET_DEST (set))
|
436 |
|
|
|| GET_CODE (SET_DEST (set)) == STRICT_LOW_PART
|
437 |
|
|
|| GET_CODE (SET_SRC (set)) == ZERO_EXTEND
|
438 |
|
|
|| GET_CODE (SET_SRC (set)) == SIGN_EXTEND)
|
439 |
|
|
; /* Continue ordinary processing. */
|
440 |
|
|
#ifdef CANNOT_CHANGE_MODE_CLASS
|
441 |
|
|
/* If the register cannot change mode to word_mode, it follows that
|
442 |
|
|
it cannot have been used in word_mode. */
|
443 |
|
|
else if (REG_P (SET_DEST (set))
|
444 |
|
|
&& CANNOT_CHANGE_MODE_CLASS (GET_MODE (SET_DEST (set)),
|
445 |
|
|
word_mode,
|
446 |
|
|
REGNO_REG_CLASS (REGNO (SET_DEST (set)))))
|
447 |
|
|
; /* Continue ordinary processing. */
|
448 |
|
|
#endif
|
449 |
|
|
/* If this is a straight load, make the extension explicit. */
|
450 |
|
|
else if (REG_P (SET_DEST (set))
|
451 |
|
|
&& recog_data.n_operands == 2
|
452 |
|
|
&& SET_SRC (set) == op
|
453 |
|
|
&& SET_DEST (set) == recog_data.operand[1-i])
|
454 |
|
|
{
|
455 |
|
|
validate_change (insn, recog_data.operand_loc[i],
|
456 |
|
|
gen_rtx_fmt_e (LOAD_EXTEND_OP (mode),
|
457 |
|
|
word_mode, op),
|
458 |
|
|
1);
|
459 |
|
|
validate_change (insn, recog_data.operand_loc[1-i],
|
460 |
|
|
gen_rtx_REG (word_mode, REGNO (SET_DEST (set))),
|
461 |
|
|
1);
|
462 |
|
|
if (! apply_change_group ())
|
463 |
|
|
return 0;
|
464 |
|
|
return reload_cse_simplify_operands (insn, testreg);
|
465 |
|
|
}
|
466 |
|
|
else
|
467 |
|
|
/* ??? There might be arithmetic operations with memory that are
|
468 |
|
|
safe to optimize, but is it worth the trouble? */
|
469 |
|
|
continue;
|
470 |
|
|
}
|
471 |
|
|
#endif /* LOAD_EXTEND_OP */
|
472 |
|
|
v = cselib_lookup (op, recog_data.operand_mode[i], 0);
|
473 |
|
|
if (! v)
|
474 |
|
|
continue;
|
475 |
|
|
|
476 |
|
|
for (l = v->locs; l; l = l->next)
|
477 |
|
|
if (REG_P (l->loc))
|
478 |
|
|
SET_HARD_REG_BIT (equiv_regs[i], REGNO (l->loc));
|
479 |
|
|
}
|
480 |
|
|
|
481 |
|
|
for (i = 0; i < recog_data.n_operands; i++)
|
482 |
|
|
{
|
483 |
|
|
enum machine_mode mode;
|
484 |
|
|
int regno;
|
485 |
|
|
const char *p;
|
486 |
|
|
|
487 |
|
|
op_alt_regno[i] = alloca (recog_data.n_alternatives * sizeof (int));
|
488 |
|
|
for (j = 0; j < recog_data.n_alternatives; j++)
|
489 |
|
|
op_alt_regno[i][j] = -1;
|
490 |
|
|
|
491 |
|
|
p = constraints[i] = recog_data.constraints[i];
|
492 |
|
|
mode = recog_data.operand_mode[i];
|
493 |
|
|
|
494 |
|
|
/* Add the reject values for each alternative given by the constraints
|
495 |
|
|
for this operand. */
|
496 |
|
|
j = 0;
|
497 |
|
|
while (*p != '\0')
|
498 |
|
|
{
|
499 |
|
|
char c = *p++;
|
500 |
|
|
if (c == ',')
|
501 |
|
|
j++;
|
502 |
|
|
else if (c == '?')
|
503 |
|
|
alternative_reject[j] += 3;
|
504 |
|
|
else if (c == '!')
|
505 |
|
|
alternative_reject[j] += 300;
|
506 |
|
|
}
|
507 |
|
|
|
508 |
|
|
/* We won't change operands which are already registers. We
|
509 |
|
|
also don't want to modify output operands. */
|
510 |
|
|
regno = true_regnum (recog_data.operand[i]);
|
511 |
|
|
if (regno >= 0
|
512 |
|
|
|| constraints[i][0] == '='
|
513 |
|
|
|| constraints[i][0] == '+')
|
514 |
|
|
continue;
|
515 |
|
|
|
516 |
|
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
517 |
|
|
{
|
518 |
|
|
int class = (int) NO_REGS;
|
519 |
|
|
|
520 |
|
|
if (! TEST_HARD_REG_BIT (equiv_regs[i], regno))
|
521 |
|
|
continue;
|
522 |
|
|
|
523 |
|
|
REGNO (testreg) = regno;
|
524 |
|
|
PUT_MODE (testreg, mode);
|
525 |
|
|
|
526 |
|
|
/* We found a register equal to this operand. Now look for all
|
527 |
|
|
alternatives that can accept this register and have not been
|
528 |
|
|
assigned a register they can use yet. */
|
529 |
|
|
j = 0;
|
530 |
|
|
p = constraints[i];
|
531 |
|
|
for (;;)
|
532 |
|
|
{
|
533 |
|
|
char c = *p;
|
534 |
|
|
|
535 |
|
|
switch (c)
|
536 |
|
|
{
|
537 |
|
|
case '=': case '+': case '?':
|
538 |
|
|
case '#': case '&': case '!':
|
539 |
|
|
case '*': case '%':
|
540 |
|
|
case '0': case '1': case '2': case '3': case '4':
|
541 |
|
|
case '5': case '6': case '7': case '8': case '9':
|
542 |
|
|
case 'm': case '<': case '>': case 'V': case 'o':
|
543 |
|
|
case 'E': case 'F': case 'G': case 'H':
|
544 |
|
|
case 's': case 'i': case 'n':
|
545 |
|
|
case 'I': case 'J': case 'K': case 'L':
|
546 |
|
|
case 'M': case 'N': case 'O': case 'P':
|
547 |
|
|
case 'p': case 'X':
|
548 |
|
|
/* These don't say anything we care about. */
|
549 |
|
|
break;
|
550 |
|
|
|
551 |
|
|
case 'g': case 'r':
|
552 |
|
|
class = reg_class_subunion[(int) class][(int) GENERAL_REGS];
|
553 |
|
|
break;
|
554 |
|
|
|
555 |
|
|
default:
|
556 |
|
|
class
|
557 |
|
|
= (reg_class_subunion
|
558 |
|
|
[(int) class]
|
559 |
|
|
[(int) REG_CLASS_FROM_CONSTRAINT ((unsigned char) c, p)]);
|
560 |
|
|
break;
|
561 |
|
|
|
562 |
|
|
case ',': case '\0':
|
563 |
|
|
/* See if REGNO fits this alternative, and set it up as the
|
564 |
|
|
replacement register if we don't have one for this
|
565 |
|
|
alternative yet and the operand being replaced is not
|
566 |
|
|
a cheap CONST_INT. */
|
567 |
|
|
if (op_alt_regno[i][j] == -1
|
568 |
|
|
&& reg_fits_class_p (testreg, class, 0, mode)
|
569 |
|
|
&& (GET_CODE (recog_data.operand[i]) != CONST_INT
|
570 |
|
|
|| (rtx_cost (recog_data.operand[i], SET)
|
571 |
|
|
> rtx_cost (testreg, SET))))
|
572 |
|
|
{
|
573 |
|
|
alternative_nregs[j]++;
|
574 |
|
|
op_alt_regno[i][j] = regno;
|
575 |
|
|
}
|
576 |
|
|
j++;
|
577 |
|
|
class = (int) NO_REGS;
|
578 |
|
|
break;
|
579 |
|
|
}
|
580 |
|
|
p += CONSTRAINT_LEN (c, p);
|
581 |
|
|
|
582 |
|
|
if (c == '\0')
|
583 |
|
|
break;
|
584 |
|
|
}
|
585 |
|
|
}
|
586 |
|
|
}
|
587 |
|
|
|
588 |
|
|
/* Record all alternatives which are better or equal to the currently
|
589 |
|
|
matching one in the alternative_order array. */
|
590 |
|
|
for (i = j = 0; i < recog_data.n_alternatives; i++)
|
591 |
|
|
if (alternative_reject[i] <= alternative_reject[which_alternative])
|
592 |
|
|
alternative_order[j++] = i;
|
593 |
|
|
recog_data.n_alternatives = j;
|
594 |
|
|
|
595 |
|
|
/* Sort it. Given a small number of alternatives, a dumb algorithm
|
596 |
|
|
won't hurt too much. */
|
597 |
|
|
for (i = 0; i < recog_data.n_alternatives - 1; i++)
|
598 |
|
|
{
|
599 |
|
|
int best = i;
|
600 |
|
|
int best_reject = alternative_reject[alternative_order[i]];
|
601 |
|
|
int best_nregs = alternative_nregs[alternative_order[i]];
|
602 |
|
|
int tmp;
|
603 |
|
|
|
604 |
|
|
for (j = i + 1; j < recog_data.n_alternatives; j++)
|
605 |
|
|
{
|
606 |
|
|
int this_reject = alternative_reject[alternative_order[j]];
|
607 |
|
|
int this_nregs = alternative_nregs[alternative_order[j]];
|
608 |
|
|
|
609 |
|
|
if (this_reject < best_reject
|
610 |
|
|
|| (this_reject == best_reject && this_nregs > best_nregs))
|
611 |
|
|
{
|
612 |
|
|
best = j;
|
613 |
|
|
best_reject = this_reject;
|
614 |
|
|
best_nregs = this_nregs;
|
615 |
|
|
}
|
616 |
|
|
}
|
617 |
|
|
|
618 |
|
|
tmp = alternative_order[best];
|
619 |
|
|
alternative_order[best] = alternative_order[i];
|
620 |
|
|
alternative_order[i] = tmp;
|
621 |
|
|
}
|
622 |
|
|
|
623 |
|
|
/* Substitute the operands as determined by op_alt_regno for the best
|
624 |
|
|
alternative. */
|
625 |
|
|
j = alternative_order[0];
|
626 |
|
|
|
627 |
|
|
for (i = 0; i < recog_data.n_operands; i++)
|
628 |
|
|
{
|
629 |
|
|
enum machine_mode mode = recog_data.operand_mode[i];
|
630 |
|
|
if (op_alt_regno[i][j] == -1)
|
631 |
|
|
continue;
|
632 |
|
|
|
633 |
|
|
validate_change (insn, recog_data.operand_loc[i],
|
634 |
|
|
gen_rtx_REG (mode, op_alt_regno[i][j]), 1);
|
635 |
|
|
}
|
636 |
|
|
|
637 |
|
|
for (i = recog_data.n_dups - 1; i >= 0; i--)
|
638 |
|
|
{
|
639 |
|
|
int op = recog_data.dup_num[i];
|
640 |
|
|
enum machine_mode mode = recog_data.operand_mode[op];
|
641 |
|
|
|
642 |
|
|
if (op_alt_regno[op][j] == -1)
|
643 |
|
|
continue;
|
644 |
|
|
|
645 |
|
|
validate_change (insn, recog_data.dup_loc[i],
|
646 |
|
|
gen_rtx_REG (mode, op_alt_regno[op][j]), 1);
|
647 |
|
|
}
|
648 |
|
|
|
649 |
|
|
return apply_change_group ();
|
650 |
|
|
}
|
651 |
|
|
|
652 |
|
|
/* If reload couldn't use reg+reg+offset addressing, try to use reg+reg
|
653 |
|
|
addressing now.
|
654 |
|
|
This code might also be useful when reload gave up on reg+reg addressing
|
655 |
|
|
because of clashes between the return register and INDEX_REG_CLASS. */
|
656 |
|
|
|
657 |
|
|
/* The maximum number of uses of a register we can keep track of to
|
658 |
|
|
replace them with reg+reg addressing. */
|
659 |
|
|
#define RELOAD_COMBINE_MAX_USES 6
|
660 |
|
|
|
661 |
|
|
/* INSN is the insn where a register has ben used, and USEP points to the
|
662 |
|
|
location of the register within the rtl. */
|
663 |
|
|
struct reg_use { rtx insn, *usep; };
|
664 |
|
|
|
665 |
|
|
/* If the register is used in some unknown fashion, USE_INDEX is negative.
|
666 |
|
|
If it is dead, USE_INDEX is RELOAD_COMBINE_MAX_USES, and STORE_RUID
|
667 |
|
|
indicates where it becomes live again.
|
668 |
|
|
Otherwise, USE_INDEX is the index of the last encountered use of the
|
669 |
|
|
register (which is first among these we have seen since we scan backwards),
|
670 |
|
|
OFFSET contains the constant offset that is added to the register in
|
671 |
|
|
all encountered uses, and USE_RUID indicates the first encountered, i.e.
|
672 |
|
|
last, of these uses.
|
673 |
|
|
STORE_RUID is always meaningful if we only want to use a value in a
|
674 |
|
|
register in a different place: it denotes the next insn in the insn
|
675 |
|
|
stream (i.e. the last encountered) that sets or clobbers the register. */
|
676 |
|
|
static struct
|
677 |
|
|
{
|
678 |
|
|
struct reg_use reg_use[RELOAD_COMBINE_MAX_USES];
|
679 |
|
|
int use_index;
|
680 |
|
|
rtx offset;
|
681 |
|
|
int store_ruid;
|
682 |
|
|
int use_ruid;
|
683 |
|
|
} reg_state[FIRST_PSEUDO_REGISTER];
|
684 |
|
|
|
685 |
|
|
/* Reverse linear uid. This is increased in reload_combine while scanning
|
686 |
|
|
the instructions from last to first. It is used to set last_label_ruid
|
687 |
|
|
and the store_ruid / use_ruid fields in reg_state. */
|
688 |
|
|
static int reload_combine_ruid;
|
689 |
|
|
|
690 |
|
|
#define LABEL_LIVE(LABEL) \
|
691 |
|
|
(label_live[CODE_LABEL_NUMBER (LABEL) - min_labelno])
|
692 |
|
|
|
693 |
|
|
static void
|
694 |
|
|
reload_combine (void)
|
695 |
|
|
{
|
696 |
|
|
rtx insn, set;
|
697 |
|
|
int first_index_reg = -1;
|
698 |
|
|
int last_index_reg = 0;
|
699 |
|
|
int i;
|
700 |
|
|
basic_block bb;
|
701 |
|
|
unsigned int r;
|
702 |
|
|
int last_label_ruid;
|
703 |
|
|
int min_labelno, n_labels;
|
704 |
|
|
HARD_REG_SET ever_live_at_start, *label_live;
|
705 |
|
|
|
706 |
|
|
/* If reg+reg can be used in offsetable memory addresses, the main chunk of
|
707 |
|
|
reload has already used it where appropriate, so there is no use in
|
708 |
|
|
trying to generate it now. */
|
709 |
|
|
if (double_reg_address_ok && INDEX_REG_CLASS != NO_REGS)
|
710 |
|
|
return;
|
711 |
|
|
|
712 |
|
|
/* To avoid wasting too much time later searching for an index register,
|
713 |
|
|
determine the minimum and maximum index register numbers. */
|
714 |
|
|
for (r = 0; r < FIRST_PSEUDO_REGISTER; r++)
|
715 |
|
|
if (TEST_HARD_REG_BIT (reg_class_contents[INDEX_REG_CLASS], r))
|
716 |
|
|
{
|
717 |
|
|
if (first_index_reg == -1)
|
718 |
|
|
first_index_reg = r;
|
719 |
|
|
|
720 |
|
|
last_index_reg = r;
|
721 |
|
|
}
|
722 |
|
|
|
723 |
|
|
/* If no index register is available, we can quit now. */
|
724 |
|
|
if (first_index_reg == -1)
|
725 |
|
|
return;
|
726 |
|
|
|
727 |
|
|
/* Set up LABEL_LIVE and EVER_LIVE_AT_START. The register lifetime
|
728 |
|
|
information is a bit fuzzy immediately after reload, but it's
|
729 |
|
|
still good enough to determine which registers are live at a jump
|
730 |
|
|
destination. */
|
731 |
|
|
min_labelno = get_first_label_num ();
|
732 |
|
|
n_labels = max_label_num () - min_labelno;
|
733 |
|
|
label_live = XNEWVEC (HARD_REG_SET, n_labels);
|
734 |
|
|
CLEAR_HARD_REG_SET (ever_live_at_start);
|
735 |
|
|
|
736 |
|
|
FOR_EACH_BB_REVERSE (bb)
|
737 |
|
|
{
|
738 |
|
|
insn = BB_HEAD (bb);
|
739 |
|
|
if (LABEL_P (insn))
|
740 |
|
|
{
|
741 |
|
|
HARD_REG_SET live;
|
742 |
|
|
|
743 |
|
|
REG_SET_TO_HARD_REG_SET (live,
|
744 |
|
|
bb->il.rtl->global_live_at_start);
|
745 |
|
|
compute_use_by_pseudos (&live,
|
746 |
|
|
bb->il.rtl->global_live_at_start);
|
747 |
|
|
COPY_HARD_REG_SET (LABEL_LIVE (insn), live);
|
748 |
|
|
IOR_HARD_REG_SET (ever_live_at_start, live);
|
749 |
|
|
}
|
750 |
|
|
}
|
751 |
|
|
|
752 |
|
|
/* Initialize last_label_ruid, reload_combine_ruid and reg_state. */
|
753 |
|
|
last_label_ruid = reload_combine_ruid = 0;
|
754 |
|
|
for (r = 0; r < FIRST_PSEUDO_REGISTER; r++)
|
755 |
|
|
{
|
756 |
|
|
reg_state[r].store_ruid = reload_combine_ruid;
|
757 |
|
|
if (fixed_regs[r])
|
758 |
|
|
reg_state[r].use_index = -1;
|
759 |
|
|
else
|
760 |
|
|
reg_state[r].use_index = RELOAD_COMBINE_MAX_USES;
|
761 |
|
|
}
|
762 |
|
|
|
763 |
|
|
for (insn = get_last_insn (); insn; insn = PREV_INSN (insn))
|
764 |
|
|
{
|
765 |
|
|
rtx note;
|
766 |
|
|
|
767 |
|
|
/* We cannot do our optimization across labels. Invalidating all the use
|
768 |
|
|
information we have would be costly, so we just note where the label
|
769 |
|
|
is and then later disable any optimization that would cross it. */
|
770 |
|
|
if (LABEL_P (insn))
|
771 |
|
|
last_label_ruid = reload_combine_ruid;
|
772 |
|
|
else if (BARRIER_P (insn))
|
773 |
|
|
for (r = 0; r < FIRST_PSEUDO_REGISTER; r++)
|
774 |
|
|
if (! fixed_regs[r])
|
775 |
|
|
reg_state[r].use_index = RELOAD_COMBINE_MAX_USES;
|
776 |
|
|
|
777 |
|
|
if (! INSN_P (insn))
|
778 |
|
|
continue;
|
779 |
|
|
|
780 |
|
|
reload_combine_ruid++;
|
781 |
|
|
|
782 |
|
|
/* Look for (set (REGX) (CONST_INT))
|
783 |
|
|
(set (REGX) (PLUS (REGX) (REGY)))
|
784 |
|
|
...
|
785 |
|
|
... (MEM (REGX)) ...
|
786 |
|
|
and convert it to
|
787 |
|
|
(set (REGZ) (CONST_INT))
|
788 |
|
|
...
|
789 |
|
|
... (MEM (PLUS (REGZ) (REGY)))... .
|
790 |
|
|
|
791 |
|
|
First, check that we have (set (REGX) (PLUS (REGX) (REGY)))
|
792 |
|
|
and that we know all uses of REGX before it dies.
|
793 |
|
|
Also, explicitly check that REGX != REGY; our life information
|
794 |
|
|
does not yet show whether REGY changes in this insn. */
|
795 |
|
|
set = single_set (insn);
|
796 |
|
|
if (set != NULL_RTX
|
797 |
|
|
&& REG_P (SET_DEST (set))
|
798 |
|
|
&& (hard_regno_nregs[REGNO (SET_DEST (set))]
|
799 |
|
|
[GET_MODE (SET_DEST (set))]
|
800 |
|
|
== 1)
|
801 |
|
|
&& GET_CODE (SET_SRC (set)) == PLUS
|
802 |
|
|
&& REG_P (XEXP (SET_SRC (set), 1))
|
803 |
|
|
&& rtx_equal_p (XEXP (SET_SRC (set), 0), SET_DEST (set))
|
804 |
|
|
&& !rtx_equal_p (XEXP (SET_SRC (set), 1), SET_DEST (set))
|
805 |
|
|
&& last_label_ruid < reg_state[REGNO (SET_DEST (set))].use_ruid)
|
806 |
|
|
{
|
807 |
|
|
rtx reg = SET_DEST (set);
|
808 |
|
|
rtx plus = SET_SRC (set);
|
809 |
|
|
rtx base = XEXP (plus, 1);
|
810 |
|
|
rtx prev = prev_nonnote_insn (insn);
|
811 |
|
|
rtx prev_set = prev ? single_set (prev) : NULL_RTX;
|
812 |
|
|
unsigned int regno = REGNO (reg);
|
813 |
|
|
rtx const_reg = NULL_RTX;
|
814 |
|
|
rtx reg_sum = NULL_RTX;
|
815 |
|
|
|
816 |
|
|
/* Now, we need an index register.
|
817 |
|
|
We'll set index_reg to this index register, const_reg to the
|
818 |
|
|
register that is to be loaded with the constant
|
819 |
|
|
(denoted as REGZ in the substitution illustration above),
|
820 |
|
|
and reg_sum to the register-register that we want to use to
|
821 |
|
|
substitute uses of REG (typically in MEMs) with.
|
822 |
|
|
First check REG and BASE for being index registers;
|
823 |
|
|
we can use them even if they are not dead. */
|
824 |
|
|
if (TEST_HARD_REG_BIT (reg_class_contents[INDEX_REG_CLASS], regno)
|
825 |
|
|
|| TEST_HARD_REG_BIT (reg_class_contents[INDEX_REG_CLASS],
|
826 |
|
|
REGNO (base)))
|
827 |
|
|
{
|
828 |
|
|
const_reg = reg;
|
829 |
|
|
reg_sum = plus;
|
830 |
|
|
}
|
831 |
|
|
else
|
832 |
|
|
{
|
833 |
|
|
/* Otherwise, look for a free index register. Since we have
|
834 |
|
|
checked above that neither REG nor BASE are index registers,
|
835 |
|
|
if we find anything at all, it will be different from these
|
836 |
|
|
two registers. */
|
837 |
|
|
for (i = first_index_reg; i <= last_index_reg; i++)
|
838 |
|
|
{
|
839 |
|
|
if (TEST_HARD_REG_BIT (reg_class_contents[INDEX_REG_CLASS],
|
840 |
|
|
i)
|
841 |
|
|
&& reg_state[i].use_index == RELOAD_COMBINE_MAX_USES
|
842 |
|
|
&& reg_state[i].store_ruid <= reg_state[regno].use_ruid
|
843 |
|
|
&& hard_regno_nregs[i][GET_MODE (reg)] == 1)
|
844 |
|
|
{
|
845 |
|
|
rtx index_reg = gen_rtx_REG (GET_MODE (reg), i);
|
846 |
|
|
|
847 |
|
|
const_reg = index_reg;
|
848 |
|
|
reg_sum = gen_rtx_PLUS (GET_MODE (reg), index_reg, base);
|
849 |
|
|
break;
|
850 |
|
|
}
|
851 |
|
|
}
|
852 |
|
|
}
|
853 |
|
|
|
854 |
|
|
/* Check that PREV_SET is indeed (set (REGX) (CONST_INT)) and that
|
855 |
|
|
(REGY), i.e. BASE, is not clobbered before the last use we'll
|
856 |
|
|
create. */
|
857 |
|
|
if (prev_set != 0
|
858 |
|
|
&& GET_CODE (SET_SRC (prev_set)) == CONST_INT
|
859 |
|
|
&& rtx_equal_p (SET_DEST (prev_set), reg)
|
860 |
|
|
&& reg_state[regno].use_index >= 0
|
861 |
|
|
&& (reg_state[REGNO (base)].store_ruid
|
862 |
|
|
<= reg_state[regno].use_ruid)
|
863 |
|
|
&& reg_sum != 0)
|
864 |
|
|
{
|
865 |
|
|
int i;
|
866 |
|
|
|
867 |
|
|
/* Change destination register and, if necessary, the
|
868 |
|
|
constant value in PREV, the constant loading instruction. */
|
869 |
|
|
validate_change (prev, &SET_DEST (prev_set), const_reg, 1);
|
870 |
|
|
if (reg_state[regno].offset != const0_rtx)
|
871 |
|
|
validate_change (prev,
|
872 |
|
|
&SET_SRC (prev_set),
|
873 |
|
|
GEN_INT (INTVAL (SET_SRC (prev_set))
|
874 |
|
|
+ INTVAL (reg_state[regno].offset)),
|
875 |
|
|
1);
|
876 |
|
|
|
877 |
|
|
/* Now for every use of REG that we have recorded, replace REG
|
878 |
|
|
with REG_SUM. */
|
879 |
|
|
for (i = reg_state[regno].use_index;
|
880 |
|
|
i < RELOAD_COMBINE_MAX_USES; i++)
|
881 |
|
|
validate_change (reg_state[regno].reg_use[i].insn,
|
882 |
|
|
reg_state[regno].reg_use[i].usep,
|
883 |
|
|
/* Each change must have its own
|
884 |
|
|
replacement. */
|
885 |
|
|
copy_rtx (reg_sum), 1);
|
886 |
|
|
|
887 |
|
|
if (apply_change_group ())
|
888 |
|
|
{
|
889 |
|
|
rtx *np;
|
890 |
|
|
|
891 |
|
|
/* Delete the reg-reg addition. */
|
892 |
|
|
delete_insn (insn);
|
893 |
|
|
|
894 |
|
|
if (reg_state[regno].offset != const0_rtx)
|
895 |
|
|
/* Previous REG_EQUIV / REG_EQUAL notes for PREV
|
896 |
|
|
are now invalid. */
|
897 |
|
|
for (np = ®_NOTES (prev); *np;)
|
898 |
|
|
{
|
899 |
|
|
if (REG_NOTE_KIND (*np) == REG_EQUAL
|
900 |
|
|
|| REG_NOTE_KIND (*np) == REG_EQUIV)
|
901 |
|
|
*np = XEXP (*np, 1);
|
902 |
|
|
else
|
903 |
|
|
np = &XEXP (*np, 1);
|
904 |
|
|
}
|
905 |
|
|
|
906 |
|
|
reg_state[regno].use_index = RELOAD_COMBINE_MAX_USES;
|
907 |
|
|
reg_state[REGNO (const_reg)].store_ruid
|
908 |
|
|
= reload_combine_ruid;
|
909 |
|
|
continue;
|
910 |
|
|
}
|
911 |
|
|
}
|
912 |
|
|
}
|
913 |
|
|
|
914 |
|
|
note_stores (PATTERN (insn), reload_combine_note_store, NULL);
|
915 |
|
|
|
916 |
|
|
if (CALL_P (insn))
|
917 |
|
|
{
|
918 |
|
|
rtx link;
|
919 |
|
|
|
920 |
|
|
for (r = 0; r < FIRST_PSEUDO_REGISTER; r++)
|
921 |
|
|
if (call_used_regs[r])
|
922 |
|
|
{
|
923 |
|
|
reg_state[r].use_index = RELOAD_COMBINE_MAX_USES;
|
924 |
|
|
reg_state[r].store_ruid = reload_combine_ruid;
|
925 |
|
|
}
|
926 |
|
|
|
927 |
|
|
for (link = CALL_INSN_FUNCTION_USAGE (insn); link;
|
928 |
|
|
link = XEXP (link, 1))
|
929 |
|
|
{
|
930 |
|
|
rtx usage_rtx = XEXP (XEXP (link, 0), 0);
|
931 |
|
|
if (REG_P (usage_rtx))
|
932 |
|
|
{
|
933 |
|
|
unsigned int i;
|
934 |
|
|
unsigned int start_reg = REGNO (usage_rtx);
|
935 |
|
|
unsigned int num_regs =
|
936 |
|
|
hard_regno_nregs[start_reg][GET_MODE (usage_rtx)];
|
937 |
|
|
unsigned int end_reg = start_reg + num_regs - 1;
|
938 |
|
|
for (i = start_reg; i <= end_reg; i++)
|
939 |
|
|
if (GET_CODE (XEXP (link, 0)) == CLOBBER)
|
940 |
|
|
{
|
941 |
|
|
reg_state[i].use_index = RELOAD_COMBINE_MAX_USES;
|
942 |
|
|
reg_state[i].store_ruid = reload_combine_ruid;
|
943 |
|
|
}
|
944 |
|
|
else
|
945 |
|
|
reg_state[i].use_index = -1;
|
946 |
|
|
}
|
947 |
|
|
}
|
948 |
|
|
|
949 |
|
|
}
|
950 |
|
|
else if (JUMP_P (insn)
|
951 |
|
|
&& GET_CODE (PATTERN (insn)) != RETURN)
|
952 |
|
|
{
|
953 |
|
|
/* Non-spill registers might be used at the call destination in
|
954 |
|
|
some unknown fashion, so we have to mark the unknown use. */
|
955 |
|
|
HARD_REG_SET *live;
|
956 |
|
|
|
957 |
|
|
if ((condjump_p (insn) || condjump_in_parallel_p (insn))
|
958 |
|
|
&& JUMP_LABEL (insn))
|
959 |
|
|
live = &LABEL_LIVE (JUMP_LABEL (insn));
|
960 |
|
|
else
|
961 |
|
|
live = &ever_live_at_start;
|
962 |
|
|
|
963 |
|
|
for (i = FIRST_PSEUDO_REGISTER - 1; i >= 0; --i)
|
964 |
|
|
if (TEST_HARD_REG_BIT (*live, i))
|
965 |
|
|
reg_state[i].use_index = -1;
|
966 |
|
|
}
|
967 |
|
|
|
968 |
|
|
reload_combine_note_use (&PATTERN (insn), insn);
|
969 |
|
|
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
|
970 |
|
|
{
|
971 |
|
|
if (REG_NOTE_KIND (note) == REG_INC
|
972 |
|
|
&& REG_P (XEXP (note, 0)))
|
973 |
|
|
{
|
974 |
|
|
int regno = REGNO (XEXP (note, 0));
|
975 |
|
|
|
976 |
|
|
reg_state[regno].store_ruid = reload_combine_ruid;
|
977 |
|
|
reg_state[regno].use_index = -1;
|
978 |
|
|
}
|
979 |
|
|
}
|
980 |
|
|
}
|
981 |
|
|
|
982 |
|
|
free (label_live);
|
983 |
|
|
}
|
984 |
|
|
|
985 |
|
|
/* Check if DST is a register or a subreg of a register; if it is,
|
986 |
|
|
update reg_state[regno].store_ruid and reg_state[regno].use_index
|
987 |
|
|
accordingly. Called via note_stores from reload_combine. */
|
988 |
|
|
|
989 |
|
|
static void
|
990 |
|
|
reload_combine_note_store (rtx dst, rtx set, void *data ATTRIBUTE_UNUSED)
|
991 |
|
|
{
|
992 |
|
|
int regno = 0;
|
993 |
|
|
int i;
|
994 |
|
|
enum machine_mode mode = GET_MODE (dst);
|
995 |
|
|
|
996 |
|
|
if (GET_CODE (dst) == SUBREG)
|
997 |
|
|
{
|
998 |
|
|
regno = subreg_regno_offset (REGNO (SUBREG_REG (dst)),
|
999 |
|
|
GET_MODE (SUBREG_REG (dst)),
|
1000 |
|
|
SUBREG_BYTE (dst),
|
1001 |
|
|
GET_MODE (dst));
|
1002 |
|
|
dst = SUBREG_REG (dst);
|
1003 |
|
|
}
|
1004 |
|
|
if (!REG_P (dst))
|
1005 |
|
|
return;
|
1006 |
|
|
regno += REGNO (dst);
|
1007 |
|
|
|
1008 |
|
|
/* note_stores might have stripped a STRICT_LOW_PART, so we have to be
|
1009 |
|
|
careful with registers / register parts that are not full words.
|
1010 |
|
|
Similarly for ZERO_EXTRACT. */
|
1011 |
|
|
if (GET_CODE (set) != SET
|
1012 |
|
|
|| GET_CODE (SET_DEST (set)) == ZERO_EXTRACT
|
1013 |
|
|
|| GET_CODE (SET_DEST (set)) == STRICT_LOW_PART)
|
1014 |
|
|
{
|
1015 |
|
|
for (i = hard_regno_nregs[regno][mode] - 1 + regno; i >= regno; i--)
|
1016 |
|
|
{
|
1017 |
|
|
reg_state[i].use_index = -1;
|
1018 |
|
|
reg_state[i].store_ruid = reload_combine_ruid;
|
1019 |
|
|
}
|
1020 |
|
|
}
|
1021 |
|
|
else
|
1022 |
|
|
{
|
1023 |
|
|
for (i = hard_regno_nregs[regno][mode] - 1 + regno; i >= regno; i--)
|
1024 |
|
|
{
|
1025 |
|
|
reg_state[i].store_ruid = reload_combine_ruid;
|
1026 |
|
|
reg_state[i].use_index = RELOAD_COMBINE_MAX_USES;
|
1027 |
|
|
}
|
1028 |
|
|
}
|
1029 |
|
|
}
|
1030 |
|
|
|
1031 |
|
|
/* XP points to a piece of rtl that has to be checked for any uses of
|
1032 |
|
|
registers.
|
1033 |
|
|
*XP is the pattern of INSN, or a part of it.
|
1034 |
|
|
Called from reload_combine, and recursively by itself. */
|
1035 |
|
|
static void
|
1036 |
|
|
reload_combine_note_use (rtx *xp, rtx insn)
|
1037 |
|
|
{
|
1038 |
|
|
rtx x = *xp;
|
1039 |
|
|
enum rtx_code code = x->code;
|
1040 |
|
|
const char *fmt;
|
1041 |
|
|
int i, j;
|
1042 |
|
|
rtx offset = const0_rtx; /* For the REG case below. */
|
1043 |
|
|
|
1044 |
|
|
switch (code)
|
1045 |
|
|
{
|
1046 |
|
|
case SET:
|
1047 |
|
|
if (REG_P (SET_DEST (x)))
|
1048 |
|
|
{
|
1049 |
|
|
reload_combine_note_use (&SET_SRC (x), insn);
|
1050 |
|
|
return;
|
1051 |
|
|
}
|
1052 |
|
|
break;
|
1053 |
|
|
|
1054 |
|
|
case USE:
|
1055 |
|
|
/* If this is the USE of a return value, we can't change it. */
|
1056 |
|
|
if (REG_P (XEXP (x, 0)) && REG_FUNCTION_VALUE_P (XEXP (x, 0)))
|
1057 |
|
|
{
|
1058 |
|
|
/* Mark the return register as used in an unknown fashion. */
|
1059 |
|
|
rtx reg = XEXP (x, 0);
|
1060 |
|
|
int regno = REGNO (reg);
|
1061 |
|
|
int nregs = hard_regno_nregs[regno][GET_MODE (reg)];
|
1062 |
|
|
|
1063 |
|
|
while (--nregs >= 0)
|
1064 |
|
|
reg_state[regno + nregs].use_index = -1;
|
1065 |
|
|
return;
|
1066 |
|
|
}
|
1067 |
|
|
break;
|
1068 |
|
|
|
1069 |
|
|
case CLOBBER:
|
1070 |
|
|
if (REG_P (SET_DEST (x)))
|
1071 |
|
|
{
|
1072 |
|
|
/* No spurious CLOBBERs of pseudo registers may remain. */
|
1073 |
|
|
gcc_assert (REGNO (SET_DEST (x)) < FIRST_PSEUDO_REGISTER);
|
1074 |
|
|
return;
|
1075 |
|
|
}
|
1076 |
|
|
break;
|
1077 |
|
|
|
1078 |
|
|
case PLUS:
|
1079 |
|
|
/* We are interested in (plus (reg) (const_int)) . */
|
1080 |
|
|
if (!REG_P (XEXP (x, 0))
|
1081 |
|
|
|| GET_CODE (XEXP (x, 1)) != CONST_INT)
|
1082 |
|
|
break;
|
1083 |
|
|
offset = XEXP (x, 1);
|
1084 |
|
|
x = XEXP (x, 0);
|
1085 |
|
|
/* Fall through. */
|
1086 |
|
|
case REG:
|
1087 |
|
|
{
|
1088 |
|
|
int regno = REGNO (x);
|
1089 |
|
|
int use_index;
|
1090 |
|
|
int nregs;
|
1091 |
|
|
|
1092 |
|
|
/* No spurious USEs of pseudo registers may remain. */
|
1093 |
|
|
gcc_assert (regno < FIRST_PSEUDO_REGISTER);
|
1094 |
|
|
|
1095 |
|
|
nregs = hard_regno_nregs[regno][GET_MODE (x)];
|
1096 |
|
|
|
1097 |
|
|
/* We can't substitute into multi-hard-reg uses. */
|
1098 |
|
|
if (nregs > 1)
|
1099 |
|
|
{
|
1100 |
|
|
while (--nregs >= 0)
|
1101 |
|
|
reg_state[regno + nregs].use_index = -1;
|
1102 |
|
|
return;
|
1103 |
|
|
}
|
1104 |
|
|
|
1105 |
|
|
/* If this register is already used in some unknown fashion, we
|
1106 |
|
|
can't do anything.
|
1107 |
|
|
If we decrement the index from zero to -1, we can't store more
|
1108 |
|
|
uses, so this register becomes used in an unknown fashion. */
|
1109 |
|
|
use_index = --reg_state[regno].use_index;
|
1110 |
|
|
if (use_index < 0)
|
1111 |
|
|
return;
|
1112 |
|
|
|
1113 |
|
|
if (use_index != RELOAD_COMBINE_MAX_USES - 1)
|
1114 |
|
|
{
|
1115 |
|
|
/* We have found another use for a register that is already
|
1116 |
|
|
used later. Check if the offsets match; if not, mark the
|
1117 |
|
|
register as used in an unknown fashion. */
|
1118 |
|
|
if (! rtx_equal_p (offset, reg_state[regno].offset))
|
1119 |
|
|
{
|
1120 |
|
|
reg_state[regno].use_index = -1;
|
1121 |
|
|
return;
|
1122 |
|
|
}
|
1123 |
|
|
}
|
1124 |
|
|
else
|
1125 |
|
|
{
|
1126 |
|
|
/* This is the first use of this register we have seen since we
|
1127 |
|
|
marked it as dead. */
|
1128 |
|
|
reg_state[regno].offset = offset;
|
1129 |
|
|
reg_state[regno].use_ruid = reload_combine_ruid;
|
1130 |
|
|
}
|
1131 |
|
|
reg_state[regno].reg_use[use_index].insn = insn;
|
1132 |
|
|
reg_state[regno].reg_use[use_index].usep = xp;
|
1133 |
|
|
return;
|
1134 |
|
|
}
|
1135 |
|
|
|
1136 |
|
|
default:
|
1137 |
|
|
break;
|
1138 |
|
|
}
|
1139 |
|
|
|
1140 |
|
|
/* Recursively process the components of X. */
|
1141 |
|
|
fmt = GET_RTX_FORMAT (code);
|
1142 |
|
|
for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
|
1143 |
|
|
{
|
1144 |
|
|
if (fmt[i] == 'e')
|
1145 |
|
|
reload_combine_note_use (&XEXP (x, i), insn);
|
1146 |
|
|
else if (fmt[i] == 'E')
|
1147 |
|
|
{
|
1148 |
|
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
1149 |
|
|
reload_combine_note_use (&XVECEXP (x, i, j), insn);
|
1150 |
|
|
}
|
1151 |
|
|
}
|
1152 |
|
|
}
|
1153 |
|
|
|
1154 |
|
|
/* See if we can reduce the cost of a constant by replacing a move
|
1155 |
|
|
with an add. We track situations in which a register is set to a
|
1156 |
|
|
constant or to a register plus a constant. */
|
1157 |
|
|
/* We cannot do our optimization across labels. Invalidating all the
|
1158 |
|
|
information about register contents we have would be costly, so we
|
1159 |
|
|
use move2add_last_label_luid to note where the label is and then
|
1160 |
|
|
later disable any optimization that would cross it.
|
1161 |
|
|
reg_offset[n] / reg_base_reg[n] / reg_mode[n] are only valid if
|
1162 |
|
|
reg_set_luid[n] is greater than move2add_last_label_luid. */
|
1163 |
|
|
static int reg_set_luid[FIRST_PSEUDO_REGISTER];
|
1164 |
|
|
|
1165 |
|
|
/* If reg_base_reg[n] is negative, register n has been set to
|
1166 |
|
|
reg_offset[n] in mode reg_mode[n] .
|
1167 |
|
|
If reg_base_reg[n] is non-negative, register n has been set to the
|
1168 |
|
|
sum of reg_offset[n] and the value of register reg_base_reg[n]
|
1169 |
|
|
before reg_set_luid[n], calculated in mode reg_mode[n] . */
|
1170 |
|
|
static HOST_WIDE_INT reg_offset[FIRST_PSEUDO_REGISTER];
|
1171 |
|
|
static int reg_base_reg[FIRST_PSEUDO_REGISTER];
|
1172 |
|
|
static enum machine_mode reg_mode[FIRST_PSEUDO_REGISTER];
|
1173 |
|
|
|
1174 |
|
|
/* move2add_luid is linearly increased while scanning the instructions
|
1175 |
|
|
from first to last. It is used to set reg_set_luid in
|
1176 |
|
|
reload_cse_move2add and move2add_note_store. */
|
1177 |
|
|
static int move2add_luid;
|
1178 |
|
|
|
1179 |
|
|
/* move2add_last_label_luid is set whenever a label is found. Labels
|
1180 |
|
|
invalidate all previously collected reg_offset data. */
|
1181 |
|
|
static int move2add_last_label_luid;
|
1182 |
|
|
|
1183 |
|
|
/* ??? We don't know how zero / sign extension is handled, hence we
|
1184 |
|
|
can't go from a narrower to a wider mode. */
|
1185 |
|
|
#define MODES_OK_FOR_MOVE2ADD(OUTMODE, INMODE) \
|
1186 |
|
|
(GET_MODE_SIZE (OUTMODE) == GET_MODE_SIZE (INMODE) \
|
1187 |
|
|
|| (GET_MODE_SIZE (OUTMODE) <= GET_MODE_SIZE (INMODE) \
|
1188 |
|
|
&& TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (OUTMODE), \
|
1189 |
|
|
GET_MODE_BITSIZE (INMODE))))
|
1190 |
|
|
|
1191 |
|
|
static void
|
1192 |
|
|
reload_cse_move2add (rtx first)
|
1193 |
|
|
{
|
1194 |
|
|
int i;
|
1195 |
|
|
rtx insn;
|
1196 |
|
|
|
1197 |
|
|
for (i = FIRST_PSEUDO_REGISTER - 1; i >= 0; i--)
|
1198 |
|
|
reg_set_luid[i] = 0;
|
1199 |
|
|
|
1200 |
|
|
move2add_last_label_luid = 0;
|
1201 |
|
|
move2add_luid = 2;
|
1202 |
|
|
for (insn = first; insn; insn = NEXT_INSN (insn), move2add_luid++)
|
1203 |
|
|
{
|
1204 |
|
|
rtx pat, note;
|
1205 |
|
|
|
1206 |
|
|
if (LABEL_P (insn))
|
1207 |
|
|
{
|
1208 |
|
|
move2add_last_label_luid = move2add_luid;
|
1209 |
|
|
/* We're going to increment move2add_luid twice after a
|
1210 |
|
|
label, so that we can use move2add_last_label_luid + 1 as
|
1211 |
|
|
the luid for constants. */
|
1212 |
|
|
move2add_luid++;
|
1213 |
|
|
continue;
|
1214 |
|
|
}
|
1215 |
|
|
if (! INSN_P (insn))
|
1216 |
|
|
continue;
|
1217 |
|
|
pat = PATTERN (insn);
|
1218 |
|
|
/* For simplicity, we only perform this optimization on
|
1219 |
|
|
straightforward SETs. */
|
1220 |
|
|
if (GET_CODE (pat) == SET
|
1221 |
|
|
&& REG_P (SET_DEST (pat)))
|
1222 |
|
|
{
|
1223 |
|
|
rtx reg = SET_DEST (pat);
|
1224 |
|
|
int regno = REGNO (reg);
|
1225 |
|
|
rtx src = SET_SRC (pat);
|
1226 |
|
|
|
1227 |
|
|
/* Check if we have valid information on the contents of this
|
1228 |
|
|
register in the mode of REG. */
|
1229 |
|
|
if (reg_set_luid[regno] > move2add_last_label_luid
|
1230 |
|
|
&& MODES_OK_FOR_MOVE2ADD (GET_MODE (reg), reg_mode[regno]))
|
1231 |
|
|
{
|
1232 |
|
|
/* Try to transform (set (REGX) (CONST_INT A))
|
1233 |
|
|
...
|
1234 |
|
|
(set (REGX) (CONST_INT B))
|
1235 |
|
|
to
|
1236 |
|
|
(set (REGX) (CONST_INT A))
|
1237 |
|
|
...
|
1238 |
|
|
(set (REGX) (plus (REGX) (CONST_INT B-A)))
|
1239 |
|
|
or
|
1240 |
|
|
(set (REGX) (CONST_INT A))
|
1241 |
|
|
...
|
1242 |
|
|
(set (STRICT_LOW_PART (REGX)) (CONST_INT B))
|
1243 |
|
|
*/
|
1244 |
|
|
|
1245 |
|
|
if (GET_CODE (src) == CONST_INT && reg_base_reg[regno] < 0)
|
1246 |
|
|
{
|
1247 |
|
|
rtx new_src = gen_int_mode (INTVAL (src) - reg_offset[regno],
|
1248 |
|
|
GET_MODE (reg));
|
1249 |
|
|
/* (set (reg) (plus (reg) (const_int 0))) is not canonical;
|
1250 |
|
|
use (set (reg) (reg)) instead.
|
1251 |
|
|
We don't delete this insn, nor do we convert it into a
|
1252 |
|
|
note, to avoid losing register notes or the return
|
1253 |
|
|
value flag. jump2 already knows how to get rid of
|
1254 |
|
|
no-op moves. */
|
1255 |
|
|
if (new_src == const0_rtx)
|
1256 |
|
|
{
|
1257 |
|
|
/* If the constants are different, this is a
|
1258 |
|
|
truncation, that, if turned into (set (reg)
|
1259 |
|
|
(reg)), would be discarded. Maybe we should
|
1260 |
|
|
try a truncMN pattern? */
|
1261 |
|
|
if (INTVAL (src) == reg_offset [regno])
|
1262 |
|
|
validate_change (insn, &SET_SRC (pat), reg, 0);
|
1263 |
|
|
}
|
1264 |
|
|
else if (rtx_cost (new_src, PLUS) < rtx_cost (src, SET)
|
1265 |
|
|
&& have_add2_insn (reg, new_src))
|
1266 |
|
|
{
|
1267 |
|
|
rtx tem = gen_rtx_PLUS (GET_MODE (reg), reg, new_src);
|
1268 |
|
|
validate_change (insn, &SET_SRC (pat), tem, 0);
|
1269 |
|
|
}
|
1270 |
|
|
else if (GET_MODE (reg) != BImode)
|
1271 |
|
|
{
|
1272 |
|
|
enum machine_mode narrow_mode;
|
1273 |
|
|
for (narrow_mode = GET_CLASS_NARROWEST_MODE (MODE_INT);
|
1274 |
|
|
narrow_mode != VOIDmode
|
1275 |
|
|
&& narrow_mode != GET_MODE (reg);
|
1276 |
|
|
narrow_mode = GET_MODE_WIDER_MODE (narrow_mode))
|
1277 |
|
|
{
|
1278 |
|
|
if (have_insn_for (STRICT_LOW_PART, narrow_mode)
|
1279 |
|
|
&& ((reg_offset[regno]
|
1280 |
|
|
& ~GET_MODE_MASK (narrow_mode))
|
1281 |
|
|
== (INTVAL (src)
|
1282 |
|
|
& ~GET_MODE_MASK (narrow_mode))))
|
1283 |
|
|
{
|
1284 |
|
|
rtx narrow_reg = gen_rtx_REG (narrow_mode,
|
1285 |
|
|
REGNO (reg));
|
1286 |
|
|
rtx narrow_src = gen_int_mode (INTVAL (src),
|
1287 |
|
|
narrow_mode);
|
1288 |
|
|
rtx new_set =
|
1289 |
|
|
gen_rtx_SET (VOIDmode,
|
1290 |
|
|
gen_rtx_STRICT_LOW_PART (VOIDmode,
|
1291 |
|
|
narrow_reg),
|
1292 |
|
|
narrow_src);
|
1293 |
|
|
if (validate_change (insn, &PATTERN (insn),
|
1294 |
|
|
new_set, 0))
|
1295 |
|
|
break;
|
1296 |
|
|
}
|
1297 |
|
|
}
|
1298 |
|
|
}
|
1299 |
|
|
reg_set_luid[regno] = move2add_luid;
|
1300 |
|
|
reg_mode[regno] = GET_MODE (reg);
|
1301 |
|
|
reg_offset[regno] = INTVAL (src);
|
1302 |
|
|
continue;
|
1303 |
|
|
}
|
1304 |
|
|
|
1305 |
|
|
/* Try to transform (set (REGX) (REGY))
|
1306 |
|
|
(set (REGX) (PLUS (REGX) (CONST_INT A)))
|
1307 |
|
|
...
|
1308 |
|
|
(set (REGX) (REGY))
|
1309 |
|
|
(set (REGX) (PLUS (REGX) (CONST_INT B)))
|
1310 |
|
|
to
|
1311 |
|
|
(set (REGX) (REGY))
|
1312 |
|
|
(set (REGX) (PLUS (REGX) (CONST_INT A)))
|
1313 |
|
|
...
|
1314 |
|
|
(set (REGX) (plus (REGX) (CONST_INT B-A))) */
|
1315 |
|
|
else if (REG_P (src)
|
1316 |
|
|
&& reg_set_luid[regno] == reg_set_luid[REGNO (src)]
|
1317 |
|
|
&& reg_base_reg[regno] == reg_base_reg[REGNO (src)]
|
1318 |
|
|
&& MODES_OK_FOR_MOVE2ADD (GET_MODE (reg),
|
1319 |
|
|
reg_mode[REGNO (src)]))
|
1320 |
|
|
{
|
1321 |
|
|
rtx next = next_nonnote_insn (insn);
|
1322 |
|
|
rtx set = NULL_RTX;
|
1323 |
|
|
if (next)
|
1324 |
|
|
set = single_set (next);
|
1325 |
|
|
if (set
|
1326 |
|
|
&& SET_DEST (set) == reg
|
1327 |
|
|
&& GET_CODE (SET_SRC (set)) == PLUS
|
1328 |
|
|
&& XEXP (SET_SRC (set), 0) == reg
|
1329 |
|
|
&& GET_CODE (XEXP (SET_SRC (set), 1)) == CONST_INT)
|
1330 |
|
|
{
|
1331 |
|
|
rtx src3 = XEXP (SET_SRC (set), 1);
|
1332 |
|
|
HOST_WIDE_INT added_offset = INTVAL (src3);
|
1333 |
|
|
HOST_WIDE_INT base_offset = reg_offset[REGNO (src)];
|
1334 |
|
|
HOST_WIDE_INT regno_offset = reg_offset[regno];
|
1335 |
|
|
rtx new_src =
|
1336 |
|
|
gen_int_mode (added_offset
|
1337 |
|
|
+ base_offset
|
1338 |
|
|
- regno_offset,
|
1339 |
|
|
GET_MODE (reg));
|
1340 |
|
|
int success = 0;
|
1341 |
|
|
|
1342 |
|
|
if (new_src == const0_rtx)
|
1343 |
|
|
/* See above why we create (set (reg) (reg)) here. */
|
1344 |
|
|
success
|
1345 |
|
|
= validate_change (next, &SET_SRC (set), reg, 0);
|
1346 |
|
|
else if ((rtx_cost (new_src, PLUS)
|
1347 |
|
|
< COSTS_N_INSNS (1) + rtx_cost (src3, SET))
|
1348 |
|
|
&& have_add2_insn (reg, new_src))
|
1349 |
|
|
{
|
1350 |
|
|
rtx newpat = gen_rtx_SET (VOIDmode,
|
1351 |
|
|
reg,
|
1352 |
|
|
gen_rtx_PLUS (GET_MODE (reg),
|
1353 |
|
|
reg,
|
1354 |
|
|
new_src));
|
1355 |
|
|
success
|
1356 |
|
|
= validate_change (next, &PATTERN (next),
|
1357 |
|
|
newpat, 0);
|
1358 |
|
|
}
|
1359 |
|
|
if (success)
|
1360 |
|
|
delete_insn (insn);
|
1361 |
|
|
insn = next;
|
1362 |
|
|
reg_mode[regno] = GET_MODE (reg);
|
1363 |
|
|
reg_offset[regno] =
|
1364 |
|
|
trunc_int_for_mode (added_offset + base_offset,
|
1365 |
|
|
GET_MODE (reg));
|
1366 |
|
|
continue;
|
1367 |
|
|
}
|
1368 |
|
|
}
|
1369 |
|
|
}
|
1370 |
|
|
}
|
1371 |
|
|
|
1372 |
|
|
for (note = REG_NOTES (insn); note; note = XEXP (note, 1))
|
1373 |
|
|
{
|
1374 |
|
|
if (REG_NOTE_KIND (note) == REG_INC
|
1375 |
|
|
&& REG_P (XEXP (note, 0)))
|
1376 |
|
|
{
|
1377 |
|
|
/* Reset the information about this register. */
|
1378 |
|
|
int regno = REGNO (XEXP (note, 0));
|
1379 |
|
|
if (regno < FIRST_PSEUDO_REGISTER)
|
1380 |
|
|
reg_set_luid[regno] = 0;
|
1381 |
|
|
}
|
1382 |
|
|
}
|
1383 |
|
|
note_stores (PATTERN (insn), move2add_note_store, NULL);
|
1384 |
|
|
|
1385 |
|
|
/* If INSN is a conditional branch, we try to extract an
|
1386 |
|
|
implicit set out of it. */
|
1387 |
|
|
if (any_condjump_p (insn))
|
1388 |
|
|
{
|
1389 |
|
|
rtx cnd = fis_get_condition (insn);
|
1390 |
|
|
|
1391 |
|
|
if (cnd != NULL_RTX
|
1392 |
|
|
&& GET_CODE (cnd) == NE
|
1393 |
|
|
&& REG_P (XEXP (cnd, 0))
|
1394 |
|
|
&& !reg_set_p (XEXP (cnd, 0), insn)
|
1395 |
|
|
/* The following two checks, which are also in
|
1396 |
|
|
move2add_note_store, are intended to reduce the
|
1397 |
|
|
number of calls to gen_rtx_SET to avoid memory
|
1398 |
|
|
allocation if possible. */
|
1399 |
|
|
&& SCALAR_INT_MODE_P (GET_MODE (XEXP (cnd, 0)))
|
1400 |
|
|
&& hard_regno_nregs[REGNO (XEXP (cnd, 0))][GET_MODE (XEXP (cnd, 0))] == 1
|
1401 |
|
|
&& GET_CODE (XEXP (cnd, 1)) == CONST_INT)
|
1402 |
|
|
{
|
1403 |
|
|
rtx implicit_set =
|
1404 |
|
|
gen_rtx_SET (VOIDmode, XEXP (cnd, 0), XEXP (cnd, 1));
|
1405 |
|
|
move2add_note_store (SET_DEST (implicit_set), implicit_set, 0);
|
1406 |
|
|
}
|
1407 |
|
|
}
|
1408 |
|
|
|
1409 |
|
|
/* If this is a CALL_INSN, all call used registers are stored with
|
1410 |
|
|
unknown values. */
|
1411 |
|
|
if (CALL_P (insn))
|
1412 |
|
|
{
|
1413 |
|
|
for (i = FIRST_PSEUDO_REGISTER - 1; i >= 0; i--)
|
1414 |
|
|
{
|
1415 |
|
|
if (call_used_regs[i])
|
1416 |
|
|
/* Reset the information about this register. */
|
1417 |
|
|
reg_set_luid[i] = 0;
|
1418 |
|
|
}
|
1419 |
|
|
}
|
1420 |
|
|
}
|
1421 |
|
|
}
|
1422 |
|
|
|
1423 |
|
|
/* SET is a SET or CLOBBER that sets DST.
|
1424 |
|
|
Update reg_set_luid, reg_offset and reg_base_reg accordingly.
|
1425 |
|
|
Called from reload_cse_move2add via note_stores. */
|
1426 |
|
|
|
1427 |
|
|
static void
|
1428 |
|
|
move2add_note_store (rtx dst, rtx set, void *data ATTRIBUTE_UNUSED)
|
1429 |
|
|
{
|
1430 |
|
|
unsigned int regno = 0;
|
1431 |
|
|
unsigned int i;
|
1432 |
|
|
enum machine_mode mode = GET_MODE (dst);
|
1433 |
|
|
|
1434 |
|
|
if (GET_CODE (dst) == SUBREG)
|
1435 |
|
|
{
|
1436 |
|
|
regno = subreg_regno_offset (REGNO (SUBREG_REG (dst)),
|
1437 |
|
|
GET_MODE (SUBREG_REG (dst)),
|
1438 |
|
|
SUBREG_BYTE (dst),
|
1439 |
|
|
GET_MODE (dst));
|
1440 |
|
|
dst = SUBREG_REG (dst);
|
1441 |
|
|
}
|
1442 |
|
|
|
1443 |
|
|
/* Some targets do argument pushes without adding REG_INC notes. */
|
1444 |
|
|
|
1445 |
|
|
if (MEM_P (dst))
|
1446 |
|
|
{
|
1447 |
|
|
dst = XEXP (dst, 0);
|
1448 |
|
|
if (GET_CODE (dst) == PRE_INC || GET_CODE (dst) == POST_INC
|
1449 |
|
|
|| GET_CODE (dst) == PRE_DEC || GET_CODE (dst) == POST_DEC)
|
1450 |
|
|
reg_set_luid[REGNO (XEXP (dst, 0))] = 0;
|
1451 |
|
|
return;
|
1452 |
|
|
}
|
1453 |
|
|
if (!REG_P (dst))
|
1454 |
|
|
return;
|
1455 |
|
|
|
1456 |
|
|
regno += REGNO (dst);
|
1457 |
|
|
|
1458 |
|
|
if (SCALAR_INT_MODE_P (GET_MODE (dst))
|
1459 |
|
|
&& hard_regno_nregs[regno][mode] == 1 && GET_CODE (set) == SET
|
1460 |
|
|
&& GET_CODE (SET_DEST (set)) != ZERO_EXTRACT
|
1461 |
|
|
&& GET_CODE (SET_DEST (set)) != STRICT_LOW_PART)
|
1462 |
|
|
{
|
1463 |
|
|
rtx src = SET_SRC (set);
|
1464 |
|
|
rtx base_reg;
|
1465 |
|
|
HOST_WIDE_INT offset;
|
1466 |
|
|
int base_regno;
|
1467 |
|
|
/* This may be different from mode, if SET_DEST (set) is a
|
1468 |
|
|
SUBREG. */
|
1469 |
|
|
enum machine_mode dst_mode = GET_MODE (dst);
|
1470 |
|
|
|
1471 |
|
|
switch (GET_CODE (src))
|
1472 |
|
|
{
|
1473 |
|
|
case PLUS:
|
1474 |
|
|
if (REG_P (XEXP (src, 0)))
|
1475 |
|
|
{
|
1476 |
|
|
base_reg = XEXP (src, 0);
|
1477 |
|
|
|
1478 |
|
|
if (GET_CODE (XEXP (src, 1)) == CONST_INT)
|
1479 |
|
|
offset = INTVAL (XEXP (src, 1));
|
1480 |
|
|
else if (REG_P (XEXP (src, 1))
|
1481 |
|
|
&& (reg_set_luid[REGNO (XEXP (src, 1))]
|
1482 |
|
|
> move2add_last_label_luid)
|
1483 |
|
|
&& (MODES_OK_FOR_MOVE2ADD
|
1484 |
|
|
(dst_mode, reg_mode[REGNO (XEXP (src, 1))])))
|
1485 |
|
|
{
|
1486 |
|
|
if (reg_base_reg[REGNO (XEXP (src, 1))] < 0)
|
1487 |
|
|
offset = reg_offset[REGNO (XEXP (src, 1))];
|
1488 |
|
|
/* Maybe the first register is known to be a
|
1489 |
|
|
constant. */
|
1490 |
|
|
else if (reg_set_luid[REGNO (base_reg)]
|
1491 |
|
|
> move2add_last_label_luid
|
1492 |
|
|
&& (MODES_OK_FOR_MOVE2ADD
|
1493 |
|
|
(dst_mode, reg_mode[REGNO (XEXP (src, 1))]))
|
1494 |
|
|
&& reg_base_reg[REGNO (base_reg)] < 0)
|
1495 |
|
|
{
|
1496 |
|
|
offset = reg_offset[REGNO (base_reg)];
|
1497 |
|
|
base_reg = XEXP (src, 1);
|
1498 |
|
|
}
|
1499 |
|
|
else
|
1500 |
|
|
goto invalidate;
|
1501 |
|
|
}
|
1502 |
|
|
else
|
1503 |
|
|
goto invalidate;
|
1504 |
|
|
|
1505 |
|
|
break;
|
1506 |
|
|
}
|
1507 |
|
|
|
1508 |
|
|
goto invalidate;
|
1509 |
|
|
|
1510 |
|
|
case REG:
|
1511 |
|
|
base_reg = src;
|
1512 |
|
|
offset = 0;
|
1513 |
|
|
break;
|
1514 |
|
|
|
1515 |
|
|
case CONST_INT:
|
1516 |
|
|
/* Start tracking the register as a constant. */
|
1517 |
|
|
reg_base_reg[regno] = -1;
|
1518 |
|
|
reg_offset[regno] = INTVAL (SET_SRC (set));
|
1519 |
|
|
/* We assign the same luid to all registers set to constants. */
|
1520 |
|
|
reg_set_luid[regno] = move2add_last_label_luid + 1;
|
1521 |
|
|
reg_mode[regno] = mode;
|
1522 |
|
|
return;
|
1523 |
|
|
|
1524 |
|
|
default:
|
1525 |
|
|
invalidate:
|
1526 |
|
|
/* Invalidate the contents of the register. */
|
1527 |
|
|
reg_set_luid[regno] = 0;
|
1528 |
|
|
return;
|
1529 |
|
|
}
|
1530 |
|
|
|
1531 |
|
|
base_regno = REGNO (base_reg);
|
1532 |
|
|
/* If information about the base register is not valid, set it
|
1533 |
|
|
up as a new base register, pretending its value is known
|
1534 |
|
|
starting from the current insn. */
|
1535 |
|
|
if (reg_set_luid[base_regno] <= move2add_last_label_luid)
|
1536 |
|
|
{
|
1537 |
|
|
reg_base_reg[base_regno] = base_regno;
|
1538 |
|
|
reg_offset[base_regno] = 0;
|
1539 |
|
|
reg_set_luid[base_regno] = move2add_luid;
|
1540 |
|
|
reg_mode[base_regno] = mode;
|
1541 |
|
|
}
|
1542 |
|
|
else if (! MODES_OK_FOR_MOVE2ADD (dst_mode,
|
1543 |
|
|
reg_mode[base_regno]))
|
1544 |
|
|
goto invalidate;
|
1545 |
|
|
|
1546 |
|
|
reg_mode[regno] = mode;
|
1547 |
|
|
|
1548 |
|
|
/* Copy base information from our base register. */
|
1549 |
|
|
reg_set_luid[regno] = reg_set_luid[base_regno];
|
1550 |
|
|
reg_base_reg[regno] = reg_base_reg[base_regno];
|
1551 |
|
|
|
1552 |
|
|
/* Compute the sum of the offsets or constants. */
|
1553 |
|
|
reg_offset[regno] = trunc_int_for_mode (offset
|
1554 |
|
|
+ reg_offset[base_regno],
|
1555 |
|
|
dst_mode);
|
1556 |
|
|
}
|
1557 |
|
|
else
|
1558 |
|
|
{
|
1559 |
|
|
unsigned int endregno = regno + hard_regno_nregs[regno][mode];
|
1560 |
|
|
|
1561 |
|
|
for (i = regno; i < endregno; i++)
|
1562 |
|
|
/* Reset the information about this register. */
|
1563 |
|
|
reg_set_luid[i] = 0;
|
1564 |
|
|
}
|
1565 |
|
|
}
|
1566 |
|
|
|
1567 |
|
|
static bool
|
1568 |
|
|
gate_handle_postreload (void)
|
1569 |
|
|
{
|
1570 |
|
|
return (optimize > 0);
|
1571 |
|
|
}
|
1572 |
|
|
|
1573 |
|
|
|
1574 |
|
|
static unsigned int
|
1575 |
|
|
rest_of_handle_postreload (void)
|
1576 |
|
|
{
|
1577 |
|
|
/* Do a very simple CSE pass over just the hard registers. */
|
1578 |
|
|
reload_cse_regs (get_insns ());
|
1579 |
|
|
/* reload_cse_regs can eliminate potentially-trapping MEMs.
|
1580 |
|
|
Remove any EH edges associated with them. */
|
1581 |
|
|
if (flag_non_call_exceptions)
|
1582 |
|
|
purge_all_dead_edges ();
|
1583 |
|
|
return 0;
|
1584 |
|
|
}
|
1585 |
|
|
|
1586 |
|
|
struct tree_opt_pass pass_postreload_cse =
|
1587 |
|
|
{
|
1588 |
|
|
"postreload", /* name */
|
1589 |
|
|
gate_handle_postreload, /* gate */
|
1590 |
|
|
rest_of_handle_postreload, /* execute */
|
1591 |
|
|
NULL, /* sub */
|
1592 |
|
|
NULL, /* next */
|
1593 |
|
|
0, /* static_pass_number */
|
1594 |
|
|
TV_RELOAD_CSE_REGS, /* tv_id */
|
1595 |
|
|
0, /* properties_required */
|
1596 |
|
|
0, /* properties_provided */
|
1597 |
|
|
0, /* properties_destroyed */
|
1598 |
|
|
0, /* todo_flags_start */
|
1599 |
|
|
TODO_dump_func, /* todo_flags_finish */
|
1600 |
|
|
'o' /* letter */
|
1601 |
|
|
};
|
1602 |
|
|
|