tgsi_exec.c revision 2083a276eb270b748d1c2668eb9faa5aadc8e700
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 align_malloc(sizeof(struct tgsi_exec_vector) * 679 TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS, 680 16); 681 struct tgsi_exec_vector *outputs = 682 align_malloc(sizeof(struct tgsi_exec_vector) * 683 TGSI_MAX_TOTAL_VERTICES, 16); 684 685 if (!inputs) 686 return; 687 if (!outputs) { 688 align_free(inputs); 689 return; 690 } 691 692 align_free(mach->Inputs); 693 align_free(mach->Outputs); 694 695 mach->Inputs = inputs; 696 mach->Outputs = outputs; 697 mach->UsedGeometryShader = TRUE; 698 } 699 700 declarations = (struct tgsi_full_declaration *) 701 MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) ); 702 703 if (!declarations) { 704 return; 705 } 706 707 instructions = (struct tgsi_full_instruction *) 708 MALLOC( maxInstructions * sizeof(struct tgsi_full_instruction) ); 709 710 if (!instructions) { 711 FREE( declarations ); 712 return; 713 } 714 715 while( !tgsi_parse_end_of_tokens( &parse ) ) { 716 uint i; 717 718 tgsi_parse_token( &parse ); 719 switch( parse.FullToken.Token.Type ) { 720 case TGSI_TOKEN_TYPE_DECLARATION: 721 /* save expanded declaration */ 722 if (numDeclarations == maxDeclarations) { 723 declarations = REALLOC(declarations, 724 maxDeclarations 725 * sizeof(struct tgsi_full_declaration), 726 (maxDeclarations + 10) 727 * sizeof(struct tgsi_full_declaration)); 728 maxDeclarations += 10; 729 } 730 if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_OUTPUT) { 731 unsigned reg; 732 for (reg = parse.FullToken.FullDeclaration.Range.First; 733 reg <= parse.FullToken.FullDeclaration.Range.Last; 734 ++reg) { 735 ++mach->NumOutputs; 736 } 737 } 738 if (parse.FullToken.FullDeclaration.Declaration.File == 739 TGSI_FILE_IMMEDIATE_ARRAY) { 740 unsigned reg; 741 struct tgsi_full_declaration *decl = 742 &parse.FullToken.FullDeclaration; 743 debug_assert(decl->Range.Last < TGSI_EXEC_NUM_IMMEDIATES); 744 for (reg = decl->Range.First; reg <= decl->Range.Last; ++reg) { 745 for( i = 0; i < 4; i++ ) { 746 int idx = reg * 4 + i; 747 mach->ImmArray[reg][i] = decl->ImmediateData.u[idx].Float; 748 } 749 } 750 } 751 memcpy(declarations + numDeclarations, 752 &parse.FullToken.FullDeclaration, 753 sizeof(declarations[0])); 754 numDeclarations++; 755 break; 756 757 case TGSI_TOKEN_TYPE_IMMEDIATE: 758 { 759 uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1; 760 assert( size <= 4 ); 761 assert( mach->ImmLimit + 1 <= TGSI_EXEC_NUM_IMMEDIATES ); 762 763 for( i = 0; i < size; i++ ) { 764 mach->Imms[mach->ImmLimit][i] = 765 parse.FullToken.FullImmediate.u[i].Float; 766 } 767 mach->ImmLimit += 1; 768 } 769 break; 770 771 case TGSI_TOKEN_TYPE_INSTRUCTION: 772 773 /* save expanded instruction */ 774 if (numInstructions == maxInstructions) { 775 instructions = REALLOC(instructions, 776 maxInstructions 777 * sizeof(struct tgsi_full_instruction), 778 (maxInstructions + 10) 779 * sizeof(struct tgsi_full_instruction)); 780 maxInstructions += 10; 781 } 782 783 memcpy(instructions + numInstructions, 784 &parse.FullToken.FullInstruction, 785 sizeof(instructions[0])); 786 787 numInstructions++; 788 break; 789 790 case TGSI_TOKEN_TYPE_PROPERTY: 791 break; 792 793 default: 794 assert( 0 ); 795 } 796 } 797 tgsi_parse_free (&parse); 798 799 if (mach->Declarations) { 800 FREE( mach->Declarations ); 801 } 802 mach->Declarations = declarations; 803 mach->NumDeclarations = numDeclarations; 804 805 if (mach->Instructions) { 806 FREE( mach->Instructions ); 807 } 808 mach->Instructions = instructions; 809 mach->NumInstructions = numInstructions; 810} 811 812 813struct tgsi_exec_machine * 814tgsi_exec_machine_create( void ) 815{ 816 struct tgsi_exec_machine *mach; 817 uint i; 818 819 mach = align_malloc( sizeof *mach, 16 ); 820 if (!mach) 821 goto fail; 822 823 memset(mach, 0, sizeof(*mach)); 824 825 mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR]; 826 mach->MaxGeometryShaderOutputs = TGSI_MAX_TOTAL_VERTICES; 827 mach->Predicates = &mach->Temps[TGSI_EXEC_TEMP_P0]; 828 829 mach->Inputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16); 830 mach->Outputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16); 831 if (!mach->Inputs || !mach->Outputs) 832 goto fail; 833 834 /* Setup constants needed by the SSE2 executor. */ 835 for( i = 0; i < 4; i++ ) { 836 mach->Temps[TGSI_EXEC_TEMP_00000000_I].xyzw[TGSI_EXEC_TEMP_00000000_C].u[i] = 0x00000000; 837 mach->Temps[TGSI_EXEC_TEMP_7FFFFFFF_I].xyzw[TGSI_EXEC_TEMP_7FFFFFFF_C].u[i] = 0x7FFFFFFF; 838 mach->Temps[TGSI_EXEC_TEMP_80000000_I].xyzw[TGSI_EXEC_TEMP_80000000_C].u[i] = 0x80000000; 839 mach->Temps[TGSI_EXEC_TEMP_FFFFFFFF_I].xyzw[TGSI_EXEC_TEMP_FFFFFFFF_C].u[i] = 0xFFFFFFFF; /* not used */ 840 mach->Temps[TGSI_EXEC_TEMP_ONE_I].xyzw[TGSI_EXEC_TEMP_ONE_C].f[i] = 1.0f; 841 mach->Temps[TGSI_EXEC_TEMP_TWO_I].xyzw[TGSI_EXEC_TEMP_TWO_C].f[i] = 2.0f; /* not used */ 842 mach->Temps[TGSI_EXEC_TEMP_128_I].xyzw[TGSI_EXEC_TEMP_128_C].f[i] = 128.0f; 843 mach->Temps[TGSI_EXEC_TEMP_MINUS_128_I].xyzw[TGSI_EXEC_TEMP_MINUS_128_C].f[i] = -128.0f; 844 mach->Temps[TGSI_EXEC_TEMP_THREE_I].xyzw[TGSI_EXEC_TEMP_THREE_C].f[i] = 3.0f; 845 mach->Temps[TGSI_EXEC_TEMP_HALF_I].xyzw[TGSI_EXEC_TEMP_HALF_C].f[i] = 0.5f; 846 } 847 848#ifdef DEBUG 849 /* silence warnings */ 850 (void) print_chan; 851 (void) print_temp; 852#endif 853 854 return mach; 855 856fail: 857 if (mach) { 858 align_free(mach->Inputs); 859 align_free(mach->Outputs); 860 align_free(mach); 861 } 862 return NULL; 863} 864 865 866void 867tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach) 868{ 869 if (mach) { 870 if (mach->Instructions) 871 FREE(mach->Instructions); 872 if (mach->Declarations) 873 FREE(mach->Declarations); 874 875 align_free(mach->Inputs); 876 align_free(mach->Outputs); 877 878 align_free(mach); 879 } 880} 881 882static void 883micro_add(union tgsi_exec_channel *dst, 884 const union tgsi_exec_channel *src0, 885 const union tgsi_exec_channel *src1) 886{ 887 dst->f[0] = src0->f[0] + src1->f[0]; 888 dst->f[1] = src0->f[1] + src1->f[1]; 889 dst->f[2] = src0->f[2] + src1->f[2]; 890 dst->f[3] = src0->f[3] + src1->f[3]; 891} 892 893static void 894micro_div( 895 union tgsi_exec_channel *dst, 896 const union tgsi_exec_channel *src0, 897 const union tgsi_exec_channel *src1 ) 898{ 899 if (src1->f[0] != 0) { 900 dst->f[0] = src0->f[0] / src1->f[0]; 901 } 902 if (src1->f[1] != 0) { 903 dst->f[1] = src0->f[1] / src1->f[1]; 904 } 905 if (src1->f[2] != 0) { 906 dst->f[2] = src0->f[2] / src1->f[2]; 907 } 908 if (src1->f[3] != 0) { 909 dst->f[3] = src0->f[3] / src1->f[3]; 910 } 911} 912 913static void 914micro_rcc(union tgsi_exec_channel *dst, 915 const union tgsi_exec_channel *src) 916{ 917 uint i; 918 919 for (i = 0; i < 4; i++) { 920 float recip = 1.0f / src->f[i]; 921 922 if (recip > 0.0f) { 923 if (recip > 1.884467e+019f) { 924 dst->f[i] = 1.884467e+019f; 925 } 926 else if (recip < 5.42101e-020f) { 927 dst->f[i] = 5.42101e-020f; 928 } 929 else { 930 dst->f[i] = recip; 931 } 932 } 933 else { 934 if (recip < -1.884467e+019f) { 935 dst->f[i] = -1.884467e+019f; 936 } 937 else if (recip > -5.42101e-020f) { 938 dst->f[i] = -5.42101e-020f; 939 } 940 else { 941 dst->f[i] = recip; 942 } 943 } 944 } 945} 946 947static void 948micro_lt( 949 union tgsi_exec_channel *dst, 950 const union tgsi_exec_channel *src0, 951 const union tgsi_exec_channel *src1, 952 const union tgsi_exec_channel *src2, 953 const union tgsi_exec_channel *src3 ) 954{ 955 dst->f[0] = src0->f[0] < src1->f[0] ? src2->f[0] : src3->f[0]; 956 dst->f[1] = src0->f[1] < src1->f[1] ? src2->f[1] : src3->f[1]; 957 dst->f[2] = src0->f[2] < src1->f[2] ? src2->f[2] : src3->f[2]; 958 dst->f[3] = src0->f[3] < src1->f[3] ? src2->f[3] : src3->f[3]; 959} 960 961static void 962micro_max(union tgsi_exec_channel *dst, 963 const union tgsi_exec_channel *src0, 964 const union tgsi_exec_channel *src1) 965{ 966 dst->f[0] = src0->f[0] > src1->f[0] ? src0->f[0] : src1->f[0]; 967 dst->f[1] = src0->f[1] > src1->f[1] ? src0->f[1] : src1->f[1]; 968 dst->f[2] = src0->f[2] > src1->f[2] ? src0->f[2] : src1->f[2]; 969 dst->f[3] = src0->f[3] > src1->f[3] ? src0->f[3] : src1->f[3]; 970} 971 972static void 973micro_min(union tgsi_exec_channel *dst, 974 const union tgsi_exec_channel *src0, 975 const union tgsi_exec_channel *src1) 976{ 977 dst->f[0] = src0->f[0] < src1->f[0] ? src0->f[0] : src1->f[0]; 978 dst->f[1] = src0->f[1] < src1->f[1] ? src0->f[1] : src1->f[1]; 979 dst->f[2] = src0->f[2] < src1->f[2] ? src0->f[2] : src1->f[2]; 980 dst->f[3] = src0->f[3] < src1->f[3] ? src0->f[3] : src1->f[3]; 981} 982 983static void 984micro_mul(union tgsi_exec_channel *dst, 985 const union tgsi_exec_channel *src0, 986 const union tgsi_exec_channel *src1) 987{ 988 dst->f[0] = src0->f[0] * src1->f[0]; 989 dst->f[1] = src0->f[1] * src1->f[1]; 990 dst->f[2] = src0->f[2] * src1->f[2]; 991 dst->f[3] = src0->f[3] * src1->f[3]; 992} 993 994static void 995micro_neg( 996 union tgsi_exec_channel *dst, 997 const union tgsi_exec_channel *src ) 998{ 999 dst->f[0] = -src->f[0]; 1000 dst->f[1] = -src->f[1]; 1001 dst->f[2] = -src->f[2]; 1002 dst->f[3] = -src->f[3]; 1003} 1004 1005static void 1006micro_pow( 1007 union tgsi_exec_channel *dst, 1008 const union tgsi_exec_channel *src0, 1009 const union tgsi_exec_channel *src1 ) 1010{ 1011#if FAST_MATH 1012 dst->f[0] = util_fast_pow( src0->f[0], src1->f[0] ); 1013 dst->f[1] = util_fast_pow( src0->f[1], src1->f[1] ); 1014 dst->f[2] = util_fast_pow( src0->f[2], src1->f[2] ); 1015 dst->f[3] = util_fast_pow( src0->f[3], src1->f[3] ); 1016#else 1017 dst->f[0] = powf( src0->f[0], src1->f[0] ); 1018 dst->f[1] = powf( src0->f[1], src1->f[1] ); 1019 dst->f[2] = powf( src0->f[2], src1->f[2] ); 1020 dst->f[3] = powf( src0->f[3], src1->f[3] ); 1021#endif 1022} 1023 1024static void 1025micro_sub(union tgsi_exec_channel *dst, 1026 const union tgsi_exec_channel *src0, 1027 const union tgsi_exec_channel *src1) 1028{ 1029 dst->f[0] = src0->f[0] - src1->f[0]; 1030 dst->f[1] = src0->f[1] - src1->f[1]; 1031 dst->f[2] = src0->f[2] - src1->f[2]; 1032 dst->f[3] = src0->f[3] - src1->f[3]; 1033} 1034 1035static void 1036fetch_src_file_channel(const struct tgsi_exec_machine *mach, 1037 const uint file, 1038 const uint swizzle, 1039 const union tgsi_exec_channel *index, 1040 const union tgsi_exec_channel *index2D, 1041 union tgsi_exec_channel *chan) 1042{ 1043 uint i; 1044 1045 assert(swizzle < 4); 1046 1047 switch (file) { 1048 case TGSI_FILE_CONSTANT: 1049 for (i = 0; i < QUAD_SIZE; i++) { 1050 assert(index2D->i[i] >= 0 && index2D->i[i] < PIPE_MAX_CONSTANT_BUFFERS); 1051 assert(mach->Consts[index2D->i[i]]); 1052 1053 if (index->i[i] < 0) { 1054 chan->u[i] = 0; 1055 } else { 1056 /* NOTE: copying the const value as a uint instead of float */ 1057 const uint constbuf = index2D->i[i]; 1058 const uint *buf = (const uint *)mach->Consts[constbuf]; 1059 const int pos = index->i[i] * 4 + swizzle; 1060 /* const buffer bounds check */ 1061 if (pos < 0 || pos >= mach->ConstsSize[constbuf]) { 1062 if (0) { 1063 /* Debug: print warning */ 1064 static int count = 0; 1065 if (count++ < 100) 1066 debug_printf("TGSI Exec: const buffer index %d" 1067 " out of bounds\n", pos); 1068 } 1069 chan->u[i] = 0; 1070 } 1071 else 1072 chan->u[i] = buf[pos]; 1073 } 1074 } 1075 break; 1076 1077 case TGSI_FILE_INPUT: 1078 for (i = 0; i < QUAD_SIZE; i++) { 1079 /* 1080 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1081 debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n", 1082 index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i], 1083 index2D->i[i], index->i[i]); 1084 }*/ 1085 int pos = index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i]; 1086 assert(pos >= 0); 1087 assert(pos < TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS); 1088 chan->u[i] = mach->Inputs[pos].xyzw[swizzle].u[i]; 1089 } 1090 break; 1091 1092 case TGSI_FILE_SYSTEM_VALUE: 1093 /* XXX no swizzling at this point. Will be needed if we put 1094 * gl_FragCoord, for example, in a sys value register. 1095 */ 1096 for (i = 0; i < QUAD_SIZE; i++) { 1097 chan->f[i] = mach->SystemValue[index->i[i]][0]; 1098 } 1099 break; 1100 1101 case TGSI_FILE_TEMPORARY: 1102 for (i = 0; i < QUAD_SIZE; i++) { 1103 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); 1104 assert(index2D->i[i] == 0); 1105 1106 chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i]; 1107 } 1108 break; 1109 1110 case TGSI_FILE_TEMPORARY_ARRAY: 1111 for (i = 0; i < QUAD_SIZE; i++) { 1112 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); 1113 assert(index2D->i[i] < TGSI_EXEC_NUM_TEMP_ARRAYS); 1114 1115 chan->u[i] = 1116 mach->TempArray[index2D->i[i]][index->i[i]].xyzw[swizzle].u[i]; 1117 } 1118 break; 1119 1120 case TGSI_FILE_IMMEDIATE: 1121 for (i = 0; i < QUAD_SIZE; i++) { 1122 assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit); 1123 assert(index2D->i[i] == 0); 1124 1125 chan->f[i] = mach->Imms[index->i[i]][swizzle]; 1126 } 1127 break; 1128 1129 case TGSI_FILE_IMMEDIATE_ARRAY: 1130 for (i = 0; i < QUAD_SIZE; i++) { 1131 assert(index2D->i[i] == 0); 1132 1133 chan->f[i] = mach->ImmArray[index->i[i]][swizzle]; 1134 } 1135 break; 1136 1137 case TGSI_FILE_ADDRESS: 1138 for (i = 0; i < QUAD_SIZE; i++) { 1139 assert(index->i[i] >= 0); 1140 assert(index2D->i[i] == 0); 1141 1142 chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i]; 1143 } 1144 break; 1145 1146 case TGSI_FILE_PREDICATE: 1147 for (i = 0; i < QUAD_SIZE; i++) { 1148 assert(index->i[i] >= 0 && index->i[i] < TGSI_EXEC_NUM_PREDS); 1149 assert(index2D->i[i] == 0); 1150 1151 chan->u[i] = mach->Predicates[0].xyzw[swizzle].u[i]; 1152 } 1153 break; 1154 1155 case TGSI_FILE_OUTPUT: 1156 /* vertex/fragment output vars can be read too */ 1157 for (i = 0; i < QUAD_SIZE; i++) { 1158 assert(index->i[i] >= 0); 1159 assert(index2D->i[i] == 0); 1160 1161 chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i]; 1162 } 1163 break; 1164 1165 default: 1166 assert(0); 1167 for (i = 0; i < QUAD_SIZE; i++) { 1168 chan->u[i] = 0; 1169 } 1170 } 1171} 1172 1173static void 1174fetch_source(const struct tgsi_exec_machine *mach, 1175 union tgsi_exec_channel *chan, 1176 const struct tgsi_full_src_register *reg, 1177 const uint chan_index, 1178 enum tgsi_exec_datatype src_datatype) 1179{ 1180 union tgsi_exec_channel index; 1181 union tgsi_exec_channel index2D; 1182 uint swizzle; 1183 1184 /* We start with a direct index into a register file. 1185 * 1186 * file[1], 1187 * where: 1188 * file = Register.File 1189 * [1] = Register.Index 1190 */ 1191 index.i[0] = 1192 index.i[1] = 1193 index.i[2] = 1194 index.i[3] = reg->Register.Index; 1195 1196 /* There is an extra source register that indirectly subscripts 1197 * a register file. The direct index now becomes an offset 1198 * that is being added to the indirect register. 1199 * 1200 * file[ind[2].x+1], 1201 * where: 1202 * ind = Indirect.File 1203 * [2] = Indirect.Index 1204 * .x = Indirect.SwizzleX 1205 */ 1206 if (reg->Register.Indirect) { 1207 union tgsi_exec_channel index2; 1208 union tgsi_exec_channel indir_index; 1209 const uint execmask = mach->ExecMask; 1210 uint i; 1211 1212 /* which address register (always zero now) */ 1213 index2.i[0] = 1214 index2.i[1] = 1215 index2.i[2] = 1216 index2.i[3] = reg->Indirect.Index; 1217 assert(reg->Indirect.File == TGSI_FILE_ADDRESS); 1218 /* get current value of address register[swizzle] */ 1219 swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, CHAN_X ); 1220 fetch_src_file_channel(mach, 1221 reg->Indirect.File, 1222 swizzle, 1223 &index2, 1224 &ZeroVec, 1225 &indir_index); 1226 1227 /* add value of address register to the offset */ 1228 index.i[0] += indir_index.i[0]; 1229 index.i[1] += indir_index.i[1]; 1230 index.i[2] += indir_index.i[2]; 1231 index.i[3] += indir_index.i[3]; 1232 1233 /* for disabled execution channels, zero-out the index to 1234 * avoid using a potential garbage value. 1235 */ 1236 for (i = 0; i < QUAD_SIZE; i++) { 1237 if ((execmask & (1 << i)) == 0) 1238 index.i[i] = 0; 1239 } 1240 } 1241 1242 /* There is an extra source register that is a second 1243 * subscript to a register file. Effectively it means that 1244 * the register file is actually a 2D array of registers. 1245 * 1246 * file[3][1], 1247 * where: 1248 * [3] = Dimension.Index 1249 */ 1250 if (reg->Register.Dimension) { 1251 index2D.i[0] = 1252 index2D.i[1] = 1253 index2D.i[2] = 1254 index2D.i[3] = reg->Dimension.Index; 1255 1256 /* Again, the second subscript index can be addressed indirectly 1257 * identically to the first one. 1258 * Nothing stops us from indirectly addressing the indirect register, 1259 * but there is no need for that, so we won't exercise it. 1260 * 1261 * file[ind[4].y+3][1], 1262 * where: 1263 * ind = DimIndirect.File 1264 * [4] = DimIndirect.Index 1265 * .y = DimIndirect.SwizzleX 1266 */ 1267 if (reg->Dimension.Indirect) { 1268 union tgsi_exec_channel index2; 1269 union tgsi_exec_channel indir_index; 1270 const uint execmask = mach->ExecMask; 1271 uint i; 1272 1273 index2.i[0] = 1274 index2.i[1] = 1275 index2.i[2] = 1276 index2.i[3] = reg->DimIndirect.Index; 1277 1278 swizzle = tgsi_util_get_src_register_swizzle( ®->DimIndirect, CHAN_X ); 1279 fetch_src_file_channel(mach, 1280 reg->DimIndirect.File, 1281 swizzle, 1282 &index2, 1283 &ZeroVec, 1284 &indir_index); 1285 1286 index2D.i[0] += indir_index.i[0]; 1287 index2D.i[1] += indir_index.i[1]; 1288 index2D.i[2] += indir_index.i[2]; 1289 index2D.i[3] += indir_index.i[3]; 1290 1291 /* for disabled execution channels, zero-out the index to 1292 * avoid using a potential garbage value. 1293 */ 1294 for (i = 0; i < QUAD_SIZE; i++) { 1295 if ((execmask & (1 << i)) == 0) { 1296 index2D.i[i] = 0; 1297 } 1298 } 1299 } 1300 1301 /* If by any chance there was a need for a 3D array of register 1302 * files, we would have to check whether Dimension is followed 1303 * by a dimension register and continue the saga. 1304 */ 1305 } else { 1306 index2D.i[0] = 1307 index2D.i[1] = 1308 index2D.i[2] = 1309 index2D.i[3] = 0; 1310 } 1311 1312 swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index ); 1313 fetch_src_file_channel(mach, 1314 reg->Register.File, 1315 swizzle, 1316 &index, 1317 &index2D, 1318 chan); 1319 1320 if (reg->Register.Absolute) { 1321 if (src_datatype == TGSI_EXEC_DATA_FLOAT) { 1322 micro_abs(chan, chan); 1323 } else { 1324 micro_iabs(chan, chan); 1325 } 1326 } 1327 1328 if (reg->Register.Negate) { 1329 if (src_datatype == TGSI_EXEC_DATA_FLOAT) { 1330 micro_neg(chan, chan); 1331 } else { 1332 micro_ineg(chan, chan); 1333 } 1334 } 1335} 1336 1337static void 1338store_dest(struct tgsi_exec_machine *mach, 1339 const union tgsi_exec_channel *chan, 1340 const struct tgsi_full_dst_register *reg, 1341 const struct tgsi_full_instruction *inst, 1342 uint chan_index, 1343 enum tgsi_exec_datatype dst_datatype) 1344{ 1345 uint i; 1346 union tgsi_exec_channel null; 1347 union tgsi_exec_channel *dst; 1348 union tgsi_exec_channel index2D; 1349 uint execmask = mach->ExecMask; 1350 int offset = 0; /* indirection offset */ 1351 int index; 1352 1353 /* for debugging */ 1354 if (0 && dst_datatype == TGSI_EXEC_DATA_FLOAT) { 1355 check_inf_or_nan(chan); 1356 } 1357 1358 /* There is an extra source register that indirectly subscripts 1359 * a register file. The direct index now becomes an offset 1360 * that is being added to the indirect register. 1361 * 1362 * file[ind[2].x+1], 1363 * where: 1364 * ind = Indirect.File 1365 * [2] = Indirect.Index 1366 * .x = Indirect.SwizzleX 1367 */ 1368 if (reg->Register.Indirect) { 1369 union tgsi_exec_channel index; 1370 union tgsi_exec_channel indir_index; 1371 uint swizzle; 1372 1373 /* which address register (always zero for now) */ 1374 index.i[0] = 1375 index.i[1] = 1376 index.i[2] = 1377 index.i[3] = reg->Indirect.Index; 1378 1379 /* get current value of address register[swizzle] */ 1380 swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, CHAN_X ); 1381 1382 /* fetch values from the address/indirection register */ 1383 fetch_src_file_channel(mach, 1384 reg->Indirect.File, 1385 swizzle, 1386 &index, 1387 &ZeroVec, 1388 &indir_index); 1389 1390 /* save indirection offset */ 1391 offset = indir_index.i[0]; 1392 } 1393 1394 /* There is an extra source register that is a second 1395 * subscript to a register file. Effectively it means that 1396 * the register file is actually a 2D array of registers. 1397 * 1398 * file[3][1], 1399 * where: 1400 * [3] = Dimension.Index 1401 */ 1402 if (reg->Register.Dimension) { 1403 index2D.i[0] = 1404 index2D.i[1] = 1405 index2D.i[2] = 1406 index2D.i[3] = reg->Dimension.Index; 1407 1408 /* Again, the second subscript index can be addressed indirectly 1409 * identically to the first one. 1410 * Nothing stops us from indirectly addressing the indirect register, 1411 * but there is no need for that, so we won't exercise it. 1412 * 1413 * file[ind[4].y+3][1], 1414 * where: 1415 * ind = DimIndirect.File 1416 * [4] = DimIndirect.Index 1417 * .y = DimIndirect.SwizzleX 1418 */ 1419 if (reg->Dimension.Indirect) { 1420 union tgsi_exec_channel index2; 1421 union tgsi_exec_channel indir_index; 1422 const uint execmask = mach->ExecMask; 1423 unsigned swizzle; 1424 uint i; 1425 1426 index2.i[0] = 1427 index2.i[1] = 1428 index2.i[2] = 1429 index2.i[3] = reg->DimIndirect.Index; 1430 1431 swizzle = tgsi_util_get_src_register_swizzle( ®->DimIndirect, CHAN_X ); 1432 fetch_src_file_channel(mach, 1433 reg->DimIndirect.File, 1434 swizzle, 1435 &index2, 1436 &ZeroVec, 1437 &indir_index); 1438 1439 index2D.i[0] += indir_index.i[0]; 1440 index2D.i[1] += indir_index.i[1]; 1441 index2D.i[2] += indir_index.i[2]; 1442 index2D.i[3] += indir_index.i[3]; 1443 1444 /* for disabled execution channels, zero-out the index to 1445 * avoid using a potential garbage value. 1446 */ 1447 for (i = 0; i < QUAD_SIZE; i++) { 1448 if ((execmask & (1 << i)) == 0) { 1449 index2D.i[i] = 0; 1450 } 1451 } 1452 } 1453 1454 /* If by any chance there was a need for a 3D array of register 1455 * files, we would have to check whether Dimension is followed 1456 * by a dimension register and continue the saga. 1457 */ 1458 } else { 1459 index2D.i[0] = 1460 index2D.i[1] = 1461 index2D.i[2] = 1462 index2D.i[3] = 0; 1463 } 1464 1465 switch (reg->Register.File) { 1466 case TGSI_FILE_NULL: 1467 dst = &null; 1468 break; 1469 1470 case TGSI_FILE_OUTPUT: 1471 index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] 1472 + reg->Register.Index; 1473 dst = &mach->Outputs[offset + index].xyzw[chan_index]; 1474#if 0 1475 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1476 fprintf(stderr, "STORING OUT[%d] mask(%d), = (", offset + index, execmask); 1477 for (i = 0; i < QUAD_SIZE; i++) 1478 if (execmask & (1 << i)) 1479 fprintf(stderr, "%f, ", chan->f[i]); 1480 fprintf(stderr, ")\n"); 1481 } 1482#endif 1483 break; 1484 1485 case TGSI_FILE_TEMPORARY: 1486 index = reg->Register.Index; 1487 assert( index < TGSI_EXEC_NUM_TEMPS ); 1488 dst = &mach->Temps[offset + index].xyzw[chan_index]; 1489 break; 1490 1491 case TGSI_FILE_TEMPORARY_ARRAY: 1492 index = reg->Register.Index; 1493 assert( index < TGSI_EXEC_NUM_TEMPS ); 1494 assert( index2D.i[0] < TGSI_EXEC_NUM_TEMP_ARRAYS ); 1495 /* XXX we use index2D.i[0] here but somehow we might 1496 * end up with someone trying to store indirectly in 1497 * different buffers */ 1498 dst = &mach->TempArray[index2D.i[0]][offset + index].xyzw[chan_index]; 1499 break; 1500 1501 case TGSI_FILE_ADDRESS: 1502 index = reg->Register.Index; 1503 dst = &mach->Addrs[index].xyzw[chan_index]; 1504 break; 1505 1506 case TGSI_FILE_PREDICATE: 1507 index = reg->Register.Index; 1508 assert(index < TGSI_EXEC_NUM_PREDS); 1509 dst = &mach->Predicates[index].xyzw[chan_index]; 1510 break; 1511 1512 default: 1513 assert( 0 ); 1514 return; 1515 } 1516 1517 if (inst->Instruction.Predicate) { 1518 uint swizzle; 1519 union tgsi_exec_channel *pred; 1520 1521 switch (chan_index) { 1522 case CHAN_X: 1523 swizzle = inst->Predicate.SwizzleX; 1524 break; 1525 case CHAN_Y: 1526 swizzle = inst->Predicate.SwizzleY; 1527 break; 1528 case CHAN_Z: 1529 swizzle = inst->Predicate.SwizzleZ; 1530 break; 1531 case CHAN_W: 1532 swizzle = inst->Predicate.SwizzleW; 1533 break; 1534 default: 1535 assert(0); 1536 return; 1537 } 1538 1539 assert(inst->Predicate.Index == 0); 1540 1541 pred = &mach->Predicates[inst->Predicate.Index].xyzw[swizzle]; 1542 1543 if (inst->Predicate.Negate) { 1544 for (i = 0; i < QUAD_SIZE; i++) { 1545 if (pred->u[i]) { 1546 execmask &= ~(1 << i); 1547 } 1548 } 1549 } else { 1550 for (i = 0; i < QUAD_SIZE; i++) { 1551 if (!pred->u[i]) { 1552 execmask &= ~(1 << i); 1553 } 1554 } 1555 } 1556 } 1557 1558 switch (inst->Instruction.Saturate) { 1559 case TGSI_SAT_NONE: 1560 for (i = 0; i < QUAD_SIZE; i++) 1561 if (execmask & (1 << i)) 1562 dst->i[i] = chan->i[i]; 1563 break; 1564 1565 case TGSI_SAT_ZERO_ONE: 1566 for (i = 0; i < QUAD_SIZE; i++) 1567 if (execmask & (1 << i)) { 1568 if (chan->f[i] < 0.0f) 1569 dst->f[i] = 0.0f; 1570 else if (chan->f[i] > 1.0f) 1571 dst->f[i] = 1.0f; 1572 else 1573 dst->i[i] = chan->i[i]; 1574 } 1575 break; 1576 1577 case TGSI_SAT_MINUS_PLUS_ONE: 1578 for (i = 0; i < QUAD_SIZE; i++) 1579 if (execmask & (1 << i)) { 1580 if (chan->f[i] < -1.0f) 1581 dst->f[i] = -1.0f; 1582 else if (chan->f[i] > 1.0f) 1583 dst->f[i] = 1.0f; 1584 else 1585 dst->i[i] = chan->i[i]; 1586 } 1587 break; 1588 1589 default: 1590 assert( 0 ); 1591 } 1592} 1593 1594#define FETCH(VAL,INDEX,CHAN)\ 1595 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT) 1596 1597#define IFETCH(VAL,INDEX,CHAN)\ 1598 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT) 1599 1600 1601/** 1602 * Execute ARB-style KIL which is predicated by a src register. 1603 * Kill fragment if any of the four values is less than zero. 1604 */ 1605static void 1606exec_kil(struct tgsi_exec_machine *mach, 1607 const struct tgsi_full_instruction *inst) 1608{ 1609 uint uniquemask; 1610 uint chan_index; 1611 uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ 1612 union tgsi_exec_channel r[1]; 1613 1614 /* This mask stores component bits that were already tested. */ 1615 uniquemask = 0; 1616 1617 for (chan_index = 0; chan_index < 4; chan_index++) 1618 { 1619 uint swizzle; 1620 uint i; 1621 1622 /* unswizzle channel */ 1623 swizzle = tgsi_util_get_full_src_register_swizzle ( 1624 &inst->Src[0], 1625 chan_index); 1626 1627 /* check if the component has not been already tested */ 1628 if (uniquemask & (1 << swizzle)) 1629 continue; 1630 uniquemask |= 1 << swizzle; 1631 1632 FETCH(&r[0], 0, chan_index); 1633 for (i = 0; i < 4; i++) 1634 if (r[0].f[i] < 0.0f) 1635 kilmask |= 1 << i; 1636 } 1637 1638 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; 1639} 1640 1641/** 1642 * Execute NVIDIA-style KIL which is predicated by a condition code. 1643 * Kill fragment if the condition code is TRUE. 1644 */ 1645static void 1646exec_kilp(struct tgsi_exec_machine *mach, 1647 const struct tgsi_full_instruction *inst) 1648{ 1649 uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ 1650 1651 /* "unconditional" kil */ 1652 kilmask = mach->ExecMask; 1653 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; 1654} 1655 1656static void 1657emit_vertex(struct tgsi_exec_machine *mach) 1658{ 1659 /* FIXME: check for exec mask correctly 1660 unsigned i; 1661 for (i = 0; i < QUAD_SIZE; ++i) { 1662 if ((mach->ExecMask & (1 << i))) 1663 */ 1664 if (mach->ExecMask) { 1665 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += mach->NumOutputs; 1666 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++; 1667 } 1668} 1669 1670static void 1671emit_primitive(struct tgsi_exec_machine *mach) 1672{ 1673 unsigned *prim_count = &mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]; 1674 /* FIXME: check for exec mask correctly 1675 unsigned i; 1676 for (i = 0; i < QUAD_SIZE; ++i) { 1677 if ((mach->ExecMask & (1 << i))) 1678 */ 1679 if (mach->ExecMask) { 1680 ++(*prim_count); 1681 debug_assert((*prim_count * mach->NumOutputs) < mach->MaxGeometryShaderOutputs); 1682 mach->Primitives[*prim_count] = 0; 1683 } 1684} 1685 1686static void 1687conditional_emit_primitive(struct tgsi_exec_machine *mach) 1688{ 1689 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1690 int emitted_verts = 1691 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]; 1692 if (emitted_verts) { 1693 emit_primitive(mach); 1694 } 1695 } 1696} 1697 1698 1699/* 1700 * Fetch four texture samples using STR texture coordinates. 1701 */ 1702static void 1703fetch_texel( struct tgsi_sampler *sampler, 1704 const union tgsi_exec_channel *s, 1705 const union tgsi_exec_channel *t, 1706 const union tgsi_exec_channel *p, 1707 const union tgsi_exec_channel *c0, 1708 enum tgsi_sampler_control control, 1709 union tgsi_exec_channel *r, 1710 union tgsi_exec_channel *g, 1711 union tgsi_exec_channel *b, 1712 union tgsi_exec_channel *a ) 1713{ 1714 uint j; 1715 float rgba[NUM_CHANNELS][QUAD_SIZE]; 1716 1717 sampler->get_samples(sampler, s->f, t->f, p->f, c0->f, control, rgba); 1718 1719 for (j = 0; j < 4; j++) { 1720 r->f[j] = rgba[0][j]; 1721 g->f[j] = rgba[1][j]; 1722 b->f[j] = rgba[2][j]; 1723 a->f[j] = rgba[3][j]; 1724 } 1725} 1726 1727 1728#define TEX_MODIFIER_NONE 0 1729#define TEX_MODIFIER_PROJECTED 1 1730#define TEX_MODIFIER_LOD_BIAS 2 1731#define TEX_MODIFIER_EXPLICIT_LOD 3 1732 1733 1734static void 1735exec_tex(struct tgsi_exec_machine *mach, 1736 const struct tgsi_full_instruction *inst, 1737 uint modifier) 1738{ 1739 const uint unit = inst->Src[1].Register.Index; 1740 union tgsi_exec_channel r[4]; 1741 const union tgsi_exec_channel *lod = &ZeroVec; 1742 enum tgsi_sampler_control control; 1743 uint chan; 1744 1745 if (modifier != TEX_MODIFIER_NONE) { 1746 FETCH(&r[3], 0, CHAN_W); 1747 if (modifier != TEX_MODIFIER_PROJECTED) { 1748 lod = &r[3]; 1749 } 1750 } 1751 1752 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) { 1753 control = tgsi_sampler_lod_explicit; 1754 } else { 1755 control = tgsi_sampler_lod_bias; 1756 } 1757 1758 switch (inst->Texture.Texture) { 1759 case TGSI_TEXTURE_1D: 1760 case TGSI_TEXTURE_SHADOW1D: 1761 FETCH(&r[0], 0, CHAN_X); 1762 1763 if (modifier == TEX_MODIFIER_PROJECTED) { 1764 micro_div(&r[0], &r[0], &r[3]); 1765 } 1766 1767 fetch_texel(mach->Samplers[unit], 1768 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */ 1769 control, 1770 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1771 break; 1772 1773 case TGSI_TEXTURE_2D: 1774 case TGSI_TEXTURE_RECT: 1775 case TGSI_TEXTURE_SHADOW2D: 1776 case TGSI_TEXTURE_SHADOWRECT: 1777 FETCH(&r[0], 0, CHAN_X); 1778 FETCH(&r[1], 0, CHAN_Y); 1779 FETCH(&r[2], 0, CHAN_Z); 1780 1781 if (modifier == TEX_MODIFIER_PROJECTED) { 1782 micro_div(&r[0], &r[0], &r[3]); 1783 micro_div(&r[1], &r[1], &r[3]); 1784 micro_div(&r[2], &r[2], &r[3]); 1785 } 1786 1787 fetch_texel(mach->Samplers[unit], 1788 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1789 control, 1790 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1791 break; 1792 1793 case TGSI_TEXTURE_1D_ARRAY: 1794 FETCH(&r[0], 0, CHAN_X); 1795 FETCH(&r[1], 0, CHAN_Y); 1796 1797 if (modifier == TEX_MODIFIER_PROJECTED) { 1798 micro_div(&r[0], &r[0], &r[3]); 1799 } 1800 1801 fetch_texel(mach->Samplers[unit], 1802 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1803 control, 1804 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1805 break; 1806 1807 case TGSI_TEXTURE_2D_ARRAY: 1808 FETCH(&r[0], 0, CHAN_X); 1809 FETCH(&r[1], 0, CHAN_Y); 1810 FETCH(&r[2], 0, CHAN_Z); 1811 1812 if (modifier == TEX_MODIFIER_PROJECTED) { 1813 micro_div(&r[0], &r[0], &r[3]); 1814 micro_div(&r[1], &r[1], &r[3]); 1815 } 1816 1817 fetch_texel(mach->Samplers[unit], 1818 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1819 control, 1820 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1821 break; 1822 1823 case TGSI_TEXTURE_3D: 1824 case TGSI_TEXTURE_CUBE: 1825 FETCH(&r[0], 0, CHAN_X); 1826 FETCH(&r[1], 0, CHAN_Y); 1827 FETCH(&r[2], 0, CHAN_Z); 1828 1829 if (modifier == TEX_MODIFIER_PROJECTED) { 1830 micro_div(&r[0], &r[0], &r[3]); 1831 micro_div(&r[1], &r[1], &r[3]); 1832 micro_div(&r[2], &r[2], &r[3]); 1833 } 1834 1835 fetch_texel(mach->Samplers[unit], 1836 &r[0], &r[1], &r[2], lod, 1837 control, 1838 &r[0], &r[1], &r[2], &r[3]); 1839 break; 1840 1841 default: 1842 assert(0); 1843 } 1844 1845#if 0 1846 debug_printf("fetch r: %g %g %g %g\n", 1847 r[0].f[0], r[0].f[1], r[0].f[2], r[0].f[3]); 1848 debug_printf("fetch g: %g %g %g %g\n", 1849 r[1].f[0], r[1].f[1], r[1].f[2], r[1].f[3]); 1850 debug_printf("fetch b: %g %g %g %g\n", 1851 r[2].f[0], r[2].f[1], r[2].f[2], r[2].f[3]); 1852 debug_printf("fetch a: %g %g %g %g\n", 1853 r[3].f[0], r[3].f[1], r[3].f[2], r[3].f[3]); 1854#endif 1855 1856 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1857 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1858 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1859 } 1860 } 1861} 1862 1863static void 1864exec_txd(struct tgsi_exec_machine *mach, 1865 const struct tgsi_full_instruction *inst) 1866{ 1867 const uint unit = inst->Src[3].Register.Index; 1868 union tgsi_exec_channel r[4]; 1869 uint chan; 1870 1871 /* 1872 * XXX: This is fake TXD -- the derivatives are not taken into account, yet. 1873 */ 1874 1875 switch (inst->Texture.Texture) { 1876 case TGSI_TEXTURE_1D: 1877 case TGSI_TEXTURE_SHADOW1D: 1878 1879 FETCH(&r[0], 0, CHAN_X); 1880 1881 fetch_texel(mach->Samplers[unit], 1882 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */ 1883 tgsi_sampler_lod_bias, 1884 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1885 break; 1886 1887 case TGSI_TEXTURE_2D: 1888 case TGSI_TEXTURE_RECT: 1889 case TGSI_TEXTURE_SHADOW2D: 1890 case TGSI_TEXTURE_SHADOWRECT: 1891 1892 FETCH(&r[0], 0, CHAN_X); 1893 FETCH(&r[1], 0, CHAN_Y); 1894 FETCH(&r[2], 0, CHAN_Z); 1895 1896 fetch_texel(mach->Samplers[unit], 1897 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */ 1898 tgsi_sampler_lod_bias, 1899 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1900 break; 1901 1902 case TGSI_TEXTURE_3D: 1903 case TGSI_TEXTURE_CUBE: 1904 1905 FETCH(&r[0], 0, CHAN_X); 1906 FETCH(&r[1], 0, CHAN_Y); 1907 FETCH(&r[2], 0, CHAN_Z); 1908 1909 fetch_texel(mach->Samplers[unit], 1910 &r[0], &r[1], &r[2], &ZeroVec, 1911 tgsi_sampler_lod_bias, 1912 &r[0], &r[1], &r[2], &r[3]); 1913 break; 1914 1915 default: 1916 assert(0); 1917 } 1918 1919 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1920 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1921 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1922 } 1923 } 1924} 1925 1926 1927static void 1928exec_txf(struct tgsi_exec_machine *mach, 1929 const struct tgsi_full_instruction *inst) 1930{ 1931 struct tgsi_sampler *sampler; 1932 const uint unit = inst->Src[2].Register.Index; 1933 union tgsi_exec_channel r[4]; 1934 union tgsi_exec_channel offset[3]; 1935 uint chan; 1936 float rgba[NUM_CHANNELS][QUAD_SIZE]; 1937 int j; 1938 int8_t offsets[3]; 1939 1940 if (inst->Texture.NumOffsets == 1) { 1941 union tgsi_exec_channel index; 1942 index.i[0] = index.i[1] = index.i[2] = index.i[3] = inst->TexOffsets[0].Index; 1943 fetch_src_file_channel(mach, inst->TexOffsets[0].File, 1944 inst->TexOffsets[0].SwizzleX, &index, &ZeroVec, &offset[0]); 1945 fetch_src_file_channel(mach, inst->TexOffsets[0].File, 1946 inst->TexOffsets[0].SwizzleY, &index, &ZeroVec, &offset[1]); 1947 fetch_src_file_channel(mach, inst->TexOffsets[0].File, 1948 inst->TexOffsets[0].SwizzleZ, &index, &ZeroVec, &offset[2]); 1949 offsets[0] = offset[0].i[0]; 1950 offsets[1] = offset[1].i[0]; 1951 offsets[2] = offset[2].i[0]; 1952 } else 1953 offsets[0] = offsets[1] = offsets[2] = 0; 1954 1955 IFETCH(&r[3], 0, CHAN_W); 1956 1957 switch(inst->Texture.Texture) { 1958 case TGSI_TEXTURE_3D: 1959 case TGSI_TEXTURE_2D_ARRAY: 1960 IFETCH(&r[2], 0, CHAN_Z); 1961 /* fallthrough */ 1962 case TGSI_TEXTURE_2D: 1963 case TGSI_TEXTURE_RECT: 1964 case TGSI_TEXTURE_SHADOW2D: 1965 case TGSI_TEXTURE_SHADOWRECT: 1966 case TGSI_TEXTURE_1D_ARRAY: 1967 IFETCH(&r[1], 0, CHAN_Y); 1968 /* fallthrough */ 1969 case TGSI_TEXTURE_1D: 1970 case TGSI_TEXTURE_SHADOW1D: 1971 IFETCH(&r[0], 0, CHAN_X); 1972 break; 1973 default: 1974 assert(0); 1975 break; 1976 } 1977 1978 sampler = mach->Samplers[unit]; 1979 sampler->get_texel(sampler, r[0].i, r[1].i, r[2].i, r[3].i, 1980 offsets, rgba); 1981 1982 for (j = 0; j < QUAD_SIZE; j++) { 1983 r[0].f[j] = rgba[0][j]; 1984 r[1].f[j] = rgba[1][j]; 1985 r[2].f[j] = rgba[2][j]; 1986 r[3].f[j] = rgba[3][j]; 1987 } 1988 1989 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1990 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1991 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1992 } 1993 } 1994} 1995 1996static void 1997exec_txq(struct tgsi_exec_machine *mach, 1998 const struct tgsi_full_instruction *inst) 1999{ 2000 struct tgsi_sampler *sampler; 2001 const uint unit = inst->Src[1].Register.Index; 2002 int result[4]; 2003 union tgsi_exec_channel r[4], src; 2004 uint chan; 2005 int i,j; 2006 2007 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_INT); 2008 sampler = mach->Samplers[unit]; 2009 2010 sampler->get_dims(sampler, src.i[0], result); 2011 2012 for (i = 0; i < QUAD_SIZE; i++) { 2013 for (j = 0; j < 4; j++) { 2014 r[j].i[i] = result[j]; 2015 } 2016 } 2017 2018 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2019 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2020 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, 2021 TGSI_EXEC_DATA_INT); 2022 } 2023 } 2024} 2025 2026static void 2027exec_sample(struct tgsi_exec_machine *mach, 2028 const struct tgsi_full_instruction *inst, 2029 uint modifier) 2030{ 2031 const uint resource_unit = inst->Src[1].Register.Index; 2032 const uint sampler_unit = inst->Src[2].Register.Index; 2033 union tgsi_exec_channel r[4]; 2034 const union tgsi_exec_channel *lod = &ZeroVec; 2035 enum tgsi_sampler_control control; 2036 uint chan; 2037 2038 if (modifier != TEX_MODIFIER_NONE) { 2039 if (modifier == TEX_MODIFIER_LOD_BIAS) 2040 FETCH(&r[3], 3, CHAN_X); 2041 else /*TEX_MODIFIER_LOD*/ 2042 FETCH(&r[3], 0, CHAN_W); 2043 2044 if (modifier != TEX_MODIFIER_PROJECTED) { 2045 lod = &r[3]; 2046 } 2047 } 2048 2049 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) { 2050 control = tgsi_sampler_lod_explicit; 2051 } else { 2052 control = tgsi_sampler_lod_bias; 2053 } 2054 2055 switch (mach->Resources[resource_unit].Resource) { 2056 case TGSI_TEXTURE_1D: 2057 case TGSI_TEXTURE_SHADOW1D: 2058 FETCH(&r[0], 0, CHAN_X); 2059 2060 if (modifier == TEX_MODIFIER_PROJECTED) { 2061 micro_div(&r[0], &r[0], &r[3]); 2062 } 2063 2064 fetch_texel(mach->Samplers[sampler_unit], 2065 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */ 2066 control, 2067 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 2068 break; 2069 2070 case TGSI_TEXTURE_2D: 2071 case TGSI_TEXTURE_RECT: 2072 case TGSI_TEXTURE_SHADOW2D: 2073 case TGSI_TEXTURE_SHADOWRECT: 2074 FETCH(&r[0], 0, CHAN_X); 2075 FETCH(&r[1], 0, CHAN_Y); 2076 FETCH(&r[2], 0, CHAN_Z); 2077 2078 if (modifier == TEX_MODIFIER_PROJECTED) { 2079 micro_div(&r[0], &r[0], &r[3]); 2080 micro_div(&r[1], &r[1], &r[3]); 2081 micro_div(&r[2], &r[2], &r[3]); 2082 } 2083 2084 fetch_texel(mach->Samplers[sampler_unit], 2085 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 2086 control, 2087 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 2088 break; 2089 2090 case TGSI_TEXTURE_3D: 2091 case TGSI_TEXTURE_CUBE: 2092 FETCH(&r[0], 0, CHAN_X); 2093 FETCH(&r[1], 0, CHAN_Y); 2094 FETCH(&r[2], 0, CHAN_Z); 2095 2096 if (modifier == TEX_MODIFIER_PROJECTED) { 2097 micro_div(&r[0], &r[0], &r[3]); 2098 micro_div(&r[1], &r[1], &r[3]); 2099 micro_div(&r[2], &r[2], &r[3]); 2100 } 2101 2102 fetch_texel(mach->Samplers[sampler_unit], 2103 &r[0], &r[1], &r[2], lod, 2104 control, 2105 &r[0], &r[1], &r[2], &r[3]); 2106 break; 2107 2108 default: 2109 assert(0); 2110 } 2111 2112 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2113 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2114 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2115 } 2116 } 2117} 2118 2119static void 2120exec_sample_d(struct tgsi_exec_machine *mach, 2121 const struct tgsi_full_instruction *inst) 2122{ 2123 const uint resource_unit = inst->Src[1].Register.Index; 2124 const uint sampler_unit = inst->Src[2].Register.Index; 2125 union tgsi_exec_channel r[4]; 2126 uint chan; 2127 /* 2128 * XXX: This is fake SAMPLE_D -- the derivatives are not taken into account, yet. 2129 */ 2130 2131 switch (mach->Resources[resource_unit].Resource) { 2132 case TGSI_TEXTURE_1D: 2133 case TGSI_TEXTURE_SHADOW1D: 2134 2135 FETCH(&r[0], 0, CHAN_X); 2136 2137 fetch_texel(mach->Samplers[sampler_unit], 2138 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */ 2139 tgsi_sampler_lod_bias, 2140 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 2141 break; 2142 2143 case TGSI_TEXTURE_2D: 2144 case TGSI_TEXTURE_RECT: 2145 case TGSI_TEXTURE_SHADOW2D: 2146 case TGSI_TEXTURE_SHADOWRECT: 2147 2148 FETCH(&r[0], 0, CHAN_X); 2149 FETCH(&r[1], 0, CHAN_Y); 2150 FETCH(&r[2], 0, CHAN_Z); 2151 2152 fetch_texel(mach->Samplers[sampler_unit], 2153 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */ 2154 tgsi_sampler_lod_bias, 2155 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 2156 break; 2157 2158 case TGSI_TEXTURE_3D: 2159 case TGSI_TEXTURE_CUBE: 2160 2161 FETCH(&r[0], 0, CHAN_X); 2162 FETCH(&r[1], 0, CHAN_Y); 2163 FETCH(&r[2], 0, CHAN_Z); 2164 2165 fetch_texel(mach->Samplers[sampler_unit], 2166 &r[0], &r[1], &r[2], &ZeroVec, 2167 tgsi_sampler_lod_bias, 2168 &r[0], &r[1], &r[2], &r[3]); 2169 break; 2170 2171 default: 2172 assert(0); 2173 } 2174 2175 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2176 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2177 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2178 } 2179 } 2180} 2181 2182 2183/** 2184 * Evaluate a constant-valued coefficient at the position of the 2185 * current quad. 2186 */ 2187static void 2188eval_constant_coef( 2189 struct tgsi_exec_machine *mach, 2190 unsigned attrib, 2191 unsigned chan ) 2192{ 2193 unsigned i; 2194 2195 for( i = 0; i < QUAD_SIZE; i++ ) { 2196 mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan]; 2197 } 2198} 2199 2200/** 2201 * Evaluate a linear-valued coefficient at the position of the 2202 * current quad. 2203 */ 2204static void 2205eval_linear_coef( 2206 struct tgsi_exec_machine *mach, 2207 unsigned attrib, 2208 unsigned chan ) 2209{ 2210 const float x = mach->QuadPos.xyzw[0].f[0]; 2211 const float y = mach->QuadPos.xyzw[1].f[0]; 2212 const float dadx = mach->InterpCoefs[attrib].dadx[chan]; 2213 const float dady = mach->InterpCoefs[attrib].dady[chan]; 2214 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; 2215 mach->Inputs[attrib].xyzw[chan].f[0] = a0; 2216 mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx; 2217 mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady; 2218 mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady; 2219} 2220 2221/** 2222 * Evaluate a perspective-valued coefficient at the position of the 2223 * current quad. 2224 */ 2225static void 2226eval_perspective_coef( 2227 struct tgsi_exec_machine *mach, 2228 unsigned attrib, 2229 unsigned chan ) 2230{ 2231 const float x = mach->QuadPos.xyzw[0].f[0]; 2232 const float y = mach->QuadPos.xyzw[1].f[0]; 2233 const float dadx = mach->InterpCoefs[attrib].dadx[chan]; 2234 const float dady = mach->InterpCoefs[attrib].dady[chan]; 2235 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; 2236 const float *w = mach->QuadPos.xyzw[3].f; 2237 /* divide by W here */ 2238 mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0]; 2239 mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1]; 2240 mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2]; 2241 mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3]; 2242} 2243 2244 2245typedef void (* eval_coef_func)( 2246 struct tgsi_exec_machine *mach, 2247 unsigned attrib, 2248 unsigned chan ); 2249 2250static void 2251exec_declaration(struct tgsi_exec_machine *mach, 2252 const struct tgsi_full_declaration *decl) 2253{ 2254 if (decl->Declaration.File == TGSI_FILE_RESOURCE) { 2255 mach->Resources[decl->Range.First] = decl->Resource; 2256 return; 2257 } 2258 2259 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { 2260 if (decl->Declaration.File == TGSI_FILE_INPUT) { 2261 uint first, last, mask; 2262 2263 first = decl->Range.First; 2264 last = decl->Range.Last; 2265 mask = decl->Declaration.UsageMask; 2266 2267 /* XXX we could remove this special-case code since 2268 * mach->InterpCoefs[first].a0 should already have the 2269 * front/back-face value. But we should first update the 2270 * ureg code to emit the right UsageMask value (WRITEMASK_X). 2271 * Then, we could remove the tgsi_exec_machine::Face field. 2272 */ 2273 /* XXX make FACE a system value */ 2274 if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) { 2275 uint i; 2276 2277 assert(decl->Semantic.Index == 0); 2278 assert(first == last); 2279 2280 for (i = 0; i < QUAD_SIZE; i++) { 2281 mach->Inputs[first].xyzw[0].f[i] = mach->Face; 2282 } 2283 } else { 2284 eval_coef_func eval; 2285 uint i, j; 2286 2287 switch (decl->Declaration.Interpolate) { 2288 case TGSI_INTERPOLATE_CONSTANT: 2289 eval = eval_constant_coef; 2290 break; 2291 2292 case TGSI_INTERPOLATE_LINEAR: 2293 eval = eval_linear_coef; 2294 break; 2295 2296 case TGSI_INTERPOLATE_PERSPECTIVE: 2297 eval = eval_perspective_coef; 2298 break; 2299 2300 default: 2301 assert(0); 2302 return; 2303 } 2304 2305 for (j = 0; j < NUM_CHANNELS; j++) { 2306 if (mask & (1 << j)) { 2307 for (i = first; i <= last; i++) { 2308 eval(mach, i, j); 2309 } 2310 } 2311 } 2312 } 2313 } 2314 } 2315 2316 if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) { 2317 mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First; 2318 } 2319} 2320 2321 2322typedef void (* micro_op)(union tgsi_exec_channel *dst); 2323 2324static void 2325exec_vector(struct tgsi_exec_machine *mach, 2326 const struct tgsi_full_instruction *inst, 2327 micro_op op, 2328 enum tgsi_exec_datatype dst_datatype) 2329{ 2330 unsigned int chan; 2331 2332 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2333 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2334 union tgsi_exec_channel dst; 2335 2336 op(&dst); 2337 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2338 } 2339 } 2340} 2341 2342typedef void (* micro_unary_op)(union tgsi_exec_channel *dst, 2343 const union tgsi_exec_channel *src); 2344 2345static void 2346exec_scalar_unary(struct tgsi_exec_machine *mach, 2347 const struct tgsi_full_instruction *inst, 2348 micro_unary_op op, 2349 enum tgsi_exec_datatype dst_datatype, 2350 enum tgsi_exec_datatype src_datatype) 2351{ 2352 unsigned int chan; 2353 union tgsi_exec_channel src; 2354 union tgsi_exec_channel dst; 2355 2356 fetch_source(mach, &src, &inst->Src[0], CHAN_X, src_datatype); 2357 op(&dst, &src); 2358 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2359 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2360 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2361 } 2362 } 2363} 2364 2365static void 2366exec_vector_unary(struct tgsi_exec_machine *mach, 2367 const struct tgsi_full_instruction *inst, 2368 micro_unary_op op, 2369 enum tgsi_exec_datatype dst_datatype, 2370 enum tgsi_exec_datatype src_datatype) 2371{ 2372 unsigned int chan; 2373 struct tgsi_exec_vector dst; 2374 2375 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2376 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2377 union tgsi_exec_channel src; 2378 2379 fetch_source(mach, &src, &inst->Src[0], chan, src_datatype); 2380 op(&dst.xyzw[chan], &src); 2381 } 2382 } 2383 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2384 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2385 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2386 } 2387 } 2388} 2389 2390typedef void (* micro_binary_op)(union tgsi_exec_channel *dst, 2391 const union tgsi_exec_channel *src0, 2392 const union tgsi_exec_channel *src1); 2393 2394static void 2395exec_scalar_binary(struct tgsi_exec_machine *mach, 2396 const struct tgsi_full_instruction *inst, 2397 micro_binary_op op, 2398 enum tgsi_exec_datatype dst_datatype, 2399 enum tgsi_exec_datatype src_datatype) 2400{ 2401 unsigned int chan; 2402 union tgsi_exec_channel src[2]; 2403 union tgsi_exec_channel dst; 2404 2405 fetch_source(mach, &src[0], &inst->Src[0], CHAN_X, src_datatype); 2406 fetch_source(mach, &src[1], &inst->Src[1], CHAN_Y, src_datatype); 2407 op(&dst, &src[0], &src[1]); 2408 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2409 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2410 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2411 } 2412 } 2413} 2414 2415static void 2416exec_vector_binary(struct tgsi_exec_machine *mach, 2417 const struct tgsi_full_instruction *inst, 2418 micro_binary_op op, 2419 enum tgsi_exec_datatype dst_datatype, 2420 enum tgsi_exec_datatype src_datatype) 2421{ 2422 unsigned int chan; 2423 struct tgsi_exec_vector dst; 2424 2425 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2426 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2427 union tgsi_exec_channel src[2]; 2428 2429 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype); 2430 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype); 2431 op(&dst.xyzw[chan], &src[0], &src[1]); 2432 } 2433 } 2434 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2435 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2436 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2437 } 2438 } 2439} 2440 2441typedef void (* micro_trinary_op)(union tgsi_exec_channel *dst, 2442 const union tgsi_exec_channel *src0, 2443 const union tgsi_exec_channel *src1, 2444 const union tgsi_exec_channel *src2); 2445 2446static void 2447exec_vector_trinary(struct tgsi_exec_machine *mach, 2448 const struct tgsi_full_instruction *inst, 2449 micro_trinary_op op, 2450 enum tgsi_exec_datatype dst_datatype, 2451 enum tgsi_exec_datatype src_datatype) 2452{ 2453 unsigned int chan; 2454 struct tgsi_exec_vector dst; 2455 2456 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2457 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2458 union tgsi_exec_channel src[3]; 2459 2460 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype); 2461 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype); 2462 fetch_source(mach, &src[2], &inst->Src[2], chan, src_datatype); 2463 op(&dst.xyzw[chan], &src[0], &src[1], &src[2]); 2464 } 2465 } 2466 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2467 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2468 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2469 } 2470 } 2471} 2472 2473static void 2474exec_dp3(struct tgsi_exec_machine *mach, 2475 const struct tgsi_full_instruction *inst) 2476{ 2477 unsigned int chan; 2478 union tgsi_exec_channel arg[3]; 2479 2480 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2481 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2482 micro_mul(&arg[2], &arg[0], &arg[1]); 2483 2484 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) { 2485 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2486 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT); 2487 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2488 } 2489 2490 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2491 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2492 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2493 } 2494 } 2495} 2496 2497static void 2498exec_dp4(struct tgsi_exec_machine *mach, 2499 const struct tgsi_full_instruction *inst) 2500{ 2501 unsigned int chan; 2502 union tgsi_exec_channel arg[3]; 2503 2504 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2505 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2506 micro_mul(&arg[2], &arg[0], &arg[1]); 2507 2508 for (chan = CHAN_Y; chan <= CHAN_W; chan++) { 2509 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2510 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT); 2511 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2512 } 2513 2514 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2515 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2516 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2517 } 2518 } 2519} 2520 2521static void 2522exec_dp2a(struct tgsi_exec_machine *mach, 2523 const struct tgsi_full_instruction *inst) 2524{ 2525 unsigned int chan; 2526 union tgsi_exec_channel arg[3]; 2527 2528 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2529 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2530 micro_mul(&arg[2], &arg[0], &arg[1]); 2531 2532 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2533 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2534 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]); 2535 2536 fetch_source(mach, &arg[1], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2537 micro_add(&arg[0], &arg[0], &arg[1]); 2538 2539 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2540 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2541 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2542 } 2543 } 2544} 2545 2546static void 2547exec_dph(struct tgsi_exec_machine *mach, 2548 const struct tgsi_full_instruction *inst) 2549{ 2550 unsigned int chan; 2551 union tgsi_exec_channel arg[3]; 2552 2553 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2554 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2555 micro_mul(&arg[2], &arg[0], &arg[1]); 2556 2557 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2558 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2559 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2560 2561 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2562 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2563 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]); 2564 2565 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2566 micro_add(&arg[0], &arg[0], &arg[1]); 2567 2568 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2569 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2570 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2571 } 2572 } 2573} 2574 2575static void 2576exec_dp2(struct tgsi_exec_machine *mach, 2577 const struct tgsi_full_instruction *inst) 2578{ 2579 unsigned int chan; 2580 union tgsi_exec_channel arg[3]; 2581 2582 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2583 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2584 micro_mul(&arg[2], &arg[0], &arg[1]); 2585 2586 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2587 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2588 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2589 2590 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2591 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2592 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2593 } 2594 } 2595} 2596 2597static void 2598exec_nrm4(struct tgsi_exec_machine *mach, 2599 const struct tgsi_full_instruction *inst) 2600{ 2601 unsigned int chan; 2602 union tgsi_exec_channel arg[4]; 2603 union tgsi_exec_channel scale; 2604 2605 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2606 micro_mul(&scale, &arg[0], &arg[0]); 2607 2608 for (chan = CHAN_Y; chan <= CHAN_W; chan++) { 2609 union tgsi_exec_channel product; 2610 2611 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2612 micro_mul(&product, &arg[chan], &arg[chan]); 2613 micro_add(&scale, &scale, &product); 2614 } 2615 2616 micro_rsq(&scale, &scale); 2617 2618 for (chan = CHAN_X; chan <= CHAN_W; chan++) { 2619 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2620 micro_mul(&arg[chan], &arg[chan], &scale); 2621 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2622 } 2623 } 2624} 2625 2626static void 2627exec_nrm3(struct tgsi_exec_machine *mach, 2628 const struct tgsi_full_instruction *inst) 2629{ 2630 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) { 2631 unsigned int chan; 2632 union tgsi_exec_channel arg[3]; 2633 union tgsi_exec_channel scale; 2634 2635 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2636 micro_mul(&scale, &arg[0], &arg[0]); 2637 2638 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) { 2639 union tgsi_exec_channel product; 2640 2641 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2642 micro_mul(&product, &arg[chan], &arg[chan]); 2643 micro_add(&scale, &scale, &product); 2644 } 2645 2646 micro_rsq(&scale, &scale); 2647 2648 for (chan = CHAN_X; chan <= CHAN_Z; chan++) { 2649 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2650 micro_mul(&arg[chan], &arg[chan], &scale); 2651 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2652 } 2653 } 2654 } 2655 2656 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2657 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2658 } 2659} 2660 2661static void 2662exec_scs(struct tgsi_exec_machine *mach, 2663 const struct tgsi_full_instruction *inst) 2664{ 2665 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XY) { 2666 union tgsi_exec_channel arg; 2667 union tgsi_exec_channel result; 2668 2669 fetch_source(mach, &arg, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2670 2671 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2672 micro_cos(&result, &arg); 2673 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2674 } 2675 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2676 micro_sin(&result, &arg); 2677 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2678 } 2679 } 2680 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2681 store_dest(mach, &ZeroVec, &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2682 } 2683 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2684 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2685 } 2686} 2687 2688static void 2689exec_x2d(struct tgsi_exec_machine *mach, 2690 const struct tgsi_full_instruction *inst) 2691{ 2692 union tgsi_exec_channel r[4]; 2693 union tgsi_exec_channel d[2]; 2694 2695 fetch_source(mach, &r[0], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2696 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2697 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XZ) { 2698 fetch_source(mach, &r[2], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2699 micro_mul(&r[2], &r[2], &r[0]); 2700 fetch_source(mach, &r[3], &inst->Src[2], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2701 micro_mul(&r[3], &r[3], &r[1]); 2702 micro_add(&r[2], &r[2], &r[3]); 2703 fetch_source(mach, &r[3], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2704 micro_add(&d[0], &r[2], &r[3]); 2705 } 2706 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YW) { 2707 fetch_source(mach, &r[2], &inst->Src[2], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2708 micro_mul(&r[2], &r[2], &r[0]); 2709 fetch_source(mach, &r[3], &inst->Src[2], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2710 micro_mul(&r[3], &r[3], &r[1]); 2711 micro_add(&r[2], &r[2], &r[3]); 2712 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2713 micro_add(&d[1], &r[2], &r[3]); 2714 } 2715 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2716 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2717 } 2718 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2719 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2720 } 2721 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2722 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2723 } 2724 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2725 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2726 } 2727} 2728 2729static void 2730exec_rfl(struct tgsi_exec_machine *mach, 2731 const struct tgsi_full_instruction *inst) 2732{ 2733 union tgsi_exec_channel r[9]; 2734 2735 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) { 2736 /* r0 = dp3(src0, src0) */ 2737 fetch_source(mach, &r[2], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2738 micro_mul(&r[0], &r[2], &r[2]); 2739 fetch_source(mach, &r[4], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2740 micro_mul(&r[8], &r[4], &r[4]); 2741 micro_add(&r[0], &r[0], &r[8]); 2742 fetch_source(mach, &r[6], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2743 micro_mul(&r[8], &r[6], &r[6]); 2744 micro_add(&r[0], &r[0], &r[8]); 2745 2746 /* r1 = dp3(src0, src1) */ 2747 fetch_source(mach, &r[3], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2748 micro_mul(&r[1], &r[2], &r[3]); 2749 fetch_source(mach, &r[5], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2750 micro_mul(&r[8], &r[4], &r[5]); 2751 micro_add(&r[1], &r[1], &r[8]); 2752 fetch_source(mach, &r[7], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2753 micro_mul(&r[8], &r[6], &r[7]); 2754 micro_add(&r[1], &r[1], &r[8]); 2755 2756 /* r1 = 2 * r1 / r0 */ 2757 micro_add(&r[1], &r[1], &r[1]); 2758 micro_div(&r[1], &r[1], &r[0]); 2759 2760 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2761 micro_mul(&r[2], &r[2], &r[1]); 2762 micro_sub(&r[2], &r[2], &r[3]); 2763 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2764 } 2765 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2766 micro_mul(&r[4], &r[4], &r[1]); 2767 micro_sub(&r[4], &r[4], &r[5]); 2768 store_dest(mach, &r[4], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2769 } 2770 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2771 micro_mul(&r[6], &r[6], &r[1]); 2772 micro_sub(&r[6], &r[6], &r[7]); 2773 store_dest(mach, &r[6], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2774 } 2775 } 2776 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2777 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2778 } 2779} 2780 2781static void 2782exec_xpd(struct tgsi_exec_machine *mach, 2783 const struct tgsi_full_instruction *inst) 2784{ 2785 union tgsi_exec_channel r[6]; 2786 union tgsi_exec_channel d[3]; 2787 2788 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2789 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2790 2791 micro_mul(&r[2], &r[0], &r[1]); 2792 2793 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2794 fetch_source(mach, &r[4], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2795 2796 micro_mul(&r[5], &r[3], &r[4] ); 2797 micro_sub(&d[CHAN_X], &r[2], &r[5]); 2798 2799 fetch_source(mach, &r[2], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2800 2801 micro_mul(&r[3], &r[3], &r[2]); 2802 2803 fetch_source(mach, &r[5], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2804 2805 micro_mul(&r[1], &r[1], &r[5]); 2806 micro_sub(&d[CHAN_Y], &r[3], &r[1]); 2807 2808 micro_mul(&r[5], &r[5], &r[4]); 2809 micro_mul(&r[0], &r[0], &r[2]); 2810 micro_sub(&d[CHAN_Z], &r[5], &r[0]); 2811 2812 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2813 store_dest(mach, &d[CHAN_X], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2814 } 2815 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2816 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2817 } 2818 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2819 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2820 } 2821 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2822 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2823 } 2824} 2825 2826static void 2827exec_dst(struct tgsi_exec_machine *mach, 2828 const struct tgsi_full_instruction *inst) 2829{ 2830 union tgsi_exec_channel r[2]; 2831 union tgsi_exec_channel d[4]; 2832 2833 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2834 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2835 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2836 micro_mul(&d[CHAN_Y], &r[0], &r[1]); 2837 } 2838 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2839 fetch_source(mach, &d[CHAN_Z], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2840 } 2841 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2842 fetch_source(mach, &d[CHAN_W], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2843 } 2844 2845 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2846 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2847 } 2848 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2849 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2850 } 2851 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2852 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2853 } 2854 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2855 store_dest(mach, &d[CHAN_W], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2856 } 2857} 2858 2859static void 2860exec_log(struct tgsi_exec_machine *mach, 2861 const struct tgsi_full_instruction *inst) 2862{ 2863 union tgsi_exec_channel r[3]; 2864 2865 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2866 micro_abs(&r[2], &r[0]); /* r2 = abs(r0) */ 2867 micro_lg2(&r[1], &r[2]); /* r1 = lg2(r2) */ 2868 micro_flr(&r[0], &r[1]); /* r0 = floor(r1) */ 2869 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2870 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2871 } 2872 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2873 micro_exp2(&r[0], &r[0]); /* r0 = 2 ^ r0 */ 2874 micro_div(&r[0], &r[2], &r[0]); /* r0 = r2 / r0 */ 2875 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2876 } 2877 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2878 store_dest(mach, &r[1], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2879 } 2880 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2881 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2882 } 2883} 2884 2885static void 2886exec_exp(struct tgsi_exec_machine *mach, 2887 const struct tgsi_full_instruction *inst) 2888{ 2889 union tgsi_exec_channel r[3]; 2890 2891 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2892 micro_flr(&r[1], &r[0]); /* r1 = floor(r0) */ 2893 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2894 micro_exp2(&r[2], &r[1]); /* r2 = 2 ^ r1 */ 2895 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2896 } 2897 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2898 micro_sub(&r[2], &r[0], &r[1]); /* r2 = r0 - r1 */ 2899 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2900 } 2901 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2902 micro_exp2(&r[2], &r[0]); /* r2 = 2 ^ r0 */ 2903 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2904 } 2905 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2906 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2907 } 2908} 2909 2910static void 2911exec_lit(struct tgsi_exec_machine *mach, 2912 const struct tgsi_full_instruction *inst) 2913{ 2914 union tgsi_exec_channel r[3]; 2915 union tgsi_exec_channel d[3]; 2916 2917 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2918 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2919 } 2920 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YZ) { 2921 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2922 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2923 micro_max(&d[CHAN_Y], &r[0], &ZeroVec); 2924 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2925 } 2926 2927 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2928 fetch_source(mach, &r[1], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2929 micro_max(&r[1], &r[1], &ZeroVec); 2930 2931 fetch_source(mach, &r[2], &inst->Src[0], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2932 micro_min(&r[2], &r[2], &P128Vec); 2933 micro_max(&r[2], &r[2], &M128Vec); 2934 micro_pow(&r[1], &r[1], &r[2]); 2935 micro_lt(&d[CHAN_Z], &ZeroVec, &r[0], &r[1], &ZeroVec); 2936 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2937 } 2938 } 2939 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2940 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2941 } 2942} 2943 2944static void 2945exec_break(struct tgsi_exec_machine *mach) 2946{ 2947 if (mach->BreakType == TGSI_EXEC_BREAK_INSIDE_LOOP) { 2948 /* turn off loop channels for each enabled exec channel */ 2949 mach->LoopMask &= ~mach->ExecMask; 2950 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 2951 UPDATE_EXEC_MASK(mach); 2952 } else { 2953 assert(mach->BreakType == TGSI_EXEC_BREAK_INSIDE_SWITCH); 2954 2955 mach->Switch.mask = 0x0; 2956 2957 UPDATE_EXEC_MASK(mach); 2958 } 2959} 2960 2961static void 2962exec_switch(struct tgsi_exec_machine *mach, 2963 const struct tgsi_full_instruction *inst) 2964{ 2965 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING); 2966 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 2967 2968 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch; 2969 fetch_source(mach, &mach->Switch.selector, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT); 2970 mach->Switch.mask = 0x0; 2971 mach->Switch.defaultMask = 0x0; 2972 2973 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 2974 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_SWITCH; 2975 2976 UPDATE_EXEC_MASK(mach); 2977} 2978 2979static void 2980exec_case(struct tgsi_exec_machine *mach, 2981 const struct tgsi_full_instruction *inst) 2982{ 2983 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask; 2984 union tgsi_exec_channel src; 2985 uint mask = 0; 2986 2987 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT); 2988 2989 if (mach->Switch.selector.u[0] == src.u[0]) { 2990 mask |= 0x1; 2991 } 2992 if (mach->Switch.selector.u[1] == src.u[1]) { 2993 mask |= 0x2; 2994 } 2995 if (mach->Switch.selector.u[2] == src.u[2]) { 2996 mask |= 0x4; 2997 } 2998 if (mach->Switch.selector.u[3] == src.u[3]) { 2999 mask |= 0x8; 3000 } 3001 3002 mach->Switch.defaultMask |= mask; 3003 3004 mach->Switch.mask |= mask & prevMask; 3005 3006 UPDATE_EXEC_MASK(mach); 3007} 3008 3009static void 3010exec_default(struct tgsi_exec_machine *mach) 3011{ 3012 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask; 3013 3014 mach->Switch.mask |= ~mach->Switch.defaultMask & prevMask; 3015 3016 UPDATE_EXEC_MASK(mach); 3017} 3018 3019static void 3020exec_endswitch(struct tgsi_exec_machine *mach) 3021{ 3022 mach->Switch = mach->SwitchStack[--mach->SwitchStackTop]; 3023 mach->BreakType = mach->BreakStack[--mach->BreakStackTop]; 3024 3025 UPDATE_EXEC_MASK(mach); 3026} 3027 3028static void 3029micro_i2f(union tgsi_exec_channel *dst, 3030 const union tgsi_exec_channel *src) 3031{ 3032 dst->f[0] = (float)src->i[0]; 3033 dst->f[1] = (float)src->i[1]; 3034 dst->f[2] = (float)src->i[2]; 3035 dst->f[3] = (float)src->i[3]; 3036} 3037 3038static void 3039micro_not(union tgsi_exec_channel *dst, 3040 const union tgsi_exec_channel *src) 3041{ 3042 dst->u[0] = ~src->u[0]; 3043 dst->u[1] = ~src->u[1]; 3044 dst->u[2] = ~src->u[2]; 3045 dst->u[3] = ~src->u[3]; 3046} 3047 3048static void 3049micro_shl(union tgsi_exec_channel *dst, 3050 const union tgsi_exec_channel *src0, 3051 const union tgsi_exec_channel *src1) 3052{ 3053 dst->u[0] = src0->u[0] << src1->u[0]; 3054 dst->u[1] = src0->u[1] << src1->u[1]; 3055 dst->u[2] = src0->u[2] << src1->u[2]; 3056 dst->u[3] = src0->u[3] << src1->u[3]; 3057} 3058 3059static void 3060micro_and(union tgsi_exec_channel *dst, 3061 const union tgsi_exec_channel *src0, 3062 const union tgsi_exec_channel *src1) 3063{ 3064 dst->u[0] = src0->u[0] & src1->u[0]; 3065 dst->u[1] = src0->u[1] & src1->u[1]; 3066 dst->u[2] = src0->u[2] & src1->u[2]; 3067 dst->u[3] = src0->u[3] & src1->u[3]; 3068} 3069 3070static void 3071micro_or(union tgsi_exec_channel *dst, 3072 const union tgsi_exec_channel *src0, 3073 const union tgsi_exec_channel *src1) 3074{ 3075 dst->u[0] = src0->u[0] | src1->u[0]; 3076 dst->u[1] = src0->u[1] | src1->u[1]; 3077 dst->u[2] = src0->u[2] | src1->u[2]; 3078 dst->u[3] = src0->u[3] | src1->u[3]; 3079} 3080 3081static void 3082micro_xor(union tgsi_exec_channel *dst, 3083 const union tgsi_exec_channel *src0, 3084 const union tgsi_exec_channel *src1) 3085{ 3086 dst->u[0] = src0->u[0] ^ src1->u[0]; 3087 dst->u[1] = src0->u[1] ^ src1->u[1]; 3088 dst->u[2] = src0->u[2] ^ src1->u[2]; 3089 dst->u[3] = src0->u[3] ^ src1->u[3]; 3090} 3091 3092static void 3093micro_mod(union tgsi_exec_channel *dst, 3094 const union tgsi_exec_channel *src0, 3095 const union tgsi_exec_channel *src1) 3096{ 3097 dst->i[0] = src0->i[0] % src1->i[0]; 3098 dst->i[1] = src0->i[1] % src1->i[1]; 3099 dst->i[2] = src0->i[2] % src1->i[2]; 3100 dst->i[3] = src0->i[3] % src1->i[3]; 3101} 3102 3103static void 3104micro_f2i(union tgsi_exec_channel *dst, 3105 const union tgsi_exec_channel *src) 3106{ 3107 dst->i[0] = (int)src->f[0]; 3108 dst->i[1] = (int)src->f[1]; 3109 dst->i[2] = (int)src->f[2]; 3110 dst->i[3] = (int)src->f[3]; 3111} 3112 3113static void 3114micro_idiv(union tgsi_exec_channel *dst, 3115 const union tgsi_exec_channel *src0, 3116 const union tgsi_exec_channel *src1) 3117{ 3118 dst->i[0] = src0->i[0] / src1->i[0]; 3119 dst->i[1] = src0->i[1] / src1->i[1]; 3120 dst->i[2] = src0->i[2] / src1->i[2]; 3121 dst->i[3] = src0->i[3] / src1->i[3]; 3122} 3123 3124static void 3125micro_imax(union tgsi_exec_channel *dst, 3126 const union tgsi_exec_channel *src0, 3127 const union tgsi_exec_channel *src1) 3128{ 3129 dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0]; 3130 dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1]; 3131 dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2]; 3132 dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3]; 3133} 3134 3135static void 3136micro_imin(union tgsi_exec_channel *dst, 3137 const union tgsi_exec_channel *src0, 3138 const union tgsi_exec_channel *src1) 3139{ 3140 dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0]; 3141 dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1]; 3142 dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2]; 3143 dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3]; 3144} 3145 3146static void 3147micro_isge(union tgsi_exec_channel *dst, 3148 const union tgsi_exec_channel *src0, 3149 const union tgsi_exec_channel *src1) 3150{ 3151 dst->i[0] = src0->i[0] >= src1->i[0] ? -1 : 0; 3152 dst->i[1] = src0->i[1] >= src1->i[1] ? -1 : 0; 3153 dst->i[2] = src0->i[2] >= src1->i[2] ? -1 : 0; 3154 dst->i[3] = src0->i[3] >= src1->i[3] ? -1 : 0; 3155} 3156 3157static void 3158micro_ishr(union tgsi_exec_channel *dst, 3159 const union tgsi_exec_channel *src0, 3160 const union tgsi_exec_channel *src1) 3161{ 3162 dst->i[0] = src0->i[0] >> src1->i[0]; 3163 dst->i[1] = src0->i[1] >> src1->i[1]; 3164 dst->i[2] = src0->i[2] >> src1->i[2]; 3165 dst->i[3] = src0->i[3] >> src1->i[3]; 3166} 3167 3168static void 3169micro_islt(union tgsi_exec_channel *dst, 3170 const union tgsi_exec_channel *src0, 3171 const union tgsi_exec_channel *src1) 3172{ 3173 dst->i[0] = src0->i[0] < src1->i[0] ? -1 : 0; 3174 dst->i[1] = src0->i[1] < src1->i[1] ? -1 : 0; 3175 dst->i[2] = src0->i[2] < src1->i[2] ? -1 : 0; 3176 dst->i[3] = src0->i[3] < src1->i[3] ? -1 : 0; 3177} 3178 3179static void 3180micro_f2u(union tgsi_exec_channel *dst, 3181 const union tgsi_exec_channel *src) 3182{ 3183 dst->u[0] = (uint)src->f[0]; 3184 dst->u[1] = (uint)src->f[1]; 3185 dst->u[2] = (uint)src->f[2]; 3186 dst->u[3] = (uint)src->f[3]; 3187} 3188 3189static void 3190micro_u2f(union tgsi_exec_channel *dst, 3191 const union tgsi_exec_channel *src) 3192{ 3193 dst->f[0] = (float)src->u[0]; 3194 dst->f[1] = (float)src->u[1]; 3195 dst->f[2] = (float)src->u[2]; 3196 dst->f[3] = (float)src->u[3]; 3197} 3198 3199static void 3200micro_uadd(union tgsi_exec_channel *dst, 3201 const union tgsi_exec_channel *src0, 3202 const union tgsi_exec_channel *src1) 3203{ 3204 dst->u[0] = src0->u[0] + src1->u[0]; 3205 dst->u[1] = src0->u[1] + src1->u[1]; 3206 dst->u[2] = src0->u[2] + src1->u[2]; 3207 dst->u[3] = src0->u[3] + src1->u[3]; 3208} 3209 3210static void 3211micro_udiv(union tgsi_exec_channel *dst, 3212 const union tgsi_exec_channel *src0, 3213 const union tgsi_exec_channel *src1) 3214{ 3215 dst->u[0] = src0->u[0] / src1->u[0]; 3216 dst->u[1] = src0->u[1] / src1->u[1]; 3217 dst->u[2] = src0->u[2] / src1->u[2]; 3218 dst->u[3] = src0->u[3] / src1->u[3]; 3219} 3220 3221static void 3222micro_umad(union tgsi_exec_channel *dst, 3223 const union tgsi_exec_channel *src0, 3224 const union tgsi_exec_channel *src1, 3225 const union tgsi_exec_channel *src2) 3226{ 3227 dst->u[0] = src0->u[0] * src1->u[0] + src2->u[0]; 3228 dst->u[1] = src0->u[1] * src1->u[1] + src2->u[1]; 3229 dst->u[2] = src0->u[2] * src1->u[2] + src2->u[2]; 3230 dst->u[3] = src0->u[3] * src1->u[3] + src2->u[3]; 3231} 3232 3233static void 3234micro_umax(union tgsi_exec_channel *dst, 3235 const union tgsi_exec_channel *src0, 3236 const union tgsi_exec_channel *src1) 3237{ 3238 dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0]; 3239 dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1]; 3240 dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2]; 3241 dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3]; 3242} 3243 3244static void 3245micro_umin(union tgsi_exec_channel *dst, 3246 const union tgsi_exec_channel *src0, 3247 const union tgsi_exec_channel *src1) 3248{ 3249 dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0]; 3250 dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1]; 3251 dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2]; 3252 dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3]; 3253} 3254 3255static void 3256micro_umod(union tgsi_exec_channel *dst, 3257 const union tgsi_exec_channel *src0, 3258 const union tgsi_exec_channel *src1) 3259{ 3260 dst->u[0] = src0->u[0] % src1->u[0]; 3261 dst->u[1] = src0->u[1] % src1->u[1]; 3262 dst->u[2] = src0->u[2] % src1->u[2]; 3263 dst->u[3] = src0->u[3] % src1->u[3]; 3264} 3265 3266static void 3267micro_umul(union tgsi_exec_channel *dst, 3268 const union tgsi_exec_channel *src0, 3269 const union tgsi_exec_channel *src1) 3270{ 3271 dst->u[0] = src0->u[0] * src1->u[0]; 3272 dst->u[1] = src0->u[1] * src1->u[1]; 3273 dst->u[2] = src0->u[2] * src1->u[2]; 3274 dst->u[3] = src0->u[3] * src1->u[3]; 3275} 3276 3277static void 3278micro_useq(union tgsi_exec_channel *dst, 3279 const union tgsi_exec_channel *src0, 3280 const union tgsi_exec_channel *src1) 3281{ 3282 dst->u[0] = src0->u[0] == src1->u[0] ? ~0 : 0; 3283 dst->u[1] = src0->u[1] == src1->u[1] ? ~0 : 0; 3284 dst->u[2] = src0->u[2] == src1->u[2] ? ~0 : 0; 3285 dst->u[3] = src0->u[3] == src1->u[3] ? ~0 : 0; 3286} 3287 3288static void 3289micro_usge(union tgsi_exec_channel *dst, 3290 const union tgsi_exec_channel *src0, 3291 const union tgsi_exec_channel *src1) 3292{ 3293 dst->u[0] = src0->u[0] >= src1->u[0] ? ~0 : 0; 3294 dst->u[1] = src0->u[1] >= src1->u[1] ? ~0 : 0; 3295 dst->u[2] = src0->u[2] >= src1->u[2] ? ~0 : 0; 3296 dst->u[3] = src0->u[3] >= src1->u[3] ? ~0 : 0; 3297} 3298 3299static void 3300micro_ushr(union tgsi_exec_channel *dst, 3301 const union tgsi_exec_channel *src0, 3302 const union tgsi_exec_channel *src1) 3303{ 3304 dst->u[0] = src0->u[0] >> src1->u[0]; 3305 dst->u[1] = src0->u[1] >> src1->u[1]; 3306 dst->u[2] = src0->u[2] >> src1->u[2]; 3307 dst->u[3] = src0->u[3] >> src1->u[3]; 3308} 3309 3310static void 3311micro_uslt(union tgsi_exec_channel *dst, 3312 const union tgsi_exec_channel *src0, 3313 const union tgsi_exec_channel *src1) 3314{ 3315 dst->u[0] = src0->u[0] < src1->u[0] ? ~0 : 0; 3316 dst->u[1] = src0->u[1] < src1->u[1] ? ~0 : 0; 3317 dst->u[2] = src0->u[2] < src1->u[2] ? ~0 : 0; 3318 dst->u[3] = src0->u[3] < src1->u[3] ? ~0 : 0; 3319} 3320 3321static void 3322micro_usne(union tgsi_exec_channel *dst, 3323 const union tgsi_exec_channel *src0, 3324 const union tgsi_exec_channel *src1) 3325{ 3326 dst->u[0] = src0->u[0] != src1->u[0] ? ~0 : 0; 3327 dst->u[1] = src0->u[1] != src1->u[1] ? ~0 : 0; 3328 dst->u[2] = src0->u[2] != src1->u[2] ? ~0 : 0; 3329 dst->u[3] = src0->u[3] != src1->u[3] ? ~0 : 0; 3330} 3331 3332static void 3333exec_instruction( 3334 struct tgsi_exec_machine *mach, 3335 const struct tgsi_full_instruction *inst, 3336 int *pc ) 3337{ 3338 union tgsi_exec_channel r[10]; 3339 3340 (*pc)++; 3341 3342 switch (inst->Instruction.Opcode) { 3343 case TGSI_OPCODE_ARL: 3344 exec_vector_unary(mach, inst, micro_arl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3345 break; 3346 3347 case TGSI_OPCODE_MOV: 3348 exec_vector_unary(mach, inst, micro_mov, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT); 3349 break; 3350 3351 case TGSI_OPCODE_LIT: 3352 exec_lit(mach, inst); 3353 break; 3354 3355 case TGSI_OPCODE_RCP: 3356 exec_scalar_unary(mach, inst, micro_rcp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3357 break; 3358 3359 case TGSI_OPCODE_RSQ: 3360 exec_scalar_unary(mach, inst, micro_rsq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3361 break; 3362 3363 case TGSI_OPCODE_EXP: 3364 exec_exp(mach, inst); 3365 break; 3366 3367 case TGSI_OPCODE_LOG: 3368 exec_log(mach, inst); 3369 break; 3370 3371 case TGSI_OPCODE_MUL: 3372 exec_vector_binary(mach, inst, micro_mul, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3373 break; 3374 3375 case TGSI_OPCODE_ADD: 3376 exec_vector_binary(mach, inst, micro_add, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3377 break; 3378 3379 case TGSI_OPCODE_DP3: 3380 exec_dp3(mach, inst); 3381 break; 3382 3383 case TGSI_OPCODE_DP4: 3384 exec_dp4(mach, inst); 3385 break; 3386 3387 case TGSI_OPCODE_DST: 3388 exec_dst(mach, inst); 3389 break; 3390 3391 case TGSI_OPCODE_MIN: 3392 exec_vector_binary(mach, inst, micro_min, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3393 break; 3394 3395 case TGSI_OPCODE_MAX: 3396 exec_vector_binary(mach, inst, micro_max, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3397 break; 3398 3399 case TGSI_OPCODE_SLT: 3400 exec_vector_binary(mach, inst, micro_slt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3401 break; 3402 3403 case TGSI_OPCODE_SGE: 3404 exec_vector_binary(mach, inst, micro_sge, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3405 break; 3406 3407 case TGSI_OPCODE_MAD: 3408 exec_vector_trinary(mach, inst, micro_mad, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3409 break; 3410 3411 case TGSI_OPCODE_SUB: 3412 exec_vector_binary(mach, inst, micro_sub, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3413 break; 3414 3415 case TGSI_OPCODE_LRP: 3416 exec_vector_trinary(mach, inst, micro_lrp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3417 break; 3418 3419 case TGSI_OPCODE_CND: 3420 exec_vector_trinary(mach, inst, micro_cnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3421 break; 3422 3423 case TGSI_OPCODE_DP2A: 3424 exec_dp2a(mach, inst); 3425 break; 3426 3427 case TGSI_OPCODE_FRC: 3428 exec_vector_unary(mach, inst, micro_frc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3429 break; 3430 3431 case TGSI_OPCODE_CLAMP: 3432 exec_vector_trinary(mach, inst, micro_clamp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3433 break; 3434 3435 case TGSI_OPCODE_FLR: 3436 exec_vector_unary(mach, inst, micro_flr, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3437 break; 3438 3439 case TGSI_OPCODE_ROUND: 3440 exec_vector_unary(mach, inst, micro_rnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3441 break; 3442 3443 case TGSI_OPCODE_EX2: 3444 exec_scalar_unary(mach, inst, micro_exp2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3445 break; 3446 3447 case TGSI_OPCODE_LG2: 3448 exec_scalar_unary(mach, inst, micro_lg2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3449 break; 3450 3451 case TGSI_OPCODE_POW: 3452 exec_scalar_binary(mach, inst, micro_pow, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3453 break; 3454 3455 case TGSI_OPCODE_XPD: 3456 exec_xpd(mach, inst); 3457 break; 3458 3459 case TGSI_OPCODE_ABS: 3460 exec_vector_unary(mach, inst, micro_abs, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3461 break; 3462 3463 case TGSI_OPCODE_RCC: 3464 exec_scalar_unary(mach, inst, micro_rcc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3465 break; 3466 3467 case TGSI_OPCODE_DPH: 3468 exec_dph(mach, inst); 3469 break; 3470 3471 case TGSI_OPCODE_COS: 3472 exec_scalar_unary(mach, inst, micro_cos, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3473 break; 3474 3475 case TGSI_OPCODE_DDX: 3476 exec_vector_unary(mach, inst, micro_ddx, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3477 break; 3478 3479 case TGSI_OPCODE_DDY: 3480 exec_vector_unary(mach, inst, micro_ddy, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3481 break; 3482 3483 case TGSI_OPCODE_KILP: 3484 exec_kilp (mach, inst); 3485 break; 3486 3487 case TGSI_OPCODE_KIL: 3488 exec_kil (mach, inst); 3489 break; 3490 3491 case TGSI_OPCODE_PK2H: 3492 assert (0); 3493 break; 3494 3495 case TGSI_OPCODE_PK2US: 3496 assert (0); 3497 break; 3498 3499 case TGSI_OPCODE_PK4B: 3500 assert (0); 3501 break; 3502 3503 case TGSI_OPCODE_PK4UB: 3504 assert (0); 3505 break; 3506 3507 case TGSI_OPCODE_RFL: 3508 exec_rfl(mach, inst); 3509 break; 3510 3511 case TGSI_OPCODE_SEQ: 3512 exec_vector_binary(mach, inst, micro_seq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3513 break; 3514 3515 case TGSI_OPCODE_SFL: 3516 exec_vector(mach, inst, micro_sfl, TGSI_EXEC_DATA_FLOAT); 3517 break; 3518 3519 case TGSI_OPCODE_SGT: 3520 exec_vector_binary(mach, inst, micro_sgt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3521 break; 3522 3523 case TGSI_OPCODE_SIN: 3524 exec_scalar_unary(mach, inst, micro_sin, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3525 break; 3526 3527 case TGSI_OPCODE_SLE: 3528 exec_vector_binary(mach, inst, micro_sle, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3529 break; 3530 3531 case TGSI_OPCODE_SNE: 3532 exec_vector_binary(mach, inst, micro_sne, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3533 break; 3534 3535 case TGSI_OPCODE_STR: 3536 exec_vector(mach, inst, micro_str, TGSI_EXEC_DATA_FLOAT); 3537 break; 3538 3539 case TGSI_OPCODE_TEX: 3540 /* simple texture lookup */ 3541 /* src[0] = texcoord */ 3542 /* src[1] = sampler unit */ 3543 exec_tex(mach, inst, TEX_MODIFIER_NONE); 3544 break; 3545 3546 case TGSI_OPCODE_TXB: 3547 /* Texture lookup with lod bias */ 3548 /* src[0] = texcoord (src[0].w = LOD bias) */ 3549 /* src[1] = sampler unit */ 3550 exec_tex(mach, inst, TEX_MODIFIER_LOD_BIAS); 3551 break; 3552 3553 case TGSI_OPCODE_TXD: 3554 /* Texture lookup with explict partial derivatives */ 3555 /* src[0] = texcoord */ 3556 /* src[1] = d[strq]/dx */ 3557 /* src[2] = d[strq]/dy */ 3558 /* src[3] = sampler unit */ 3559 exec_txd(mach, inst); 3560 break; 3561 3562 case TGSI_OPCODE_TXL: 3563 /* Texture lookup with explit LOD */ 3564 /* src[0] = texcoord (src[0].w = LOD) */ 3565 /* src[1] = sampler unit */ 3566 exec_tex(mach, inst, TEX_MODIFIER_EXPLICIT_LOD); 3567 break; 3568 3569 case TGSI_OPCODE_TXP: 3570 /* Texture lookup with projection */ 3571 /* src[0] = texcoord (src[0].w = projection) */ 3572 /* src[1] = sampler unit */ 3573 exec_tex(mach, inst, TEX_MODIFIER_PROJECTED); 3574 break; 3575 3576 case TGSI_OPCODE_UP2H: 3577 assert (0); 3578 break; 3579 3580 case TGSI_OPCODE_UP2US: 3581 assert (0); 3582 break; 3583 3584 case TGSI_OPCODE_UP4B: 3585 assert (0); 3586 break; 3587 3588 case TGSI_OPCODE_UP4UB: 3589 assert (0); 3590 break; 3591 3592 case TGSI_OPCODE_X2D: 3593 exec_x2d(mach, inst); 3594 break; 3595 3596 case TGSI_OPCODE_ARA: 3597 assert (0); 3598 break; 3599 3600 case TGSI_OPCODE_ARR: 3601 exec_vector_unary(mach, inst, micro_arr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3602 break; 3603 3604 case TGSI_OPCODE_BRA: 3605 assert (0); 3606 break; 3607 3608 case TGSI_OPCODE_CAL: 3609 /* skip the call if no execution channels are enabled */ 3610 if (mach->ExecMask) { 3611 /* do the call */ 3612 3613 /* First, record the depths of the execution stacks. 3614 * This is important for deeply nested/looped return statements. 3615 * We have to unwind the stacks by the correct amount. For a 3616 * real code generator, we could determine the number of entries 3617 * to pop off each stack with simple static analysis and avoid 3618 * implementing this data structure at run time. 3619 */ 3620 mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop; 3621 mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop; 3622 mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop; 3623 mach->CallStack[mach->CallStackTop].SwitchStackTop = mach->SwitchStackTop; 3624 mach->CallStack[mach->CallStackTop].BreakStackTop = mach->BreakStackTop; 3625 /* note that PC was already incremented above */ 3626 mach->CallStack[mach->CallStackTop].ReturnAddr = *pc; 3627 3628 mach->CallStackTop++; 3629 3630 /* Second, push the Cond, Loop, Cont, Func stacks */ 3631 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); 3632 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3633 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3634 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING); 3635 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3636 assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING); 3637 3638 mach->CondStack[mach->CondStackTop++] = mach->CondMask; 3639 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; 3640 mach->ContStack[mach->ContStackTop++] = mach->ContMask; 3641 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch; 3642 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3643 mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask; 3644 3645 /* Finally, jump to the subroutine */ 3646 *pc = inst->Label.Label; 3647 } 3648 break; 3649 3650 case TGSI_OPCODE_RET: 3651 mach->FuncMask &= ~mach->ExecMask; 3652 UPDATE_EXEC_MASK(mach); 3653 3654 if (mach->FuncMask == 0x0) { 3655 /* really return now (otherwise, keep executing */ 3656 3657 if (mach->CallStackTop == 0) { 3658 /* returning from main() */ 3659 mach->CondStackTop = 0; 3660 mach->LoopStackTop = 0; 3661 *pc = -1; 3662 return; 3663 } 3664 3665 assert(mach->CallStackTop > 0); 3666 mach->CallStackTop--; 3667 3668 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; 3669 mach->CondMask = mach->CondStack[mach->CondStackTop]; 3670 3671 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; 3672 mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; 3673 3674 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; 3675 mach->ContMask = mach->ContStack[mach->ContStackTop]; 3676 3677 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop; 3678 mach->Switch = mach->SwitchStack[mach->SwitchStackTop]; 3679 3680 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop; 3681 mach->BreakType = mach->BreakStack[mach->BreakStackTop]; 3682 3683 assert(mach->FuncStackTop > 0); 3684 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; 3685 3686 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; 3687 3688 UPDATE_EXEC_MASK(mach); 3689 } 3690 break; 3691 3692 case TGSI_OPCODE_SSG: 3693 exec_vector_unary(mach, inst, micro_sgn, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3694 break; 3695 3696 case TGSI_OPCODE_CMP: 3697 exec_vector_trinary(mach, inst, micro_cmp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3698 break; 3699 3700 case TGSI_OPCODE_SCS: 3701 exec_scs(mach, inst); 3702 break; 3703 3704 case TGSI_OPCODE_NRM: 3705 exec_nrm3(mach, inst); 3706 break; 3707 3708 case TGSI_OPCODE_NRM4: 3709 exec_nrm4(mach, inst); 3710 break; 3711 3712 case TGSI_OPCODE_DIV: 3713 exec_vector_binary(mach, inst, micro_div, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3714 break; 3715 3716 case TGSI_OPCODE_DP2: 3717 exec_dp2(mach, inst); 3718 break; 3719 3720 case TGSI_OPCODE_IF: 3721 /* push CondMask */ 3722 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); 3723 mach->CondStack[mach->CondStackTop++] = mach->CondMask; 3724 FETCH( &r[0], 0, CHAN_X ); 3725 /* update CondMask */ 3726 if( ! r[0].u[0] ) { 3727 mach->CondMask &= ~0x1; 3728 } 3729 if( ! r[0].u[1] ) { 3730 mach->CondMask &= ~0x2; 3731 } 3732 if( ! r[0].u[2] ) { 3733 mach->CondMask &= ~0x4; 3734 } 3735 if( ! r[0].u[3] ) { 3736 mach->CondMask &= ~0x8; 3737 } 3738 UPDATE_EXEC_MASK(mach); 3739 /* Todo: If CondMask==0, jump to ELSE */ 3740 break; 3741 3742 case TGSI_OPCODE_ELSE: 3743 /* invert CondMask wrt previous mask */ 3744 { 3745 uint prevMask; 3746 assert(mach->CondStackTop > 0); 3747 prevMask = mach->CondStack[mach->CondStackTop - 1]; 3748 mach->CondMask = ~mach->CondMask & prevMask; 3749 UPDATE_EXEC_MASK(mach); 3750 /* Todo: If CondMask==0, jump to ENDIF */ 3751 } 3752 break; 3753 3754 case TGSI_OPCODE_ENDIF: 3755 /* pop CondMask */ 3756 assert(mach->CondStackTop > 0); 3757 mach->CondMask = mach->CondStack[--mach->CondStackTop]; 3758 UPDATE_EXEC_MASK(mach); 3759 break; 3760 3761 case TGSI_OPCODE_END: 3762 /* make sure we end primitives which haven't 3763 * been explicitly emitted */ 3764 conditional_emit_primitive(mach); 3765 /* halt execution */ 3766 *pc = -1; 3767 break; 3768 3769 case TGSI_OPCODE_PUSHA: 3770 assert (0); 3771 break; 3772 3773 case TGSI_OPCODE_POPA: 3774 assert (0); 3775 break; 3776 3777 case TGSI_OPCODE_CEIL: 3778 exec_vector_unary(mach, inst, micro_ceil, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3779 break; 3780 3781 case TGSI_OPCODE_I2F: 3782 exec_vector_unary(mach, inst, micro_i2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_INT); 3783 break; 3784 3785 case TGSI_OPCODE_NOT: 3786 exec_vector_unary(mach, inst, micro_not, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3787 break; 3788 3789 case TGSI_OPCODE_TRUNC: 3790 exec_vector_unary(mach, inst, micro_trunc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3791 break; 3792 3793 case TGSI_OPCODE_SHL: 3794 exec_vector_binary(mach, inst, micro_shl, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3795 break; 3796 3797 case TGSI_OPCODE_AND: 3798 exec_vector_binary(mach, inst, micro_and, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3799 break; 3800 3801 case TGSI_OPCODE_OR: 3802 exec_vector_binary(mach, inst, micro_or, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3803 break; 3804 3805 case TGSI_OPCODE_MOD: 3806 exec_vector_binary(mach, inst, micro_mod, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3807 break; 3808 3809 case TGSI_OPCODE_XOR: 3810 exec_vector_binary(mach, inst, micro_xor, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3811 break; 3812 3813 case TGSI_OPCODE_SAD: 3814 assert (0); 3815 break; 3816 3817 case TGSI_OPCODE_TXF: 3818 exec_txf(mach, inst); 3819 break; 3820 3821 case TGSI_OPCODE_TXQ: 3822 exec_txq(mach, inst); 3823 break; 3824 3825 case TGSI_OPCODE_EMIT: 3826 emit_vertex(mach); 3827 break; 3828 3829 case TGSI_OPCODE_ENDPRIM: 3830 emit_primitive(mach); 3831 break; 3832 3833 case TGSI_OPCODE_BGNLOOP: 3834 /* push LoopMask and ContMasks */ 3835 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3836 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3837 assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3838 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3839 3840 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; 3841 mach->ContStack[mach->ContStackTop++] = mach->ContMask; 3842 mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1; 3843 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3844 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP; 3845 break; 3846 3847 case TGSI_OPCODE_ENDLOOP: 3848 /* Restore ContMask, but don't pop */ 3849 assert(mach->ContStackTop > 0); 3850 mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; 3851 UPDATE_EXEC_MASK(mach); 3852 if (mach->ExecMask) { 3853 /* repeat loop: jump to instruction just past BGNLOOP */ 3854 assert(mach->LoopLabelStackTop > 0); 3855 *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1; 3856 } 3857 else { 3858 /* exit loop: pop LoopMask */ 3859 assert(mach->LoopStackTop > 0); 3860 mach->LoopMask = mach->LoopStack[--mach->LoopStackTop]; 3861 /* pop ContMask */ 3862 assert(mach->ContStackTop > 0); 3863 mach->ContMask = mach->ContStack[--mach->ContStackTop]; 3864 assert(mach->LoopLabelStackTop > 0); 3865 --mach->LoopLabelStackTop; 3866 3867 mach->BreakType = mach->BreakStack[--mach->BreakStackTop]; 3868 } 3869 UPDATE_EXEC_MASK(mach); 3870 break; 3871 3872 case TGSI_OPCODE_BRK: 3873 exec_break(mach); 3874 break; 3875 3876 case TGSI_OPCODE_CONT: 3877 /* turn off cont channels for each enabled exec channel */ 3878 mach->ContMask &= ~mach->ExecMask; 3879 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 3880 UPDATE_EXEC_MASK(mach); 3881 break; 3882 3883 case TGSI_OPCODE_BGNSUB: 3884 /* no-op */ 3885 break; 3886 3887 case TGSI_OPCODE_ENDSUB: 3888 /* 3889 * XXX: This really should be a no-op. We should never reach this opcode. 3890 */ 3891 3892 assert(mach->CallStackTop > 0); 3893 mach->CallStackTop--; 3894 3895 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; 3896 mach->CondMask = mach->CondStack[mach->CondStackTop]; 3897 3898 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; 3899 mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; 3900 3901 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; 3902 mach->ContMask = mach->ContStack[mach->ContStackTop]; 3903 3904 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop; 3905 mach->Switch = mach->SwitchStack[mach->SwitchStackTop]; 3906 3907 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop; 3908 mach->BreakType = mach->BreakStack[mach->BreakStackTop]; 3909 3910 assert(mach->FuncStackTop > 0); 3911 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; 3912 3913 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; 3914 3915 UPDATE_EXEC_MASK(mach); 3916 break; 3917 3918 case TGSI_OPCODE_NOP: 3919 break; 3920 3921 case TGSI_OPCODE_BREAKC: 3922 FETCH(&r[0], 0, CHAN_X); 3923 /* update CondMask */ 3924 if (r[0].u[0] && (mach->ExecMask & 0x1)) { 3925 mach->LoopMask &= ~0x1; 3926 } 3927 if (r[0].u[1] && (mach->ExecMask & 0x2)) { 3928 mach->LoopMask &= ~0x2; 3929 } 3930 if (r[0].u[2] && (mach->ExecMask & 0x4)) { 3931 mach->LoopMask &= ~0x4; 3932 } 3933 if (r[0].u[3] && (mach->ExecMask & 0x8)) { 3934 mach->LoopMask &= ~0x8; 3935 } 3936 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 3937 UPDATE_EXEC_MASK(mach); 3938 break; 3939 3940 case TGSI_OPCODE_F2I: 3941 exec_vector_unary(mach, inst, micro_f2i, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3942 break; 3943 3944 case TGSI_OPCODE_IDIV: 3945 exec_vector_binary(mach, inst, micro_idiv, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3946 break; 3947 3948 case TGSI_OPCODE_IMAX: 3949 exec_vector_binary(mach, inst, micro_imax, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3950 break; 3951 3952 case TGSI_OPCODE_IMIN: 3953 exec_vector_binary(mach, inst, micro_imin, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3954 break; 3955 3956 case TGSI_OPCODE_INEG: 3957 exec_vector_unary(mach, inst, micro_ineg, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3958 break; 3959 3960 case TGSI_OPCODE_ISGE: 3961 exec_vector_binary(mach, inst, micro_isge, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3962 break; 3963 3964 case TGSI_OPCODE_ISHR: 3965 exec_vector_binary(mach, inst, micro_ishr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3966 break; 3967 3968 case TGSI_OPCODE_ISLT: 3969 exec_vector_binary(mach, inst, micro_islt, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3970 break; 3971 3972 case TGSI_OPCODE_F2U: 3973 exec_vector_unary(mach, inst, micro_f2u, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT); 3974 break; 3975 3976 case TGSI_OPCODE_U2F: 3977 exec_vector_unary(mach, inst, micro_u2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_UINT); 3978 break; 3979 3980 case TGSI_OPCODE_UADD: 3981 exec_vector_binary(mach, inst, micro_uadd, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3982 break; 3983 3984 case TGSI_OPCODE_UDIV: 3985 exec_vector_binary(mach, inst, micro_udiv, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3986 break; 3987 3988 case TGSI_OPCODE_UMAD: 3989 exec_vector_trinary(mach, inst, micro_umad, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3990 break; 3991 3992 case TGSI_OPCODE_UMAX: 3993 exec_vector_binary(mach, inst, micro_umax, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3994 break; 3995 3996 case TGSI_OPCODE_UMIN: 3997 exec_vector_binary(mach, inst, micro_umin, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3998 break; 3999 4000 case TGSI_OPCODE_UMOD: 4001 exec_vector_binary(mach, inst, micro_umod, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4002 break; 4003 4004 case TGSI_OPCODE_UMUL: 4005 exec_vector_binary(mach, inst, micro_umul, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4006 break; 4007 4008 case TGSI_OPCODE_USEQ: 4009 exec_vector_binary(mach, inst, micro_useq, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4010 break; 4011 4012 case TGSI_OPCODE_USGE: 4013 exec_vector_binary(mach, inst, micro_usge, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4014 break; 4015 4016 case TGSI_OPCODE_USHR: 4017 exec_vector_binary(mach, inst, micro_ushr, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4018 break; 4019 4020 case TGSI_OPCODE_USLT: 4021 exec_vector_binary(mach, inst, micro_uslt, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4022 break; 4023 4024 case TGSI_OPCODE_USNE: 4025 exec_vector_binary(mach, inst, micro_usne, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4026 break; 4027 4028 case TGSI_OPCODE_SWITCH: 4029 exec_switch(mach, inst); 4030 break; 4031 4032 case TGSI_OPCODE_CASE: 4033 exec_case(mach, inst); 4034 break; 4035 4036 case TGSI_OPCODE_DEFAULT: 4037 exec_default(mach); 4038 break; 4039 4040 case TGSI_OPCODE_ENDSWITCH: 4041 exec_endswitch(mach); 4042 break; 4043 4044 case TGSI_OPCODE_LOAD: 4045 assert(0); 4046 break; 4047 4048 case TGSI_OPCODE_LOAD_MS: 4049 assert(0); 4050 break; 4051 4052 case TGSI_OPCODE_SAMPLE: 4053 exec_sample(mach, inst, TEX_MODIFIER_NONE); 4054 break; 4055 4056 case TGSI_OPCODE_SAMPLE_B: 4057 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS); 4058 break; 4059 4060 case TGSI_OPCODE_SAMPLE_C: 4061 exec_sample(mach, inst, TEX_MODIFIER_NONE); 4062 break; 4063 4064 case TGSI_OPCODE_SAMPLE_C_LZ: 4065 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS); 4066 break; 4067 4068 case TGSI_OPCODE_SAMPLE_D: 4069 exec_sample_d(mach, inst); 4070 break; 4071 4072 case TGSI_OPCODE_SAMPLE_L: 4073 exec_sample(mach, inst, TEX_MODIFIER_EXPLICIT_LOD); 4074 break; 4075 4076 case TGSI_OPCODE_GATHER4: 4077 assert(0); 4078 break; 4079 4080 case TGSI_OPCODE_RESINFO: 4081 assert(0); 4082 break; 4083 4084 case TGSI_OPCODE_SAMPLE_POS: 4085 assert(0); 4086 break; 4087 4088 case TGSI_OPCODE_SAMPLE_INFO: 4089 assert(0); 4090 break; 4091 4092 default: 4093 assert( 0 ); 4094 } 4095} 4096 4097 4098#define DEBUG_EXECUTION 0 4099 4100 4101/** 4102 * Run TGSI interpreter. 4103 * \return bitmask of "alive" quad components 4104 */ 4105uint 4106tgsi_exec_machine_run( struct tgsi_exec_machine *mach ) 4107{ 4108 uint i; 4109 int pc = 0; 4110 4111 mach->CondMask = 0xf; 4112 mach->LoopMask = 0xf; 4113 mach->ContMask = 0xf; 4114 mach->FuncMask = 0xf; 4115 mach->ExecMask = 0xf; 4116 4117 mach->Switch.mask = 0xf; 4118 4119 assert(mach->CondStackTop == 0); 4120 assert(mach->LoopStackTop == 0); 4121 assert(mach->ContStackTop == 0); 4122 assert(mach->SwitchStackTop == 0); 4123 assert(mach->BreakStackTop == 0); 4124 assert(mach->CallStackTop == 0); 4125 4126 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0; 4127 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0; 4128 4129 if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) { 4130 mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0; 4131 mach->Primitives[0] = 0; 4132 } 4133 4134 /* execute declarations (interpolants) */ 4135 for (i = 0; i < mach->NumDeclarations; i++) { 4136 exec_declaration( mach, mach->Declarations+i ); 4137 } 4138 4139 { 4140#if DEBUG_EXECUTION 4141 struct tgsi_exec_vector temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS]; 4142 struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS]; 4143 uint inst = 1; 4144 4145 memcpy(temps, mach->Temps, sizeof(temps)); 4146 memcpy(outputs, mach->Outputs, sizeof(outputs)); 4147#endif 4148 4149 /* execute instructions, until pc is set to -1 */ 4150 while (pc != -1) { 4151 4152#if DEBUG_EXECUTION 4153 uint i; 4154 4155 tgsi_dump_instruction(&mach->Instructions[pc], inst++); 4156#endif 4157 4158 assert(pc < (int) mach->NumInstructions); 4159 exec_instruction(mach, mach->Instructions + pc, &pc); 4160 4161#if DEBUG_EXECUTION 4162 for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) { 4163 if (memcmp(&temps[i], &mach->Temps[i], sizeof(temps[i]))) { 4164 uint j; 4165 4166 memcpy(&temps[i], &mach->Temps[i], sizeof(temps[i])); 4167 debug_printf("TEMP[%2u] = ", i); 4168 for (j = 0; j < 4; j++) { 4169 if (j > 0) { 4170 debug_printf(" "); 4171 } 4172 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n", 4173 temps[i].xyzw[0].f[j], temps[i].xyzw[0].u[j], 4174 temps[i].xyzw[1].f[j], temps[i].xyzw[1].u[j], 4175 temps[i].xyzw[2].f[j], temps[i].xyzw[2].u[j], 4176 temps[i].xyzw[3].f[j], temps[i].xyzw[3].u[j]); 4177 } 4178 } 4179 } 4180 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { 4181 if (memcmp(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]))) { 4182 uint j; 4183 4184 memcpy(&outputs[i], &mach->Outputs[i], sizeof(outputs[i])); 4185 debug_printf("OUT[%2u] = ", i); 4186 for (j = 0; j < 4; j++) { 4187 if (j > 0) { 4188 debug_printf(" "); 4189 } 4190 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n", 4191 outputs[i].xyzw[0].f[j], outputs[i].xyzw[0].u[j], 4192 outputs[i].xyzw[1].f[j], outputs[i].xyzw[1].u[j], 4193 outputs[i].xyzw[2].f[j], outputs[i].xyzw[2].u[j], 4194 outputs[i].xyzw[3].f[j], outputs[i].xyzw[3].u[j]); 4195 } 4196 } 4197 } 4198#endif 4199 } 4200 } 4201 4202#if 0 4203 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */ 4204 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { 4205 /* 4206 * Scale back depth component. 4207 */ 4208 for (i = 0; i < 4; i++) 4209 mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF; 4210 } 4211#endif 4212 4213 /* Strictly speaking, these assertions aren't really needed but they 4214 * can potentially catch some bugs in the control flow code. 4215 */ 4216 assert(mach->CondStackTop == 0); 4217 assert(mach->LoopStackTop == 0); 4218 assert(mach->ContStackTop == 0); 4219 assert(mach->SwitchStackTop == 0); 4220 assert(mach->BreakStackTop == 0); 4221 assert(mach->CallStackTop == 0); 4222 4223 return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0]; 4224} 4225