]>
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, | |
48b56ce1 | 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 | ||
48b56ce1 | 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) { | |
48b56ce1 | 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 | } | |
48b56ce1 | 83 | temps[temp].state = TCG_TEMP_UNDEF; |
22613af4 | 84 | if (new_base != (TCGArg)-1 && temps[new_base].next_copy == new_base) { |
48b56ce1 | 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 | ||
b80bb016 AJ |
109 | static void tcg_opt_gen_mov(TCGContext *s, TCGArg *gen_args, |
110 | TCGArg dst, TCGArg src) | |
22613af4 | 111 | { |
b80bb016 | 112 | reset_temp(dst, s->nb_temps, s->nb_globals); |
22613af4 | 113 | assert(temps[src].state != TCG_TEMP_COPY); |
b80bb016 AJ |
114 | /* Only consider temps with the same type (width) as copies. */ |
115 | if (src >= s->nb_globals && s->temps[dst].type == s->temps[src].type) { | |
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 | ||
f8dd19e5 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; |
5d8f5363 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; | |
65a7cce1 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; | |
fa01a208 | 399 | CASE_OP_32_64(movcond): |
5d8f5363 | 400 | cond = args[5]; |
fa01a208 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; | |
5d8f5363 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); | |
fa01a208 | 416 | } |
5d8f5363 | 417 | args[5] = cond; |
53108fb5 KB |
418 | default: |
419 | break; | |
420 | } | |
421 | ||
01ee5282 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 | ||
56e49438 | 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): | |
38ee188b 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); | |
b80bb016 | 465 | tcg_opt_gen_mov(s, gen_args, args[0], args[1]); |
53108fb5 | 466 | gen_args += 2; |
53108fb5 | 467 | } |
fedc0da2 | 468 | args += 3; |
53108fb5 KB |
469 | continue; |
470 | } | |
471 | break; | |
56e49438 AJ |
472 | default: |
473 | break; | |
474 | } | |
475 | ||
476 | /* Simplify expression for "op r, a, 0 => movi r, 0" cases */ | |
477 | switch (op) { | |
61251c0c | 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; | |
56e49438 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); | |
b80bb016 | 502 | tcg_opt_gen_mov(s, gen_args, args[0], args[1]); |
9a81090b | 503 | gen_args += 2; |
9a81090b | 504 | } |
fedc0da2 | 505 | args += 3; |
9a81090b KB |
506 | continue; |
507 | } | |
508 | break; | |
fe0de7aa BS |
509 | default: |
510 | break; | |
53108fb5 KB |
511 | } |
512 | ||
22613af4 KB |
513 | /* Propagate constants through copy operations and do constant |
514 | folding. Constants will be substituted to arguments by register | |
515 | allocator where needed and possible. Also detect copies. */ | |
8f2e8c07 | 516 | switch (op) { |
22613af4 KB |
517 | CASE_OP_32_64(mov): |
518 | if ((temps[args[1]].state == TCG_TEMP_COPY | |
519 | && temps[args[1]].val == args[0]) | |
520 | || args[0] == args[1]) { | |
521 | args += 2; | |
522 | gen_opc_buf[op_index] = INDEX_op_nop; | |
523 | break; | |
524 | } | |
525 | if (temps[args[1]].state != TCG_TEMP_CONST) { | |
b80bb016 | 526 | tcg_opt_gen_mov(s, gen_args, args[0], args[1]); |
22613af4 KB |
527 | gen_args += 2; |
528 | args += 2; | |
529 | break; | |
530 | } | |
531 | /* Source argument is constant. Rewrite the operation and | |
532 | let movi case handle it. */ | |
533 | op = op_to_movi(op); | |
534 | gen_opc_buf[op_index] = op; | |
535 | args[1] = temps[args[1]].val; | |
536 | /* fallthrough */ | |
537 | CASE_OP_32_64(movi): | |
538 | tcg_opt_gen_movi(gen_args, args[0], args[1], nb_temps, nb_globals); | |
539 | gen_args += 2; | |
540 | args += 2; | |
541 | break; | |
a640f031 | 542 | CASE_OP_32_64(not): |
cb25c80a | 543 | CASE_OP_32_64(neg): |
25c4d9cc RH |
544 | CASE_OP_32_64(ext8s): |
545 | CASE_OP_32_64(ext8u): | |
546 | CASE_OP_32_64(ext16s): | |
547 | CASE_OP_32_64(ext16u): | |
a640f031 KB |
548 | case INDEX_op_ext32s_i64: |
549 | case INDEX_op_ext32u_i64: | |
a640f031 KB |
550 | if (temps[args[1]].state == TCG_TEMP_CONST) { |
551 | gen_opc_buf[op_index] = op_to_movi(op); | |
552 | tmp = do_constant_folding(op, temps[args[1]].val, 0); | |
553 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
a640f031 KB |
554 | } else { |
555 | reset_temp(args[0], nb_temps, nb_globals); | |
556 | gen_args[0] = args[0]; | |
557 | gen_args[1] = args[1]; | |
a640f031 | 558 | } |
fedc0da2 AJ |
559 | gen_args += 2; |
560 | args += 2; | |
561 | break; | |
53108fb5 KB |
562 | CASE_OP_32_64(add): |
563 | CASE_OP_32_64(sub): | |
564 | CASE_OP_32_64(mul): | |
9a81090b KB |
565 | CASE_OP_32_64(or): |
566 | CASE_OP_32_64(and): | |
567 | CASE_OP_32_64(xor): | |
55c0975c KB |
568 | CASE_OP_32_64(shl): |
569 | CASE_OP_32_64(shr): | |
570 | CASE_OP_32_64(sar): | |
25c4d9cc RH |
571 | CASE_OP_32_64(rotl): |
572 | CASE_OP_32_64(rotr): | |
cb25c80a RH |
573 | CASE_OP_32_64(andc): |
574 | CASE_OP_32_64(orc): | |
575 | CASE_OP_32_64(eqv): | |
576 | CASE_OP_32_64(nand): | |
577 | CASE_OP_32_64(nor): | |
53108fb5 KB |
578 | if (temps[args[1]].state == TCG_TEMP_CONST |
579 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
580 | gen_opc_buf[op_index] = op_to_movi(op); | |
581 | tmp = do_constant_folding(op, temps[args[1]].val, | |
582 | temps[args[2]].val); | |
583 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
584 | gen_args += 2; | |
53108fb5 KB |
585 | } else { |
586 | reset_temp(args[0], nb_temps, nb_globals); | |
587 | gen_args[0] = args[0]; | |
588 | gen_args[1] = args[1]; | |
589 | gen_args[2] = args[2]; | |
590 | gen_args += 3; | |
53108fb5 | 591 | } |
fedc0da2 AJ |
592 | args += 3; |
593 | break; | |
f8dd19e5 AJ |
594 | CASE_OP_32_64(setcond): |
595 | if (temps[args[1]].state == TCG_TEMP_CONST | |
596 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
597 | gen_opc_buf[op_index] = op_to_movi(op); | |
598 | tmp = do_constant_folding_cond(op, temps[args[1]].val, | |
599 | temps[args[2]].val, args[3]); | |
600 | tcg_opt_gen_movi(gen_args, args[0], tmp, nb_temps, nb_globals); | |
601 | gen_args += 2; | |
f8dd19e5 AJ |
602 | } else { |
603 | reset_temp(args[0], nb_temps, nb_globals); | |
604 | gen_args[0] = args[0]; | |
605 | gen_args[1] = args[1]; | |
606 | gen_args[2] = args[2]; | |
607 | gen_args[3] = args[3]; | |
608 | gen_args += 4; | |
f8dd19e5 | 609 | } |
fedc0da2 AJ |
610 | args += 4; |
611 | break; | |
fbeaa26c AJ |
612 | CASE_OP_32_64(brcond): |
613 | if (temps[args[0]].state == TCG_TEMP_CONST | |
614 | && temps[args[1]].state == TCG_TEMP_CONST) { | |
615 | if (do_constant_folding_cond(op, temps[args[0]].val, | |
616 | temps[args[1]].val, args[2])) { | |
617 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
618 | gen_opc_buf[op_index] = INDEX_op_br; | |
619 | gen_args[0] = args[3]; | |
620 | gen_args += 1; | |
fbeaa26c AJ |
621 | } else { |
622 | gen_opc_buf[op_index] = INDEX_op_nop; | |
fbeaa26c | 623 | } |
fbeaa26c AJ |
624 | } else { |
625 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
626 | reset_temp(args[0], nb_temps, nb_globals); | |
627 | gen_args[0] = args[0]; | |
628 | gen_args[1] = args[1]; | |
629 | gen_args[2] = args[2]; | |
630 | gen_args[3] = args[3]; | |
631 | gen_args += 4; | |
fbeaa26c | 632 | } |
fedc0da2 AJ |
633 | args += 4; |
634 | break; | |
fa01a208 RH |
635 | CASE_OP_32_64(movcond): |
636 | if (temps[args[1]].state == TCG_TEMP_CONST | |
637 | && temps[args[2]].state == TCG_TEMP_CONST) { | |
638 | tmp = do_constant_folding_cond(op, temps[args[1]].val, | |
639 | temps[args[2]].val, args[5]); | |
640 | if (args[0] == args[4-tmp] | |
641 | || (temps[args[4-tmp]].state == TCG_TEMP_COPY | |
642 | && temps[args[4-tmp]].val == args[0])) { | |
643 | gen_opc_buf[op_index] = INDEX_op_nop; | |
644 | } else if (temps[args[4-tmp]].state == TCG_TEMP_CONST) { | |
645 | gen_opc_buf[op_index] = op_to_movi(op); | |
646 | tcg_opt_gen_movi(gen_args, args[0], temps[args[4-tmp]].val, | |
647 | nb_temps, nb_globals); | |
648 | gen_args += 2; | |
649 | } else { | |
650 | gen_opc_buf[op_index] = op_to_mov(op); | |
651 | tcg_opt_gen_mov(gen_args, args[0], args[4-tmp], | |
652 | nb_temps, nb_globals); | |
653 | gen_args += 2; | |
654 | } | |
655 | } else { | |
656 | reset_temp(args[0], nb_temps, nb_globals); | |
657 | gen_args[0] = args[0]; | |
658 | gen_args[1] = args[1]; | |
659 | gen_args[2] = args[2]; | |
660 | gen_args[3] = args[3]; | |
661 | gen_args[4] = args[4]; | |
662 | gen_args[5] = args[5]; | |
663 | gen_args += 6; | |
664 | } | |
665 | args += 6; | |
666 | break; | |
8f2e8c07 | 667 | case INDEX_op_call: |
22613af4 KB |
668 | nb_call_args = (args[0] >> 16) + (args[0] & 0xffff); |
669 | if (!(args[nb_call_args + 1] & (TCG_CALL_CONST | TCG_CALL_PURE))) { | |
670 | for (i = 0; i < nb_globals; i++) { | |
671 | reset_temp(i, nb_temps, nb_globals); | |
672 | } | |
673 | } | |
674 | for (i = 0; i < (args[0] >> 16); i++) { | |
675 | reset_temp(args[i + 1], nb_temps, nb_globals); | |
676 | } | |
677 | i = nb_call_args + 3; | |
8f2e8c07 KB |
678 | while (i) { |
679 | *gen_args = *args; | |
680 | args++; | |
681 | gen_args++; | |
682 | i--; | |
683 | } | |
684 | break; | |
8f2e8c07 | 685 | default: |
22613af4 | 686 | /* Default case: we do know nothing about operation so no |
a2550660 AJ |
687 | propagation is done. We trash everything if the operation |
688 | is the end of a basic block, otherwise we only trash the | |
689 | output args. */ | |
690 | if (def->flags & TCG_OPF_BB_END) { | |
691 | memset(temps, 0, nb_temps * sizeof(struct tcg_temp_info)); | |
692 | } else { | |
693 | for (i = 0; i < def->nb_oargs; i++) { | |
694 | reset_temp(args[i], nb_temps, nb_globals); | |
695 | } | |
22613af4 | 696 | } |
8f2e8c07 KB |
697 | for (i = 0; i < def->nb_args; i++) { |
698 | gen_args[i] = args[i]; | |
699 | } | |
700 | args += def->nb_args; | |
701 | gen_args += def->nb_args; | |
702 | break; | |
703 | } | |
704 | } | |
705 | ||
706 | return gen_args; | |
707 | } | |
708 | ||
709 | TCGArg *tcg_optimize(TCGContext *s, uint16_t *tcg_opc_ptr, | |
710 | TCGArg *args, TCGOpDef *tcg_op_defs) | |
711 | { | |
712 | TCGArg *res; | |
713 | res = tcg_constant_folding(s, tcg_opc_ptr, args, tcg_op_defs); | |
714 | return res; | |
715 | } |