]>
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, | |
ad797145 | 42 | TCG_TEMP_HAS_COPY |
22613af4 KB |
43 | } tcg_temp_state; |
44 | ||
45 | struct tcg_temp_info { | |
46 | tcg_temp_state state; | |
47 | uint16_t prev_copy; | |
48 | uint16_t next_copy; | |
49 | tcg_target_ulong val; | |
50 | }; | |
51 | ||
52 | static struct tcg_temp_info temps[TCG_MAX_TEMPS]; | |
53 | ||
ad797145 | 54 | /* Reset TEMP's state to TCG_TEMP_UNDEF. If TEMP was a representative of some |
22613af4 KB |
55 | class of equivalent temp's, a new representative should be chosen in this |
56 | class. */ | |
57 | static void reset_temp(TCGArg temp, int nb_temps, int nb_globals) | |
58 | { | |
59 | int i; | |
60 | TCGArg new_base = (TCGArg)-1; | |
61 | if (temps[temp].state == TCG_TEMP_HAS_COPY) { | |
62 | for (i = temps[temp].next_copy; i != temp; i = temps[i].next_copy) { | |
63 | if (i >= nb_globals) { | |
64 | temps[i].state = TCG_TEMP_HAS_COPY; | |
65 | new_base = i; | |
66 | break; | |
67 | } | |
68 | } | |
69 | for (i = temps[temp].next_copy; i != temp; i = temps[i].next_copy) { | |
70 | if (new_base == (TCGArg)-1) { | |
ad797145 | 71 | temps[i].state = TCG_TEMP_UNDEF; |
22613af4 KB |
72 | } else { |
73 | temps[i].val = new_base; | |
74 | } | |
75 | } | |
76 | temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy; | |
77 | temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy; | |
78 | } else if (temps[temp].state == TCG_TEMP_COPY) { | |
79 | temps[temps[temp].next_copy].prev_copy = temps[temp].prev_copy; | |
80 | temps[temps[temp].prev_copy].next_copy = temps[temp].next_copy; | |
81 | new_base = temps[temp].val; | |
82 | } | |
ad797145 | 83 | temps[temp].state = TCG_TEMP_UNDEF; |
22613af4 | 84 | if (new_base != (TCGArg)-1 && temps[new_base].next_copy == new_base) { |
ad797145 | 85 | temps[new_base].state = TCG_TEMP_UNDEF; |
22613af4 KB |
86 | } |
87 | } | |
88 | ||
fe0de7aa | 89 | static int op_bits(TCGOpcode op) |
22613af4 | 90 | { |
8399ad59 RH |
91 | const TCGOpDef *def = &tcg_op_defs[op]; |
92 | return def->flags & TCG_OPF_64BIT ? 64 : 32; | |
22613af4 KB |
93 | } |
94 | ||
fe0de7aa | 95 | static TCGOpcode op_to_movi(TCGOpcode op) |
22613af4 KB |
96 | { |
97 | switch (op_bits(op)) { | |
98 | case 32: | |
99 | return INDEX_op_movi_i32; | |
22613af4 KB |
100 | case 64: |
101 | return INDEX_op_movi_i64; | |
22613af4 KB |
102 | default: |
103 | fprintf(stderr, "op_to_movi: unexpected return value of " | |
104 | "function op_bits.\n"); | |
105 | tcg_abort(); | |
106 | } | |
107 | } | |
108 | ||
4f17bc1e AJ |
109 | static void tcg_opt_gen_mov(TCGArg *gen_args, TCGArg dst, TCGArg src, |
110 | int nb_temps, int nb_globals) | |
22613af4 KB |
111 | { |
112 | reset_temp(dst, nb_temps, nb_globals); | |
113 | assert(temps[src].state != TCG_TEMP_COPY); | |
4f17bc1e | 114 | if (src >= nb_globals) { |
22613af4 KB |
115 | assert(temps[src].state != TCG_TEMP_CONST); |
116 | if (temps[src].state != TCG_TEMP_HAS_COPY) { | |
117 | temps[src].state = TCG_TEMP_HAS_COPY; | |
118 | temps[src].next_copy = src; | |
119 | temps[src].prev_copy = src; | |
120 | } | |
121 | temps[dst].state = TCG_TEMP_COPY; | |
122 | temps[dst].val = src; | |
123 | temps[dst].next_copy = temps[src].next_copy; | |
124 | temps[dst].prev_copy = src; | |
125 | temps[temps[dst].next_copy].prev_copy = dst; | |
126 | temps[src].next_copy = dst; | |
127 | } | |
128 | gen_args[0] = dst; | |
129 | gen_args[1] = src; | |
130 | } | |
131 | ||
132 | static void tcg_opt_gen_movi(TCGArg *gen_args, TCGArg dst, TCGArg val, | |
133 | int nb_temps, int nb_globals) | |
134 | { | |
135 | reset_temp(dst, nb_temps, nb_globals); | |
136 | temps[dst].state = TCG_TEMP_CONST; | |
137 | temps[dst].val = val; | |
138 | gen_args[0] = dst; | |
139 | gen_args[1] = val; | |
140 | } | |
141 | ||
fe0de7aa | 142 | static TCGOpcode op_to_mov(TCGOpcode op) |
53108fb5 KB |
143 | { |
144 | switch (op_bits(op)) { | |
145 | case 32: | |
146 | return INDEX_op_mov_i32; | |
53108fb5 KB |
147 | case 64: |
148 | return INDEX_op_mov_i64; | |
53108fb5 KB |
149 | default: |
150 | fprintf(stderr, "op_to_mov: unexpected return value of " | |
151 | "function op_bits.\n"); | |
152 | tcg_abort(); | |
153 | } | |
154 | } | |
155 | ||
fe0de7aa | 156 | static TCGArg do_constant_folding_2(TCGOpcode op, TCGArg x, TCGArg y) |
53108fb5 KB |
157 | { |
158 | switch (op) { | |
159 | CASE_OP_32_64(add): | |
160 | return x + y; | |
161 | ||
162 | CASE_OP_32_64(sub): | |
163 | return x - y; | |
164 | ||
165 | CASE_OP_32_64(mul): | |
166 | return x * y; | |
167 | ||
9a81090b KB |
168 | CASE_OP_32_64(and): |
169 | return x & y; | |
170 | ||
171 | CASE_OP_32_64(or): | |
172 | return x | y; | |
173 | ||
174 | CASE_OP_32_64(xor): | |
175 | return x ^ y; | |
176 | ||
55c0975c KB |
177 | case INDEX_op_shl_i32: |
178 | return (uint32_t)x << (uint32_t)y; | |
179 | ||
55c0975c KB |
180 | case INDEX_op_shl_i64: |
181 | return (uint64_t)x << (uint64_t)y; | |
55c0975c KB |
182 | |
183 | case INDEX_op_shr_i32: | |
184 | return (uint32_t)x >> (uint32_t)y; | |
185 | ||
55c0975c KB |
186 | case INDEX_op_shr_i64: |
187 | return (uint64_t)x >> (uint64_t)y; | |
55c0975c KB |
188 | |
189 | case INDEX_op_sar_i32: | |
190 | return (int32_t)x >> (int32_t)y; | |
191 | ||
55c0975c KB |
192 | case INDEX_op_sar_i64: |
193 | return (int64_t)x >> (int64_t)y; | |
55c0975c KB |
194 | |
195 | case INDEX_op_rotr_i32: | |
25c4d9cc | 196 | x = ((uint32_t)x << (32 - y)) | ((uint32_t)x >> y); |
55c0975c KB |
197 | return x; |
198 | ||
55c0975c | 199 | case INDEX_op_rotr_i64: |
25c4d9cc | 200 | x = ((uint64_t)x << (64 - y)) | ((uint64_t)x >> y); |
55c0975c | 201 | return x; |
55c0975c KB |
202 | |
203 | case INDEX_op_rotl_i32: | |
25c4d9cc | 204 | x = ((uint32_t)x << y) | ((uint32_t)x >> (32 - y)); |
55c0975c KB |
205 | return x; |
206 | ||
55c0975c | 207 | case INDEX_op_rotl_i64: |
25c4d9cc | 208 | x = ((uint64_t)x << y) | ((uint64_t)x >> (64 - y)); |
55c0975c | 209 | return x; |
25c4d9cc RH |
210 | |
211 | CASE_OP_32_64(not): | |
a640f031 | 212 | return ~x; |
25c4d9cc | 213 | |
cb25c80a RH |
214 | CASE_OP_32_64(neg): |
215 | return -x; | |
216 | ||
217 | CASE_OP_32_64(andc): | |
218 | return x & ~y; | |
219 | ||
220 | CASE_OP_32_64(orc): | |
221 | return x | ~y; | |
222 | ||
223 | CASE_OP_32_64(eqv): | |
224 | return ~(x ^ y); | |
225 | ||
226 | CASE_OP_32_64(nand): | |
227 | return ~(x & y); | |
228 | ||
229 | CASE_OP_32_64(nor): | |
230 | return ~(x | y); | |
231 | ||
25c4d9cc | 232 | CASE_OP_32_64(ext8s): |
a640f031 | 233 | return (int8_t)x; |
25c4d9cc RH |
234 | |
235 | CASE_OP_32_64(ext16s): | |
a640f031 | 236 | return (int16_t)x; |
25c4d9cc RH |
237 | |
238 | CASE_OP_32_64(ext8u): | |
a640f031 | 239 | return (uint8_t)x; |
25c4d9cc RH |
240 | |
241 | CASE_OP_32_64(ext16u): | |
a640f031 KB |
242 | return (uint16_t)x; |
243 | ||
a640f031 KB |
244 | case INDEX_op_ext32s_i64: |
245 | return (int32_t)x; | |
246 | ||
247 | case INDEX_op_ext32u_i64: | |
248 | return (uint32_t)x; | |
a640f031 | 249 | |
53108fb5 KB |
250 | default: |
251 | fprintf(stderr, | |
252 | "Unrecognized operation %d in do_constant_folding.\n", op); | |
253 | tcg_abort(); | |
254 | } | |
255 | } | |
256 | ||
fe0de7aa | 257 | static TCGArg do_constant_folding(TCGOpcode op, TCGArg x, TCGArg y) |
53108fb5 KB |
258 | { |
259 | TCGArg res = do_constant_folding_2(op, x, y); | |
53108fb5 KB |
260 | if (op_bits(op) == 32) { |
261 | res &= 0xffffffff; | |
262 | } | |
53108fb5 KB |
263 | return res; |
264 | } | |
265 | ||
2108a786 AJ |
266 | static TCGArg do_constant_folding_cond(TCGOpcode op, TCGArg x, |
267 | TCGArg y, TCGCond c) | |
268 | { | |
269 | switch (op_bits(op)) { | |
270 | case 32: | |
271 | switch (c) { | |
272 | case TCG_COND_EQ: | |
273 | return (uint32_t)x == (uint32_t)y; | |
274 | case TCG_COND_NE: | |
275 | return (uint32_t)x != (uint32_t)y; | |
276 | case TCG_COND_LT: | |
277 | return (int32_t)x < (int32_t)y; | |
278 | case TCG_COND_GE: | |
279 | return (int32_t)x >= (int32_t)y; | |
280 | case TCG_COND_LE: | |
281 | return (int32_t)x <= (int32_t)y; | |
282 | case TCG_COND_GT: | |
283 | return (int32_t)x > (int32_t)y; | |
284 | case TCG_COND_LTU: | |
285 | return (uint32_t)x < (uint32_t)y; | |
286 | case TCG_COND_GEU: | |
287 | return (uint32_t)x >= (uint32_t)y; | |
288 | case TCG_COND_LEU: | |
289 | return (uint32_t)x <= (uint32_t)y; | |
290 | case TCG_COND_GTU: | |
291 | return (uint32_t)x > (uint32_t)y; | |
292 | } | |
293 | break; | |
294 | case 64: | |
295 | switch (c) { | |
296 | case TCG_COND_EQ: | |
297 | return (uint64_t)x == (uint64_t)y; | |
298 | case TCG_COND_NE: | |
299 | return (uint64_t)x != (uint64_t)y; | |
300 | case TCG_COND_LT: | |
301 | return (int64_t)x < (int64_t)y; | |
302 | case TCG_COND_GE: | |
303 | return (int64_t)x >= (int64_t)y; | |
304 | case TCG_COND_LE: | |
305 | return (int64_t)x <= (int64_t)y; | |
306 | case TCG_COND_GT: | |
307 | return (int64_t)x > (int64_t)y; | |
308 | case TCG_COND_LTU: | |
309 | return (uint64_t)x < (uint64_t)y; | |
310 | case TCG_COND_GEU: | |
311 | return (uint64_t)x >= (uint64_t)y; | |
312 | case TCG_COND_LEU: | |
313 | return (uint64_t)x <= (uint64_t)y; | |
314 | case TCG_COND_GTU: | |
315 | return (uint64_t)x > (uint64_t)y; | |
316 | } | |
317 | break; | |
318 | } | |
319 | ||
320 | fprintf(stderr, | |
321 | "Unrecognized bitness %d or condition %d in " | |
322 | "do_constant_folding_cond.\n", op_bits(op), c); | |
323 | tcg_abort(); | |
324 | } | |
325 | ||
326 | ||
22613af4 | 327 | /* Propagate constants and copies, fold constant expressions. */ |
8f2e8c07 KB |
328 | static TCGArg *tcg_constant_folding(TCGContext *s, uint16_t *tcg_opc_ptr, |
329 | TCGArg *args, TCGOpDef *tcg_op_defs) | |
330 | { | |
fe0de7aa BS |
331 | int i, nb_ops, op_index, nb_temps, nb_globals, nb_call_args; |
332 | TCGOpcode op; | |
8f2e8c07 KB |
333 | const TCGOpDef *def; |
334 | TCGArg *gen_args; | |
53108fb5 | 335 | TCGArg tmp; |
1d465a94 RH |
336 | TCGCond cond; |
337 | ||
22613af4 KB |
338 | /* Array VALS has an element for each temp. |
339 | If this temp holds a constant then its value is kept in VALS' element. | |
340 | If this temp is a copy of other ones then this equivalence class' | |
341 | representative is kept in VALS' element. | |
342 | If this temp is neither copy nor constant then corresponding VALS' | |
343 | element is unused. */ | |
8f2e8c07 KB |
344 | |
345 | nb_temps = s->nb_temps; | |
346 | nb_globals = s->nb_globals; | |
22613af4 | 347 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); |
8f2e8c07 KB |
348 | |
349 | nb_ops = tcg_opc_ptr - gen_opc_buf; | |
350 | gen_args = args; | |
351 | for (op_index = 0; op_index < nb_ops; op_index++) { | |
352 | op = gen_opc_buf[op_index]; | |
353 | def = &tcg_op_defs[op]; | |
22613af4 KB |
354 | /* Do copy propagation */ |
355 | if (!(def->flags & (TCG_OPF_CALL_CLOBBER | TCG_OPF_SIDE_EFFECTS))) { | |
356 | assert(op != INDEX_op_call); | |
357 | for (i = def->nb_oargs; i < def->nb_oargs + def->nb_iargs; i++) { | |
358 | if (temps[args[i]].state == TCG_TEMP_COPY) { | |
359 | args[i] = temps[args[i]].val; | |
360 | } | |
361 | } | |
362 | } | |
363 | ||
53108fb5 KB |
364 | /* For commutative operations make constant second argument */ |
365 | switch (op) { | |
366 | CASE_OP_32_64(add): | |
367 | CASE_OP_32_64(mul): | |
9a81090b KB |
368 | CASE_OP_32_64(and): |
369 | CASE_OP_32_64(or): | |
370 | CASE_OP_32_64(xor): | |
cb25c80a RH |
371 | CASE_OP_32_64(eqv): |
372 | CASE_OP_32_64(nand): | |
373 | CASE_OP_32_64(nor): | |
53108fb5 KB |
374 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
375 | tmp = args[1]; | |
376 | args[1] = args[2]; | |
377 | args[2] = tmp; | |
378 | } | |
379 | break; | |
decddd90 AJ |
380 | CASE_OP_32_64(brcond): |
381 | if (temps[args[0]].state == TCG_TEMP_CONST | |
382 | && temps[args[1]].state != TCG_TEMP_CONST) { | |
383 | tmp = args[0]; | |
384 | args[0] = args[1]; | |
385 | args[1] = tmp; | |
386 | args[2] = tcg_swap_cond(args[2]); | |
387 | } | |
388 | break; | |
389 | CASE_OP_32_64(setcond): | |
390 | if (temps[args[1]].state == TCG_TEMP_CONST | |
391 | && temps[args[2]].state != TCG_TEMP_CONST) { | |
392 | tmp = args[1]; | |
393 | args[1] = args[2]; | |
394 | args[2] = tmp; | |
395 | args[3] = tcg_swap_cond(args[3]); | |
396 | } | |
397 | break; | |
b5b6bb16 | 398 | CASE_OP_32_64(movcond): |
1d465a94 | 399 | cond = args[5]; |
b5b6bb16 RH |
400 | if (temps[args[1]].state == TCG_TEMP_CONST |
401 | && temps[args[2]].state != TCG_TEMP_CONST) { | |
402 | tmp = args[1]; | |
403 | args[1] = args[2]; | |
404 | args[2] = tmp; | |
1d465a94 RH |
405 | cond = tcg_swap_cond(cond); |
406 | } | |
407 | /* For movcond, we canonicalize the "false" input reg to match | |
408 | the destination reg so that the tcg backend can implement | |
409 | a "move if true" operation. */ | |
410 | if (args[0] == args[3]) { | |
411 | tmp = args[3]; | |
412 | args[3] = args[4]; | |
413 | args[4] = tmp; | |
414 | cond = tcg_invert_cond(cond); | |
b5b6bb16 | 415 | } |
1d465a94 | 416 | args[5] = cond; |
53108fb5 KB |
417 | default: |
418 | break; | |
419 | } | |
420 | ||
52173bab AJ |
421 | /* Simplify expressions for "shift/rot r, 0, a => movi r, 0" */ |
422 | switch (op) { | |
423 | CASE_OP_32_64(shl): | |
424 | CASE_OP_32_64(shr): | |
425 | CASE_OP_32_64(sar): | |
426 | CASE_OP_32_64(rotl): | |
427 | CASE_OP_32_64(rotr): | |
428 | if (temps[args[1]].state == TCG_TEMP_CONST | |
429 | && temps[args[1]].val == 0) { | |
430 | gen_opc_buf[op_index] = op_to_movi(op); | |
431 | tcg_opt_gen_movi(gen_args, args[0], 0, nb_temps, nb_globals); | |
432 | args += 3; | |
433 | gen_args += 2; | |
434 | continue; | |
435 | } | |
436 | break; | |
437 | default: | |
438 | break; | |
439 | } | |
440 | ||
d9693072 | 441 | /* Simplify expression for "op r, a, 0 => mov r, a" cases */ |
53108fb5 KB |
442 | switch (op) { |
443 | CASE_OP_32_64(add): | |
444 | CASE_OP_32_64(sub): | |
55c0975c KB |
445 | CASE_OP_32_64(shl): |
446 | CASE_OP_32_64(shr): | |
447 | CASE_OP_32_64(sar): | |
25c4d9cc RH |
448 | CASE_OP_32_64(rotl): |
449 | CASE_OP_32_64(rotr): | |
6c9785fb AJ |
450 | CASE_OP_32_64(or): |
451 | CASE_OP_32_64(xor): | |
53108fb5 KB |
452 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
453 | /* Proceed with possible constant folding. */ | |
454 | break; | |
455 | } | |
456 | if (temps[args[2]].state == TCG_TEMP_CONST | |
457 | && temps[args[2]].val == 0) { | |
458 | if ((temps[args[0]].state == TCG_TEMP_COPY | |
459 | && temps[args[0]].val == args[1]) | |
460 | || args[0] == args[1]) { | |
53108fb5 KB |
461 | gen_opc_buf[op_index] = INDEX_op_nop; |
462 | } else { | |
463 | gen_opc_buf[op_index] = op_to_mov(op); | |
4f17bc1e | 464 | tcg_opt_gen_mov(gen_args, args[0], args[1], |
53108fb5 KB |
465 | nb_temps, nb_globals); |
466 | gen_args += 2; | |
53108fb5 | 467 | } |
578c7582 | 468 | args += 3; |
53108fb5 KB |
469 | continue; |
470 | } | |
471 | break; | |
d9693072 AJ |
472 | default: |
473 | break; | |
474 | } | |
475 | ||
476 | /* Simplify expression for "op r, a, 0 => movi r, 0" cases */ | |
477 | switch (op) { | |
f66f1dab | 478 | CASE_OP_32_64(and): |
53108fb5 KB |
479 | CASE_OP_32_64(mul): |
480 | if ((temps[args[2]].state == TCG_TEMP_CONST | |
481 | && temps[args[2]].val == 0)) { | |
482 | gen_opc_buf[op_index] = op_to_movi(op); | |
483 | tcg_opt_gen_movi(gen_args, args[0], 0, nb_temps, nb_globals); | |
484 | args += 3; | |
485 | gen_args += 2; | |
486 | continue; | |
487 | } | |
488 | break; | |
d9693072 AJ |
489 | default: |
490 | break; | |
491 | } | |
492 | ||
493 | /* Simplify expression for "op r, a, a => mov r, a" cases */ | |
494 | switch (op) { | |
9a81090b KB |
495 | CASE_OP_32_64(or): |
496 | CASE_OP_32_64(and): | |
497 | if (args[1] == args[2]) { | |
498 | if (args[1] == args[0]) { | |
9a81090b KB |
499 | gen_opc_buf[op_index] = INDEX_op_nop; |
500 | } else { | |
501 | gen_opc_buf[op_index] = op_to_mov(op); | |
4f17bc1e | 502 | tcg_opt_gen_mov(gen_args, args[0], args[1], nb_temps, |
9a81090b KB |
503 | nb_globals); |
504 | gen_args += 2; | |
9a81090b | 505 | } |
578c7582 | 506 | args += 3; |
9a81090b KB |
507 | continue; |
508 | } | |
509 | break; | |
fe0de7aa BS |
510 | default: |
511 | break; | |
53108fb5 KB |
512 | } |
513 | ||
22613af4 KB |
514 | /* Propagate constants through copy operations and do constant |
515 | folding. Constants will be substituted to arguments by register | |
516 | allocator where needed and possible. Also detect copies. */ | |
8f2e8c07 | 517 | switch (op) { |
22613af4 KB |
518 | CASE_OP_32_64(mov): |
519 | if ((temps[args[1]].state == TCG_TEMP_COPY | |
520 | && temps[args[1]].val == args[0]) | |
521 | || args[0] == args[1]) { | |
522 | args += 2; | |
523 | gen_opc_buf[op_index] = INDEX_op_nop; | |
524 | break; | |
525 | } | |
526 | if (temps[args[1]].state != TCG_TEMP_CONST) { | |
4f17bc1e | 527 | tcg_opt_gen_mov(gen_args, args[0], args[1], |
22613af4 KB |
528 | nb_temps, nb_globals); |
529 | gen_args += 2; | |
530 | args += 2; | |
531 | break; | |
532 | } | |
533 | /* Source argument is constant. Rewrite the operation and | |
534 | let movi case handle it. */ | |
535 | op = op_to_movi(op); | |
536 | gen_opc_buf[op_index] = op; | |
537 | args[1] = temps[args[1]].val; | |
538 | /* fallthrough */ | |
539 | CASE_OP_32_64(movi): | |
540 | tcg_opt_gen_movi(gen_args, args[0], args[1], nb_temps, nb_globals); | |
541 | gen_args += 2; | |
542 | args += 2; | |
543 | break; | |
a640f031 | 544 | CASE_OP_32_64(not): |
cb25c80a | 545 | CASE_OP_32_64(neg): |
25c4d9cc RH |
546 | CASE_OP_32_64(ext8s): |
547 | CASE_OP_32_64(ext8u): | |
548 | CASE_OP_32_64(ext16s): | |
549 | CASE_OP_32_64(ext16u): | |
a640f031 KB |
550 | case INDEX_op_ext32s_i64: |
551 | case INDEX_op_ext32u_i64: | |
a640f031 KB |
552 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
553 | gen_opc_buf[op_index] = op_to_movi(op); | |
554 | tmp = do_constant_folding(op, temps[args[1]].val, 0); | |
555 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
a640f031 KB |
556 | } else { |
557 | reset_temp(args[0], nb_temps, nb_globals); | |
558 | gen_args[0] = args[0]; | |
559 | gen_args[1] = args[1]; | |
a640f031 | 560 | } |
578c7582 AJ |
561 | gen_args += 2; |
562 | args += 2; | |
563 | break; | |
53108fb5 KB |
564 | CASE_OP_32_64(add): |
565 | CASE_OP_32_64(sub): | |
566 | CASE_OP_32_64(mul): | |
9a81090b KB |
567 | CASE_OP_32_64(or): |
568 | CASE_OP_32_64(and): | |
569 | CASE_OP_32_64(xor): | |
55c0975c KB |
570 | CASE_OP_32_64(shl): |
571 | CASE_OP_32_64(shr): | |
572 | CASE_OP_32_64(sar): | |
25c4d9cc RH |
573 | CASE_OP_32_64(rotl): |
574 | CASE_OP_32_64(rotr): | |
cb25c80a RH |
575 | CASE_OP_32_64(andc): |
576 | CASE_OP_32_64(orc): | |
577 | CASE_OP_32_64(eqv): | |
578 | CASE_OP_32_64(nand): | |
579 | CASE_OP_32_64(nor): | |
53108fb5 KB |
580 | if (temps[args[1]].state == TCG_TEMP_CONST |
581 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
582 | gen_opc_buf[op_index] = op_to_movi(op); | |
583 | tmp = do_constant_folding(op, temps[args[1]].val, | |
584 | temps[args[2]].val); | |
585 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
586 | gen_args += 2; | |
53108fb5 KB |
587 | } else { |
588 | reset_temp(args[0], nb_temps, nb_globals); | |
589 | gen_args[0] = args[0]; | |
590 | gen_args[1] = args[1]; | |
591 | gen_args[2] = args[2]; | |
592 | gen_args += 3; | |
53108fb5 | 593 | } |
578c7582 AJ |
594 | args += 3; |
595 | break; | |
2108a786 AJ |
596 | CASE_OP_32_64(setcond): |
597 | if (temps[args[1]].state == TCG_TEMP_CONST | |
598 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
599 | gen_opc_buf[op_index] = op_to_movi(op); | |
600 | tmp = do_constant_folding_cond(op, temps[args[1]].val, | |
601 | temps[args[2]].val, args[3]); | |
602 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
603 | gen_args += 2; | |
2108a786 AJ |
604 | } else { |
605 | reset_temp(args[0], nb_temps, nb_globals); | |
606 | gen_args[0] = args[0]; | |
607 | gen_args[1] = args[1]; | |
608 | gen_args[2] = args[2]; | |
609 | gen_args[3] = args[3]; | |
610 | gen_args += 4; | |
2108a786 | 611 | } |
578c7582 AJ |
612 | args += 4; |
613 | break; | |
97c012c7 AJ |
614 | CASE_OP_32_64(brcond): |
615 | if (temps[args[0]].state == TCG_TEMP_CONST | |
616 | && temps[args[1]].state == TCG_TEMP_CONST) { | |
617 | if (do_constant_folding_cond(op, temps[args[0]].val, | |
618 | temps[args[1]].val, args[2])) { | |
619 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
620 | gen_opc_buf[op_index] = INDEX_op_br; | |
621 | gen_args[0] = args[3]; | |
622 | gen_args += 1; | |
97c012c7 AJ |
623 | } else { |
624 | gen_opc_buf[op_index] = INDEX_op_nop; | |
97c012c7 | 625 | } |
97c012c7 AJ |
626 | } else { |
627 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
628 | reset_temp(args[0], nb_temps, nb_globals); | |
629 | gen_args[0] = args[0]; | |
630 | gen_args[1] = args[1]; | |
631 | gen_args[2] = args[2]; | |
632 | gen_args[3] = args[3]; | |
633 | gen_args += 4; | |
97c012c7 | 634 | } |
578c7582 AJ |
635 | args += 4; |
636 | break; | |
b5b6bb16 RH |
637 | CASE_OP_32_64(movcond): |
638 | if (temps[args[1]].state == TCG_TEMP_CONST | |
639 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
640 | tmp = do_constant_folding_cond(op, temps[args[1]].val, | |
641 | temps[args[2]].val, args[5]); | |
642 | if (args[0] == args[4-tmp] | |
643 | || (temps[args[4-tmp]].state == TCG_TEMP_COPY | |
644 | && temps[args[4-tmp]].val == args[0])) { | |
645 | gen_opc_buf[op_index] = INDEX_op_nop; | |
646 | } else if (temps[args[4-tmp]].state == TCG_TEMP_CONST) { | |
647 | gen_opc_buf[op_index] = op_to_movi(op); | |
648 | tcg_opt_gen_movi(gen_args, args[0], temps[args[4-tmp]].val, | |
649 | nb_temps, nb_globals); | |
650 | gen_args += 2; | |
651 | } else { | |
652 | gen_opc_buf[op_index] = op_to_mov(op); | |
653 | tcg_opt_gen_mov(gen_args, args[0], args[4-tmp], | |
654 | nb_temps, nb_globals); | |
655 | gen_args += 2; | |
656 | } | |
657 | } else { | |
658 | reset_temp(args[0], nb_temps, nb_globals); | |
659 | gen_args[0] = args[0]; | |
660 | gen_args[1] = args[1]; | |
661 | gen_args[2] = args[2]; | |
662 | gen_args[3] = args[3]; | |
663 | gen_args[4] = args[4]; | |
664 | gen_args[5] = args[5]; | |
665 | gen_args += 6; | |
666 | } | |
667 | args += 6; | |
668 | break; | |
8f2e8c07 | 669 | case INDEX_op_call: |
22613af4 KB |
670 | nb_call_args = (args[0] >> 16) + (args[0] & 0xffff); |
671 | if (!(args[nb_call_args + 1] & (TCG_CALL_CONST | TCG_CALL_PURE))) { | |
672 | for (i = 0; i < nb_globals; i++) { | |
673 | reset_temp(i, nb_temps, nb_globals); | |
674 | } | |
675 | } | |
676 | for (i = 0; i < (args[0] >> 16); i++) { | |
677 | reset_temp(args[i + 1], nb_temps, nb_globals); | |
678 | } | |
679 | i = nb_call_args + 3; | |
8f2e8c07 KB |
680 | while (i) { |
681 | *gen_args = *args; | |
682 | args++; | |
683 | gen_args++; | |
684 | i--; | |
685 | } | |
686 | break; | |
8f2e8c07 | 687 | default: |
22613af4 | 688 | /* Default case: we do know nothing about operation so no |
5b681fff AJ |
689 | propagation is done. We trash everything if the operation |
690 | is the end of a basic block, otherwise we only trash the | |
691 | output args. */ | |
692 | if (def->flags & TCG_OPF_BB_END) { | |
693 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
694 | } else { | |
695 | for (i = 0; i < def->nb_oargs; i++) { | |
696 | reset_temp(args[i], nb_temps, nb_globals); | |
697 | } | |
22613af4 | 698 | } |
8f2e8c07 KB |
699 | for (i = 0; i < def->nb_args; i++) { |
700 | gen_args[i] = args[i]; | |
701 | } | |
702 | args += def->nb_args; | |
703 | gen_args += def->nb_args; | |
704 | break; | |
705 | } | |
706 | } | |
707 | ||
708 | return gen_args; | |
709 | } | |
710 | ||
711 | TCGArg *tcg_optimize(TCGContext *s, uint16_t *tcg_opc_ptr, | |
712 | TCGArg *args, TCGOpDef *tcg_op_defs) | |
713 | { | |
714 | TCGArg *res; | |
715 | res = tcg_constant_folding(s, tcg_opc_ptr, args, tcg_op_defs); | |
716 | return res; | |
717 | } |