tgsi_exec.c revision 34a78b7ef6b0edf217acf221eab4b63542be5552
1/************************************************************************** 2 * 3 * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas. 4 * All Rights Reserved. 5 * Copyright 2009-2010 VMware, Inc. All rights Reserved. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the 9 * "Software"), to deal in the Software without restriction, including 10 * without limitation the rights to use, copy, modify, merge, publish, 11 * distribute, sub license, and/or sell copies of the Software, and to 12 * permit persons to whom the Software is furnished to do so, subject to 13 * the following conditions: 14 * 15 * The above copyright notice and this permission notice (including the 16 * next paragraph) shall be included in all copies or substantial portions 17 * of the Software. 18 * 19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 26 * 27 **************************************************************************/ 28 29/** 30 * TGSI interpreter/executor. 31 * 32 * Flow control information: 33 * 34 * Since we operate on 'quads' (4 pixels or 4 vertices in parallel) 35 * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special 36 * care since a condition may be true for some quad components but false 37 * for other components. 38 * 39 * We basically execute all statements (even if they're in the part of 40 * an IF/ELSE clause that's "not taken") and use a special mask to 41 * control writing to destination registers. This is the ExecMask. 42 * See store_dest(). 43 * 44 * The ExecMask is computed from three other masks (CondMask, LoopMask and 45 * ContMask) which are controlled by the flow control instructions (namely: 46 * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT). 47 * 48 * 49 * Authors: 50 * Michal Krol 51 * Brian Paul 52 */ 53 54#include "pipe/p_compiler.h" 55#include "pipe/p_state.h" 56#include "pipe/p_shader_tokens.h" 57#include "tgsi/tgsi_dump.h" 58#include "tgsi/tgsi_parse.h" 59#include "tgsi/tgsi_util.h" 60#include "tgsi_exec.h" 61#include "util/u_memory.h" 62#include "util/u_math.h" 63 64 65#define FAST_MATH 0 66 67#define TILE_TOP_LEFT 0 68#define TILE_TOP_RIGHT 1 69#define TILE_BOTTOM_LEFT 2 70#define TILE_BOTTOM_RIGHT 3 71 72static void 73micro_abs(union tgsi_exec_channel *dst, 74 const union tgsi_exec_channel *src) 75{ 76 dst->f[0] = fabsf(src->f[0]); 77 dst->f[1] = fabsf(src->f[1]); 78 dst->f[2] = fabsf(src->f[2]); 79 dst->f[3] = fabsf(src->f[3]); 80} 81 82static void 83micro_arl(union tgsi_exec_channel *dst, 84 const union tgsi_exec_channel *src) 85{ 86 dst->i[0] = (int)floorf(src->f[0]); 87 dst->i[1] = (int)floorf(src->f[1]); 88 dst->i[2] = (int)floorf(src->f[2]); 89 dst->i[3] = (int)floorf(src->f[3]); 90} 91 92static void 93micro_arr(union tgsi_exec_channel *dst, 94 const union tgsi_exec_channel *src) 95{ 96 dst->i[0] = (int)floorf(src->f[0] + 0.5f); 97 dst->i[1] = (int)floorf(src->f[1] + 0.5f); 98 dst->i[2] = (int)floorf(src->f[2] + 0.5f); 99 dst->i[3] = (int)floorf(src->f[3] + 0.5f); 100} 101 102static void 103micro_ceil(union tgsi_exec_channel *dst, 104 const union tgsi_exec_channel *src) 105{ 106 dst->f[0] = ceilf(src->f[0]); 107 dst->f[1] = ceilf(src->f[1]); 108 dst->f[2] = ceilf(src->f[2]); 109 dst->f[3] = ceilf(src->f[3]); 110} 111 112static void 113micro_clamp(union tgsi_exec_channel *dst, 114 const union tgsi_exec_channel *src0, 115 const union tgsi_exec_channel *src1, 116 const union tgsi_exec_channel *src2) 117{ 118 dst->f[0] = src0->f[0] < src1->f[0] ? src1->f[0] : src0->f[0] > src2->f[0] ? src2->f[0] : src0->f[0]; 119 dst->f[1] = src0->f[1] < src1->f[1] ? src1->f[1] : src0->f[1] > src2->f[1] ? src2->f[1] : src0->f[1]; 120 dst->f[2] = src0->f[2] < src1->f[2] ? src1->f[2] : src0->f[2] > src2->f[2] ? src2->f[2] : src0->f[2]; 121 dst->f[3] = src0->f[3] < src1->f[3] ? src1->f[3] : src0->f[3] > src2->f[3] ? src2->f[3] : src0->f[3]; 122} 123 124static void 125micro_cmp(union tgsi_exec_channel *dst, 126 const union tgsi_exec_channel *src0, 127 const union tgsi_exec_channel *src1, 128 const union tgsi_exec_channel *src2) 129{ 130 dst->f[0] = src0->f[0] < 0.0f ? src1->f[0] : src2->f[0]; 131 dst->f[1] = src0->f[1] < 0.0f ? src1->f[1] : src2->f[1]; 132 dst->f[2] = src0->f[2] < 0.0f ? src1->f[2] : src2->f[2]; 133 dst->f[3] = src0->f[3] < 0.0f ? src1->f[3] : src2->f[3]; 134} 135 136static void 137micro_cnd(union tgsi_exec_channel *dst, 138 const union tgsi_exec_channel *src0, 139 const union tgsi_exec_channel *src1, 140 const union tgsi_exec_channel *src2) 141{ 142 dst->f[0] = src2->f[0] > 0.5f ? src0->f[0] : src1->f[0]; 143 dst->f[1] = src2->f[1] > 0.5f ? src0->f[1] : src1->f[1]; 144 dst->f[2] = src2->f[2] > 0.5f ? src0->f[2] : src1->f[2]; 145 dst->f[3] = src2->f[3] > 0.5f ? src0->f[3] : src1->f[3]; 146} 147 148static void 149micro_cos(union tgsi_exec_channel *dst, 150 const union tgsi_exec_channel *src) 151{ 152 dst->f[0] = cosf(src->f[0]); 153 dst->f[1] = cosf(src->f[1]); 154 dst->f[2] = cosf(src->f[2]); 155 dst->f[3] = cosf(src->f[3]); 156} 157 158static void 159micro_ddx(union tgsi_exec_channel *dst, 160 const union tgsi_exec_channel *src) 161{ 162 dst->f[0] = 163 dst->f[1] = 164 dst->f[2] = 165 dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_BOTTOM_LEFT]; 166} 167 168static void 169micro_ddy(union tgsi_exec_channel *dst, 170 const union tgsi_exec_channel *src) 171{ 172 dst->f[0] = 173 dst->f[1] = 174 dst->f[2] = 175 dst->f[3] = src->f[TILE_BOTTOM_LEFT] - src->f[TILE_TOP_LEFT]; 176} 177 178static void 179micro_exp2(union tgsi_exec_channel *dst, 180 const union tgsi_exec_channel *src) 181{ 182#if FAST_MATH 183 dst->f[0] = util_fast_exp2(src->f[0]); 184 dst->f[1] = util_fast_exp2(src->f[1]); 185 dst->f[2] = util_fast_exp2(src->f[2]); 186 dst->f[3] = util_fast_exp2(src->f[3]); 187#else 188#if DEBUG 189 /* Inf is okay for this instruction, so clamp it to silence assertions. */ 190 uint i; 191 union tgsi_exec_channel clamped; 192 193 for (i = 0; i < 4; i++) { 194 if (src->f[i] > 127.99999f) { 195 clamped.f[i] = 127.99999f; 196 } else if (src->f[i] < -126.99999f) { 197 clamped.f[i] = -126.99999f; 198 } else { 199 clamped.f[i] = src->f[i]; 200 } 201 } 202 src = &clamped; 203#endif /* DEBUG */ 204 205 dst->f[0] = powf(2.0f, src->f[0]); 206 dst->f[1] = powf(2.0f, src->f[1]); 207 dst->f[2] = powf(2.0f, src->f[2]); 208 dst->f[3] = powf(2.0f, src->f[3]); 209#endif /* FAST_MATH */ 210} 211 212static void 213micro_flr(union tgsi_exec_channel *dst, 214 const union tgsi_exec_channel *src) 215{ 216 dst->f[0] = floorf(src->f[0]); 217 dst->f[1] = floorf(src->f[1]); 218 dst->f[2] = floorf(src->f[2]); 219 dst->f[3] = floorf(src->f[3]); 220} 221 222static void 223micro_frc(union tgsi_exec_channel *dst, 224 const union tgsi_exec_channel *src) 225{ 226 dst->f[0] = src->f[0] - floorf(src->f[0]); 227 dst->f[1] = src->f[1] - floorf(src->f[1]); 228 dst->f[2] = src->f[2] - floorf(src->f[2]); 229 dst->f[3] = src->f[3] - floorf(src->f[3]); 230} 231 232static void 233micro_iabs(union tgsi_exec_channel *dst, 234 const union tgsi_exec_channel *src) 235{ 236 dst->i[0] = src->i[0] >= 0 ? src->i[0] : -src->i[0]; 237 dst->i[1] = src->i[1] >= 0 ? src->i[1] : -src->i[1]; 238 dst->i[2] = src->i[2] >= 0 ? src->i[2] : -src->i[2]; 239 dst->i[3] = src->i[3] >= 0 ? src->i[3] : -src->i[3]; 240} 241 242static void 243micro_ineg(union tgsi_exec_channel *dst, 244 const union tgsi_exec_channel *src) 245{ 246 dst->i[0] = -src->i[0]; 247 dst->i[1] = -src->i[1]; 248 dst->i[2] = -src->i[2]; 249 dst->i[3] = -src->i[3]; 250} 251 252static void 253micro_lg2(union tgsi_exec_channel *dst, 254 const union tgsi_exec_channel *src) 255{ 256#if FAST_MATH 257 dst->f[0] = util_fast_log2(src->f[0]); 258 dst->f[1] = util_fast_log2(src->f[1]); 259 dst->f[2] = util_fast_log2(src->f[2]); 260 dst->f[3] = util_fast_log2(src->f[3]); 261#else 262 dst->f[0] = logf(src->f[0]) * 1.442695f; 263 dst->f[1] = logf(src->f[1]) * 1.442695f; 264 dst->f[2] = logf(src->f[2]) * 1.442695f; 265 dst->f[3] = logf(src->f[3]) * 1.442695f; 266#endif 267} 268 269static void 270micro_lrp(union tgsi_exec_channel *dst, 271 const union tgsi_exec_channel *src0, 272 const union tgsi_exec_channel *src1, 273 const union tgsi_exec_channel *src2) 274{ 275 dst->f[0] = src0->f[0] * (src1->f[0] - src2->f[0]) + src2->f[0]; 276 dst->f[1] = src0->f[1] * (src1->f[1] - src2->f[1]) + src2->f[1]; 277 dst->f[2] = src0->f[2] * (src1->f[2] - src2->f[2]) + src2->f[2]; 278 dst->f[3] = src0->f[3] * (src1->f[3] - src2->f[3]) + src2->f[3]; 279} 280 281static void 282micro_mad(union tgsi_exec_channel *dst, 283 const union tgsi_exec_channel *src0, 284 const union tgsi_exec_channel *src1, 285 const union tgsi_exec_channel *src2) 286{ 287 dst->f[0] = src0->f[0] * src1->f[0] + src2->f[0]; 288 dst->f[1] = src0->f[1] * src1->f[1] + src2->f[1]; 289 dst->f[2] = src0->f[2] * src1->f[2] + src2->f[2]; 290 dst->f[3] = src0->f[3] * src1->f[3] + src2->f[3]; 291} 292 293static void 294micro_mov(union tgsi_exec_channel *dst, 295 const union tgsi_exec_channel *src) 296{ 297 dst->u[0] = src->u[0]; 298 dst->u[1] = src->u[1]; 299 dst->u[2] = src->u[2]; 300 dst->u[3] = src->u[3]; 301} 302 303static void 304micro_rcp(union tgsi_exec_channel *dst, 305 const union tgsi_exec_channel *src) 306{ 307#if 0 /* for debugging */ 308 assert(src->f[0] != 0.0f); 309 assert(src->f[1] != 0.0f); 310 assert(src->f[2] != 0.0f); 311 assert(src->f[3] != 0.0f); 312#endif 313 dst->f[0] = 1.0f / src->f[0]; 314 dst->f[1] = 1.0f / src->f[1]; 315 dst->f[2] = 1.0f / src->f[2]; 316 dst->f[3] = 1.0f / src->f[3]; 317} 318 319static void 320micro_rnd(union tgsi_exec_channel *dst, 321 const union tgsi_exec_channel *src) 322{ 323 dst->f[0] = floorf(src->f[0] + 0.5f); 324 dst->f[1] = floorf(src->f[1] + 0.5f); 325 dst->f[2] = floorf(src->f[2] + 0.5f); 326 dst->f[3] = floorf(src->f[3] + 0.5f); 327} 328 329static void 330micro_rsq(union tgsi_exec_channel *dst, 331 const union tgsi_exec_channel *src) 332{ 333#if 0 /* for debugging */ 334 assert(src->f[0] != 0.0f); 335 assert(src->f[1] != 0.0f); 336 assert(src->f[2] != 0.0f); 337 assert(src->f[3] != 0.0f); 338#endif 339 dst->f[0] = 1.0f / sqrtf(fabsf(src->f[0])); 340 dst->f[1] = 1.0f / sqrtf(fabsf(src->f[1])); 341 dst->f[2] = 1.0f / sqrtf(fabsf(src->f[2])); 342 dst->f[3] = 1.0f / sqrtf(fabsf(src->f[3])); 343} 344 345static void 346micro_seq(union tgsi_exec_channel *dst, 347 const union tgsi_exec_channel *src0, 348 const union tgsi_exec_channel *src1) 349{ 350 dst->f[0] = src0->f[0] == src1->f[0] ? 1.0f : 0.0f; 351 dst->f[1] = src0->f[1] == src1->f[1] ? 1.0f : 0.0f; 352 dst->f[2] = src0->f[2] == src1->f[2] ? 1.0f : 0.0f; 353 dst->f[3] = src0->f[3] == src1->f[3] ? 1.0f : 0.0f; 354} 355 356static void 357micro_sge(union tgsi_exec_channel *dst, 358 const union tgsi_exec_channel *src0, 359 const union tgsi_exec_channel *src1) 360{ 361 dst->f[0] = src0->f[0] >= src1->f[0] ? 1.0f : 0.0f; 362 dst->f[1] = src0->f[1] >= src1->f[1] ? 1.0f : 0.0f; 363 dst->f[2] = src0->f[2] >= src1->f[2] ? 1.0f : 0.0f; 364 dst->f[3] = src0->f[3] >= src1->f[3] ? 1.0f : 0.0f; 365} 366 367static void 368micro_sgn(union tgsi_exec_channel *dst, 369 const union tgsi_exec_channel *src) 370{ 371 dst->f[0] = src->f[0] < 0.0f ? -1.0f : src->f[0] > 0.0f ? 1.0f : 0.0f; 372 dst->f[1] = src->f[1] < 0.0f ? -1.0f : src->f[1] > 0.0f ? 1.0f : 0.0f; 373 dst->f[2] = src->f[2] < 0.0f ? -1.0f : src->f[2] > 0.0f ? 1.0f : 0.0f; 374 dst->f[3] = src->f[3] < 0.0f ? -1.0f : src->f[3] > 0.0f ? 1.0f : 0.0f; 375} 376 377static void 378micro_sgt(union tgsi_exec_channel *dst, 379 const union tgsi_exec_channel *src0, 380 const union tgsi_exec_channel *src1) 381{ 382 dst->f[0] = src0->f[0] > src1->f[0] ? 1.0f : 0.0f; 383 dst->f[1] = src0->f[1] > src1->f[1] ? 1.0f : 0.0f; 384 dst->f[2] = src0->f[2] > src1->f[2] ? 1.0f : 0.0f; 385 dst->f[3] = src0->f[3] > src1->f[3] ? 1.0f : 0.0f; 386} 387 388static void 389micro_sin(union tgsi_exec_channel *dst, 390 const union tgsi_exec_channel *src) 391{ 392 dst->f[0] = sinf(src->f[0]); 393 dst->f[1] = sinf(src->f[1]); 394 dst->f[2] = sinf(src->f[2]); 395 dst->f[3] = sinf(src->f[3]); 396} 397 398static void 399micro_sle(union tgsi_exec_channel *dst, 400 const union tgsi_exec_channel *src0, 401 const union tgsi_exec_channel *src1) 402{ 403 dst->f[0] = src0->f[0] <= src1->f[0] ? 1.0f : 0.0f; 404 dst->f[1] = src0->f[1] <= src1->f[1] ? 1.0f : 0.0f; 405 dst->f[2] = src0->f[2] <= src1->f[2] ? 1.0f : 0.0f; 406 dst->f[3] = src0->f[3] <= src1->f[3] ? 1.0f : 0.0f; 407} 408 409static void 410micro_slt(union tgsi_exec_channel *dst, 411 const union tgsi_exec_channel *src0, 412 const union tgsi_exec_channel *src1) 413{ 414 dst->f[0] = src0->f[0] < src1->f[0] ? 1.0f : 0.0f; 415 dst->f[1] = src0->f[1] < src1->f[1] ? 1.0f : 0.0f; 416 dst->f[2] = src0->f[2] < src1->f[2] ? 1.0f : 0.0f; 417 dst->f[3] = src0->f[3] < src1->f[3] ? 1.0f : 0.0f; 418} 419 420static void 421micro_sne(union tgsi_exec_channel *dst, 422 const union tgsi_exec_channel *src0, 423 const union tgsi_exec_channel *src1) 424{ 425 dst->f[0] = src0->f[0] != src1->f[0] ? 1.0f : 0.0f; 426 dst->f[1] = src0->f[1] != src1->f[1] ? 1.0f : 0.0f; 427 dst->f[2] = src0->f[2] != src1->f[2] ? 1.0f : 0.0f; 428 dst->f[3] = src0->f[3] != src1->f[3] ? 1.0f : 0.0f; 429} 430 431static void 432micro_sfl(union tgsi_exec_channel *dst) 433{ 434 dst->f[0] = 0.0f; 435 dst->f[1] = 0.0f; 436 dst->f[2] = 0.0f; 437 dst->f[3] = 0.0f; 438} 439 440static void 441micro_str(union tgsi_exec_channel *dst) 442{ 443 dst->f[0] = 1.0f; 444 dst->f[1] = 1.0f; 445 dst->f[2] = 1.0f; 446 dst->f[3] = 1.0f; 447} 448 449static void 450micro_trunc(union tgsi_exec_channel *dst, 451 const union tgsi_exec_channel *src) 452{ 453 dst->f[0] = (float)(int)src->f[0]; 454 dst->f[1] = (float)(int)src->f[1]; 455 dst->f[2] = (float)(int)src->f[2]; 456 dst->f[3] = (float)(int)src->f[3]; 457} 458 459 460#define CHAN_X 0 461#define CHAN_Y 1 462#define CHAN_Z 2 463#define CHAN_W 3 464 465enum tgsi_exec_datatype { 466 TGSI_EXEC_DATA_FLOAT, 467 TGSI_EXEC_DATA_INT, 468 TGSI_EXEC_DATA_UINT 469}; 470 471/* 472 * Shorthand locations of various utility registers (_I = Index, _C = Channel) 473 */ 474#define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I 475#define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C 476#define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I 477#define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C 478#define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I 479#define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C 480 481 482/** The execution mask depends on the conditional mask and the loop mask */ 483#define UPDATE_EXEC_MASK(MACH) \ 484 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask 485 486 487static const union tgsi_exec_channel ZeroVec = 488 { { 0.0, 0.0, 0.0, 0.0 } }; 489 490static const union tgsi_exec_channel OneVec = { 491 {1.0f, 1.0f, 1.0f, 1.0f} 492}; 493 494static const union tgsi_exec_channel P128Vec = { 495 {128.0f, 128.0f, 128.0f, 128.0f} 496}; 497 498static const union tgsi_exec_channel M128Vec = { 499 {-128.0f, -128.0f, -128.0f, -128.0f} 500}; 501 502 503/** 504 * Assert that none of the float values in 'chan' are infinite or NaN. 505 * NaN and Inf may occur normally during program execution and should 506 * not lead to crashes, etc. But when debugging, it's helpful to catch 507 * them. 508 */ 509static INLINE void 510check_inf_or_nan(const union tgsi_exec_channel *chan) 511{ 512 assert(!util_is_inf_or_nan((chan)->f[0])); 513 assert(!util_is_inf_or_nan((chan)->f[1])); 514 assert(!util_is_inf_or_nan((chan)->f[2])); 515 assert(!util_is_inf_or_nan((chan)->f[3])); 516} 517 518 519#ifdef DEBUG 520static void 521print_chan(const char *msg, const union tgsi_exec_channel *chan) 522{ 523 debug_printf("%s = {%f, %f, %f, %f}\n", 524 msg, chan->f[0], chan->f[1], chan->f[2], chan->f[3]); 525} 526#endif 527 528 529#ifdef DEBUG 530static void 531print_temp(const struct tgsi_exec_machine *mach, uint index) 532{ 533 const struct tgsi_exec_vector *tmp = &mach->Temps[index]; 534 int i; 535 debug_printf("Temp[%u] =\n", index); 536 for (i = 0; i < 4; i++) { 537 debug_printf(" %c: { %f, %f, %f, %f }\n", 538 "XYZW"[i], 539 tmp->xyzw[i].f[0], 540 tmp->xyzw[i].f[1], 541 tmp->xyzw[i].f[2], 542 tmp->xyzw[i].f[3]); 543 } 544} 545#endif 546 547 548void 549tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach, 550 unsigned num_bufs, 551 const void **bufs, 552 const unsigned *buf_sizes) 553{ 554 unsigned i; 555 556 for (i = 0; i < num_bufs; i++) { 557 mach->Consts[i] = bufs[i]; 558 mach->ConstsSize[i] = buf_sizes[i]; 559 } 560} 561 562 563/** 564 * Check if there's a potential src/dst register data dependency when 565 * using SOA execution. 566 * Example: 567 * MOV T, T.yxwz; 568 * This would expand into: 569 * MOV t0, t1; 570 * MOV t1, t0; 571 * MOV t2, t3; 572 * MOV t3, t2; 573 * The second instruction will have the wrong value for t0 if executed as-is. 574 */ 575boolean 576tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst) 577{ 578 uint i, chan; 579 580 uint writemask = inst->Dst[0].Register.WriteMask; 581 if (writemask == TGSI_WRITEMASK_X || 582 writemask == TGSI_WRITEMASK_Y || 583 writemask == TGSI_WRITEMASK_Z || 584 writemask == TGSI_WRITEMASK_W || 585 writemask == TGSI_WRITEMASK_NONE) { 586 /* no chance of data dependency */ 587 return FALSE; 588 } 589 590 /* loop over src regs */ 591 for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { 592 if ((inst->Src[i].Register.File == 593 inst->Dst[0].Register.File) && 594 ((inst->Src[i].Register.Index == 595 inst->Dst[0].Register.Index) || 596 inst->Src[i].Register.Indirect || 597 inst->Dst[0].Register.Indirect)) { 598 /* loop over dest channels */ 599 uint channelsWritten = 0x0; 600 for (chan = 0; chan < NUM_CHANNELS; chan++) { 601 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 602 /* check if we're reading a channel that's been written */ 603 uint swizzle = tgsi_util_get_full_src_register_swizzle(&inst->Src[i], chan); 604 if (channelsWritten & (1 << swizzle)) { 605 return TRUE; 606 } 607 608 channelsWritten |= (1 << chan); 609 } 610 } 611 } 612 } 613 return FALSE; 614} 615 616 617/** 618 * Initialize machine state by expanding tokens to full instructions, 619 * allocating temporary storage, setting up constants, etc. 620 * After this, we can call tgsi_exec_machine_run() many times. 621 */ 622void 623tgsi_exec_machine_bind_shader( 624 struct tgsi_exec_machine *mach, 625 const struct tgsi_token *tokens, 626 uint numSamplers, 627 struct tgsi_sampler **samplers) 628{ 629 uint k; 630 struct tgsi_parse_context parse; 631 struct tgsi_full_instruction *instructions; 632 struct tgsi_full_declaration *declarations; 633 uint maxInstructions = 10, numInstructions = 0; 634 uint maxDeclarations = 10, numDeclarations = 0; 635 636#if 0 637 tgsi_dump(tokens, 0); 638#endif 639 640 util_init_math(); 641 642 if (numSamplers) { 643 assert(samplers); 644 } 645 646 mach->Tokens = tokens; 647 mach->Samplers = samplers; 648 649 if (!tokens) { 650 /* unbind and free all */ 651 if (mach->Declarations) { 652 FREE( mach->Declarations ); 653 } 654 mach->Declarations = NULL; 655 mach->NumDeclarations = 0; 656 657 if (mach->Instructions) { 658 FREE( mach->Instructions ); 659 } 660 mach->Instructions = NULL; 661 mach->NumInstructions = 0; 662 663 return; 664 } 665 666 k = tgsi_parse_init (&parse, mach->Tokens); 667 if (k != TGSI_PARSE_OK) { 668 debug_printf( "Problem parsing!\n" ); 669 return; 670 } 671 672 mach->Processor = parse.FullHeader.Processor.Processor; 673 mach->ImmLimit = 0; 674 675 if (mach->Processor == TGSI_PROCESSOR_GEOMETRY && 676 !mach->UsedGeometryShader) { 677 struct tgsi_exec_vector *inputs; 678 struct tgsi_exec_vector *outputs; 679 680 inputs = align_malloc(sizeof(struct tgsi_exec_vector) * 681 TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS, 682 16); 683 684 if (!inputs) 685 return; 686 687 outputs = align_malloc(sizeof(struct tgsi_exec_vector) * 688 TGSI_MAX_TOTAL_VERTICES, 16); 689 690 if (!outputs) { 691 align_free(inputs); 692 return; 693 } 694 695 align_free(mach->Inputs); 696 align_free(mach->Outputs); 697 698 mach->Inputs = inputs; 699 mach->Outputs = outputs; 700 mach->UsedGeometryShader = TRUE; 701 } 702 703 declarations = (struct tgsi_full_declaration *) 704 MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) ); 705 706 if (!declarations) { 707 return; 708 } 709 710 instructions = (struct tgsi_full_instruction *) 711 MALLOC( maxInstructions * sizeof(struct tgsi_full_instruction) ); 712 713 if (!instructions) { 714 FREE( declarations ); 715 return; 716 } 717 718 while( !tgsi_parse_end_of_tokens( &parse ) ) { 719 uint i; 720 721 tgsi_parse_token( &parse ); 722 switch( parse.FullToken.Token.Type ) { 723 case TGSI_TOKEN_TYPE_DECLARATION: 724 /* save expanded declaration */ 725 if (numDeclarations == maxDeclarations) { 726 declarations = REALLOC(declarations, 727 maxDeclarations 728 * sizeof(struct tgsi_full_declaration), 729 (maxDeclarations + 10) 730 * sizeof(struct tgsi_full_declaration)); 731 maxDeclarations += 10; 732 } 733 if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_OUTPUT) { 734 unsigned reg; 735 for (reg = parse.FullToken.FullDeclaration.Range.First; 736 reg <= parse.FullToken.FullDeclaration.Range.Last; 737 ++reg) { 738 ++mach->NumOutputs; 739 } 740 } 741 if (parse.FullToken.FullDeclaration.Declaration.File == 742 TGSI_FILE_IMMEDIATE_ARRAY) { 743 unsigned reg; 744 struct tgsi_full_declaration *decl = 745 &parse.FullToken.FullDeclaration; 746 debug_assert(decl->Range.Last < TGSI_EXEC_NUM_IMMEDIATES); 747 for (reg = decl->Range.First; reg <= decl->Range.Last; ++reg) { 748 for( i = 0; i < 4; i++ ) { 749 int idx = reg * 4 + i; 750 mach->ImmArray[reg][i] = decl->ImmediateData.u[idx].Float; 751 } 752 } 753 } 754 memcpy(declarations + numDeclarations, 755 &parse.FullToken.FullDeclaration, 756 sizeof(declarations[0])); 757 numDeclarations++; 758 break; 759 760 case TGSI_TOKEN_TYPE_IMMEDIATE: 761 { 762 uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1; 763 assert( size <= 4 ); 764 assert( mach->ImmLimit + 1 <= TGSI_EXEC_NUM_IMMEDIATES ); 765 766 for( i = 0; i < size; i++ ) { 767 mach->Imms[mach->ImmLimit][i] = 768 parse.FullToken.FullImmediate.u[i].Float; 769 } 770 mach->ImmLimit += 1; 771 } 772 break; 773 774 case TGSI_TOKEN_TYPE_INSTRUCTION: 775 776 /* save expanded instruction */ 777 if (numInstructions == maxInstructions) { 778 instructions = REALLOC(instructions, 779 maxInstructions 780 * sizeof(struct tgsi_full_instruction), 781 (maxInstructions + 10) 782 * sizeof(struct tgsi_full_instruction)); 783 maxInstructions += 10; 784 } 785 786 memcpy(instructions + numInstructions, 787 &parse.FullToken.FullInstruction, 788 sizeof(instructions[0])); 789 790 numInstructions++; 791 break; 792 793 case TGSI_TOKEN_TYPE_PROPERTY: 794 break; 795 796 default: 797 assert( 0 ); 798 } 799 } 800 tgsi_parse_free (&parse); 801 802 if (mach->Declarations) { 803 FREE( mach->Declarations ); 804 } 805 mach->Declarations = declarations; 806 mach->NumDeclarations = numDeclarations; 807 808 if (mach->Instructions) { 809 FREE( mach->Instructions ); 810 } 811 mach->Instructions = instructions; 812 mach->NumInstructions = numInstructions; 813} 814 815 816struct tgsi_exec_machine * 817tgsi_exec_machine_create( void ) 818{ 819 struct tgsi_exec_machine *mach; 820 uint i; 821 822 mach = align_malloc( sizeof *mach, 16 ); 823 if (!mach) 824 goto fail; 825 826 memset(mach, 0, sizeof(*mach)); 827 828 mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR]; 829 mach->MaxGeometryShaderOutputs = TGSI_MAX_TOTAL_VERTICES; 830 mach->Predicates = &mach->Temps[TGSI_EXEC_TEMP_P0]; 831 832 mach->Inputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16); 833 mach->Outputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16); 834 if (!mach->Inputs || !mach->Outputs) 835 goto fail; 836 837 /* Setup constants needed by the SSE2 executor. */ 838 for( i = 0; i < 4; i++ ) { 839 mach->Temps[TGSI_EXEC_TEMP_00000000_I].xyzw[TGSI_EXEC_TEMP_00000000_C].u[i] = 0x00000000; 840 mach->Temps[TGSI_EXEC_TEMP_7FFFFFFF_I].xyzw[TGSI_EXEC_TEMP_7FFFFFFF_C].u[i] = 0x7FFFFFFF; 841 mach->Temps[TGSI_EXEC_TEMP_80000000_I].xyzw[TGSI_EXEC_TEMP_80000000_C].u[i] = 0x80000000; 842 mach->Temps[TGSI_EXEC_TEMP_FFFFFFFF_I].xyzw[TGSI_EXEC_TEMP_FFFFFFFF_C].u[i] = 0xFFFFFFFF; /* not used */ 843 mach->Temps[TGSI_EXEC_TEMP_ONE_I].xyzw[TGSI_EXEC_TEMP_ONE_C].f[i] = 1.0f; 844 mach->Temps[TGSI_EXEC_TEMP_TWO_I].xyzw[TGSI_EXEC_TEMP_TWO_C].f[i] = 2.0f; /* not used */ 845 mach->Temps[TGSI_EXEC_TEMP_128_I].xyzw[TGSI_EXEC_TEMP_128_C].f[i] = 128.0f; 846 mach->Temps[TGSI_EXEC_TEMP_MINUS_128_I].xyzw[TGSI_EXEC_TEMP_MINUS_128_C].f[i] = -128.0f; 847 mach->Temps[TGSI_EXEC_TEMP_THREE_I].xyzw[TGSI_EXEC_TEMP_THREE_C].f[i] = 3.0f; 848 mach->Temps[TGSI_EXEC_TEMP_HALF_I].xyzw[TGSI_EXEC_TEMP_HALF_C].f[i] = 0.5f; 849 } 850 851#ifdef DEBUG 852 /* silence warnings */ 853 (void) print_chan; 854 (void) print_temp; 855#endif 856 857 return mach; 858 859fail: 860 if (mach) { 861 align_free(mach->Inputs); 862 align_free(mach->Outputs); 863 align_free(mach); 864 } 865 return NULL; 866} 867 868 869void 870tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach) 871{ 872 if (mach) { 873 if (mach->Instructions) 874 FREE(mach->Instructions); 875 if (mach->Declarations) 876 FREE(mach->Declarations); 877 878 align_free(mach->Inputs); 879 align_free(mach->Outputs); 880 881 align_free(mach); 882 } 883} 884 885static void 886micro_add(union tgsi_exec_channel *dst, 887 const union tgsi_exec_channel *src0, 888 const union tgsi_exec_channel *src1) 889{ 890 dst->f[0] = src0->f[0] + src1->f[0]; 891 dst->f[1] = src0->f[1] + src1->f[1]; 892 dst->f[2] = src0->f[2] + src1->f[2]; 893 dst->f[3] = src0->f[3] + src1->f[3]; 894} 895 896static void 897micro_div( 898 union tgsi_exec_channel *dst, 899 const union tgsi_exec_channel *src0, 900 const union tgsi_exec_channel *src1 ) 901{ 902 if (src1->f[0] != 0) { 903 dst->f[0] = src0->f[0] / src1->f[0]; 904 } 905 if (src1->f[1] != 0) { 906 dst->f[1] = src0->f[1] / src1->f[1]; 907 } 908 if (src1->f[2] != 0) { 909 dst->f[2] = src0->f[2] / src1->f[2]; 910 } 911 if (src1->f[3] != 0) { 912 dst->f[3] = src0->f[3] / src1->f[3]; 913 } 914} 915 916static void 917micro_rcc(union tgsi_exec_channel *dst, 918 const union tgsi_exec_channel *src) 919{ 920 uint i; 921 922 for (i = 0; i < 4; i++) { 923 float recip = 1.0f / src->f[i]; 924 925 if (recip > 0.0f) { 926 if (recip > 1.884467e+019f) { 927 dst->f[i] = 1.884467e+019f; 928 } 929 else if (recip < 5.42101e-020f) { 930 dst->f[i] = 5.42101e-020f; 931 } 932 else { 933 dst->f[i] = recip; 934 } 935 } 936 else { 937 if (recip < -1.884467e+019f) { 938 dst->f[i] = -1.884467e+019f; 939 } 940 else if (recip > -5.42101e-020f) { 941 dst->f[i] = -5.42101e-020f; 942 } 943 else { 944 dst->f[i] = recip; 945 } 946 } 947 } 948} 949 950static void 951micro_lt( 952 union tgsi_exec_channel *dst, 953 const union tgsi_exec_channel *src0, 954 const union tgsi_exec_channel *src1, 955 const union tgsi_exec_channel *src2, 956 const union tgsi_exec_channel *src3 ) 957{ 958 dst->f[0] = src0->f[0] < src1->f[0] ? src2->f[0] : src3->f[0]; 959 dst->f[1] = src0->f[1] < src1->f[1] ? src2->f[1] : src3->f[1]; 960 dst->f[2] = src0->f[2] < src1->f[2] ? src2->f[2] : src3->f[2]; 961 dst->f[3] = src0->f[3] < src1->f[3] ? src2->f[3] : src3->f[3]; 962} 963 964static void 965micro_max(union tgsi_exec_channel *dst, 966 const union tgsi_exec_channel *src0, 967 const union tgsi_exec_channel *src1) 968{ 969 dst->f[0] = src0->f[0] > src1->f[0] ? src0->f[0] : src1->f[0]; 970 dst->f[1] = src0->f[1] > src1->f[1] ? src0->f[1] : src1->f[1]; 971 dst->f[2] = src0->f[2] > src1->f[2] ? src0->f[2] : src1->f[2]; 972 dst->f[3] = src0->f[3] > src1->f[3] ? src0->f[3] : src1->f[3]; 973} 974 975static void 976micro_min(union tgsi_exec_channel *dst, 977 const union tgsi_exec_channel *src0, 978 const union tgsi_exec_channel *src1) 979{ 980 dst->f[0] = src0->f[0] < src1->f[0] ? src0->f[0] : src1->f[0]; 981 dst->f[1] = src0->f[1] < src1->f[1] ? src0->f[1] : src1->f[1]; 982 dst->f[2] = src0->f[2] < src1->f[2] ? src0->f[2] : src1->f[2]; 983 dst->f[3] = src0->f[3] < src1->f[3] ? src0->f[3] : src1->f[3]; 984} 985 986static void 987micro_mul(union tgsi_exec_channel *dst, 988 const union tgsi_exec_channel *src0, 989 const union tgsi_exec_channel *src1) 990{ 991 dst->f[0] = src0->f[0] * src1->f[0]; 992 dst->f[1] = src0->f[1] * src1->f[1]; 993 dst->f[2] = src0->f[2] * src1->f[2]; 994 dst->f[3] = src0->f[3] * src1->f[3]; 995} 996 997static void 998micro_neg( 999 union tgsi_exec_channel *dst, 1000 const union tgsi_exec_channel *src ) 1001{ 1002 dst->f[0] = -src->f[0]; 1003 dst->f[1] = -src->f[1]; 1004 dst->f[2] = -src->f[2]; 1005 dst->f[3] = -src->f[3]; 1006} 1007 1008static void 1009micro_pow( 1010 union tgsi_exec_channel *dst, 1011 const union tgsi_exec_channel *src0, 1012 const union tgsi_exec_channel *src1 ) 1013{ 1014#if FAST_MATH 1015 dst->f[0] = util_fast_pow( src0->f[0], src1->f[0] ); 1016 dst->f[1] = util_fast_pow( src0->f[1], src1->f[1] ); 1017 dst->f[2] = util_fast_pow( src0->f[2], src1->f[2] ); 1018 dst->f[3] = util_fast_pow( src0->f[3], src1->f[3] ); 1019#else 1020 dst->f[0] = powf( src0->f[0], src1->f[0] ); 1021 dst->f[1] = powf( src0->f[1], src1->f[1] ); 1022 dst->f[2] = powf( src0->f[2], src1->f[2] ); 1023 dst->f[3] = powf( src0->f[3], src1->f[3] ); 1024#endif 1025} 1026 1027static void 1028micro_sub(union tgsi_exec_channel *dst, 1029 const union tgsi_exec_channel *src0, 1030 const union tgsi_exec_channel *src1) 1031{ 1032 dst->f[0] = src0->f[0] - src1->f[0]; 1033 dst->f[1] = src0->f[1] - src1->f[1]; 1034 dst->f[2] = src0->f[2] - src1->f[2]; 1035 dst->f[3] = src0->f[3] - src1->f[3]; 1036} 1037 1038static void 1039fetch_src_file_channel(const struct tgsi_exec_machine *mach, 1040 const uint chan_index, 1041 const uint file, 1042 const uint swizzle, 1043 const union tgsi_exec_channel *index, 1044 const union tgsi_exec_channel *index2D, 1045 union tgsi_exec_channel *chan) 1046{ 1047 uint i; 1048 1049 assert(swizzle < 4); 1050 1051 switch (file) { 1052 case TGSI_FILE_CONSTANT: 1053 for (i = 0; i < QUAD_SIZE; i++) { 1054 assert(index2D->i[i] >= 0 && index2D->i[i] < PIPE_MAX_CONSTANT_BUFFERS); 1055 assert(mach->Consts[index2D->i[i]]); 1056 1057 if (index->i[i] < 0) { 1058 chan->u[i] = 0; 1059 } else { 1060 /* NOTE: copying the const value as a uint instead of float */ 1061 const uint constbuf = index2D->i[i]; 1062 const uint *buf = (const uint *)mach->Consts[constbuf]; 1063 const int pos = index->i[i] * 4 + swizzle; 1064 /* const buffer bounds check */ 1065 if (pos < 0 || pos >= mach->ConstsSize[constbuf]) { 1066 if (0) { 1067 /* Debug: print warning */ 1068 static int count = 0; 1069 if (count++ < 100) 1070 debug_printf("TGSI Exec: const buffer index %d" 1071 " out of bounds\n", pos); 1072 } 1073 chan->u[i] = 0; 1074 } 1075 else 1076 chan->u[i] = buf[pos]; 1077 } 1078 } 1079 break; 1080 1081 case TGSI_FILE_INPUT: 1082 for (i = 0; i < QUAD_SIZE; i++) { 1083 /* 1084 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1085 debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n", 1086 index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i], 1087 index2D->i[i], index->i[i]); 1088 }*/ 1089 int pos = index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i]; 1090 assert(pos >= 0); 1091 assert(pos < TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS); 1092 chan->u[i] = mach->Inputs[pos].xyzw[swizzle].u[i]; 1093 } 1094 break; 1095 1096 case TGSI_FILE_SYSTEM_VALUE: 1097 /* XXX no swizzling at this point. Will be needed if we put 1098 * gl_FragCoord, for example, in a sys value register. 1099 */ 1100 for (i = 0; i < QUAD_SIZE; i++) { 1101 chan->u[i] = mach->SystemValue[index->i[i]].u[i]; 1102 } 1103 break; 1104 1105 case TGSI_FILE_TEMPORARY: 1106 for (i = 0; i < QUAD_SIZE; i++) { 1107 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); 1108 assert(index2D->i[i] == 0); 1109 1110 chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i]; 1111 } 1112 break; 1113 1114 case TGSI_FILE_TEMPORARY_ARRAY: 1115 for (i = 0; i < QUAD_SIZE; i++) { 1116 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); 1117 assert(index2D->i[i] < TGSI_EXEC_NUM_TEMP_ARRAYS); 1118 1119 chan->u[i] = 1120 mach->TempArray[index2D->i[i]][index->i[i]].xyzw[swizzle].u[i]; 1121 } 1122 break; 1123 1124 case TGSI_FILE_IMMEDIATE: 1125 for (i = 0; i < QUAD_SIZE; i++) { 1126 assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit); 1127 assert(index2D->i[i] == 0); 1128 1129 chan->f[i] = mach->Imms[index->i[i]][swizzle]; 1130 } 1131 break; 1132 1133 case TGSI_FILE_IMMEDIATE_ARRAY: 1134 for (i = 0; i < QUAD_SIZE; i++) { 1135 assert(index2D->i[i] == 0); 1136 1137 chan->f[i] = mach->ImmArray[index->i[i]][swizzle]; 1138 } 1139 break; 1140 1141 case TGSI_FILE_ADDRESS: 1142 for (i = 0; i < QUAD_SIZE; i++) { 1143 assert(index->i[i] >= 0); 1144 assert(index2D->i[i] == 0); 1145 1146 chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i]; 1147 } 1148 break; 1149 1150 case TGSI_FILE_PREDICATE: 1151 for (i = 0; i < QUAD_SIZE; i++) { 1152 assert(index->i[i] >= 0 && index->i[i] < TGSI_EXEC_NUM_PREDS); 1153 assert(index2D->i[i] == 0); 1154 1155 chan->u[i] = mach->Predicates[0].xyzw[swizzle].u[i]; 1156 } 1157 break; 1158 1159 case TGSI_FILE_OUTPUT: 1160 /* vertex/fragment output vars can be read too */ 1161 for (i = 0; i < QUAD_SIZE; i++) { 1162 assert(index->i[i] >= 0); 1163 assert(index2D->i[i] == 0); 1164 1165 chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i]; 1166 } 1167 break; 1168 1169 default: 1170 assert(0); 1171 for (i = 0; i < QUAD_SIZE; i++) { 1172 chan->u[i] = 0; 1173 } 1174 } 1175} 1176 1177static void 1178fetch_source(const struct tgsi_exec_machine *mach, 1179 union tgsi_exec_channel *chan, 1180 const struct tgsi_full_src_register *reg, 1181 const uint chan_index, 1182 enum tgsi_exec_datatype src_datatype) 1183{ 1184 union tgsi_exec_channel index; 1185 union tgsi_exec_channel index2D; 1186 uint swizzle; 1187 1188 /* We start with a direct index into a register file. 1189 * 1190 * file[1], 1191 * where: 1192 * file = Register.File 1193 * [1] = Register.Index 1194 */ 1195 index.i[0] = 1196 index.i[1] = 1197 index.i[2] = 1198 index.i[3] = reg->Register.Index; 1199 1200 /* There is an extra source register that indirectly subscripts 1201 * a register file. The direct index now becomes an offset 1202 * that is being added to the indirect register. 1203 * 1204 * file[ind[2].x+1], 1205 * where: 1206 * ind = Indirect.File 1207 * [2] = Indirect.Index 1208 * .x = Indirect.SwizzleX 1209 */ 1210 if (reg->Register.Indirect) { 1211 union tgsi_exec_channel index2; 1212 union tgsi_exec_channel indir_index; 1213 const uint execmask = mach->ExecMask; 1214 uint i; 1215 1216 /* which address register (always zero now) */ 1217 index2.i[0] = 1218 index2.i[1] = 1219 index2.i[2] = 1220 index2.i[3] = reg->Indirect.Index; 1221 assert(reg->Indirect.File == TGSI_FILE_ADDRESS); 1222 /* get current value of address register[swizzle] */ 1223 swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, CHAN_X ); 1224 fetch_src_file_channel(mach, 1225 chan_index, 1226 reg->Indirect.File, 1227 swizzle, 1228 &index2, 1229 &ZeroVec, 1230 &indir_index); 1231 1232 /* add value of address register to the offset */ 1233 index.i[0] += indir_index.i[0]; 1234 index.i[1] += indir_index.i[1]; 1235 index.i[2] += indir_index.i[2]; 1236 index.i[3] += indir_index.i[3]; 1237 1238 /* for disabled execution channels, zero-out the index to 1239 * avoid using a potential garbage value. 1240 */ 1241 for (i = 0; i < QUAD_SIZE; i++) { 1242 if ((execmask & (1 << i)) == 0) 1243 index.i[i] = 0; 1244 } 1245 } 1246 1247 /* There is an extra source register that is a second 1248 * subscript to a register file. Effectively it means that 1249 * the register file is actually a 2D array of registers. 1250 * 1251 * file[3][1], 1252 * where: 1253 * [3] = Dimension.Index 1254 */ 1255 if (reg->Register.Dimension) { 1256 index2D.i[0] = 1257 index2D.i[1] = 1258 index2D.i[2] = 1259 index2D.i[3] = reg->Dimension.Index; 1260 1261 /* Again, the second subscript index can be addressed indirectly 1262 * identically to the first one. 1263 * Nothing stops us from indirectly addressing the indirect register, 1264 * but there is no need for that, so we won't exercise it. 1265 * 1266 * file[ind[4].y+3][1], 1267 * where: 1268 * ind = DimIndirect.File 1269 * [4] = DimIndirect.Index 1270 * .y = DimIndirect.SwizzleX 1271 */ 1272 if (reg->Dimension.Indirect) { 1273 union tgsi_exec_channel index2; 1274 union tgsi_exec_channel indir_index; 1275 const uint execmask = mach->ExecMask; 1276 uint i; 1277 1278 index2.i[0] = 1279 index2.i[1] = 1280 index2.i[2] = 1281 index2.i[3] = reg->DimIndirect.Index; 1282 1283 swizzle = tgsi_util_get_src_register_swizzle( ®->DimIndirect, CHAN_X ); 1284 fetch_src_file_channel(mach, 1285 chan_index, 1286 reg->DimIndirect.File, 1287 swizzle, 1288 &index2, 1289 &ZeroVec, 1290 &indir_index); 1291 1292 index2D.i[0] += indir_index.i[0]; 1293 index2D.i[1] += indir_index.i[1]; 1294 index2D.i[2] += indir_index.i[2]; 1295 index2D.i[3] += indir_index.i[3]; 1296 1297 /* for disabled execution channels, zero-out the index to 1298 * avoid using a potential garbage value. 1299 */ 1300 for (i = 0; i < QUAD_SIZE; i++) { 1301 if ((execmask & (1 << i)) == 0) { 1302 index2D.i[i] = 0; 1303 } 1304 } 1305 } 1306 1307 /* If by any chance there was a need for a 3D array of register 1308 * files, we would have to check whether Dimension is followed 1309 * by a dimension register and continue the saga. 1310 */ 1311 } else { 1312 index2D.i[0] = 1313 index2D.i[1] = 1314 index2D.i[2] = 1315 index2D.i[3] = 0; 1316 } 1317 1318 swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index ); 1319 fetch_src_file_channel(mach, 1320 chan_index, 1321 reg->Register.File, 1322 swizzle, 1323 &index, 1324 &index2D, 1325 chan); 1326 1327 if (reg->Register.Absolute) { 1328 if (src_datatype == TGSI_EXEC_DATA_FLOAT) { 1329 micro_abs(chan, chan); 1330 } else { 1331 micro_iabs(chan, chan); 1332 } 1333 } 1334 1335 if (reg->Register.Negate) { 1336 if (src_datatype == TGSI_EXEC_DATA_FLOAT) { 1337 micro_neg(chan, chan); 1338 } else { 1339 micro_ineg(chan, chan); 1340 } 1341 } 1342} 1343 1344static void 1345store_dest(struct tgsi_exec_machine *mach, 1346 const union tgsi_exec_channel *chan, 1347 const struct tgsi_full_dst_register *reg, 1348 const struct tgsi_full_instruction *inst, 1349 uint chan_index, 1350 enum tgsi_exec_datatype dst_datatype) 1351{ 1352 uint i; 1353 union tgsi_exec_channel null; 1354 union tgsi_exec_channel *dst; 1355 union tgsi_exec_channel index2D; 1356 uint execmask = mach->ExecMask; 1357 int offset = 0; /* indirection offset */ 1358 int index; 1359 1360 /* for debugging */ 1361 if (0 && dst_datatype == TGSI_EXEC_DATA_FLOAT) { 1362 check_inf_or_nan(chan); 1363 } 1364 1365 /* There is an extra source register that indirectly subscripts 1366 * a register file. The direct index now becomes an offset 1367 * that is being added to the indirect register. 1368 * 1369 * file[ind[2].x+1], 1370 * where: 1371 * ind = Indirect.File 1372 * [2] = Indirect.Index 1373 * .x = Indirect.SwizzleX 1374 */ 1375 if (reg->Register.Indirect) { 1376 union tgsi_exec_channel index; 1377 union tgsi_exec_channel indir_index; 1378 uint swizzle; 1379 1380 /* which address register (always zero for now) */ 1381 index.i[0] = 1382 index.i[1] = 1383 index.i[2] = 1384 index.i[3] = reg->Indirect.Index; 1385 1386 /* get current value of address register[swizzle] */ 1387 swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, CHAN_X ); 1388 1389 /* fetch values from the address/indirection register */ 1390 fetch_src_file_channel(mach, 1391 chan_index, 1392 reg->Indirect.File, 1393 swizzle, 1394 &index, 1395 &ZeroVec, 1396 &indir_index); 1397 1398 /* save indirection offset */ 1399 offset = indir_index.i[0]; 1400 } 1401 1402 /* There is an extra source register that is a second 1403 * subscript to a register file. Effectively it means that 1404 * the register file is actually a 2D array of registers. 1405 * 1406 * file[3][1], 1407 * where: 1408 * [3] = Dimension.Index 1409 */ 1410 if (reg->Register.Dimension) { 1411 index2D.i[0] = 1412 index2D.i[1] = 1413 index2D.i[2] = 1414 index2D.i[3] = reg->Dimension.Index; 1415 1416 /* Again, the second subscript index can be addressed indirectly 1417 * identically to the first one. 1418 * Nothing stops us from indirectly addressing the indirect register, 1419 * but there is no need for that, so we won't exercise it. 1420 * 1421 * file[ind[4].y+3][1], 1422 * where: 1423 * ind = DimIndirect.File 1424 * [4] = DimIndirect.Index 1425 * .y = DimIndirect.SwizzleX 1426 */ 1427 if (reg->Dimension.Indirect) { 1428 union tgsi_exec_channel index2; 1429 union tgsi_exec_channel indir_index; 1430 const uint execmask = mach->ExecMask; 1431 unsigned swizzle; 1432 uint i; 1433 1434 index2.i[0] = 1435 index2.i[1] = 1436 index2.i[2] = 1437 index2.i[3] = reg->DimIndirect.Index; 1438 1439 swizzle = tgsi_util_get_src_register_swizzle( ®->DimIndirect, CHAN_X ); 1440 fetch_src_file_channel(mach, 1441 chan_index, 1442 reg->DimIndirect.File, 1443 swizzle, 1444 &index2, 1445 &ZeroVec, 1446 &indir_index); 1447 1448 index2D.i[0] += indir_index.i[0]; 1449 index2D.i[1] += indir_index.i[1]; 1450 index2D.i[2] += indir_index.i[2]; 1451 index2D.i[3] += indir_index.i[3]; 1452 1453 /* for disabled execution channels, zero-out the index to 1454 * avoid using a potential garbage value. 1455 */ 1456 for (i = 0; i < QUAD_SIZE; i++) { 1457 if ((execmask & (1 << i)) == 0) { 1458 index2D.i[i] = 0; 1459 } 1460 } 1461 } 1462 1463 /* If by any chance there was a need for a 3D array of register 1464 * files, we would have to check whether Dimension is followed 1465 * by a dimension register and continue the saga. 1466 */ 1467 } else { 1468 index2D.i[0] = 1469 index2D.i[1] = 1470 index2D.i[2] = 1471 index2D.i[3] = 0; 1472 } 1473 1474 switch (reg->Register.File) { 1475 case TGSI_FILE_NULL: 1476 dst = &null; 1477 break; 1478 1479 case TGSI_FILE_OUTPUT: 1480 index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] 1481 + reg->Register.Index; 1482 dst = &mach->Outputs[offset + index].xyzw[chan_index]; 1483#if 0 1484 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1485 fprintf(stderr, "STORING OUT[%d] mask(%d), = (", offset + index, execmask); 1486 for (i = 0; i < QUAD_SIZE; i++) 1487 if (execmask & (1 << i)) 1488 fprintf(stderr, "%f, ", chan->f[i]); 1489 fprintf(stderr, ")\n"); 1490 } 1491#endif 1492 break; 1493 1494 case TGSI_FILE_TEMPORARY: 1495 index = reg->Register.Index; 1496 assert( index < TGSI_EXEC_NUM_TEMPS ); 1497 dst = &mach->Temps[offset + index].xyzw[chan_index]; 1498 break; 1499 1500 case TGSI_FILE_TEMPORARY_ARRAY: 1501 index = reg->Register.Index; 1502 assert( index < TGSI_EXEC_NUM_TEMPS ); 1503 assert( index2D.i[0] < TGSI_EXEC_NUM_TEMP_ARRAYS ); 1504 /* XXX we use index2D.i[0] here but somehow we might 1505 * end up with someone trying to store indirectly in 1506 * different buffers */ 1507 dst = &mach->TempArray[index2D.i[0]][offset + index].xyzw[chan_index]; 1508 break; 1509 1510 case TGSI_FILE_ADDRESS: 1511 index = reg->Register.Index; 1512 dst = &mach->Addrs[index].xyzw[chan_index]; 1513 break; 1514 1515 case TGSI_FILE_PREDICATE: 1516 index = reg->Register.Index; 1517 assert(index < TGSI_EXEC_NUM_PREDS); 1518 dst = &mach->Predicates[index].xyzw[chan_index]; 1519 break; 1520 1521 default: 1522 assert( 0 ); 1523 return; 1524 } 1525 1526 if (inst->Instruction.Predicate) { 1527 uint swizzle; 1528 union tgsi_exec_channel *pred; 1529 1530 switch (chan_index) { 1531 case CHAN_X: 1532 swizzle = inst->Predicate.SwizzleX; 1533 break; 1534 case CHAN_Y: 1535 swizzle = inst->Predicate.SwizzleY; 1536 break; 1537 case CHAN_Z: 1538 swizzle = inst->Predicate.SwizzleZ; 1539 break; 1540 case CHAN_W: 1541 swizzle = inst->Predicate.SwizzleW; 1542 break; 1543 default: 1544 assert(0); 1545 return; 1546 } 1547 1548 assert(inst->Predicate.Index == 0); 1549 1550 pred = &mach->Predicates[inst->Predicate.Index].xyzw[swizzle]; 1551 1552 if (inst->Predicate.Negate) { 1553 for (i = 0; i < QUAD_SIZE; i++) { 1554 if (pred->u[i]) { 1555 execmask &= ~(1 << i); 1556 } 1557 } 1558 } else { 1559 for (i = 0; i < QUAD_SIZE; i++) { 1560 if (!pred->u[i]) { 1561 execmask &= ~(1 << i); 1562 } 1563 } 1564 } 1565 } 1566 1567 switch (inst->Instruction.Saturate) { 1568 case TGSI_SAT_NONE: 1569 for (i = 0; i < QUAD_SIZE; i++) 1570 if (execmask & (1 << i)) 1571 dst->i[i] = chan->i[i]; 1572 break; 1573 1574 case TGSI_SAT_ZERO_ONE: 1575 for (i = 0; i < QUAD_SIZE; i++) 1576 if (execmask & (1 << i)) { 1577 if (chan->f[i] < 0.0f) 1578 dst->f[i] = 0.0f; 1579 else if (chan->f[i] > 1.0f) 1580 dst->f[i] = 1.0f; 1581 else 1582 dst->i[i] = chan->i[i]; 1583 } 1584 break; 1585 1586 case TGSI_SAT_MINUS_PLUS_ONE: 1587 for (i = 0; i < QUAD_SIZE; i++) 1588 if (execmask & (1 << i)) { 1589 if (chan->f[i] < -1.0f) 1590 dst->f[i] = -1.0f; 1591 else if (chan->f[i] > 1.0f) 1592 dst->f[i] = 1.0f; 1593 else 1594 dst->i[i] = chan->i[i]; 1595 } 1596 break; 1597 1598 default: 1599 assert( 0 ); 1600 } 1601} 1602 1603#define FETCH(VAL,INDEX,CHAN)\ 1604 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT) 1605 1606#define IFETCH(VAL,INDEX,CHAN)\ 1607 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT) 1608 1609 1610/** 1611 * Execute ARB-style KIL which is predicated by a src register. 1612 * Kill fragment if any of the four values is less than zero. 1613 */ 1614static void 1615exec_kil(struct tgsi_exec_machine *mach, 1616 const struct tgsi_full_instruction *inst) 1617{ 1618 uint uniquemask; 1619 uint chan_index; 1620 uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ 1621 union tgsi_exec_channel r[1]; 1622 1623 /* This mask stores component bits that were already tested. */ 1624 uniquemask = 0; 1625 1626 for (chan_index = 0; chan_index < 4; chan_index++) 1627 { 1628 uint swizzle; 1629 uint i; 1630 1631 /* unswizzle channel */ 1632 swizzle = tgsi_util_get_full_src_register_swizzle ( 1633 &inst->Src[0], 1634 chan_index); 1635 1636 /* check if the component has not been already tested */ 1637 if (uniquemask & (1 << swizzle)) 1638 continue; 1639 uniquemask |= 1 << swizzle; 1640 1641 FETCH(&r[0], 0, chan_index); 1642 for (i = 0; i < 4; i++) 1643 if (r[0].f[i] < 0.0f) 1644 kilmask |= 1 << i; 1645 } 1646 1647 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; 1648} 1649 1650/** 1651 * Execute NVIDIA-style KIL which is predicated by a condition code. 1652 * Kill fragment if the condition code is TRUE. 1653 */ 1654static void 1655exec_kilp(struct tgsi_exec_machine *mach, 1656 const struct tgsi_full_instruction *inst) 1657{ 1658 uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ 1659 1660 /* "unconditional" kil */ 1661 kilmask = mach->ExecMask; 1662 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; 1663} 1664 1665static void 1666emit_vertex(struct tgsi_exec_machine *mach) 1667{ 1668 /* FIXME: check for exec mask correctly 1669 unsigned i; 1670 for (i = 0; i < QUAD_SIZE; ++i) { 1671 if ((mach->ExecMask & (1 << i))) 1672 */ 1673 if (mach->ExecMask) { 1674 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += mach->NumOutputs; 1675 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++; 1676 } 1677} 1678 1679static void 1680emit_primitive(struct tgsi_exec_machine *mach) 1681{ 1682 unsigned *prim_count = &mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]; 1683 /* FIXME: check for exec mask correctly 1684 unsigned i; 1685 for (i = 0; i < QUAD_SIZE; ++i) { 1686 if ((mach->ExecMask & (1 << i))) 1687 */ 1688 if (mach->ExecMask) { 1689 ++(*prim_count); 1690 debug_assert((*prim_count * mach->NumOutputs) < mach->MaxGeometryShaderOutputs); 1691 mach->Primitives[*prim_count] = 0; 1692 } 1693} 1694 1695static void 1696conditional_emit_primitive(struct tgsi_exec_machine *mach) 1697{ 1698 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1699 int emitted_verts = 1700 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]; 1701 if (emitted_verts) { 1702 emit_primitive(mach); 1703 } 1704 } 1705} 1706 1707 1708/* 1709 * Fetch four texture samples using STR texture coordinates. 1710 */ 1711static void 1712fetch_texel( struct tgsi_sampler *sampler, 1713 const union tgsi_exec_channel *s, 1714 const union tgsi_exec_channel *t, 1715 const union tgsi_exec_channel *p, 1716 const union tgsi_exec_channel *c0, 1717 enum tgsi_sampler_control control, 1718 union tgsi_exec_channel *r, 1719 union tgsi_exec_channel *g, 1720 union tgsi_exec_channel *b, 1721 union tgsi_exec_channel *a ) 1722{ 1723 uint j; 1724 float rgba[NUM_CHANNELS][QUAD_SIZE]; 1725 1726 sampler->get_samples(sampler, s->f, t->f, p->f, c0->f, control, rgba); 1727 1728 for (j = 0; j < 4; j++) { 1729 r->f[j] = rgba[0][j]; 1730 g->f[j] = rgba[1][j]; 1731 b->f[j] = rgba[2][j]; 1732 a->f[j] = rgba[3][j]; 1733 } 1734} 1735 1736 1737#define TEX_MODIFIER_NONE 0 1738#define TEX_MODIFIER_PROJECTED 1 1739#define TEX_MODIFIER_LOD_BIAS 2 1740#define TEX_MODIFIER_EXPLICIT_LOD 3 1741 1742 1743static void 1744exec_tex(struct tgsi_exec_machine *mach, 1745 const struct tgsi_full_instruction *inst, 1746 uint modifier) 1747{ 1748 const uint unit = inst->Src[1].Register.Index; 1749 union tgsi_exec_channel r[4]; 1750 const union tgsi_exec_channel *lod = &ZeroVec; 1751 enum tgsi_sampler_control control; 1752 uint chan; 1753 1754 if (modifier != TEX_MODIFIER_NONE) { 1755 FETCH(&r[3], 0, CHAN_W); 1756 if (modifier != TEX_MODIFIER_PROJECTED) { 1757 lod = &r[3]; 1758 } 1759 } 1760 1761 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) { 1762 control = tgsi_sampler_lod_explicit; 1763 } else { 1764 control = tgsi_sampler_lod_bias; 1765 } 1766 1767 switch (inst->Texture.Texture) { 1768 case TGSI_TEXTURE_1D: 1769 FETCH(&r[0], 0, CHAN_X); 1770 1771 if (modifier == TEX_MODIFIER_PROJECTED) { 1772 micro_div(&r[0], &r[0], &r[3]); 1773 } 1774 1775 fetch_texel(mach->Samplers[unit], 1776 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */ 1777 control, 1778 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1779 break; 1780 case TGSI_TEXTURE_SHADOW1D: 1781 FETCH(&r[0], 0, CHAN_X); 1782 FETCH(&r[2], 0, CHAN_Z); 1783 1784 if (modifier == TEX_MODIFIER_PROJECTED) { 1785 micro_div(&r[0], &r[0], &r[3]); 1786 } 1787 1788 fetch_texel(mach->Samplers[unit], 1789 &r[0], &ZeroVec, &r[2], lod, /* S, T, P, LOD */ 1790 control, 1791 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1792 break; 1793 1794 case TGSI_TEXTURE_2D: 1795 case TGSI_TEXTURE_RECT: 1796 case TGSI_TEXTURE_SHADOW2D: 1797 case TGSI_TEXTURE_SHADOWRECT: 1798 FETCH(&r[0], 0, CHAN_X); 1799 FETCH(&r[1], 0, CHAN_Y); 1800 FETCH(&r[2], 0, CHAN_Z); 1801 1802 if (modifier == TEX_MODIFIER_PROJECTED) { 1803 micro_div(&r[0], &r[0], &r[3]); 1804 micro_div(&r[1], &r[1], &r[3]); 1805 micro_div(&r[2], &r[2], &r[3]); 1806 } 1807 1808 fetch_texel(mach->Samplers[unit], 1809 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1810 control, 1811 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1812 break; 1813 1814 case TGSI_TEXTURE_1D_ARRAY: 1815 FETCH(&r[0], 0, CHAN_X); 1816 FETCH(&r[1], 0, CHAN_Y); 1817 1818 if (modifier == TEX_MODIFIER_PROJECTED) { 1819 micro_div(&r[0], &r[0], &r[3]); 1820 } 1821 1822 fetch_texel(mach->Samplers[unit], 1823 &r[0], &r[1], &ZeroVec, lod, /* S, T, P, LOD */ 1824 control, 1825 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1826 break; 1827 case TGSI_TEXTURE_SHADOW1D_ARRAY: 1828 FETCH(&r[0], 0, CHAN_X); 1829 FETCH(&r[1], 0, CHAN_Y); 1830 FETCH(&r[2], 0, CHAN_Z); 1831 1832 if (modifier == TEX_MODIFIER_PROJECTED) { 1833 micro_div(&r[0], &r[0], &r[3]); 1834 } 1835 1836 fetch_texel(mach->Samplers[unit], 1837 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1838 control, 1839 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1840 break; 1841 1842 case TGSI_TEXTURE_2D_ARRAY: 1843 FETCH(&r[0], 0, CHAN_X); 1844 FETCH(&r[1], 0, CHAN_Y); 1845 FETCH(&r[2], 0, CHAN_Z); 1846 1847 if (modifier == TEX_MODIFIER_PROJECTED) { 1848 micro_div(&r[0], &r[0], &r[3]); 1849 micro_div(&r[1], &r[1], &r[3]); 1850 } 1851 1852 fetch_texel(mach->Samplers[unit], 1853 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1854 control, 1855 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1856 break; 1857 case TGSI_TEXTURE_SHADOW2D_ARRAY: 1858 FETCH(&r[0], 0, CHAN_X); 1859 FETCH(&r[1], 0, CHAN_Y); 1860 FETCH(&r[2], 0, CHAN_Z); 1861 FETCH(&r[3], 0, CHAN_W); 1862 1863 fetch_texel(mach->Samplers[unit], 1864 &r[0], &r[1], &r[2], &r[3], /* S, T, P, LOD */ 1865 control, 1866 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1867 break; 1868 case TGSI_TEXTURE_3D: 1869 case TGSI_TEXTURE_CUBE: 1870 FETCH(&r[0], 0, CHAN_X); 1871 FETCH(&r[1], 0, CHAN_Y); 1872 FETCH(&r[2], 0, CHAN_Z); 1873 1874 if (modifier == TEX_MODIFIER_PROJECTED) { 1875 micro_div(&r[0], &r[0], &r[3]); 1876 micro_div(&r[1], &r[1], &r[3]); 1877 micro_div(&r[2], &r[2], &r[3]); 1878 } 1879 1880 fetch_texel(mach->Samplers[unit], 1881 &r[0], &r[1], &r[2], lod, 1882 control, 1883 &r[0], &r[1], &r[2], &r[3]); 1884 break; 1885 1886 default: 1887 assert(0); 1888 } 1889 1890#if 0 1891 debug_printf("fetch r: %g %g %g %g\n", 1892 r[0].f[0], r[0].f[1], r[0].f[2], r[0].f[3]); 1893 debug_printf("fetch g: %g %g %g %g\n", 1894 r[1].f[0], r[1].f[1], r[1].f[2], r[1].f[3]); 1895 debug_printf("fetch b: %g %g %g %g\n", 1896 r[2].f[0], r[2].f[1], r[2].f[2], r[2].f[3]); 1897 debug_printf("fetch a: %g %g %g %g\n", 1898 r[3].f[0], r[3].f[1], r[3].f[2], r[3].f[3]); 1899#endif 1900 1901 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1902 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1903 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1904 } 1905 } 1906} 1907 1908static void 1909exec_txd(struct tgsi_exec_machine *mach, 1910 const struct tgsi_full_instruction *inst) 1911{ 1912 const uint unit = inst->Src[3].Register.Index; 1913 union tgsi_exec_channel r[4]; 1914 uint chan; 1915 1916 /* 1917 * XXX: This is fake TXD -- the derivatives are not taken into account, yet. 1918 */ 1919 1920 switch (inst->Texture.Texture) { 1921 case TGSI_TEXTURE_1D: 1922 case TGSI_TEXTURE_SHADOW1D: 1923 1924 FETCH(&r[0], 0, CHAN_X); 1925 1926 fetch_texel(mach->Samplers[unit], 1927 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */ 1928 tgsi_sampler_lod_bias, 1929 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1930 break; 1931 1932 case TGSI_TEXTURE_1D_ARRAY: 1933 case TGSI_TEXTURE_2D: 1934 case TGSI_TEXTURE_RECT: 1935 case TGSI_TEXTURE_SHADOW1D_ARRAY: 1936 case TGSI_TEXTURE_SHADOW2D: 1937 case TGSI_TEXTURE_SHADOWRECT: 1938 1939 FETCH(&r[0], 0, CHAN_X); 1940 FETCH(&r[1], 0, CHAN_Y); 1941 FETCH(&r[2], 0, CHAN_Z); 1942 1943 fetch_texel(mach->Samplers[unit], 1944 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */ 1945 tgsi_sampler_lod_bias, 1946 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1947 break; 1948 1949 case TGSI_TEXTURE_2D_ARRAY: 1950 case TGSI_TEXTURE_3D: 1951 case TGSI_TEXTURE_CUBE: 1952 1953 FETCH(&r[0], 0, CHAN_X); 1954 FETCH(&r[1], 0, CHAN_Y); 1955 FETCH(&r[2], 0, CHAN_Z); 1956 1957 fetch_texel(mach->Samplers[unit], 1958 &r[0], &r[1], &r[2], &ZeroVec, 1959 tgsi_sampler_lod_bias, 1960 &r[0], &r[1], &r[2], &r[3]); 1961 break; 1962 1963 case TGSI_TEXTURE_SHADOW2D_ARRAY: 1964 1965 FETCH(&r[0], 0, CHAN_X); 1966 FETCH(&r[1], 0, CHAN_Y); 1967 FETCH(&r[2], 0, CHAN_Z); 1968 FETCH(&r[3], 0, CHAN_W); 1969 1970 fetch_texel(mach->Samplers[unit], 1971 &r[0], &r[1], &r[2], &r[3], 1972 tgsi_sampler_lod_bias, 1973 &r[0], &r[1], &r[2], &r[3]); 1974 break; 1975 1976 default: 1977 assert(0); 1978 } 1979 1980 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1981 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1982 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1983 } 1984 } 1985} 1986 1987 1988static void 1989exec_txf(struct tgsi_exec_machine *mach, 1990 const struct tgsi_full_instruction *inst) 1991{ 1992 struct tgsi_sampler *sampler; 1993 const uint unit = inst->Src[2].Register.Index; 1994 union tgsi_exec_channel r[4]; 1995 union tgsi_exec_channel offset[3]; 1996 uint chan; 1997 float rgba[NUM_CHANNELS][QUAD_SIZE]; 1998 int j; 1999 int8_t offsets[3]; 2000 2001 if (inst->Texture.NumOffsets == 1) { 2002 union tgsi_exec_channel index; 2003 index.i[0] = index.i[1] = index.i[2] = index.i[3] = inst->TexOffsets[0].Index; 2004 fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File, 2005 inst->TexOffsets[0].SwizzleX, &index, &ZeroVec, &offset[0]); 2006 fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File, 2007 inst->TexOffsets[0].SwizzleY, &index, &ZeroVec, &offset[1]); 2008 fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File, 2009 inst->TexOffsets[0].SwizzleZ, &index, &ZeroVec, &offset[2]); 2010 offsets[0] = offset[0].i[0]; 2011 offsets[1] = offset[1].i[0]; 2012 offsets[2] = offset[2].i[0]; 2013 } else 2014 offsets[0] = offsets[1] = offsets[2] = 0; 2015 2016 IFETCH(&r[3], 0, CHAN_W); 2017 2018 switch(inst->Texture.Texture) { 2019 case TGSI_TEXTURE_3D: 2020 case TGSI_TEXTURE_2D_ARRAY: 2021 case TGSI_TEXTURE_SHADOW2D_ARRAY: 2022 IFETCH(&r[2], 0, CHAN_Z); 2023 /* fallthrough */ 2024 case TGSI_TEXTURE_2D: 2025 case TGSI_TEXTURE_RECT: 2026 case TGSI_TEXTURE_SHADOW1D_ARRAY: 2027 case TGSI_TEXTURE_SHADOW2D: 2028 case TGSI_TEXTURE_SHADOWRECT: 2029 case TGSI_TEXTURE_1D_ARRAY: 2030 IFETCH(&r[1], 0, CHAN_Y); 2031 /* fallthrough */ 2032 case TGSI_TEXTURE_1D: 2033 case TGSI_TEXTURE_SHADOW1D: 2034 IFETCH(&r[0], 0, CHAN_X); 2035 break; 2036 default: 2037 assert(0); 2038 break; 2039 } 2040 2041 sampler = mach->Samplers[unit]; 2042 sampler->get_texel(sampler, r[0].i, r[1].i, r[2].i, r[3].i, 2043 offsets, rgba); 2044 2045 for (j = 0; j < QUAD_SIZE; j++) { 2046 r[0].f[j] = rgba[0][j]; 2047 r[1].f[j] = rgba[1][j]; 2048 r[2].f[j] = rgba[2][j]; 2049 r[3].f[j] = rgba[3][j]; 2050 } 2051 2052 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2053 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2054 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2055 } 2056 } 2057} 2058 2059static void 2060exec_txq(struct tgsi_exec_machine *mach, 2061 const struct tgsi_full_instruction *inst) 2062{ 2063 struct tgsi_sampler *sampler; 2064 const uint unit = inst->Src[1].Register.Index; 2065 int result[4]; 2066 union tgsi_exec_channel r[4], src; 2067 uint chan; 2068 int i,j; 2069 2070 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_INT); 2071 sampler = mach->Samplers[unit]; 2072 2073 sampler->get_dims(sampler, src.i[0], result); 2074 2075 for (i = 0; i < QUAD_SIZE; i++) { 2076 for (j = 0; j < 4; j++) { 2077 r[j].i[i] = result[j]; 2078 } 2079 } 2080 2081 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2082 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2083 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, 2084 TGSI_EXEC_DATA_INT); 2085 } 2086 } 2087} 2088 2089static void 2090exec_sample(struct tgsi_exec_machine *mach, 2091 const struct tgsi_full_instruction *inst, 2092 uint modifier) 2093{ 2094 const uint resource_unit = inst->Src[1].Register.Index; 2095 const uint sampler_unit = inst->Src[2].Register.Index; 2096 union tgsi_exec_channel r[4]; 2097 const union tgsi_exec_channel *lod = &ZeroVec; 2098 enum tgsi_sampler_control control; 2099 uint chan; 2100 2101 if (modifier != TEX_MODIFIER_NONE) { 2102 if (modifier == TEX_MODIFIER_LOD_BIAS) 2103 FETCH(&r[3], 3, CHAN_X); 2104 else /*TEX_MODIFIER_LOD*/ 2105 FETCH(&r[3], 0, CHAN_W); 2106 2107 if (modifier != TEX_MODIFIER_PROJECTED) { 2108 lod = &r[3]; 2109 } 2110 } 2111 2112 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) { 2113 control = tgsi_sampler_lod_explicit; 2114 } else { 2115 control = tgsi_sampler_lod_bias; 2116 } 2117 2118 switch (mach->Resources[resource_unit].Resource) { 2119 case TGSI_TEXTURE_1D: 2120 case TGSI_TEXTURE_SHADOW1D: 2121 FETCH(&r[0], 0, CHAN_X); 2122 2123 if (modifier == TEX_MODIFIER_PROJECTED) { 2124 micro_div(&r[0], &r[0], &r[3]); 2125 } 2126 2127 fetch_texel(mach->Samplers[sampler_unit], 2128 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */ 2129 control, 2130 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 2131 break; 2132 2133 case TGSI_TEXTURE_1D_ARRAY: 2134 case TGSI_TEXTURE_2D: 2135 case TGSI_TEXTURE_RECT: 2136 case TGSI_TEXTURE_SHADOW1D_ARRAY: 2137 case TGSI_TEXTURE_SHADOW2D: 2138 case TGSI_TEXTURE_SHADOWRECT: 2139 FETCH(&r[0], 0, CHAN_X); 2140 FETCH(&r[1], 0, CHAN_Y); 2141 FETCH(&r[2], 0, CHAN_Z); 2142 2143 if (modifier == TEX_MODIFIER_PROJECTED) { 2144 micro_div(&r[0], &r[0], &r[3]); 2145 micro_div(&r[1], &r[1], &r[3]); 2146 micro_div(&r[2], &r[2], &r[3]); 2147 } 2148 2149 fetch_texel(mach->Samplers[sampler_unit], 2150 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 2151 control, 2152 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 2153 break; 2154 2155 case TGSI_TEXTURE_2D_ARRAY: 2156 case TGSI_TEXTURE_3D: 2157 case TGSI_TEXTURE_CUBE: 2158 FETCH(&r[0], 0, CHAN_X); 2159 FETCH(&r[1], 0, CHAN_Y); 2160 FETCH(&r[2], 0, CHAN_Z); 2161 2162 if (modifier == TEX_MODIFIER_PROJECTED) { 2163 micro_div(&r[0], &r[0], &r[3]); 2164 micro_div(&r[1], &r[1], &r[3]); 2165 micro_div(&r[2], &r[2], &r[3]); 2166 } 2167 2168 fetch_texel(mach->Samplers[sampler_unit], 2169 &r[0], &r[1], &r[2], lod, 2170 control, 2171 &r[0], &r[1], &r[2], &r[3]); 2172 break; 2173 2174 case TGSI_TEXTURE_SHADOW2D_ARRAY: 2175 FETCH(&r[0], 0, CHAN_X); 2176 FETCH(&r[1], 0, CHAN_Y); 2177 FETCH(&r[2], 0, CHAN_Z); 2178 FETCH(&r[3], 0, CHAN_W); 2179 2180 assert(modifier != TEX_MODIFIER_PROJECTED); 2181 2182 fetch_texel(mach->Samplers[sampler_unit], 2183 &r[0], &r[1], &r[2], &r[3], 2184 control, 2185 &r[0], &r[1], &r[2], &r[3]); 2186 break; 2187 2188 default: 2189 assert(0); 2190 } 2191 2192 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2193 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2194 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2195 } 2196 } 2197} 2198 2199static void 2200exec_sample_d(struct tgsi_exec_machine *mach, 2201 const struct tgsi_full_instruction *inst) 2202{ 2203 const uint resource_unit = inst->Src[1].Register.Index; 2204 const uint sampler_unit = inst->Src[2].Register.Index; 2205 union tgsi_exec_channel r[4]; 2206 uint chan; 2207 /* 2208 * XXX: This is fake SAMPLE_D -- the derivatives are not taken into account, yet. 2209 */ 2210 2211 switch (mach->Resources[resource_unit].Resource) { 2212 case TGSI_TEXTURE_1D: 2213 case TGSI_TEXTURE_SHADOW1D: 2214 2215 FETCH(&r[0], 0, CHAN_X); 2216 2217 fetch_texel(mach->Samplers[sampler_unit], 2218 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */ 2219 tgsi_sampler_lod_bias, 2220 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 2221 break; 2222 2223 case TGSI_TEXTURE_2D: 2224 case TGSI_TEXTURE_RECT: 2225 case TGSI_TEXTURE_SHADOW2D: 2226 case TGSI_TEXTURE_SHADOWRECT: 2227 2228 FETCH(&r[0], 0, CHAN_X); 2229 FETCH(&r[1], 0, CHAN_Y); 2230 FETCH(&r[2], 0, CHAN_Z); 2231 2232 fetch_texel(mach->Samplers[sampler_unit], 2233 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */ 2234 tgsi_sampler_lod_bias, 2235 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 2236 break; 2237 2238 case TGSI_TEXTURE_3D: 2239 case TGSI_TEXTURE_CUBE: 2240 2241 FETCH(&r[0], 0, CHAN_X); 2242 FETCH(&r[1], 0, CHAN_Y); 2243 FETCH(&r[2], 0, CHAN_Z); 2244 2245 fetch_texel(mach->Samplers[sampler_unit], 2246 &r[0], &r[1], &r[2], &ZeroVec, 2247 tgsi_sampler_lod_bias, 2248 &r[0], &r[1], &r[2], &r[3]); 2249 break; 2250 2251 default: 2252 assert(0); 2253 } 2254 2255 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2256 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2257 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2258 } 2259 } 2260} 2261 2262 2263/** 2264 * Evaluate a constant-valued coefficient at the position of the 2265 * current quad. 2266 */ 2267static void 2268eval_constant_coef( 2269 struct tgsi_exec_machine *mach, 2270 unsigned attrib, 2271 unsigned chan ) 2272{ 2273 unsigned i; 2274 2275 for( i = 0; i < QUAD_SIZE; i++ ) { 2276 mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan]; 2277 } 2278} 2279 2280/** 2281 * Evaluate a linear-valued coefficient at the position of the 2282 * current quad. 2283 */ 2284static void 2285eval_linear_coef( 2286 struct tgsi_exec_machine *mach, 2287 unsigned attrib, 2288 unsigned chan ) 2289{ 2290 const float x = mach->QuadPos.xyzw[0].f[0]; 2291 const float y = mach->QuadPos.xyzw[1].f[0]; 2292 const float dadx = mach->InterpCoefs[attrib].dadx[chan]; 2293 const float dady = mach->InterpCoefs[attrib].dady[chan]; 2294 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; 2295 mach->Inputs[attrib].xyzw[chan].f[0] = a0; 2296 mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx; 2297 mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady; 2298 mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady; 2299} 2300 2301/** 2302 * Evaluate a perspective-valued coefficient at the position of the 2303 * current quad. 2304 */ 2305static void 2306eval_perspective_coef( 2307 struct tgsi_exec_machine *mach, 2308 unsigned attrib, 2309 unsigned chan ) 2310{ 2311 const float x = mach->QuadPos.xyzw[0].f[0]; 2312 const float y = mach->QuadPos.xyzw[1].f[0]; 2313 const float dadx = mach->InterpCoefs[attrib].dadx[chan]; 2314 const float dady = mach->InterpCoefs[attrib].dady[chan]; 2315 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; 2316 const float *w = mach->QuadPos.xyzw[3].f; 2317 /* divide by W here */ 2318 mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0]; 2319 mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1]; 2320 mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2]; 2321 mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3]; 2322} 2323 2324 2325typedef void (* eval_coef_func)( 2326 struct tgsi_exec_machine *mach, 2327 unsigned attrib, 2328 unsigned chan ); 2329 2330static void 2331exec_declaration(struct tgsi_exec_machine *mach, 2332 const struct tgsi_full_declaration *decl) 2333{ 2334 if (decl->Declaration.File == TGSI_FILE_RESOURCE) { 2335 mach->Resources[decl->Range.First] = decl->Resource; 2336 return; 2337 } 2338 2339 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { 2340 if (decl->Declaration.File == TGSI_FILE_INPUT) { 2341 uint first, last, mask; 2342 2343 first = decl->Range.First; 2344 last = decl->Range.Last; 2345 mask = decl->Declaration.UsageMask; 2346 2347 /* XXX we could remove this special-case code since 2348 * mach->InterpCoefs[first].a0 should already have the 2349 * front/back-face value. But we should first update the 2350 * ureg code to emit the right UsageMask value (WRITEMASK_X). 2351 * Then, we could remove the tgsi_exec_machine::Face field. 2352 */ 2353 /* XXX make FACE a system value */ 2354 if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) { 2355 uint i; 2356 2357 assert(decl->Semantic.Index == 0); 2358 assert(first == last); 2359 2360 for (i = 0; i < QUAD_SIZE; i++) { 2361 mach->Inputs[first].xyzw[0].f[i] = mach->Face; 2362 } 2363 } else { 2364 eval_coef_func eval; 2365 uint i, j; 2366 2367 switch (decl->Declaration.Interpolate) { 2368 case TGSI_INTERPOLATE_CONSTANT: 2369 eval = eval_constant_coef; 2370 break; 2371 2372 case TGSI_INTERPOLATE_LINEAR: 2373 eval = eval_linear_coef; 2374 break; 2375 2376 case TGSI_INTERPOLATE_PERSPECTIVE: 2377 eval = eval_perspective_coef; 2378 break; 2379 2380 case TGSI_INTERPOLATE_COLOR: 2381 eval = mach->flatshade_color ? eval_constant_coef : eval_perspective_coef; 2382 break; 2383 2384 default: 2385 assert(0); 2386 return; 2387 } 2388 2389 for (j = 0; j < NUM_CHANNELS; j++) { 2390 if (mask & (1 << j)) { 2391 for (i = first; i <= last; i++) { 2392 eval(mach, i, j); 2393 } 2394 } 2395 } 2396 } 2397 } 2398 } 2399 2400 if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) { 2401 mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First; 2402 } 2403} 2404 2405 2406typedef void (* micro_op)(union tgsi_exec_channel *dst); 2407 2408static void 2409exec_vector(struct tgsi_exec_machine *mach, 2410 const struct tgsi_full_instruction *inst, 2411 micro_op op, 2412 enum tgsi_exec_datatype dst_datatype) 2413{ 2414 unsigned int chan; 2415 2416 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2417 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2418 union tgsi_exec_channel dst; 2419 2420 op(&dst); 2421 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2422 } 2423 } 2424} 2425 2426typedef void (* micro_unary_op)(union tgsi_exec_channel *dst, 2427 const union tgsi_exec_channel *src); 2428 2429static void 2430exec_scalar_unary(struct tgsi_exec_machine *mach, 2431 const struct tgsi_full_instruction *inst, 2432 micro_unary_op op, 2433 enum tgsi_exec_datatype dst_datatype, 2434 enum tgsi_exec_datatype src_datatype) 2435{ 2436 unsigned int chan; 2437 union tgsi_exec_channel src; 2438 union tgsi_exec_channel dst; 2439 2440 fetch_source(mach, &src, &inst->Src[0], CHAN_X, src_datatype); 2441 op(&dst, &src); 2442 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2443 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2444 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2445 } 2446 } 2447} 2448 2449static void 2450exec_vector_unary(struct tgsi_exec_machine *mach, 2451 const struct tgsi_full_instruction *inst, 2452 micro_unary_op op, 2453 enum tgsi_exec_datatype dst_datatype, 2454 enum tgsi_exec_datatype src_datatype) 2455{ 2456 unsigned int chan; 2457 struct tgsi_exec_vector dst; 2458 2459 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2460 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2461 union tgsi_exec_channel src; 2462 2463 fetch_source(mach, &src, &inst->Src[0], chan, src_datatype); 2464 op(&dst.xyzw[chan], &src); 2465 } 2466 } 2467 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2468 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2469 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2470 } 2471 } 2472} 2473 2474typedef void (* micro_binary_op)(union tgsi_exec_channel *dst, 2475 const union tgsi_exec_channel *src0, 2476 const union tgsi_exec_channel *src1); 2477 2478static void 2479exec_scalar_binary(struct tgsi_exec_machine *mach, 2480 const struct tgsi_full_instruction *inst, 2481 micro_binary_op op, 2482 enum tgsi_exec_datatype dst_datatype, 2483 enum tgsi_exec_datatype src_datatype) 2484{ 2485 unsigned int chan; 2486 union tgsi_exec_channel src[2]; 2487 union tgsi_exec_channel dst; 2488 2489 fetch_source(mach, &src[0], &inst->Src[0], CHAN_X, src_datatype); 2490 fetch_source(mach, &src[1], &inst->Src[1], CHAN_Y, src_datatype); 2491 op(&dst, &src[0], &src[1]); 2492 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2493 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2494 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2495 } 2496 } 2497} 2498 2499static void 2500exec_vector_binary(struct tgsi_exec_machine *mach, 2501 const struct tgsi_full_instruction *inst, 2502 micro_binary_op op, 2503 enum tgsi_exec_datatype dst_datatype, 2504 enum tgsi_exec_datatype src_datatype) 2505{ 2506 unsigned int chan; 2507 struct tgsi_exec_vector dst; 2508 2509 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2510 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2511 union tgsi_exec_channel src[2]; 2512 2513 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype); 2514 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype); 2515 op(&dst.xyzw[chan], &src[0], &src[1]); 2516 } 2517 } 2518 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2519 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2520 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2521 } 2522 } 2523} 2524 2525typedef void (* micro_trinary_op)(union tgsi_exec_channel *dst, 2526 const union tgsi_exec_channel *src0, 2527 const union tgsi_exec_channel *src1, 2528 const union tgsi_exec_channel *src2); 2529 2530static void 2531exec_vector_trinary(struct tgsi_exec_machine *mach, 2532 const struct tgsi_full_instruction *inst, 2533 micro_trinary_op op, 2534 enum tgsi_exec_datatype dst_datatype, 2535 enum tgsi_exec_datatype src_datatype) 2536{ 2537 unsigned int chan; 2538 struct tgsi_exec_vector dst; 2539 2540 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2541 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2542 union tgsi_exec_channel src[3]; 2543 2544 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype); 2545 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype); 2546 fetch_source(mach, &src[2], &inst->Src[2], chan, src_datatype); 2547 op(&dst.xyzw[chan], &src[0], &src[1], &src[2]); 2548 } 2549 } 2550 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2551 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2552 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2553 } 2554 } 2555} 2556 2557static void 2558exec_dp3(struct tgsi_exec_machine *mach, 2559 const struct tgsi_full_instruction *inst) 2560{ 2561 unsigned int chan; 2562 union tgsi_exec_channel arg[3]; 2563 2564 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2565 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2566 micro_mul(&arg[2], &arg[0], &arg[1]); 2567 2568 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) { 2569 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2570 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT); 2571 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2572 } 2573 2574 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2575 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2576 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2577 } 2578 } 2579} 2580 2581static void 2582exec_dp4(struct tgsi_exec_machine *mach, 2583 const struct tgsi_full_instruction *inst) 2584{ 2585 unsigned int chan; 2586 union tgsi_exec_channel arg[3]; 2587 2588 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2589 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2590 micro_mul(&arg[2], &arg[0], &arg[1]); 2591 2592 for (chan = CHAN_Y; chan <= CHAN_W; chan++) { 2593 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2594 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT); 2595 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2596 } 2597 2598 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2599 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2600 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2601 } 2602 } 2603} 2604 2605static void 2606exec_dp2a(struct tgsi_exec_machine *mach, 2607 const struct tgsi_full_instruction *inst) 2608{ 2609 unsigned int chan; 2610 union tgsi_exec_channel arg[3]; 2611 2612 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2613 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2614 micro_mul(&arg[2], &arg[0], &arg[1]); 2615 2616 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2617 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2618 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]); 2619 2620 fetch_source(mach, &arg[1], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2621 micro_add(&arg[0], &arg[0], &arg[1]); 2622 2623 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2624 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2625 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2626 } 2627 } 2628} 2629 2630static void 2631exec_dph(struct tgsi_exec_machine *mach, 2632 const struct tgsi_full_instruction *inst) 2633{ 2634 unsigned int chan; 2635 union tgsi_exec_channel arg[3]; 2636 2637 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2638 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2639 micro_mul(&arg[2], &arg[0], &arg[1]); 2640 2641 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2642 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2643 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2644 2645 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2646 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2647 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]); 2648 2649 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2650 micro_add(&arg[0], &arg[0], &arg[1]); 2651 2652 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2653 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2654 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2655 } 2656 } 2657} 2658 2659static void 2660exec_dp2(struct tgsi_exec_machine *mach, 2661 const struct tgsi_full_instruction *inst) 2662{ 2663 unsigned int chan; 2664 union tgsi_exec_channel arg[3]; 2665 2666 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2667 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2668 micro_mul(&arg[2], &arg[0], &arg[1]); 2669 2670 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2671 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2672 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2673 2674 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2675 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2676 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2677 } 2678 } 2679} 2680 2681static void 2682exec_nrm4(struct tgsi_exec_machine *mach, 2683 const struct tgsi_full_instruction *inst) 2684{ 2685 unsigned int chan; 2686 union tgsi_exec_channel arg[4]; 2687 union tgsi_exec_channel scale; 2688 2689 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2690 micro_mul(&scale, &arg[0], &arg[0]); 2691 2692 for (chan = CHAN_Y; chan <= CHAN_W; chan++) { 2693 union tgsi_exec_channel product; 2694 2695 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2696 micro_mul(&product, &arg[chan], &arg[chan]); 2697 micro_add(&scale, &scale, &product); 2698 } 2699 2700 micro_rsq(&scale, &scale); 2701 2702 for (chan = CHAN_X; chan <= CHAN_W; chan++) { 2703 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2704 micro_mul(&arg[chan], &arg[chan], &scale); 2705 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2706 } 2707 } 2708} 2709 2710static void 2711exec_nrm3(struct tgsi_exec_machine *mach, 2712 const struct tgsi_full_instruction *inst) 2713{ 2714 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) { 2715 unsigned int chan; 2716 union tgsi_exec_channel arg[3]; 2717 union tgsi_exec_channel scale; 2718 2719 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2720 micro_mul(&scale, &arg[0], &arg[0]); 2721 2722 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) { 2723 union tgsi_exec_channel product; 2724 2725 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2726 micro_mul(&product, &arg[chan], &arg[chan]); 2727 micro_add(&scale, &scale, &product); 2728 } 2729 2730 micro_rsq(&scale, &scale); 2731 2732 for (chan = CHAN_X; chan <= CHAN_Z; chan++) { 2733 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2734 micro_mul(&arg[chan], &arg[chan], &scale); 2735 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2736 } 2737 } 2738 } 2739 2740 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2741 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2742 } 2743} 2744 2745static void 2746exec_scs(struct tgsi_exec_machine *mach, 2747 const struct tgsi_full_instruction *inst) 2748{ 2749 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XY) { 2750 union tgsi_exec_channel arg; 2751 union tgsi_exec_channel result; 2752 2753 fetch_source(mach, &arg, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2754 2755 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2756 micro_cos(&result, &arg); 2757 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2758 } 2759 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2760 micro_sin(&result, &arg); 2761 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2762 } 2763 } 2764 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2765 store_dest(mach, &ZeroVec, &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2766 } 2767 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2768 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2769 } 2770} 2771 2772static void 2773exec_x2d(struct tgsi_exec_machine *mach, 2774 const struct tgsi_full_instruction *inst) 2775{ 2776 union tgsi_exec_channel r[4]; 2777 union tgsi_exec_channel d[2]; 2778 2779 fetch_source(mach, &r[0], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2780 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2781 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XZ) { 2782 fetch_source(mach, &r[2], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2783 micro_mul(&r[2], &r[2], &r[0]); 2784 fetch_source(mach, &r[3], &inst->Src[2], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2785 micro_mul(&r[3], &r[3], &r[1]); 2786 micro_add(&r[2], &r[2], &r[3]); 2787 fetch_source(mach, &r[3], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2788 micro_add(&d[0], &r[2], &r[3]); 2789 } 2790 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YW) { 2791 fetch_source(mach, &r[2], &inst->Src[2], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2792 micro_mul(&r[2], &r[2], &r[0]); 2793 fetch_source(mach, &r[3], &inst->Src[2], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2794 micro_mul(&r[3], &r[3], &r[1]); 2795 micro_add(&r[2], &r[2], &r[3]); 2796 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2797 micro_add(&d[1], &r[2], &r[3]); 2798 } 2799 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2800 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2801 } 2802 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2803 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2804 } 2805 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2806 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2807 } 2808 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2809 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2810 } 2811} 2812 2813static void 2814exec_rfl(struct tgsi_exec_machine *mach, 2815 const struct tgsi_full_instruction *inst) 2816{ 2817 union tgsi_exec_channel r[9]; 2818 2819 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) { 2820 /* r0 = dp3(src0, src0) */ 2821 fetch_source(mach, &r[2], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2822 micro_mul(&r[0], &r[2], &r[2]); 2823 fetch_source(mach, &r[4], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2824 micro_mul(&r[8], &r[4], &r[4]); 2825 micro_add(&r[0], &r[0], &r[8]); 2826 fetch_source(mach, &r[6], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2827 micro_mul(&r[8], &r[6], &r[6]); 2828 micro_add(&r[0], &r[0], &r[8]); 2829 2830 /* r1 = dp3(src0, src1) */ 2831 fetch_source(mach, &r[3], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2832 micro_mul(&r[1], &r[2], &r[3]); 2833 fetch_source(mach, &r[5], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2834 micro_mul(&r[8], &r[4], &r[5]); 2835 micro_add(&r[1], &r[1], &r[8]); 2836 fetch_source(mach, &r[7], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2837 micro_mul(&r[8], &r[6], &r[7]); 2838 micro_add(&r[1], &r[1], &r[8]); 2839 2840 /* r1 = 2 * r1 / r0 */ 2841 micro_add(&r[1], &r[1], &r[1]); 2842 micro_div(&r[1], &r[1], &r[0]); 2843 2844 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2845 micro_mul(&r[2], &r[2], &r[1]); 2846 micro_sub(&r[2], &r[2], &r[3]); 2847 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2848 } 2849 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2850 micro_mul(&r[4], &r[4], &r[1]); 2851 micro_sub(&r[4], &r[4], &r[5]); 2852 store_dest(mach, &r[4], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2853 } 2854 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2855 micro_mul(&r[6], &r[6], &r[1]); 2856 micro_sub(&r[6], &r[6], &r[7]); 2857 store_dest(mach, &r[6], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2858 } 2859 } 2860 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2861 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2862 } 2863} 2864 2865static void 2866exec_xpd(struct tgsi_exec_machine *mach, 2867 const struct tgsi_full_instruction *inst) 2868{ 2869 union tgsi_exec_channel r[6]; 2870 union tgsi_exec_channel d[3]; 2871 2872 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2873 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2874 2875 micro_mul(&r[2], &r[0], &r[1]); 2876 2877 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2878 fetch_source(mach, &r[4], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2879 2880 micro_mul(&r[5], &r[3], &r[4] ); 2881 micro_sub(&d[CHAN_X], &r[2], &r[5]); 2882 2883 fetch_source(mach, &r[2], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2884 2885 micro_mul(&r[3], &r[3], &r[2]); 2886 2887 fetch_source(mach, &r[5], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2888 2889 micro_mul(&r[1], &r[1], &r[5]); 2890 micro_sub(&d[CHAN_Y], &r[3], &r[1]); 2891 2892 micro_mul(&r[5], &r[5], &r[4]); 2893 micro_mul(&r[0], &r[0], &r[2]); 2894 micro_sub(&d[CHAN_Z], &r[5], &r[0]); 2895 2896 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2897 store_dest(mach, &d[CHAN_X], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2898 } 2899 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2900 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2901 } 2902 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2903 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2904 } 2905 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2906 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2907 } 2908} 2909 2910static void 2911exec_dst(struct tgsi_exec_machine *mach, 2912 const struct tgsi_full_instruction *inst) 2913{ 2914 union tgsi_exec_channel r[2]; 2915 union tgsi_exec_channel d[4]; 2916 2917 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2918 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2919 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2920 micro_mul(&d[CHAN_Y], &r[0], &r[1]); 2921 } 2922 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2923 fetch_source(mach, &d[CHAN_Z], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2924 } 2925 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2926 fetch_source(mach, &d[CHAN_W], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2927 } 2928 2929 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2930 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2931 } 2932 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2933 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2934 } 2935 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2936 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2937 } 2938 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2939 store_dest(mach, &d[CHAN_W], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2940 } 2941} 2942 2943static void 2944exec_log(struct tgsi_exec_machine *mach, 2945 const struct tgsi_full_instruction *inst) 2946{ 2947 union tgsi_exec_channel r[3]; 2948 2949 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2950 micro_abs(&r[2], &r[0]); /* r2 = abs(r0) */ 2951 micro_lg2(&r[1], &r[2]); /* r1 = lg2(r2) */ 2952 micro_flr(&r[0], &r[1]); /* r0 = floor(r1) */ 2953 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2954 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2955 } 2956 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2957 micro_exp2(&r[0], &r[0]); /* r0 = 2 ^ r0 */ 2958 micro_div(&r[0], &r[2], &r[0]); /* r0 = r2 / r0 */ 2959 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2960 } 2961 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2962 store_dest(mach, &r[1], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2963 } 2964 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2965 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2966 } 2967} 2968 2969static void 2970exec_exp(struct tgsi_exec_machine *mach, 2971 const struct tgsi_full_instruction *inst) 2972{ 2973 union tgsi_exec_channel r[3]; 2974 2975 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2976 micro_flr(&r[1], &r[0]); /* r1 = floor(r0) */ 2977 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2978 micro_exp2(&r[2], &r[1]); /* r2 = 2 ^ r1 */ 2979 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2980 } 2981 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2982 micro_sub(&r[2], &r[0], &r[1]); /* r2 = r0 - r1 */ 2983 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2984 } 2985 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2986 micro_exp2(&r[2], &r[0]); /* r2 = 2 ^ r0 */ 2987 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2988 } 2989 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2990 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2991 } 2992} 2993 2994static void 2995exec_lit(struct tgsi_exec_machine *mach, 2996 const struct tgsi_full_instruction *inst) 2997{ 2998 union tgsi_exec_channel r[3]; 2999 union tgsi_exec_channel d[3]; 3000 3001 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YZ) { 3002 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 3003 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 3004 fetch_source(mach, &r[1], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 3005 micro_max(&r[1], &r[1], &ZeroVec); 3006 3007 fetch_source(mach, &r[2], &inst->Src[0], CHAN_W, TGSI_EXEC_DATA_FLOAT); 3008 micro_min(&r[2], &r[2], &P128Vec); 3009 micro_max(&r[2], &r[2], &M128Vec); 3010 micro_pow(&r[1], &r[1], &r[2]); 3011 micro_lt(&d[CHAN_Z], &ZeroVec, &r[0], &r[1], &ZeroVec); 3012 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 3013 } 3014 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 3015 micro_max(&d[CHAN_Y], &r[0], &ZeroVec); 3016 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 3017 } 3018 } 3019 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 3020 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 3021 } 3022 3023 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 3024 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 3025 } 3026} 3027 3028static void 3029exec_break(struct tgsi_exec_machine *mach) 3030{ 3031 if (mach->BreakType == TGSI_EXEC_BREAK_INSIDE_LOOP) { 3032 /* turn off loop channels for each enabled exec channel */ 3033 mach->LoopMask &= ~mach->ExecMask; 3034 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 3035 UPDATE_EXEC_MASK(mach); 3036 } else { 3037 assert(mach->BreakType == TGSI_EXEC_BREAK_INSIDE_SWITCH); 3038 3039 mach->Switch.mask = 0x0; 3040 3041 UPDATE_EXEC_MASK(mach); 3042 } 3043} 3044 3045static void 3046exec_switch(struct tgsi_exec_machine *mach, 3047 const struct tgsi_full_instruction *inst) 3048{ 3049 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING); 3050 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3051 3052 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch; 3053 fetch_source(mach, &mach->Switch.selector, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT); 3054 mach->Switch.mask = 0x0; 3055 mach->Switch.defaultMask = 0x0; 3056 3057 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3058 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_SWITCH; 3059 3060 UPDATE_EXEC_MASK(mach); 3061} 3062 3063static void 3064exec_case(struct tgsi_exec_machine *mach, 3065 const struct tgsi_full_instruction *inst) 3066{ 3067 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask; 3068 union tgsi_exec_channel src; 3069 uint mask = 0; 3070 3071 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT); 3072 3073 if (mach->Switch.selector.u[0] == src.u[0]) { 3074 mask |= 0x1; 3075 } 3076 if (mach->Switch.selector.u[1] == src.u[1]) { 3077 mask |= 0x2; 3078 } 3079 if (mach->Switch.selector.u[2] == src.u[2]) { 3080 mask |= 0x4; 3081 } 3082 if (mach->Switch.selector.u[3] == src.u[3]) { 3083 mask |= 0x8; 3084 } 3085 3086 mach->Switch.defaultMask |= mask; 3087 3088 mach->Switch.mask |= mask & prevMask; 3089 3090 UPDATE_EXEC_MASK(mach); 3091} 3092 3093static void 3094exec_default(struct tgsi_exec_machine *mach) 3095{ 3096 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask; 3097 3098 mach->Switch.mask |= ~mach->Switch.defaultMask & prevMask; 3099 3100 UPDATE_EXEC_MASK(mach); 3101} 3102 3103static void 3104exec_endswitch(struct tgsi_exec_machine *mach) 3105{ 3106 mach->Switch = mach->SwitchStack[--mach->SwitchStackTop]; 3107 mach->BreakType = mach->BreakStack[--mach->BreakStackTop]; 3108 3109 UPDATE_EXEC_MASK(mach); 3110} 3111 3112static void 3113micro_i2f(union tgsi_exec_channel *dst, 3114 const union tgsi_exec_channel *src) 3115{ 3116 dst->f[0] = (float)src->i[0]; 3117 dst->f[1] = (float)src->i[1]; 3118 dst->f[2] = (float)src->i[2]; 3119 dst->f[3] = (float)src->i[3]; 3120} 3121 3122static void 3123micro_not(union tgsi_exec_channel *dst, 3124 const union tgsi_exec_channel *src) 3125{ 3126 dst->u[0] = ~src->u[0]; 3127 dst->u[1] = ~src->u[1]; 3128 dst->u[2] = ~src->u[2]; 3129 dst->u[3] = ~src->u[3]; 3130} 3131 3132static void 3133micro_shl(union tgsi_exec_channel *dst, 3134 const union tgsi_exec_channel *src0, 3135 const union tgsi_exec_channel *src1) 3136{ 3137 dst->u[0] = src0->u[0] << src1->u[0]; 3138 dst->u[1] = src0->u[1] << src1->u[1]; 3139 dst->u[2] = src0->u[2] << src1->u[2]; 3140 dst->u[3] = src0->u[3] << src1->u[3]; 3141} 3142 3143static void 3144micro_and(union tgsi_exec_channel *dst, 3145 const union tgsi_exec_channel *src0, 3146 const union tgsi_exec_channel *src1) 3147{ 3148 dst->u[0] = src0->u[0] & src1->u[0]; 3149 dst->u[1] = src0->u[1] & src1->u[1]; 3150 dst->u[2] = src0->u[2] & src1->u[2]; 3151 dst->u[3] = src0->u[3] & src1->u[3]; 3152} 3153 3154static void 3155micro_or(union tgsi_exec_channel *dst, 3156 const union tgsi_exec_channel *src0, 3157 const union tgsi_exec_channel *src1) 3158{ 3159 dst->u[0] = src0->u[0] | src1->u[0]; 3160 dst->u[1] = src0->u[1] | src1->u[1]; 3161 dst->u[2] = src0->u[2] | src1->u[2]; 3162 dst->u[3] = src0->u[3] | src1->u[3]; 3163} 3164 3165static void 3166micro_xor(union tgsi_exec_channel *dst, 3167 const union tgsi_exec_channel *src0, 3168 const union tgsi_exec_channel *src1) 3169{ 3170 dst->u[0] = src0->u[0] ^ src1->u[0]; 3171 dst->u[1] = src0->u[1] ^ src1->u[1]; 3172 dst->u[2] = src0->u[2] ^ src1->u[2]; 3173 dst->u[3] = src0->u[3] ^ src1->u[3]; 3174} 3175 3176static void 3177micro_mod(union tgsi_exec_channel *dst, 3178 const union tgsi_exec_channel *src0, 3179 const union tgsi_exec_channel *src1) 3180{ 3181 dst->i[0] = src0->i[0] % src1->i[0]; 3182 dst->i[1] = src0->i[1] % src1->i[1]; 3183 dst->i[2] = src0->i[2] % src1->i[2]; 3184 dst->i[3] = src0->i[3] % src1->i[3]; 3185} 3186 3187static void 3188micro_f2i(union tgsi_exec_channel *dst, 3189 const union tgsi_exec_channel *src) 3190{ 3191 dst->i[0] = (int)src->f[0]; 3192 dst->i[1] = (int)src->f[1]; 3193 dst->i[2] = (int)src->f[2]; 3194 dst->i[3] = (int)src->f[3]; 3195} 3196 3197static void 3198micro_idiv(union tgsi_exec_channel *dst, 3199 const union tgsi_exec_channel *src0, 3200 const union tgsi_exec_channel *src1) 3201{ 3202 dst->i[0] = src0->i[0] / src1->i[0]; 3203 dst->i[1] = src0->i[1] / src1->i[1]; 3204 dst->i[2] = src0->i[2] / src1->i[2]; 3205 dst->i[3] = src0->i[3] / src1->i[3]; 3206} 3207 3208static void 3209micro_imax(union tgsi_exec_channel *dst, 3210 const union tgsi_exec_channel *src0, 3211 const union tgsi_exec_channel *src1) 3212{ 3213 dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0]; 3214 dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1]; 3215 dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2]; 3216 dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3]; 3217} 3218 3219static void 3220micro_imin(union tgsi_exec_channel *dst, 3221 const union tgsi_exec_channel *src0, 3222 const union tgsi_exec_channel *src1) 3223{ 3224 dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0]; 3225 dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1]; 3226 dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2]; 3227 dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3]; 3228} 3229 3230static void 3231micro_isge(union tgsi_exec_channel *dst, 3232 const union tgsi_exec_channel *src0, 3233 const union tgsi_exec_channel *src1) 3234{ 3235 dst->i[0] = src0->i[0] >= src1->i[0] ? -1 : 0; 3236 dst->i[1] = src0->i[1] >= src1->i[1] ? -1 : 0; 3237 dst->i[2] = src0->i[2] >= src1->i[2] ? -1 : 0; 3238 dst->i[3] = src0->i[3] >= src1->i[3] ? -1 : 0; 3239} 3240 3241static void 3242micro_ishr(union tgsi_exec_channel *dst, 3243 const union tgsi_exec_channel *src0, 3244 const union tgsi_exec_channel *src1) 3245{ 3246 dst->i[0] = src0->i[0] >> src1->i[0]; 3247 dst->i[1] = src0->i[1] >> src1->i[1]; 3248 dst->i[2] = src0->i[2] >> src1->i[2]; 3249 dst->i[3] = src0->i[3] >> src1->i[3]; 3250} 3251 3252static void 3253micro_islt(union tgsi_exec_channel *dst, 3254 const union tgsi_exec_channel *src0, 3255 const union tgsi_exec_channel *src1) 3256{ 3257 dst->i[0] = src0->i[0] < src1->i[0] ? -1 : 0; 3258 dst->i[1] = src0->i[1] < src1->i[1] ? -1 : 0; 3259 dst->i[2] = src0->i[2] < src1->i[2] ? -1 : 0; 3260 dst->i[3] = src0->i[3] < src1->i[3] ? -1 : 0; 3261} 3262 3263static void 3264micro_f2u(union tgsi_exec_channel *dst, 3265 const union tgsi_exec_channel *src) 3266{ 3267 dst->u[0] = (uint)src->f[0]; 3268 dst->u[1] = (uint)src->f[1]; 3269 dst->u[2] = (uint)src->f[2]; 3270 dst->u[3] = (uint)src->f[3]; 3271} 3272 3273static void 3274micro_u2f(union tgsi_exec_channel *dst, 3275 const union tgsi_exec_channel *src) 3276{ 3277 dst->f[0] = (float)src->u[0]; 3278 dst->f[1] = (float)src->u[1]; 3279 dst->f[2] = (float)src->u[2]; 3280 dst->f[3] = (float)src->u[3]; 3281} 3282 3283static void 3284micro_uadd(union tgsi_exec_channel *dst, 3285 const union tgsi_exec_channel *src0, 3286 const union tgsi_exec_channel *src1) 3287{ 3288 dst->u[0] = src0->u[0] + src1->u[0]; 3289 dst->u[1] = src0->u[1] + src1->u[1]; 3290 dst->u[2] = src0->u[2] + src1->u[2]; 3291 dst->u[3] = src0->u[3] + src1->u[3]; 3292} 3293 3294static void 3295micro_udiv(union tgsi_exec_channel *dst, 3296 const union tgsi_exec_channel *src0, 3297 const union tgsi_exec_channel *src1) 3298{ 3299 dst->u[0] = src0->u[0] / src1->u[0]; 3300 dst->u[1] = src0->u[1] / src1->u[1]; 3301 dst->u[2] = src0->u[2] / src1->u[2]; 3302 dst->u[3] = src0->u[3] / src1->u[3]; 3303} 3304 3305static void 3306micro_umad(union tgsi_exec_channel *dst, 3307 const union tgsi_exec_channel *src0, 3308 const union tgsi_exec_channel *src1, 3309 const union tgsi_exec_channel *src2) 3310{ 3311 dst->u[0] = src0->u[0] * src1->u[0] + src2->u[0]; 3312 dst->u[1] = src0->u[1] * src1->u[1] + src2->u[1]; 3313 dst->u[2] = src0->u[2] * src1->u[2] + src2->u[2]; 3314 dst->u[3] = src0->u[3] * src1->u[3] + src2->u[3]; 3315} 3316 3317static void 3318micro_umax(union tgsi_exec_channel *dst, 3319 const union tgsi_exec_channel *src0, 3320 const union tgsi_exec_channel *src1) 3321{ 3322 dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0]; 3323 dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1]; 3324 dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2]; 3325 dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3]; 3326} 3327 3328static void 3329micro_umin(union tgsi_exec_channel *dst, 3330 const union tgsi_exec_channel *src0, 3331 const union tgsi_exec_channel *src1) 3332{ 3333 dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0]; 3334 dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1]; 3335 dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2]; 3336 dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3]; 3337} 3338 3339static void 3340micro_umod(union tgsi_exec_channel *dst, 3341 const union tgsi_exec_channel *src0, 3342 const union tgsi_exec_channel *src1) 3343{ 3344 dst->u[0] = src0->u[0] % src1->u[0]; 3345 dst->u[1] = src0->u[1] % src1->u[1]; 3346 dst->u[2] = src0->u[2] % src1->u[2]; 3347 dst->u[3] = src0->u[3] % src1->u[3]; 3348} 3349 3350static void 3351micro_umul(union tgsi_exec_channel *dst, 3352 const union tgsi_exec_channel *src0, 3353 const union tgsi_exec_channel *src1) 3354{ 3355 dst->u[0] = src0->u[0] * src1->u[0]; 3356 dst->u[1] = src0->u[1] * src1->u[1]; 3357 dst->u[2] = src0->u[2] * src1->u[2]; 3358 dst->u[3] = src0->u[3] * src1->u[3]; 3359} 3360 3361static void 3362micro_useq(union tgsi_exec_channel *dst, 3363 const union tgsi_exec_channel *src0, 3364 const union tgsi_exec_channel *src1) 3365{ 3366 dst->u[0] = src0->u[0] == src1->u[0] ? ~0 : 0; 3367 dst->u[1] = src0->u[1] == src1->u[1] ? ~0 : 0; 3368 dst->u[2] = src0->u[2] == src1->u[2] ? ~0 : 0; 3369 dst->u[3] = src0->u[3] == src1->u[3] ? ~0 : 0; 3370} 3371 3372static void 3373micro_usge(union tgsi_exec_channel *dst, 3374 const union tgsi_exec_channel *src0, 3375 const union tgsi_exec_channel *src1) 3376{ 3377 dst->u[0] = src0->u[0] >= src1->u[0] ? ~0 : 0; 3378 dst->u[1] = src0->u[1] >= src1->u[1] ? ~0 : 0; 3379 dst->u[2] = src0->u[2] >= src1->u[2] ? ~0 : 0; 3380 dst->u[3] = src0->u[3] >= src1->u[3] ? ~0 : 0; 3381} 3382 3383static void 3384micro_ushr(union tgsi_exec_channel *dst, 3385 const union tgsi_exec_channel *src0, 3386 const union tgsi_exec_channel *src1) 3387{ 3388 dst->u[0] = src0->u[0] >> src1->u[0]; 3389 dst->u[1] = src0->u[1] >> src1->u[1]; 3390 dst->u[2] = src0->u[2] >> src1->u[2]; 3391 dst->u[3] = src0->u[3] >> src1->u[3]; 3392} 3393 3394static void 3395micro_uslt(union tgsi_exec_channel *dst, 3396 const union tgsi_exec_channel *src0, 3397 const union tgsi_exec_channel *src1) 3398{ 3399 dst->u[0] = src0->u[0] < src1->u[0] ? ~0 : 0; 3400 dst->u[1] = src0->u[1] < src1->u[1] ? ~0 : 0; 3401 dst->u[2] = src0->u[2] < src1->u[2] ? ~0 : 0; 3402 dst->u[3] = src0->u[3] < src1->u[3] ? ~0 : 0; 3403} 3404 3405static void 3406micro_usne(union tgsi_exec_channel *dst, 3407 const union tgsi_exec_channel *src0, 3408 const union tgsi_exec_channel *src1) 3409{ 3410 dst->u[0] = src0->u[0] != src1->u[0] ? ~0 : 0; 3411 dst->u[1] = src0->u[1] != src1->u[1] ? ~0 : 0; 3412 dst->u[2] = src0->u[2] != src1->u[2] ? ~0 : 0; 3413 dst->u[3] = src0->u[3] != src1->u[3] ? ~0 : 0; 3414} 3415 3416static void 3417micro_uarl(union tgsi_exec_channel *dst, 3418 const union tgsi_exec_channel *src) 3419{ 3420 dst->i[0] = src->u[0]; 3421 dst->i[1] = src->u[1]; 3422 dst->i[2] = src->u[2]; 3423 dst->i[3] = src->u[3]; 3424} 3425 3426static void 3427micro_ucmp(union tgsi_exec_channel *dst, 3428 const union tgsi_exec_channel *src0, 3429 const union tgsi_exec_channel *src1, 3430 const union tgsi_exec_channel *src2) 3431{ 3432 dst->u[0] = src0->u[0] ? src1->u[0] : src2->u[0]; 3433 dst->u[1] = src0->u[1] ? src1->u[1] : src2->u[1]; 3434 dst->u[2] = src0->u[2] ? src1->u[2] : src2->u[2]; 3435 dst->u[3] = src0->u[3] ? src1->u[3] : src2->u[3]; 3436} 3437 3438static void 3439exec_instruction( 3440 struct tgsi_exec_machine *mach, 3441 const struct tgsi_full_instruction *inst, 3442 int *pc ) 3443{ 3444 union tgsi_exec_channel r[10]; 3445 3446 (*pc)++; 3447 3448 switch (inst->Instruction.Opcode) { 3449 case TGSI_OPCODE_ARL: 3450 exec_vector_unary(mach, inst, micro_arl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3451 break; 3452 3453 case TGSI_OPCODE_MOV: 3454 exec_vector_unary(mach, inst, micro_mov, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT); 3455 break; 3456 3457 case TGSI_OPCODE_LIT: 3458 exec_lit(mach, inst); 3459 break; 3460 3461 case TGSI_OPCODE_RCP: 3462 exec_scalar_unary(mach, inst, micro_rcp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3463 break; 3464 3465 case TGSI_OPCODE_RSQ: 3466 exec_scalar_unary(mach, inst, micro_rsq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3467 break; 3468 3469 case TGSI_OPCODE_EXP: 3470 exec_exp(mach, inst); 3471 break; 3472 3473 case TGSI_OPCODE_LOG: 3474 exec_log(mach, inst); 3475 break; 3476 3477 case TGSI_OPCODE_MUL: 3478 exec_vector_binary(mach, inst, micro_mul, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3479 break; 3480 3481 case TGSI_OPCODE_ADD: 3482 exec_vector_binary(mach, inst, micro_add, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3483 break; 3484 3485 case TGSI_OPCODE_DP3: 3486 exec_dp3(mach, inst); 3487 break; 3488 3489 case TGSI_OPCODE_DP4: 3490 exec_dp4(mach, inst); 3491 break; 3492 3493 case TGSI_OPCODE_DST: 3494 exec_dst(mach, inst); 3495 break; 3496 3497 case TGSI_OPCODE_MIN: 3498 exec_vector_binary(mach, inst, micro_min, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3499 break; 3500 3501 case TGSI_OPCODE_MAX: 3502 exec_vector_binary(mach, inst, micro_max, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3503 break; 3504 3505 case TGSI_OPCODE_SLT: 3506 exec_vector_binary(mach, inst, micro_slt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3507 break; 3508 3509 case TGSI_OPCODE_SGE: 3510 exec_vector_binary(mach, inst, micro_sge, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3511 break; 3512 3513 case TGSI_OPCODE_MAD: 3514 exec_vector_trinary(mach, inst, micro_mad, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3515 break; 3516 3517 case TGSI_OPCODE_SUB: 3518 exec_vector_binary(mach, inst, micro_sub, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3519 break; 3520 3521 case TGSI_OPCODE_LRP: 3522 exec_vector_trinary(mach, inst, micro_lrp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3523 break; 3524 3525 case TGSI_OPCODE_CND: 3526 exec_vector_trinary(mach, inst, micro_cnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3527 break; 3528 3529 case TGSI_OPCODE_DP2A: 3530 exec_dp2a(mach, inst); 3531 break; 3532 3533 case TGSI_OPCODE_FRC: 3534 exec_vector_unary(mach, inst, micro_frc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3535 break; 3536 3537 case TGSI_OPCODE_CLAMP: 3538 exec_vector_trinary(mach, inst, micro_clamp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3539 break; 3540 3541 case TGSI_OPCODE_FLR: 3542 exec_vector_unary(mach, inst, micro_flr, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3543 break; 3544 3545 case TGSI_OPCODE_ROUND: 3546 exec_vector_unary(mach, inst, micro_rnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3547 break; 3548 3549 case TGSI_OPCODE_EX2: 3550 exec_scalar_unary(mach, inst, micro_exp2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3551 break; 3552 3553 case TGSI_OPCODE_LG2: 3554 exec_scalar_unary(mach, inst, micro_lg2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3555 break; 3556 3557 case TGSI_OPCODE_POW: 3558 exec_scalar_binary(mach, inst, micro_pow, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3559 break; 3560 3561 case TGSI_OPCODE_XPD: 3562 exec_xpd(mach, inst); 3563 break; 3564 3565 case TGSI_OPCODE_ABS: 3566 exec_vector_unary(mach, inst, micro_abs, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3567 break; 3568 3569 case TGSI_OPCODE_RCC: 3570 exec_scalar_unary(mach, inst, micro_rcc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3571 break; 3572 3573 case TGSI_OPCODE_DPH: 3574 exec_dph(mach, inst); 3575 break; 3576 3577 case TGSI_OPCODE_COS: 3578 exec_scalar_unary(mach, inst, micro_cos, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3579 break; 3580 3581 case TGSI_OPCODE_DDX: 3582 exec_vector_unary(mach, inst, micro_ddx, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3583 break; 3584 3585 case TGSI_OPCODE_DDY: 3586 exec_vector_unary(mach, inst, micro_ddy, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3587 break; 3588 3589 case TGSI_OPCODE_KILP: 3590 exec_kilp (mach, inst); 3591 break; 3592 3593 case TGSI_OPCODE_KIL: 3594 exec_kil (mach, inst); 3595 break; 3596 3597 case TGSI_OPCODE_PK2H: 3598 assert (0); 3599 break; 3600 3601 case TGSI_OPCODE_PK2US: 3602 assert (0); 3603 break; 3604 3605 case TGSI_OPCODE_PK4B: 3606 assert (0); 3607 break; 3608 3609 case TGSI_OPCODE_PK4UB: 3610 assert (0); 3611 break; 3612 3613 case TGSI_OPCODE_RFL: 3614 exec_rfl(mach, inst); 3615 break; 3616 3617 case TGSI_OPCODE_SEQ: 3618 exec_vector_binary(mach, inst, micro_seq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3619 break; 3620 3621 case TGSI_OPCODE_SFL: 3622 exec_vector(mach, inst, micro_sfl, TGSI_EXEC_DATA_FLOAT); 3623 break; 3624 3625 case TGSI_OPCODE_SGT: 3626 exec_vector_binary(mach, inst, micro_sgt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3627 break; 3628 3629 case TGSI_OPCODE_SIN: 3630 exec_scalar_unary(mach, inst, micro_sin, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3631 break; 3632 3633 case TGSI_OPCODE_SLE: 3634 exec_vector_binary(mach, inst, micro_sle, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3635 break; 3636 3637 case TGSI_OPCODE_SNE: 3638 exec_vector_binary(mach, inst, micro_sne, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3639 break; 3640 3641 case TGSI_OPCODE_STR: 3642 exec_vector(mach, inst, micro_str, TGSI_EXEC_DATA_FLOAT); 3643 break; 3644 3645 case TGSI_OPCODE_TEX: 3646 /* simple texture lookup */ 3647 /* src[0] = texcoord */ 3648 /* src[1] = sampler unit */ 3649 exec_tex(mach, inst, TEX_MODIFIER_NONE); 3650 break; 3651 3652 case TGSI_OPCODE_TXB: 3653 /* Texture lookup with lod bias */ 3654 /* src[0] = texcoord (src[0].w = LOD bias) */ 3655 /* src[1] = sampler unit */ 3656 exec_tex(mach, inst, TEX_MODIFIER_LOD_BIAS); 3657 break; 3658 3659 case TGSI_OPCODE_TXD: 3660 /* Texture lookup with explict partial derivatives */ 3661 /* src[0] = texcoord */ 3662 /* src[1] = d[strq]/dx */ 3663 /* src[2] = d[strq]/dy */ 3664 /* src[3] = sampler unit */ 3665 exec_txd(mach, inst); 3666 break; 3667 3668 case TGSI_OPCODE_TXL: 3669 /* Texture lookup with explit LOD */ 3670 /* src[0] = texcoord (src[0].w = LOD) */ 3671 /* src[1] = sampler unit */ 3672 exec_tex(mach, inst, TEX_MODIFIER_EXPLICIT_LOD); 3673 break; 3674 3675 case TGSI_OPCODE_TXP: 3676 /* Texture lookup with projection */ 3677 /* src[0] = texcoord (src[0].w = projection) */ 3678 /* src[1] = sampler unit */ 3679 exec_tex(mach, inst, TEX_MODIFIER_PROJECTED); 3680 break; 3681 3682 case TGSI_OPCODE_UP2H: 3683 assert (0); 3684 break; 3685 3686 case TGSI_OPCODE_UP2US: 3687 assert (0); 3688 break; 3689 3690 case TGSI_OPCODE_UP4B: 3691 assert (0); 3692 break; 3693 3694 case TGSI_OPCODE_UP4UB: 3695 assert (0); 3696 break; 3697 3698 case TGSI_OPCODE_X2D: 3699 exec_x2d(mach, inst); 3700 break; 3701 3702 case TGSI_OPCODE_ARA: 3703 assert (0); 3704 break; 3705 3706 case TGSI_OPCODE_ARR: 3707 exec_vector_unary(mach, inst, micro_arr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3708 break; 3709 3710 case TGSI_OPCODE_BRA: 3711 assert (0); 3712 break; 3713 3714 case TGSI_OPCODE_CAL: 3715 /* skip the call if no execution channels are enabled */ 3716 if (mach->ExecMask) { 3717 /* do the call */ 3718 3719 /* First, record the depths of the execution stacks. 3720 * This is important for deeply nested/looped return statements. 3721 * We have to unwind the stacks by the correct amount. For a 3722 * real code generator, we could determine the number of entries 3723 * to pop off each stack with simple static analysis and avoid 3724 * implementing this data structure at run time. 3725 */ 3726 mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop; 3727 mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop; 3728 mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop; 3729 mach->CallStack[mach->CallStackTop].SwitchStackTop = mach->SwitchStackTop; 3730 mach->CallStack[mach->CallStackTop].BreakStackTop = mach->BreakStackTop; 3731 /* note that PC was already incremented above */ 3732 mach->CallStack[mach->CallStackTop].ReturnAddr = *pc; 3733 3734 mach->CallStackTop++; 3735 3736 /* Second, push the Cond, Loop, Cont, Func stacks */ 3737 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); 3738 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3739 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3740 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING); 3741 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3742 assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING); 3743 3744 mach->CondStack[mach->CondStackTop++] = mach->CondMask; 3745 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; 3746 mach->ContStack[mach->ContStackTop++] = mach->ContMask; 3747 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch; 3748 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3749 mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask; 3750 3751 /* Finally, jump to the subroutine */ 3752 *pc = inst->Label.Label; 3753 } 3754 break; 3755 3756 case TGSI_OPCODE_RET: 3757 mach->FuncMask &= ~mach->ExecMask; 3758 UPDATE_EXEC_MASK(mach); 3759 3760 if (mach->FuncMask == 0x0) { 3761 /* really return now (otherwise, keep executing */ 3762 3763 if (mach->CallStackTop == 0) { 3764 /* returning from main() */ 3765 mach->CondStackTop = 0; 3766 mach->LoopStackTop = 0; 3767 *pc = -1; 3768 return; 3769 } 3770 3771 assert(mach->CallStackTop > 0); 3772 mach->CallStackTop--; 3773 3774 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; 3775 mach->CondMask = mach->CondStack[mach->CondStackTop]; 3776 3777 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; 3778 mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; 3779 3780 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; 3781 mach->ContMask = mach->ContStack[mach->ContStackTop]; 3782 3783 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop; 3784 mach->Switch = mach->SwitchStack[mach->SwitchStackTop]; 3785 3786 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop; 3787 mach->BreakType = mach->BreakStack[mach->BreakStackTop]; 3788 3789 assert(mach->FuncStackTop > 0); 3790 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; 3791 3792 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; 3793 3794 UPDATE_EXEC_MASK(mach); 3795 } 3796 break; 3797 3798 case TGSI_OPCODE_SSG: 3799 exec_vector_unary(mach, inst, micro_sgn, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3800 break; 3801 3802 case TGSI_OPCODE_CMP: 3803 exec_vector_trinary(mach, inst, micro_cmp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3804 break; 3805 3806 case TGSI_OPCODE_SCS: 3807 exec_scs(mach, inst); 3808 break; 3809 3810 case TGSI_OPCODE_NRM: 3811 exec_nrm3(mach, inst); 3812 break; 3813 3814 case TGSI_OPCODE_NRM4: 3815 exec_nrm4(mach, inst); 3816 break; 3817 3818 case TGSI_OPCODE_DIV: 3819 exec_vector_binary(mach, inst, micro_div, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3820 break; 3821 3822 case TGSI_OPCODE_DP2: 3823 exec_dp2(mach, inst); 3824 break; 3825 3826 case TGSI_OPCODE_IF: 3827 /* push CondMask */ 3828 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); 3829 mach->CondStack[mach->CondStackTop++] = mach->CondMask; 3830 FETCH( &r[0], 0, CHAN_X ); 3831 /* update CondMask */ 3832 if( ! r[0].u[0] ) { 3833 mach->CondMask &= ~0x1; 3834 } 3835 if( ! r[0].u[1] ) { 3836 mach->CondMask &= ~0x2; 3837 } 3838 if( ! r[0].u[2] ) { 3839 mach->CondMask &= ~0x4; 3840 } 3841 if( ! r[0].u[3] ) { 3842 mach->CondMask &= ~0x8; 3843 } 3844 UPDATE_EXEC_MASK(mach); 3845 /* Todo: If CondMask==0, jump to ELSE */ 3846 break; 3847 3848 case TGSI_OPCODE_ELSE: 3849 /* invert CondMask wrt previous mask */ 3850 { 3851 uint prevMask; 3852 assert(mach->CondStackTop > 0); 3853 prevMask = mach->CondStack[mach->CondStackTop - 1]; 3854 mach->CondMask = ~mach->CondMask & prevMask; 3855 UPDATE_EXEC_MASK(mach); 3856 /* Todo: If CondMask==0, jump to ENDIF */ 3857 } 3858 break; 3859 3860 case TGSI_OPCODE_ENDIF: 3861 /* pop CondMask */ 3862 assert(mach->CondStackTop > 0); 3863 mach->CondMask = mach->CondStack[--mach->CondStackTop]; 3864 UPDATE_EXEC_MASK(mach); 3865 break; 3866 3867 case TGSI_OPCODE_END: 3868 /* make sure we end primitives which haven't 3869 * been explicitly emitted */ 3870 conditional_emit_primitive(mach); 3871 /* halt execution */ 3872 *pc = -1; 3873 break; 3874 3875 case TGSI_OPCODE_PUSHA: 3876 assert (0); 3877 break; 3878 3879 case TGSI_OPCODE_POPA: 3880 assert (0); 3881 break; 3882 3883 case TGSI_OPCODE_CEIL: 3884 exec_vector_unary(mach, inst, micro_ceil, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3885 break; 3886 3887 case TGSI_OPCODE_I2F: 3888 exec_vector_unary(mach, inst, micro_i2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_INT); 3889 break; 3890 3891 case TGSI_OPCODE_NOT: 3892 exec_vector_unary(mach, inst, micro_not, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3893 break; 3894 3895 case TGSI_OPCODE_TRUNC: 3896 exec_vector_unary(mach, inst, micro_trunc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3897 break; 3898 3899 case TGSI_OPCODE_SHL: 3900 exec_vector_binary(mach, inst, micro_shl, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3901 break; 3902 3903 case TGSI_OPCODE_AND: 3904 exec_vector_binary(mach, inst, micro_and, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3905 break; 3906 3907 case TGSI_OPCODE_OR: 3908 exec_vector_binary(mach, inst, micro_or, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3909 break; 3910 3911 case TGSI_OPCODE_MOD: 3912 exec_vector_binary(mach, inst, micro_mod, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3913 break; 3914 3915 case TGSI_OPCODE_XOR: 3916 exec_vector_binary(mach, inst, micro_xor, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3917 break; 3918 3919 case TGSI_OPCODE_SAD: 3920 assert (0); 3921 break; 3922 3923 case TGSI_OPCODE_TXF: 3924 exec_txf(mach, inst); 3925 break; 3926 3927 case TGSI_OPCODE_TXQ: 3928 exec_txq(mach, inst); 3929 break; 3930 3931 case TGSI_OPCODE_EMIT: 3932 emit_vertex(mach); 3933 break; 3934 3935 case TGSI_OPCODE_ENDPRIM: 3936 emit_primitive(mach); 3937 break; 3938 3939 case TGSI_OPCODE_BGNLOOP: 3940 /* push LoopMask and ContMasks */ 3941 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3942 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3943 assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3944 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3945 3946 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; 3947 mach->ContStack[mach->ContStackTop++] = mach->ContMask; 3948 mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1; 3949 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3950 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP; 3951 break; 3952 3953 case TGSI_OPCODE_ENDLOOP: 3954 /* Restore ContMask, but don't pop */ 3955 assert(mach->ContStackTop > 0); 3956 mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; 3957 UPDATE_EXEC_MASK(mach); 3958 if (mach->ExecMask) { 3959 /* repeat loop: jump to instruction just past BGNLOOP */ 3960 assert(mach->LoopLabelStackTop > 0); 3961 *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1; 3962 } 3963 else { 3964 /* exit loop: pop LoopMask */ 3965 assert(mach->LoopStackTop > 0); 3966 mach->LoopMask = mach->LoopStack[--mach->LoopStackTop]; 3967 /* pop ContMask */ 3968 assert(mach->ContStackTop > 0); 3969 mach->ContMask = mach->ContStack[--mach->ContStackTop]; 3970 assert(mach->LoopLabelStackTop > 0); 3971 --mach->LoopLabelStackTop; 3972 3973 mach->BreakType = mach->BreakStack[--mach->BreakStackTop]; 3974 } 3975 UPDATE_EXEC_MASK(mach); 3976 break; 3977 3978 case TGSI_OPCODE_BRK: 3979 exec_break(mach); 3980 break; 3981 3982 case TGSI_OPCODE_CONT: 3983 /* turn off cont channels for each enabled exec channel */ 3984 mach->ContMask &= ~mach->ExecMask; 3985 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 3986 UPDATE_EXEC_MASK(mach); 3987 break; 3988 3989 case TGSI_OPCODE_BGNSUB: 3990 /* no-op */ 3991 break; 3992 3993 case TGSI_OPCODE_ENDSUB: 3994 /* 3995 * XXX: This really should be a no-op. We should never reach this opcode. 3996 */ 3997 3998 assert(mach->CallStackTop > 0); 3999 mach->CallStackTop--; 4000 4001 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; 4002 mach->CondMask = mach->CondStack[mach->CondStackTop]; 4003 4004 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; 4005 mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; 4006 4007 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; 4008 mach->ContMask = mach->ContStack[mach->ContStackTop]; 4009 4010 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop; 4011 mach->Switch = mach->SwitchStack[mach->SwitchStackTop]; 4012 4013 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop; 4014 mach->BreakType = mach->BreakStack[mach->BreakStackTop]; 4015 4016 assert(mach->FuncStackTop > 0); 4017 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; 4018 4019 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; 4020 4021 UPDATE_EXEC_MASK(mach); 4022 break; 4023 4024 case TGSI_OPCODE_NOP: 4025 break; 4026 4027 case TGSI_OPCODE_BREAKC: 4028 FETCH(&r[0], 0, CHAN_X); 4029 /* update CondMask */ 4030 if (r[0].u[0] && (mach->ExecMask & 0x1)) { 4031 mach->LoopMask &= ~0x1; 4032 } 4033 if (r[0].u[1] && (mach->ExecMask & 0x2)) { 4034 mach->LoopMask &= ~0x2; 4035 } 4036 if (r[0].u[2] && (mach->ExecMask & 0x4)) { 4037 mach->LoopMask &= ~0x4; 4038 } 4039 if (r[0].u[3] && (mach->ExecMask & 0x8)) { 4040 mach->LoopMask &= ~0x8; 4041 } 4042 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 4043 UPDATE_EXEC_MASK(mach); 4044 break; 4045 4046 case TGSI_OPCODE_F2I: 4047 exec_vector_unary(mach, inst, micro_f2i, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 4048 break; 4049 4050 case TGSI_OPCODE_IDIV: 4051 exec_vector_binary(mach, inst, micro_idiv, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4052 break; 4053 4054 case TGSI_OPCODE_IMAX: 4055 exec_vector_binary(mach, inst, micro_imax, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4056 break; 4057 4058 case TGSI_OPCODE_IMIN: 4059 exec_vector_binary(mach, inst, micro_imin, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4060 break; 4061 4062 case TGSI_OPCODE_INEG: 4063 exec_vector_unary(mach, inst, micro_ineg, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4064 break; 4065 4066 case TGSI_OPCODE_ISGE: 4067 exec_vector_binary(mach, inst, micro_isge, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4068 break; 4069 4070 case TGSI_OPCODE_ISHR: 4071 exec_vector_binary(mach, inst, micro_ishr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4072 break; 4073 4074 case TGSI_OPCODE_ISLT: 4075 exec_vector_binary(mach, inst, micro_islt, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4076 break; 4077 4078 case TGSI_OPCODE_F2U: 4079 exec_vector_unary(mach, inst, micro_f2u, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT); 4080 break; 4081 4082 case TGSI_OPCODE_U2F: 4083 exec_vector_unary(mach, inst, micro_u2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_UINT); 4084 break; 4085 4086 case TGSI_OPCODE_UADD: 4087 exec_vector_binary(mach, inst, micro_uadd, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4088 break; 4089 4090 case TGSI_OPCODE_UDIV: 4091 exec_vector_binary(mach, inst, micro_udiv, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4092 break; 4093 4094 case TGSI_OPCODE_UMAD: 4095 exec_vector_trinary(mach, inst, micro_umad, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4096 break; 4097 4098 case TGSI_OPCODE_UMAX: 4099 exec_vector_binary(mach, inst, micro_umax, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4100 break; 4101 4102 case TGSI_OPCODE_UMIN: 4103 exec_vector_binary(mach, inst, micro_umin, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4104 break; 4105 4106 case TGSI_OPCODE_UMOD: 4107 exec_vector_binary(mach, inst, micro_umod, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4108 break; 4109 4110 case TGSI_OPCODE_UMUL: 4111 exec_vector_binary(mach, inst, micro_umul, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4112 break; 4113 4114 case TGSI_OPCODE_USEQ: 4115 exec_vector_binary(mach, inst, micro_useq, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4116 break; 4117 4118 case TGSI_OPCODE_USGE: 4119 exec_vector_binary(mach, inst, micro_usge, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4120 break; 4121 4122 case TGSI_OPCODE_USHR: 4123 exec_vector_binary(mach, inst, micro_ushr, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4124 break; 4125 4126 case TGSI_OPCODE_USLT: 4127 exec_vector_binary(mach, inst, micro_uslt, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4128 break; 4129 4130 case TGSI_OPCODE_USNE: 4131 exec_vector_binary(mach, inst, micro_usne, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4132 break; 4133 4134 case TGSI_OPCODE_SWITCH: 4135 exec_switch(mach, inst); 4136 break; 4137 4138 case TGSI_OPCODE_CASE: 4139 exec_case(mach, inst); 4140 break; 4141 4142 case TGSI_OPCODE_DEFAULT: 4143 exec_default(mach); 4144 break; 4145 4146 case TGSI_OPCODE_ENDSWITCH: 4147 exec_endswitch(mach); 4148 break; 4149 4150 case TGSI_OPCODE_LOAD: 4151 assert(0); 4152 break; 4153 4154 case TGSI_OPCODE_LOAD_MS: 4155 assert(0); 4156 break; 4157 4158 case TGSI_OPCODE_SAMPLE: 4159 exec_sample(mach, inst, TEX_MODIFIER_NONE); 4160 break; 4161 4162 case TGSI_OPCODE_SAMPLE_B: 4163 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS); 4164 break; 4165 4166 case TGSI_OPCODE_SAMPLE_C: 4167 exec_sample(mach, inst, TEX_MODIFIER_NONE); 4168 break; 4169 4170 case TGSI_OPCODE_SAMPLE_C_LZ: 4171 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS); 4172 break; 4173 4174 case TGSI_OPCODE_SAMPLE_D: 4175 exec_sample_d(mach, inst); 4176 break; 4177 4178 case TGSI_OPCODE_SAMPLE_L: 4179 exec_sample(mach, inst, TEX_MODIFIER_EXPLICIT_LOD); 4180 break; 4181 4182 case TGSI_OPCODE_GATHER4: 4183 assert(0); 4184 break; 4185 4186 case TGSI_OPCODE_RESINFO: 4187 assert(0); 4188 break; 4189 4190 case TGSI_OPCODE_SAMPLE_POS: 4191 assert(0); 4192 break; 4193 4194 case TGSI_OPCODE_SAMPLE_INFO: 4195 assert(0); 4196 break; 4197 4198 case TGSI_OPCODE_UARL: 4199 exec_vector_unary(mach, inst, micro_uarl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_UINT); 4200 break; 4201 4202 case TGSI_OPCODE_UCMP: 4203 exec_vector_trinary(mach, inst, micro_ucmp, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4204 break; 4205 4206 case TGSI_OPCODE_IABS: 4207 exec_vector_unary(mach, inst, micro_iabs, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4208 break; 4209 4210 default: 4211 assert( 0 ); 4212 } 4213} 4214 4215 4216#define DEBUG_EXECUTION 0 4217 4218 4219/** 4220 * Run TGSI interpreter. 4221 * \return bitmask of "alive" quad components 4222 */ 4223uint 4224tgsi_exec_machine_run( struct tgsi_exec_machine *mach ) 4225{ 4226 uint i; 4227 int pc = 0; 4228 4229 mach->CondMask = 0xf; 4230 mach->LoopMask = 0xf; 4231 mach->ContMask = 0xf; 4232 mach->FuncMask = 0xf; 4233 mach->ExecMask = 0xf; 4234 4235 mach->Switch.mask = 0xf; 4236 4237 assert(mach->CondStackTop == 0); 4238 assert(mach->LoopStackTop == 0); 4239 assert(mach->ContStackTop == 0); 4240 assert(mach->SwitchStackTop == 0); 4241 assert(mach->BreakStackTop == 0); 4242 assert(mach->CallStackTop == 0); 4243 4244 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0; 4245 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0; 4246 4247 if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) { 4248 mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0; 4249 mach->Primitives[0] = 0; 4250 } 4251 4252 /* execute declarations (interpolants) */ 4253 for (i = 0; i < mach->NumDeclarations; i++) { 4254 exec_declaration( mach, mach->Declarations+i ); 4255 } 4256 4257 { 4258#if DEBUG_EXECUTION 4259 struct tgsi_exec_vector temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS]; 4260 struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS]; 4261 uint inst = 1; 4262 4263 memcpy(temps, mach->Temps, sizeof(temps)); 4264 memcpy(outputs, mach->Outputs, sizeof(outputs)); 4265#endif 4266 4267 /* execute instructions, until pc is set to -1 */ 4268 while (pc != -1) { 4269 4270#if DEBUG_EXECUTION 4271 uint i; 4272 4273 tgsi_dump_instruction(&mach->Instructions[pc], inst++); 4274#endif 4275 4276 assert(pc < (int) mach->NumInstructions); 4277 exec_instruction(mach, mach->Instructions + pc, &pc); 4278 4279#if DEBUG_EXECUTION 4280 for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) { 4281 if (memcmp(&temps[i], &mach->Temps[i], sizeof(temps[i]))) { 4282 uint j; 4283 4284 memcpy(&temps[i], &mach->Temps[i], sizeof(temps[i])); 4285 debug_printf("TEMP[%2u] = ", i); 4286 for (j = 0; j < 4; j++) { 4287 if (j > 0) { 4288 debug_printf(" "); 4289 } 4290 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n", 4291 temps[i].xyzw[0].f[j], temps[i].xyzw[0].u[j], 4292 temps[i].xyzw[1].f[j], temps[i].xyzw[1].u[j], 4293 temps[i].xyzw[2].f[j], temps[i].xyzw[2].u[j], 4294 temps[i].xyzw[3].f[j], temps[i].xyzw[3].u[j]); 4295 } 4296 } 4297 } 4298 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { 4299 if (memcmp(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]))) { 4300 uint j; 4301 4302 memcpy(&outputs[i], &mach->Outputs[i], sizeof(outputs[i])); 4303 debug_printf("OUT[%2u] = ", i); 4304 for (j = 0; j < 4; j++) { 4305 if (j > 0) { 4306 debug_printf(" "); 4307 } 4308 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n", 4309 outputs[i].xyzw[0].f[j], outputs[i].xyzw[0].u[j], 4310 outputs[i].xyzw[1].f[j], outputs[i].xyzw[1].u[j], 4311 outputs[i].xyzw[2].f[j], outputs[i].xyzw[2].u[j], 4312 outputs[i].xyzw[3].f[j], outputs[i].xyzw[3].u[j]); 4313 } 4314 } 4315 } 4316#endif 4317 } 4318 } 4319 4320#if 0 4321 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */ 4322 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { 4323 /* 4324 * Scale back depth component. 4325 */ 4326 for (i = 0; i < 4; i++) 4327 mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF; 4328 } 4329#endif 4330 4331 /* Strictly speaking, these assertions aren't really needed but they 4332 * can potentially catch some bugs in the control flow code. 4333 */ 4334 assert(mach->CondStackTop == 0); 4335 assert(mach->LoopStackTop == 0); 4336 assert(mach->ContStackTop == 0); 4337 assert(mach->SwitchStackTop == 0); 4338 assert(mach->BreakStackTop == 0); 4339 assert(mach->CallStackTop == 0); 4340 4341 return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0]; 4342} 4343