]>
Commit | Line | Data |
---|---|---|
8f2e8c07 KB |
1 | /* |
2 | * Optimizations for Tiny Code Generator for QEMU | |
3 | * | |
4 | * Copyright (c) 2010 Samsung Electronics. | |
5 | * Contributed by Kirill Batuzov <batuzovk@ispras.ru> | |
6 | * | |
7 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
8 | * of this software and associated documentation files (the "Software"), to deal | |
9 | * in the Software without restriction, including without limitation the rights | |
10 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
11 | * copies of the Software, and to permit persons to whom the Software is | |
12 | * furnished to do so, subject to the following conditions: | |
13 | * | |
14 | * The above copyright notice and this permission notice shall be included in | |
15 | * all copies or substantial portions of the Software. | |
16 | * | |
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
18 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
19 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
20 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
21 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
22 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
23 | * THE SOFTWARE. | |
24 | */ | |
25 | ||
26 | #include "config.h" | |
27 | ||
28 | #include <stdlib.h> | |
29 | #include <stdio.h> | |
30 | ||
31 | #include "qemu-common.h" | |
32 | #include "tcg-op.h" | |
33 | ||
8f2e8c07 KB |
34 | #define CASE_OP_32_64(x) \ |
35 | glue(glue(case INDEX_op_, x), _i32): \ | |
36 | glue(glue(case INDEX_op_, x), _i64) | |
8f2e8c07 | 37 | |
22613af4 KB |
38 | typedef enum { |
39 | TCG_TEMP_UNDEF = 0, | |
40 | TCG_TEMP_CONST, | |
41 | TCG_TEMP_COPY, | |
42 | TCG_TEMP_HAS_COPY, | |
43 | TCG_TEMP_ANY | |
44 | } tcg_temp_state; | |
45 | ||
46 | struct tcg_temp_info { | |
47 | tcg_temp_state state; | |
48 | uint16_t prev_copy; | |
49 | uint16_t next_copy; | |
50 | tcg_target_ulong val; | |
51 | }; | |
52 | ||
53 | static struct tcg_temp_info temps[TCG_MAX_TEMPS]; | |
54 | ||
55 | /* Reset TEMP's state to TCG_TEMP_ANY. If TEMP was a representative of some | |
56 | class of equivalent temp's, a new representative should be chosen in this | |
57 | class. */ | |
58 | static void reset_temp(TCGArg temp, int nb_temps, int nb_globals) | |
59 | { | |
60 | int i; | |
61 | TCGArg new_base = (TCGArg)-1; | |
62 | if (temps[temp].state == TCG_TEMP_HAS_COPY) { | |
63 | for (i = temps[temp].next_copy; i != temp; i = temps[i].next_copy) { | |
64 | if (i >= nb_globals) { | |
65 | temps[i].state = TCG_TEMP_HAS_COPY; | |
66 | new_base = i; | |
67 | break; | |
68 | } | |
69 | } | |
70 | for (i = temps[temp].next_copy; i != temp; i = temps[i].next_copy) { | |
71 | if (new_base == (TCGArg)-1) { | |
72 | temps[i].state = TCG_TEMP_ANY; | |
73 | } else { | |
74 | temps[i].val = new_base; | |
75 | } | |
76 | } | |
77 | temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy; | |
78 | temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy; | |
79 | } else if (temps[temp].state == TCG_TEMP_COPY) { | |
80 | temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy; | |
81 | temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy; | |
82 | new_base = temps[temp].val; | |
83 | } | |
84 | temps[temp].state = TCG_TEMP_ANY; | |
85 | if (new_base != (TCGArg)-1 && temps[new_base].next_copy == new_base) { | |
86 | temps[new_base].state = TCG_TEMP_ANY; | |
87 | } | |
88 | } | |
89 | ||
fe0de7aa | 90 | static int op_bits(TCGOpcode op) |
22613af4 | 91 | { |
8399ad59 RH |
92 | const TCGOpDef *def = &tcg_op_defs[op]; |
93 | return def->flags & TCG_OPF_64BIT ? 64 : 32; | |
22613af4 KB |
94 | } |
95 | ||
fe0de7aa | 96 | static TCGOpcode op_to_movi(TCGOpcode op) |
22613af4 KB |
97 | { |
98 | switch (op_bits(op)) { | |
99 | case 32: | |
100 | return INDEX_op_movi_i32; | |
22613af4 KB |
101 | case 64: |
102 | return INDEX_op_movi_i64; | |
22613af4 KB |
103 | default: |
104 | fprintf(stderr, "op_to_movi: unexpected return value of " | |
105 | "function op_bits.\n"); | |
106 | tcg_abort(); | |
107 | } | |
108 | } | |
109 | ||
4f17bc1e AJ |
110 | static void tcg_opt_gen_mov(TCGArg *gen_args, TCGArg dst, TCGArg src, |
111 | int nb_temps, int nb_globals) | |
22613af4 KB |
112 | { |
113 | reset_temp(dst, nb_temps, nb_globals); | |
114 | assert(temps[src].state != TCG_TEMP_COPY); | |
4f17bc1e | 115 | if (src >= nb_globals) { |
22613af4 KB |
116 | assert(temps[src].state != TCG_TEMP_CONST); |
117 | if (temps[src].state != TCG_TEMP_HAS_COPY) { | |
118 | temps[src].state = TCG_TEMP_HAS_COPY; | |
119 | temps[src].next_copy = src; | |
120 | temps[src].prev_copy = src; | |
121 | } | |
122 | temps[dst].state = TCG_TEMP_COPY; | |
123 | temps[dst].val = src; | |
124 | temps[dst].next_copy = temps[src].next_copy; | |
125 | temps[dst].prev_copy = src; | |
126 | temps[temps[dst].next_copy].prev_copy = dst; | |
127 | temps[src].next_copy = dst; | |
128 | } | |
129 | gen_args[0] = dst; | |
130 | gen_args[1] = src; | |
131 | } | |
132 | ||
133 | static void tcg_opt_gen_movi(TCGArg *gen_args, TCGArg dst, TCGArg val, | |
134 | int nb_temps, int nb_globals) | |
135 | { | |
136 | reset_temp(dst, nb_temps, nb_globals); | |
137 | temps[dst].state = TCG_TEMP_CONST; | |
138 | temps[dst].val = val; | |
139 | gen_args[0] = dst; | |
140 | gen_args[1] = val; | |
141 | } | |
142 | ||
fe0de7aa | 143 | static TCGOpcode op_to_mov(TCGOpcode op) |
53108fb5 KB |
144 | { |
145 | switch (op_bits(op)) { | |
146 | case 32: | |
147 | return INDEX_op_mov_i32; | |
53108fb5 KB |
148 | case 64: |
149 | return INDEX_op_mov_i64; | |
53108fb5 KB |
150 | default: |
151 | fprintf(stderr, "op_to_mov: unexpected return value of " | |
152 | "function op_bits.\n"); | |
153 | tcg_abort(); | |
154 | } | |
155 | } | |
156 | ||
fe0de7aa | 157 | static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y) |
53108fb5 KB |
158 | { |
159 | switch (op) { | |
160 | CASE_OP_32_64(add): | |
161 | return x + y; | |
162 | ||
163 | CASE_OP_32_64(sub): | |
164 | return x - y; | |
165 | ||
166 | CASE_OP_32_64(mul): | |
167 | return x * y; | |
168 | ||
9a81090b KB |
169 | CASE_OP_32_64(and): |
170 | return x & y; | |
171 | ||
172 | CASE_OP_32_64(or): | |
173 | return x | y; | |
174 | ||
175 | CASE_OP_32_64(xor): | |
176 | return x ^ y; | |
177 | ||
55c0975c KB |
178 | case INDEX_op_shl_i32: |
179 | return (uint32_t)x << (uint32_t)y; | |
180 | ||
55c0975c KB |
181 | case INDEX_op_shl_i64: |
182 | return (uint64_t)x << (uint64_t)y; | |
55c0975c KB |
183 | |
184 | case INDEX_op_shr_i32: | |
185 | return (uint32_t)x >> (uint32_t)y; | |
186 | ||
55c0975c KB |
187 | case INDEX_op_shr_i64: |
188 | return (uint64_t)x >> (uint64_t)y; | |
55c0975c KB |
189 | |
190 | case INDEX_op_sar_i32: | |
191 | return (int32_t)x >> (int32_t)y; | |
192 | ||
55c0975c KB |
193 | case INDEX_op_sar_i64: |
194 | return (int64_t)x >> (int64_t)y; | |
55c0975c KB |
195 | |
196 | case INDEX_op_rotr_i32: | |
25c4d9cc | 197 | x = ((uint32_t)x << (32 - y)) | ((uint32_t)x >> y); |
55c0975c KB |
198 | return x; |
199 | ||
55c0975c | 200 | case INDEX_op_rotr_i64: |
25c4d9cc | 201 | x = ((uint64_t)x << (64 - y)) | ((uint64_t)x >> y); |
55c0975c | 202 | return x; |
55c0975c KB |
203 | |
204 | case INDEX_op_rotl_i32: | |
25c4d9cc | 205 | x = ((uint32_t)x << y) | ((uint32_t)x >> (32 - y)); |
55c0975c KB |
206 | return x; |
207 | ||
55c0975c | 208 | case INDEX_op_rotl_i64: |
25c4d9cc | 209 | x = ((uint64_t)x << y) | ((uint64_t)x >> (64 - y)); |
55c0975c | 210 | return x; |
25c4d9cc RH |
211 | |
212 | CASE_OP_32_64(not): | |
a640f031 | 213 | return ~x; |
25c4d9cc | 214 | |
cb25c80a RH |
215 | CASE_OP_32_64(neg): |
216 | return -x; | |
217 | ||
218 | CASE_OP_32_64(andc): | |
219 | return x & ~y; | |
220 | ||
221 | CASE_OP_32_64(orc): | |
222 | return x | ~y; | |
223 | ||
224 | CASE_OP_32_64(eqv): | |
225 | return ~(x ^ y); | |
226 | ||
227 | CASE_OP_32_64(nand): | |
228 | return ~(x & y); | |
229 | ||
230 | CASE_OP_32_64(nor): | |
231 | return ~(x | y); | |
232 | ||
25c4d9cc | 233 | CASE_OP_32_64(ext8s): |
a640f031 | 234 | return (int8_t)x; |
25c4d9cc RH |
235 | |
236 | CASE_OP_32_64(ext16s): | |
a640f031 | 237 | return (int16_t)x; |
25c4d9cc RH |
238 | |
239 | CASE_OP_32_64(ext8u): | |
a640f031 | 240 | return (uint8_t)x; |
25c4d9cc RH |
241 | |
242 | CASE_OP_32_64(ext16u): | |
a640f031 KB |
243 | return (uint16_t)x; |
244 | ||
a640f031 KB |
245 | case INDEX_op_ext32s_i64: |
246 | return (int32_t)x; | |
247 | ||
248 | case INDEX_op_ext32u_i64: | |
249 | return (uint32_t)x; | |
a640f031 | 250 | |
53108fb5 KB |
251 | default: |
252 | fprintf(stderr, | |
253 | "Unrecognized operation %d in do_constant_folding.\n", op); | |
254 | tcg_abort(); | |
255 | } | |
256 | } | |
257 | ||
fe0de7aa | 258 | static TCGArg do_constant_folding(TCGOpcode op, TCGArg x, TCGArg y) |
53108fb5 KB |
259 | { |
260 | TCGArg res = do_constant_folding_2(op, x, y); | |
53108fb5 KB |
261 | if (op_bits(op) == 32) { |
262 | res &= 0xffffffff; | |
263 | } | |
53108fb5 KB |
264 | return res; |
265 | } | |
266 | ||
2108a786 AJ |
267 | static TCGArg do_constant_folding_cond(TCGOpcode op, TCGArg x, |
268 | TCGArg y, TCGCond c) | |
269 | { | |
270 | switch (op_bits(op)) { | |
271 | case 32: | |
272 | switch (c) { | |
273 | case TCG_COND_EQ: | |
274 | return (uint32_t)x == (uint32_t)y; | |
275 | case TCG_COND_NE: | |
276 | return (uint32_t)x != (uint32_t)y; | |
277 | case TCG_COND_LT: | |
278 | return (int32_t)x < (int32_t)y; | |
279 | case TCG_COND_GE: | |
280 | return (int32_t)x >= (int32_t)y; | |
281 | case TCG_COND_LE: | |
282 | return (int32_t)x <= (int32_t)y; | |
283 | case TCG_COND_GT: | |
284 | return (int32_t)x > (int32_t)y; | |
285 | case TCG_COND_LTU: | |
286 | return (uint32_t)x < (uint32_t)y; | |
287 | case TCG_COND_GEU: | |
288 | return (uint32_t)x >= (uint32_t)y; | |
289 | case TCG_COND_LEU: | |
290 | return (uint32_t)x <= (uint32_t)y; | |
291 | case TCG_COND_GTU: | |
292 | return (uint32_t)x > (uint32_t)y; | |
293 | } | |
294 | break; | |
295 | case 64: | |
296 | switch (c) { | |
297 | case TCG_COND_EQ: | |
298 | return (uint64_t)x == (uint64_t)y; | |
299 | case TCG_COND_NE: | |
300 | return (uint64_t)x != (uint64_t)y; | |
301 | case TCG_COND_LT: | |
302 | return (int64_t)x < (int64_t)y; | |
303 | case TCG_COND_GE: | |
304 | return (int64_t)x >= (int64_t)y; | |
305 | case TCG_COND_LE: | |
306 | return (int64_t)x <= (int64_t)y; | |
307 | case TCG_COND_GT: | |
308 | return (int64_t)x > (int64_t)y; | |
309 | case TCG_COND_LTU: | |
310 | return (uint64_t)x < (uint64_t)y; | |
311 | case TCG_COND_GEU: | |
312 | return (uint64_t)x >= (uint64_t)y; | |
313 | case TCG_COND_LEU: | |
314 | return (uint64_t)x <= (uint64_t)y; | |
315 | case TCG_COND_GTU: | |
316 | return (uint64_t)x > (uint64_t)y; | |
317 | } | |
318 | break; | |
319 | } | |
320 | ||
321 | fprintf(stderr, | |
322 | "Unrecognized bitness %d or condition %d in " | |
323 | "do_constant_folding_cond.\n", op_bits(op), c); | |
324 | tcg_abort(); | |
325 | } | |
326 | ||
327 | ||
22613af4 | 328 | /* Propagate constants and copies, fold constant expressions. */ |
8f2e8c07 KB |
329 | static TCGArg *tcg_constant_folding(TCGContext *s, uint16_t *tcg_opc_ptr, |
330 | TCGArg *args, TCGOpDef *tcg_op_defs) | |
331 | { | |
fe0de7aa BS |
332 | int i, nb_ops, op_index, nb_temps, nb_globals, nb_call_args; |
333 | TCGOpcode op; | |
8f2e8c07 KB |
334 | const TCGOpDef *def; |
335 | TCGArg *gen_args; | |
53108fb5 | 336 | TCGArg tmp; |
1d465a94 RH |
337 | TCGCond cond; |
338 | ||
22613af4 KB |
339 | /* Array VALS has an element for each temp. |
340 | If this temp holds a constant then its value is kept in VALS' element. | |
341 | If this temp is a copy of other ones then this equivalence class' | |
342 | representative is kept in VALS' element. | |
343 | If this temp is neither copy nor constant then corresponding VALS' | |
344 | element is unused. */ | |
8f2e8c07 KB |
345 | |
346 | nb_temps = s->nb_temps; | |
347 | nb_globals = s->nb_globals; | |
22613af4 | 348 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); |
8f2e8c07 KB |
349 | |
350 | nb_ops = tcg_opc_ptr - gen_opc_buf; | |
351 | gen_args = args; | |
352 | for (op_index = 0; op_index < nb_ops; op_index++) { | |
353 | op = gen_opc_buf[op_index]; | |
354 | def = &tcg_op_defs[op]; | |
22613af4 KB |
355 | /* Do copy propagation */ |
356 | if (!(def->flags & (TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS))) { | |
357 | assert(op != INDEX_op_call); | |
358 | for (i = def->nb_oargs; i < def->nb_oargs + def->nb_iargs; i++) { | |
359 | if (temps[args[i]].state == TCG_TEMP_COPY) { | |
360 | args[i] = temps[args[i]].val; | |
361 | } | |
362 | } | |
363 | } | |
364 | ||
53108fb5 KB |
365 | /* For commutative operations make constant second argument */ |
366 | switch (op) { | |
367 | CASE_OP_32_64(add): | |
368 | CASE_OP_32_64(mul): | |
9a81090b KB |
369 | CASE_OP_32_64(and): |
370 | CASE_OP_32_64(or): | |
371 | CASE_OP_32_64(xor): | |
cb25c80a RH |
372 | CASE_OP_32_64(eqv): |
373 | CASE_OP_32_64(nand): | |
374 | CASE_OP_32_64(nor): | |
53108fb5 KB |
375 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
376 | tmp = args[1]; | |
377 | args[1] = args[2]; | |
378 | args[2] = tmp; | |
379 | } | |
380 | break; | |
decddd90 AJ |
381 | CASE_OP_32_64(brcond): |
382 | if (temps[args[0]].state == TCG_TEMP_CONST | |
383 | && temps[args[1]].state != TCG_TEMP_CONST) { | |
384 | tmp = args[0]; | |
385 | args[0] = args[1]; | |
386 | args[1] = tmp; | |
387 | args[2] = tcg_swap_cond(args[2]); | |
388 | } | |
389 | break; | |
390 | CASE_OP_32_64(setcond): | |
391 | if (temps[args[1]].state == TCG_TEMP_CONST | |
392 | && temps[args[2]].state != TCG_TEMP_CONST) { | |
393 | tmp = args[1]; | |
394 | args[1] = args[2]; | |
395 | args[2] = tmp; | |
396 | args[3] = tcg_swap_cond(args[3]); | |
397 | } | |
398 | break; | |
b5b6bb16 | 399 | CASE_OP_32_64(movcond): |
1d465a94 | 400 | cond = args[5]; |
b5b6bb16 RH |
401 | if (temps[args[1]].state == TCG_TEMP_CONST |
402 | && temps[args[2]].state != TCG_TEMP_CONST) { | |
403 | tmp = args[1]; | |
404 | args[1] = args[2]; | |
405 | args[2] = tmp; | |
1d465a94 RH |
406 | cond = tcg_swap_cond(cond); |
407 | } | |
408 | /* For movcond, we canonicalize the "false" input reg to match | |
409 | the destination reg so that the tcg backend can implement | |
410 | a "move if true" operation. */ | |
411 | if (args[0] == args[3]) { | |
412 | tmp = args[3]; | |
413 | args[3] = args[4]; | |
414 | args[4] = tmp; | |
415 | cond = tcg_invert_cond(cond); | |
b5b6bb16 | 416 | } |
1d465a94 | 417 | args[5] = cond; |
53108fb5 KB |
418 | default: |
419 | break; | |
420 | } | |
421 | ||
52173bab AJ |
422 | /* Simplify expressions for "shift/rot r, 0, a => movi r, 0" */ |
423 | switch (op) { | |
424 | CASE_OP_32_64(shl): | |
425 | CASE_OP_32_64(shr): | |
426 | CASE_OP_32_64(sar): | |
427 | CASE_OP_32_64(rotl): | |
428 | CASE_OP_32_64(rotr): | |
429 | if (temps[args[1]].state == TCG_TEMP_CONST | |
430 | && temps[args[1]].val == 0) { | |
431 | gen_opc_buf[op_index] = op_to_movi(op); | |
432 | tcg_opt_gen_movi(gen_args, args[0], 0, nb_temps, nb_globals); | |
433 | args += 3; | |
434 | gen_args += 2; | |
435 | continue; | |
436 | } | |
437 | break; | |
438 | default: | |
439 | break; | |
440 | } | |
441 | ||
d9693072 | 442 | /* Simplify expression for "op r, a, 0 => mov r, a" cases */ |
53108fb5 KB |
443 | switch (op) { |
444 | CASE_OP_32_64(add): | |
445 | CASE_OP_32_64(sub): | |
55c0975c KB |
446 | CASE_OP_32_64(shl): |
447 | CASE_OP_32_64(shr): | |
448 | CASE_OP_32_64(sar): | |
25c4d9cc RH |
449 | CASE_OP_32_64(rotl): |
450 | CASE_OP_32_64(rotr): | |
6c9785fb AJ |
451 | CASE_OP_32_64(or): |
452 | CASE_OP_32_64(xor): | |
53108fb5 KB |
453 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
454 | /* Proceed with possible constant folding. */ | |
455 | break; | |
456 | } | |
457 | if (temps[args[2]].state == TCG_TEMP_CONST | |
458 | && temps[args[2]].val == 0) { | |
459 | if ((temps[args[0]].state == TCG_TEMP_COPY | |
460 | && temps[args[0]].val == args[1]) | |
461 | || args[0] == args[1]) { | |
53108fb5 KB |
462 | gen_opc_buf[op_index] = INDEX_op_nop; |
463 | } else { | |
464 | gen_opc_buf[op_index] = op_to_mov(op); | |
4f17bc1e | 465 | tcg_opt_gen_mov(gen_args, args[0], args[1], |
53108fb5 KB |
466 | nb_temps, nb_globals); |
467 | gen_args += 2; | |
53108fb5 | 468 | } |
578c7582 | 469 | args += 3; |
53108fb5 KB |
470 | continue; |
471 | } | |
472 | break; | |
d9693072 AJ |
473 | default: |
474 | break; | |
475 | } | |
476 | ||
477 | /* Simplify expression for "op r, a, 0 => movi r, 0" cases */ | |
478 | switch (op) { | |
f66f1dab | 479 | CASE_OP_32_64(and): |
53108fb5 KB |
480 | CASE_OP_32_64(mul): |
481 | if ((temps[args[2]].state == TCG_TEMP_CONST | |
482 | && temps[args[2]].val == 0)) { | |
483 | gen_opc_buf[op_index] = op_to_movi(op); | |
484 | tcg_opt_gen_movi(gen_args, args[0], 0, nb_temps, nb_globals); | |
485 | args += 3; | |
486 | gen_args += 2; | |
487 | continue; | |
488 | } | |
489 | break; | |
d9693072 AJ |
490 | default: |
491 | break; | |
492 | } | |
493 | ||
494 | /* Simplify expression for "op r, a, a => mov r, a" cases */ | |
495 | switch (op) { | |
9a81090b KB |
496 | CASE_OP_32_64(or): |
497 | CASE_OP_32_64(and): | |
498 | if (args[1] == args[2]) { | |
499 | if (args[1] == args[0]) { | |
9a81090b KB |
500 | gen_opc_buf[op_index] = INDEX_op_nop; |
501 | } else { | |
502 | gen_opc_buf[op_index] = op_to_mov(op); | |
4f17bc1e | 503 | tcg_opt_gen_mov(gen_args, args[0], args[1], nb_temps, |
9a81090b KB |
504 | nb_globals); |
505 | gen_args += 2; | |
9a81090b | 506 | } |
578c7582 | 507 | args += 3; |
9a81090b KB |
508 | continue; |
509 | } | |
510 | break; | |
fe0de7aa BS |
511 | default: |
512 | break; | |
53108fb5 KB |
513 | } |
514 | ||
22613af4 KB |
515 | /* Propagate constants through copy operations and do constant |
516 | folding. Constants will be substituted to arguments by register | |
517 | allocator where needed and possible. Also detect copies. */ | |
8f2e8c07 | 518 | switch (op) { |
22613af4 KB |
519 | CASE_OP_32_64(mov): |
520 | if ((temps[args[1]].state == TCG_TEMP_COPY | |
521 | && temps[args[1]].val == args[0]) | |
522 | || args[0] == args[1]) { | |
523 | args += 2; | |
524 | gen_opc_buf[op_index] = INDEX_op_nop; | |
525 | break; | |
526 | } | |
527 | if (temps[args[1]].state != TCG_TEMP_CONST) { | |
4f17bc1e | 528 | tcg_opt_gen_mov(gen_args, args[0], args[1], |
22613af4 KB |
529 | nb_temps, nb_globals); |
530 | gen_args += 2; | |
531 | args += 2; | |
532 | break; | |
533 | } | |
534 | /* Source argument is constant. Rewrite the operation and | |
535 | let movi case handle it. */ | |
536 | op = op_to_movi(op); | |
537 | gen_opc_buf[op_index] = op; | |
538 | args[1] = temps[args[1]].val; | |
539 | /* fallthrough */ | |
540 | CASE_OP_32_64(movi): | |
541 | tcg_opt_gen_movi(gen_args, args[0], args[1], nb_temps, nb_globals); | |
542 | gen_args += 2; | |
543 | args += 2; | |
544 | break; | |
a640f031 | 545 | CASE_OP_32_64(not): |
cb25c80a | 546 | CASE_OP_32_64(neg): |
25c4d9cc RH |
547 | CASE_OP_32_64(ext8s): |
548 | CASE_OP_32_64(ext8u): | |
549 | CASE_OP_32_64(ext16s): | |
550 | CASE_OP_32_64(ext16u): | |
a640f031 KB |
551 | case INDEX_op_ext32s_i64: |
552 | case INDEX_op_ext32u_i64: | |
a640f031 KB |
553 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
554 | gen_opc_buf[op_index] = op_to_movi(op); | |
555 | tmp = do_constant_folding(op, temps[args[1]].val, 0); | |
556 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
a640f031 KB |
557 | } else { |
558 | reset_temp(args[0], nb_temps, nb_globals); | |
559 | gen_args[0] = args[0]; | |
560 | gen_args[1] = args[1]; | |
a640f031 | 561 | } |
578c7582 AJ |
562 | gen_args += 2; |
563 | args += 2; | |
564 | break; | |
53108fb5 KB |
565 | CASE_OP_32_64(add): |
566 | CASE_OP_32_64(sub): | |
567 | CASE_OP_32_64(mul): | |
9a81090b KB |
568 | CASE_OP_32_64(or): |
569 | CASE_OP_32_64(and): | |
570 | CASE_OP_32_64(xor): | |
55c0975c KB |
571 | CASE_OP_32_64(shl): |
572 | CASE_OP_32_64(shr): | |
573 | CASE_OP_32_64(sar): | |
25c4d9cc RH |
574 | CASE_OP_32_64(rotl): |
575 | CASE_OP_32_64(rotr): | |
cb25c80a RH |
576 | CASE_OP_32_64(andc): |
577 | CASE_OP_32_64(orc): | |
578 | CASE_OP_32_64(eqv): | |
579 | CASE_OP_32_64(nand): | |
580 | CASE_OP_32_64(nor): | |
53108fb5 KB |
581 | if (temps[args[1]].state == TCG_TEMP_CONST |
582 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
583 | gen_opc_buf[op_index] = op_to_movi(op); | |
584 | tmp = do_constant_folding(op, temps[args[1]].val, | |
585 | temps[args[2]].val); | |
586 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
587 | gen_args += 2; | |
53108fb5 KB |
588 | } else { |
589 | reset_temp(args[0], nb_temps, nb_globals); | |
590 | gen_args[0] = args[0]; | |
591 | gen_args[1] = args[1]; | |
592 | gen_args[2] = args[2]; | |
593 | gen_args += 3; | |
53108fb5 | 594 | } |
578c7582 AJ |
595 | args += 3; |
596 | break; | |
2108a786 AJ |
597 | CASE_OP_32_64(setcond): |
598 | if (temps[args[1]].state == TCG_TEMP_CONST | |
599 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
600 | gen_opc_buf[op_index] = op_to_movi(op); | |
601 | tmp = do_constant_folding_cond(op, temps[args[1]].val, | |
602 | temps[args[2]].val, args[3]); | |
603 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
604 | gen_args += 2; | |
2108a786 AJ |
605 | } else { |
606 | reset_temp(args[0], nb_temps, nb_globals); | |
607 | gen_args[0] = args[0]; | |
608 | gen_args[1] = args[1]; | |
609 | gen_args[2] = args[2]; | |
610 | gen_args[3] = args[3]; | |
611 | gen_args += 4; | |
2108a786 | 612 | } |
578c7582 AJ |
613 | args += 4; |
614 | break; | |
97c012c7 AJ |
615 | CASE_OP_32_64(brcond): |
616 | if (temps[args[0]].state == TCG_TEMP_CONST | |
617 | && temps[args[1]].state == TCG_TEMP_CONST) { | |
618 | if (do_constant_folding_cond(op, temps[args[0]].val, | |
619 | temps[args[1]].val, args[2])) { | |
620 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
621 | gen_opc_buf[op_index] = INDEX_op_br; | |
622 | gen_args[0] = args[3]; | |
623 | gen_args += 1; | |
97c012c7 AJ |
624 | } else { |
625 | gen_opc_buf[op_index] = INDEX_op_nop; | |
97c012c7 | 626 | } |
97c012c7 AJ |
627 | } else { |
628 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
629 | reset_temp(args[0], nb_temps, nb_globals); | |
630 | gen_args[0] = args[0]; | |
631 | gen_args[1] = args[1]; | |
632 | gen_args[2] = args[2]; | |
633 | gen_args[3] = args[3]; | |
634 | gen_args += 4; | |
97c012c7 | 635 | } |
578c7582 AJ |
636 | args += 4; |
637 | break; | |
b5b6bb16 RH |
638 | CASE_OP_32_64(movcond): |
639 | if (temps[args[1]].state == TCG_TEMP_CONST | |
640 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
641 | tmp = do_constant_folding_cond(op, temps[args[1]].val, | |
642 | temps[args[2]].val, args[5]); | |
643 | if (args[0] == args[4-tmp] | |
644 | || (temps[args[4-tmp]].state == TCG_TEMP_COPY | |
645 | && temps[args[4-tmp]].val == args[0])) { | |
646 | gen_opc_buf[op_index] = INDEX_op_nop; | |
647 | } else if (temps[args[4-tmp]].state == TCG_TEMP_CONST) { | |
648 | gen_opc_buf[op_index] = op_to_movi(op); | |
649 | tcg_opt_gen_movi(gen_args, args[0], temps[args[4-tmp]].val, | |
650 | nb_temps, nb_globals); | |
651 | gen_args += 2; | |
652 | } else { | |
653 | gen_opc_buf[op_index] = op_to_mov(op); | |
654 | tcg_opt_gen_mov(gen_args, args[0], args[4-tmp], | |
655 | nb_temps, nb_globals); | |
656 | gen_args += 2; | |
657 | } | |
658 | } else { | |
659 | reset_temp(args[0], nb_temps, nb_globals); | |
660 | gen_args[0] = args[0]; | |
661 | gen_args[1] = args[1]; | |
662 | gen_args[2] = args[2]; | |
663 | gen_args[3] = args[3]; | |
664 | gen_args[4] = args[4]; | |
665 | gen_args[5] = args[5]; | |
666 | gen_args += 6; | |
667 | } | |
668 | args += 6; | |
669 | break; | |
8f2e8c07 | 670 | case INDEX_op_call: |
22613af4 KB |
671 | nb_call_args = (args[0] >> 16) + (args[0] & 0xffff); |
672 | if (!(args[nb_call_args + 1] & (TCG_CALL_CONST | TCG_CALL_PURE))) { | |
673 | for (i = 0; i < nb_globals; i++) { | |
674 | reset_temp(i, nb_temps, nb_globals); | |
675 | } | |
676 | } | |
677 | for (i = 0; i < (args[0] >> 16); i++) { | |
678 | reset_temp(args[i + 1], nb_temps, nb_globals); | |
679 | } | |
680 | i = nb_call_args + 3; | |
8f2e8c07 KB |
681 | while (i) { |
682 | *gen_args = *args; | |
683 | args++; | |
684 | gen_args++; | |
685 | i--; | |
686 | } | |
687 | break; | |
8f2e8c07 | 688 | default: |
22613af4 | 689 | /* Default case: we do know nothing about operation so no |
5b681fff AJ |
690 | propagation is done. We trash everything if the operation |
691 | is the end of a basic block, otherwise we only trash the | |
692 | output args. */ | |
693 | if (def->flags & TCG_OPF_BB_END) { | |
694 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
695 | } else { | |
696 | for (i = 0; i < def->nb_oargs; i++) { | |
697 | reset_temp(args[i], nb_temps, nb_globals); | |
698 | } | |
22613af4 | 699 | } |
8f2e8c07 KB |
700 | for (i = 0; i < def->nb_args; i++) { |
701 | gen_args[i] = args[i]; | |
702 | } | |
703 | args += def->nb_args; | |
704 | gen_args += def->nb_args; | |
705 | break; | |
706 | } | |
707 | } | |
708 | ||
709 | return gen_args; | |
710 | } | |
711 | ||
712 | TCGArg *tcg_optimize(TCGContext *s, uint16_t *tcg_opc_ptr, | |
713 | TCGArg *args, TCGOpDef *tcg_op_defs) | |
714 | { | |
715 | TCGArg *res; | |
716 | res = tcg_constant_folding(s, tcg_opc_ptr, args, tcg_op_defs); | |
717 | return res; | |
718 | } |