2 * Copyright (C) 2012, 2013, 2014
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
32 /* The options to use for inexact and arithmetic exceptions */
33 #define FOLD_ROUNDING SFLOAT_ROUND_NEAREST_EVEN
34 #define FOLD_TINYNESS SFLOAT_TBEFORE
37 * The constant folder is also responsible for validating if the constant
38 * expressions produce valid results. We cannot trust the FPU control
39 * unit for these exceptions because setting FPU control words might not
40 * work. Systems can set and enforce FPU modes of operation. It's also valid
41 * for libc's to simply ignore FPU exceptions. For instance ARM CPUs in
42 * glibc. We implement some trivial and IEE 754 conformant functions which
43 * emulate those operations. This is an entierly optional compiler feature
44 * which shouldn't be enabled for anything other than performing strict
45 * passes on constant expressions since it's quite slow.
47 typedef uint32_t sfloat_t;
58 SFLOAT_UNDERFLOW = 16,
60 } sfloat_exceptionflags_t;
63 SFLOAT_ROUND_NEAREST_EVEN,
67 } sfloat_roundingmode_t;
75 sfloat_roundingmode_t roundingmode;
76 sfloat_exceptionflags_t exceptionflags;
77 sfloat_tdetect_t tiny;
80 /* Count of leading zero bits before the most-significand 1 bit. */
82 /* MSVC has an intrinsic for this */
83 static GMQCC_INLINE uint32_t sfloat_clz(uint32_t x) {
85 _BitScanForward(&r, x);
88 # define SFLOAT_CLZ(X, SUB) \
89 (sfloat_clz((X)) - (SUB))
90 #elif defined(__GNUC__) || defined(__CLANG__)
91 /* Clang and GCC have a builtin for this */
92 # define SFLOAT_CLZ(X, SUB) \
93 (__builtin_clz((X)) - (SUB))
96 static GMQCC_INLINE uint32_t sfloat_popcnt(uint32_t x) {
97 x -= ((x >> 1) & 0x55555555);
98 x = (((x >> 2) & 0x33333333) + (x & 0x33333333));
99 x = (((x >> 4) + x) & 0x0F0F0F0F);
102 return x & 0x0000003F;
104 static GMQCC_INLINE uint32_t sfloat_clz(uint32_t x) {
110 return 32 - sfloat_popcnt(x);
112 # define SFLOAT_CLZ(X, SUB) \
113 (sfloat_clz((X) - (SUB)))
116 /* The value of a NaN */
117 #define SFLOAT_NAN 0xFFC00000
119 #define SFLOAT_ISNAN(A) \
120 (0xFF000000 < (uint32_t)((A) << 1))
121 /* Test if signaling NaN */
122 #define SFLOAT_ISSNAN(A) \
123 (((((A) >> 22) & 0x1FF) == 0x1FE) && ((A) & 0x003FFFFF))
124 /* Raise exception */
125 #define SFLOAT_RAISE(STATE, FLAGS) \
126 ((STATE)->exceptionflags |= (FLAGS))
128 * Shifts `A' right `COUNT' bits. Non-zero bits are stored in LSB. Size
129 * sets the arbitrarly-large limit.
131 #define SFLOAT_SHIFT(SIZE, A, COUNT, Z) \
132 *(Z) = ((COUNT) == 0) \
134 : (((COUNT) < (SIZE)) \
135 ? ((A) >> (COUNT)) | (((A) << ((-(COUNT)) & ((SIZE) - 1))) != 0) \
137 /* Extract fractional component */
138 #define SFLOAT_EXTRACT_FRAC(X) \
139 ((uint32_t)((X) & 0x007FFFFF))
140 /* Extract exponent component */
141 #define SFLOAT_EXTRACT_EXP(X) \
142 ((int16_t)((X) >> 23) & 0xFF)
143 /* Extract sign bit */
144 #define SFLOAT_EXTRACT_SIGN(X) \
146 /* Normalize a subnormal */
147 #define SFLOAT_SUBNORMALIZE(SA, Z, SZ) \
148 (void)(*(SZ) = (SA) << SFLOAT_CLZ((SA), 8), *(SZ) = 1 - SFLOAT_CLZ((SA), 8))
150 * Pack sign, exponent and significand and produce a float.
152 * Integer portions of the significand are added to the exponent. The
153 * exponent input should be one less than the result exponent whenever
154 * the significand is normalized since normalized significand will
155 * always have an integer portion of value one.
157 #define SFLOAT_PACK(SIGN, EXP, SIG) \
158 (sfloat_t)((((uint32_t)(SIGN)) << 31) + (((uint32_t)(EXP)) << 23) + (SIG))
160 /* Calculate NaN. If either operands are signaling then raise invalid */
161 static sfloat_t sfloat_propagate_nan(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
162 bool isnan_a = SFLOAT_ISNAN(a);
163 bool issnan_a = SFLOAT_ISSNAN(a);
164 bool isnan_b = SFLOAT_ISNAN(b);
165 bool issnan_b = SFLOAT_ISSNAN(b);
170 if (issnan_a | issnan_b)
171 SFLOAT_RAISE(state, SFLOAT_INEXACT);
175 return isnan_b ? b : a;
176 } else if (isnan_a) {
177 if (issnan_b | !isnan_b)
180 if ((uint32_t)(a << 1) < (uint32_t)(b << 1)) return b;
181 if ((uint32_t)(b << 1) < (uint32_t)(a << 1)) return a;
182 return (a < b) ? a : b;
188 static sfloat_t SFLOAT_PACK_round(sfloat_state_t *state, bool sign_z, int16_t exp_z, uint32_t sig_z) {
189 sfloat_roundingmode_t mode = state->roundingmode;
190 bool even = !!(mode == SFLOAT_ROUND_NEAREST_EVEN);
191 unsigned char increment = 0x40;
192 unsigned char bits = sig_z & 0x7F;
195 if (mode == SFLOAT_ROUND_TO_ZERO)
200 if (mode == SFLOAT_ROUND_UP)
203 if (mode == SFLOAT_ROUND_DOWN)
209 if (0xFD <= (uint16_t)exp_z) {
210 if ((0xFD < exp_z) || ((exp_z == 0xFD) && ((int32_t)(sig_z + increment) < 0))) {
211 SFLOAT_RAISE(state, SFLOAT_OVERFLOW | SFLOAT_INEXACT);
212 return SFLOAT_PACK(sign_z, 0xFF, 0) - (increment == 0);
215 /* Check for underflow */
216 bool tiny = (state->tiny == SFLOAT_TBEFORE) || (exp_z < -1) || (sig_z + increment < 0x80000000);
217 SFLOAT_SHIFT(32, sig_z, -exp_z, &sig_z);
221 SFLOAT_RAISE(state, SFLOAT_UNDERFLOW);
226 * Significand has point between bits 30 and 29, 7 bits to the left of
227 * the usual place. This shifted significand has to be normalized
228 * or smaller, if it isn't the exponent must be zero, in which case
229 * no rounding occurs since the result will be a subnormal.
232 SFLOAT_RAISE(state, SFLOAT_INEXACT);
233 sig_z = (sig_z + increment) >> 7;
234 sig_z &= ~(((bits ^ 0x40) == 0) & even);
237 return SFLOAT_PACK(sign_z, exp_z, sig_z);
240 /* Normalized round and pack */
241 static sfloat_t SFLOAT_PACK_normal(sfloat_state_t *state, bool sign_z, int16_t exp_z, uint32_t sig_z) {
242 unsigned char c = SFLOAT_CLZ(sig_z, 1);
243 return SFLOAT_PACK_round(state, sign_z, exp_z - c, sig_z << c);
246 static sfloat_t sfloat_add_impl(sfloat_state_t *state, sfloat_t a, sfloat_t b, bool sign_z) {
247 int16_t exp_a = SFLOAT_EXTRACT_EXP(a);
248 int16_t exp_b = SFLOAT_EXTRACT_EXP(b);
250 int16_t exp_d = exp_a - exp_b;
251 uint32_t sig_a = SFLOAT_EXTRACT_FRAC(a) << 6;
252 uint32_t sig_b = SFLOAT_EXTRACT_FRAC(b) << 6;
257 return sig_a ? sfloat_propagate_nan(state, a, b) : a;
262 SFLOAT_SHIFT(32, sig_b, exp_d, &sig_b);
264 } else if (exp_d < 0) {
266 return sig_b ? sfloat_propagate_nan(state, a, b) : SFLOAT_PACK(sign_z, 0xFF, 0);
271 SFLOAT_SHIFT(32, sig_a, -exp_d, &sig_a);
275 return (sig_a | sig_b) ? sfloat_propagate_nan(state, a, b) : a;
277 return SFLOAT_PACK(sign_z, 0, (sig_a + sig_b) >> 6);
278 sig_z = 0x40000000 + sig_a + sig_b;
283 sig_z = (sig_a + sig_b) << 1;
285 if ((int32_t)sig_z < 0) {
286 sig_z = sig_a + sig_b;
290 return SFLOAT_PACK_round(state, sign_z, exp_z, sig_z);
293 static sfloat_t sfloat_sub_impl(sfloat_state_t *state, sfloat_t a, sfloat_t b, bool sign_z) {
294 int16_t exp_a = SFLOAT_EXTRACT_EXP(a);
295 int16_t exp_b = SFLOAT_EXTRACT_EXP(b);
297 int16_t exp_d = exp_a - exp_b;
298 uint32_t sig_a = SFLOAT_EXTRACT_FRAC(a) << 7;
299 uint32_t sig_b = SFLOAT_EXTRACT_FRAC(b) << 7;
302 if (0 < exp_d) goto exp_greater_a;
303 if (exp_d < 0) goto exp_greater_b;
307 return sfloat_propagate_nan(state, a, b);
308 SFLOAT_RAISE(state, SFLOAT_INVALID);
315 if (sig_b < sig_a) goto greater_a;
316 if (sig_a < sig_b) goto greater_b;
318 return SFLOAT_PACK(state->roundingmode == SFLOAT_ROUND_DOWN, 0, 0);
322 return (sig_b) ? sfloat_propagate_nan(state, a, b) : SFLOAT_PACK(sign_z ^ 1, 0xFF, 0);
327 SFLOAT_SHIFT(32, sig_a, -exp_d, &sig_a);
330 sig_z = sig_b - sig_a;
337 return (sig_a) ? sfloat_propagate_nan(state, a, b) : a;
342 SFLOAT_SHIFT(32, sig_b, exp_d, &sig_b);
345 sig_z = sig_a - sig_b;
350 return SFLOAT_PACK_normal(state, sign_z, exp_z, sig_z);
353 static GMQCC_INLINE sfloat_t sfloat_add(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
354 bool sign_a = SFLOAT_EXTRACT_SIGN(a);
355 bool sign_b = SFLOAT_EXTRACT_SIGN(b);
356 return (sign_a == sign_b) ? sfloat_add_impl(state, a, b, sign_a)
357 : sfloat_sub_impl(state, a, b, sign_a);
360 static GMQCC_INLINE sfloat_t sfloat_sub(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
361 bool sign_a = SFLOAT_EXTRACT_SIGN(a);
362 bool sign_b = SFLOAT_EXTRACT_SIGN(b);
363 return (sign_a == sign_b) ? sfloat_sub_impl(state, a, b, sign_a)
364 : sfloat_add_impl(state, a, b, sign_a);
367 static sfloat_t sfloat_mul(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
368 int16_t exp_a = SFLOAT_EXTRACT_EXP(a);
369 int16_t exp_b = SFLOAT_EXTRACT_EXP(b);
371 uint32_t sig_a = SFLOAT_EXTRACT_FRAC(a);
372 uint32_t sig_b = SFLOAT_EXTRACT_FRAC(b);
374 uint64_t sig_z64 = 0;
375 bool sign_a = SFLOAT_EXTRACT_SIGN(a);
376 bool sign_b = SFLOAT_EXTRACT_SIGN(b);
377 bool sign_z = sign_a ^ sign_b;
380 if (sig_a || ((exp_b == 0xFF) && sig_b))
381 return sfloat_propagate_nan(state, a, b);
382 if ((exp_b | sig_b) == 0) {
383 SFLOAT_RAISE(state, SFLOAT_INVALID);
386 return SFLOAT_PACK(sign_z, 0xFF, 0);
390 return sfloat_propagate_nan(state, a, b);
391 if ((exp_a | sig_a) == 0) {
392 SFLOAT_RAISE(state, SFLOAT_INVALID);
395 return SFLOAT_PACK(sign_z, 0xFF, 0);
399 return SFLOAT_PACK(sign_z, 0, 0);
400 SFLOAT_SUBNORMALIZE(sig_a, &exp_a, &sig_a);
404 return SFLOAT_PACK(sign_z, 0, 0);
405 SFLOAT_SUBNORMALIZE(sig_b, &exp_b, &sig_b);
407 exp_z = exp_a + exp_b - 0x7F;
408 sig_a = (sig_a | 0x00800000) << 7;
409 sig_b = (sig_b | 0x00800000) << 8;
410 SFLOAT_SHIFT(64, ((uint64_t)sig_a) * sig_b, 32, &sig_z64);
412 if (0 <= (int32_t)(sig_z << 1)) {
416 return SFLOAT_PACK_round(state, sign_z, exp_z, sig_z);
419 static sfloat_t sfloat_div(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
420 int16_t exp_a = SFLOAT_EXTRACT_EXP(a);
421 int16_t exp_b = SFLOAT_EXTRACT_EXP(b);
423 uint32_t sig_a = SFLOAT_EXTRACT_FRAC(a);
424 uint32_t sig_b = SFLOAT_EXTRACT_FRAC(b);
426 bool sign_a = SFLOAT_EXTRACT_SIGN(a);
427 bool sign_b = SFLOAT_EXTRACT_SIGN(b);
428 bool sign_z = sign_a ^ sign_b;
432 return sfloat_propagate_nan(state, a, b);
435 return sfloat_propagate_nan(state, a, b);
436 SFLOAT_RAISE(state, SFLOAT_INVALID);
439 return SFLOAT_PACK(sign_z, 0xFF, 0);
442 return (sig_b) ? sfloat_propagate_nan(state, a, b) : SFLOAT_PACK(sign_z, 0, 0);
445 if ((exp_a | sig_a) == 0) {
446 SFLOAT_RAISE(state, SFLOAT_INVALID);
449 SFLOAT_RAISE(state, SFLOAT_DIVBYZERO);
450 return SFLOAT_PACK(sign_z, 0xFF, 0);
452 SFLOAT_SUBNORMALIZE(sig_b, &exp_b, &sig_b);
456 return SFLOAT_PACK(sign_z, 0, 0);
457 SFLOAT_SUBNORMALIZE(sig_a, &exp_a, &sig_a);
459 exp_z = exp_a - exp_b + 0x7D;
460 sig_a = (sig_a | 0x00800000) << 7;
461 sig_b = (sig_b | 0x00800000) << 8;
462 if (sig_b <= (sig_a + sig_a)) {
466 sig_z = (((uint64_t)sig_a) << 32) / sig_b;
467 if ((sig_z & 0x3F) == 0)
468 sig_z |= ((uint64_t)sig_b * sig_z != ((uint64_t)sig_a) << 32);
469 return SFLOAT_PACK_round(state, sign_z, exp_z, sig_z);
472 static GMQCC_INLINE void sfloat_check(lex_ctx_t ctx, sfloat_state_t *state, const char *vec) {
473 /* Exception comes from vector component */
475 if (state->exceptionflags & SFLOAT_DIVBYZERO)
476 compile_error(ctx, "division by zero in `%s' component", vec);
477 if (state->exceptionflags & SFLOAT_INVALID)
478 compile_error(ctx, "undefined (inf) in `%s' component", vec);
479 if (state->exceptionflags & SFLOAT_OVERFLOW)
480 compile_error(ctx, "arithmetic overflow in `%s' component", vec);
481 if (state->exceptionflags & SFLOAT_UNDERFLOW)
482 compile_error(ctx, "arithmetic underflow in `%s' component", vec);
485 if (state->exceptionflags & SFLOAT_DIVBYZERO)
486 compile_error(ctx, "division by zero");
487 if (state->exceptionflags & SFLOAT_INVALID)
488 compile_error(ctx, "undefined (inf)");
489 if (state->exceptionflags & SFLOAT_OVERFLOW)
490 compile_error(ctx, "arithmetic overflow");
491 if (state->exceptionflags & SFLOAT_UNDERFLOW)
492 compile_error(ctx, "arithmetic underflow");
495 static GMQCC_INLINE void sfloat_init(sfloat_state_t *state) {
496 state->exceptionflags = 0;
497 state->roundingmode = FOLD_ROUNDING;
498 state->tiny = FOLD_TINYNESS;
502 * There is two stages to constant folding in GMQCC: there is the parse
503 * stage constant folding, where, witht he help of the AST, operator
504 * usages can be constant folded. Then there is the constant folding
505 * in the IR for things like eliding if statements, can occur.
507 * This file is thus, split into two parts.
510 #define isfloat(X) (((ast_expression*)(X))->vtype == TYPE_FLOAT)
511 #define isvector(X) (((ast_expression*)(X))->vtype == TYPE_VECTOR)
512 #define isstring(X) (((ast_expression*)(X))->vtype == TYPE_STRING)
513 #define isfloats(X,Y) (isfloat (X) && isfloat (Y))
516 * Implementation of basic vector math for vec3_t, for trivial constant
519 * TODO: gcc/clang hinting for autovectorization
535 sfloat_state_t state[3];
538 static GMQCC_INLINE vec3_soft_t vec3_soft_convert(vec3_t vec) {
546 static GMQCC_INLINE bool vec3_soft_exception(vec3_soft_state_t *vstate, size_t index) {
547 sfloat_exceptionflags_t flags = vstate->state[index].exceptionflags;
548 if (flags & SFLOAT_DIVBYZERO) return true;
549 if (flags & SFLOAT_INVALID) return true;
550 if (flags & SFLOAT_OVERFLOW) return true;
551 if (flags & SFLOAT_UNDERFLOW) return true;
555 static GMQCC_INLINE void vec3_soft_eval(vec3_soft_state_t *state,
556 sfloat_t (*callback)(sfloat_state_t *, sfloat_t, sfloat_t),
560 vec3_soft_t sa = vec3_soft_convert(a);
561 vec3_soft_t sb = vec3_soft_convert(b);
562 callback(&state->state[0], sa.x.s, sb.x.s);
563 if (vec3_soft_exception(state, 0)) state->faults |= VEC_COMP_X;
564 callback(&state->state[1], sa.y.s, sb.y.s);
565 if (vec3_soft_exception(state, 1)) state->faults |= VEC_COMP_Y;
566 callback(&state->state[2], sa.z.s, sb.z.s);
567 if (vec3_soft_exception(state, 2)) state->faults |= VEC_COMP_Z;
570 static GMQCC_INLINE void vec3_check_except(vec3_t a,
573 sfloat_t (*callback)(sfloat_state_t *, sfloat_t, sfloat_t))
575 vec3_soft_state_t state;
577 if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
580 sfloat_init(&state.state[0]);
581 sfloat_init(&state.state[1]);
582 sfloat_init(&state.state[2]);
584 vec3_soft_eval(&state, callback, a, b);
585 if (state.faults & VEC_COMP_X) sfloat_check(ctx, &state.state[0], "x");
586 if (state.faults & VEC_COMP_Y) sfloat_check(ctx, &state.state[1], "y");
587 if (state.faults & VEC_COMP_Z) sfloat_check(ctx, &state.state[2], "z");
590 static GMQCC_INLINE vec3_t vec3_add(lex_ctx_t ctx, vec3_t a, vec3_t b) {
592 vec3_check_except(a, b, ctx, &sfloat_add);
599 static GMQCC_INLINE vec3_t vec3_sub(lex_ctx_t ctx, vec3_t a, vec3_t b) {
601 vec3_check_except(a, b, ctx, &sfloat_sub);
608 static GMQCC_INLINE vec3_t vec3_neg(vec3_t a) {
616 static GMQCC_INLINE vec3_t vec3_or(vec3_t a, vec3_t b) {
618 out.x = (qcfloat_t)(((qcint_t)a.x) | ((qcint_t)b.x));
619 out.y = (qcfloat_t)(((qcint_t)a.y) | ((qcint_t)b.y));
620 out.z = (qcfloat_t)(((qcint_t)a.z) | ((qcint_t)b.z));
624 static GMQCC_INLINE vec3_t vec3_orvf(vec3_t a, qcfloat_t b) {
626 out.x = (qcfloat_t)(((qcint_t)a.x) | ((qcint_t)b));
627 out.y = (qcfloat_t)(((qcint_t)a.y) | ((qcint_t)b));
628 out.z = (qcfloat_t)(((qcint_t)a.z) | ((qcint_t)b));
632 static GMQCC_INLINE vec3_t vec3_and(vec3_t a, vec3_t b) {
634 out.x = (qcfloat_t)(((qcint_t)a.x) & ((qcint_t)b.x));
635 out.y = (qcfloat_t)(((qcint_t)a.y) & ((qcint_t)b.y));
636 out.z = (qcfloat_t)(((qcint_t)a.z) & ((qcint_t)b.z));
640 static GMQCC_INLINE vec3_t vec3_andvf(vec3_t a, qcfloat_t b) {
642 out.x = (qcfloat_t)(((qcint_t)a.x) & ((qcint_t)b));
643 out.y = (qcfloat_t)(((qcint_t)a.y) & ((qcint_t)b));
644 out.z = (qcfloat_t)(((qcint_t)a.z) & ((qcint_t)b));
648 static GMQCC_INLINE vec3_t vec3_xor(vec3_t a, vec3_t b) {
650 out.x = (qcfloat_t)(((qcint_t)a.x) ^ ((qcint_t)b.x));
651 out.y = (qcfloat_t)(((qcint_t)a.y) ^ ((qcint_t)b.y));
652 out.z = (qcfloat_t)(((qcint_t)a.z) ^ ((qcint_t)b.z));
656 static GMQCC_INLINE vec3_t vec3_xorvf(vec3_t a, qcfloat_t b) {
658 out.x = (qcfloat_t)(((qcint_t)a.x) ^ ((qcint_t)b));
659 out.y = (qcfloat_t)(((qcint_t)a.y) ^ ((qcint_t)b));
660 out.z = (qcfloat_t)(((qcint_t)a.z) ^ ((qcint_t)b));
664 static GMQCC_INLINE vec3_t vec3_not(vec3_t a) {
672 static GMQCC_INLINE qcfloat_t vec3_mulvv(lex_ctx_t ctx, vec3_t a, vec3_t b) {
678 if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
681 sa = vec3_soft_convert(a);
682 sb = vec3_soft_convert(b);
690 r[0] = sfloat_mul(&s[0], sa.x.s, sb.x.s);
691 r[1] = sfloat_mul(&s[1], sa.y.s, sb.y.s);
692 r[2] = sfloat_mul(&s[2], sa.z.s, sb.z.s);
693 r[3] = sfloat_add(&s[3], r[0], r[1]);
694 r[4] = sfloat_add(&s[4], r[3], r[2]);
696 sfloat_check(ctx, &s[0], NULL);
697 sfloat_check(ctx, &s[1], NULL);
698 sfloat_check(ctx, &s[2], NULL);
699 sfloat_check(ctx, &s[3], NULL);
700 sfloat_check(ctx, &s[4], NULL);
703 return (a.x * b.x + a.y * b.y + a.z * b.z);
706 static GMQCC_INLINE vec3_t vec3_mulvf(lex_ctx_t ctx, vec3_t a, qcfloat_t b) {
712 if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
715 sa = vec3_soft_convert(a);
721 sfloat_mul(&s[0], sa.x.s, sb.s);
722 sfloat_mul(&s[1], sa.y.s, sb.s);
723 sfloat_mul(&s[2], sa.z.s, sb.s);
725 sfloat_check(ctx, &s[0], "x");
726 sfloat_check(ctx, &s[1], "y");
727 sfloat_check(ctx, &s[2], "z");
736 static GMQCC_INLINE bool vec3_cmp(vec3_t a, vec3_t b) {
742 static GMQCC_INLINE vec3_t vec3_create(float x, float y, float z) {
750 static GMQCC_INLINE qcfloat_t vec3_notf(vec3_t a) {
751 return (!a.x && !a.y && !a.z);
754 static GMQCC_INLINE bool vec3_pbool(vec3_t a) {
755 return (a.x || a.y || a.z);
758 static GMQCC_INLINE vec3_t vec3_cross(lex_ctx_t ctx, vec3_t a, vec3_t b) {
765 if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
768 sa = vec3_soft_convert(a);
769 sb = vec3_soft_convert(b);
781 r[0] = sfloat_mul(&s[0], sa.y.s, sb.z.s);
782 r[1] = sfloat_mul(&s[1], sa.z.s, sb.y.s);
783 r[2] = sfloat_mul(&s[2], sa.z.s, sb.x.s);
784 r[3] = sfloat_mul(&s[3], sa.x.s, sb.z.s);
785 r[4] = sfloat_mul(&s[4], sa.x.s, sb.y.s);
786 r[5] = sfloat_mul(&s[5], sa.y.s, sb.x.s);
787 r[6] = sfloat_sub(&s[6], r[0], r[1]);
788 r[7] = sfloat_sub(&s[7], r[2], r[3]);
789 r[8] = sfloat_sub(&s[8], r[4], r[5]);
791 sfloat_check(ctx, &s[0], NULL);
792 sfloat_check(ctx, &s[1], NULL);
793 sfloat_check(ctx, &s[2], NULL);
794 sfloat_check(ctx, &s[3], NULL);
795 sfloat_check(ctx, &s[4], NULL);
796 sfloat_check(ctx, &s[5], NULL);
797 sfloat_check(ctx, &s[6], "x");
798 sfloat_check(ctx, &s[7], "y");
799 sfloat_check(ctx, &s[8], "z");
802 out.x = a.y * b.z - a.z * b.y;
803 out.y = a.z * b.x - a.x * b.z;
804 out.z = a.x * b.y - a.y * b.x;
808 static lex_ctx_t fold_ctx(fold_t *fold) {
810 if (fold->parser->lex)
811 return parser_ctx(fold->parser);
813 memset(&ctx, 0, sizeof(ctx));
817 static GMQCC_INLINE bool fold_immediate_true(fold_t *fold, ast_value *v) {
818 switch (v->expression.vtype) {
820 return !!v->constval.vfloat;
822 return !!v->constval.vint;
824 if (OPTS_FLAG(CORRECT_LOGIC))
825 return vec3_pbool(v->constval.vvec);
826 return !!(v->constval.vvec.x);
828 if (!v->constval.vstring)
830 if (OPTS_FLAG(TRUE_EMPTY_STRINGS))
832 return !!v->constval.vstring[0];
834 compile_error(fold_ctx(fold), "internal error: fold_immediate_true on invalid type");
837 return !!v->constval.vfunc;
840 /* Handy macros to determine if an ast_value can be constant folded. */
841 #define fold_can_1(X) \
842 (ast_istype(((ast_expression*)(X)), ast_value) && (X)->hasvalue && ((X)->cvq == CV_CONST) && \
843 ((ast_expression*)(X))->vtype != TYPE_FUNCTION)
845 #define fold_can_2(X, Y) (fold_can_1(X) && fold_can_1(Y))
847 #define fold_immvalue_float(E) ((E)->constval.vfloat)
848 #define fold_immvalue_vector(E) ((E)->constval.vvec)
849 #define fold_immvalue_string(E) ((E)->constval.vstring)
851 fold_t *fold_init(parser_t *parser) {
852 fold_t *fold = (fold_t*)mem_a(sizeof(fold_t));
853 fold->parser = parser;
854 fold->imm_float = NULL;
855 fold->imm_vector = NULL;
856 fold->imm_string = NULL;
857 fold->imm_string_untranslate = util_htnew(FOLD_STRING_UNTRANSLATE_HTSIZE);
858 fold->imm_string_dotranslate = util_htnew(FOLD_STRING_DOTRANSLATE_HTSIZE);
861 * prime the tables with common constant values at constant
864 (void)fold_constgen_float (fold, 0.0f, false);
865 (void)fold_constgen_float (fold, 1.0f, false);
866 (void)fold_constgen_float (fold, -1.0f, false);
867 (void)fold_constgen_float (fold, 2.0f, false);
869 (void)fold_constgen_vector(fold, vec3_create(0.0f, 0.0f, 0.0f));
870 (void)fold_constgen_vector(fold, vec3_create(-1.0f, -1.0f, -1.0f));
875 bool fold_generate(fold_t *fold, ir_builder *ir) {
876 /* generate globals for immediate folded values */
880 for (i = 0; i < vec_size(fold->imm_float); ++i)
881 if (!ast_global_codegen ((cur = fold->imm_float[i]), ir, false)) goto err;
882 for (i = 0; i < vec_size(fold->imm_vector); ++i)
883 if (!ast_global_codegen((cur = fold->imm_vector[i]), ir, false)) goto err;
884 for (i = 0; i < vec_size(fold->imm_string); ++i)
885 if (!ast_global_codegen((cur = fold->imm_string[i]), ir, false)) goto err;
890 con_out("failed to generate global %s\n", cur->name);
891 ir_builder_delete(ir);
895 void fold_cleanup(fold_t *fold) {
898 for (i = 0; i < vec_size(fold->imm_float); ++i) ast_delete(fold->imm_float[i]);
899 for (i = 0; i < vec_size(fold->imm_vector); ++i) ast_delete(fold->imm_vector[i]);
900 for (i = 0; i < vec_size(fold->imm_string); ++i) ast_delete(fold->imm_string[i]);
902 vec_free(fold->imm_float);
903 vec_free(fold->imm_vector);
904 vec_free(fold->imm_string);
906 util_htdel(fold->imm_string_untranslate);
907 util_htdel(fold->imm_string_dotranslate);
912 ast_expression *fold_constgen_float(fold_t *fold, qcfloat_t value, bool inexact) {
913 ast_value *out = NULL;
916 for (i = 0; i < vec_size(fold->imm_float); i++) {
917 if (!memcmp(&fold->imm_float[i]->constval.vfloat, &value, sizeof(qcfloat_t)))
918 return (ast_expression*)fold->imm_float[i];
921 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_FLOAT);
923 out->hasvalue = true;
924 out->inexact = inexact;
925 out->constval.vfloat = value;
927 vec_push(fold->imm_float, out);
929 return (ast_expression*)out;
932 ast_expression *fold_constgen_vector(fold_t *fold, vec3_t value) {
936 for (i = 0; i < vec_size(fold->imm_vector); i++) {
937 if (vec3_cmp(fold->imm_vector[i]->constval.vvec, value))
938 return (ast_expression*)fold->imm_vector[i];
941 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_VECTOR);
943 out->hasvalue = true;
944 out->constval.vvec = value;
946 vec_push(fold->imm_vector, out);
948 return (ast_expression*)out;
951 ast_expression *fold_constgen_string(fold_t *fold, const char *str, bool translate) {
952 hash_table_t *table = (translate) ? fold->imm_string_untranslate : fold->imm_string_dotranslate;
953 ast_value *out = NULL;
954 size_t hash = util_hthash(table, str);
956 if ((out = (ast_value*)util_htgeth(table, str, hash)))
957 return (ast_expression*)out;
961 util_snprintf(name, sizeof(name), "dotranslate_%lu", (unsigned long)(fold->parser->translated++));
962 out = ast_value_new(parser_ctx(fold->parser), name, TYPE_STRING);
963 out->expression.flags |= AST_FLAG_INCLUDE_DEF; /* def needs to be included for translatables */
965 out = ast_value_new(fold_ctx(fold), "#IMMEDIATE", TYPE_STRING);
968 out->hasvalue = true;
970 out->constval.vstring = parser_strdup(str);
972 vec_push(fold->imm_string, out);
973 util_htseth(table, str, hash, out);
975 return (ast_expression*)out;
979 static GMQCC_INLINE ast_expression *fold_op_mul_vec(fold_t *fold, vec3_t vec, ast_value *sel, const char *set) {
981 * vector-component constant folding works by matching the component sets
982 * to eliminate expensive operations on whole-vectors (3 components at runtime).
983 * to achive this effect in a clean manner this function generalizes the
984 * values through the use of a set paramater, which is used as an indexing method
985 * for creating the elided ast binary expression.
987 * Consider 'n 0 0' where y, and z need to be tested for 0, and x is
988 * used as the value in a binary operation generating an INSTR_MUL instruction,
989 * to acomplish the indexing of the correct component value we use set[0], set[1], set[2]
990 * as x, y, z, where the values of those operations return 'x', 'y', 'z'. Because
991 * of how ASCII works we can easily deliniate:
992 * vec.z is the same as set[2]-'x' for when set[2] is 'z', 'z'-'x' results in a
993 * literal value of 2, using this 2, we know that taking the address of vec->x (float)
994 * and indxing it with this literal will yeild the immediate address of that component
996 * Of course more work needs to be done to generate the correct index for the ast_member_new
997 * call, which is no problem: set[0]-'x' suffices that job.
999 qcfloat_t x = (&vec.x)[set[0]-'x'];
1000 qcfloat_t y = (&vec.x)[set[1]-'x'];
1001 qcfloat_t z = (&vec.x)[set[2]-'x'];
1004 ast_expression *out;
1005 ++opts_optimizationcount[OPTIM_VECTOR_COMPONENTS];
1006 out = (ast_expression*)ast_member_new(fold_ctx(fold), (ast_expression*)sel, set[0]-'x', NULL);
1007 out->node.keep = false;
1008 ((ast_member*)out)->rvalue = true;
1010 return (ast_expression*)ast_binary_new(fold_ctx(fold), INSTR_MUL_F, fold_constgen_float(fold, x, false), out);
1016 static GMQCC_INLINE ast_expression *fold_op_neg(fold_t *fold, ast_value *a) {
1019 return fold_constgen_float(fold, -fold_immvalue_float(a), false);
1020 } else if (isvector(a)) {
1022 return fold_constgen_vector(fold, vec3_neg(fold_immvalue_vector(a)));
1027 static GMQCC_INLINE ast_expression *fold_op_not(fold_t *fold, ast_value *a) {
1030 return fold_constgen_float(fold, !fold_immvalue_float(a), false);
1031 } else if (isvector(a)) {
1033 return fold_constgen_float(fold, vec3_notf(fold_immvalue_vector(a)), false);
1034 } else if (isstring(a)) {
1035 if (fold_can_1(a)) {
1036 if (OPTS_FLAG(TRUE_EMPTY_STRINGS))
1037 return fold_constgen_float(fold, !fold_immvalue_string(a), false);
1039 return fold_constgen_float(fold, !fold_immvalue_string(a) || !*fold_immvalue_string(a), false);
1045 static bool fold_check_except_float(sfloat_t (*callback)(sfloat_state_t *, sfloat_t, sfloat_t),
1054 if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS) && !OPTS_WARN(WARN_INEXACT_COMPARES))
1058 ca.f = fold_immvalue_float(a);
1059 cb.f = fold_immvalue_float(b);
1061 callback(&s, ca.s, cb.s);
1062 if (s.exceptionflags == 0)
1065 if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
1066 goto inexact_possible;
1068 sfloat_check(fold_ctx(fold), &s, NULL);
1071 return s.exceptionflags & SFLOAT_INEXACT;
1074 static bool fold_check_inexact_float(fold_t *fold, ast_value *a, ast_value *b) {
1075 lex_ctx_t ctx = fold_ctx(fold);
1076 if (!OPTS_WARN(WARN_INEXACT_COMPARES))
1078 if (!a->inexact && !b->inexact)
1080 return compile_warning(ctx, WARN_INEXACT_COMPARES, "inexact value in comparison");
1083 static GMQCC_INLINE ast_expression *fold_op_add(fold_t *fold, ast_value *a, ast_value *b) {
1085 if (fold_can_2(a, b)) {
1086 bool inexact = fold_check_except_float(&sfloat_add, fold, a, b);
1087 return fold_constgen_float(fold, fold_immvalue_float(a) + fold_immvalue_float(b), inexact);
1089 } else if (isvector(a)) {
1090 if (fold_can_2(a, b))
1091 return fold_constgen_vector(fold, vec3_add(fold_ctx(fold),
1092 fold_immvalue_vector(a),
1093 fold_immvalue_vector(b)));
1098 static GMQCC_INLINE ast_expression *fold_op_sub(fold_t *fold, ast_value *a, ast_value *b) {
1100 if (fold_can_2(a, b)) {
1101 bool inexact = fold_check_except_float(&sfloat_sub, fold, a, b);
1102 return fold_constgen_float(fold, fold_immvalue_float(a) - fold_immvalue_float(b), inexact);
1104 } else if (isvector(a)) {
1105 if (fold_can_2(a, b))
1106 return fold_constgen_vector(fold, vec3_sub(fold_ctx(fold),
1107 fold_immvalue_vector(a),
1108 fold_immvalue_vector(b)));
1113 static GMQCC_INLINE ast_expression *fold_op_mul(fold_t *fold, ast_value *a, ast_value *b) {
1116 if (fold_can_2(a, b))
1117 return fold_constgen_vector(fold, vec3_mulvf(fold_ctx(fold), fold_immvalue_vector(b), fold_immvalue_float(a)));
1119 if (fold_can_2(a, b)) {
1120 bool inexact = fold_check_except_float(&sfloat_mul, fold, a, b);
1121 return fold_constgen_float(fold, fold_immvalue_float(a) * fold_immvalue_float(b), inexact);
1124 } else if (isvector(a)) {
1126 if (fold_can_2(a, b))
1127 return fold_constgen_vector(fold, vec3_mulvf(fold_ctx(fold), fold_immvalue_vector(a), fold_immvalue_float(b)));
1129 if (fold_can_2(a, b)) {
1130 return fold_constgen_float(fold, vec3_mulvv(fold_ctx(fold), fold_immvalue_vector(a), fold_immvalue_vector(b)), false);
1131 } else if (OPTS_OPTIMIZATION(OPTIM_VECTOR_COMPONENTS) && fold_can_1(a)) {
1132 ast_expression *out;
1133 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(a), b, "xyz"))) return out;
1134 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(a), b, "yxz"))) return out;
1135 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(a), b, "zxy"))) return out;
1136 } else if (OPTS_OPTIMIZATION(OPTIM_VECTOR_COMPONENTS) && fold_can_1(b)) {
1137 ast_expression *out;
1138 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(b), a, "xyz"))) return out;
1139 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(b), a, "yxz"))) return out;
1140 if ((out = fold_op_mul_vec(fold, fold_immvalue_vector(b), a, "zxy"))) return out;
1147 static GMQCC_INLINE ast_expression *fold_op_div(fold_t *fold, ast_value *a, ast_value *b) {
1149 if (fold_can_2(a, b)) {
1150 bool inexact = fold_check_except_float(&sfloat_div, fold, a, b);
1151 return fold_constgen_float(fold, fold_immvalue_float(a) / fold_immvalue_float(b), inexact);
1152 } else if (fold_can_1(b)) {
1153 return (ast_expression*)ast_binary_new(
1157 fold_constgen_float(fold, 1.0f / fold_immvalue_float(b), false)
1160 } else if (isvector(a)) {
1161 if (fold_can_2(a, b)) {
1162 return fold_constgen_vector(fold, vec3_mulvf(fold_ctx(fold), fold_immvalue_vector(a), 1.0f / fold_immvalue_float(b)));
1164 return (ast_expression*)ast_binary_new(
1169 ? (ast_expression*)fold_constgen_float(fold, 1.0f / fold_immvalue_float(b), false)
1170 : (ast_expression*)ast_binary_new(
1173 (ast_expression*)fold->imm_float[1],
1182 static GMQCC_INLINE ast_expression *fold_op_mod(fold_t *fold, ast_value *a, ast_value *b) {
1183 return (fold_can_2(a, b))
1184 ? fold_constgen_float(fold, fmod(fold_immvalue_float(a), fold_immvalue_float(b)), false)
1188 static GMQCC_INLINE ast_expression *fold_op_bor(fold_t *fold, ast_value *a, ast_value *b) {
1190 if (fold_can_2(a, b))
1191 return fold_constgen_float(fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) | ((qcint_t)fold_immvalue_float(b))), false);
1194 if (fold_can_2(a, b))
1195 return fold_constgen_vector(fold, vec3_or(fold_immvalue_vector(a), fold_immvalue_vector(b)));
1197 if (fold_can_2(a, b))
1198 return fold_constgen_vector(fold, vec3_orvf(fold_immvalue_vector(a), fold_immvalue_float(b)));
1204 static GMQCC_INLINE ast_expression *fold_op_band(fold_t *fold, ast_value *a, ast_value *b) {
1206 if (fold_can_2(a, b))
1207 return fold_constgen_float(fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) & ((qcint_t)fold_immvalue_float(b))), false);
1210 if (fold_can_2(a, b))
1211 return fold_constgen_vector(fold, vec3_and(fold_immvalue_vector(a), fold_immvalue_vector(b)));
1213 if (fold_can_2(a, b))
1214 return fold_constgen_vector(fold, vec3_andvf(fold_immvalue_vector(a), fold_immvalue_float(b)));
1220 static GMQCC_INLINE ast_expression *fold_op_xor(fold_t *fold, ast_value *a, ast_value *b) {
1222 if (fold_can_2(a, b))
1223 return fold_constgen_float(fold, (qcfloat_t)(((qcint_t)fold_immvalue_float(a)) ^ ((qcint_t)fold_immvalue_float(b))), false);
1225 if (fold_can_2(a, b)) {
1227 return fold_constgen_vector(fold, vec3_xor(fold_immvalue_vector(a), fold_immvalue_vector(b)));
1229 return fold_constgen_vector(fold, vec3_xorvf(fold_immvalue_vector(a), fold_immvalue_float(b)));
1235 static GMQCC_INLINE ast_expression *fold_op_lshift(fold_t *fold, ast_value *a, ast_value *b) {
1236 if (fold_can_2(a, b) && isfloats(a, b))
1237 return fold_constgen_float(fold, (qcfloat_t)floorf(fold_immvalue_float(a) * powf(2.0f, fold_immvalue_float(b))), false);
1241 static GMQCC_INLINE ast_expression *fold_op_rshift(fold_t *fold, ast_value *a, ast_value *b) {
1242 if (fold_can_2(a, b) && isfloats(a, b))
1243 return fold_constgen_float(fold, (qcfloat_t)floorf(fold_immvalue_float(a) / powf(2.0f, fold_immvalue_float(b))), false);
1247 static GMQCC_INLINE ast_expression *fold_op_andor(fold_t *fold, ast_value *a, ast_value *b, float expr) {
1248 if (fold_can_2(a, b)) {
1249 if (OPTS_FLAG(PERL_LOGIC)) {
1251 return (fold_immediate_true(fold, a)) ? (ast_expression*)a : (ast_expression*)b;
1253 return (fold_immediate_true(fold, a)) ? (ast_expression*)b : (ast_expression*)a;
1255 return fold_constgen_float (
1257 ((expr) ? (fold_immediate_true(fold, a) || fold_immediate_true(fold, b))
1258 : (fold_immediate_true(fold, a) && fold_immediate_true(fold, b)))
1268 static GMQCC_INLINE ast_expression *fold_op_tern(fold_t *fold, ast_value *a, ast_value *b, ast_value *c) {
1269 if (fold_can_1(a)) {
1270 return fold_immediate_true(fold, a)
1271 ? (ast_expression*)b
1272 : (ast_expression*)c;
1277 static GMQCC_INLINE ast_expression *fold_op_exp(fold_t *fold, ast_value *a, ast_value *b) {
1278 if (fold_can_2(a, b))
1279 return fold_constgen_float(fold, (qcfloat_t)powf(fold_immvalue_float(a), fold_immvalue_float(b)), false);
1283 static GMQCC_INLINE ast_expression *fold_op_lteqgt(fold_t *fold, ast_value *a, ast_value *b) {
1284 if (fold_can_2(a,b)) {
1285 fold_check_inexact_float(fold, a, b);
1286 if (fold_immvalue_float(a) < fold_immvalue_float(b)) return (ast_expression*)fold->imm_float[2];
1287 if (fold_immvalue_float(a) == fold_immvalue_float(b)) return (ast_expression*)fold->imm_float[0];
1288 if (fold_immvalue_float(a) > fold_immvalue_float(b)) return (ast_expression*)fold->imm_float[1];
1293 static GMQCC_INLINE ast_expression *fold_op_ltgt(fold_t *fold, ast_value *a, ast_value *b, bool lt) {
1294 if (fold_can_2(a, b)) {
1295 fold_check_inexact_float(fold, a, b);
1296 return (lt) ? (ast_expression*)fold->imm_float[!!(fold_immvalue_float(a) < fold_immvalue_float(b))]
1297 : (ast_expression*)fold->imm_float[!!(fold_immvalue_float(a) > fold_immvalue_float(b))];
1302 static GMQCC_INLINE ast_expression *fold_op_cmp(fold_t *fold, ast_value *a, ast_value *b, bool ne) {
1303 if (fold_can_2(a, b)) {
1304 if (isfloat(a) && isfloat(b)) {
1305 float la = fold_immvalue_float(a);
1306 float lb = fold_immvalue_float(b);
1307 fold_check_inexact_float(fold, a, b);
1308 return (ast_expression*)fold->imm_float[!(ne ? la == lb : la != lb)];
1309 } if (isvector(a) && isvector(b)) {
1310 vec3_t la = fold_immvalue_vector(a);
1311 vec3_t lb = fold_immvalue_vector(b);
1312 return (ast_expression*)fold->imm_float[!(ne ? vec3_cmp(la, lb) : !vec3_cmp(la, lb))];
1318 static GMQCC_INLINE ast_expression *fold_op_bnot(fold_t *fold, ast_value *a) {
1321 return fold_constgen_float(fold, -1-fold_immvalue_float(a), false);
1325 return fold_constgen_vector(fold, vec3_not(fold_immvalue_vector(a)));
1331 static GMQCC_INLINE ast_expression *fold_op_cross(fold_t *fold, ast_value *a, ast_value *b) {
1332 if (fold_can_2(a, b))
1333 return fold_constgen_vector(fold, vec3_cross(fold_ctx(fold),
1334 fold_immvalue_vector(a),
1335 fold_immvalue_vector(b)));
1339 ast_expression *fold_op(fold_t *fold, const oper_info *info, ast_expression **opexprs) {
1340 ast_value *a = (ast_value*)opexprs[0];
1341 ast_value *b = (ast_value*)opexprs[1];
1342 ast_value *c = (ast_value*)opexprs[2];
1343 ast_expression *e = NULL;
1345 /* can a fold operation be applied to this operator usage? */
1349 switch(info->operands) {
1350 case 3: if(!c) return NULL;
1351 case 2: if(!b) return NULL;
1354 compile_error(fold_ctx(fold), "internal error: fold_op no operands to fold\n");
1360 * we could use a boolean and default case but ironically gcc produces
1361 * invalid broken assembly from that operation. clang/tcc get it right,
1362 * but interestingly ignore compiling this to a jump-table when I do that,
1363 * this happens to be the most efficent method, since you have per-level
1364 * granularity on the pointer check happening only for the case you check
1365 * it in. Opposed to the default method which would involve a boolean and
1366 * pointer check after wards.
1368 #define fold_op_case(ARGS, ARGS_OPID, OP, ARGS_FOLD) \
1369 case opid##ARGS ARGS_OPID: \
1370 if ((e = fold_op_##OP ARGS_FOLD)) { \
1371 ++opts_optimizationcount[OPTIM_CONST_FOLD]; \
1376 fold_op_case(2, ('-', 'P'), neg, (fold, a));
1377 fold_op_case(2, ('!', 'P'), not, (fold, a));
1378 fold_op_case(1, ('+'), add, (fold, a, b));
1379 fold_op_case(1, ('-'), sub, (fold, a, b));
1380 fold_op_case(1, ('*'), mul, (fold, a, b));
1381 fold_op_case(1, ('/'), div, (fold, a, b));
1382 fold_op_case(1, ('%'), mod, (fold, a, b));
1383 fold_op_case(1, ('|'), bor, (fold, a, b));
1384 fold_op_case(1, ('&'), band, (fold, a, b));
1385 fold_op_case(1, ('^'), xor, (fold, a, b));
1386 fold_op_case(1, ('<'), ltgt, (fold, a, b, true));
1387 fold_op_case(1, ('>'), ltgt, (fold, a, b, false));
1388 fold_op_case(2, ('<', '<'), lshift, (fold, a, b));
1389 fold_op_case(2, ('>', '>'), rshift, (fold, a, b));
1390 fold_op_case(2, ('|', '|'), andor, (fold, a, b, true));
1391 fold_op_case(2, ('&', '&'), andor, (fold, a, b, false));
1392 fold_op_case(2, ('?', ':'), tern, (fold, a, b, c));
1393 fold_op_case(2, ('*', '*'), exp, (fold, a, b));
1394 fold_op_case(3, ('<','=','>'), lteqgt, (fold, a, b));
1395 fold_op_case(2, ('!', '='), cmp, (fold, a, b, true));
1396 fold_op_case(2, ('=', '='), cmp, (fold, a, b, false));
1397 fold_op_case(2, ('~', 'P'), bnot, (fold, a));
1398 fold_op_case(2, ('>', '<'), cross, (fold, a, b));
1401 compile_error(fold_ctx(fold), "internal error: attempted to constant-fold for unsupported operator");
1406 * Constant folding for compiler intrinsics, simaler approach to operator
1407 * folding, primarly: individual functions for each intrinsics to fold,
1408 * and a generic selection function.
1410 static GMQCC_INLINE ast_expression *fold_intrin_isfinite(fold_t *fold, ast_value *a) {
1411 return fold_constgen_float(fold, isfinite(fold_immvalue_float(a)), false);
1413 static GMQCC_INLINE ast_expression *fold_intrin_isinf(fold_t *fold, ast_value *a) {
1414 return fold_constgen_float(fold, isinf(fold_immvalue_float(a)), false);
1416 static GMQCC_INLINE ast_expression *fold_intrin_isnan(fold_t *fold, ast_value *a) {
1417 return fold_constgen_float(fold, isnan(fold_immvalue_float(a)), false);
1419 static GMQCC_INLINE ast_expression *fold_intrin_isnormal(fold_t *fold, ast_value *a) {
1420 return fold_constgen_float(fold, isnormal(fold_immvalue_float(a)), false);
1422 static GMQCC_INLINE ast_expression *fold_intrin_signbit(fold_t *fold, ast_value *a) {
1423 return fold_constgen_float(fold, signbit(fold_immvalue_float(a)), false);
1425 static GMQCC_INLINE ast_expression *fold_intirn_acosh(fold_t *fold, ast_value *a) {
1426 return fold_constgen_float(fold, acoshf(fold_immvalue_float(a)), false);
1428 static GMQCC_INLINE ast_expression *fold_intrin_asinh(fold_t *fold, ast_value *a) {
1429 return fold_constgen_float(fold, asinhf(fold_immvalue_float(a)), false);
1431 static GMQCC_INLINE ast_expression *fold_intrin_atanh(fold_t *fold, ast_value *a) {
1432 return fold_constgen_float(fold, (float)atanh(fold_immvalue_float(a)), false);
1434 static GMQCC_INLINE ast_expression *fold_intrin_exp(fold_t *fold, ast_value *a) {
1435 return fold_constgen_float(fold, expf(fold_immvalue_float(a)), false);
1437 static GMQCC_INLINE ast_expression *fold_intrin_exp2(fold_t *fold, ast_value *a) {
1438 return fold_constgen_float(fold, exp2f(fold_immvalue_float(a)), false);
1440 static GMQCC_INLINE ast_expression *fold_intrin_expm1(fold_t *fold, ast_value *a) {
1441 return fold_constgen_float(fold, expm1f(fold_immvalue_float(a)), false);
1443 static GMQCC_INLINE ast_expression *fold_intrin_mod(fold_t *fold, ast_value *lhs, ast_value *rhs) {
1444 return fold_constgen_float(fold, fmodf(fold_immvalue_float(lhs), fold_immvalue_float(rhs)), false);
1446 static GMQCC_INLINE ast_expression *fold_intrin_pow(fold_t *fold, ast_value *lhs, ast_value *rhs) {
1447 return fold_constgen_float(fold, powf(fold_immvalue_float(lhs), fold_immvalue_float(rhs)), false);
1449 static GMQCC_INLINE ast_expression *fold_intrin_fabs(fold_t *fold, ast_value *a) {
1450 return fold_constgen_float(fold, fabsf(fold_immvalue_float(a)), false);
1454 ast_expression *fold_intrin(fold_t *fold, const char *intrin, ast_expression **arg) {
1455 ast_expression *ret = NULL;
1456 ast_value *a = (ast_value*)arg[0];
1457 ast_value *b = (ast_value*)arg[1];
1459 if (!strcmp(intrin, "isfinite")) ret = fold_intrin_isfinite(fold, a);
1460 if (!strcmp(intrin, "isinf")) ret = fold_intrin_isinf(fold, a);
1461 if (!strcmp(intrin, "isnan")) ret = fold_intrin_isnan(fold, a);
1462 if (!strcmp(intrin, "isnormal")) ret = fold_intrin_isnormal(fold, a);
1463 if (!strcmp(intrin, "signbit")) ret = fold_intrin_signbit(fold, a);
1464 if (!strcmp(intrin, "acosh")) ret = fold_intirn_acosh(fold, a);
1465 if (!strcmp(intrin, "asinh")) ret = fold_intrin_asinh(fold, a);
1466 if (!strcmp(intrin, "atanh")) ret = fold_intrin_atanh(fold, a);
1467 if (!strcmp(intrin, "exp")) ret = fold_intrin_exp(fold, a);
1468 if (!strcmp(intrin, "exp2")) ret = fold_intrin_exp2(fold, a);
1469 if (!strcmp(intrin, "expm1")) ret = fold_intrin_expm1(fold, a);
1470 if (!strcmp(intrin, "mod")) ret = fold_intrin_mod(fold, a, b);
1471 if (!strcmp(intrin, "pow")) ret = fold_intrin_pow(fold, a, b);
1472 if (!strcmp(intrin, "fabs")) ret = fold_intrin_fabs(fold, a);
1475 ++opts_optimizationcount[OPTIM_CONST_FOLD];
1481 * These are all the actual constant folding methods that happen in between
1482 * the AST/IR stage of the compiler , i.e eliminating branches for const
1483 * expressions, which is the only supported thing so far. We undefine the
1484 * testing macros here because an ir_value is differant than an ast_value.
1490 #undef fold_immvalue_float
1491 #undef fold_immvalue_string
1492 #undef fold_immvalue_vector
1496 #define isfloat(X) ((X)->vtype == TYPE_FLOAT)
1497 /*#define isstring(X) ((X)->vtype == TYPE_STRING)*/
1498 /*#define isvector(X) ((X)->vtype == TYPE_VECTOR)*/
1499 #define fold_immvalue_float(X) ((X)->constval.vfloat)
1500 #define fold_immvalue_vector(X) ((X)->constval.vvec)
1501 /*#define fold_immvalue_string(X) ((X)->constval.vstring)*/
1502 #define fold_can_1(X) ((X)->hasvalue && (X)->cvq == CV_CONST)
1503 /*#define fold_can_2(X,Y) (fold_can_1(X) && fold_can_1(Y))*/
1505 static ast_expression *fold_superfluous(ast_expression *left, ast_expression *right, int op) {
1506 ast_expression *swapped = NULL; /* using this as bool */
1509 if (!ast_istype(right, ast_value) || !fold_can_1((load = (ast_value*)right))) {
1515 if (!ast_istype(right, ast_value) || !fold_can_1((load = (ast_value*)right)))
1523 if (fold_immvalue_float(load) == 1.0f) {
1524 ++opts_optimizationcount[OPTIM_PEEPHOLE];
1535 if (fold_immvalue_float(load) == 0.0f) {
1536 ++opts_optimizationcount[OPTIM_PEEPHOLE];
1543 if (vec3_cmp(fold_immvalue_vector(load), vec3_create(1, 1, 1))) {
1544 ++opts_optimizationcount[OPTIM_PEEPHOLE];
1554 if (vec3_cmp(fold_immvalue_vector(load), vec3_create(0, 0, 0))) {
1555 ++opts_optimizationcount[OPTIM_PEEPHOLE];
1565 ast_expression *fold_binary(lex_ctx_t ctx, int op, ast_expression *left, ast_expression *right) {
1566 ast_expression *ret = fold_superfluous(left, right, op);
1569 return (ast_expression*)ast_binary_new(ctx, op, left, right);
1572 static GMQCC_INLINE int fold_cond(ir_value *condval, ast_function *func, ast_ifthen *branch) {
1573 if (isfloat(condval) && fold_can_1(condval) && OPTS_OPTIMIZATION(OPTIM_CONST_FOLD_DCE)) {
1574 ast_expression_codegen *cgen;
1577 bool istrue = (fold_immvalue_float(condval) != 0.0f && branch->on_true);
1578 bool isfalse = (fold_immvalue_float(condval) == 0.0f && branch->on_false);
1579 ast_expression *path = (istrue) ? branch->on_true :
1580 (isfalse) ? branch->on_false : NULL;
1583 * no path to take implies that the evaluation is if(0) and there
1584 * is no else block. so eliminate all the code.
1586 ++opts_optimizationcount[OPTIM_CONST_FOLD_DCE];
1590 if (!(elide = ir_function_create_block(ast_ctx(branch), func->ir_func, ast_function_label(func, ((istrue) ? "ontrue" : "onfalse")))))
1592 if (!(*(cgen = path->codegen))((ast_expression*)path, func, false, &dummy))
1594 if (!ir_block_create_jump(func->curblock, ast_ctx(branch), elide))
1597 * now the branch has been eliminated and the correct block for the constant evaluation
1598 * is expanded into the current block for the function.
1600 func->curblock = elide;
1601 ++opts_optimizationcount[OPTIM_CONST_FOLD_DCE];
1604 return -1; /* nothing done */
1607 int fold_cond_ternary(ir_value *condval, ast_function *func, ast_ternary *branch) {
1608 return fold_cond(condval, func, (ast_ifthen*)branch);
1611 int fold_cond_ifthen(ir_value *condval, ast_function *func, ast_ifthen *branch) {
1612 return fold_cond(condval, func, branch);