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.
40 ast_expression **fold_const_values = NULL;
42 static GMQCC_INLINE bool fold_possible(const ast_value *val) {
43 return ast_istype((ast_expression*)val, ast_value) &&
44 val->hasvalue && (val->cvq == CV_CONST) &&
45 ((ast_expression*)val)->vtype != TYPE_FUNCTION; /* why not for functions? */
48 #define isfloatonly(X) (((ast_expression*)(X))->vtype == TYPE_FLOAT)
49 #define isvectoronly(X) (((ast_expression*)(X))->vtype == TYPE_VECTOR)
50 #define isstringonly(X) (((ast_expression*)(X))->vtype == TYPE_STRING)
51 #define isfloat(X) (isfloatonly (X) && fold_possible(X))
52 #define isvector(X) (isvectoronly(X) && fold_possible(X))
53 #define isstring(X) (isstringonly(X) && fold_possible(X))
54 #define isfloats(X,Y) (isfloat (X) && isfloat (Y))
55 #define isvectors(X,Y) (isvector (X) && isvector(Y))
56 #define isstrings(X,Y) (isstring (X) && isstring(Y))
59 * Implementation of basic vector math for vec3_t, for trivial constant
62 * TODO: gcc/clang hinting for autovectorization
64 static GMQCC_INLINE vec3_t vec3_add(vec3_t a, vec3_t b) {
72 static GMQCC_INLINE vec3_t vec3_sub(vec3_t a, vec3_t b) {
80 static GMQCC_INLINE vec3_t vec3_not(vec3_t a) {
88 static GMQCC_INLINE vec3_t vec3_neg(vec3_t a) {
96 static GMQCC_INLINE vec3_t vec3_xor(vec3_t a, vec3_t b) {
98 out.x = (qcfloat_t)((qcint_t)a.x ^ (qcint_t)b.x);
99 out.y = (qcfloat_t)((qcint_t)a.y ^ (qcint_t)b.y);
100 out.z = (qcfloat_t)((qcint_t)a.z ^ (qcint_t)b.z);
104 static GMQCC_INLINE vec3_t vec3_xorvf(vec3_t a, qcfloat_t b) {
106 out.x = (qcfloat_t)((qcint_t)a.x ^ (qcint_t)b);
107 out.y = (qcfloat_t)((qcint_t)a.y ^ (qcint_t)b);
108 out.z = (qcfloat_t)((qcint_t)a.z ^ (qcint_t)b);
112 static GMQCC_INLINE qcfloat_t vec3_mulvv(vec3_t a, vec3_t b) {
113 return (a.x * b.x + a.y * b.y + a.z * b.z);
117 static GMQCC_INLINE vec3_t vec3_mulvf(vec3_t a, qcfloat_t b) {
125 static GMQCC_INLINE bool vec3_cmp(vec3_t a, vec3_t b) {
131 static GMQCC_INLINE vec3_t vec3_create(float x, float y, float z) {
140 static GMQCC_INLINE float fold_immvalue_float(ast_value *expr) {
141 return expr->constval.vfloat;
143 static GMQCC_INLINE vec3_t fold_immvalue_vector(ast_value *expr) {
144 return expr->constval.vvec;
146 static GMQCC_INLINE const char *fold_immvalue_string(ast_value *expr) {
147 return expr->constval.vstring;
151 fold_t *fold_init(parser_t *parser) {
152 fold_t *fold = (fold_t*)mem_a(sizeof(fold_t));
153 fold->parser = parser;
154 fold->imm_float = NULL;
155 fold->imm_vector = NULL;
156 fold->imm_string = NULL;
157 fold->imm_string_untranslate = util_htnew(FOLD_STRING_UNTRANSLATE_HTSIZE);
158 fold->imm_string_dotranslate = util_htnew(FOLD_STRING_DOTRANSLATE_HTSIZE);
161 * prime the tables with common constant values at constant
164 (void)fold_constgen_float (fold, 0.0f);
165 (void)fold_constgen_float (fold, 1.0f);
166 (void)fold_constgen_float (fold, -1.0f);
168 (void)fold_constgen_vector(fold, vec3_create(0.0f, 0.0f, 0.0f));
173 bool fold_generate(fold_t *fold, ir_builder *ir) {
174 /* generate globals for immediate folded values */
178 for (i = 0; i < vec_size(fold->imm_float); ++i)
179 if (!ast_global_codegen ((cur = fold->imm_float[i]), ir, false)) goto err;
180 for (i = 0; i < vec_size(fold->imm_vector); ++i)
181 if (!ast_global_codegen((cur = fold->imm_vector[i]), ir, false)) goto err;
182 for (i = 0; i < vec_size(fold->imm_string); ++i)
183 if (!ast_global_codegen((cur = fold->imm_string[i]), ir, false)) goto err;
188 con_out("failed to generate global %s\n", cur->name);
189 ir_builder_delete(ir);
193 void fold_cleanup(fold_t *fold) {
196 for (i = 0; i < vec_size(fold->imm_float); ++i) ast_delete(fold->imm_float[i]);
197 for (i = 0; i < vec_size(fold->imm_vector); ++i) ast_delete(fold->imm_vector[i]);
198 for (i = 0; i < vec_size(fold->imm_string); ++i) ast_delete(fold->imm_string[i]);
200 vec_free(fold->imm_float);
201 vec_free(fold->imm_vector);
202 vec_free(fold->imm_string);
204 util_htdel(fold->imm_string_untranslate);
205 util_htdel(fold->imm_string_dotranslate);
210 static lex_ctx_t fold_ctx(fold_t *fold) {
212 if (fold->parser->lex)
213 return parser_ctx(fold->parser);
215 memset(&ctx, 0, sizeof(ctx));
219 ast_expression *fold_constgen_float(fold_t *fold, qcfloat_t value) {
220 ast_value *out = NULL;
223 for (i = 0; i < vec_size(fold->imm_float); i++) {
224 if (fold->imm_float[i]->constval.vfloat == value)
225 return (ast_expression*)fold->imm_float[i];
228 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_FLOAT);
230 out->hasvalue = true;
231 out->constval.vfloat = value;
233 vec_push(fold->imm_float, out);
235 return (ast_expression*)out;
238 ast_expression *fold_constgen_vector(fold_t *fold, vec3_t value) {
242 for (i = 0; i < vec_size(fold->imm_vector); i++) {
243 if (vec3_cmp(fold->imm_vector[i]->constval.vvec, value))
244 return (ast_expression*)fold->imm_vector[i];
247 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_VECTOR);
249 out->hasvalue = true;
250 out->constval.vvec = value;
252 vec_push(fold->imm_vector, out);
254 return (ast_expression*)out;
257 ast_expression *fold_constgen_string(fold_t *fold, const char *str, bool translate) {
258 hash_table_t *table = (translate) ? fold->imm_string_untranslate : fold->imm_string_dotranslate;
259 ast_value *out = NULL;
260 size_t hash = util_hthash(table, str);
262 if ((out = (ast_value*)util_htgeth(table, str, hash)))
263 return (ast_expression*)out;
267 util_snprintf(name, sizeof(name), "dotranslate_%lu", (unsigned long)(fold->parser->translated++));
268 out = ast_value_new(parser_ctx(fold->parser), name, TYPE_STRING);
269 out->expression.flags |= AST_FLAG_INCLUDE_DEF; /* def needs to be included for translatables */
271 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_STRING);
274 out->hasvalue = true;
276 out->constval.vstring = parser_strdup(str);
278 vec_push(fold->imm_string, out);
279 util_htseth(table, str, hash, out);
281 return (ast_expression*)out;
284 static GMQCC_INLINE ast_expression *fold_op_mul_vec(fold_t *fold, vec3_t *vec, ast_value *sel, const char *set) {
286 * vector-component constant folding works by matching the component sets
287 * to eliminate expensive operations on whole-vectors (3 components at runtime).
288 * to achive this effect in a clean manner this function generalizes the
289 * values through the use of a set paramater, which is used as an indexing method
290 * for creating the elided ast binary expression.
292 * Consider 'n 0 0' where y, and z need to be tested for 0, and x is
293 * used as the value in a binary operation generating an INSTR_MUL instruction
294 * to acomplish the indexing of the correct component value we use set[0], set[1], set[2]
295 * as x, y, z, where the values of those operations return 'x', 'y', 'z'. Because
296 * of how ASCII works we can easily deliniate:
297 * vec.z is the same as set[2]-'x' for when set[2] is 'z', 'z'-'x' results in a
298 * literal value of 2, using this 2, we know that taking the address of vec->x (float)
299 * and indxing it with this literal will yeild the immediate address of that component
301 * Of course more work needs to be done to generate the correct index for the ast_member_new
302 * call, which is no problem: set[0]-'x' suffices that job.
304 qcfloat_t x = (&vec->x)[set[0]-'x'];
305 qcfloat_t y = (&vec->x)[set[1]-'x'];
306 qcfloat_t z = (&vec->x)[set[2]-'x'];
310 ++opts_optimizationcount[OPTIM_VECTOR_COMPONENTS];
311 out = (ast_expression*)ast_member_new(fold_ctx(fold), (ast_expression*)sel, set[0]-'x', NULL);
312 out->node.keep = false;
313 ((ast_member*)out)->rvalue = true;
315 return (ast_expression*)ast_binary_new(fold_ctx(fold), INSTR_MUL_F, fold_constgen_float(fold, x), out);
322 static GMQCC_INLINE ast_expression *fold_op_mul(fold_t *fold, ast_value *a, ast_value *b) {
323 if (isfloatonly(a)) {
324 return (fold_possible(a) && fold_possible(b))
325 ? fold_constgen_vector(fold, vec3_mulvf(fold_immvalue_vector(b), fold_immvalue_float(a))) /* a=float, b=vector */
326 : NULL; /* cannot fold them */
327 } else if (isfloats(a, b)) {
328 return fold_constgen_float(fold, fold_immvalue_float(a) * fold_immvalue_float(b)); /* a=float, b=float */
329 } else if (isvectoronly(a)) {
330 if (isfloat(b) && fold_possible(a))
331 return fold_constgen_vector(fold, vec3_mulvf(fold_immvalue_vector(a), fold_immvalue_float(b))); /* a=vector, b=float */
332 else if (isvector(b)) {
334 * if we made it here the two ast values are both vectors. However because vectors are represented as
335 * three float values, constant folding can still occur within reason of the individual const-qualification
336 * of the components the vector is composed of.
338 if (fold_possible(a) && fold_possible(b))
339 return fold_constgen_float(fold, vec3_mulvv(fold_immvalue_vector(a), fold_immvalue_vector(b)));
340 else if (OPTS_OPTIMIZATION(OPTIM_VECTOR_COMPONENTS) && fold_possible(a)) {
341 vec3_t vec = fold_immvalue_vector(a);
343 if ((out = fold_op_mul_vec(fold, &vec, b, "xyz"))) return out;
344 if ((out = fold_op_mul_vec(fold, &vec, b, "yxz"))) return out;
345 if ((out = fold_op_mul_vec(fold, &vec, b, "zxy"))) return out;
347 } else if (OPTS_OPTIMIZATION(OPTIM_VECTOR_COMPONENTS) && fold_possible(b)) {
348 vec3_t vec = fold_immvalue_vector(b);
350 if ((out = fold_op_mul_vec(fold, &vec, a, "xyz"))) return out;
351 if ((out = fold_op_mul_vec(fold, &vec, a, "yxz"))) return out;
352 if ((out = fold_op_mul_vec(fold, &vec, a, "zxy"))) return out;
360 ast_expression *fold_op(fold_t *fold, const oper_info *info, ast_expression **opexprs) {
361 ast_value *a = (ast_value*)opexprs[0];
362 ast_value *b = (ast_value*)opexprs[1];
363 ast_value *c = (ast_value*)opexprs[2];
365 /* can a fold operation be applied to this operator usage? */
369 switch(info->operands) {
370 case 3: if(!c) return NULL;
371 case 2: if(!b) return NULL;
375 case opid2('-', 'P'):
376 return isfloat (a) ? fold_constgen_float (fold, fold_immvalue_float(a))
377 : isvector(a) ? fold_constgen_vector(fold, vec3_neg(fold_immvalue_vector(a)))
379 case opid2('!', 'P'):
380 return isfloat (a) ? fold_constgen_float (fold, !fold_immvalue_float(a))
381 : isvector(a) ? fold_constgen_vector(fold, vec3_not(fold_immvalue_vector(a)))
382 : isstring(a) ? fold_constgen_float (fold, !fold_immvalue_string(a) || OPTS_FLAG(TRUE_EMPTY_STRINGS) ? 0 : !*fold_immvalue_string(a))
385 return isfloats(a,b) ? fold_constgen_float (fold, fold_immvalue_float(a) + fold_immvalue_float(b))
386 : isvectors(a,b) ? fold_constgen_vector(fold, vec3_add(fold_immvalue_vector(a), fold_immvalue_vector(b)))
389 return isfloats(a,b) ? fold_constgen_float (fold, fold_immvalue_float(a) - fold_immvalue_float(b))
390 : isvectors(a,b) ? fold_constgen_vector(fold, vec3_sub(fold_immvalue_vector(a), fold_immvalue_vector(b)))
393 return isfloats(a,b) ? fold_constgen_float (fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) % ((qcint_t)fold_immvalue_float(b))))
396 return isfloats(a,b) ? fold_constgen_float (fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) | ((qcint_t)fold_immvalue_float(b))))
399 return isfloats(a,b) ? fold_constgen_float (fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) & ((qcint_t)fold_immvalue_float(b))))
402 return isfloats(a,b) ? fold_constgen_float (fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) ^ ((qcint_t)fold_immvalue_float(b))))
403 : isvectors(a,b) ? fold_constgen_vector(fold, vec3_xor (fold_immvalue_vector(a), fold_immvalue_vector(b)))
404 : isvector(a)&&isfloat(b) ? fold_constgen_vector(fold, vec3_xorvf(fold_immvalue_vector(a), fold_immvalue_float (b)))
407 return isfloats(a,b) ? fold_constgen_float (fold, (qcfloat_t)(((qcuint_t)(fold_immvalue_float(a)) << ((qcuint_t)fold_immvalue_float(b)))))
410 return isfloats(a,b) ? fold_constgen_float (fold, (qcfloat_t)(((qcuint_t)(fold_immvalue_float(a)) >> ((qcuint_t)fold_immvalue_float(b)))))
413 return isfloats(a,b) ? fold_constgen_float (fold, (qcfloat_t)powf(fold_immvalue_float(a), fold_immvalue_float(b)))
416 return isfloats(a,b) ? fold_constgen_float (fold, fold_immvalue_float(a) != fold_immvalue_float(b))
419 return isfloats(a,b) ? fold_constgen_float (fold, fold_immvalue_float(a) == fold_immvalue_float(b))
422 return isfloat(a) ? fold_constgen_float (fold, ~(qcint_t)fold_immvalue_float(a))
425 case opid1('*'): return fold_op_mul(fold, a, b);
427 /* TODO: seperate function for this case */
430 /* TODO: seperate function for this case */
433 /* TODO: seperate function for this case */
436 /* TODO: seperate function for this case */
438 case opid3('<','=','>'):
439 /* TODO: seperate function for this case */