tgsi_exec.c revision 265f55e6273aafc8e7607cd70a4b9756f7cb6bff
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 1 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 file, 1041 const uint swizzle, 1042 const union tgsi_exec_channel *index, 1043 const union tgsi_exec_channel *index2D, 1044 union tgsi_exec_channel *chan) 1045{ 1046 uint i; 1047 1048 assert(swizzle < 4); 1049 1050 switch (file) { 1051 case TGSI_FILE_CONSTANT: 1052 for (i = 0; i < QUAD_SIZE; i++) { 1053 assert(index2D->i[i] >= 0 && index2D->i[i] < PIPE_MAX_CONSTANT_BUFFERS); 1054 assert(mach->Consts[index2D->i[i]]); 1055 1056 if (index->i[i] < 0) { 1057 chan->u[i] = 0; 1058 } else { 1059 /* NOTE: copying the const value as a uint instead of float */ 1060 const uint constbuf = index2D->i[i]; 1061 const uint *buf = (const uint *)mach->Consts[constbuf]; 1062 const int pos = index->i[i] * 4 + swizzle; 1063 /* const buffer bounds check */ 1064 if (pos < 0 || pos >= mach->ConstsSize[constbuf]) { 1065 if (0) { 1066 /* Debug: print warning */ 1067 static int count = 0; 1068 if (count++ < 100) 1069 debug_printf("TGSI Exec: const buffer index %d" 1070 " out of bounds\n", pos); 1071 } 1072 chan->u[i] = 0; 1073 } 1074 else 1075 chan->u[i] = buf[pos]; 1076 } 1077 } 1078 break; 1079 1080 case TGSI_FILE_INPUT: 1081 for (i = 0; i < QUAD_SIZE; i++) { 1082 /* 1083 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1084 debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n", 1085 index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i], 1086 index2D->i[i], index->i[i]); 1087 }*/ 1088 int pos = index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i]; 1089 assert(pos >= 0); 1090 assert(pos < TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS); 1091 chan->u[i] = mach->Inputs[pos].xyzw[swizzle].u[i]; 1092 } 1093 break; 1094 1095 case TGSI_FILE_SYSTEM_VALUE: 1096 /* XXX no swizzling at this point. Will be needed if we put 1097 * gl_FragCoord, for example, in a sys value register. 1098 */ 1099 for (i = 0; i < QUAD_SIZE; i++) { 1100 chan->f[i] = mach->SystemValue[index->i[i]][0]; 1101 } 1102 break; 1103 1104 case TGSI_FILE_TEMPORARY: 1105 for (i = 0; i < QUAD_SIZE; i++) { 1106 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); 1107 assert(index2D->i[i] == 0); 1108 1109 chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i]; 1110 } 1111 break; 1112 1113 case TGSI_FILE_TEMPORARY_ARRAY: 1114 for (i = 0; i < QUAD_SIZE; i++) { 1115 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); 1116 assert(index2D->i[i] < TGSI_EXEC_NUM_TEMP_ARRAYS); 1117 1118 chan->u[i] = 1119 mach->TempArray[index2D->i[i]][index->i[i]].xyzw[swizzle].u[i]; 1120 } 1121 break; 1122 1123 case TGSI_FILE_IMMEDIATE: 1124 for (i = 0; i < QUAD_SIZE; i++) { 1125 assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit); 1126 assert(index2D->i[i] == 0); 1127 1128 chan->f[i] = mach->Imms[index->i[i]][swizzle]; 1129 } 1130 break; 1131 1132 case TGSI_FILE_IMMEDIATE_ARRAY: 1133 for (i = 0; i < QUAD_SIZE; i++) { 1134 assert(index2D->i[i] == 0); 1135 1136 chan->f[i] = mach->ImmArray[index->i[i]][swizzle]; 1137 } 1138 break; 1139 1140 case TGSI_FILE_ADDRESS: 1141 for (i = 0; i < QUAD_SIZE; i++) { 1142 assert(index->i[i] >= 0); 1143 assert(index2D->i[i] == 0); 1144 1145 chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i]; 1146 } 1147 break; 1148 1149 case TGSI_FILE_PREDICATE: 1150 for (i = 0; i < QUAD_SIZE; i++) { 1151 assert(index->i[i] >= 0 && index->i[i] < TGSI_EXEC_NUM_PREDS); 1152 assert(index2D->i[i] == 0); 1153 1154 chan->u[i] = mach->Predicates[0].xyzw[swizzle].u[i]; 1155 } 1156 break; 1157 1158 case TGSI_FILE_OUTPUT: 1159 /* vertex/fragment output vars can be read too */ 1160 for (i = 0; i < QUAD_SIZE; i++) { 1161 assert(index->i[i] >= 0); 1162 assert(index2D->i[i] == 0); 1163 1164 chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i]; 1165 } 1166 break; 1167 1168 default: 1169 assert(0); 1170 for (i = 0; i < QUAD_SIZE; i++) { 1171 chan->u[i] = 0; 1172 } 1173 } 1174} 1175 1176static void 1177fetch_source(const struct tgsi_exec_machine *mach, 1178 union tgsi_exec_channel *chan, 1179 const struct tgsi_full_src_register *reg, 1180 const uint chan_index, 1181 enum tgsi_exec_datatype src_datatype) 1182{ 1183 union tgsi_exec_channel index; 1184 union tgsi_exec_channel index2D; 1185 uint swizzle; 1186 1187 /* We start with a direct index into a register file. 1188 * 1189 * file[1], 1190 * where: 1191 * file = Register.File 1192 * [1] = Register.Index 1193 */ 1194 index.i[0] = 1195 index.i[1] = 1196 index.i[2] = 1197 index.i[3] = reg->Register.Index; 1198 1199 /* There is an extra source register that indirectly subscripts 1200 * a register file. The direct index now becomes an offset 1201 * that is being added to the indirect register. 1202 * 1203 * file[ind[2].x+1], 1204 * where: 1205 * ind = Indirect.File 1206 * [2] = Indirect.Index 1207 * .x = Indirect.SwizzleX 1208 */ 1209 if (reg->Register.Indirect) { 1210 union tgsi_exec_channel index2; 1211 union tgsi_exec_channel indir_index; 1212 const uint execmask = mach->ExecMask; 1213 uint i; 1214 1215 /* which address register (always zero now) */ 1216 index2.i[0] = 1217 index2.i[1] = 1218 index2.i[2] = 1219 index2.i[3] = reg->Indirect.Index; 1220 assert(reg->Indirect.File == TGSI_FILE_ADDRESS); 1221 /* get current value of address register[swizzle] */ 1222 swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, CHAN_X ); 1223 fetch_src_file_channel(mach, 1224 reg->Indirect.File, 1225 swizzle, 1226 &index2, 1227 &ZeroVec, 1228 &indir_index); 1229 1230 /* add value of address register to the offset */ 1231 index.i[0] += indir_index.i[0]; 1232 index.i[1] += indir_index.i[1]; 1233 index.i[2] += indir_index.i[2]; 1234 index.i[3] += indir_index.i[3]; 1235 1236 /* for disabled execution channels, zero-out the index to 1237 * avoid using a potential garbage value. 1238 */ 1239 for (i = 0; i < QUAD_SIZE; i++) { 1240 if ((execmask & (1 << i)) == 0) 1241 index.i[i] = 0; 1242 } 1243 } 1244 1245 /* There is an extra source register that is a second 1246 * subscript to a register file. Effectively it means that 1247 * the register file is actually a 2D array of registers. 1248 * 1249 * file[3][1], 1250 * where: 1251 * [3] = Dimension.Index 1252 */ 1253 if (reg->Register.Dimension) { 1254 index2D.i[0] = 1255 index2D.i[1] = 1256 index2D.i[2] = 1257 index2D.i[3] = reg->Dimension.Index; 1258 1259 /* Again, the second subscript index can be addressed indirectly 1260 * identically to the first one. 1261 * Nothing stops us from indirectly addressing the indirect register, 1262 * but there is no need for that, so we won't exercise it. 1263 * 1264 * file[ind[4].y+3][1], 1265 * where: 1266 * ind = DimIndirect.File 1267 * [4] = DimIndirect.Index 1268 * .y = DimIndirect.SwizzleX 1269 */ 1270 if (reg->Dimension.Indirect) { 1271 union tgsi_exec_channel index2; 1272 union tgsi_exec_channel indir_index; 1273 const uint execmask = mach->ExecMask; 1274 uint i; 1275 1276 index2.i[0] = 1277 index2.i[1] = 1278 index2.i[2] = 1279 index2.i[3] = reg->DimIndirect.Index; 1280 1281 swizzle = tgsi_util_get_src_register_swizzle( ®->DimIndirect, CHAN_X ); 1282 fetch_src_file_channel(mach, 1283 reg->DimIndirect.File, 1284 swizzle, 1285 &index2, 1286 &ZeroVec, 1287 &indir_index); 1288 1289 index2D.i[0] += indir_index.i[0]; 1290 index2D.i[1] += indir_index.i[1]; 1291 index2D.i[2] += indir_index.i[2]; 1292 index2D.i[3] += indir_index.i[3]; 1293 1294 /* for disabled execution channels, zero-out the index to 1295 * avoid using a potential garbage value. 1296 */ 1297 for (i = 0; i < QUAD_SIZE; i++) { 1298 if ((execmask & (1 << i)) == 0) { 1299 index2D.i[i] = 0; 1300 } 1301 } 1302 } 1303 1304 /* If by any chance there was a need for a 3D array of register 1305 * files, we would have to check whether Dimension is followed 1306 * by a dimension register and continue the saga. 1307 */ 1308 } else { 1309 index2D.i[0] = 1310 index2D.i[1] = 1311 index2D.i[2] = 1312 index2D.i[3] = 0; 1313 } 1314 1315 swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index ); 1316 fetch_src_file_channel(mach, 1317 reg->Register.File, 1318 swizzle, 1319 &index, 1320 &index2D, 1321 chan); 1322 1323 if (reg->Register.Absolute) { 1324 if (src_datatype == TGSI_EXEC_DATA_FLOAT) { 1325 micro_abs(chan, chan); 1326 } else { 1327 micro_iabs(chan, chan); 1328 } 1329 } 1330 1331 if (reg->Register.Negate) { 1332 if (src_datatype == TGSI_EXEC_DATA_FLOAT) { 1333 micro_neg(chan, chan); 1334 } else { 1335 micro_ineg(chan, chan); 1336 } 1337 } 1338} 1339 1340static void 1341store_dest(struct tgsi_exec_machine *mach, 1342 const union tgsi_exec_channel *chan, 1343 const struct tgsi_full_dst_register *reg, 1344 const struct tgsi_full_instruction *inst, 1345 uint chan_index, 1346 enum tgsi_exec_datatype dst_datatype) 1347{ 1348 uint i; 1349 union tgsi_exec_channel null; 1350 union tgsi_exec_channel *dst; 1351 union tgsi_exec_channel index2D; 1352 uint execmask = mach->ExecMask; 1353 int offset = 0; /* indirection offset */ 1354 int index; 1355 1356 /* for debugging */ 1357 if (0 && dst_datatype == TGSI_EXEC_DATA_FLOAT) { 1358 check_inf_or_nan(chan); 1359 } 1360 1361 /* There is an extra source register that indirectly subscripts 1362 * a register file. The direct index now becomes an offset 1363 * that is being added to the indirect register. 1364 * 1365 * file[ind[2].x+1], 1366 * where: 1367 * ind = Indirect.File 1368 * [2] = Indirect.Index 1369 * .x = Indirect.SwizzleX 1370 */ 1371 if (reg->Register.Indirect) { 1372 union tgsi_exec_channel index; 1373 union tgsi_exec_channel indir_index; 1374 uint swizzle; 1375 1376 /* which address register (always zero for now) */ 1377 index.i[0] = 1378 index.i[1] = 1379 index.i[2] = 1380 index.i[3] = reg->Indirect.Index; 1381 1382 /* get current value of address register[swizzle] */ 1383 swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, CHAN_X ); 1384 1385 /* fetch values from the address/indirection register */ 1386 fetch_src_file_channel(mach, 1387 reg->Indirect.File, 1388 swizzle, 1389 &index, 1390 &ZeroVec, 1391 &indir_index); 1392 1393 /* save indirection offset */ 1394 offset = indir_index.i[0]; 1395 } 1396 1397 /* There is an extra source register that is a second 1398 * subscript to a register file. Effectively it means that 1399 * the register file is actually a 2D array of registers. 1400 * 1401 * file[3][1], 1402 * where: 1403 * [3] = Dimension.Index 1404 */ 1405 if (reg->Register.Dimension) { 1406 index2D.i[0] = 1407 index2D.i[1] = 1408 index2D.i[2] = 1409 index2D.i[3] = reg->Dimension.Index; 1410 1411 /* Again, the second subscript index can be addressed indirectly 1412 * identically to the first one. 1413 * Nothing stops us from indirectly addressing the indirect register, 1414 * but there is no need for that, so we won't exercise it. 1415 * 1416 * file[ind[4].y+3][1], 1417 * where: 1418 * ind = DimIndirect.File 1419 * [4] = DimIndirect.Index 1420 * .y = DimIndirect.SwizzleX 1421 */ 1422 if (reg->Dimension.Indirect) { 1423 union tgsi_exec_channel index2; 1424 union tgsi_exec_channel indir_index; 1425 const uint execmask = mach->ExecMask; 1426 unsigned swizzle; 1427 uint i; 1428 1429 index2.i[0] = 1430 index2.i[1] = 1431 index2.i[2] = 1432 index2.i[3] = reg->DimIndirect.Index; 1433 1434 swizzle = tgsi_util_get_src_register_swizzle( ®->DimIndirect, CHAN_X ); 1435 fetch_src_file_channel(mach, 1436 reg->DimIndirect.File, 1437 swizzle, 1438 &index2, 1439 &ZeroVec, 1440 &indir_index); 1441 1442 index2D.i[0] += indir_index.i[0]; 1443 index2D.i[1] += indir_index.i[1]; 1444 index2D.i[2] += indir_index.i[2]; 1445 index2D.i[3] += indir_index.i[3]; 1446 1447 /* for disabled execution channels, zero-out the index to 1448 * avoid using a potential garbage value. 1449 */ 1450 for (i = 0; i < QUAD_SIZE; i++) { 1451 if ((execmask & (1 << i)) == 0) { 1452 index2D.i[i] = 0; 1453 } 1454 } 1455 } 1456 1457 /* If by any chance there was a need for a 3D array of register 1458 * files, we would have to check whether Dimension is followed 1459 * by a dimension register and continue the saga. 1460 */ 1461 } else { 1462 index2D.i[0] = 1463 index2D.i[1] = 1464 index2D.i[2] = 1465 index2D.i[3] = 0; 1466 } 1467 1468 switch (reg->Register.File) { 1469 case TGSI_FILE_NULL: 1470 dst = &null; 1471 break; 1472 1473 case TGSI_FILE_OUTPUT: 1474 index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] 1475 + reg->Register.Index; 1476 dst = &mach->Outputs[offset + index].xyzw[chan_index]; 1477#if 0 1478 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1479 fprintf(stderr, "STORING OUT[%d] mask(%d), = (", offset + index, execmask); 1480 for (i = 0; i < QUAD_SIZE; i++) 1481 if (execmask & (1 << i)) 1482 fprintf(stderr, "%f, ", chan->f[i]); 1483 fprintf(stderr, ")\n"); 1484 } 1485#endif 1486 break; 1487 1488 case TGSI_FILE_TEMPORARY: 1489 index = reg->Register.Index; 1490 assert( index < TGSI_EXEC_NUM_TEMPS ); 1491 dst = &mach->Temps[offset + index].xyzw[chan_index]; 1492 break; 1493 1494 case TGSI_FILE_TEMPORARY_ARRAY: 1495 index = reg->Register.Index; 1496 assert( index < TGSI_EXEC_NUM_TEMPS ); 1497 assert( index2D.i[0] < TGSI_EXEC_NUM_TEMP_ARRAYS ); 1498 /* XXX we use index2D.i[0] here but somehow we might 1499 * end up with someone trying to store indirectly in 1500 * different buffers */ 1501 dst = &mach->TempArray[index2D.i[0]][offset + index].xyzw[chan_index]; 1502 break; 1503 1504 case TGSI_FILE_ADDRESS: 1505 index = reg->Register.Index; 1506 dst = &mach->Addrs[index].xyzw[chan_index]; 1507 break; 1508 1509 case TGSI_FILE_PREDICATE: 1510 index = reg->Register.Index; 1511 assert(index < TGSI_EXEC_NUM_PREDS); 1512 dst = &mach->Predicates[index].xyzw[chan_index]; 1513 break; 1514 1515 default: 1516 assert( 0 ); 1517 return; 1518 } 1519 1520 if (inst->Instruction.Predicate) { 1521 uint swizzle; 1522 union tgsi_exec_channel *pred; 1523 1524 switch (chan_index) { 1525 case CHAN_X: 1526 swizzle = inst->Predicate.SwizzleX; 1527 break; 1528 case CHAN_Y: 1529 swizzle = inst->Predicate.SwizzleY; 1530 break; 1531 case CHAN_Z: 1532 swizzle = inst->Predicate.SwizzleZ; 1533 break; 1534 case CHAN_W: 1535 swizzle = inst->Predicate.SwizzleW; 1536 break; 1537 default: 1538 assert(0); 1539 return; 1540 } 1541 1542 assert(inst->Predicate.Index == 0); 1543 1544 pred = &mach->Predicates[inst->Predicate.Index].xyzw[swizzle]; 1545 1546 if (inst->Predicate.Negate) { 1547 for (i = 0; i < QUAD_SIZE; i++) { 1548 if (pred->u[i]) { 1549 execmask &= ~(1 << i); 1550 } 1551 } 1552 } else { 1553 for (i = 0; i < QUAD_SIZE; i++) { 1554 if (!pred->u[i]) { 1555 execmask &= ~(1 << i); 1556 } 1557 } 1558 } 1559 } 1560 1561 switch (inst->Instruction.Saturate) { 1562 case TGSI_SAT_NONE: 1563 for (i = 0; i < QUAD_SIZE; i++) 1564 if (execmask & (1 << i)) 1565 dst->i[i] = chan->i[i]; 1566 break; 1567 1568 case TGSI_SAT_ZERO_ONE: 1569 for (i = 0; i < QUAD_SIZE; i++) 1570 if (execmask & (1 << i)) { 1571 if (chan->f[i] < 0.0f) 1572 dst->f[i] = 0.0f; 1573 else if (chan->f[i] > 1.0f) 1574 dst->f[i] = 1.0f; 1575 else 1576 dst->i[i] = chan->i[i]; 1577 } 1578 break; 1579 1580 case TGSI_SAT_MINUS_PLUS_ONE: 1581 for (i = 0; i < QUAD_SIZE; i++) 1582 if (execmask & (1 << i)) { 1583 if (chan->f[i] < -1.0f) 1584 dst->f[i] = -1.0f; 1585 else if (chan->f[i] > 1.0f) 1586 dst->f[i] = 1.0f; 1587 else 1588 dst->i[i] = chan->i[i]; 1589 } 1590 break; 1591 1592 default: 1593 assert( 0 ); 1594 } 1595} 1596 1597#define FETCH(VAL,INDEX,CHAN)\ 1598 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT) 1599 1600#define IFETCH(VAL,INDEX,CHAN)\ 1601 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT) 1602 1603 1604/** 1605 * Execute ARB-style KIL which is predicated by a src register. 1606 * Kill fragment if any of the four values is less than zero. 1607 */ 1608static void 1609exec_kil(struct tgsi_exec_machine *mach, 1610 const struct tgsi_full_instruction *inst) 1611{ 1612 uint uniquemask; 1613 uint chan_index; 1614 uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ 1615 union tgsi_exec_channel r[1]; 1616 1617 /* This mask stores component bits that were already tested. */ 1618 uniquemask = 0; 1619 1620 for (chan_index = 0; chan_index < 4; chan_index++) 1621 { 1622 uint swizzle; 1623 uint i; 1624 1625 /* unswizzle channel */ 1626 swizzle = tgsi_util_get_full_src_register_swizzle ( 1627 &inst->Src[0], 1628 chan_index); 1629 1630 /* check if the component has not been already tested */ 1631 if (uniquemask & (1 << swizzle)) 1632 continue; 1633 uniquemask |= 1 << swizzle; 1634 1635 FETCH(&r[0], 0, chan_index); 1636 for (i = 0; i < 4; i++) 1637 if (r[0].f[i] < 0.0f) 1638 kilmask |= 1 << i; 1639 } 1640 1641 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; 1642} 1643 1644/** 1645 * Execute NVIDIA-style KIL which is predicated by a condition code. 1646 * Kill fragment if the condition code is TRUE. 1647 */ 1648static void 1649exec_kilp(struct tgsi_exec_machine *mach, 1650 const struct tgsi_full_instruction *inst) 1651{ 1652 uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ 1653 1654 /* "unconditional" kil */ 1655 kilmask = mach->ExecMask; 1656 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; 1657} 1658 1659static void 1660emit_vertex(struct tgsi_exec_machine *mach) 1661{ 1662 /* FIXME: check for exec mask correctly 1663 unsigned i; 1664 for (i = 0; i < QUAD_SIZE; ++i) { 1665 if ((mach->ExecMask & (1 << i))) 1666 */ 1667 if (mach->ExecMask) { 1668 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += mach->NumOutputs; 1669 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++; 1670 } 1671} 1672 1673static void 1674emit_primitive(struct tgsi_exec_machine *mach) 1675{ 1676 unsigned *prim_count = &mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]; 1677 /* FIXME: check for exec mask correctly 1678 unsigned i; 1679 for (i = 0; i < QUAD_SIZE; ++i) { 1680 if ((mach->ExecMask & (1 << i))) 1681 */ 1682 if (mach->ExecMask) { 1683 ++(*prim_count); 1684 debug_assert((*prim_count * mach->NumOutputs) < mach->MaxGeometryShaderOutputs); 1685 mach->Primitives[*prim_count] = 0; 1686 } 1687} 1688 1689static void 1690conditional_emit_primitive(struct tgsi_exec_machine *mach) 1691{ 1692 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1693 int emitted_verts = 1694 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]; 1695 if (emitted_verts) { 1696 emit_primitive(mach); 1697 } 1698 } 1699} 1700 1701 1702/* 1703 * Fetch four texture samples using STR texture coordinates. 1704 */ 1705static void 1706fetch_texel( struct tgsi_sampler *sampler, 1707 const union tgsi_exec_channel *s, 1708 const union tgsi_exec_channel *t, 1709 const union tgsi_exec_channel *p, 1710 const union tgsi_exec_channel *c0, 1711 enum tgsi_sampler_control control, 1712 union tgsi_exec_channel *r, 1713 union tgsi_exec_channel *g, 1714 union tgsi_exec_channel *b, 1715 union tgsi_exec_channel *a ) 1716{ 1717 uint j; 1718 float rgba[NUM_CHANNELS][QUAD_SIZE]; 1719 1720 sampler->get_samples(sampler, s->f, t->f, p->f, c0->f, control, rgba); 1721 1722 for (j = 0; j < 4; j++) { 1723 r->f[j] = rgba[0][j]; 1724 g->f[j] = rgba[1][j]; 1725 b->f[j] = rgba[2][j]; 1726 a->f[j] = rgba[3][j]; 1727 } 1728} 1729 1730 1731#define TEX_MODIFIER_NONE 0 1732#define TEX_MODIFIER_PROJECTED 1 1733#define TEX_MODIFIER_LOD_BIAS 2 1734#define TEX_MODIFIER_EXPLICIT_LOD 3 1735 1736 1737static void 1738exec_tex(struct tgsi_exec_machine *mach, 1739 const struct tgsi_full_instruction *inst, 1740 uint modifier) 1741{ 1742 const uint unit = inst->Src[1].Register.Index; 1743 union tgsi_exec_channel r[4]; 1744 const union tgsi_exec_channel *lod = &ZeroVec; 1745 enum tgsi_sampler_control control; 1746 uint chan; 1747 1748 if (modifier != TEX_MODIFIER_NONE) { 1749 FETCH(&r[3], 0, CHAN_W); 1750 if (modifier != TEX_MODIFIER_PROJECTED) { 1751 lod = &r[3]; 1752 } 1753 } 1754 1755 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) { 1756 control = tgsi_sampler_lod_explicit; 1757 } else { 1758 control = tgsi_sampler_lod_bias; 1759 } 1760 1761 switch (inst->Texture.Texture) { 1762 case TGSI_TEXTURE_1D: 1763 case TGSI_TEXTURE_SHADOW1D: 1764 FETCH(&r[0], 0, CHAN_X); 1765 1766 if (modifier == TEX_MODIFIER_PROJECTED) { 1767 micro_div(&r[0], &r[0], &r[3]); 1768 } 1769 1770 fetch_texel(mach->Samplers[unit], 1771 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */ 1772 control, 1773 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1774 break; 1775 1776 case TGSI_TEXTURE_2D: 1777 case TGSI_TEXTURE_RECT: 1778 case TGSI_TEXTURE_SHADOW2D: 1779 case TGSI_TEXTURE_SHADOWRECT: 1780 FETCH(&r[0], 0, CHAN_X); 1781 FETCH(&r[1], 0, CHAN_Y); 1782 FETCH(&r[2], 0, CHAN_Z); 1783 1784 if (modifier == TEX_MODIFIER_PROJECTED) { 1785 micro_div(&r[0], &r[0], &r[3]); 1786 micro_div(&r[1], &r[1], &r[3]); 1787 micro_div(&r[2], &r[2], &r[3]); 1788 } 1789 1790 fetch_texel(mach->Samplers[unit], 1791 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1792 control, 1793 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1794 break; 1795 1796 case TGSI_TEXTURE_1D_ARRAY: 1797 case TGSI_TEXTURE_SHADOW1D_ARRAY: 1798 FETCH(&r[0], 0, CHAN_X); 1799 FETCH(&r[1], 0, CHAN_Y); 1800 1801 if (modifier == TEX_MODIFIER_PROJECTED) { 1802 micro_div(&r[0], &r[0], &r[3]); 1803 } 1804 1805 fetch_texel(mach->Samplers[unit], 1806 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1807 control, 1808 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1809 break; 1810 1811 case TGSI_TEXTURE_2D_ARRAY: 1812 case TGSI_TEXTURE_SHADOW2D_ARRAY: 1813 FETCH(&r[0], 0, CHAN_X); 1814 FETCH(&r[1], 0, CHAN_Y); 1815 FETCH(&r[2], 0, CHAN_Z); 1816 1817 if (modifier == TEX_MODIFIER_PROJECTED) { 1818 micro_div(&r[0], &r[0], &r[3]); 1819 micro_div(&r[1], &r[1], &r[3]); 1820 } 1821 1822 fetch_texel(mach->Samplers[unit], 1823 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1824 control, 1825 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1826 break; 1827 1828 case TGSI_TEXTURE_3D: 1829 case TGSI_TEXTURE_CUBE: 1830 FETCH(&r[0], 0, CHAN_X); 1831 FETCH(&r[1], 0, CHAN_Y); 1832 FETCH(&r[2], 0, CHAN_Z); 1833 1834 if (modifier == TEX_MODIFIER_PROJECTED) { 1835 micro_div(&r[0], &r[0], &r[3]); 1836 micro_div(&r[1], &r[1], &r[3]); 1837 micro_div(&r[2], &r[2], &r[3]); 1838 } 1839 1840 fetch_texel(mach->Samplers[unit], 1841 &r[0], &r[1], &r[2], lod, 1842 control, 1843 &r[0], &r[1], &r[2], &r[3]); 1844 break; 1845 1846 default: 1847 assert(0); 1848 } 1849 1850#if 0 1851 debug_printf("fetch r: %g %g %g %g\n", 1852 r[0].f[0], r[0].f[1], r[0].f[2], r[0].f[3]); 1853 debug_printf("fetch g: %g %g %g %g\n", 1854 r[1].f[0], r[1].f[1], r[1].f[2], r[1].f[3]); 1855 debug_printf("fetch b: %g %g %g %g\n", 1856 r[2].f[0], r[2].f[1], r[2].f[2], r[2].f[3]); 1857 debug_printf("fetch a: %g %g %g %g\n", 1858 r[3].f[0], r[3].f[1], r[3].f[2], r[3].f[3]); 1859#endif 1860 1861 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1862 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1863 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1864 } 1865 } 1866} 1867 1868static void 1869exec_txd(struct tgsi_exec_machine *mach, 1870 const struct tgsi_full_instruction *inst) 1871{ 1872 const uint unit = inst->Src[3].Register.Index; 1873 union tgsi_exec_channel r[4]; 1874 uint chan; 1875 1876 /* 1877 * XXX: This is fake TXD -- the derivatives are not taken into account, yet. 1878 */ 1879 1880 switch (inst->Texture.Texture) { 1881 case TGSI_TEXTURE_1D: 1882 case TGSI_TEXTURE_SHADOW1D: 1883 1884 FETCH(&r[0], 0, CHAN_X); 1885 1886 fetch_texel(mach->Samplers[unit], 1887 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */ 1888 tgsi_sampler_lod_bias, 1889 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1890 break; 1891 1892 case TGSI_TEXTURE_1D_ARRAY: 1893 case TGSI_TEXTURE_2D: 1894 case TGSI_TEXTURE_RECT: 1895 case TGSI_TEXTURE_SHADOW1D_ARRAY: 1896 case TGSI_TEXTURE_SHADOW2D: 1897 case TGSI_TEXTURE_SHADOWRECT: 1898 1899 FETCH(&r[0], 0, CHAN_X); 1900 FETCH(&r[1], 0, CHAN_Y); 1901 FETCH(&r[2], 0, CHAN_Z); 1902 1903 fetch_texel(mach->Samplers[unit], 1904 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */ 1905 tgsi_sampler_lod_bias, 1906 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1907 break; 1908 1909 case TGSI_TEXTURE_2D_ARRAY: 1910 case TGSI_TEXTURE_3D: 1911 case TGSI_TEXTURE_CUBE: 1912 1913 FETCH(&r[0], 0, CHAN_X); 1914 FETCH(&r[1], 0, CHAN_Y); 1915 FETCH(&r[2], 0, CHAN_Z); 1916 1917 fetch_texel(mach->Samplers[unit], 1918 &r[0], &r[1], &r[2], &ZeroVec, 1919 tgsi_sampler_lod_bias, 1920 &r[0], &r[1], &r[2], &r[3]); 1921 break; 1922 1923 case TGSI_TEXTURE_SHADOW2D_ARRAY: 1924 1925 FETCH(&r[0], 0, CHAN_X); 1926 FETCH(&r[1], 0, CHAN_Y); 1927 FETCH(&r[2], 0, CHAN_Z); 1928 FETCH(&r[3], 0, CHAN_W); 1929 1930 fetch_texel(mach->Samplers[unit], 1931 &r[0], &r[1], &r[2], &r[3], 1932 tgsi_sampler_lod_bias, 1933 &r[0], &r[1], &r[2], &r[3]); 1934 break; 1935 1936 default: 1937 assert(0); 1938 } 1939 1940 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1941 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1942 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1943 } 1944 } 1945} 1946 1947 1948static void 1949exec_txf(struct tgsi_exec_machine *mach, 1950 const struct tgsi_full_instruction *inst) 1951{ 1952 struct tgsi_sampler *sampler; 1953 const uint unit = inst->Src[2].Register.Index; 1954 union tgsi_exec_channel r[4]; 1955 union tgsi_exec_channel offset[3]; 1956 uint chan; 1957 float rgba[NUM_CHANNELS][QUAD_SIZE]; 1958 int j; 1959 int8_t offsets[3]; 1960 1961 if (inst->Texture.NumOffsets == 1) { 1962 union tgsi_exec_channel index; 1963 index.i[0] = index.i[1] = index.i[2] = index.i[3] = inst->TexOffsets[0].Index; 1964 fetch_src_file_channel(mach, inst->TexOffsets[0].File, 1965 inst->TexOffsets[0].SwizzleX, &index, &ZeroVec, &offset[0]); 1966 fetch_src_file_channel(mach, inst->TexOffsets[0].File, 1967 inst->TexOffsets[0].SwizzleY, &index, &ZeroVec, &offset[1]); 1968 fetch_src_file_channel(mach, inst->TexOffsets[0].File, 1969 inst->TexOffsets[0].SwizzleZ, &index, &ZeroVec, &offset[2]); 1970 offsets[0] = offset[0].i[0]; 1971 offsets[1] = offset[1].i[0]; 1972 offsets[2] = offset[2].i[0]; 1973 } else 1974 offsets[0] = offsets[1] = offsets[2] = 0; 1975 1976 IFETCH(&r[3], 0, CHAN_W); 1977 1978 switch(inst->Texture.Texture) { 1979 case TGSI_TEXTURE_3D: 1980 case TGSI_TEXTURE_2D_ARRAY: 1981 case TGSI_TEXTURE_SHADOW2D_ARRAY: 1982 IFETCH(&r[2], 0, CHAN_Z); 1983 /* fallthrough */ 1984 case TGSI_TEXTURE_2D: 1985 case TGSI_TEXTURE_RECT: 1986 case TGSI_TEXTURE_SHADOW1D_ARRAY: 1987 case TGSI_TEXTURE_SHADOW2D: 1988 case TGSI_TEXTURE_SHADOWRECT: 1989 case TGSI_TEXTURE_1D_ARRAY: 1990 IFETCH(&r[1], 0, CHAN_Y); 1991 /* fallthrough */ 1992 case TGSI_TEXTURE_1D: 1993 case TGSI_TEXTURE_SHADOW1D: 1994 IFETCH(&r[0], 0, CHAN_X); 1995 break; 1996 default: 1997 assert(0); 1998 break; 1999 } 2000 2001 sampler = mach->Samplers[unit]; 2002 sampler->get_texel(sampler, r[0].i, r[1].i, r[2].i, r[3].i, 2003 offsets, rgba); 2004 2005 for (j = 0; j < QUAD_SIZE; j++) { 2006 r[0].f[j] = rgba[0][j]; 2007 r[1].f[j] = rgba[1][j]; 2008 r[2].f[j] = rgba[2][j]; 2009 r[3].f[j] = rgba[3][j]; 2010 } 2011 2012 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2013 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2014 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2015 } 2016 } 2017} 2018 2019static void 2020exec_txq(struct tgsi_exec_machine *mach, 2021 const struct tgsi_full_instruction *inst) 2022{ 2023 struct tgsi_sampler *sampler; 2024 const uint unit = inst->Src[1].Register.Index; 2025 int result[4]; 2026 union tgsi_exec_channel r[4], src; 2027 uint chan; 2028 int i,j; 2029 2030 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_INT); 2031 sampler = mach->Samplers[unit]; 2032 2033 sampler->get_dims(sampler, src.i[0], result); 2034 2035 for (i = 0; i < QUAD_SIZE; i++) { 2036 for (j = 0; j < 4; j++) { 2037 r[j].i[i] = result[j]; 2038 } 2039 } 2040 2041 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2042 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2043 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, 2044 TGSI_EXEC_DATA_INT); 2045 } 2046 } 2047} 2048 2049static void 2050exec_sample(struct tgsi_exec_machine *mach, 2051 const struct tgsi_full_instruction *inst, 2052 uint modifier) 2053{ 2054 const uint resource_unit = inst->Src[1].Register.Index; 2055 const uint sampler_unit = inst->Src[2].Register.Index; 2056 union tgsi_exec_channel r[4]; 2057 const union tgsi_exec_channel *lod = &ZeroVec; 2058 enum tgsi_sampler_control control; 2059 uint chan; 2060 2061 if (modifier != TEX_MODIFIER_NONE) { 2062 if (modifier == TEX_MODIFIER_LOD_BIAS) 2063 FETCH(&r[3], 3, CHAN_X); 2064 else /*TEX_MODIFIER_LOD*/ 2065 FETCH(&r[3], 0, CHAN_W); 2066 2067 if (modifier != TEX_MODIFIER_PROJECTED) { 2068 lod = &r[3]; 2069 } 2070 } 2071 2072 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) { 2073 control = tgsi_sampler_lod_explicit; 2074 } else { 2075 control = tgsi_sampler_lod_bias; 2076 } 2077 2078 switch (mach->Resources[resource_unit].Resource) { 2079 case TGSI_TEXTURE_1D: 2080 case TGSI_TEXTURE_SHADOW1D: 2081 FETCH(&r[0], 0, CHAN_X); 2082 2083 if (modifier == TEX_MODIFIER_PROJECTED) { 2084 micro_div(&r[0], &r[0], &r[3]); 2085 } 2086 2087 fetch_texel(mach->Samplers[sampler_unit], 2088 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */ 2089 control, 2090 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 2091 break; 2092 2093 case TGSI_TEXTURE_1D_ARRAY: 2094 case TGSI_TEXTURE_2D: 2095 case TGSI_TEXTURE_RECT: 2096 case TGSI_TEXTURE_SHADOW1D_ARRAY: 2097 case TGSI_TEXTURE_SHADOW2D: 2098 case TGSI_TEXTURE_SHADOWRECT: 2099 FETCH(&r[0], 0, CHAN_X); 2100 FETCH(&r[1], 0, CHAN_Y); 2101 FETCH(&r[2], 0, CHAN_Z); 2102 2103 if (modifier == TEX_MODIFIER_PROJECTED) { 2104 micro_div(&r[0], &r[0], &r[3]); 2105 micro_div(&r[1], &r[1], &r[3]); 2106 micro_div(&r[2], &r[2], &r[3]); 2107 } 2108 2109 fetch_texel(mach->Samplers[sampler_unit], 2110 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 2111 control, 2112 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 2113 break; 2114 2115 case TGSI_TEXTURE_2D_ARRAY: 2116 case TGSI_TEXTURE_3D: 2117 case TGSI_TEXTURE_CUBE: 2118 FETCH(&r[0], 0, CHAN_X); 2119 FETCH(&r[1], 0, CHAN_Y); 2120 FETCH(&r[2], 0, CHAN_Z); 2121 2122 if (modifier == TEX_MODIFIER_PROJECTED) { 2123 micro_div(&r[0], &r[0], &r[3]); 2124 micro_div(&r[1], &r[1], &r[3]); 2125 micro_div(&r[2], &r[2], &r[3]); 2126 } 2127 2128 fetch_texel(mach->Samplers[sampler_unit], 2129 &r[0], &r[1], &r[2], lod, 2130 control, 2131 &r[0], &r[1], &r[2], &r[3]); 2132 break; 2133 2134 case TGSI_TEXTURE_SHADOW2D_ARRAY: 2135 FETCH(&r[0], 0, CHAN_X); 2136 FETCH(&r[1], 0, CHAN_Y); 2137 FETCH(&r[2], 0, CHAN_Z); 2138 FETCH(&r[3], 0, CHAN_W); 2139 2140 assert(modifier != TEX_MODIFIER_PROJECTED); 2141 2142 fetch_texel(mach->Samplers[sampler_unit], 2143 &r[0], &r[1], &r[2], &r[3], 2144 control, 2145 &r[0], &r[1], &r[2], &r[3]); 2146 break; 2147 2148 default: 2149 assert(0); 2150 } 2151 2152 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2153 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2154 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2155 } 2156 } 2157} 2158 2159static void 2160exec_sample_d(struct tgsi_exec_machine *mach, 2161 const struct tgsi_full_instruction *inst) 2162{ 2163 const uint resource_unit = inst->Src[1].Register.Index; 2164 const uint sampler_unit = inst->Src[2].Register.Index; 2165 union tgsi_exec_channel r[4]; 2166 uint chan; 2167 /* 2168 * XXX: This is fake SAMPLE_D -- the derivatives are not taken into account, yet. 2169 */ 2170 2171 switch (mach->Resources[resource_unit].Resource) { 2172 case TGSI_TEXTURE_1D: 2173 case TGSI_TEXTURE_SHADOW1D: 2174 2175 FETCH(&r[0], 0, CHAN_X); 2176 2177 fetch_texel(mach->Samplers[sampler_unit], 2178 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */ 2179 tgsi_sampler_lod_bias, 2180 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 2181 break; 2182 2183 case TGSI_TEXTURE_2D: 2184 case TGSI_TEXTURE_RECT: 2185 case TGSI_TEXTURE_SHADOW2D: 2186 case TGSI_TEXTURE_SHADOWRECT: 2187 2188 FETCH(&r[0], 0, CHAN_X); 2189 FETCH(&r[1], 0, CHAN_Y); 2190 FETCH(&r[2], 0, CHAN_Z); 2191 2192 fetch_texel(mach->Samplers[sampler_unit], 2193 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */ 2194 tgsi_sampler_lod_bias, 2195 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 2196 break; 2197 2198 case TGSI_TEXTURE_3D: 2199 case TGSI_TEXTURE_CUBE: 2200 2201 FETCH(&r[0], 0, CHAN_X); 2202 FETCH(&r[1], 0, CHAN_Y); 2203 FETCH(&r[2], 0, CHAN_Z); 2204 2205 fetch_texel(mach->Samplers[sampler_unit], 2206 &r[0], &r[1], &r[2], &ZeroVec, 2207 tgsi_sampler_lod_bias, 2208 &r[0], &r[1], &r[2], &r[3]); 2209 break; 2210 2211 default: 2212 assert(0); 2213 } 2214 2215 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2216 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2217 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2218 } 2219 } 2220} 2221 2222 2223/** 2224 * Evaluate a constant-valued coefficient at the position of the 2225 * current quad. 2226 */ 2227static void 2228eval_constant_coef( 2229 struct tgsi_exec_machine *mach, 2230 unsigned attrib, 2231 unsigned chan ) 2232{ 2233 unsigned i; 2234 2235 for( i = 0; i < QUAD_SIZE; i++ ) { 2236 mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan]; 2237 } 2238} 2239 2240/** 2241 * Evaluate a linear-valued coefficient at the position of the 2242 * current quad. 2243 */ 2244static void 2245eval_linear_coef( 2246 struct tgsi_exec_machine *mach, 2247 unsigned attrib, 2248 unsigned chan ) 2249{ 2250 const float x = mach->QuadPos.xyzw[0].f[0]; 2251 const float y = mach->QuadPos.xyzw[1].f[0]; 2252 const float dadx = mach->InterpCoefs[attrib].dadx[chan]; 2253 const float dady = mach->InterpCoefs[attrib].dady[chan]; 2254 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; 2255 mach->Inputs[attrib].xyzw[chan].f[0] = a0; 2256 mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx; 2257 mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady; 2258 mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady; 2259} 2260 2261/** 2262 * Evaluate a perspective-valued coefficient at the position of the 2263 * current quad. 2264 */ 2265static void 2266eval_perspective_coef( 2267 struct tgsi_exec_machine *mach, 2268 unsigned attrib, 2269 unsigned chan ) 2270{ 2271 const float x = mach->QuadPos.xyzw[0].f[0]; 2272 const float y = mach->QuadPos.xyzw[1].f[0]; 2273 const float dadx = mach->InterpCoefs[attrib].dadx[chan]; 2274 const float dady = mach->InterpCoefs[attrib].dady[chan]; 2275 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; 2276 const float *w = mach->QuadPos.xyzw[3].f; 2277 /* divide by W here */ 2278 mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0]; 2279 mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1]; 2280 mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2]; 2281 mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3]; 2282} 2283 2284 2285typedef void (* eval_coef_func)( 2286 struct tgsi_exec_machine *mach, 2287 unsigned attrib, 2288 unsigned chan ); 2289 2290static void 2291exec_declaration(struct tgsi_exec_machine *mach, 2292 const struct tgsi_full_declaration *decl) 2293{ 2294 if (decl->Declaration.File == TGSI_FILE_RESOURCE) { 2295 mach->Resources[decl->Range.First] = decl->Resource; 2296 return; 2297 } 2298 2299 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { 2300 if (decl->Declaration.File == TGSI_FILE_INPUT) { 2301 uint first, last, mask; 2302 2303 first = decl->Range.First; 2304 last = decl->Range.Last; 2305 mask = decl->Declaration.UsageMask; 2306 2307 /* XXX we could remove this special-case code since 2308 * mach->InterpCoefs[first].a0 should already have the 2309 * front/back-face value. But we should first update the 2310 * ureg code to emit the right UsageMask value (WRITEMASK_X). 2311 * Then, we could remove the tgsi_exec_machine::Face field. 2312 */ 2313 /* XXX make FACE a system value */ 2314 if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) { 2315 uint i; 2316 2317 assert(decl->Semantic.Index == 0); 2318 assert(first == last); 2319 2320 for (i = 0; i < QUAD_SIZE; i++) { 2321 mach->Inputs[first].xyzw[0].f[i] = mach->Face; 2322 } 2323 } else { 2324 eval_coef_func eval; 2325 uint i, j; 2326 2327 switch (decl->Declaration.Interpolate) { 2328 case TGSI_INTERPOLATE_CONSTANT: 2329 eval = eval_constant_coef; 2330 break; 2331 2332 case TGSI_INTERPOLATE_LINEAR: 2333 eval = eval_linear_coef; 2334 break; 2335 2336 case TGSI_INTERPOLATE_PERSPECTIVE: 2337 eval = eval_perspective_coef; 2338 break; 2339 2340 default: 2341 assert(0); 2342 return; 2343 } 2344 2345 for (j = 0; j < NUM_CHANNELS; j++) { 2346 if (mask & (1 << j)) { 2347 for (i = first; i <= last; i++) { 2348 eval(mach, i, j); 2349 } 2350 } 2351 } 2352 } 2353 } 2354 } 2355 2356 if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) { 2357 mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First; 2358 } 2359} 2360 2361 2362typedef void (* micro_op)(union tgsi_exec_channel *dst); 2363 2364static void 2365exec_vector(struct tgsi_exec_machine *mach, 2366 const struct tgsi_full_instruction *inst, 2367 micro_op op, 2368 enum tgsi_exec_datatype dst_datatype) 2369{ 2370 unsigned int chan; 2371 2372 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2373 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2374 union tgsi_exec_channel dst; 2375 2376 op(&dst); 2377 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2378 } 2379 } 2380} 2381 2382typedef void (* micro_unary_op)(union tgsi_exec_channel *dst, 2383 const union tgsi_exec_channel *src); 2384 2385static void 2386exec_scalar_unary(struct tgsi_exec_machine *mach, 2387 const struct tgsi_full_instruction *inst, 2388 micro_unary_op op, 2389 enum tgsi_exec_datatype dst_datatype, 2390 enum tgsi_exec_datatype src_datatype) 2391{ 2392 unsigned int chan; 2393 union tgsi_exec_channel src; 2394 union tgsi_exec_channel dst; 2395 2396 fetch_source(mach, &src, &inst->Src[0], CHAN_X, src_datatype); 2397 op(&dst, &src); 2398 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2399 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2400 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2401 } 2402 } 2403} 2404 2405static void 2406exec_vector_unary(struct tgsi_exec_machine *mach, 2407 const struct tgsi_full_instruction *inst, 2408 micro_unary_op op, 2409 enum tgsi_exec_datatype dst_datatype, 2410 enum tgsi_exec_datatype src_datatype) 2411{ 2412 unsigned int chan; 2413 struct tgsi_exec_vector dst; 2414 2415 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2416 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2417 union tgsi_exec_channel src; 2418 2419 fetch_source(mach, &src, &inst->Src[0], chan, src_datatype); 2420 op(&dst.xyzw[chan], &src); 2421 } 2422 } 2423 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2424 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2425 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2426 } 2427 } 2428} 2429 2430typedef void (* micro_binary_op)(union tgsi_exec_channel *dst, 2431 const union tgsi_exec_channel *src0, 2432 const union tgsi_exec_channel *src1); 2433 2434static void 2435exec_scalar_binary(struct tgsi_exec_machine *mach, 2436 const struct tgsi_full_instruction *inst, 2437 micro_binary_op op, 2438 enum tgsi_exec_datatype dst_datatype, 2439 enum tgsi_exec_datatype src_datatype) 2440{ 2441 unsigned int chan; 2442 union tgsi_exec_channel src[2]; 2443 union tgsi_exec_channel dst; 2444 2445 fetch_source(mach, &src[0], &inst->Src[0], CHAN_X, src_datatype); 2446 fetch_source(mach, &src[1], &inst->Src[1], CHAN_Y, src_datatype); 2447 op(&dst, &src[0], &src[1]); 2448 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2449 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2450 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2451 } 2452 } 2453} 2454 2455static void 2456exec_vector_binary(struct tgsi_exec_machine *mach, 2457 const struct tgsi_full_instruction *inst, 2458 micro_binary_op op, 2459 enum tgsi_exec_datatype dst_datatype, 2460 enum tgsi_exec_datatype src_datatype) 2461{ 2462 unsigned int chan; 2463 struct tgsi_exec_vector dst; 2464 2465 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2466 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2467 union tgsi_exec_channel src[2]; 2468 2469 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype); 2470 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype); 2471 op(&dst.xyzw[chan], &src[0], &src[1]); 2472 } 2473 } 2474 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2475 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2476 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2477 } 2478 } 2479} 2480 2481typedef void (* micro_trinary_op)(union tgsi_exec_channel *dst, 2482 const union tgsi_exec_channel *src0, 2483 const union tgsi_exec_channel *src1, 2484 const union tgsi_exec_channel *src2); 2485 2486static void 2487exec_vector_trinary(struct tgsi_exec_machine *mach, 2488 const struct tgsi_full_instruction *inst, 2489 micro_trinary_op op, 2490 enum tgsi_exec_datatype dst_datatype, 2491 enum tgsi_exec_datatype src_datatype) 2492{ 2493 unsigned int chan; 2494 struct tgsi_exec_vector dst; 2495 2496 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2497 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2498 union tgsi_exec_channel src[3]; 2499 2500 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype); 2501 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype); 2502 fetch_source(mach, &src[2], &inst->Src[2], chan, src_datatype); 2503 op(&dst.xyzw[chan], &src[0], &src[1], &src[2]); 2504 } 2505 } 2506 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2507 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2508 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2509 } 2510 } 2511} 2512 2513static void 2514exec_dp3(struct tgsi_exec_machine *mach, 2515 const struct tgsi_full_instruction *inst) 2516{ 2517 unsigned int chan; 2518 union tgsi_exec_channel arg[3]; 2519 2520 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2521 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2522 micro_mul(&arg[2], &arg[0], &arg[1]); 2523 2524 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) { 2525 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2526 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT); 2527 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2528 } 2529 2530 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2531 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2532 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2533 } 2534 } 2535} 2536 2537static void 2538exec_dp4(struct tgsi_exec_machine *mach, 2539 const struct tgsi_full_instruction *inst) 2540{ 2541 unsigned int chan; 2542 union tgsi_exec_channel arg[3]; 2543 2544 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2545 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2546 micro_mul(&arg[2], &arg[0], &arg[1]); 2547 2548 for (chan = CHAN_Y; chan <= CHAN_W; chan++) { 2549 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2550 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT); 2551 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2552 } 2553 2554 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2555 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2556 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2557 } 2558 } 2559} 2560 2561static void 2562exec_dp2a(struct tgsi_exec_machine *mach, 2563 const struct tgsi_full_instruction *inst) 2564{ 2565 unsigned int chan; 2566 union tgsi_exec_channel arg[3]; 2567 2568 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2569 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2570 micro_mul(&arg[2], &arg[0], &arg[1]); 2571 2572 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2573 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2574 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]); 2575 2576 fetch_source(mach, &arg[1], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2577 micro_add(&arg[0], &arg[0], &arg[1]); 2578 2579 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2580 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2581 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2582 } 2583 } 2584} 2585 2586static void 2587exec_dph(struct tgsi_exec_machine *mach, 2588 const struct tgsi_full_instruction *inst) 2589{ 2590 unsigned int chan; 2591 union tgsi_exec_channel arg[3]; 2592 2593 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2594 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2595 micro_mul(&arg[2], &arg[0], &arg[1]); 2596 2597 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2598 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2599 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2600 2601 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2602 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2603 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]); 2604 2605 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2606 micro_add(&arg[0], &arg[0], &arg[1]); 2607 2608 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2609 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2610 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2611 } 2612 } 2613} 2614 2615static void 2616exec_dp2(struct tgsi_exec_machine *mach, 2617 const struct tgsi_full_instruction *inst) 2618{ 2619 unsigned int chan; 2620 union tgsi_exec_channel arg[3]; 2621 2622 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2623 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2624 micro_mul(&arg[2], &arg[0], &arg[1]); 2625 2626 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2627 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2628 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2629 2630 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2631 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2632 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2633 } 2634 } 2635} 2636 2637static void 2638exec_nrm4(struct tgsi_exec_machine *mach, 2639 const struct tgsi_full_instruction *inst) 2640{ 2641 unsigned int chan; 2642 union tgsi_exec_channel arg[4]; 2643 union tgsi_exec_channel scale; 2644 2645 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2646 micro_mul(&scale, &arg[0], &arg[0]); 2647 2648 for (chan = CHAN_Y; chan <= CHAN_W; chan++) { 2649 union tgsi_exec_channel product; 2650 2651 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2652 micro_mul(&product, &arg[chan], &arg[chan]); 2653 micro_add(&scale, &scale, &product); 2654 } 2655 2656 micro_rsq(&scale, &scale); 2657 2658 for (chan = CHAN_X; chan <= CHAN_W; chan++) { 2659 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2660 micro_mul(&arg[chan], &arg[chan], &scale); 2661 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2662 } 2663 } 2664} 2665 2666static void 2667exec_nrm3(struct tgsi_exec_machine *mach, 2668 const struct tgsi_full_instruction *inst) 2669{ 2670 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) { 2671 unsigned int chan; 2672 union tgsi_exec_channel arg[3]; 2673 union tgsi_exec_channel scale; 2674 2675 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2676 micro_mul(&scale, &arg[0], &arg[0]); 2677 2678 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) { 2679 union tgsi_exec_channel product; 2680 2681 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2682 micro_mul(&product, &arg[chan], &arg[chan]); 2683 micro_add(&scale, &scale, &product); 2684 } 2685 2686 micro_rsq(&scale, &scale); 2687 2688 for (chan = CHAN_X; chan <= CHAN_Z; chan++) { 2689 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2690 micro_mul(&arg[chan], &arg[chan], &scale); 2691 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2692 } 2693 } 2694 } 2695 2696 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2697 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2698 } 2699} 2700 2701static void 2702exec_scs(struct tgsi_exec_machine *mach, 2703 const struct tgsi_full_instruction *inst) 2704{ 2705 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XY) { 2706 union tgsi_exec_channel arg; 2707 union tgsi_exec_channel result; 2708 2709 fetch_source(mach, &arg, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2710 2711 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2712 micro_cos(&result, &arg); 2713 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2714 } 2715 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2716 micro_sin(&result, &arg); 2717 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2718 } 2719 } 2720 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2721 store_dest(mach, &ZeroVec, &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2722 } 2723 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2724 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2725 } 2726} 2727 2728static void 2729exec_x2d(struct tgsi_exec_machine *mach, 2730 const struct tgsi_full_instruction *inst) 2731{ 2732 union tgsi_exec_channel r[4]; 2733 union tgsi_exec_channel d[2]; 2734 2735 fetch_source(mach, &r[0], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2736 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2737 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XZ) { 2738 fetch_source(mach, &r[2], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2739 micro_mul(&r[2], &r[2], &r[0]); 2740 fetch_source(mach, &r[3], &inst->Src[2], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2741 micro_mul(&r[3], &r[3], &r[1]); 2742 micro_add(&r[2], &r[2], &r[3]); 2743 fetch_source(mach, &r[3], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2744 micro_add(&d[0], &r[2], &r[3]); 2745 } 2746 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YW) { 2747 fetch_source(mach, &r[2], &inst->Src[2], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2748 micro_mul(&r[2], &r[2], &r[0]); 2749 fetch_source(mach, &r[3], &inst->Src[2], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2750 micro_mul(&r[3], &r[3], &r[1]); 2751 micro_add(&r[2], &r[2], &r[3]); 2752 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2753 micro_add(&d[1], &r[2], &r[3]); 2754 } 2755 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2756 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2757 } 2758 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2759 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2760 } 2761 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2762 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2763 } 2764 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2765 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2766 } 2767} 2768 2769static void 2770exec_rfl(struct tgsi_exec_machine *mach, 2771 const struct tgsi_full_instruction *inst) 2772{ 2773 union tgsi_exec_channel r[9]; 2774 2775 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) { 2776 /* r0 = dp3(src0, src0) */ 2777 fetch_source(mach, &r[2], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2778 micro_mul(&r[0], &r[2], &r[2]); 2779 fetch_source(mach, &r[4], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2780 micro_mul(&r[8], &r[4], &r[4]); 2781 micro_add(&r[0], &r[0], &r[8]); 2782 fetch_source(mach, &r[6], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2783 micro_mul(&r[8], &r[6], &r[6]); 2784 micro_add(&r[0], &r[0], &r[8]); 2785 2786 /* r1 = dp3(src0, src1) */ 2787 fetch_source(mach, &r[3], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2788 micro_mul(&r[1], &r[2], &r[3]); 2789 fetch_source(mach, &r[5], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2790 micro_mul(&r[8], &r[4], &r[5]); 2791 micro_add(&r[1], &r[1], &r[8]); 2792 fetch_source(mach, &r[7], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2793 micro_mul(&r[8], &r[6], &r[7]); 2794 micro_add(&r[1], &r[1], &r[8]); 2795 2796 /* r1 = 2 * r1 / r0 */ 2797 micro_add(&r[1], &r[1], &r[1]); 2798 micro_div(&r[1], &r[1], &r[0]); 2799 2800 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2801 micro_mul(&r[2], &r[2], &r[1]); 2802 micro_sub(&r[2], &r[2], &r[3]); 2803 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2804 } 2805 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2806 micro_mul(&r[4], &r[4], &r[1]); 2807 micro_sub(&r[4], &r[4], &r[5]); 2808 store_dest(mach, &r[4], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2809 } 2810 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2811 micro_mul(&r[6], &r[6], &r[1]); 2812 micro_sub(&r[6], &r[6], &r[7]); 2813 store_dest(mach, &r[6], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2814 } 2815 } 2816 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2817 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2818 } 2819} 2820 2821static void 2822exec_xpd(struct tgsi_exec_machine *mach, 2823 const struct tgsi_full_instruction *inst) 2824{ 2825 union tgsi_exec_channel r[6]; 2826 union tgsi_exec_channel d[3]; 2827 2828 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2829 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2830 2831 micro_mul(&r[2], &r[0], &r[1]); 2832 2833 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2834 fetch_source(mach, &r[4], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2835 2836 micro_mul(&r[5], &r[3], &r[4] ); 2837 micro_sub(&d[CHAN_X], &r[2], &r[5]); 2838 2839 fetch_source(mach, &r[2], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2840 2841 micro_mul(&r[3], &r[3], &r[2]); 2842 2843 fetch_source(mach, &r[5], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2844 2845 micro_mul(&r[1], &r[1], &r[5]); 2846 micro_sub(&d[CHAN_Y], &r[3], &r[1]); 2847 2848 micro_mul(&r[5], &r[5], &r[4]); 2849 micro_mul(&r[0], &r[0], &r[2]); 2850 micro_sub(&d[CHAN_Z], &r[5], &r[0]); 2851 2852 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2853 store_dest(mach, &d[CHAN_X], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2854 } 2855 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2856 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2857 } 2858 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2859 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2860 } 2861 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2862 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2863 } 2864} 2865 2866static void 2867exec_dst(struct tgsi_exec_machine *mach, 2868 const struct tgsi_full_instruction *inst) 2869{ 2870 union tgsi_exec_channel r[2]; 2871 union tgsi_exec_channel d[4]; 2872 2873 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2874 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2875 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2876 micro_mul(&d[CHAN_Y], &r[0], &r[1]); 2877 } 2878 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2879 fetch_source(mach, &d[CHAN_Z], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2880 } 2881 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2882 fetch_source(mach, &d[CHAN_W], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2883 } 2884 2885 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2886 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2887 } 2888 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2889 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2890 } 2891 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2892 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2893 } 2894 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2895 store_dest(mach, &d[CHAN_W], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2896 } 2897} 2898 2899static void 2900exec_log(struct tgsi_exec_machine *mach, 2901 const struct tgsi_full_instruction *inst) 2902{ 2903 union tgsi_exec_channel r[3]; 2904 2905 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2906 micro_abs(&r[2], &r[0]); /* r2 = abs(r0) */ 2907 micro_lg2(&r[1], &r[2]); /* r1 = lg2(r2) */ 2908 micro_flr(&r[0], &r[1]); /* r0 = floor(r1) */ 2909 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2910 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2911 } 2912 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2913 micro_exp2(&r[0], &r[0]); /* r0 = 2 ^ r0 */ 2914 micro_div(&r[0], &r[2], &r[0]); /* r0 = r2 / r0 */ 2915 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2916 } 2917 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2918 store_dest(mach, &r[1], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2919 } 2920 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2921 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2922 } 2923} 2924 2925static void 2926exec_exp(struct tgsi_exec_machine *mach, 2927 const struct tgsi_full_instruction *inst) 2928{ 2929 union tgsi_exec_channel r[3]; 2930 2931 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2932 micro_flr(&r[1], &r[0]); /* r1 = floor(r0) */ 2933 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2934 micro_exp2(&r[2], &r[1]); /* r2 = 2 ^ r1 */ 2935 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2936 } 2937 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2938 micro_sub(&r[2], &r[0], &r[1]); /* r2 = r0 - r1 */ 2939 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2940 } 2941 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2942 micro_exp2(&r[2], &r[0]); /* r2 = 2 ^ r0 */ 2943 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2944 } 2945 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2946 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2947 } 2948} 2949 2950static void 2951exec_lit(struct tgsi_exec_machine *mach, 2952 const struct tgsi_full_instruction *inst) 2953{ 2954 union tgsi_exec_channel r[3]; 2955 union tgsi_exec_channel d[3]; 2956 2957 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2958 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2959 } 2960 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YZ) { 2961 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2962 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2963 micro_max(&d[CHAN_Y], &r[0], &ZeroVec); 2964 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2965 } 2966 2967 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2968 fetch_source(mach, &r[1], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2969 micro_max(&r[1], &r[1], &ZeroVec); 2970 2971 fetch_source(mach, &r[2], &inst->Src[0], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2972 micro_min(&r[2], &r[2], &P128Vec); 2973 micro_max(&r[2], &r[2], &M128Vec); 2974 micro_pow(&r[1], &r[1], &r[2]); 2975 micro_lt(&d[CHAN_Z], &ZeroVec, &r[0], &r[1], &ZeroVec); 2976 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2977 } 2978 } 2979 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2980 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2981 } 2982} 2983 2984static void 2985exec_break(struct tgsi_exec_machine *mach) 2986{ 2987 if (mach->BreakType == TGSI_EXEC_BREAK_INSIDE_LOOP) { 2988 /* turn off loop channels for each enabled exec channel */ 2989 mach->LoopMask &= ~mach->ExecMask; 2990 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 2991 UPDATE_EXEC_MASK(mach); 2992 } else { 2993 assert(mach->BreakType == TGSI_EXEC_BREAK_INSIDE_SWITCH); 2994 2995 mach->Switch.mask = 0x0; 2996 2997 UPDATE_EXEC_MASK(mach); 2998 } 2999} 3000 3001static void 3002exec_switch(struct tgsi_exec_machine *mach, 3003 const struct tgsi_full_instruction *inst) 3004{ 3005 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING); 3006 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3007 3008 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch; 3009 fetch_source(mach, &mach->Switch.selector, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT); 3010 mach->Switch.mask = 0x0; 3011 mach->Switch.defaultMask = 0x0; 3012 3013 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3014 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_SWITCH; 3015 3016 UPDATE_EXEC_MASK(mach); 3017} 3018 3019static void 3020exec_case(struct tgsi_exec_machine *mach, 3021 const struct tgsi_full_instruction *inst) 3022{ 3023 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask; 3024 union tgsi_exec_channel src; 3025 uint mask = 0; 3026 3027 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT); 3028 3029 if (mach->Switch.selector.u[0] == src.u[0]) { 3030 mask |= 0x1; 3031 } 3032 if (mach->Switch.selector.u[1] == src.u[1]) { 3033 mask |= 0x2; 3034 } 3035 if (mach->Switch.selector.u[2] == src.u[2]) { 3036 mask |= 0x4; 3037 } 3038 if (mach->Switch.selector.u[3] == src.u[3]) { 3039 mask |= 0x8; 3040 } 3041 3042 mach->Switch.defaultMask |= mask; 3043 3044 mach->Switch.mask |= mask & prevMask; 3045 3046 UPDATE_EXEC_MASK(mach); 3047} 3048 3049static void 3050exec_default(struct tgsi_exec_machine *mach) 3051{ 3052 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask; 3053 3054 mach->Switch.mask |= ~mach->Switch.defaultMask & prevMask; 3055 3056 UPDATE_EXEC_MASK(mach); 3057} 3058 3059static void 3060exec_endswitch(struct tgsi_exec_machine *mach) 3061{ 3062 mach->Switch = mach->SwitchStack[--mach->SwitchStackTop]; 3063 mach->BreakType = mach->BreakStack[--mach->BreakStackTop]; 3064 3065 UPDATE_EXEC_MASK(mach); 3066} 3067 3068static void 3069micro_i2f(union tgsi_exec_channel *dst, 3070 const union tgsi_exec_channel *src) 3071{ 3072 dst->f[0] = (float)src->i[0]; 3073 dst->f[1] = (float)src->i[1]; 3074 dst->f[2] = (float)src->i[2]; 3075 dst->f[3] = (float)src->i[3]; 3076} 3077 3078static void 3079micro_not(union tgsi_exec_channel *dst, 3080 const union tgsi_exec_channel *src) 3081{ 3082 dst->u[0] = ~src->u[0]; 3083 dst->u[1] = ~src->u[1]; 3084 dst->u[2] = ~src->u[2]; 3085 dst->u[3] = ~src->u[3]; 3086} 3087 3088static void 3089micro_shl(union tgsi_exec_channel *dst, 3090 const union tgsi_exec_channel *src0, 3091 const union tgsi_exec_channel *src1) 3092{ 3093 dst->u[0] = src0->u[0] << src1->u[0]; 3094 dst->u[1] = src0->u[1] << src1->u[1]; 3095 dst->u[2] = src0->u[2] << src1->u[2]; 3096 dst->u[3] = src0->u[3] << src1->u[3]; 3097} 3098 3099static void 3100micro_and(union tgsi_exec_channel *dst, 3101 const union tgsi_exec_channel *src0, 3102 const union tgsi_exec_channel *src1) 3103{ 3104 dst->u[0] = src0->u[0] & src1->u[0]; 3105 dst->u[1] = src0->u[1] & src1->u[1]; 3106 dst->u[2] = src0->u[2] & src1->u[2]; 3107 dst->u[3] = src0->u[3] & src1->u[3]; 3108} 3109 3110static void 3111micro_or(union tgsi_exec_channel *dst, 3112 const union tgsi_exec_channel *src0, 3113 const union tgsi_exec_channel *src1) 3114{ 3115 dst->u[0] = src0->u[0] | src1->u[0]; 3116 dst->u[1] = src0->u[1] | src1->u[1]; 3117 dst->u[2] = src0->u[2] | src1->u[2]; 3118 dst->u[3] = src0->u[3] | src1->u[3]; 3119} 3120 3121static void 3122micro_xor(union tgsi_exec_channel *dst, 3123 const union tgsi_exec_channel *src0, 3124 const union tgsi_exec_channel *src1) 3125{ 3126 dst->u[0] = src0->u[0] ^ src1->u[0]; 3127 dst->u[1] = src0->u[1] ^ src1->u[1]; 3128 dst->u[2] = src0->u[2] ^ src1->u[2]; 3129 dst->u[3] = src0->u[3] ^ src1->u[3]; 3130} 3131 3132static void 3133micro_mod(union tgsi_exec_channel *dst, 3134 const union tgsi_exec_channel *src0, 3135 const union tgsi_exec_channel *src1) 3136{ 3137 dst->i[0] = src0->i[0] % src1->i[0]; 3138 dst->i[1] = src0->i[1] % src1->i[1]; 3139 dst->i[2] = src0->i[2] % src1->i[2]; 3140 dst->i[3] = src0->i[3] % src1->i[3]; 3141} 3142 3143static void 3144micro_f2i(union tgsi_exec_channel *dst, 3145 const union tgsi_exec_channel *src) 3146{ 3147 dst->i[0] = (int)src->f[0]; 3148 dst->i[1] = (int)src->f[1]; 3149 dst->i[2] = (int)src->f[2]; 3150 dst->i[3] = (int)src->f[3]; 3151} 3152 3153static void 3154micro_idiv(union tgsi_exec_channel *dst, 3155 const union tgsi_exec_channel *src0, 3156 const union tgsi_exec_channel *src1) 3157{ 3158 dst->i[0] = src0->i[0] / src1->i[0]; 3159 dst->i[1] = src0->i[1] / src1->i[1]; 3160 dst->i[2] = src0->i[2] / src1->i[2]; 3161 dst->i[3] = src0->i[3] / src1->i[3]; 3162} 3163 3164static void 3165micro_imax(union tgsi_exec_channel *dst, 3166 const union tgsi_exec_channel *src0, 3167 const union tgsi_exec_channel *src1) 3168{ 3169 dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0]; 3170 dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1]; 3171 dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2]; 3172 dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3]; 3173} 3174 3175static void 3176micro_imin(union tgsi_exec_channel *dst, 3177 const union tgsi_exec_channel *src0, 3178 const union tgsi_exec_channel *src1) 3179{ 3180 dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0]; 3181 dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1]; 3182 dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2]; 3183 dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3]; 3184} 3185 3186static void 3187micro_isge(union tgsi_exec_channel *dst, 3188 const union tgsi_exec_channel *src0, 3189 const union tgsi_exec_channel *src1) 3190{ 3191 dst->i[0] = src0->i[0] >= src1->i[0] ? -1 : 0; 3192 dst->i[1] = src0->i[1] >= src1->i[1] ? -1 : 0; 3193 dst->i[2] = src0->i[2] >= src1->i[2] ? -1 : 0; 3194 dst->i[3] = src0->i[3] >= src1->i[3] ? -1 : 0; 3195} 3196 3197static void 3198micro_ishr(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_islt(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] ? -1 : 0; 3214 dst->i[1] = src0->i[1] < src1->i[1] ? -1 : 0; 3215 dst->i[2] = src0->i[2] < src1->i[2] ? -1 : 0; 3216 dst->i[3] = src0->i[3] < src1->i[3] ? -1 : 0; 3217} 3218 3219static void 3220micro_f2u(union tgsi_exec_channel *dst, 3221 const union tgsi_exec_channel *src) 3222{ 3223 dst->u[0] = (uint)src->f[0]; 3224 dst->u[1] = (uint)src->f[1]; 3225 dst->u[2] = (uint)src->f[2]; 3226 dst->u[3] = (uint)src->f[3]; 3227} 3228 3229static void 3230micro_u2f(union tgsi_exec_channel *dst, 3231 const union tgsi_exec_channel *src) 3232{ 3233 dst->f[0] = (float)src->u[0]; 3234 dst->f[1] = (float)src->u[1]; 3235 dst->f[2] = (float)src->u[2]; 3236 dst->f[3] = (float)src->u[3]; 3237} 3238 3239static void 3240micro_uadd(union tgsi_exec_channel *dst, 3241 const union tgsi_exec_channel *src0, 3242 const union tgsi_exec_channel *src1) 3243{ 3244 dst->u[0] = src0->u[0] + src1->u[0]; 3245 dst->u[1] = src0->u[1] + src1->u[1]; 3246 dst->u[2] = src0->u[2] + src1->u[2]; 3247 dst->u[3] = src0->u[3] + src1->u[3]; 3248} 3249 3250static void 3251micro_udiv(union tgsi_exec_channel *dst, 3252 const union tgsi_exec_channel *src0, 3253 const union tgsi_exec_channel *src1) 3254{ 3255 dst->u[0] = src0->u[0] / src1->u[0]; 3256 dst->u[1] = src0->u[1] / src1->u[1]; 3257 dst->u[2] = src0->u[2] / src1->u[2]; 3258 dst->u[3] = src0->u[3] / src1->u[3]; 3259} 3260 3261static void 3262micro_umad(union tgsi_exec_channel *dst, 3263 const union tgsi_exec_channel *src0, 3264 const union tgsi_exec_channel *src1, 3265 const union tgsi_exec_channel *src2) 3266{ 3267 dst->u[0] = src0->u[0] * src1->u[0] + src2->u[0]; 3268 dst->u[1] = src0->u[1] * src1->u[1] + src2->u[1]; 3269 dst->u[2] = src0->u[2] * src1->u[2] + src2->u[2]; 3270 dst->u[3] = src0->u[3] * src1->u[3] + src2->u[3]; 3271} 3272 3273static void 3274micro_umax(union tgsi_exec_channel *dst, 3275 const union tgsi_exec_channel *src0, 3276 const union tgsi_exec_channel *src1) 3277{ 3278 dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0]; 3279 dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1]; 3280 dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2]; 3281 dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3]; 3282} 3283 3284static void 3285micro_umin(union tgsi_exec_channel *dst, 3286 const union tgsi_exec_channel *src0, 3287 const union tgsi_exec_channel *src1) 3288{ 3289 dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0]; 3290 dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1]; 3291 dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2]; 3292 dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3]; 3293} 3294 3295static void 3296micro_umod(union tgsi_exec_channel *dst, 3297 const union tgsi_exec_channel *src0, 3298 const union tgsi_exec_channel *src1) 3299{ 3300 dst->u[0] = src0->u[0] % src1->u[0]; 3301 dst->u[1] = src0->u[1] % src1->u[1]; 3302 dst->u[2] = src0->u[2] % src1->u[2]; 3303 dst->u[3] = src0->u[3] % src1->u[3]; 3304} 3305 3306static void 3307micro_umul(union tgsi_exec_channel *dst, 3308 const union tgsi_exec_channel *src0, 3309 const union tgsi_exec_channel *src1) 3310{ 3311 dst->u[0] = src0->u[0] * src1->u[0]; 3312 dst->u[1] = src0->u[1] * src1->u[1]; 3313 dst->u[2] = src0->u[2] * src1->u[2]; 3314 dst->u[3] = src0->u[3] * src1->u[3]; 3315} 3316 3317static void 3318micro_useq(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] ? ~0 : 0; 3323 dst->u[1] = src0->u[1] == src1->u[1] ? ~0 : 0; 3324 dst->u[2] = src0->u[2] == src1->u[2] ? ~0 : 0; 3325 dst->u[3] = src0->u[3] == src1->u[3] ? ~0 : 0; 3326} 3327 3328static void 3329micro_usge(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] ? ~0 : 0; 3334 dst->u[1] = src0->u[1] >= src1->u[1] ? ~0 : 0; 3335 dst->u[2] = src0->u[2] >= src1->u[2] ? ~0 : 0; 3336 dst->u[3] = src0->u[3] >= src1->u[3] ? ~0 : 0; 3337} 3338 3339static void 3340micro_ushr(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_uslt(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] ? ~0 : 0; 3356 dst->u[1] = src0->u[1] < src1->u[1] ? ~0 : 0; 3357 dst->u[2] = src0->u[2] < src1->u[2] ? ~0 : 0; 3358 dst->u[3] = src0->u[3] < src1->u[3] ? ~0 : 0; 3359} 3360 3361static void 3362micro_usne(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_uarl(union tgsi_exec_channel *dst, 3374 const union tgsi_exec_channel *src) 3375{ 3376 dst->i[0] = src->u[0]; 3377 dst->i[1] = src->u[1]; 3378 dst->i[2] = src->u[2]; 3379 dst->i[3] = src->u[3]; 3380} 3381 3382static void 3383micro_ucmp(union tgsi_exec_channel *dst, 3384 const union tgsi_exec_channel *src0, 3385 const union tgsi_exec_channel *src1, 3386 const union tgsi_exec_channel *src2) 3387{ 3388 dst->u[0] = src0->u[0] ? src1->u[0] : src2->u[0]; 3389 dst->u[1] = src0->u[1] ? src1->u[1] : src2->u[1]; 3390 dst->u[2] = src0->u[2] ? src1->u[2] : src2->u[2]; 3391 dst->u[3] = src0->u[3] ? src1->u[3] : src2->u[3]; 3392} 3393 3394static void 3395exec_instruction( 3396 struct tgsi_exec_machine *mach, 3397 const struct tgsi_full_instruction *inst, 3398 int *pc ) 3399{ 3400 union tgsi_exec_channel r[10]; 3401 3402 (*pc)++; 3403 3404 switch (inst->Instruction.Opcode) { 3405 case TGSI_OPCODE_ARL: 3406 exec_vector_unary(mach, inst, micro_arl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3407 break; 3408 3409 case TGSI_OPCODE_MOV: 3410 exec_vector_unary(mach, inst, micro_mov, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT); 3411 break; 3412 3413 case TGSI_OPCODE_LIT: 3414 exec_lit(mach, inst); 3415 break; 3416 3417 case TGSI_OPCODE_RCP: 3418 exec_scalar_unary(mach, inst, micro_rcp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3419 break; 3420 3421 case TGSI_OPCODE_RSQ: 3422 exec_scalar_unary(mach, inst, micro_rsq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3423 break; 3424 3425 case TGSI_OPCODE_EXP: 3426 exec_exp(mach, inst); 3427 break; 3428 3429 case TGSI_OPCODE_LOG: 3430 exec_log(mach, inst); 3431 break; 3432 3433 case TGSI_OPCODE_MUL: 3434 exec_vector_binary(mach, inst, micro_mul, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3435 break; 3436 3437 case TGSI_OPCODE_ADD: 3438 exec_vector_binary(mach, inst, micro_add, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3439 break; 3440 3441 case TGSI_OPCODE_DP3: 3442 exec_dp3(mach, inst); 3443 break; 3444 3445 case TGSI_OPCODE_DP4: 3446 exec_dp4(mach, inst); 3447 break; 3448 3449 case TGSI_OPCODE_DST: 3450 exec_dst(mach, inst); 3451 break; 3452 3453 case TGSI_OPCODE_MIN: 3454 exec_vector_binary(mach, inst, micro_min, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3455 break; 3456 3457 case TGSI_OPCODE_MAX: 3458 exec_vector_binary(mach, inst, micro_max, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3459 break; 3460 3461 case TGSI_OPCODE_SLT: 3462 exec_vector_binary(mach, inst, micro_slt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3463 break; 3464 3465 case TGSI_OPCODE_SGE: 3466 exec_vector_binary(mach, inst, micro_sge, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3467 break; 3468 3469 case TGSI_OPCODE_MAD: 3470 exec_vector_trinary(mach, inst, micro_mad, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3471 break; 3472 3473 case TGSI_OPCODE_SUB: 3474 exec_vector_binary(mach, inst, micro_sub, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3475 break; 3476 3477 case TGSI_OPCODE_LRP: 3478 exec_vector_trinary(mach, inst, micro_lrp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3479 break; 3480 3481 case TGSI_OPCODE_CND: 3482 exec_vector_trinary(mach, inst, micro_cnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3483 break; 3484 3485 case TGSI_OPCODE_DP2A: 3486 exec_dp2a(mach, inst); 3487 break; 3488 3489 case TGSI_OPCODE_FRC: 3490 exec_vector_unary(mach, inst, micro_frc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3491 break; 3492 3493 case TGSI_OPCODE_CLAMP: 3494 exec_vector_trinary(mach, inst, micro_clamp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3495 break; 3496 3497 case TGSI_OPCODE_FLR: 3498 exec_vector_unary(mach, inst, micro_flr, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3499 break; 3500 3501 case TGSI_OPCODE_ROUND: 3502 exec_vector_unary(mach, inst, micro_rnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3503 break; 3504 3505 case TGSI_OPCODE_EX2: 3506 exec_scalar_unary(mach, inst, micro_exp2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3507 break; 3508 3509 case TGSI_OPCODE_LG2: 3510 exec_scalar_unary(mach, inst, micro_lg2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3511 break; 3512 3513 case TGSI_OPCODE_POW: 3514 exec_scalar_binary(mach, inst, micro_pow, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3515 break; 3516 3517 case TGSI_OPCODE_XPD: 3518 exec_xpd(mach, inst); 3519 break; 3520 3521 case TGSI_OPCODE_ABS: 3522 exec_vector_unary(mach, inst, micro_abs, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3523 break; 3524 3525 case TGSI_OPCODE_RCC: 3526 exec_scalar_unary(mach, inst, micro_rcc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3527 break; 3528 3529 case TGSI_OPCODE_DPH: 3530 exec_dph(mach, inst); 3531 break; 3532 3533 case TGSI_OPCODE_COS: 3534 exec_scalar_unary(mach, inst, micro_cos, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3535 break; 3536 3537 case TGSI_OPCODE_DDX: 3538 exec_vector_unary(mach, inst, micro_ddx, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3539 break; 3540 3541 case TGSI_OPCODE_DDY: 3542 exec_vector_unary(mach, inst, micro_ddy, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3543 break; 3544 3545 case TGSI_OPCODE_KILP: 3546 exec_kilp (mach, inst); 3547 break; 3548 3549 case TGSI_OPCODE_KIL: 3550 exec_kil (mach, inst); 3551 break; 3552 3553 case TGSI_OPCODE_PK2H: 3554 assert (0); 3555 break; 3556 3557 case TGSI_OPCODE_PK2US: 3558 assert (0); 3559 break; 3560 3561 case TGSI_OPCODE_PK4B: 3562 assert (0); 3563 break; 3564 3565 case TGSI_OPCODE_PK4UB: 3566 assert (0); 3567 break; 3568 3569 case TGSI_OPCODE_RFL: 3570 exec_rfl(mach, inst); 3571 break; 3572 3573 case TGSI_OPCODE_SEQ: 3574 exec_vector_binary(mach, inst, micro_seq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3575 break; 3576 3577 case TGSI_OPCODE_SFL: 3578 exec_vector(mach, inst, micro_sfl, TGSI_EXEC_DATA_FLOAT); 3579 break; 3580 3581 case TGSI_OPCODE_SGT: 3582 exec_vector_binary(mach, inst, micro_sgt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3583 break; 3584 3585 case TGSI_OPCODE_SIN: 3586 exec_scalar_unary(mach, inst, micro_sin, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3587 break; 3588 3589 case TGSI_OPCODE_SLE: 3590 exec_vector_binary(mach, inst, micro_sle, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3591 break; 3592 3593 case TGSI_OPCODE_SNE: 3594 exec_vector_binary(mach, inst, micro_sne, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3595 break; 3596 3597 case TGSI_OPCODE_STR: 3598 exec_vector(mach, inst, micro_str, TGSI_EXEC_DATA_FLOAT); 3599 break; 3600 3601 case TGSI_OPCODE_TEX: 3602 /* simple texture lookup */ 3603 /* src[0] = texcoord */ 3604 /* src[1] = sampler unit */ 3605 exec_tex(mach, inst, TEX_MODIFIER_NONE); 3606 break; 3607 3608 case TGSI_OPCODE_TXB: 3609 /* Texture lookup with lod bias */ 3610 /* src[0] = texcoord (src[0].w = LOD bias) */ 3611 /* src[1] = sampler unit */ 3612 exec_tex(mach, inst, TEX_MODIFIER_LOD_BIAS); 3613 break; 3614 3615 case TGSI_OPCODE_TXD: 3616 /* Texture lookup with explict partial derivatives */ 3617 /* src[0] = texcoord */ 3618 /* src[1] = d[strq]/dx */ 3619 /* src[2] = d[strq]/dy */ 3620 /* src[3] = sampler unit */ 3621 exec_txd(mach, inst); 3622 break; 3623 3624 case TGSI_OPCODE_TXL: 3625 /* Texture lookup with explit LOD */ 3626 /* src[0] = texcoord (src[0].w = LOD) */ 3627 /* src[1] = sampler unit */ 3628 exec_tex(mach, inst, TEX_MODIFIER_EXPLICIT_LOD); 3629 break; 3630 3631 case TGSI_OPCODE_TXP: 3632 /* Texture lookup with projection */ 3633 /* src[0] = texcoord (src[0].w = projection) */ 3634 /* src[1] = sampler unit */ 3635 exec_tex(mach, inst, TEX_MODIFIER_PROJECTED); 3636 break; 3637 3638 case TGSI_OPCODE_UP2H: 3639 assert (0); 3640 break; 3641 3642 case TGSI_OPCODE_UP2US: 3643 assert (0); 3644 break; 3645 3646 case TGSI_OPCODE_UP4B: 3647 assert (0); 3648 break; 3649 3650 case TGSI_OPCODE_UP4UB: 3651 assert (0); 3652 break; 3653 3654 case TGSI_OPCODE_X2D: 3655 exec_x2d(mach, inst); 3656 break; 3657 3658 case TGSI_OPCODE_ARA: 3659 assert (0); 3660 break; 3661 3662 case TGSI_OPCODE_ARR: 3663 exec_vector_unary(mach, inst, micro_arr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3664 break; 3665 3666 case TGSI_OPCODE_BRA: 3667 assert (0); 3668 break; 3669 3670 case TGSI_OPCODE_CAL: 3671 /* skip the call if no execution channels are enabled */ 3672 if (mach->ExecMask) { 3673 /* do the call */ 3674 3675 /* First, record the depths of the execution stacks. 3676 * This is important for deeply nested/looped return statements. 3677 * We have to unwind the stacks by the correct amount. For a 3678 * real code generator, we could determine the number of entries 3679 * to pop off each stack with simple static analysis and avoid 3680 * implementing this data structure at run time. 3681 */ 3682 mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop; 3683 mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop; 3684 mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop; 3685 mach->CallStack[mach->CallStackTop].SwitchStackTop = mach->SwitchStackTop; 3686 mach->CallStack[mach->CallStackTop].BreakStackTop = mach->BreakStackTop; 3687 /* note that PC was already incremented above */ 3688 mach->CallStack[mach->CallStackTop].ReturnAddr = *pc; 3689 3690 mach->CallStackTop++; 3691 3692 /* Second, push the Cond, Loop, Cont, Func stacks */ 3693 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); 3694 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3695 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3696 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING); 3697 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3698 assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING); 3699 3700 mach->CondStack[mach->CondStackTop++] = mach->CondMask; 3701 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; 3702 mach->ContStack[mach->ContStackTop++] = mach->ContMask; 3703 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch; 3704 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3705 mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask; 3706 3707 /* Finally, jump to the subroutine */ 3708 *pc = inst->Label.Label; 3709 } 3710 break; 3711 3712 case TGSI_OPCODE_RET: 3713 mach->FuncMask &= ~mach->ExecMask; 3714 UPDATE_EXEC_MASK(mach); 3715 3716 if (mach->FuncMask == 0x0) { 3717 /* really return now (otherwise, keep executing */ 3718 3719 if (mach->CallStackTop == 0) { 3720 /* returning from main() */ 3721 mach->CondStackTop = 0; 3722 mach->LoopStackTop = 0; 3723 *pc = -1; 3724 return; 3725 } 3726 3727 assert(mach->CallStackTop > 0); 3728 mach->CallStackTop--; 3729 3730 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; 3731 mach->CondMask = mach->CondStack[mach->CondStackTop]; 3732 3733 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; 3734 mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; 3735 3736 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; 3737 mach->ContMask = mach->ContStack[mach->ContStackTop]; 3738 3739 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop; 3740 mach->Switch = mach->SwitchStack[mach->SwitchStackTop]; 3741 3742 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop; 3743 mach->BreakType = mach->BreakStack[mach->BreakStackTop]; 3744 3745 assert(mach->FuncStackTop > 0); 3746 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; 3747 3748 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; 3749 3750 UPDATE_EXEC_MASK(mach); 3751 } 3752 break; 3753 3754 case TGSI_OPCODE_SSG: 3755 exec_vector_unary(mach, inst, micro_sgn, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3756 break; 3757 3758 case TGSI_OPCODE_CMP: 3759 exec_vector_trinary(mach, inst, micro_cmp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3760 break; 3761 3762 case TGSI_OPCODE_SCS: 3763 exec_scs(mach, inst); 3764 break; 3765 3766 case TGSI_OPCODE_NRM: 3767 exec_nrm3(mach, inst); 3768 break; 3769 3770 case TGSI_OPCODE_NRM4: 3771 exec_nrm4(mach, inst); 3772 break; 3773 3774 case TGSI_OPCODE_DIV: 3775 exec_vector_binary(mach, inst, micro_div, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3776 break; 3777 3778 case TGSI_OPCODE_DP2: 3779 exec_dp2(mach, inst); 3780 break; 3781 3782 case TGSI_OPCODE_IF: 3783 /* push CondMask */ 3784 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); 3785 mach->CondStack[mach->CondStackTop++] = mach->CondMask; 3786 FETCH( &r[0], 0, CHAN_X ); 3787 /* update CondMask */ 3788 if( ! r[0].u[0] ) { 3789 mach->CondMask &= ~0x1; 3790 } 3791 if( ! r[0].u[1] ) { 3792 mach->CondMask &= ~0x2; 3793 } 3794 if( ! r[0].u[2] ) { 3795 mach->CondMask &= ~0x4; 3796 } 3797 if( ! r[0].u[3] ) { 3798 mach->CondMask &= ~0x8; 3799 } 3800 UPDATE_EXEC_MASK(mach); 3801 /* Todo: If CondMask==0, jump to ELSE */ 3802 break; 3803 3804 case TGSI_OPCODE_ELSE: 3805 /* invert CondMask wrt previous mask */ 3806 { 3807 uint prevMask; 3808 assert(mach->CondStackTop > 0); 3809 prevMask = mach->CondStack[mach->CondStackTop - 1]; 3810 mach->CondMask = ~mach->CondMask & prevMask; 3811 UPDATE_EXEC_MASK(mach); 3812 /* Todo: If CondMask==0, jump to ENDIF */ 3813 } 3814 break; 3815 3816 case TGSI_OPCODE_ENDIF: 3817 /* pop CondMask */ 3818 assert(mach->CondStackTop > 0); 3819 mach->CondMask = mach->CondStack[--mach->CondStackTop]; 3820 UPDATE_EXEC_MASK(mach); 3821 break; 3822 3823 case TGSI_OPCODE_END: 3824 /* make sure we end primitives which haven't 3825 * been explicitly emitted */ 3826 conditional_emit_primitive(mach); 3827 /* halt execution */ 3828 *pc = -1; 3829 break; 3830 3831 case TGSI_OPCODE_PUSHA: 3832 assert (0); 3833 break; 3834 3835 case TGSI_OPCODE_POPA: 3836 assert (0); 3837 break; 3838 3839 case TGSI_OPCODE_CEIL: 3840 exec_vector_unary(mach, inst, micro_ceil, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3841 break; 3842 3843 case TGSI_OPCODE_I2F: 3844 exec_vector_unary(mach, inst, micro_i2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_INT); 3845 break; 3846 3847 case TGSI_OPCODE_NOT: 3848 exec_vector_unary(mach, inst, micro_not, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3849 break; 3850 3851 case TGSI_OPCODE_TRUNC: 3852 exec_vector_unary(mach, inst, micro_trunc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3853 break; 3854 3855 case TGSI_OPCODE_SHL: 3856 exec_vector_binary(mach, inst, micro_shl, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3857 break; 3858 3859 case TGSI_OPCODE_AND: 3860 exec_vector_binary(mach, inst, micro_and, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3861 break; 3862 3863 case TGSI_OPCODE_OR: 3864 exec_vector_binary(mach, inst, micro_or, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3865 break; 3866 3867 case TGSI_OPCODE_MOD: 3868 exec_vector_binary(mach, inst, micro_mod, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3869 break; 3870 3871 case TGSI_OPCODE_XOR: 3872 exec_vector_binary(mach, inst, micro_xor, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3873 break; 3874 3875 case TGSI_OPCODE_SAD: 3876 assert (0); 3877 break; 3878 3879 case TGSI_OPCODE_TXF: 3880 exec_txf(mach, inst); 3881 break; 3882 3883 case TGSI_OPCODE_TXQ: 3884 exec_txq(mach, inst); 3885 break; 3886 3887 case TGSI_OPCODE_EMIT: 3888 emit_vertex(mach); 3889 break; 3890 3891 case TGSI_OPCODE_ENDPRIM: 3892 emit_primitive(mach); 3893 break; 3894 3895 case TGSI_OPCODE_BGNLOOP: 3896 /* push LoopMask and ContMasks */ 3897 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3898 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3899 assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3900 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3901 3902 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; 3903 mach->ContStack[mach->ContStackTop++] = mach->ContMask; 3904 mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1; 3905 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3906 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP; 3907 break; 3908 3909 case TGSI_OPCODE_ENDLOOP: 3910 /* Restore ContMask, but don't pop */ 3911 assert(mach->ContStackTop > 0); 3912 mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; 3913 UPDATE_EXEC_MASK(mach); 3914 if (mach->ExecMask) { 3915 /* repeat loop: jump to instruction just past BGNLOOP */ 3916 assert(mach->LoopLabelStackTop > 0); 3917 *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1; 3918 } 3919 else { 3920 /* exit loop: pop LoopMask */ 3921 assert(mach->LoopStackTop > 0); 3922 mach->LoopMask = mach->LoopStack[--mach->LoopStackTop]; 3923 /* pop ContMask */ 3924 assert(mach->ContStackTop > 0); 3925 mach->ContMask = mach->ContStack[--mach->ContStackTop]; 3926 assert(mach->LoopLabelStackTop > 0); 3927 --mach->LoopLabelStackTop; 3928 3929 mach->BreakType = mach->BreakStack[--mach->BreakStackTop]; 3930 } 3931 UPDATE_EXEC_MASK(mach); 3932 break; 3933 3934 case TGSI_OPCODE_BRK: 3935 exec_break(mach); 3936 break; 3937 3938 case TGSI_OPCODE_CONT: 3939 /* turn off cont channels for each enabled exec channel */ 3940 mach->ContMask &= ~mach->ExecMask; 3941 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 3942 UPDATE_EXEC_MASK(mach); 3943 break; 3944 3945 case TGSI_OPCODE_BGNSUB: 3946 /* no-op */ 3947 break; 3948 3949 case TGSI_OPCODE_ENDSUB: 3950 /* 3951 * XXX: This really should be a no-op. We should never reach this opcode. 3952 */ 3953 3954 assert(mach->CallStackTop > 0); 3955 mach->CallStackTop--; 3956 3957 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; 3958 mach->CondMask = mach->CondStack[mach->CondStackTop]; 3959 3960 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; 3961 mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; 3962 3963 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; 3964 mach->ContMask = mach->ContStack[mach->ContStackTop]; 3965 3966 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop; 3967 mach->Switch = mach->SwitchStack[mach->SwitchStackTop]; 3968 3969 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop; 3970 mach->BreakType = mach->BreakStack[mach->BreakStackTop]; 3971 3972 assert(mach->FuncStackTop > 0); 3973 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; 3974 3975 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; 3976 3977 UPDATE_EXEC_MASK(mach); 3978 break; 3979 3980 case TGSI_OPCODE_NOP: 3981 break; 3982 3983 case TGSI_OPCODE_BREAKC: 3984 FETCH(&r[0], 0, CHAN_X); 3985 /* update CondMask */ 3986 if (r[0].u[0] && (mach->ExecMask & 0x1)) { 3987 mach->LoopMask &= ~0x1; 3988 } 3989 if (r[0].u[1] && (mach->ExecMask & 0x2)) { 3990 mach->LoopMask &= ~0x2; 3991 } 3992 if (r[0].u[2] && (mach->ExecMask & 0x4)) { 3993 mach->LoopMask &= ~0x4; 3994 } 3995 if (r[0].u[3] && (mach->ExecMask & 0x8)) { 3996 mach->LoopMask &= ~0x8; 3997 } 3998 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 3999 UPDATE_EXEC_MASK(mach); 4000 break; 4001 4002 case TGSI_OPCODE_F2I: 4003 exec_vector_unary(mach, inst, micro_f2i, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 4004 break; 4005 4006 case TGSI_OPCODE_IDIV: 4007 exec_vector_binary(mach, inst, micro_idiv, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4008 break; 4009 4010 case TGSI_OPCODE_IMAX: 4011 exec_vector_binary(mach, inst, micro_imax, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4012 break; 4013 4014 case TGSI_OPCODE_IMIN: 4015 exec_vector_binary(mach, inst, micro_imin, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4016 break; 4017 4018 case TGSI_OPCODE_INEG: 4019 exec_vector_unary(mach, inst, micro_ineg, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4020 break; 4021 4022 case TGSI_OPCODE_ISGE: 4023 exec_vector_binary(mach, inst, micro_isge, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4024 break; 4025 4026 case TGSI_OPCODE_ISHR: 4027 exec_vector_binary(mach, inst, micro_ishr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4028 break; 4029 4030 case TGSI_OPCODE_ISLT: 4031 exec_vector_binary(mach, inst, micro_islt, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4032 break; 4033 4034 case TGSI_OPCODE_F2U: 4035 exec_vector_unary(mach, inst, micro_f2u, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT); 4036 break; 4037 4038 case TGSI_OPCODE_U2F: 4039 exec_vector_unary(mach, inst, micro_u2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_UINT); 4040 break; 4041 4042 case TGSI_OPCODE_UADD: 4043 exec_vector_binary(mach, inst, micro_uadd, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4044 break; 4045 4046 case TGSI_OPCODE_UDIV: 4047 exec_vector_binary(mach, inst, micro_udiv, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4048 break; 4049 4050 case TGSI_OPCODE_UMAD: 4051 exec_vector_trinary(mach, inst, micro_umad, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4052 break; 4053 4054 case TGSI_OPCODE_UMAX: 4055 exec_vector_binary(mach, inst, micro_umax, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4056 break; 4057 4058 case TGSI_OPCODE_UMIN: 4059 exec_vector_binary(mach, inst, micro_umin, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4060 break; 4061 4062 case TGSI_OPCODE_UMOD: 4063 exec_vector_binary(mach, inst, micro_umod, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4064 break; 4065 4066 case TGSI_OPCODE_UMUL: 4067 exec_vector_binary(mach, inst, micro_umul, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4068 break; 4069 4070 case TGSI_OPCODE_USEQ: 4071 exec_vector_binary(mach, inst, micro_useq, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4072 break; 4073 4074 case TGSI_OPCODE_USGE: 4075 exec_vector_binary(mach, inst, micro_usge, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4076 break; 4077 4078 case TGSI_OPCODE_USHR: 4079 exec_vector_binary(mach, inst, micro_ushr, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4080 break; 4081 4082 case TGSI_OPCODE_USLT: 4083 exec_vector_binary(mach, inst, micro_uslt, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4084 break; 4085 4086 case TGSI_OPCODE_USNE: 4087 exec_vector_binary(mach, inst, micro_usne, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4088 break; 4089 4090 case TGSI_OPCODE_SWITCH: 4091 exec_switch(mach, inst); 4092 break; 4093 4094 case TGSI_OPCODE_CASE: 4095 exec_case(mach, inst); 4096 break; 4097 4098 case TGSI_OPCODE_DEFAULT: 4099 exec_default(mach); 4100 break; 4101 4102 case TGSI_OPCODE_ENDSWITCH: 4103 exec_endswitch(mach); 4104 break; 4105 4106 case TGSI_OPCODE_LOAD: 4107 assert(0); 4108 break; 4109 4110 case TGSI_OPCODE_LOAD_MS: 4111 assert(0); 4112 break; 4113 4114 case TGSI_OPCODE_SAMPLE: 4115 exec_sample(mach, inst, TEX_MODIFIER_NONE); 4116 break; 4117 4118 case TGSI_OPCODE_SAMPLE_B: 4119 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS); 4120 break; 4121 4122 case TGSI_OPCODE_SAMPLE_C: 4123 exec_sample(mach, inst, TEX_MODIFIER_NONE); 4124 break; 4125 4126 case TGSI_OPCODE_SAMPLE_C_LZ: 4127 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS); 4128 break; 4129 4130 case TGSI_OPCODE_SAMPLE_D: 4131 exec_sample_d(mach, inst); 4132 break; 4133 4134 case TGSI_OPCODE_SAMPLE_L: 4135 exec_sample(mach, inst, TEX_MODIFIER_EXPLICIT_LOD); 4136 break; 4137 4138 case TGSI_OPCODE_GATHER4: 4139 assert(0); 4140 break; 4141 4142 case TGSI_OPCODE_RESINFO: 4143 assert(0); 4144 break; 4145 4146 case TGSI_OPCODE_SAMPLE_POS: 4147 assert(0); 4148 break; 4149 4150 case TGSI_OPCODE_SAMPLE_INFO: 4151 assert(0); 4152 break; 4153 4154 case TGSI_OPCODE_UARL: 4155 exec_vector_unary(mach, inst, micro_uarl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_UINT); 4156 break; 4157 4158 case TGSI_OPCODE_UCMP: 4159 exec_vector_trinary(mach, inst, micro_ucmp, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4160 break; 4161 4162 default: 4163 assert( 0 ); 4164 } 4165} 4166 4167 4168#define DEBUG_EXECUTION 0 4169 4170 4171/** 4172 * Run TGSI interpreter. 4173 * \return bitmask of "alive" quad components 4174 */ 4175uint 4176tgsi_exec_machine_run( struct tgsi_exec_machine *mach ) 4177{ 4178 uint i; 4179 int pc = 0; 4180 4181 mach->CondMask = 0xf; 4182 mach->LoopMask = 0xf; 4183 mach->ContMask = 0xf; 4184 mach->FuncMask = 0xf; 4185 mach->ExecMask = 0xf; 4186 4187 mach->Switch.mask = 0xf; 4188 4189 assert(mach->CondStackTop == 0); 4190 assert(mach->LoopStackTop == 0); 4191 assert(mach->ContStackTop == 0); 4192 assert(mach->SwitchStackTop == 0); 4193 assert(mach->BreakStackTop == 0); 4194 assert(mach->CallStackTop == 0); 4195 4196 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0; 4197 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0; 4198 4199 if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) { 4200 mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0; 4201 mach->Primitives[0] = 0; 4202 } 4203 4204 /* execute declarations (interpolants) */ 4205 for (i = 0; i < mach->NumDeclarations; i++) { 4206 exec_declaration( mach, mach->Declarations+i ); 4207 } 4208 4209 { 4210#if DEBUG_EXECUTION 4211 struct tgsi_exec_vector temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS]; 4212 struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS]; 4213 uint inst = 1; 4214 4215 memcpy(temps, mach->Temps, sizeof(temps)); 4216 memcpy(outputs, mach->Outputs, sizeof(outputs)); 4217#endif 4218 4219 /* execute instructions, until pc is set to -1 */ 4220 while (pc != -1) { 4221 4222#if DEBUG_EXECUTION 4223 uint i; 4224 4225 tgsi_dump_instruction(&mach->Instructions[pc], inst++); 4226#endif 4227 4228 assert(pc < (int) mach->NumInstructions); 4229 exec_instruction(mach, mach->Instructions + pc, &pc); 4230 4231#if DEBUG_EXECUTION 4232 for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) { 4233 if (memcmp(&temps[i], &mach->Temps[i], sizeof(temps[i]))) { 4234 uint j; 4235 4236 memcpy(&temps[i], &mach->Temps[i], sizeof(temps[i])); 4237 debug_printf("TEMP[%2u] = ", i); 4238 for (j = 0; j < 4; j++) { 4239 if (j > 0) { 4240 debug_printf(" "); 4241 } 4242 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n", 4243 temps[i].xyzw[0].f[j], temps[i].xyzw[0].u[j], 4244 temps[i].xyzw[1].f[j], temps[i].xyzw[1].u[j], 4245 temps[i].xyzw[2].f[j], temps[i].xyzw[2].u[j], 4246 temps[i].xyzw[3].f[j], temps[i].xyzw[3].u[j]); 4247 } 4248 } 4249 } 4250 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { 4251 if (memcmp(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]))) { 4252 uint j; 4253 4254 memcpy(&outputs[i], &mach->Outputs[i], sizeof(outputs[i])); 4255 debug_printf("OUT[%2u] = ", i); 4256 for (j = 0; j < 4; j++) { 4257 if (j > 0) { 4258 debug_printf(" "); 4259 } 4260 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n", 4261 outputs[i].xyzw[0].f[j], outputs[i].xyzw[0].u[j], 4262 outputs[i].xyzw[1].f[j], outputs[i].xyzw[1].u[j], 4263 outputs[i].xyzw[2].f[j], outputs[i].xyzw[2].u[j], 4264 outputs[i].xyzw[3].f[j], outputs[i].xyzw[3].u[j]); 4265 } 4266 } 4267 } 4268#endif 4269 } 4270 } 4271 4272#if 0 4273 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */ 4274 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { 4275 /* 4276 * Scale back depth component. 4277 */ 4278 for (i = 0; i < 4; i++) 4279 mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF; 4280 } 4281#endif 4282 4283 /* Strictly speaking, these assertions aren't really needed but they 4284 * can potentially catch some bugs in the control flow code. 4285 */ 4286 assert(mach->CondStackTop == 0); 4287 assert(mach->LoopStackTop == 0); 4288 assert(mach->ContStackTop == 0); 4289 assert(mach->SwitchStackTop == 0); 4290 assert(mach->BreakStackTop == 0); 4291 assert(mach->CallStackTop == 0); 4292 4293 return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0]; 4294} 4295