]>
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, | |
22613af4 KB |
42 | } tcg_temp_state; |
43 | ||
44 | struct tcg_temp_info { | |
45 | tcg_temp_state state; | |
46 | uint16_t prev_copy; | |
47 | uint16_t next_copy; | |
48 | tcg_target_ulong val; | |
49 | }; | |
50 | ||
51 | static struct tcg_temp_info temps[TCG_MAX_TEMPS]; | |
52 | ||
e590d4e6 AJ |
53 | /* Reset TEMP's state to TCG_TEMP_UNDEF. If TEMP only had one copy, remove |
54 | the copy flag from the left temp. */ | |
55 | static void reset_temp(TCGArg temp) | |
22613af4 | 56 | { |
e590d4e6 AJ |
57 | if (temps[temp].state == TCG_TEMP_COPY) { |
58 | if (temps[temp].prev_copy == temps[temp].next_copy) { | |
59 | temps[temps[temp].next_copy].state = TCG_TEMP_UNDEF; | |
60 | } else { | |
61 | temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy; | |
62 | temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy; | |
22613af4 | 63 | } |
22613af4 | 64 | } |
48b56ce1 | 65 | temps[temp].state = TCG_TEMP_UNDEF; |
22613af4 KB |
66 | } |
67 | ||
fe0de7aa | 68 | static int op_bits(TCGOpcode op) |
22613af4 | 69 | { |
8399ad59 RH |
70 | const TCGOpDef *def = &tcg_op_defs[op]; |
71 | return def->flags & TCG_OPF_64BIT ? 64 : 32; | |
22613af4 KB |
72 | } |
73 | ||
fe0de7aa | 74 | static TCGOpcode op_to_movi(TCGOpcode op) |
22613af4 KB |
75 | { |
76 | switch (op_bits(op)) { | |
77 | case 32: | |
78 | return INDEX_op_movi_i32; | |
22613af4 KB |
79 | case 64: |
80 | return INDEX_op_movi_i64; | |
22613af4 KB |
81 | default: |
82 | fprintf(stderr, "op_to_movi: unexpected return value of " | |
83 | "function op_bits.\n"); | |
84 | tcg_abort(); | |
85 | } | |
86 | } | |
87 | ||
e590d4e6 AJ |
88 | static TCGArg find_better_copy(TCGContext *s, TCGArg temp) |
89 | { | |
90 | TCGArg i; | |
91 | ||
92 | /* If this is already a global, we can't do better. */ | |
93 | if (temp < s->nb_globals) { | |
94 | return temp; | |
95 | } | |
96 | ||
97 | /* Search for a global first. */ | |
98 | for (i = temps[temp].next_copy ; i != temp ; i = temps[i].next_copy) { | |
99 | if (i < s->nb_globals) { | |
100 | return i; | |
101 | } | |
102 | } | |
103 | ||
104 | /* If it is a temp, search for a temp local. */ | |
105 | if (!s->temps[temp].temp_local) { | |
106 | for (i = temps[temp].next_copy ; i != temp ; i = temps[i].next_copy) { | |
107 | if (s->temps[i].temp_local) { | |
108 | return i; | |
109 | } | |
110 | } | |
111 | } | |
112 | ||
113 | /* Failure to find a better representation, return the same temp. */ | |
114 | return temp; | |
115 | } | |
116 | ||
117 | static bool temps_are_copies(TCGArg arg1, TCGArg arg2) | |
118 | { | |
119 | TCGArg i; | |
120 | ||
121 | if (arg1 == arg2) { | |
122 | return true; | |
123 | } | |
124 | ||
125 | if (temps[arg1].state != TCG_TEMP_COPY | |
126 | || temps[arg2].state != TCG_TEMP_COPY) { | |
127 | return false; | |
128 | } | |
129 | ||
130 | for (i = temps[arg1].next_copy ; i != arg1 ; i = temps[i].next_copy) { | |
131 | if (i == arg2) { | |
132 | return true; | |
133 | } | |
134 | } | |
135 | ||
136 | return false; | |
137 | } | |
138 | ||
b80bb016 AJ |
139 | static void tcg_opt_gen_mov(TCGContext *s, TCGArg *gen_args, |
140 | TCGArg dst, TCGArg src) | |
22613af4 | 141 | { |
e590d4e6 AJ |
142 | reset_temp(dst); |
143 | assert(temps[src].state != TCG_TEMP_CONST); | |
144 | ||
145 | if (s->temps[src].type == s->temps[dst].type) { | |
146 | if (temps[src].state != TCG_TEMP_COPY) { | |
147 | temps[src].state = TCG_TEMP_COPY; | |
22613af4 KB |
148 | temps[src].next_copy = src; |
149 | temps[src].prev_copy = src; | |
150 | } | |
151 | temps[dst].state = TCG_TEMP_COPY; | |
22613af4 KB |
152 | temps[dst].next_copy = temps[src].next_copy; |
153 | temps[dst].prev_copy = src; | |
154 | temps[temps[dst].next_copy].prev_copy = dst; | |
155 | temps[src].next_copy = dst; | |
156 | } | |
e590d4e6 | 157 | |
22613af4 KB |
158 | gen_args[0] = dst; |
159 | gen_args[1] = src; | |
160 | } | |
161 | ||
e590d4e6 | 162 | static void tcg_opt_gen_movi(TCGArg *gen_args, TCGArg dst, TCGArg val) |
22613af4 | 163 | { |
e590d4e6 | 164 | reset_temp(dst); |
22613af4 KB |
165 | temps[dst].state = TCG_TEMP_CONST; |
166 | temps[dst].val = val; | |
167 | gen_args[0] = dst; | |
168 | gen_args[1] = val; | |
169 | } | |
170 | ||
fe0de7aa | 171 | static TCGOpcode op_to_mov(TCGOpcode op) |
53108fb5 KB |
172 | { |
173 | switch (op_bits(op)) { | |
174 | case 32: | |
175 | return INDEX_op_mov_i32; | |
53108fb5 KB |
176 | case 64: |
177 | return INDEX_op_mov_i64; | |
53108fb5 KB |
178 | default: |
179 | fprintf(stderr, "op_to_mov: unexpected return value of " | |
180 | "function op_bits.\n"); | |
181 | tcg_abort(); | |
182 | } | |
183 | } | |
184 | ||
fe0de7aa | 185 | static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y) |
53108fb5 KB |
186 | { |
187 | switch (op) { | |
188 | CASE_OP_32_64(add): | |
189 | return x + y; | |
190 | ||
191 | CASE_OP_32_64(sub): | |
192 | return x - y; | |
193 | ||
194 | CASE_OP_32_64(mul): | |
195 | return x * y; | |
196 | ||
9a81090b KB |
197 | CASE_OP_32_64(and): |
198 | return x & y; | |
199 | ||
200 | CASE_OP_32_64(or): | |
201 | return x | y; | |
202 | ||
203 | CASE_OP_32_64(xor): | |
204 | return x ^ y; | |
205 | ||
55c0975c KB |
206 | case INDEX_op_shl_i32: |
207 | return (uint32_t)x << (uint32_t)y; | |
208 | ||
55c0975c KB |
209 | case INDEX_op_shl_i64: |
210 | return (uint64_t)x << (uint64_t)y; | |
55c0975c KB |
211 | |
212 | case INDEX_op_shr_i32: | |
213 | return (uint32_t)x >> (uint32_t)y; | |
214 | ||
55c0975c KB |
215 | case INDEX_op_shr_i64: |
216 | return (uint64_t)x >> (uint64_t)y; | |
55c0975c KB |
217 | |
218 | case INDEX_op_sar_i32: | |
219 | return (int32_t)x >> (int32_t)y; | |
220 | ||
55c0975c KB |
221 | case INDEX_op_sar_i64: |
222 | return (int64_t)x >> (int64_t)y; | |
55c0975c KB |
223 | |
224 | case INDEX_op_rotr_i32: | |
25c4d9cc | 225 | x = ((uint32_t)x << (32 - y)) | ((uint32_t)x >> y); |
55c0975c KB |
226 | return x; |
227 | ||
55c0975c | 228 | case INDEX_op_rotr_i64: |
25c4d9cc | 229 | x = ((uint64_t)x << (64 - y)) | ((uint64_t)x >> y); |
55c0975c | 230 | return x; |
55c0975c KB |
231 | |
232 | case INDEX_op_rotl_i32: | |
25c4d9cc | 233 | x = ((uint32_t)x << y) | ((uint32_t)x >> (32 - y)); |
55c0975c KB |
234 | return x; |
235 | ||
55c0975c | 236 | case INDEX_op_rotl_i64: |
25c4d9cc | 237 | x = ((uint64_t)x << y) | ((uint64_t)x >> (64 - y)); |
55c0975c | 238 | return x; |
25c4d9cc RH |
239 | |
240 | CASE_OP_32_64(not): | |
a640f031 | 241 | return ~x; |
25c4d9cc | 242 | |
cb25c80a RH |
243 | CASE_OP_32_64(neg): |
244 | return -x; | |
245 | ||
246 | CASE_OP_32_64(andc): | |
247 | return x & ~y; | |
248 | ||
249 | CASE_OP_32_64(orc): | |
250 | return x | ~y; | |
251 | ||
252 | CASE_OP_32_64(eqv): | |
253 | return ~(x ^ y); | |
254 | ||
255 | CASE_OP_32_64(nand): | |
256 | return ~(x & y); | |
257 | ||
258 | CASE_OP_32_64(nor): | |
259 | return ~(x | y); | |
260 | ||
25c4d9cc | 261 | CASE_OP_32_64(ext8s): |
a640f031 | 262 | return (int8_t)x; |
25c4d9cc RH |
263 | |
264 | CASE_OP_32_64(ext16s): | |
a640f031 | 265 | return (int16_t)x; |
25c4d9cc RH |
266 | |
267 | CASE_OP_32_64(ext8u): | |
a640f031 | 268 | return (uint8_t)x; |
25c4d9cc RH |
269 | |
270 | CASE_OP_32_64(ext16u): | |
a640f031 KB |
271 | return (uint16_t)x; |
272 | ||
a640f031 KB |
273 | case INDEX_op_ext32s_i64: |
274 | return (int32_t)x; | |
275 | ||
276 | case INDEX_op_ext32u_i64: | |
277 | return (uint32_t)x; | |
a640f031 | 278 | |
53108fb5 KB |
279 | default: |
280 | fprintf(stderr, | |
281 | "Unrecognized operation %d in do_constant_folding.\n", op); | |
282 | tcg_abort(); | |
283 | } | |
284 | } | |
285 | ||
fe0de7aa | 286 | static TCGArg do_constant_folding(TCGOpcode op, TCGArg x, TCGArg y) |
53108fb5 KB |
287 | { |
288 | TCGArg res = do_constant_folding_2(op, x, y); | |
53108fb5 KB |
289 | if (op_bits(op) == 32) { |
290 | res &= 0xffffffff; | |
291 | } | |
53108fb5 KB |
292 | return res; |
293 | } | |
294 | ||
b336ceb6 AJ |
295 | /* Return 2 if the condition can't be simplified, and the result |
296 | of the condition (0 or 1) if it can */ | |
f8dd19e5 AJ |
297 | static TCGArg do_constant_folding_cond(TCGOpcode op, TCGArg x, |
298 | TCGArg y, TCGCond c) | |
299 | { | |
b336ceb6 AJ |
300 | if (temps[x].state == TCG_TEMP_CONST && temps[y].state == TCG_TEMP_CONST) { |
301 | switch (op_bits(op)) { | |
302 | case 32: | |
303 | switch (c) { | |
304 | case TCG_COND_EQ: | |
305 | return (uint32_t)temps[x].val == (uint32_t)temps[y].val; | |
306 | case TCG_COND_NE: | |
307 | return (uint32_t)temps[x].val != (uint32_t)temps[y].val; | |
308 | case TCG_COND_LT: | |
309 | return (int32_t)temps[x].val < (int32_t)temps[y].val; | |
310 | case TCG_COND_GE: | |
311 | return (int32_t)temps[x].val >= (int32_t)temps[y].val; | |
312 | case TCG_COND_LE: | |
313 | return (int32_t)temps[x].val <= (int32_t)temps[y].val; | |
314 | case TCG_COND_GT: | |
315 | return (int32_t)temps[x].val > (int32_t)temps[y].val; | |
316 | case TCG_COND_LTU: | |
317 | return (uint32_t)temps[x].val < (uint32_t)temps[y].val; | |
318 | case TCG_COND_GEU: | |
319 | return (uint32_t)temps[x].val >= (uint32_t)temps[y].val; | |
320 | case TCG_COND_LEU: | |
321 | return (uint32_t)temps[x].val <= (uint32_t)temps[y].val; | |
322 | case TCG_COND_GTU: | |
323 | return (uint32_t)temps[x].val > (uint32_t)temps[y].val; | |
0aed257f RH |
324 | default: |
325 | break; | |
b336ceb6 AJ |
326 | } |
327 | break; | |
328 | case 64: | |
329 | switch (c) { | |
330 | case TCG_COND_EQ: | |
331 | return (uint64_t)temps[x].val == (uint64_t)temps[y].val; | |
332 | case TCG_COND_NE: | |
333 | return (uint64_t)temps[x].val != (uint64_t)temps[y].val; | |
334 | case TCG_COND_LT: | |
335 | return (int64_t)temps[x].val < (int64_t)temps[y].val; | |
336 | case TCG_COND_GE: | |
337 | return (int64_t)temps[x].val >= (int64_t)temps[y].val; | |
338 | case TCG_COND_LE: | |
339 | return (int64_t)temps[x].val <= (int64_t)temps[y].val; | |
340 | case TCG_COND_GT: | |
341 | return (int64_t)temps[x].val > (int64_t)temps[y].val; | |
342 | case TCG_COND_LTU: | |
343 | return (uint64_t)temps[x].val < (uint64_t)temps[y].val; | |
344 | case TCG_COND_GEU: | |
345 | return (uint64_t)temps[x].val >= (uint64_t)temps[y].val; | |
346 | case TCG_COND_LEU: | |
347 | return (uint64_t)temps[x].val <= (uint64_t)temps[y].val; | |
348 | case TCG_COND_GTU: | |
349 | return (uint64_t)temps[x].val > (uint64_t)temps[y].val; | |
0aed257f RH |
350 | default: |
351 | break; | |
b336ceb6 AJ |
352 | } |
353 | break; | |
354 | } | |
355 | } else if (temps_are_copies(x, y)) { | |
f8dd19e5 | 356 | switch (c) { |
f8dd19e5 | 357 | case TCG_COND_GT: |
f8dd19e5 | 358 | case TCG_COND_LTU: |
b336ceb6 | 359 | case TCG_COND_LT: |
f8dd19e5 | 360 | case TCG_COND_GTU: |
f8dd19e5 | 361 | case TCG_COND_NE: |
b336ceb6 | 362 | return 0; |
f8dd19e5 | 363 | case TCG_COND_GE: |
b336ceb6 | 364 | case TCG_COND_GEU: |
f8dd19e5 | 365 | case TCG_COND_LE: |
b336ceb6 AJ |
366 | case TCG_COND_LEU: |
367 | case TCG_COND_EQ: | |
368 | return 1; | |
0aed257f RH |
369 | default: |
370 | break; | |
b336ceb6 AJ |
371 | } |
372 | } else if (temps[y].state == TCG_TEMP_CONST && temps[y].val == 0) { | |
373 | switch (c) { | |
f8dd19e5 | 374 | case TCG_COND_LTU: |
b336ceb6 | 375 | return 0; |
f8dd19e5 | 376 | case TCG_COND_GEU: |
b336ceb6 AJ |
377 | return 1; |
378 | default: | |
379 | return 2; | |
f8dd19e5 | 380 | } |
b336ceb6 AJ |
381 | } else { |
382 | return 2; | |
f8dd19e5 AJ |
383 | } |
384 | ||
385 | fprintf(stderr, | |
386 | "Unrecognized bitness %d or condition %d in " | |
387 | "do_constant_folding_cond.\n", op_bits(op), c); | |
388 | tcg_abort(); | |
389 | } | |
390 | ||
24c9ae4e RH |
391 | static bool swap_commutative(TCGArg dest, TCGArg *p1, TCGArg *p2) |
392 | { | |
393 | TCGArg a1 = *p1, a2 = *p2; | |
394 | int sum = 0; | |
395 | sum += temps[a1].state == TCG_TEMP_CONST; | |
396 | sum -= temps[a2].state == TCG_TEMP_CONST; | |
397 | ||
398 | /* Prefer the constant in second argument, and then the form | |
399 | op a, a, b, which is better handled on non-RISC hosts. */ | |
400 | if (sum > 0 || (sum == 0 && dest == a2)) { | |
401 | *p1 = a2; | |
402 | *p2 = a1; | |
403 | return true; | |
404 | } | |
405 | return false; | |
406 | } | |
407 | ||
22613af4 | 408 | /* Propagate constants and copies, fold constant expressions. */ |
8f2e8c07 KB |
409 | static TCGArg *tcg_constant_folding(TCGContext *s, uint16_t *tcg_opc_ptr, |
410 | TCGArg *args, TCGOpDef *tcg_op_defs) | |
411 | { | |
fe0de7aa BS |
412 | int i, nb_ops, op_index, nb_temps, nb_globals, nb_call_args; |
413 | TCGOpcode op; | |
8f2e8c07 KB |
414 | const TCGOpDef *def; |
415 | TCGArg *gen_args; | |
53108fb5 | 416 | TCGArg tmp; |
5d8f5363 | 417 | |
22613af4 KB |
418 | /* Array VALS has an element for each temp. |
419 | If this temp holds a constant then its value is kept in VALS' element. | |
e590d4e6 AJ |
420 | If this temp is a copy of other ones then the other copies are |
421 | available through the doubly linked circular list. */ | |
8f2e8c07 KB |
422 | |
423 | nb_temps = s->nb_temps; | |
424 | nb_globals = s->nb_globals; | |
22613af4 | 425 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); |
8f2e8c07 KB |
426 | |
427 | nb_ops = tcg_opc_ptr - gen_opc_buf; | |
428 | gen_args = args; | |
429 | for (op_index = 0; op_index < nb_ops; op_index++) { | |
430 | op = gen_opc_buf[op_index]; | |
431 | def = &tcg_op_defs[op]; | |
22613af4 | 432 | /* Do copy propagation */ |
1ff8c541 AJ |
433 | if (op == INDEX_op_call) { |
434 | int nb_oargs = args[0] >> 16; | |
435 | int nb_iargs = args[0] & 0xffff; | |
436 | for (i = nb_oargs + 1; i < nb_oargs + nb_iargs + 1; i++) { | |
437 | if (temps[args[i]].state == TCG_TEMP_COPY) { | |
438 | args[i] = find_better_copy(s, args[i]); | |
439 | } | |
440 | } | |
441 | } else { | |
22613af4 KB |
442 | for (i = def->nb_oargs; i < def->nb_oargs + def->nb_iargs; i++) { |
443 | if (temps[args[i]].state == TCG_TEMP_COPY) { | |
e590d4e6 | 444 | args[i] = find_better_copy(s, args[i]); |
22613af4 KB |
445 | } |
446 | } | |
447 | } | |
448 | ||
53108fb5 KB |
449 | /* For commutative operations make constant second argument */ |
450 | switch (op) { | |
451 | CASE_OP_32_64(add): | |
452 | CASE_OP_32_64(mul): | |
9a81090b KB |
453 | CASE_OP_32_64(and): |
454 | CASE_OP_32_64(or): | |
455 | CASE_OP_32_64(xor): | |
cb25c80a RH |
456 | CASE_OP_32_64(eqv): |
457 | CASE_OP_32_64(nand): | |
458 | CASE_OP_32_64(nor): | |
24c9ae4e | 459 | swap_commutative(args[0], &args[1], &args[2]); |
53108fb5 | 460 | break; |
65a7cce1 | 461 | CASE_OP_32_64(brcond): |
24c9ae4e | 462 | if (swap_commutative(-1, &args[0], &args[1])) { |
65a7cce1 AJ |
463 | args[2] = tcg_swap_cond(args[2]); |
464 | } | |
465 | break; | |
466 | CASE_OP_32_64(setcond): | |
24c9ae4e | 467 | if (swap_commutative(args[0], &args[1], &args[2])) { |
65a7cce1 AJ |
468 | args[3] = tcg_swap_cond(args[3]); |
469 | } | |
470 | break; | |
fa01a208 | 471 | CASE_OP_32_64(movcond): |
24c9ae4e RH |
472 | if (swap_commutative(-1, &args[1], &args[2])) { |
473 | args[5] = tcg_swap_cond(args[5]); | |
5d8f5363 RH |
474 | } |
475 | /* For movcond, we canonicalize the "false" input reg to match | |
476 | the destination reg so that the tcg backend can implement | |
477 | a "move if true" operation. */ | |
24c9ae4e RH |
478 | if (swap_commutative(args[0], &args[4], &args[3])) { |
479 | args[5] = tcg_invert_cond(args[5]); | |
fa01a208 | 480 | } |
53108fb5 KB |
481 | default: |
482 | break; | |
483 | } | |
484 | ||
01ee5282 AJ |
485 | /* Simplify expressions for "shift/rot r, 0, a => movi r, 0" */ |
486 | switch (op) { | |
487 | CASE_OP_32_64(shl): | |
488 | CASE_OP_32_64(shr): | |
489 | CASE_OP_32_64(sar): | |
490 | CASE_OP_32_64(rotl): | |
491 | CASE_OP_32_64(rotr): | |
492 | if (temps[args[1]].state == TCG_TEMP_CONST | |
493 | && temps[args[1]].val == 0) { | |
494 | gen_opc_buf[op_index] = op_to_movi(op); | |
e590d4e6 | 495 | tcg_opt_gen_movi(gen_args, args[0], 0); |
01ee5282 AJ |
496 | args += 3; |
497 | gen_args += 2; | |
498 | continue; | |
499 | } | |
500 | break; | |
501 | default: | |
502 | break; | |
503 | } | |
504 | ||
56e49438 | 505 | /* Simplify expression for "op r, a, 0 => mov r, a" cases */ |
53108fb5 KB |
506 | switch (op) { |
507 | CASE_OP_32_64(add): | |
508 | CASE_OP_32_64(sub): | |
55c0975c KB |
509 | CASE_OP_32_64(shl): |
510 | CASE_OP_32_64(shr): | |
511 | CASE_OP_32_64(sar): | |
25c4d9cc RH |
512 | CASE_OP_32_64(rotl): |
513 | CASE_OP_32_64(rotr): | |
38ee188b AJ |
514 | CASE_OP_32_64(or): |
515 | CASE_OP_32_64(xor): | |
53108fb5 KB |
516 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
517 | /* Proceed with possible constant folding. */ | |
518 | break; | |
519 | } | |
520 | if (temps[args[2]].state == TCG_TEMP_CONST | |
521 | && temps[args[2]].val == 0) { | |
e590d4e6 | 522 | if (temps_are_copies(args[0], args[1])) { |
53108fb5 KB |
523 | gen_opc_buf[op_index] = INDEX_op_nop; |
524 | } else { | |
525 | gen_opc_buf[op_index] = op_to_mov(op); | |
b80bb016 | 526 | tcg_opt_gen_mov(s, gen_args, args[0], args[1]); |
53108fb5 | 527 | gen_args += 2; |
53108fb5 | 528 | } |
fedc0da2 | 529 | args += 3; |
53108fb5 KB |
530 | continue; |
531 | } | |
532 | break; | |
56e49438 AJ |
533 | default: |
534 | break; | |
535 | } | |
536 | ||
537 | /* Simplify expression for "op r, a, 0 => movi r, 0" cases */ | |
538 | switch (op) { | |
61251c0c | 539 | CASE_OP_32_64(and): |
53108fb5 KB |
540 | CASE_OP_32_64(mul): |
541 | if ((temps[args[2]].state == TCG_TEMP_CONST | |
542 | && temps[args[2]].val == 0)) { | |
543 | gen_opc_buf[op_index] = op_to_movi(op); | |
e590d4e6 | 544 | tcg_opt_gen_movi(gen_args, args[0], 0); |
53108fb5 KB |
545 | args += 3; |
546 | gen_args += 2; | |
547 | continue; | |
548 | } | |
549 | break; | |
56e49438 AJ |
550 | default: |
551 | break; | |
552 | } | |
553 | ||
554 | /* Simplify expression for "op r, a, a => mov r, a" cases */ | |
555 | switch (op) { | |
9a81090b KB |
556 | CASE_OP_32_64(or): |
557 | CASE_OP_32_64(and): | |
0aba1c73 | 558 | if (temps_are_copies(args[1], args[2])) { |
e590d4e6 | 559 | if (temps_are_copies(args[0], args[1])) { |
9a81090b KB |
560 | gen_opc_buf[op_index] = INDEX_op_nop; |
561 | } else { | |
562 | gen_opc_buf[op_index] = op_to_mov(op); | |
b80bb016 | 563 | tcg_opt_gen_mov(s, gen_args, args[0], args[1]); |
9a81090b | 564 | gen_args += 2; |
9a81090b | 565 | } |
fedc0da2 | 566 | args += 3; |
9a81090b KB |
567 | continue; |
568 | } | |
569 | break; | |
fe0de7aa BS |
570 | default: |
571 | break; | |
53108fb5 KB |
572 | } |
573 | ||
3c94193e AJ |
574 | /* Simplify expression for "op r, a, a => movi r, 0" cases */ |
575 | switch (op) { | |
576 | CASE_OP_32_64(sub): | |
577 | CASE_OP_32_64(xor): | |
578 | if (temps_are_copies(args[1], args[2])) { | |
579 | gen_opc_buf[op_index] = op_to_movi(op); | |
580 | tcg_opt_gen_movi(gen_args, args[0], 0); | |
581 | gen_args += 2; | |
582 | args += 3; | |
583 | continue; | |
584 | } | |
585 | break; | |
586 | default: | |
587 | break; | |
588 | } | |
589 | ||
22613af4 KB |
590 | /* Propagate constants through copy operations and do constant |
591 | folding. Constants will be substituted to arguments by register | |
592 | allocator where needed and possible. Also detect copies. */ | |
8f2e8c07 | 593 | switch (op) { |
22613af4 | 594 | CASE_OP_32_64(mov): |
e590d4e6 | 595 | if (temps_are_copies(args[0], args[1])) { |
22613af4 KB |
596 | args += 2; |
597 | gen_opc_buf[op_index] = INDEX_op_nop; | |
598 | break; | |
599 | } | |
600 | if (temps[args[1]].state != TCG_TEMP_CONST) { | |
b80bb016 | 601 | tcg_opt_gen_mov(s, gen_args, args[0], args[1]); |
22613af4 KB |
602 | gen_args += 2; |
603 | args += 2; | |
604 | break; | |
605 | } | |
606 | /* Source argument is constant. Rewrite the operation and | |
607 | let movi case handle it. */ | |
608 | op = op_to_movi(op); | |
609 | gen_opc_buf[op_index] = op; | |
610 | args[1] = temps[args[1]].val; | |
611 | /* fallthrough */ | |
612 | CASE_OP_32_64(movi): | |
e590d4e6 | 613 | tcg_opt_gen_movi(gen_args, args[0], args[1]); |
22613af4 KB |
614 | gen_args += 2; |
615 | args += 2; | |
616 | break; | |
a640f031 | 617 | CASE_OP_32_64(not): |
cb25c80a | 618 | CASE_OP_32_64(neg): |
25c4d9cc RH |
619 | CASE_OP_32_64(ext8s): |
620 | CASE_OP_32_64(ext8u): | |
621 | CASE_OP_32_64(ext16s): | |
622 | CASE_OP_32_64(ext16u): | |
a640f031 KB |
623 | case INDEX_op_ext32s_i64: |
624 | case INDEX_op_ext32u_i64: | |
a640f031 KB |
625 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
626 | gen_opc_buf[op_index] = op_to_movi(op); | |
627 | tmp = do_constant_folding(op, temps[args[1]].val, 0); | |
e590d4e6 | 628 | tcg_opt_gen_movi(gen_args, args[0], tmp); |
a640f031 | 629 | } else { |
e590d4e6 | 630 | reset_temp(args[0]); |
a640f031 KB |
631 | gen_args[0] = args[0]; |
632 | gen_args[1] = args[1]; | |
a640f031 | 633 | } |
fedc0da2 AJ |
634 | gen_args += 2; |
635 | args += 2; | |
636 | break; | |
53108fb5 KB |
637 | CASE_OP_32_64(add): |
638 | CASE_OP_32_64(sub): | |
639 | CASE_OP_32_64(mul): | |
9a81090b KB |
640 | CASE_OP_32_64(or): |
641 | CASE_OP_32_64(and): | |
642 | CASE_OP_32_64(xor): | |
55c0975c KB |
643 | CASE_OP_32_64(shl): |
644 | CASE_OP_32_64(shr): | |
645 | CASE_OP_32_64(sar): | |
25c4d9cc RH |
646 | CASE_OP_32_64(rotl): |
647 | CASE_OP_32_64(rotr): | |
cb25c80a RH |
648 | CASE_OP_32_64(andc): |
649 | CASE_OP_32_64(orc): | |
650 | CASE_OP_32_64(eqv): | |
651 | CASE_OP_32_64(nand): | |
652 | CASE_OP_32_64(nor): | |
53108fb5 KB |
653 | if (temps[args[1]].state == TCG_TEMP_CONST |
654 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
655 | gen_opc_buf[op_index] = op_to_movi(op); | |
656 | tmp = do_constant_folding(op, temps[args[1]].val, | |
657 | temps[args[2]].val); | |
e590d4e6 | 658 | tcg_opt_gen_movi(gen_args, args[0], tmp); |
53108fb5 | 659 | gen_args += 2; |
53108fb5 | 660 | } else { |
e590d4e6 | 661 | reset_temp(args[0]); |
53108fb5 KB |
662 | gen_args[0] = args[0]; |
663 | gen_args[1] = args[1]; | |
664 | gen_args[2] = args[2]; | |
665 | gen_args += 3; | |
53108fb5 | 666 | } |
fedc0da2 AJ |
667 | args += 3; |
668 | break; | |
7ef55fc9 AJ |
669 | CASE_OP_32_64(deposit): |
670 | if (temps[args[1]].state == TCG_TEMP_CONST | |
671 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
672 | gen_opc_buf[op_index] = op_to_movi(op); | |
673 | tmp = ((1ull << args[4]) - 1); | |
674 | tmp = (temps[args[1]].val & ~(tmp << args[3])) | |
675 | | ((temps[args[2]].val & tmp) << args[3]); | |
676 | tcg_opt_gen_movi(gen_args, args[0], tmp); | |
677 | gen_args += 2; | |
678 | } else { | |
679 | reset_temp(args[0]); | |
680 | gen_args[0] = args[0]; | |
681 | gen_args[1] = args[1]; | |
682 | gen_args[2] = args[2]; | |
683 | gen_args[3] = args[3]; | |
684 | gen_args[4] = args[4]; | |
685 | gen_args += 5; | |
686 | } | |
687 | args += 5; | |
688 | break; | |
f8dd19e5 | 689 | CASE_OP_32_64(setcond): |
b336ceb6 AJ |
690 | tmp = do_constant_folding_cond(op, args[1], args[2], args[3]); |
691 | if (tmp != 2) { | |
f8dd19e5 | 692 | gen_opc_buf[op_index] = op_to_movi(op); |
e590d4e6 | 693 | tcg_opt_gen_movi(gen_args, args[0], tmp); |
f8dd19e5 | 694 | gen_args += 2; |
f8dd19e5 | 695 | } else { |
e590d4e6 | 696 | reset_temp(args[0]); |
f8dd19e5 AJ |
697 | gen_args[0] = args[0]; |
698 | gen_args[1] = args[1]; | |
699 | gen_args[2] = args[2]; | |
700 | gen_args[3] = args[3]; | |
701 | gen_args += 4; | |
f8dd19e5 | 702 | } |
fedc0da2 AJ |
703 | args += 4; |
704 | break; | |
fbeaa26c | 705 | CASE_OP_32_64(brcond): |
b336ceb6 AJ |
706 | tmp = do_constant_folding_cond(op, args[0], args[1], args[2]); |
707 | if (tmp != 2) { | |
708 | if (tmp) { | |
fbeaa26c AJ |
709 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); |
710 | gen_opc_buf[op_index] = INDEX_op_br; | |
711 | gen_args[0] = args[3]; | |
712 | gen_args += 1; | |
fbeaa26c AJ |
713 | } else { |
714 | gen_opc_buf[op_index] = INDEX_op_nop; | |
fbeaa26c | 715 | } |
fbeaa26c AJ |
716 | } else { |
717 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
e590d4e6 | 718 | reset_temp(args[0]); |
fbeaa26c AJ |
719 | gen_args[0] = args[0]; |
720 | gen_args[1] = args[1]; | |
721 | gen_args[2] = args[2]; | |
722 | gen_args[3] = args[3]; | |
723 | gen_args += 4; | |
fbeaa26c | 724 | } |
fedc0da2 AJ |
725 | args += 4; |
726 | break; | |
fa01a208 | 727 | CASE_OP_32_64(movcond): |
b336ceb6 AJ |
728 | tmp = do_constant_folding_cond(op, args[1], args[2], args[5]); |
729 | if (tmp != 2) { | |
e590d4e6 | 730 | if (temps_are_copies(args[0], args[4-tmp])) { |
fa01a208 RH |
731 | gen_opc_buf[op_index] = INDEX_op_nop; |
732 | } else if (temps[args[4-tmp]].state == TCG_TEMP_CONST) { | |
733 | gen_opc_buf[op_index] = op_to_movi(op); | |
e590d4e6 | 734 | tcg_opt_gen_movi(gen_args, args[0], temps[args[4-tmp]].val); |
fa01a208 RH |
735 | gen_args += 2; |
736 | } else { | |
737 | gen_opc_buf[op_index] = op_to_mov(op); | |
e590d4e6 | 738 | tcg_opt_gen_mov(s, gen_args, args[0], args[4-tmp]); |
fa01a208 RH |
739 | gen_args += 2; |
740 | } | |
741 | } else { | |
e590d4e6 | 742 | reset_temp(args[0]); |
fa01a208 RH |
743 | gen_args[0] = args[0]; |
744 | gen_args[1] = args[1]; | |
745 | gen_args[2] = args[2]; | |
746 | gen_args[3] = args[3]; | |
747 | gen_args[4] = args[4]; | |
748 | gen_args[5] = args[5]; | |
749 | gen_args += 6; | |
750 | } | |
751 | args += 6; | |
752 | break; | |
8f2e8c07 | 753 | case INDEX_op_call: |
22613af4 KB |
754 | nb_call_args = (args[0] >> 16) + (args[0] & 0xffff); |
755 | if (!(args[nb_call_args + 1] & (TCG_CALL_CONST | TCG_CALL_PURE))) { | |
756 | for (i = 0; i < nb_globals; i++) { | |
e590d4e6 | 757 | reset_temp(i); |
22613af4 KB |
758 | } |
759 | } | |
760 | for (i = 0; i < (args[0] >> 16); i++) { | |
e590d4e6 | 761 | reset_temp(args[i + 1]); |
22613af4 KB |
762 | } |
763 | i = nb_call_args + 3; | |
8f2e8c07 KB |
764 | while (i) { |
765 | *gen_args = *args; | |
766 | args++; | |
767 | gen_args++; | |
768 | i--; | |
769 | } | |
770 | break; | |
8f2e8c07 | 771 | default: |
22613af4 | 772 | /* Default case: we do know nothing about operation so no |
a2550660 AJ |
773 | propagation is done. We trash everything if the operation |
774 | is the end of a basic block, otherwise we only trash the | |
775 | output args. */ | |
776 | if (def->flags & TCG_OPF_BB_END) { | |
777 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
778 | } else { | |
779 | for (i = 0; i < def->nb_oargs; i++) { | |
e590d4e6 | 780 | reset_temp(args[i]); |
a2550660 | 781 | } |
22613af4 | 782 | } |
8f2e8c07 KB |
783 | for (i = 0; i < def->nb_args; i++) { |
784 | gen_args[i] = args[i]; | |
785 | } | |
786 | args += def->nb_args; | |
787 | gen_args += def->nb_args; | |
788 | break; | |
789 | } | |
790 | } | |
791 | ||
792 | return gen_args; | |
793 | } | |
794 | ||
795 | TCGArg *tcg_optimize(TCGContext *s, uint16_t *tcg_opc_ptr, | |
796 | TCGArg *args, TCGOpDef *tcg_op_defs) | |
797 | { | |
798 | TCGArg *res; | |
799 | res = tcg_constant_folding(s, tcg_opc_ptr, args, tcg_op_defs); | |
800 | return res; | |
801 | } |