2 * Copyright (C) 2012, 2013
5 * Permission is hereby granted, free of charge, to any person obtaining a copy of
6 * this software and associated documentation files (the "Software"), to deal in
7 * the Software without restriction, including without limitation the rights to
8 * use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is furnished to do
10 * so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be included in all
13 * copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
29 #define FOLD_STRING_UNTRANSLATE_HTSIZE 1024
30 #define FOLD_STRING_DOTRANSLATE_HTSIZE 1024
33 * There is two stages to constant folding in GMQCC: there is the parse
34 * stage constant folding, where, witht he help of the AST, operator
35 * usages can be constant folded. Then there is the constant folding
36 * in the IR for things like eliding if statements, can occur.
38 * This file is thus, split into two parts.
41 #define isfloat(X) (((ast_expression*)(X))->vtype == TYPE_FLOAT)
42 #define isvector(X) (((ast_expression*)(X))->vtype == TYPE_VECTOR)
43 #define isstring(X) (((ast_expression*)(X))->vtype == TYPE_STRING)
44 #define isfloats(X,Y) (isfloat (X) && isfloat (Y))
47 * Implementation of basic vector math for vec3_t, for trivial constant
50 * TODO: gcc/clang hinting for autovectorization
52 static GMQCC_INLINE vec3_t vec3_add(vec3_t a, vec3_t b) {
60 static GMQCC_INLINE vec3_t vec3_sub(vec3_t a, vec3_t b) {
68 static GMQCC_INLINE vec3_t vec3_neg(vec3_t a) {
76 static GMQCC_INLINE vec3_t vec3_xor(vec3_t a, vec3_t b) {
78 out.x = (qcfloat_t)((qcint_t)a.x ^ (qcint_t)b.x);
79 out.y = (qcfloat_t)((qcint_t)a.y ^ (qcint_t)b.y);
80 out.z = (qcfloat_t)((qcint_t)a.z ^ (qcint_t)b.z);
84 static GMQCC_INLINE vec3_t vec3_xorvf(vec3_t a, qcfloat_t b) {
86 out.x = (qcfloat_t)((qcint_t)a.x ^ (qcint_t)b);
87 out.y = (qcfloat_t)((qcint_t)a.y ^ (qcint_t)b);
88 out.z = (qcfloat_t)((qcint_t)a.z ^ (qcint_t)b);
92 static GMQCC_INLINE qcfloat_t vec3_mulvv(vec3_t a, vec3_t b) {
93 return (a.x * b.x + a.y * b.y + a.z * b.z);
96 static GMQCC_INLINE vec3_t vec3_mulvf(vec3_t a, qcfloat_t b) {
104 static GMQCC_INLINE bool vec3_cmp(vec3_t a, vec3_t b) {
110 static GMQCC_INLINE vec3_t vec3_create(float x, float y, float z) {
118 static GMQCC_INLINE qcfloat_t vec3_notf(vec3_t a) {
119 return (!a.x && !a.y && !a.z);
122 static GMQCC_INLINE bool vec3_pbool(vec3_t a) {
123 return (a.x && a.y && a.z);
126 static GMQCC_INLINE bool fold_can_1(const ast_value *val) {
127 return (ast_istype((ast_expression*)val, ast_value) && val->hasvalue && val->cvq == CV_CONST && ((ast_expression*)val)->vtype != TYPE_FUNCTION);
130 static GMQCC_INLINE bool fold_can_2(const ast_value *v1, const ast_value *v2) {
131 return fold_can_1(v1) && fold_can_1(v2);
134 static lex_ctx_t fold_ctx(fold_t *fold) {
136 if (fold->parser->lex)
137 return parser_ctx(fold->parser);
139 memset(&ctx, 0, sizeof(ctx));
143 static GMQCC_INLINE bool fold_immediate_true(fold_t *fold, ast_value *v) {
144 switch (v->expression.vtype) {
146 return !!v->constval.vfloat;
148 return !!v->constval.vint;
150 if (OPTS_FLAG(CORRECT_LOGIC))
151 return vec3_pbool(v->constval.vvec);
152 return !!v->constval.vvec.x;
154 if (!v->constval.vstring)
156 if (OPTS_FLAG(TRUE_EMPTY_STRINGS))
158 return !!v->constval.vstring[0];
160 compile_error(fold_ctx(fold), "internal error: fold_immediate_true on invalid type");
163 return !!v->constval.vfunc;
166 #define fold_immvalue_float(E) ((E)->constval.vfloat)
167 #define fold_immvalue_vector(E) ((E)->constval.vvec)
168 #define fold_immvalue_string(E) ((E)->constval.vstring)
170 fold_t *fold_init(parser_t *parser) {
171 fold_t *fold = (fold_t*)mem_a(sizeof(fold_t));
172 fold->parser = parser;
173 fold->imm_float = NULL;
174 fold->imm_vector = NULL;
175 fold->imm_string = NULL;
176 fold->imm_string_untranslate = util_htnew(FOLD_STRING_UNTRANSLATE_HTSIZE);
177 fold->imm_string_dotranslate = util_htnew(FOLD_STRING_DOTRANSLATE_HTSIZE);
180 * prime the tables with common constant values at constant
183 (void)fold_constgen_float (fold, 0.0f);
184 (void)fold_constgen_float (fold, 1.0f);
185 (void)fold_constgen_float (fold, -1.0f);
187 (void)fold_constgen_vector(fold, vec3_create(0.0f, 0.0f, 0.0f));
192 bool fold_generate(fold_t *fold, ir_builder *ir) {
193 /* generate globals for immediate folded values */
197 for (i = 0; i < vec_size(fold->imm_float); ++i)
198 if (!ast_global_codegen ((cur = fold->imm_float[i]), ir, false)) goto err;
199 for (i = 0; i < vec_size(fold->imm_vector); ++i)
200 if (!ast_global_codegen((cur = fold->imm_vector[i]), ir, false)) goto err;
201 for (i = 0; i < vec_size(fold->imm_string); ++i)
202 if (!ast_global_codegen((cur = fold->imm_string[i]), ir, false)) goto err;
207 con_out("failed to generate global %s\n", cur->name);
208 ir_builder_delete(ir);
212 void fold_cleanup(fold_t *fold) {
215 for (i = 0; i < vec_size(fold->imm_float); ++i) ast_delete(fold->imm_float[i]);
216 for (i = 0; i < vec_size(fold->imm_vector); ++i) ast_delete(fold->imm_vector[i]);
217 for (i = 0; i < vec_size(fold->imm_string); ++i) ast_delete(fold->imm_string[i]);
219 vec_free(fold->imm_float);
220 vec_free(fold->imm_vector);
221 vec_free(fold->imm_string);
223 util_htdel(fold->imm_string_untranslate);
224 util_htdel(fold->imm_string_dotranslate);
229 ast_expression *fold_constgen_float(fold_t *fold, qcfloat_t value) {
230 ast_value *out = NULL;
233 for (i = 0; i < vec_size(fold->imm_float); i++) {
234 if (fold->imm_float[i]->constval.vfloat == value)
235 return (ast_expression*)fold->imm_float[i];
238 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_FLOAT);
240 out->hasvalue = true;
241 out->constval.vfloat = value;
243 vec_push(fold->imm_float, out);
245 return (ast_expression*)out;
248 ast_expression *fold_constgen_vector(fold_t *fold, vec3_t value) {
252 for (i = 0; i < vec_size(fold->imm_vector); i++) {
253 if (vec3_cmp(fold->imm_vector[i]->constval.vvec, value))
254 return (ast_expression*)fold->imm_vector[i];
257 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_VECTOR);
259 out->hasvalue = true;
260 out->constval.vvec = value;
262 vec_push(fold->imm_vector, out);
264 return (ast_expression*)out;
267 ast_expression *fold_constgen_string(fold_t *fold, const char *str, bool translate) {
268 hash_table_t *table = (translate) ? fold->imm_string_untranslate : fold->imm_string_dotranslate;
269 ast_value *out = NULL;
270 size_t hash = util_hthash(table, str);
272 if ((out = (ast_value*)util_htgeth(table, str, hash)))
273 return (ast_expression*)out;
277 util_snprintf(name, sizeof(name), "dotranslate_%lu", (unsigned long)(fold->parser->translated++));
278 out = ast_value_new(parser_ctx(fold->parser), name, TYPE_STRING);
279 out->expression.flags |= AST_FLAG_INCLUDE_DEF; /* def needs to be included for translatables */
281 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_STRING);
284 out->hasvalue = true;
286 out->constval.vstring = parser_strdup(str);
288 vec_push(fold->imm_string, out);
289 util_htseth(table, str, hash, out);
291 return (ast_expression*)out;
295 static GMQCC_INLINE ast_expression *fold_op_mul_vec(fold_t *fold, vec3_t vec, ast_value *sel, const char *set) {
297 * vector-component constant folding works by matching the component sets
298 * to eliminate expensive operations on whole-vectors (3 components at runtime).
299 * to achive this effect in a clean manner this function generalizes the
300 * values through the use of a set paramater, which is used as an indexing method
301 * for creating the elided ast binary expression.
303 * Consider 'n 0 0' where y, and z need to be tested for 0, and x is
304 * used as the value in a binary operation generating an INSTR_MUL instruction
305 * to acomplish the indexing of the correct component value we use set[0], set[1], set[2]
306 * as x, y, z, where the values of those operations return 'x', 'y', 'z'. Because
307 * of how ASCII works we can easily deliniate:
308 * vec.z is the same as set[2]-'x' for when set[2] is 'z', 'z'-'x' results in a
309 * literal value of 2, using this 2, we know that taking the address of vec->x (float)
310 * and indxing it with this literal will yeild the immediate address of that component
312 * Of course more work needs to be done to generate the correct index for the ast_member_new
313 * call, which is no problem: set[0]-'x' suffices that job.
315 qcfloat_t x = (&vec.x)[set[0]-'x'];
316 qcfloat_t y = (&vec.x)[set[1]-'x'];
317 qcfloat_t z = (&vec.x)[set[2]-'x'];
321 ++opts_optimizationcount[OPTIM_VECTOR_COMPONENTS];
322 out = (ast_expression*)ast_member_new(fold_ctx(fold), (ast_expression*)sel, set[0]-'x', NULL);
323 out->node.keep = false;
324 ((ast_member*)out)->rvalue = true;
326 return (ast_expression*)ast_binary_new(fold_ctx(fold), INSTR_MUL_F, fold_constgen_float(fold, x), out);
332 static GMQCC_INLINE ast_expression *fold_op_neg(fold_t *fold, ast_value *a) {
335 return fold_constgen_float(fold, -fold_immvalue_float(a));
336 } else if (isvector(a)) {
338 return fold_constgen_vector(fold, vec3_neg(fold_immvalue_vector(a)));
343 static GMQCC_INLINE ast_expression *fold_op_not(fold_t *fold, ast_value *a) {
346 return fold_constgen_float(fold, !fold_immvalue_float(a));
347 } else if (isvector(a)) {
349 return fold_constgen_float(fold, vec3_notf(fold_immvalue_vector(a)));
350 } else if (isstring(a)) {
352 if (OPTS_FLAG(TRUE_EMPTY_STRINGS))
353 return fold_constgen_float(fold, !fold_immvalue_string(a));
355 return fold_constgen_float(fold, !fold_immvalue_string(a) || !*fold_immvalue_string(a));
361 static GMQCC_INLINE ast_expression *fold_op_add(fold_t *fold, ast_value *a, ast_value *b) {
363 if (fold_can_2(a, b))
364 return fold_constgen_float(fold, fold_immvalue_float(a) + fold_immvalue_float(b));
365 } else if (isvector(a)) {
366 if (fold_can_2(a, b))
367 return fold_constgen_vector(fold, vec3_add(fold_immvalue_vector(a), fold_immvalue_vector(b)));
372 static GMQCC_INLINE ast_expression *fold_op_sub(fold_t *fold, ast_value *a, ast_value *b) {
374 if (fold_can_2(a, b))
375 return fold_constgen_float(fold, fold_immvalue_float(a) - fold_immvalue_float(b));
376 } else if (isvector(a)) {
377 if (fold_can_2(a, b))
378 return fold_constgen_vector(fold, vec3_sub(fold_immvalue_vector(a), fold_immvalue_vector(b)));
383 static GMQCC_INLINE ast_expression *fold_op_mul(fold_t *fold, ast_value *a, ast_value *b) {
385 if (isfloat(b) && fold_can_2(a, b))
386 return fold_constgen_vector(fold, vec3_mulvf(fold_immvalue_vector(b), fold_immvalue_float(a)));
387 else if (fold_can_2(a, b))
388 return fold_constgen_float(fold, fold_immvalue_float(a) * fold_immvalue_float(b));
389 } else if (isvector(a)) {
390 if (isfloat(b) && fold_can_2(a, b)) {
391 return fold_constgen_vector(fold, vec3_mulvf(fold_immvalue_vector(a), fold_immvalue_float(b)));
393 if (fold_can_2(a, b)) {
394 return fold_constgen_float(fold, vec3_mulvv(fold_immvalue_vector(a), fold_immvalue_vector(b)));
395 } else if (OPTS_OPTIMIZATION(OPTIM_VECTOR_COMPONENTS) && fold_can_1(a)) {
397 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(a), b, "xyz"))) return out;
398 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(a), b, "yxz"))) return out;
399 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(a), b, "zxy"))) return out;
400 } else if (OPTS_OPTIMIZATION(OPTIM_VECTOR_COMPONENTS) && fold_can_1(b)) {
402 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(b), a, "xyz"))) return out;
403 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(b), a, "yxz"))) return out;
404 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(b), a, "zxy"))) return out;
411 static GMQCC_INLINE ast_expression *fold_op_div(fold_t *fold, ast_value *a, ast_value *b) {
413 if (fold_can_2(a, b))
414 return fold_constgen_float(fold, fold_immvalue_float(a) / fold_immvalue_float(b));
415 } else if (isvector(a)) {
416 if (fold_can_2(a, b))
417 return fold_constgen_vector(fold, vec3_mulvf(fold_immvalue_vector(a), 1.0f / fold_immvalue_float(b)));
418 else if (fold_can_1(b))
419 return fold_constgen_float (fold, 1.0f / fold_immvalue_float(b));
424 static GMQCC_INLINE ast_expression *fold_op_mod(fold_t *fold, ast_value *a, ast_value *b) {
425 if (fold_can_2(a, b))
426 return fold_constgen_float(fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) % ((qcint_t)fold_immvalue_float(b))));
430 static GMQCC_INLINE ast_expression *fold_op_bor(fold_t *fold, ast_value *a, ast_value *b) {
431 if (fold_can_2(a, b))
432 return fold_constgen_float(fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) | ((qcint_t)fold_immvalue_float(b))));
436 static GMQCC_INLINE ast_expression *fold_op_band(fold_t *fold, ast_value *a, ast_value *b) {
437 if (fold_can_2(a, b))
438 return fold_constgen_float(fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) & ((qcint_t)fold_immvalue_float(b))));
442 static GMQCC_INLINE ast_expression *fold_op_xor(fold_t *fold, ast_value *a, ast_value *b) {
444 if (fold_can_2(a, b))
445 return fold_constgen_float(fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) ^ ((qcint_t)fold_immvalue_float(b))));
448 if (fold_can_2(a, b))
449 return fold_constgen_vector(fold, vec3_xor(fold_immvalue_vector(a), fold_immvalue_vector(b)));
451 if (fold_can_2(a, b))
452 return fold_constgen_vector(fold, vec3_xorvf(fold_immvalue_vector(a), fold_immvalue_float(b)));
458 static GMQCC_INLINE ast_expression *fold_op_lshift(fold_t *fold, ast_value *a, ast_value *b) {
459 if (fold_can_2(a, b) && isfloats(a, b))
460 return fold_constgen_float(fold, (qcfloat_t)((qcuint_t)(fold_immvalue_float(a)) << (qcuint_t)(fold_immvalue_float(b))));
464 static GMQCC_INLINE ast_expression *fold_op_rshift(fold_t *fold, ast_value *a, ast_value *b) {
465 if (fold_can_2(a, b) && isfloats(a, b))
466 return fold_constgen_float(fold, (qcfloat_t)((qcuint_t)(fold_immvalue_float(a)) >> (qcuint_t)(fold_immvalue_float(b))));
470 static GMQCC_INLINE ast_expression *fold_op_andor(fold_t *fold, ast_value *a, ast_value *b, float or) {
471 if (fold_can_2(a, b)) {
472 if (OPTS_FLAG(PERL_LOGIC)) {
473 if (fold_immediate_true(fold, a))
474 return (ast_expression*)b;
476 return fold_constgen_float (
478 ((or) ? (fold_immediate_true(fold, a) || fold_immediate_true(fold, b))
479 : (fold_immediate_true(fold, a) && fold_immediate_true(fold, b)))
488 static GMQCC_INLINE ast_expression *fold_op_tern(fold_t *fold, ast_value *a, ast_value *b, ast_value *c) {
490 return fold_immediate_true(fold, a)
492 : (ast_expression*)c;
497 static GMQCC_INLINE ast_expression *fold_op_exp(fold_t *fold, ast_value *a, ast_value *b) {
498 if (fold_can_2(a, b))
499 return fold_constgen_float(fold, (qcfloat_t)powf(fold_immvalue_float(a), fold_immvalue_float(b)));
503 static GMQCC_INLINE ast_expression *fold_op_lteqgt(fold_t *fold, ast_value *a, ast_value *b) {
504 if (fold_can_2(a,b)) {
505 if (fold_immvalue_float(a) < fold_immvalue_float(b)) return (ast_expression*)fold->imm_float[2];
506 if (fold_immvalue_float(a) == fold_immvalue_float(b)) return (ast_expression*)fold->imm_float[0];
507 if (fold_immvalue_float(a) > fold_immvalue_float(b)) return (ast_expression*)fold->imm_float[1];
512 static GMQCC_INLINE ast_expression *fold_op_cmp(fold_t *fold, ast_value *a, ast_value *b, bool ne) {
513 if (fold_can_2(a, b)) {
514 return fold_constgen_float(
516 (ne) ? (fold_immvalue_float(a) != fold_immvalue_float(b))
517 : (fold_immvalue_float(a) == fold_immvalue_float(b))
523 static GMQCC_INLINE ast_expression *fold_op_bnot(fold_t *fold, ast_value *a) {
525 return fold_constgen_float(fold, ~((qcint_t)fold_immvalue_float(a)));
529 ast_expression *fold_op(fold_t *fold, const oper_info *info, ast_expression **opexprs) {
530 ast_value *a = (ast_value*)opexprs[0];
531 ast_value *b = (ast_value*)opexprs[1];
532 ast_value *c = (ast_value*)opexprs[2];
534 /* can a fold operation be applied to this operator usage? */
538 switch(info->operands) {
539 case 3: if(!c) return NULL;
540 case 2: if(!b) return NULL;
543 compile_error(fold_ctx(fold), "interal error: fold_op no operands to fold\n");
549 case opid2('-','P'): return fold_op_neg (fold, a);
550 case opid2('!','P'): return fold_op_not (fold, a);
551 case opid1('+'): return fold_op_add (fold, a, b);
552 case opid1('-'): return fold_op_sub (fold, a, b);
553 case opid1('*'): return fold_op_mul (fold, a, b);
554 case opid1('/'): return fold_op_div (fold, a, b);
555 case opid1('%'): return fold_op_mod (fold, a, b);
556 case opid1('|'): return fold_op_bor (fold, a, b);
557 case opid1('&'): return fold_op_band (fold, a, b);
558 case opid1('^'): return fold_op_xor (fold, a, b);
559 case opid2('<','<'): return fold_op_lshift (fold, a, b);
560 case opid2('>','>'): return fold_op_rshift (fold, a, b);
561 case opid2('|','|'): return fold_op_andor (fold, a, b, true);
562 case opid2('&','&'): return fold_op_andor (fold, a, b, false);
563 case opid2('?',':'): return fold_op_tern (fold, a, b, c);
564 case opid2('*','*'): return fold_op_exp (fold, a, b);
565 case opid3('<','=','>'): return fold_op_lteqgt (fold, a, b);
566 case opid2('!','='): return fold_op_cmp (fold, a, b, true);
567 case opid2('=','='): return fold_op_cmp (fold, a, b, false);
568 case opid2('~','P'): return fold_op_bnot (fold, a);