tgsi_exec.c revision b6cbc28533a3fd68dbfe694c0774735233df8758
1/************************************************************************** 2 * 3 * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas. 4 * All Rights Reserved. 5 * Copyright 2009-2010 VMware, Inc. All rights Reserved. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a 8 * copy of this software and associated documentation files (the 9 * "Software"), to deal in the Software without restriction, including 10 * without limitation the rights to use, copy, modify, merge, publish, 11 * distribute, sub license, and/or sell copies of the Software, and to 12 * permit persons to whom the Software is furnished to do so, subject to 13 * the following conditions: 14 * 15 * The above copyright notice and this permission notice (including the 16 * next paragraph) shall be included in all copies or substantial portions 17 * of the Software. 18 * 19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. 22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR 23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 26 * 27 **************************************************************************/ 28 29/** 30 * TGSI interpreter/executor. 31 * 32 * Flow control information: 33 * 34 * Since we operate on 'quads' (4 pixels or 4 vertices in parallel) 35 * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special 36 * care since a condition may be true for some quad components but false 37 * for other components. 38 * 39 * We basically execute all statements (even if they're in the part of 40 * an IF/ELSE clause that's "not taken") and use a special mask to 41 * control writing to destination registers. This is the ExecMask. 42 * See store_dest(). 43 * 44 * The ExecMask is computed from three other masks (CondMask, LoopMask and 45 * ContMask) which are controlled by the flow control instructions (namely: 46 * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT). 47 * 48 * 49 * Authors: 50 * Michal Krol 51 * Brian Paul 52 */ 53 54#include "pipe/p_compiler.h" 55#include "pipe/p_state.h" 56#include "pipe/p_shader_tokens.h" 57#include "tgsi/tgsi_dump.h" 58#include "tgsi/tgsi_parse.h" 59#include "tgsi/tgsi_util.h" 60#include "tgsi_exec.h" 61#include "util/u_memory.h" 62#include "util/u_math.h" 63 64 65#define FAST_MATH 0 66 67#define TILE_TOP_LEFT 0 68#define TILE_TOP_RIGHT 1 69#define TILE_BOTTOM_LEFT 2 70#define TILE_BOTTOM_RIGHT 3 71 72static void 73micro_abs(union tgsi_exec_channel *dst, 74 const union tgsi_exec_channel *src) 75{ 76 dst->f[0] = fabsf(src->f[0]); 77 dst->f[1] = fabsf(src->f[1]); 78 dst->f[2] = fabsf(src->f[2]); 79 dst->f[3] = fabsf(src->f[3]); 80} 81 82static void 83micro_arl(union tgsi_exec_channel *dst, 84 const union tgsi_exec_channel *src) 85{ 86 dst->i[0] = (int)floorf(src->f[0]); 87 dst->i[1] = (int)floorf(src->f[1]); 88 dst->i[2] = (int)floorf(src->f[2]); 89 dst->i[3] = (int)floorf(src->f[3]); 90} 91 92static void 93micro_arr(union tgsi_exec_channel *dst, 94 const union tgsi_exec_channel *src) 95{ 96 dst->i[0] = (int)floorf(src->f[0] + 0.5f); 97 dst->i[1] = (int)floorf(src->f[1] + 0.5f); 98 dst->i[2] = (int)floorf(src->f[2] + 0.5f); 99 dst->i[3] = (int)floorf(src->f[3] + 0.5f); 100} 101 102static void 103micro_ceil(union tgsi_exec_channel *dst, 104 const union tgsi_exec_channel *src) 105{ 106 dst->f[0] = ceilf(src->f[0]); 107 dst->f[1] = ceilf(src->f[1]); 108 dst->f[2] = ceilf(src->f[2]); 109 dst->f[3] = ceilf(src->f[3]); 110} 111 112static void 113micro_clamp(union tgsi_exec_channel *dst, 114 const union tgsi_exec_channel *src0, 115 const union tgsi_exec_channel *src1, 116 const union tgsi_exec_channel *src2) 117{ 118 dst->f[0] = src0->f[0] < src1->f[0] ? src1->f[0] : src0->f[0] > src2->f[0] ? src2->f[0] : src0->f[0]; 119 dst->f[1] = src0->f[1] < src1->f[1] ? src1->f[1] : src0->f[1] > src2->f[1] ? src2->f[1] : src0->f[1]; 120 dst->f[2] = src0->f[2] < src1->f[2] ? src1->f[2] : src0->f[2] > src2->f[2] ? src2->f[2] : src0->f[2]; 121 dst->f[3] = src0->f[3] < src1->f[3] ? src1->f[3] : src0->f[3] > src2->f[3] ? src2->f[3] : src0->f[3]; 122} 123 124static void 125micro_cmp(union tgsi_exec_channel *dst, 126 const union tgsi_exec_channel *src0, 127 const union tgsi_exec_channel *src1, 128 const union tgsi_exec_channel *src2) 129{ 130 dst->f[0] = src0->f[0] < 0.0f ? src1->f[0] : src2->f[0]; 131 dst->f[1] = src0->f[1] < 0.0f ? src1->f[1] : src2->f[1]; 132 dst->f[2] = src0->f[2] < 0.0f ? src1->f[2] : src2->f[2]; 133 dst->f[3] = src0->f[3] < 0.0f ? src1->f[3] : src2->f[3]; 134} 135 136static void 137micro_cnd(union tgsi_exec_channel *dst, 138 const union tgsi_exec_channel *src0, 139 const union tgsi_exec_channel *src1, 140 const union tgsi_exec_channel *src2) 141{ 142 dst->f[0] = src2->f[0] > 0.5f ? src0->f[0] : src1->f[0]; 143 dst->f[1] = src2->f[1] > 0.5f ? src0->f[1] : src1->f[1]; 144 dst->f[2] = src2->f[2] > 0.5f ? src0->f[2] : src1->f[2]; 145 dst->f[3] = src2->f[3] > 0.5f ? src0->f[3] : src1->f[3]; 146} 147 148static void 149micro_cos(union tgsi_exec_channel *dst, 150 const union tgsi_exec_channel *src) 151{ 152 dst->f[0] = cosf(src->f[0]); 153 dst->f[1] = cosf(src->f[1]); 154 dst->f[2] = cosf(src->f[2]); 155 dst->f[3] = cosf(src->f[3]); 156} 157 158static void 159micro_ddx(union tgsi_exec_channel *dst, 160 const union tgsi_exec_channel *src) 161{ 162 dst->f[0] = 163 dst->f[1] = 164 dst->f[2] = 165 dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_BOTTOM_LEFT]; 166} 167 168static void 169micro_ddy(union tgsi_exec_channel *dst, 170 const union tgsi_exec_channel *src) 171{ 172 dst->f[0] = 173 dst->f[1] = 174 dst->f[2] = 175 dst->f[3] = src->f[TILE_BOTTOM_LEFT] - src->f[TILE_TOP_LEFT]; 176} 177 178static void 179micro_exp2(union tgsi_exec_channel *dst, 180 const union tgsi_exec_channel *src) 181{ 182#if FAST_MATH 183 dst->f[0] = util_fast_exp2(src->f[0]); 184 dst->f[1] = util_fast_exp2(src->f[1]); 185 dst->f[2] = util_fast_exp2(src->f[2]); 186 dst->f[3] = util_fast_exp2(src->f[3]); 187#else 188#if DEBUG 189 /* Inf is okay for this instruction, so clamp it to silence assertions. */ 190 uint i; 191 union tgsi_exec_channel clamped; 192 193 for (i = 0; i < 4; i++) { 194 if (src->f[i] > 127.99999f) { 195 clamped.f[i] = 127.99999f; 196 } else if (src->f[i] < -126.99999f) { 197 clamped.f[i] = -126.99999f; 198 } else { 199 clamped.f[i] = src->f[i]; 200 } 201 } 202 src = &clamped; 203#endif /* DEBUG */ 204 205 dst->f[0] = powf(2.0f, src->f[0]); 206 dst->f[1] = powf(2.0f, src->f[1]); 207 dst->f[2] = powf(2.0f, src->f[2]); 208 dst->f[3] = powf(2.0f, src->f[3]); 209#endif /* FAST_MATH */ 210} 211 212static void 213micro_flr(union tgsi_exec_channel *dst, 214 const union tgsi_exec_channel *src) 215{ 216 dst->f[0] = floorf(src->f[0]); 217 dst->f[1] = floorf(src->f[1]); 218 dst->f[2] = floorf(src->f[2]); 219 dst->f[3] = floorf(src->f[3]); 220} 221 222static void 223micro_frc(union tgsi_exec_channel *dst, 224 const union tgsi_exec_channel *src) 225{ 226 dst->f[0] = src->f[0] - floorf(src->f[0]); 227 dst->f[1] = src->f[1] - floorf(src->f[1]); 228 dst->f[2] = src->f[2] - floorf(src->f[2]); 229 dst->f[3] = src->f[3] - floorf(src->f[3]); 230} 231 232static void 233micro_iabs(union tgsi_exec_channel *dst, 234 const union tgsi_exec_channel *src) 235{ 236 dst->i[0] = src->i[0] >= 0 ? src->i[0] : -src->i[0]; 237 dst->i[1] = src->i[1] >= 0 ? src->i[1] : -src->i[1]; 238 dst->i[2] = src->i[2] >= 0 ? src->i[2] : -src->i[2]; 239 dst->i[3] = src->i[3] >= 0 ? src->i[3] : -src->i[3]; 240} 241 242static void 243micro_ineg(union tgsi_exec_channel *dst, 244 const union tgsi_exec_channel *src) 245{ 246 dst->i[0] = -src->i[0]; 247 dst->i[1] = -src->i[1]; 248 dst->i[2] = -src->i[2]; 249 dst->i[3] = -src->i[3]; 250} 251 252static void 253micro_lg2(union tgsi_exec_channel *dst, 254 const union tgsi_exec_channel *src) 255{ 256#if FAST_MATH 257 dst->f[0] = util_fast_log2(src->f[0]); 258 dst->f[1] = util_fast_log2(src->f[1]); 259 dst->f[2] = util_fast_log2(src->f[2]); 260 dst->f[3] = util_fast_log2(src->f[3]); 261#else 262 dst->f[0] = logf(src->f[0]) * 1.442695f; 263 dst->f[1] = logf(src->f[1]) * 1.442695f; 264 dst->f[2] = logf(src->f[2]) * 1.442695f; 265 dst->f[3] = logf(src->f[3]) * 1.442695f; 266#endif 267} 268 269static void 270micro_lrp(union tgsi_exec_channel *dst, 271 const union tgsi_exec_channel *src0, 272 const union tgsi_exec_channel *src1, 273 const union tgsi_exec_channel *src2) 274{ 275 dst->f[0] = src0->f[0] * (src1->f[0] - src2->f[0]) + src2->f[0]; 276 dst->f[1] = src0->f[1] * (src1->f[1] - src2->f[1]) + src2->f[1]; 277 dst->f[2] = src0->f[2] * (src1->f[2] - src2->f[2]) + src2->f[2]; 278 dst->f[3] = src0->f[3] * (src1->f[3] - src2->f[3]) + src2->f[3]; 279} 280 281static void 282micro_mad(union tgsi_exec_channel *dst, 283 const union tgsi_exec_channel *src0, 284 const union tgsi_exec_channel *src1, 285 const union tgsi_exec_channel *src2) 286{ 287 dst->f[0] = src0->f[0] * src1->f[0] + src2->f[0]; 288 dst->f[1] = src0->f[1] * src1->f[1] + src2->f[1]; 289 dst->f[2] = src0->f[2] * src1->f[2] + src2->f[2]; 290 dst->f[3] = src0->f[3] * src1->f[3] + src2->f[3]; 291} 292 293static void 294micro_mov(union tgsi_exec_channel *dst, 295 const union tgsi_exec_channel *src) 296{ 297 dst->u[0] = src->u[0]; 298 dst->u[1] = src->u[1]; 299 dst->u[2] = src->u[2]; 300 dst->u[3] = src->u[3]; 301} 302 303static void 304micro_rcp(union tgsi_exec_channel *dst, 305 const union tgsi_exec_channel *src) 306{ 307#if 0 /* for debugging */ 308 assert(src->f[0] != 0.0f); 309 assert(src->f[1] != 0.0f); 310 assert(src->f[2] != 0.0f); 311 assert(src->f[3] != 0.0f); 312#endif 313 dst->f[0] = 1.0f / src->f[0]; 314 dst->f[1] = 1.0f / src->f[1]; 315 dst->f[2] = 1.0f / src->f[2]; 316 dst->f[3] = 1.0f / src->f[3]; 317} 318 319static void 320micro_rnd(union tgsi_exec_channel *dst, 321 const union tgsi_exec_channel *src) 322{ 323 dst->f[0] = floorf(src->f[0] + 0.5f); 324 dst->f[1] = floorf(src->f[1] + 0.5f); 325 dst->f[2] = floorf(src->f[2] + 0.5f); 326 dst->f[3] = floorf(src->f[3] + 0.5f); 327} 328 329static void 330micro_rsq(union tgsi_exec_channel *dst, 331 const union tgsi_exec_channel *src) 332{ 333#if 0 /* for debugging */ 334 assert(src->f[0] != 0.0f); 335 assert(src->f[1] != 0.0f); 336 assert(src->f[2] != 0.0f); 337 assert(src->f[3] != 0.0f); 338#endif 339 dst->f[0] = 1.0f / sqrtf(fabsf(src->f[0])); 340 dst->f[1] = 1.0f / sqrtf(fabsf(src->f[1])); 341 dst->f[2] = 1.0f / sqrtf(fabsf(src->f[2])); 342 dst->f[3] = 1.0f / sqrtf(fabsf(src->f[3])); 343} 344 345static void 346micro_seq(union tgsi_exec_channel *dst, 347 const union tgsi_exec_channel *src0, 348 const union tgsi_exec_channel *src1) 349{ 350 dst->f[0] = src0->f[0] == src1->f[0] ? 1.0f : 0.0f; 351 dst->f[1] = src0->f[1] == src1->f[1] ? 1.0f : 0.0f; 352 dst->f[2] = src0->f[2] == src1->f[2] ? 1.0f : 0.0f; 353 dst->f[3] = src0->f[3] == src1->f[3] ? 1.0f : 0.0f; 354} 355 356static void 357micro_sge(union tgsi_exec_channel *dst, 358 const union tgsi_exec_channel *src0, 359 const union tgsi_exec_channel *src1) 360{ 361 dst->f[0] = src0->f[0] >= src1->f[0] ? 1.0f : 0.0f; 362 dst->f[1] = src0->f[1] >= src1->f[1] ? 1.0f : 0.0f; 363 dst->f[2] = src0->f[2] >= src1->f[2] ? 1.0f : 0.0f; 364 dst->f[3] = src0->f[3] >= src1->f[3] ? 1.0f : 0.0f; 365} 366 367static void 368micro_sgn(union tgsi_exec_channel *dst, 369 const union tgsi_exec_channel *src) 370{ 371 dst->f[0] = src->f[0] < 0.0f ? -1.0f : src->f[0] > 0.0f ? 1.0f : 0.0f; 372 dst->f[1] = src->f[1] < 0.0f ? -1.0f : src->f[1] > 0.0f ? 1.0f : 0.0f; 373 dst->f[2] = src->f[2] < 0.0f ? -1.0f : src->f[2] > 0.0f ? 1.0f : 0.0f; 374 dst->f[3] = src->f[3] < 0.0f ? -1.0f : src->f[3] > 0.0f ? 1.0f : 0.0f; 375} 376 377static void 378micro_isgn(union tgsi_exec_channel *dst, 379 const union tgsi_exec_channel *src) 380{ 381 dst->i[0] = src->i[0] < 0 ? -1 : src->i[0] > 0 ? 1 : 0; 382 dst->i[1] = src->i[1] < 0 ? -1 : src->i[1] > 0 ? 1 : 0; 383 dst->i[2] = src->i[2] < 0 ? -1 : src->i[2] > 0 ? 1 : 0; 384 dst->i[3] = src->i[3] < 0 ? -1 : src->i[3] > 0 ? 1 : 0; 385} 386 387static void 388micro_sgt(union tgsi_exec_channel *dst, 389 const union tgsi_exec_channel *src0, 390 const union tgsi_exec_channel *src1) 391{ 392 dst->f[0] = src0->f[0] > src1->f[0] ? 1.0f : 0.0f; 393 dst->f[1] = src0->f[1] > src1->f[1] ? 1.0f : 0.0f; 394 dst->f[2] = src0->f[2] > src1->f[2] ? 1.0f : 0.0f; 395 dst->f[3] = src0->f[3] > src1->f[3] ? 1.0f : 0.0f; 396} 397 398static void 399micro_sin(union tgsi_exec_channel *dst, 400 const union tgsi_exec_channel *src) 401{ 402 dst->f[0] = sinf(src->f[0]); 403 dst->f[1] = sinf(src->f[1]); 404 dst->f[2] = sinf(src->f[2]); 405 dst->f[3] = sinf(src->f[3]); 406} 407 408static void 409micro_sle(union tgsi_exec_channel *dst, 410 const union tgsi_exec_channel *src0, 411 const union tgsi_exec_channel *src1) 412{ 413 dst->f[0] = src0->f[0] <= src1->f[0] ? 1.0f : 0.0f; 414 dst->f[1] = src0->f[1] <= src1->f[1] ? 1.0f : 0.0f; 415 dst->f[2] = src0->f[2] <= src1->f[2] ? 1.0f : 0.0f; 416 dst->f[3] = src0->f[3] <= src1->f[3] ? 1.0f : 0.0f; 417} 418 419static void 420micro_slt(union tgsi_exec_channel *dst, 421 const union tgsi_exec_channel *src0, 422 const union tgsi_exec_channel *src1) 423{ 424 dst->f[0] = src0->f[0] < src1->f[0] ? 1.0f : 0.0f; 425 dst->f[1] = src0->f[1] < src1->f[1] ? 1.0f : 0.0f; 426 dst->f[2] = src0->f[2] < src1->f[2] ? 1.0f : 0.0f; 427 dst->f[3] = src0->f[3] < src1->f[3] ? 1.0f : 0.0f; 428} 429 430static void 431micro_sne(union tgsi_exec_channel *dst, 432 const union tgsi_exec_channel *src0, 433 const union tgsi_exec_channel *src1) 434{ 435 dst->f[0] = src0->f[0] != src1->f[0] ? 1.0f : 0.0f; 436 dst->f[1] = src0->f[1] != src1->f[1] ? 1.0f : 0.0f; 437 dst->f[2] = src0->f[2] != src1->f[2] ? 1.0f : 0.0f; 438 dst->f[3] = src0->f[3] != src1->f[3] ? 1.0f : 0.0f; 439} 440 441static void 442micro_sfl(union tgsi_exec_channel *dst) 443{ 444 dst->f[0] = 0.0f; 445 dst->f[1] = 0.0f; 446 dst->f[2] = 0.0f; 447 dst->f[3] = 0.0f; 448} 449 450static void 451micro_str(union tgsi_exec_channel *dst) 452{ 453 dst->f[0] = 1.0f; 454 dst->f[1] = 1.0f; 455 dst->f[2] = 1.0f; 456 dst->f[3] = 1.0f; 457} 458 459static void 460micro_trunc(union tgsi_exec_channel *dst, 461 const union tgsi_exec_channel *src) 462{ 463 dst->f[0] = (float)(int)src->f[0]; 464 dst->f[1] = (float)(int)src->f[1]; 465 dst->f[2] = (float)(int)src->f[2]; 466 dst->f[3] = (float)(int)src->f[3]; 467} 468 469 470#define CHAN_X 0 471#define CHAN_Y 1 472#define CHAN_Z 2 473#define CHAN_W 3 474 475enum tgsi_exec_datatype { 476 TGSI_EXEC_DATA_FLOAT, 477 TGSI_EXEC_DATA_INT, 478 TGSI_EXEC_DATA_UINT 479}; 480 481/* 482 * Shorthand locations of various utility registers (_I = Index, _C = Channel) 483 */ 484#define TEMP_KILMASK_I TGSI_EXEC_TEMP_KILMASK_I 485#define TEMP_KILMASK_C TGSI_EXEC_TEMP_KILMASK_C 486#define TEMP_OUTPUT_I TGSI_EXEC_TEMP_OUTPUT_I 487#define TEMP_OUTPUT_C TGSI_EXEC_TEMP_OUTPUT_C 488#define TEMP_PRIMITIVE_I TGSI_EXEC_TEMP_PRIMITIVE_I 489#define TEMP_PRIMITIVE_C TGSI_EXEC_TEMP_PRIMITIVE_C 490 491 492/** The execution mask depends on the conditional mask and the loop mask */ 493#define UPDATE_EXEC_MASK(MACH) \ 494 MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->Switch.mask & MACH->FuncMask 495 496 497static const union tgsi_exec_channel ZeroVec = 498 { { 0.0, 0.0, 0.0, 0.0 } }; 499 500static const union tgsi_exec_channel OneVec = { 501 {1.0f, 1.0f, 1.0f, 1.0f} 502}; 503 504static const union tgsi_exec_channel P128Vec = { 505 {128.0f, 128.0f, 128.0f, 128.0f} 506}; 507 508static const union tgsi_exec_channel M128Vec = { 509 {-128.0f, -128.0f, -128.0f, -128.0f} 510}; 511 512 513/** 514 * Assert that none of the float values in 'chan' are infinite or NaN. 515 * NaN and Inf may occur normally during program execution and should 516 * not lead to crashes, etc. But when debugging, it's helpful to catch 517 * them. 518 */ 519static INLINE void 520check_inf_or_nan(const union tgsi_exec_channel *chan) 521{ 522 assert(!util_is_inf_or_nan((chan)->f[0])); 523 assert(!util_is_inf_or_nan((chan)->f[1])); 524 assert(!util_is_inf_or_nan((chan)->f[2])); 525 assert(!util_is_inf_or_nan((chan)->f[3])); 526} 527 528 529#ifdef DEBUG 530static void 531print_chan(const char *msg, const union tgsi_exec_channel *chan) 532{ 533 debug_printf("%s = {%f, %f, %f, %f}\n", 534 msg, chan->f[0], chan->f[1], chan->f[2], chan->f[3]); 535} 536#endif 537 538 539#ifdef DEBUG 540static void 541print_temp(const struct tgsi_exec_machine *mach, uint index) 542{ 543 const struct tgsi_exec_vector *tmp = &mach->Temps[index]; 544 int i; 545 debug_printf("Temp[%u] =\n", index); 546 for (i = 0; i < 4; i++) { 547 debug_printf(" %c: { %f, %f, %f, %f }\n", 548 "XYZW"[i], 549 tmp->xyzw[i].f[0], 550 tmp->xyzw[i].f[1], 551 tmp->xyzw[i].f[2], 552 tmp->xyzw[i].f[3]); 553 } 554} 555#endif 556 557 558void 559tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach, 560 unsigned num_bufs, 561 const void **bufs, 562 const unsigned *buf_sizes) 563{ 564 unsigned i; 565 566 for (i = 0; i < num_bufs; i++) { 567 mach->Consts[i] = bufs[i]; 568 mach->ConstsSize[i] = buf_sizes[i]; 569 } 570} 571 572 573/** 574 * Check if there's a potential src/dst register data dependency when 575 * using SOA execution. 576 * Example: 577 * MOV T, T.yxwz; 578 * This would expand into: 579 * MOV t0, t1; 580 * MOV t1, t0; 581 * MOV t2, t3; 582 * MOV t3, t2; 583 * The second instruction will have the wrong value for t0 if executed as-is. 584 */ 585boolean 586tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst) 587{ 588 uint i, chan; 589 590 uint writemask = inst->Dst[0].Register.WriteMask; 591 if (writemask == TGSI_WRITEMASK_X || 592 writemask == TGSI_WRITEMASK_Y || 593 writemask == TGSI_WRITEMASK_Z || 594 writemask == TGSI_WRITEMASK_W || 595 writemask == TGSI_WRITEMASK_NONE) { 596 /* no chance of data dependency */ 597 return FALSE; 598 } 599 600 /* loop over src regs */ 601 for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { 602 if ((inst->Src[i].Register.File == 603 inst->Dst[0].Register.File) && 604 ((inst->Src[i].Register.Index == 605 inst->Dst[0].Register.Index) || 606 inst->Src[i].Register.Indirect || 607 inst->Dst[0].Register.Indirect)) { 608 /* loop over dest channels */ 609 uint channelsWritten = 0x0; 610 for (chan = 0; chan < NUM_CHANNELS; chan++) { 611 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 612 /* check if we're reading a channel that's been written */ 613 uint swizzle = tgsi_util_get_full_src_register_swizzle(&inst->Src[i], chan); 614 if (channelsWritten & (1 << swizzle)) { 615 return TRUE; 616 } 617 618 channelsWritten |= (1 << chan); 619 } 620 } 621 } 622 } 623 return FALSE; 624} 625 626 627/** 628 * Initialize machine state by expanding tokens to full instructions, 629 * allocating temporary storage, setting up constants, etc. 630 * After this, we can call tgsi_exec_machine_run() many times. 631 */ 632void 633tgsi_exec_machine_bind_shader( 634 struct tgsi_exec_machine *mach, 635 const struct tgsi_token *tokens, 636 uint numSamplers, 637 struct tgsi_sampler **samplers) 638{ 639 uint k; 640 struct tgsi_parse_context parse; 641 struct tgsi_full_instruction *instructions; 642 struct tgsi_full_declaration *declarations; 643 uint maxInstructions = 10, numInstructions = 0; 644 uint maxDeclarations = 10, numDeclarations = 0; 645 646#if 0 647 tgsi_dump(tokens, 0); 648#endif 649 650 util_init_math(); 651 652 if (numSamplers) { 653 assert(samplers); 654 } 655 656 mach->Tokens = tokens; 657 mach->Samplers = samplers; 658 659 if (!tokens) { 660 /* unbind and free all */ 661 if (mach->Declarations) { 662 FREE( mach->Declarations ); 663 } 664 mach->Declarations = NULL; 665 mach->NumDeclarations = 0; 666 667 if (mach->Instructions) { 668 FREE( mach->Instructions ); 669 } 670 mach->Instructions = NULL; 671 mach->NumInstructions = 0; 672 673 return; 674 } 675 676 k = tgsi_parse_init (&parse, mach->Tokens); 677 if (k != TGSI_PARSE_OK) { 678 debug_printf( "Problem parsing!\n" ); 679 return; 680 } 681 682 mach->Processor = parse.FullHeader.Processor.Processor; 683 mach->ImmLimit = 0; 684 685 if (mach->Processor == TGSI_PROCESSOR_GEOMETRY && 686 !mach->UsedGeometryShader) { 687 struct tgsi_exec_vector *inputs; 688 struct tgsi_exec_vector *outputs; 689 690 inputs = align_malloc(sizeof(struct tgsi_exec_vector) * 691 TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS, 692 16); 693 694 if (!inputs) 695 return; 696 697 outputs = align_malloc(sizeof(struct tgsi_exec_vector) * 698 TGSI_MAX_TOTAL_VERTICES, 16); 699 700 if (!outputs) { 701 align_free(inputs); 702 return; 703 } 704 705 align_free(mach->Inputs); 706 align_free(mach->Outputs); 707 708 mach->Inputs = inputs; 709 mach->Outputs = outputs; 710 mach->UsedGeometryShader = TRUE; 711 } 712 713 declarations = (struct tgsi_full_declaration *) 714 MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) ); 715 716 if (!declarations) { 717 return; 718 } 719 720 instructions = (struct tgsi_full_instruction *) 721 MALLOC( maxInstructions * sizeof(struct tgsi_full_instruction) ); 722 723 if (!instructions) { 724 FREE( declarations ); 725 return; 726 } 727 728 while( !tgsi_parse_end_of_tokens( &parse ) ) { 729 uint i; 730 731 tgsi_parse_token( &parse ); 732 switch( parse.FullToken.Token.Type ) { 733 case TGSI_TOKEN_TYPE_DECLARATION: 734 /* save expanded declaration */ 735 if (numDeclarations == maxDeclarations) { 736 declarations = REALLOC(declarations, 737 maxDeclarations 738 * sizeof(struct tgsi_full_declaration), 739 (maxDeclarations + 10) 740 * sizeof(struct tgsi_full_declaration)); 741 maxDeclarations += 10; 742 } 743 if (parse.FullToken.FullDeclaration.Declaration.File == TGSI_FILE_OUTPUT) { 744 unsigned reg; 745 for (reg = parse.FullToken.FullDeclaration.Range.First; 746 reg <= parse.FullToken.FullDeclaration.Range.Last; 747 ++reg) { 748 ++mach->NumOutputs; 749 } 750 } 751 if (parse.FullToken.FullDeclaration.Declaration.File == 752 TGSI_FILE_IMMEDIATE_ARRAY) { 753 unsigned reg; 754 struct tgsi_full_declaration *decl = 755 &parse.FullToken.FullDeclaration; 756 debug_assert(decl->Range.Last < TGSI_EXEC_NUM_IMMEDIATES); 757 for (reg = decl->Range.First; reg <= decl->Range.Last; ++reg) { 758 for( i = 0; i < 4; i++ ) { 759 int idx = reg * 4 + i; 760 mach->ImmArray[reg][i] = decl->ImmediateData.u[idx].Float; 761 } 762 } 763 } 764 memcpy(declarations + numDeclarations, 765 &parse.FullToken.FullDeclaration, 766 sizeof(declarations[0])); 767 numDeclarations++; 768 break; 769 770 case TGSI_TOKEN_TYPE_IMMEDIATE: 771 { 772 uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1; 773 assert( size <= 4 ); 774 assert( mach->ImmLimit + 1 <= TGSI_EXEC_NUM_IMMEDIATES ); 775 776 for( i = 0; i < size; i++ ) { 777 mach->Imms[mach->ImmLimit][i] = 778 parse.FullToken.FullImmediate.u[i].Float; 779 } 780 mach->ImmLimit += 1; 781 } 782 break; 783 784 case TGSI_TOKEN_TYPE_INSTRUCTION: 785 786 /* save expanded instruction */ 787 if (numInstructions == maxInstructions) { 788 instructions = REALLOC(instructions, 789 maxInstructions 790 * sizeof(struct tgsi_full_instruction), 791 (maxInstructions + 10) 792 * sizeof(struct tgsi_full_instruction)); 793 maxInstructions += 10; 794 } 795 796 memcpy(instructions + numInstructions, 797 &parse.FullToken.FullInstruction, 798 sizeof(instructions[0])); 799 800 numInstructions++; 801 break; 802 803 case TGSI_TOKEN_TYPE_PROPERTY: 804 break; 805 806 default: 807 assert( 0 ); 808 } 809 } 810 tgsi_parse_free (&parse); 811 812 if (mach->Declarations) { 813 FREE( mach->Declarations ); 814 } 815 mach->Declarations = declarations; 816 mach->NumDeclarations = numDeclarations; 817 818 if (mach->Instructions) { 819 FREE( mach->Instructions ); 820 } 821 mach->Instructions = instructions; 822 mach->NumInstructions = numInstructions; 823} 824 825 826struct tgsi_exec_machine * 827tgsi_exec_machine_create( void ) 828{ 829 struct tgsi_exec_machine *mach; 830 uint i; 831 832 mach = align_malloc( sizeof *mach, 16 ); 833 if (!mach) 834 goto fail; 835 836 memset(mach, 0, sizeof(*mach)); 837 838 mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR]; 839 mach->MaxGeometryShaderOutputs = TGSI_MAX_TOTAL_VERTICES; 840 mach->Predicates = &mach->Temps[TGSI_EXEC_TEMP_P0]; 841 842 mach->Inputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16); 843 mach->Outputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16); 844 if (!mach->Inputs || !mach->Outputs) 845 goto fail; 846 847 /* Setup constants needed by the SSE2 executor. */ 848 for( i = 0; i < 4; i++ ) { 849 mach->Temps[TGSI_EXEC_TEMP_00000000_I].xyzw[TGSI_EXEC_TEMP_00000000_C].u[i] = 0x00000000; 850 mach->Temps[TGSI_EXEC_TEMP_7FFFFFFF_I].xyzw[TGSI_EXEC_TEMP_7FFFFFFF_C].u[i] = 0x7FFFFFFF; 851 mach->Temps[TGSI_EXEC_TEMP_80000000_I].xyzw[TGSI_EXEC_TEMP_80000000_C].u[i] = 0x80000000; 852 mach->Temps[TGSI_EXEC_TEMP_FFFFFFFF_I].xyzw[TGSI_EXEC_TEMP_FFFFFFFF_C].u[i] = 0xFFFFFFFF; /* not used */ 853 mach->Temps[TGSI_EXEC_TEMP_ONE_I].xyzw[TGSI_EXEC_TEMP_ONE_C].f[i] = 1.0f; 854 mach->Temps[TGSI_EXEC_TEMP_TWO_I].xyzw[TGSI_EXEC_TEMP_TWO_C].f[i] = 2.0f; /* not used */ 855 mach->Temps[TGSI_EXEC_TEMP_128_I].xyzw[TGSI_EXEC_TEMP_128_C].f[i] = 128.0f; 856 mach->Temps[TGSI_EXEC_TEMP_MINUS_128_I].xyzw[TGSI_EXEC_TEMP_MINUS_128_C].f[i] = -128.0f; 857 mach->Temps[TGSI_EXEC_TEMP_THREE_I].xyzw[TGSI_EXEC_TEMP_THREE_C].f[i] = 3.0f; 858 mach->Temps[TGSI_EXEC_TEMP_HALF_I].xyzw[TGSI_EXEC_TEMP_HALF_C].f[i] = 0.5f; 859 } 860 861#ifdef DEBUG 862 /* silence warnings */ 863 (void) print_chan; 864 (void) print_temp; 865#endif 866 867 return mach; 868 869fail: 870 if (mach) { 871 align_free(mach->Inputs); 872 align_free(mach->Outputs); 873 align_free(mach); 874 } 875 return NULL; 876} 877 878 879void 880tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach) 881{ 882 if (mach) { 883 if (mach->Instructions) 884 FREE(mach->Instructions); 885 if (mach->Declarations) 886 FREE(mach->Declarations); 887 888 align_free(mach->Inputs); 889 align_free(mach->Outputs); 890 891 align_free(mach); 892 } 893} 894 895static void 896micro_add(union tgsi_exec_channel *dst, 897 const union tgsi_exec_channel *src0, 898 const union tgsi_exec_channel *src1) 899{ 900 dst->f[0] = src0->f[0] + src1->f[0]; 901 dst->f[1] = src0->f[1] + src1->f[1]; 902 dst->f[2] = src0->f[2] + src1->f[2]; 903 dst->f[3] = src0->f[3] + src1->f[3]; 904} 905 906static void 907micro_div( 908 union tgsi_exec_channel *dst, 909 const union tgsi_exec_channel *src0, 910 const union tgsi_exec_channel *src1 ) 911{ 912 if (src1->f[0] != 0) { 913 dst->f[0] = src0->f[0] / src1->f[0]; 914 } 915 if (src1->f[1] != 0) { 916 dst->f[1] = src0->f[1] / src1->f[1]; 917 } 918 if (src1->f[2] != 0) { 919 dst->f[2] = src0->f[2] / src1->f[2]; 920 } 921 if (src1->f[3] != 0) { 922 dst->f[3] = src0->f[3] / src1->f[3]; 923 } 924} 925 926static void 927micro_rcc(union tgsi_exec_channel *dst, 928 const union tgsi_exec_channel *src) 929{ 930 uint i; 931 932 for (i = 0; i < 4; i++) { 933 float recip = 1.0f / src->f[i]; 934 935 if (recip > 0.0f) { 936 if (recip > 1.884467e+019f) { 937 dst->f[i] = 1.884467e+019f; 938 } 939 else if (recip < 5.42101e-020f) { 940 dst->f[i] = 5.42101e-020f; 941 } 942 else { 943 dst->f[i] = recip; 944 } 945 } 946 else { 947 if (recip < -1.884467e+019f) { 948 dst->f[i] = -1.884467e+019f; 949 } 950 else if (recip > -5.42101e-020f) { 951 dst->f[i] = -5.42101e-020f; 952 } 953 else { 954 dst->f[i] = recip; 955 } 956 } 957 } 958} 959 960static void 961micro_lt( 962 union tgsi_exec_channel *dst, 963 const union tgsi_exec_channel *src0, 964 const union tgsi_exec_channel *src1, 965 const union tgsi_exec_channel *src2, 966 const union tgsi_exec_channel *src3 ) 967{ 968 dst->f[0] = src0->f[0] < src1->f[0] ? src2->f[0] : src3->f[0]; 969 dst->f[1] = src0->f[1] < src1->f[1] ? src2->f[1] : src3->f[1]; 970 dst->f[2] = src0->f[2] < src1->f[2] ? src2->f[2] : src3->f[2]; 971 dst->f[3] = src0->f[3] < src1->f[3] ? src2->f[3] : src3->f[3]; 972} 973 974static void 975micro_max(union tgsi_exec_channel *dst, 976 const union tgsi_exec_channel *src0, 977 const union tgsi_exec_channel *src1) 978{ 979 dst->f[0] = src0->f[0] > src1->f[0] ? src0->f[0] : src1->f[0]; 980 dst->f[1] = src0->f[1] > src1->f[1] ? src0->f[1] : src1->f[1]; 981 dst->f[2] = src0->f[2] > src1->f[2] ? src0->f[2] : src1->f[2]; 982 dst->f[3] = src0->f[3] > src1->f[3] ? src0->f[3] : src1->f[3]; 983} 984 985static void 986micro_min(union tgsi_exec_channel *dst, 987 const union tgsi_exec_channel *src0, 988 const union tgsi_exec_channel *src1) 989{ 990 dst->f[0] = src0->f[0] < src1->f[0] ? src0->f[0] : src1->f[0]; 991 dst->f[1] = src0->f[1] < src1->f[1] ? src0->f[1] : src1->f[1]; 992 dst->f[2] = src0->f[2] < src1->f[2] ? src0->f[2] : src1->f[2]; 993 dst->f[3] = src0->f[3] < src1->f[3] ? src0->f[3] : src1->f[3]; 994} 995 996static void 997micro_mul(union tgsi_exec_channel *dst, 998 const union tgsi_exec_channel *src0, 999 const union tgsi_exec_channel *src1) 1000{ 1001 dst->f[0] = src0->f[0] * src1->f[0]; 1002 dst->f[1] = src0->f[1] * src1->f[1]; 1003 dst->f[2] = src0->f[2] * src1->f[2]; 1004 dst->f[3] = src0->f[3] * src1->f[3]; 1005} 1006 1007static void 1008micro_neg( 1009 union tgsi_exec_channel *dst, 1010 const union tgsi_exec_channel *src ) 1011{ 1012 dst->f[0] = -src->f[0]; 1013 dst->f[1] = -src->f[1]; 1014 dst->f[2] = -src->f[2]; 1015 dst->f[3] = -src->f[3]; 1016} 1017 1018static void 1019micro_pow( 1020 union tgsi_exec_channel *dst, 1021 const union tgsi_exec_channel *src0, 1022 const union tgsi_exec_channel *src1 ) 1023{ 1024#if FAST_MATH 1025 dst->f[0] = util_fast_pow( src0->f[0], src1->f[0] ); 1026 dst->f[1] = util_fast_pow( src0->f[1], src1->f[1] ); 1027 dst->f[2] = util_fast_pow( src0->f[2], src1->f[2] ); 1028 dst->f[3] = util_fast_pow( src0->f[3], src1->f[3] ); 1029#else 1030 dst->f[0] = powf( src0->f[0], src1->f[0] ); 1031 dst->f[1] = powf( src0->f[1], src1->f[1] ); 1032 dst->f[2] = powf( src0->f[2], src1->f[2] ); 1033 dst->f[3] = powf( src0->f[3], src1->f[3] ); 1034#endif 1035} 1036 1037static void 1038micro_sub(union tgsi_exec_channel *dst, 1039 const union tgsi_exec_channel *src0, 1040 const union tgsi_exec_channel *src1) 1041{ 1042 dst->f[0] = src0->f[0] - src1->f[0]; 1043 dst->f[1] = src0->f[1] - src1->f[1]; 1044 dst->f[2] = src0->f[2] - src1->f[2]; 1045 dst->f[3] = src0->f[3] - src1->f[3]; 1046} 1047 1048static void 1049fetch_src_file_channel(const struct tgsi_exec_machine *mach, 1050 const uint chan_index, 1051 const uint file, 1052 const uint swizzle, 1053 const union tgsi_exec_channel *index, 1054 const union tgsi_exec_channel *index2D, 1055 union tgsi_exec_channel *chan) 1056{ 1057 uint i; 1058 1059 assert(swizzle < 4); 1060 1061 switch (file) { 1062 case TGSI_FILE_CONSTANT: 1063 for (i = 0; i < QUAD_SIZE; i++) { 1064 assert(index2D->i[i] >= 0 && index2D->i[i] < PIPE_MAX_CONSTANT_BUFFERS); 1065 assert(mach->Consts[index2D->i[i]]); 1066 1067 if (index->i[i] < 0) { 1068 chan->u[i] = 0; 1069 } else { 1070 /* NOTE: copying the const value as a uint instead of float */ 1071 const uint constbuf = index2D->i[i]; 1072 const uint *buf = (const uint *)mach->Consts[constbuf]; 1073 const int pos = index->i[i] * 4 + swizzle; 1074 /* const buffer bounds check */ 1075 if (pos < 0 || pos >= mach->ConstsSize[constbuf]) { 1076 if (0) { 1077 /* Debug: print warning */ 1078 static int count = 0; 1079 if (count++ < 100) 1080 debug_printf("TGSI Exec: const buffer index %d" 1081 " out of bounds\n", pos); 1082 } 1083 chan->u[i] = 0; 1084 } 1085 else 1086 chan->u[i] = buf[pos]; 1087 } 1088 } 1089 break; 1090 1091 case TGSI_FILE_INPUT: 1092 for (i = 0; i < QUAD_SIZE; i++) { 1093 /* 1094 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1095 debug_printf("Fetching Input[%d] (2d=%d, 1d=%d)\n", 1096 index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i], 1097 index2D->i[i], index->i[i]); 1098 }*/ 1099 int pos = index2D->i[i] * TGSI_EXEC_MAX_INPUT_ATTRIBS + index->i[i]; 1100 assert(pos >= 0); 1101 assert(pos < TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS); 1102 chan->u[i] = mach->Inputs[pos].xyzw[swizzle].u[i]; 1103 } 1104 break; 1105 1106 case TGSI_FILE_SYSTEM_VALUE: 1107 /* XXX no swizzling at this point. Will be needed if we put 1108 * gl_FragCoord, for example, in a sys value register. 1109 */ 1110 for (i = 0; i < QUAD_SIZE; i++) { 1111 chan->u[i] = mach->SystemValue[index->i[i]].u[i]; 1112 } 1113 break; 1114 1115 case TGSI_FILE_TEMPORARY: 1116 for (i = 0; i < QUAD_SIZE; i++) { 1117 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); 1118 assert(index2D->i[i] == 0); 1119 1120 chan->u[i] = mach->Temps[index->i[i]].xyzw[swizzle].u[i]; 1121 } 1122 break; 1123 1124 case TGSI_FILE_TEMPORARY_ARRAY: 1125 for (i = 0; i < QUAD_SIZE; i++) { 1126 assert(index->i[i] < TGSI_EXEC_NUM_TEMPS); 1127 assert(index2D->i[i] < TGSI_EXEC_NUM_TEMP_ARRAYS); 1128 1129 chan->u[i] = 1130 mach->TempArray[index2D->i[i]][index->i[i]].xyzw[swizzle].u[i]; 1131 } 1132 break; 1133 1134 case TGSI_FILE_IMMEDIATE: 1135 for (i = 0; i < QUAD_SIZE; i++) { 1136 assert(index->i[i] >= 0 && index->i[i] < (int)mach->ImmLimit); 1137 assert(index2D->i[i] == 0); 1138 1139 chan->f[i] = mach->Imms[index->i[i]][swizzle]; 1140 } 1141 break; 1142 1143 case TGSI_FILE_IMMEDIATE_ARRAY: 1144 for (i = 0; i < QUAD_SIZE; i++) { 1145 assert(index2D->i[i] == 0); 1146 1147 chan->f[i] = mach->ImmArray[index->i[i]][swizzle]; 1148 } 1149 break; 1150 1151 case TGSI_FILE_ADDRESS: 1152 for (i = 0; i < QUAD_SIZE; i++) { 1153 assert(index->i[i] >= 0); 1154 assert(index2D->i[i] == 0); 1155 1156 chan->u[i] = mach->Addrs[index->i[i]].xyzw[swizzle].u[i]; 1157 } 1158 break; 1159 1160 case TGSI_FILE_PREDICATE: 1161 for (i = 0; i < QUAD_SIZE; i++) { 1162 assert(index->i[i] >= 0 && index->i[i] < TGSI_EXEC_NUM_PREDS); 1163 assert(index2D->i[i] == 0); 1164 1165 chan->u[i] = mach->Predicates[0].xyzw[swizzle].u[i]; 1166 } 1167 break; 1168 1169 case TGSI_FILE_OUTPUT: 1170 /* vertex/fragment output vars can be read too */ 1171 for (i = 0; i < QUAD_SIZE; i++) { 1172 assert(index->i[i] >= 0); 1173 assert(index2D->i[i] == 0); 1174 1175 chan->u[i] = mach->Outputs[index->i[i]].xyzw[swizzle].u[i]; 1176 } 1177 break; 1178 1179 default: 1180 assert(0); 1181 for (i = 0; i < QUAD_SIZE; i++) { 1182 chan->u[i] = 0; 1183 } 1184 } 1185} 1186 1187static void 1188fetch_source(const struct tgsi_exec_machine *mach, 1189 union tgsi_exec_channel *chan, 1190 const struct tgsi_full_src_register *reg, 1191 const uint chan_index, 1192 enum tgsi_exec_datatype src_datatype) 1193{ 1194 union tgsi_exec_channel index; 1195 union tgsi_exec_channel index2D; 1196 uint swizzle; 1197 1198 /* We start with a direct index into a register file. 1199 * 1200 * file[1], 1201 * where: 1202 * file = Register.File 1203 * [1] = Register.Index 1204 */ 1205 index.i[0] = 1206 index.i[1] = 1207 index.i[2] = 1208 index.i[3] = reg->Register.Index; 1209 1210 /* There is an extra source register that indirectly subscripts 1211 * a register file. The direct index now becomes an offset 1212 * that is being added to the indirect register. 1213 * 1214 * file[ind[2].x+1], 1215 * where: 1216 * ind = Indirect.File 1217 * [2] = Indirect.Index 1218 * .x = Indirect.SwizzleX 1219 */ 1220 if (reg->Register.Indirect) { 1221 union tgsi_exec_channel index2; 1222 union tgsi_exec_channel indir_index; 1223 const uint execmask = mach->ExecMask; 1224 uint i; 1225 1226 /* which address register (always zero now) */ 1227 index2.i[0] = 1228 index2.i[1] = 1229 index2.i[2] = 1230 index2.i[3] = reg->Indirect.Index; 1231 assert(reg->Indirect.File == TGSI_FILE_ADDRESS); 1232 /* get current value of address register[swizzle] */ 1233 swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, CHAN_X ); 1234 fetch_src_file_channel(mach, 1235 chan_index, 1236 reg->Indirect.File, 1237 swizzle, 1238 &index2, 1239 &ZeroVec, 1240 &indir_index); 1241 1242 /* add value of address register to the offset */ 1243 index.i[0] += indir_index.i[0]; 1244 index.i[1] += indir_index.i[1]; 1245 index.i[2] += indir_index.i[2]; 1246 index.i[3] += indir_index.i[3]; 1247 1248 /* for disabled execution channels, zero-out the index to 1249 * avoid using a potential garbage value. 1250 */ 1251 for (i = 0; i < QUAD_SIZE; i++) { 1252 if ((execmask & (1 << i)) == 0) 1253 index.i[i] = 0; 1254 } 1255 } 1256 1257 /* There is an extra source register that is a second 1258 * subscript to a register file. Effectively it means that 1259 * the register file is actually a 2D array of registers. 1260 * 1261 * file[3][1], 1262 * where: 1263 * [3] = Dimension.Index 1264 */ 1265 if (reg->Register.Dimension) { 1266 index2D.i[0] = 1267 index2D.i[1] = 1268 index2D.i[2] = 1269 index2D.i[3] = reg->Dimension.Index; 1270 1271 /* Again, the second subscript index can be addressed indirectly 1272 * identically to the first one. 1273 * Nothing stops us from indirectly addressing the indirect register, 1274 * but there is no need for that, so we won't exercise it. 1275 * 1276 * file[ind[4].y+3][1], 1277 * where: 1278 * ind = DimIndirect.File 1279 * [4] = DimIndirect.Index 1280 * .y = DimIndirect.SwizzleX 1281 */ 1282 if (reg->Dimension.Indirect) { 1283 union tgsi_exec_channel index2; 1284 union tgsi_exec_channel indir_index; 1285 const uint execmask = mach->ExecMask; 1286 uint i; 1287 1288 index2.i[0] = 1289 index2.i[1] = 1290 index2.i[2] = 1291 index2.i[3] = reg->DimIndirect.Index; 1292 1293 swizzle = tgsi_util_get_src_register_swizzle( ®->DimIndirect, CHAN_X ); 1294 fetch_src_file_channel(mach, 1295 chan_index, 1296 reg->DimIndirect.File, 1297 swizzle, 1298 &index2, 1299 &ZeroVec, 1300 &indir_index); 1301 1302 index2D.i[0] += indir_index.i[0]; 1303 index2D.i[1] += indir_index.i[1]; 1304 index2D.i[2] += indir_index.i[2]; 1305 index2D.i[3] += indir_index.i[3]; 1306 1307 /* for disabled execution channels, zero-out the index to 1308 * avoid using a potential garbage value. 1309 */ 1310 for (i = 0; i < QUAD_SIZE; i++) { 1311 if ((execmask & (1 << i)) == 0) { 1312 index2D.i[i] = 0; 1313 } 1314 } 1315 } 1316 1317 /* If by any chance there was a need for a 3D array of register 1318 * files, we would have to check whether Dimension is followed 1319 * by a dimension register and continue the saga. 1320 */ 1321 } else { 1322 index2D.i[0] = 1323 index2D.i[1] = 1324 index2D.i[2] = 1325 index2D.i[3] = 0; 1326 } 1327 1328 swizzle = tgsi_util_get_full_src_register_swizzle( reg, chan_index ); 1329 fetch_src_file_channel(mach, 1330 chan_index, 1331 reg->Register.File, 1332 swizzle, 1333 &index, 1334 &index2D, 1335 chan); 1336 1337 if (reg->Register.Absolute) { 1338 if (src_datatype == TGSI_EXEC_DATA_FLOAT) { 1339 micro_abs(chan, chan); 1340 } else { 1341 micro_iabs(chan, chan); 1342 } 1343 } 1344 1345 if (reg->Register.Negate) { 1346 if (src_datatype == TGSI_EXEC_DATA_FLOAT) { 1347 micro_neg(chan, chan); 1348 } else { 1349 micro_ineg(chan, chan); 1350 } 1351 } 1352} 1353 1354static void 1355store_dest(struct tgsi_exec_machine *mach, 1356 const union tgsi_exec_channel *chan, 1357 const struct tgsi_full_dst_register *reg, 1358 const struct tgsi_full_instruction *inst, 1359 uint chan_index, 1360 enum tgsi_exec_datatype dst_datatype) 1361{ 1362 uint i; 1363 union tgsi_exec_channel null; 1364 union tgsi_exec_channel *dst; 1365 union tgsi_exec_channel index2D; 1366 uint execmask = mach->ExecMask; 1367 int offset = 0; /* indirection offset */ 1368 int index; 1369 1370 /* for debugging */ 1371 if (0 && dst_datatype == TGSI_EXEC_DATA_FLOAT) { 1372 check_inf_or_nan(chan); 1373 } 1374 1375 /* There is an extra source register that indirectly subscripts 1376 * a register file. The direct index now becomes an offset 1377 * that is being added to the indirect register. 1378 * 1379 * file[ind[2].x+1], 1380 * where: 1381 * ind = Indirect.File 1382 * [2] = Indirect.Index 1383 * .x = Indirect.SwizzleX 1384 */ 1385 if (reg->Register.Indirect) { 1386 union tgsi_exec_channel index; 1387 union tgsi_exec_channel indir_index; 1388 uint swizzle; 1389 1390 /* which address register (always zero for now) */ 1391 index.i[0] = 1392 index.i[1] = 1393 index.i[2] = 1394 index.i[3] = reg->Indirect.Index; 1395 1396 /* get current value of address register[swizzle] */ 1397 swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, CHAN_X ); 1398 1399 /* fetch values from the address/indirection register */ 1400 fetch_src_file_channel(mach, 1401 chan_index, 1402 reg->Indirect.File, 1403 swizzle, 1404 &index, 1405 &ZeroVec, 1406 &indir_index); 1407 1408 /* save indirection offset */ 1409 offset = indir_index.i[0]; 1410 } 1411 1412 /* There is an extra source register that is a second 1413 * subscript to a register file. Effectively it means that 1414 * the register file is actually a 2D array of registers. 1415 * 1416 * file[3][1], 1417 * where: 1418 * [3] = Dimension.Index 1419 */ 1420 if (reg->Register.Dimension) { 1421 index2D.i[0] = 1422 index2D.i[1] = 1423 index2D.i[2] = 1424 index2D.i[3] = reg->Dimension.Index; 1425 1426 /* Again, the second subscript index can be addressed indirectly 1427 * identically to the first one. 1428 * Nothing stops us from indirectly addressing the indirect register, 1429 * but there is no need for that, so we won't exercise it. 1430 * 1431 * file[ind[4].y+3][1], 1432 * where: 1433 * ind = DimIndirect.File 1434 * [4] = DimIndirect.Index 1435 * .y = DimIndirect.SwizzleX 1436 */ 1437 if (reg->Dimension.Indirect) { 1438 union tgsi_exec_channel index2; 1439 union tgsi_exec_channel indir_index; 1440 const uint execmask = mach->ExecMask; 1441 unsigned swizzle; 1442 uint i; 1443 1444 index2.i[0] = 1445 index2.i[1] = 1446 index2.i[2] = 1447 index2.i[3] = reg->DimIndirect.Index; 1448 1449 swizzle = tgsi_util_get_src_register_swizzle( ®->DimIndirect, CHAN_X ); 1450 fetch_src_file_channel(mach, 1451 chan_index, 1452 reg->DimIndirect.File, 1453 swizzle, 1454 &index2, 1455 &ZeroVec, 1456 &indir_index); 1457 1458 index2D.i[0] += indir_index.i[0]; 1459 index2D.i[1] += indir_index.i[1]; 1460 index2D.i[2] += indir_index.i[2]; 1461 index2D.i[3] += indir_index.i[3]; 1462 1463 /* for disabled execution channels, zero-out the index to 1464 * avoid using a potential garbage value. 1465 */ 1466 for (i = 0; i < QUAD_SIZE; i++) { 1467 if ((execmask & (1 << i)) == 0) { 1468 index2D.i[i] = 0; 1469 } 1470 } 1471 } 1472 1473 /* If by any chance there was a need for a 3D array of register 1474 * files, we would have to check whether Dimension is followed 1475 * by a dimension register and continue the saga. 1476 */ 1477 } else { 1478 index2D.i[0] = 1479 index2D.i[1] = 1480 index2D.i[2] = 1481 index2D.i[3] = 0; 1482 } 1483 1484 switch (reg->Register.File) { 1485 case TGSI_FILE_NULL: 1486 dst = &null; 1487 break; 1488 1489 case TGSI_FILE_OUTPUT: 1490 index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] 1491 + reg->Register.Index; 1492 dst = &mach->Outputs[offset + index].xyzw[chan_index]; 1493#if 0 1494 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1495 fprintf(stderr, "STORING OUT[%d] mask(%d), = (", offset + index, execmask); 1496 for (i = 0; i < QUAD_SIZE; i++) 1497 if (execmask & (1 << i)) 1498 fprintf(stderr, "%f, ", chan->f[i]); 1499 fprintf(stderr, ")\n"); 1500 } 1501#endif 1502 break; 1503 1504 case TGSI_FILE_TEMPORARY: 1505 index = reg->Register.Index; 1506 assert( index < TGSI_EXEC_NUM_TEMPS ); 1507 dst = &mach->Temps[offset + index].xyzw[chan_index]; 1508 break; 1509 1510 case TGSI_FILE_TEMPORARY_ARRAY: 1511 index = reg->Register.Index; 1512 assert( index < TGSI_EXEC_NUM_TEMPS ); 1513 assert( index2D.i[0] < TGSI_EXEC_NUM_TEMP_ARRAYS ); 1514 /* XXX we use index2D.i[0] here but somehow we might 1515 * end up with someone trying to store indirectly in 1516 * different buffers */ 1517 dst = &mach->TempArray[index2D.i[0]][offset + index].xyzw[chan_index]; 1518 break; 1519 1520 case TGSI_FILE_ADDRESS: 1521 index = reg->Register.Index; 1522 dst = &mach->Addrs[index].xyzw[chan_index]; 1523 break; 1524 1525 case TGSI_FILE_PREDICATE: 1526 index = reg->Register.Index; 1527 assert(index < TGSI_EXEC_NUM_PREDS); 1528 dst = &mach->Predicates[index].xyzw[chan_index]; 1529 break; 1530 1531 default: 1532 assert( 0 ); 1533 return; 1534 } 1535 1536 if (inst->Instruction.Predicate) { 1537 uint swizzle; 1538 union tgsi_exec_channel *pred; 1539 1540 switch (chan_index) { 1541 case CHAN_X: 1542 swizzle = inst->Predicate.SwizzleX; 1543 break; 1544 case CHAN_Y: 1545 swizzle = inst->Predicate.SwizzleY; 1546 break; 1547 case CHAN_Z: 1548 swizzle = inst->Predicate.SwizzleZ; 1549 break; 1550 case CHAN_W: 1551 swizzle = inst->Predicate.SwizzleW; 1552 break; 1553 default: 1554 assert(0); 1555 return; 1556 } 1557 1558 assert(inst->Predicate.Index == 0); 1559 1560 pred = &mach->Predicates[inst->Predicate.Index].xyzw[swizzle]; 1561 1562 if (inst->Predicate.Negate) { 1563 for (i = 0; i < QUAD_SIZE; i++) { 1564 if (pred->u[i]) { 1565 execmask &= ~(1 << i); 1566 } 1567 } 1568 } else { 1569 for (i = 0; i < QUAD_SIZE; i++) { 1570 if (!pred->u[i]) { 1571 execmask &= ~(1 << i); 1572 } 1573 } 1574 } 1575 } 1576 1577 switch (inst->Instruction.Saturate) { 1578 case TGSI_SAT_NONE: 1579 for (i = 0; i < QUAD_SIZE; i++) 1580 if (execmask & (1 << i)) 1581 dst->i[i] = chan->i[i]; 1582 break; 1583 1584 case TGSI_SAT_ZERO_ONE: 1585 for (i = 0; i < QUAD_SIZE; i++) 1586 if (execmask & (1 << i)) { 1587 if (chan->f[i] < 0.0f) 1588 dst->f[i] = 0.0f; 1589 else if (chan->f[i] > 1.0f) 1590 dst->f[i] = 1.0f; 1591 else 1592 dst->i[i] = chan->i[i]; 1593 } 1594 break; 1595 1596 case TGSI_SAT_MINUS_PLUS_ONE: 1597 for (i = 0; i < QUAD_SIZE; i++) 1598 if (execmask & (1 << i)) { 1599 if (chan->f[i] < -1.0f) 1600 dst->f[i] = -1.0f; 1601 else if (chan->f[i] > 1.0f) 1602 dst->f[i] = 1.0f; 1603 else 1604 dst->i[i] = chan->i[i]; 1605 } 1606 break; 1607 1608 default: 1609 assert( 0 ); 1610 } 1611} 1612 1613#define FETCH(VAL,INDEX,CHAN)\ 1614 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_FLOAT) 1615 1616#define IFETCH(VAL,INDEX,CHAN)\ 1617 fetch_source(mach, VAL, &inst->Src[INDEX], CHAN, TGSI_EXEC_DATA_INT) 1618 1619 1620/** 1621 * Execute ARB-style KIL which is predicated by a src register. 1622 * Kill fragment if any of the four values is less than zero. 1623 */ 1624static void 1625exec_kil(struct tgsi_exec_machine *mach, 1626 const struct tgsi_full_instruction *inst) 1627{ 1628 uint uniquemask; 1629 uint chan_index; 1630 uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ 1631 union tgsi_exec_channel r[1]; 1632 1633 /* This mask stores component bits that were already tested. */ 1634 uniquemask = 0; 1635 1636 for (chan_index = 0; chan_index < 4; chan_index++) 1637 { 1638 uint swizzle; 1639 uint i; 1640 1641 /* unswizzle channel */ 1642 swizzle = tgsi_util_get_full_src_register_swizzle ( 1643 &inst->Src[0], 1644 chan_index); 1645 1646 /* check if the component has not been already tested */ 1647 if (uniquemask & (1 << swizzle)) 1648 continue; 1649 uniquemask |= 1 << swizzle; 1650 1651 FETCH(&r[0], 0, chan_index); 1652 for (i = 0; i < 4; i++) 1653 if (r[0].f[i] < 0.0f) 1654 kilmask |= 1 << i; 1655 } 1656 1657 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; 1658} 1659 1660/** 1661 * Execute NVIDIA-style KIL which is predicated by a condition code. 1662 * Kill fragment if the condition code is TRUE. 1663 */ 1664static void 1665exec_kilp(struct tgsi_exec_machine *mach, 1666 const struct tgsi_full_instruction *inst) 1667{ 1668 uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */ 1669 1670 /* "unconditional" kil */ 1671 kilmask = mach->ExecMask; 1672 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask; 1673} 1674 1675static void 1676emit_vertex(struct tgsi_exec_machine *mach) 1677{ 1678 /* FIXME: check for exec mask correctly 1679 unsigned i; 1680 for (i = 0; i < QUAD_SIZE; ++i) { 1681 if ((mach->ExecMask & (1 << i))) 1682 */ 1683 if (mach->ExecMask) { 1684 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += mach->NumOutputs; 1685 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++; 1686 } 1687} 1688 1689static void 1690emit_primitive(struct tgsi_exec_machine *mach) 1691{ 1692 unsigned *prim_count = &mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]; 1693 /* FIXME: check for exec mask correctly 1694 unsigned i; 1695 for (i = 0; i < QUAD_SIZE; ++i) { 1696 if ((mach->ExecMask & (1 << i))) 1697 */ 1698 if (mach->ExecMask) { 1699 ++(*prim_count); 1700 debug_assert((*prim_count * mach->NumOutputs) < mach->MaxGeometryShaderOutputs); 1701 mach->Primitives[*prim_count] = 0; 1702 } 1703} 1704 1705static void 1706conditional_emit_primitive(struct tgsi_exec_machine *mach) 1707{ 1708 if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) { 1709 int emitted_verts = 1710 mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]; 1711 if (emitted_verts) { 1712 emit_primitive(mach); 1713 } 1714 } 1715} 1716 1717 1718/* 1719 * Fetch four texture samples using STR texture coordinates. 1720 */ 1721static void 1722fetch_texel( struct tgsi_sampler *sampler, 1723 const union tgsi_exec_channel *s, 1724 const union tgsi_exec_channel *t, 1725 const union tgsi_exec_channel *p, 1726 const union tgsi_exec_channel *c0, 1727 enum tgsi_sampler_control control, 1728 union tgsi_exec_channel *r, 1729 union tgsi_exec_channel *g, 1730 union tgsi_exec_channel *b, 1731 union tgsi_exec_channel *a ) 1732{ 1733 uint j; 1734 float rgba[NUM_CHANNELS][QUAD_SIZE]; 1735 1736 sampler->get_samples(sampler, s->f, t->f, p->f, c0->f, control, rgba); 1737 1738 for (j = 0; j < 4; j++) { 1739 r->f[j] = rgba[0][j]; 1740 g->f[j] = rgba[1][j]; 1741 b->f[j] = rgba[2][j]; 1742 a->f[j] = rgba[3][j]; 1743 } 1744} 1745 1746 1747#define TEX_MODIFIER_NONE 0 1748#define TEX_MODIFIER_PROJECTED 1 1749#define TEX_MODIFIER_LOD_BIAS 2 1750#define TEX_MODIFIER_EXPLICIT_LOD 3 1751 1752 1753static void 1754exec_tex(struct tgsi_exec_machine *mach, 1755 const struct tgsi_full_instruction *inst, 1756 uint modifier) 1757{ 1758 const uint unit = inst->Src[1].Register.Index; 1759 union tgsi_exec_channel r[4]; 1760 const union tgsi_exec_channel *lod = &ZeroVec; 1761 enum tgsi_sampler_control control; 1762 uint chan; 1763 1764 if (modifier != TEX_MODIFIER_NONE) { 1765 FETCH(&r[3], 0, CHAN_W); 1766 if (modifier != TEX_MODIFIER_PROJECTED) { 1767 lod = &r[3]; 1768 } 1769 } 1770 1771 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) { 1772 control = tgsi_sampler_lod_explicit; 1773 } else { 1774 control = tgsi_sampler_lod_bias; 1775 } 1776 1777 switch (inst->Texture.Texture) { 1778 case TGSI_TEXTURE_1D: 1779 FETCH(&r[0], 0, CHAN_X); 1780 1781 if (modifier == TEX_MODIFIER_PROJECTED) { 1782 micro_div(&r[0], &r[0], &r[3]); 1783 } 1784 1785 fetch_texel(mach->Samplers[unit], 1786 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */ 1787 control, 1788 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1789 break; 1790 case TGSI_TEXTURE_SHADOW1D: 1791 FETCH(&r[0], 0, CHAN_X); 1792 FETCH(&r[2], 0, CHAN_Z); 1793 1794 if (modifier == TEX_MODIFIER_PROJECTED) { 1795 micro_div(&r[0], &r[0], &r[3]); 1796 } 1797 1798 fetch_texel(mach->Samplers[unit], 1799 &r[0], &ZeroVec, &r[2], lod, /* S, T, P, LOD */ 1800 control, 1801 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1802 break; 1803 1804 case TGSI_TEXTURE_2D: 1805 case TGSI_TEXTURE_RECT: 1806 case TGSI_TEXTURE_SHADOW2D: 1807 case TGSI_TEXTURE_SHADOWRECT: 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 micro_div(&r[2], &r[2], &r[3]); 1816 } 1817 1818 fetch_texel(mach->Samplers[unit], 1819 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1820 control, 1821 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1822 break; 1823 1824 case TGSI_TEXTURE_1D_ARRAY: 1825 FETCH(&r[0], 0, CHAN_X); 1826 FETCH(&r[1], 0, CHAN_Y); 1827 1828 if (modifier == TEX_MODIFIER_PROJECTED) { 1829 micro_div(&r[0], &r[0], &r[3]); 1830 } 1831 1832 fetch_texel(mach->Samplers[unit], 1833 &r[0], &r[1], &ZeroVec, lod, /* S, T, P, LOD */ 1834 control, 1835 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1836 break; 1837 case TGSI_TEXTURE_SHADOW1D_ARRAY: 1838 FETCH(&r[0], 0, CHAN_X); 1839 FETCH(&r[1], 0, CHAN_Y); 1840 FETCH(&r[2], 0, CHAN_Z); 1841 1842 if (modifier == TEX_MODIFIER_PROJECTED) { 1843 micro_div(&r[0], &r[0], &r[3]); 1844 } 1845 1846 fetch_texel(mach->Samplers[unit], 1847 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1848 control, 1849 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1850 break; 1851 1852 case TGSI_TEXTURE_2D_ARRAY: 1853 FETCH(&r[0], 0, CHAN_X); 1854 FETCH(&r[1], 0, CHAN_Y); 1855 FETCH(&r[2], 0, CHAN_Z); 1856 1857 if (modifier == TEX_MODIFIER_PROJECTED) { 1858 micro_div(&r[0], &r[0], &r[3]); 1859 micro_div(&r[1], &r[1], &r[3]); 1860 } 1861 1862 fetch_texel(mach->Samplers[unit], 1863 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 1864 control, 1865 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1866 break; 1867 case TGSI_TEXTURE_SHADOW2D_ARRAY: 1868 FETCH(&r[0], 0, CHAN_X); 1869 FETCH(&r[1], 0, CHAN_Y); 1870 FETCH(&r[2], 0, CHAN_Z); 1871 FETCH(&r[3], 0, CHAN_W); 1872 1873 fetch_texel(mach->Samplers[unit], 1874 &r[0], &r[1], &r[2], &r[3], /* S, T, P, LOD */ 1875 control, 1876 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1877 break; 1878 case TGSI_TEXTURE_3D: 1879 case TGSI_TEXTURE_CUBE: 1880 FETCH(&r[0], 0, CHAN_X); 1881 FETCH(&r[1], 0, CHAN_Y); 1882 FETCH(&r[2], 0, CHAN_Z); 1883 1884 if (modifier == TEX_MODIFIER_PROJECTED) { 1885 micro_div(&r[0], &r[0], &r[3]); 1886 micro_div(&r[1], &r[1], &r[3]); 1887 micro_div(&r[2], &r[2], &r[3]); 1888 } 1889 1890 fetch_texel(mach->Samplers[unit], 1891 &r[0], &r[1], &r[2], lod, 1892 control, 1893 &r[0], &r[1], &r[2], &r[3]); 1894 break; 1895 1896 default: 1897 assert(0); 1898 } 1899 1900#if 0 1901 debug_printf("fetch r: %g %g %g %g\n", 1902 r[0].f[0], r[0].f[1], r[0].f[2], r[0].f[3]); 1903 debug_printf("fetch g: %g %g %g %g\n", 1904 r[1].f[0], r[1].f[1], r[1].f[2], r[1].f[3]); 1905 debug_printf("fetch b: %g %g %g %g\n", 1906 r[2].f[0], r[2].f[1], r[2].f[2], r[2].f[3]); 1907 debug_printf("fetch a: %g %g %g %g\n", 1908 r[3].f[0], r[3].f[1], r[3].f[2], r[3].f[3]); 1909#endif 1910 1911 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1912 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1913 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1914 } 1915 } 1916} 1917 1918static void 1919exec_txd(struct tgsi_exec_machine *mach, 1920 const struct tgsi_full_instruction *inst) 1921{ 1922 const uint unit = inst->Src[3].Register.Index; 1923 union tgsi_exec_channel r[4]; 1924 uint chan; 1925 1926 /* 1927 * XXX: This is fake TXD -- the derivatives are not taken into account, yet. 1928 */ 1929 1930 switch (inst->Texture.Texture) { 1931 case TGSI_TEXTURE_1D: 1932 case TGSI_TEXTURE_SHADOW1D: 1933 1934 FETCH(&r[0], 0, CHAN_X); 1935 1936 fetch_texel(mach->Samplers[unit], 1937 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */ 1938 tgsi_sampler_lod_bias, 1939 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 1940 break; 1941 1942 case TGSI_TEXTURE_1D_ARRAY: 1943 case TGSI_TEXTURE_2D: 1944 case TGSI_TEXTURE_RECT: 1945 case TGSI_TEXTURE_SHADOW1D_ARRAY: 1946 case TGSI_TEXTURE_SHADOW2D: 1947 case TGSI_TEXTURE_SHADOWRECT: 1948 1949 FETCH(&r[0], 0, CHAN_X); 1950 FETCH(&r[1], 0, CHAN_Y); 1951 FETCH(&r[2], 0, CHAN_Z); 1952 1953 fetch_texel(mach->Samplers[unit], 1954 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */ 1955 tgsi_sampler_lod_bias, 1956 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 1957 break; 1958 1959 case TGSI_TEXTURE_2D_ARRAY: 1960 case TGSI_TEXTURE_3D: 1961 case TGSI_TEXTURE_CUBE: 1962 1963 FETCH(&r[0], 0, CHAN_X); 1964 FETCH(&r[1], 0, CHAN_Y); 1965 FETCH(&r[2], 0, CHAN_Z); 1966 1967 fetch_texel(mach->Samplers[unit], 1968 &r[0], &r[1], &r[2], &ZeroVec, 1969 tgsi_sampler_lod_bias, 1970 &r[0], &r[1], &r[2], &r[3]); 1971 break; 1972 1973 case TGSI_TEXTURE_SHADOW2D_ARRAY: 1974 1975 FETCH(&r[0], 0, CHAN_X); 1976 FETCH(&r[1], 0, CHAN_Y); 1977 FETCH(&r[2], 0, CHAN_Z); 1978 FETCH(&r[3], 0, CHAN_W); 1979 1980 fetch_texel(mach->Samplers[unit], 1981 &r[0], &r[1], &r[2], &r[3], 1982 tgsi_sampler_lod_bias, 1983 &r[0], &r[1], &r[2], &r[3]); 1984 break; 1985 1986 default: 1987 assert(0); 1988 } 1989 1990 for (chan = 0; chan < NUM_CHANNELS; chan++) { 1991 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 1992 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 1993 } 1994 } 1995} 1996 1997 1998static void 1999exec_txf(struct tgsi_exec_machine *mach, 2000 const struct tgsi_full_instruction *inst) 2001{ 2002 struct tgsi_sampler *sampler; 2003 const uint unit = inst->Src[2].Register.Index; 2004 union tgsi_exec_channel r[4]; 2005 union tgsi_exec_channel offset[3]; 2006 uint chan; 2007 float rgba[NUM_CHANNELS][QUAD_SIZE]; 2008 int j; 2009 int8_t offsets[3]; 2010 2011 if (inst->Texture.NumOffsets == 1) { 2012 union tgsi_exec_channel index; 2013 index.i[0] = index.i[1] = index.i[2] = index.i[3] = inst->TexOffsets[0].Index; 2014 fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File, 2015 inst->TexOffsets[0].SwizzleX, &index, &ZeroVec, &offset[0]); 2016 fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File, 2017 inst->TexOffsets[0].SwizzleY, &index, &ZeroVec, &offset[1]); 2018 fetch_src_file_channel(mach, 0, inst->TexOffsets[0].File, 2019 inst->TexOffsets[0].SwizzleZ, &index, &ZeroVec, &offset[2]); 2020 offsets[0] = offset[0].i[0]; 2021 offsets[1] = offset[1].i[0]; 2022 offsets[2] = offset[2].i[0]; 2023 } else 2024 offsets[0] = offsets[1] = offsets[2] = 0; 2025 2026 IFETCH(&r[3], 0, CHAN_W); 2027 2028 switch(inst->Texture.Texture) { 2029 case TGSI_TEXTURE_3D: 2030 case TGSI_TEXTURE_2D_ARRAY: 2031 case TGSI_TEXTURE_SHADOW2D_ARRAY: 2032 IFETCH(&r[2], 0, CHAN_Z); 2033 /* fallthrough */ 2034 case TGSI_TEXTURE_2D: 2035 case TGSI_TEXTURE_RECT: 2036 case TGSI_TEXTURE_SHADOW1D_ARRAY: 2037 case TGSI_TEXTURE_SHADOW2D: 2038 case TGSI_TEXTURE_SHADOWRECT: 2039 case TGSI_TEXTURE_1D_ARRAY: 2040 IFETCH(&r[1], 0, CHAN_Y); 2041 /* fallthrough */ 2042 case TGSI_TEXTURE_1D: 2043 case TGSI_TEXTURE_SHADOW1D: 2044 IFETCH(&r[0], 0, CHAN_X); 2045 break; 2046 default: 2047 assert(0); 2048 break; 2049 } 2050 2051 sampler = mach->Samplers[unit]; 2052 sampler->get_texel(sampler, r[0].i, r[1].i, r[2].i, r[3].i, 2053 offsets, rgba); 2054 2055 for (j = 0; j < QUAD_SIZE; j++) { 2056 r[0].f[j] = rgba[0][j]; 2057 r[1].f[j] = rgba[1][j]; 2058 r[2].f[j] = rgba[2][j]; 2059 r[3].f[j] = rgba[3][j]; 2060 } 2061 2062 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2063 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2064 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2065 } 2066 } 2067} 2068 2069static void 2070exec_txq(struct tgsi_exec_machine *mach, 2071 const struct tgsi_full_instruction *inst) 2072{ 2073 struct tgsi_sampler *sampler; 2074 const uint unit = inst->Src[1].Register.Index; 2075 int result[4]; 2076 union tgsi_exec_channel r[4], src; 2077 uint chan; 2078 int i,j; 2079 2080 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_INT); 2081 sampler = mach->Samplers[unit]; 2082 2083 sampler->get_dims(sampler, src.i[0], result); 2084 2085 for (i = 0; i < QUAD_SIZE; i++) { 2086 for (j = 0; j < 4; j++) { 2087 r[j].i[i] = result[j]; 2088 } 2089 } 2090 2091 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2092 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2093 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, 2094 TGSI_EXEC_DATA_INT); 2095 } 2096 } 2097} 2098 2099static void 2100exec_sample(struct tgsi_exec_machine *mach, 2101 const struct tgsi_full_instruction *inst, 2102 uint modifier) 2103{ 2104 const uint resource_unit = inst->Src[1].Register.Index; 2105 const uint sampler_unit = inst->Src[2].Register.Index; 2106 union tgsi_exec_channel r[4]; 2107 const union tgsi_exec_channel *lod = &ZeroVec; 2108 enum tgsi_sampler_control control; 2109 uint chan; 2110 2111 if (modifier != TEX_MODIFIER_NONE) { 2112 if (modifier == TEX_MODIFIER_LOD_BIAS) 2113 FETCH(&r[3], 3, CHAN_X); 2114 else /*TEX_MODIFIER_LOD*/ 2115 FETCH(&r[3], 0, CHAN_W); 2116 2117 if (modifier != TEX_MODIFIER_PROJECTED) { 2118 lod = &r[3]; 2119 } 2120 } 2121 2122 if (modifier == TEX_MODIFIER_EXPLICIT_LOD) { 2123 control = tgsi_sampler_lod_explicit; 2124 } else { 2125 control = tgsi_sampler_lod_bias; 2126 } 2127 2128 switch (mach->Resources[resource_unit].Resource) { 2129 case TGSI_TEXTURE_1D: 2130 case TGSI_TEXTURE_SHADOW1D: 2131 FETCH(&r[0], 0, CHAN_X); 2132 2133 if (modifier == TEX_MODIFIER_PROJECTED) { 2134 micro_div(&r[0], &r[0], &r[3]); 2135 } 2136 2137 fetch_texel(mach->Samplers[sampler_unit], 2138 &r[0], &ZeroVec, &ZeroVec, lod, /* S, T, P, LOD */ 2139 control, 2140 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 2141 break; 2142 2143 case TGSI_TEXTURE_1D_ARRAY: 2144 case TGSI_TEXTURE_2D: 2145 case TGSI_TEXTURE_RECT: 2146 case TGSI_TEXTURE_SHADOW1D_ARRAY: 2147 case TGSI_TEXTURE_SHADOW2D: 2148 case TGSI_TEXTURE_SHADOWRECT: 2149 FETCH(&r[0], 0, CHAN_X); 2150 FETCH(&r[1], 0, CHAN_Y); 2151 FETCH(&r[2], 0, CHAN_Z); 2152 2153 if (modifier == TEX_MODIFIER_PROJECTED) { 2154 micro_div(&r[0], &r[0], &r[3]); 2155 micro_div(&r[1], &r[1], &r[3]); 2156 micro_div(&r[2], &r[2], &r[3]); 2157 } 2158 2159 fetch_texel(mach->Samplers[sampler_unit], 2160 &r[0], &r[1], &r[2], lod, /* S, T, P, LOD */ 2161 control, 2162 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 2163 break; 2164 2165 case TGSI_TEXTURE_2D_ARRAY: 2166 case TGSI_TEXTURE_3D: 2167 case TGSI_TEXTURE_CUBE: 2168 FETCH(&r[0], 0, CHAN_X); 2169 FETCH(&r[1], 0, CHAN_Y); 2170 FETCH(&r[2], 0, CHAN_Z); 2171 2172 if (modifier == TEX_MODIFIER_PROJECTED) { 2173 micro_div(&r[0], &r[0], &r[3]); 2174 micro_div(&r[1], &r[1], &r[3]); 2175 micro_div(&r[2], &r[2], &r[3]); 2176 } 2177 2178 fetch_texel(mach->Samplers[sampler_unit], 2179 &r[0], &r[1], &r[2], lod, 2180 control, 2181 &r[0], &r[1], &r[2], &r[3]); 2182 break; 2183 2184 case TGSI_TEXTURE_SHADOW2D_ARRAY: 2185 FETCH(&r[0], 0, CHAN_X); 2186 FETCH(&r[1], 0, CHAN_Y); 2187 FETCH(&r[2], 0, CHAN_Z); 2188 FETCH(&r[3], 0, CHAN_W); 2189 2190 assert(modifier != TEX_MODIFIER_PROJECTED); 2191 2192 fetch_texel(mach->Samplers[sampler_unit], 2193 &r[0], &r[1], &r[2], &r[3], 2194 control, 2195 &r[0], &r[1], &r[2], &r[3]); 2196 break; 2197 2198 default: 2199 assert(0); 2200 } 2201 2202 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2203 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2204 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2205 } 2206 } 2207} 2208 2209static void 2210exec_sample_d(struct tgsi_exec_machine *mach, 2211 const struct tgsi_full_instruction *inst) 2212{ 2213 const uint resource_unit = inst->Src[1].Register.Index; 2214 const uint sampler_unit = inst->Src[2].Register.Index; 2215 union tgsi_exec_channel r[4]; 2216 uint chan; 2217 /* 2218 * XXX: This is fake SAMPLE_D -- the derivatives are not taken into account, yet. 2219 */ 2220 2221 switch (mach->Resources[resource_unit].Resource) { 2222 case TGSI_TEXTURE_1D: 2223 case TGSI_TEXTURE_SHADOW1D: 2224 2225 FETCH(&r[0], 0, CHAN_X); 2226 2227 fetch_texel(mach->Samplers[sampler_unit], 2228 &r[0], &ZeroVec, &ZeroVec, &ZeroVec, /* S, T, P, BIAS */ 2229 tgsi_sampler_lod_bias, 2230 &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */ 2231 break; 2232 2233 case TGSI_TEXTURE_2D: 2234 case TGSI_TEXTURE_RECT: 2235 case TGSI_TEXTURE_SHADOW2D: 2236 case TGSI_TEXTURE_SHADOWRECT: 2237 2238 FETCH(&r[0], 0, CHAN_X); 2239 FETCH(&r[1], 0, CHAN_Y); 2240 FETCH(&r[2], 0, CHAN_Z); 2241 2242 fetch_texel(mach->Samplers[sampler_unit], 2243 &r[0], &r[1], &r[2], &ZeroVec, /* inputs */ 2244 tgsi_sampler_lod_bias, 2245 &r[0], &r[1], &r[2], &r[3]); /* outputs */ 2246 break; 2247 2248 case TGSI_TEXTURE_3D: 2249 case TGSI_TEXTURE_CUBE: 2250 2251 FETCH(&r[0], 0, CHAN_X); 2252 FETCH(&r[1], 0, CHAN_Y); 2253 FETCH(&r[2], 0, CHAN_Z); 2254 2255 fetch_texel(mach->Samplers[sampler_unit], 2256 &r[0], &r[1], &r[2], &ZeroVec, 2257 tgsi_sampler_lod_bias, 2258 &r[0], &r[1], &r[2], &r[3]); 2259 break; 2260 2261 default: 2262 assert(0); 2263 } 2264 2265 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2266 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2267 store_dest(mach, &r[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2268 } 2269 } 2270} 2271 2272 2273/** 2274 * Evaluate a constant-valued coefficient at the position of the 2275 * current quad. 2276 */ 2277static void 2278eval_constant_coef( 2279 struct tgsi_exec_machine *mach, 2280 unsigned attrib, 2281 unsigned chan ) 2282{ 2283 unsigned i; 2284 2285 for( i = 0; i < QUAD_SIZE; i++ ) { 2286 mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan]; 2287 } 2288} 2289 2290/** 2291 * Evaluate a linear-valued coefficient at the position of the 2292 * current quad. 2293 */ 2294static void 2295eval_linear_coef( 2296 struct tgsi_exec_machine *mach, 2297 unsigned attrib, 2298 unsigned chan ) 2299{ 2300 const float x = mach->QuadPos.xyzw[0].f[0]; 2301 const float y = mach->QuadPos.xyzw[1].f[0]; 2302 const float dadx = mach->InterpCoefs[attrib].dadx[chan]; 2303 const float dady = mach->InterpCoefs[attrib].dady[chan]; 2304 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; 2305 mach->Inputs[attrib].xyzw[chan].f[0] = a0; 2306 mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx; 2307 mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady; 2308 mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady; 2309} 2310 2311/** 2312 * Evaluate a perspective-valued coefficient at the position of the 2313 * current quad. 2314 */ 2315static void 2316eval_perspective_coef( 2317 struct tgsi_exec_machine *mach, 2318 unsigned attrib, 2319 unsigned chan ) 2320{ 2321 const float x = mach->QuadPos.xyzw[0].f[0]; 2322 const float y = mach->QuadPos.xyzw[1].f[0]; 2323 const float dadx = mach->InterpCoefs[attrib].dadx[chan]; 2324 const float dady = mach->InterpCoefs[attrib].dady[chan]; 2325 const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y; 2326 const float *w = mach->QuadPos.xyzw[3].f; 2327 /* divide by W here */ 2328 mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0]; 2329 mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1]; 2330 mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2]; 2331 mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3]; 2332} 2333 2334 2335typedef void (* eval_coef_func)( 2336 struct tgsi_exec_machine *mach, 2337 unsigned attrib, 2338 unsigned chan ); 2339 2340static void 2341exec_declaration(struct tgsi_exec_machine *mach, 2342 const struct tgsi_full_declaration *decl) 2343{ 2344 if (decl->Declaration.File == TGSI_FILE_RESOURCE) { 2345 mach->Resources[decl->Range.First] = decl->Resource; 2346 return; 2347 } 2348 2349 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { 2350 if (decl->Declaration.File == TGSI_FILE_INPUT) { 2351 uint first, last, mask; 2352 2353 first = decl->Range.First; 2354 last = decl->Range.Last; 2355 mask = decl->Declaration.UsageMask; 2356 2357 /* XXX we could remove this special-case code since 2358 * mach->InterpCoefs[first].a0 should already have the 2359 * front/back-face value. But we should first update the 2360 * ureg code to emit the right UsageMask value (WRITEMASK_X). 2361 * Then, we could remove the tgsi_exec_machine::Face field. 2362 */ 2363 /* XXX make FACE a system value */ 2364 if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) { 2365 uint i; 2366 2367 assert(decl->Semantic.Index == 0); 2368 assert(first == last); 2369 2370 for (i = 0; i < QUAD_SIZE; i++) { 2371 mach->Inputs[first].xyzw[0].f[i] = mach->Face; 2372 } 2373 } else { 2374 eval_coef_func eval; 2375 uint i, j; 2376 2377 switch (decl->Declaration.Interpolate) { 2378 case TGSI_INTERPOLATE_CONSTANT: 2379 eval = eval_constant_coef; 2380 break; 2381 2382 case TGSI_INTERPOLATE_LINEAR: 2383 eval = eval_linear_coef; 2384 break; 2385 2386 case TGSI_INTERPOLATE_PERSPECTIVE: 2387 eval = eval_perspective_coef; 2388 break; 2389 2390 case TGSI_INTERPOLATE_COLOR: 2391 eval = mach->flatshade_color ? eval_constant_coef : eval_perspective_coef; 2392 break; 2393 2394 default: 2395 assert(0); 2396 return; 2397 } 2398 2399 for (j = 0; j < NUM_CHANNELS; j++) { 2400 if (mask & (1 << j)) { 2401 for (i = first; i <= last; i++) { 2402 eval(mach, i, j); 2403 } 2404 } 2405 } 2406 } 2407 } 2408 } 2409 2410 if (decl->Declaration.File == TGSI_FILE_SYSTEM_VALUE) { 2411 mach->SysSemanticToIndex[decl->Declaration.Semantic] = decl->Range.First; 2412 } 2413} 2414 2415 2416typedef void (* micro_op)(union tgsi_exec_channel *dst); 2417 2418static void 2419exec_vector(struct tgsi_exec_machine *mach, 2420 const struct tgsi_full_instruction *inst, 2421 micro_op op, 2422 enum tgsi_exec_datatype dst_datatype) 2423{ 2424 unsigned int chan; 2425 2426 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2427 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2428 union tgsi_exec_channel dst; 2429 2430 op(&dst); 2431 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2432 } 2433 } 2434} 2435 2436typedef void (* micro_unary_op)(union tgsi_exec_channel *dst, 2437 const union tgsi_exec_channel *src); 2438 2439static void 2440exec_scalar_unary(struct tgsi_exec_machine *mach, 2441 const struct tgsi_full_instruction *inst, 2442 micro_unary_op op, 2443 enum tgsi_exec_datatype dst_datatype, 2444 enum tgsi_exec_datatype src_datatype) 2445{ 2446 unsigned int chan; 2447 union tgsi_exec_channel src; 2448 union tgsi_exec_channel dst; 2449 2450 fetch_source(mach, &src, &inst->Src[0], CHAN_X, src_datatype); 2451 op(&dst, &src); 2452 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2453 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2454 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2455 } 2456 } 2457} 2458 2459static void 2460exec_vector_unary(struct tgsi_exec_machine *mach, 2461 const struct tgsi_full_instruction *inst, 2462 micro_unary_op op, 2463 enum tgsi_exec_datatype dst_datatype, 2464 enum tgsi_exec_datatype src_datatype) 2465{ 2466 unsigned int chan; 2467 struct tgsi_exec_vector dst; 2468 2469 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2470 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2471 union tgsi_exec_channel src; 2472 2473 fetch_source(mach, &src, &inst->Src[0], chan, src_datatype); 2474 op(&dst.xyzw[chan], &src); 2475 } 2476 } 2477 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2478 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2479 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2480 } 2481 } 2482} 2483 2484typedef void (* micro_binary_op)(union tgsi_exec_channel *dst, 2485 const union tgsi_exec_channel *src0, 2486 const union tgsi_exec_channel *src1); 2487 2488static void 2489exec_scalar_binary(struct tgsi_exec_machine *mach, 2490 const struct tgsi_full_instruction *inst, 2491 micro_binary_op op, 2492 enum tgsi_exec_datatype dst_datatype, 2493 enum tgsi_exec_datatype src_datatype) 2494{ 2495 unsigned int chan; 2496 union tgsi_exec_channel src[2]; 2497 union tgsi_exec_channel dst; 2498 2499 fetch_source(mach, &src[0], &inst->Src[0], CHAN_X, src_datatype); 2500 fetch_source(mach, &src[1], &inst->Src[1], CHAN_Y, src_datatype); 2501 op(&dst, &src[0], &src[1]); 2502 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2503 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2504 store_dest(mach, &dst, &inst->Dst[0], inst, chan, dst_datatype); 2505 } 2506 } 2507} 2508 2509static void 2510exec_vector_binary(struct tgsi_exec_machine *mach, 2511 const struct tgsi_full_instruction *inst, 2512 micro_binary_op op, 2513 enum tgsi_exec_datatype dst_datatype, 2514 enum tgsi_exec_datatype src_datatype) 2515{ 2516 unsigned int chan; 2517 struct tgsi_exec_vector dst; 2518 2519 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2520 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2521 union tgsi_exec_channel src[2]; 2522 2523 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype); 2524 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype); 2525 op(&dst.xyzw[chan], &src[0], &src[1]); 2526 } 2527 } 2528 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2529 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2530 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2531 } 2532 } 2533} 2534 2535typedef void (* micro_trinary_op)(union tgsi_exec_channel *dst, 2536 const union tgsi_exec_channel *src0, 2537 const union tgsi_exec_channel *src1, 2538 const union tgsi_exec_channel *src2); 2539 2540static void 2541exec_vector_trinary(struct tgsi_exec_machine *mach, 2542 const struct tgsi_full_instruction *inst, 2543 micro_trinary_op op, 2544 enum tgsi_exec_datatype dst_datatype, 2545 enum tgsi_exec_datatype src_datatype) 2546{ 2547 unsigned int chan; 2548 struct tgsi_exec_vector dst; 2549 2550 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2551 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2552 union tgsi_exec_channel src[3]; 2553 2554 fetch_source(mach, &src[0], &inst->Src[0], chan, src_datatype); 2555 fetch_source(mach, &src[1], &inst->Src[1], chan, src_datatype); 2556 fetch_source(mach, &src[2], &inst->Src[2], chan, src_datatype); 2557 op(&dst.xyzw[chan], &src[0], &src[1], &src[2]); 2558 } 2559 } 2560 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2561 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2562 store_dest(mach, &dst.xyzw[chan], &inst->Dst[0], inst, chan, dst_datatype); 2563 } 2564 } 2565} 2566 2567static void 2568exec_dp3(struct tgsi_exec_machine *mach, 2569 const struct tgsi_full_instruction *inst) 2570{ 2571 unsigned int chan; 2572 union tgsi_exec_channel arg[3]; 2573 2574 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2575 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2576 micro_mul(&arg[2], &arg[0], &arg[1]); 2577 2578 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) { 2579 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2580 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT); 2581 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2582 } 2583 2584 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2585 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2586 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2587 } 2588 } 2589} 2590 2591static void 2592exec_dp4(struct tgsi_exec_machine *mach, 2593 const struct tgsi_full_instruction *inst) 2594{ 2595 unsigned int chan; 2596 union tgsi_exec_channel arg[3]; 2597 2598 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2599 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2600 micro_mul(&arg[2], &arg[0], &arg[1]); 2601 2602 for (chan = CHAN_Y; chan <= CHAN_W; chan++) { 2603 fetch_source(mach, &arg[0], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2604 fetch_source(mach, &arg[1], &inst->Src[1], chan, TGSI_EXEC_DATA_FLOAT); 2605 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2606 } 2607 2608 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2609 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2610 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2611 } 2612 } 2613} 2614 2615static void 2616exec_dp2a(struct tgsi_exec_machine *mach, 2617 const struct tgsi_full_instruction *inst) 2618{ 2619 unsigned int chan; 2620 union tgsi_exec_channel arg[3]; 2621 2622 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2623 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2624 micro_mul(&arg[2], &arg[0], &arg[1]); 2625 2626 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2627 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2628 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]); 2629 2630 fetch_source(mach, &arg[1], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2631 micro_add(&arg[0], &arg[0], &arg[1]); 2632 2633 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2634 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2635 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2636 } 2637 } 2638} 2639 2640static void 2641exec_dph(struct tgsi_exec_machine *mach, 2642 const struct tgsi_full_instruction *inst) 2643{ 2644 unsigned int chan; 2645 union tgsi_exec_channel arg[3]; 2646 2647 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2648 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2649 micro_mul(&arg[2], &arg[0], &arg[1]); 2650 2651 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2652 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2653 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2654 2655 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2656 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2657 micro_mad(&arg[0], &arg[0], &arg[1], &arg[2]); 2658 2659 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2660 micro_add(&arg[0], &arg[0], &arg[1]); 2661 2662 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2663 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2664 store_dest(mach, &arg[0], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2665 } 2666 } 2667} 2668 2669static void 2670exec_dp2(struct tgsi_exec_machine *mach, 2671 const struct tgsi_full_instruction *inst) 2672{ 2673 unsigned int chan; 2674 union tgsi_exec_channel arg[3]; 2675 2676 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2677 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2678 micro_mul(&arg[2], &arg[0], &arg[1]); 2679 2680 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2681 fetch_source(mach, &arg[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2682 micro_mad(&arg[2], &arg[0], &arg[1], &arg[2]); 2683 2684 for (chan = 0; chan < NUM_CHANNELS; chan++) { 2685 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2686 store_dest(mach, &arg[2], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2687 } 2688 } 2689} 2690 2691static void 2692exec_nrm4(struct tgsi_exec_machine *mach, 2693 const struct tgsi_full_instruction *inst) 2694{ 2695 unsigned int chan; 2696 union tgsi_exec_channel arg[4]; 2697 union tgsi_exec_channel scale; 2698 2699 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2700 micro_mul(&scale, &arg[0], &arg[0]); 2701 2702 for (chan = CHAN_Y; chan <= CHAN_W; chan++) { 2703 union tgsi_exec_channel product; 2704 2705 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2706 micro_mul(&product, &arg[chan], &arg[chan]); 2707 micro_add(&scale, &scale, &product); 2708 } 2709 2710 micro_rsq(&scale, &scale); 2711 2712 for (chan = CHAN_X; chan <= CHAN_W; chan++) { 2713 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2714 micro_mul(&arg[chan], &arg[chan], &scale); 2715 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2716 } 2717 } 2718} 2719 2720static void 2721exec_nrm3(struct tgsi_exec_machine *mach, 2722 const struct tgsi_full_instruction *inst) 2723{ 2724 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) { 2725 unsigned int chan; 2726 union tgsi_exec_channel arg[3]; 2727 union tgsi_exec_channel scale; 2728 2729 fetch_source(mach, &arg[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2730 micro_mul(&scale, &arg[0], &arg[0]); 2731 2732 for (chan = CHAN_Y; chan <= CHAN_Z; chan++) { 2733 union tgsi_exec_channel product; 2734 2735 fetch_source(mach, &arg[chan], &inst->Src[0], chan, TGSI_EXEC_DATA_FLOAT); 2736 micro_mul(&product, &arg[chan], &arg[chan]); 2737 micro_add(&scale, &scale, &product); 2738 } 2739 2740 micro_rsq(&scale, &scale); 2741 2742 for (chan = CHAN_X; chan <= CHAN_Z; chan++) { 2743 if (inst->Dst[0].Register.WriteMask & (1 << chan)) { 2744 micro_mul(&arg[chan], &arg[chan], &scale); 2745 store_dest(mach, &arg[chan], &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT); 2746 } 2747 } 2748 } 2749 2750 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2751 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2752 } 2753} 2754 2755static void 2756exec_scs(struct tgsi_exec_machine *mach, 2757 const struct tgsi_full_instruction *inst) 2758{ 2759 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XY) { 2760 union tgsi_exec_channel arg; 2761 union tgsi_exec_channel result; 2762 2763 fetch_source(mach, &arg, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2764 2765 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2766 micro_cos(&result, &arg); 2767 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2768 } 2769 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2770 micro_sin(&result, &arg); 2771 store_dest(mach, &result, &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2772 } 2773 } 2774 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2775 store_dest(mach, &ZeroVec, &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2776 } 2777 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2778 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2779 } 2780} 2781 2782static void 2783exec_x2d(struct tgsi_exec_machine *mach, 2784 const struct tgsi_full_instruction *inst) 2785{ 2786 union tgsi_exec_channel r[4]; 2787 union tgsi_exec_channel d[2]; 2788 2789 fetch_source(mach, &r[0], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2790 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2791 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XZ) { 2792 fetch_source(mach, &r[2], &inst->Src[2], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2793 micro_mul(&r[2], &r[2], &r[0]); 2794 fetch_source(mach, &r[3], &inst->Src[2], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2795 micro_mul(&r[3], &r[3], &r[1]); 2796 micro_add(&r[2], &r[2], &r[3]); 2797 fetch_source(mach, &r[3], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2798 micro_add(&d[0], &r[2], &r[3]); 2799 } 2800 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YW) { 2801 fetch_source(mach, &r[2], &inst->Src[2], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2802 micro_mul(&r[2], &r[2], &r[0]); 2803 fetch_source(mach, &r[3], &inst->Src[2], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2804 micro_mul(&r[3], &r[3], &r[1]); 2805 micro_add(&r[2], &r[2], &r[3]); 2806 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2807 micro_add(&d[1], &r[2], &r[3]); 2808 } 2809 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2810 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2811 } 2812 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2813 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2814 } 2815 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2816 store_dest(mach, &d[0], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2817 } 2818 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2819 store_dest(mach, &d[1], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2820 } 2821} 2822 2823static void 2824exec_rfl(struct tgsi_exec_machine *mach, 2825 const struct tgsi_full_instruction *inst) 2826{ 2827 union tgsi_exec_channel r[9]; 2828 2829 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XYZ) { 2830 /* r0 = dp3(src0, src0) */ 2831 fetch_source(mach, &r[2], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2832 micro_mul(&r[0], &r[2], &r[2]); 2833 fetch_source(mach, &r[4], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2834 micro_mul(&r[8], &r[4], &r[4]); 2835 micro_add(&r[0], &r[0], &r[8]); 2836 fetch_source(mach, &r[6], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2837 micro_mul(&r[8], &r[6], &r[6]); 2838 micro_add(&r[0], &r[0], &r[8]); 2839 2840 /* r1 = dp3(src0, src1) */ 2841 fetch_source(mach, &r[3], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2842 micro_mul(&r[1], &r[2], &r[3]); 2843 fetch_source(mach, &r[5], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2844 micro_mul(&r[8], &r[4], &r[5]); 2845 micro_add(&r[1], &r[1], &r[8]); 2846 fetch_source(mach, &r[7], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2847 micro_mul(&r[8], &r[6], &r[7]); 2848 micro_add(&r[1], &r[1], &r[8]); 2849 2850 /* r1 = 2 * r1 / r0 */ 2851 micro_add(&r[1], &r[1], &r[1]); 2852 micro_div(&r[1], &r[1], &r[0]); 2853 2854 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2855 micro_mul(&r[2], &r[2], &r[1]); 2856 micro_sub(&r[2], &r[2], &r[3]); 2857 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2858 } 2859 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2860 micro_mul(&r[4], &r[4], &r[1]); 2861 micro_sub(&r[4], &r[4], &r[5]); 2862 store_dest(mach, &r[4], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2863 } 2864 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2865 micro_mul(&r[6], &r[6], &r[1]); 2866 micro_sub(&r[6], &r[6], &r[7]); 2867 store_dest(mach, &r[6], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2868 } 2869 } 2870 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2871 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2872 } 2873} 2874 2875static void 2876exec_xpd(struct tgsi_exec_machine *mach, 2877 const struct tgsi_full_instruction *inst) 2878{ 2879 union tgsi_exec_channel r[6]; 2880 union tgsi_exec_channel d[3]; 2881 2882 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2883 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2884 2885 micro_mul(&r[2], &r[0], &r[1]); 2886 2887 fetch_source(mach, &r[3], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2888 fetch_source(mach, &r[4], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2889 2890 micro_mul(&r[5], &r[3], &r[4] ); 2891 micro_sub(&d[CHAN_X], &r[2], &r[5]); 2892 2893 fetch_source(mach, &r[2], &inst->Src[1], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2894 2895 micro_mul(&r[3], &r[3], &r[2]); 2896 2897 fetch_source(mach, &r[5], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2898 2899 micro_mul(&r[1], &r[1], &r[5]); 2900 micro_sub(&d[CHAN_Y], &r[3], &r[1]); 2901 2902 micro_mul(&r[5], &r[5], &r[4]); 2903 micro_mul(&r[0], &r[0], &r[2]); 2904 micro_sub(&d[CHAN_Z], &r[5], &r[0]); 2905 2906 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2907 store_dest(mach, &d[CHAN_X], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2908 } 2909 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2910 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2911 } 2912 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2913 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2914 } 2915 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2916 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2917 } 2918} 2919 2920static void 2921exec_dst(struct tgsi_exec_machine *mach, 2922 const struct tgsi_full_instruction *inst) 2923{ 2924 union tgsi_exec_channel r[2]; 2925 union tgsi_exec_channel d[4]; 2926 2927 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2928 fetch_source(mach, &r[0], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2929 fetch_source(mach, &r[1], &inst->Src[1], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2930 micro_mul(&d[CHAN_Y], &r[0], &r[1]); 2931 } 2932 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2933 fetch_source(mach, &d[CHAN_Z], &inst->Src[0], CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2934 } 2935 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2936 fetch_source(mach, &d[CHAN_W], &inst->Src[1], CHAN_W, TGSI_EXEC_DATA_FLOAT); 2937 } 2938 2939 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2940 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2941 } 2942 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2943 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2944 } 2945 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2946 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2947 } 2948 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2949 store_dest(mach, &d[CHAN_W], &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2950 } 2951} 2952 2953static void 2954exec_log(struct tgsi_exec_machine *mach, 2955 const struct tgsi_full_instruction *inst) 2956{ 2957 union tgsi_exec_channel r[3]; 2958 2959 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2960 micro_abs(&r[2], &r[0]); /* r2 = abs(r0) */ 2961 micro_lg2(&r[1], &r[2]); /* r1 = lg2(r2) */ 2962 micro_flr(&r[0], &r[1]); /* r0 = floor(r1) */ 2963 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2964 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2965 } 2966 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2967 micro_exp2(&r[0], &r[0]); /* r0 = 2 ^ r0 */ 2968 micro_div(&r[0], &r[2], &r[0]); /* r0 = r2 / r0 */ 2969 store_dest(mach, &r[0], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2970 } 2971 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2972 store_dest(mach, &r[1], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2973 } 2974 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 2975 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 2976 } 2977} 2978 2979static void 2980exec_exp(struct tgsi_exec_machine *mach, 2981 const struct tgsi_full_instruction *inst) 2982{ 2983 union tgsi_exec_channel r[3]; 2984 2985 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 2986 micro_flr(&r[1], &r[0]); /* r1 = floor(r0) */ 2987 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 2988 micro_exp2(&r[2], &r[1]); /* r2 = 2 ^ r1 */ 2989 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 2990 } 2991 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 2992 micro_sub(&r[2], &r[0], &r[1]); /* r2 = r0 - r1 */ 2993 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 2994 } 2995 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 2996 micro_exp2(&r[2], &r[0]); /* r2 = 2 ^ r0 */ 2997 store_dest(mach, &r[2], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 2998 } 2999 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 3000 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 3001 } 3002} 3003 3004static void 3005exec_lit(struct tgsi_exec_machine *mach, 3006 const struct tgsi_full_instruction *inst) 3007{ 3008 union tgsi_exec_channel r[3]; 3009 union tgsi_exec_channel d[3]; 3010 3011 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_YZ) { 3012 fetch_source(mach, &r[0], &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_FLOAT); 3013 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) { 3014 fetch_source(mach, &r[1], &inst->Src[0], CHAN_Y, TGSI_EXEC_DATA_FLOAT); 3015 micro_max(&r[1], &r[1], &ZeroVec); 3016 3017 fetch_source(mach, &r[2], &inst->Src[0], CHAN_W, TGSI_EXEC_DATA_FLOAT); 3018 micro_min(&r[2], &r[2], &P128Vec); 3019 micro_max(&r[2], &r[2], &M128Vec); 3020 micro_pow(&r[1], &r[1], &r[2]); 3021 micro_lt(&d[CHAN_Z], &ZeroVec, &r[0], &r[1], &ZeroVec); 3022 store_dest(mach, &d[CHAN_Z], &inst->Dst[0], inst, CHAN_Z, TGSI_EXEC_DATA_FLOAT); 3023 } 3024 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { 3025 micro_max(&d[CHAN_Y], &r[0], &ZeroVec); 3026 store_dest(mach, &d[CHAN_Y], &inst->Dst[0], inst, CHAN_Y, TGSI_EXEC_DATA_FLOAT); 3027 } 3028 } 3029 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { 3030 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_X, TGSI_EXEC_DATA_FLOAT); 3031 } 3032 3033 if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) { 3034 store_dest(mach, &OneVec, &inst->Dst[0], inst, CHAN_W, TGSI_EXEC_DATA_FLOAT); 3035 } 3036} 3037 3038static void 3039exec_break(struct tgsi_exec_machine *mach) 3040{ 3041 if (mach->BreakType == TGSI_EXEC_BREAK_INSIDE_LOOP) { 3042 /* turn off loop channels for each enabled exec channel */ 3043 mach->LoopMask &= ~mach->ExecMask; 3044 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 3045 UPDATE_EXEC_MASK(mach); 3046 } else { 3047 assert(mach->BreakType == TGSI_EXEC_BREAK_INSIDE_SWITCH); 3048 3049 mach->Switch.mask = 0x0; 3050 3051 UPDATE_EXEC_MASK(mach); 3052 } 3053} 3054 3055static void 3056exec_switch(struct tgsi_exec_machine *mach, 3057 const struct tgsi_full_instruction *inst) 3058{ 3059 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING); 3060 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3061 3062 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch; 3063 fetch_source(mach, &mach->Switch.selector, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT); 3064 mach->Switch.mask = 0x0; 3065 mach->Switch.defaultMask = 0x0; 3066 3067 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3068 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_SWITCH; 3069 3070 UPDATE_EXEC_MASK(mach); 3071} 3072 3073static void 3074exec_case(struct tgsi_exec_machine *mach, 3075 const struct tgsi_full_instruction *inst) 3076{ 3077 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask; 3078 union tgsi_exec_channel src; 3079 uint mask = 0; 3080 3081 fetch_source(mach, &src, &inst->Src[0], CHAN_X, TGSI_EXEC_DATA_UINT); 3082 3083 if (mach->Switch.selector.u[0] == src.u[0]) { 3084 mask |= 0x1; 3085 } 3086 if (mach->Switch.selector.u[1] == src.u[1]) { 3087 mask |= 0x2; 3088 } 3089 if (mach->Switch.selector.u[2] == src.u[2]) { 3090 mask |= 0x4; 3091 } 3092 if (mach->Switch.selector.u[3] == src.u[3]) { 3093 mask |= 0x8; 3094 } 3095 3096 mach->Switch.defaultMask |= mask; 3097 3098 mach->Switch.mask |= mask & prevMask; 3099 3100 UPDATE_EXEC_MASK(mach); 3101} 3102 3103static void 3104exec_default(struct tgsi_exec_machine *mach) 3105{ 3106 uint prevMask = mach->SwitchStack[mach->SwitchStackTop - 1].mask; 3107 3108 mach->Switch.mask |= ~mach->Switch.defaultMask & prevMask; 3109 3110 UPDATE_EXEC_MASK(mach); 3111} 3112 3113static void 3114exec_endswitch(struct tgsi_exec_machine *mach) 3115{ 3116 mach->Switch = mach->SwitchStack[--mach->SwitchStackTop]; 3117 mach->BreakType = mach->BreakStack[--mach->BreakStackTop]; 3118 3119 UPDATE_EXEC_MASK(mach); 3120} 3121 3122static void 3123micro_i2f(union tgsi_exec_channel *dst, 3124 const union tgsi_exec_channel *src) 3125{ 3126 dst->f[0] = (float)src->i[0]; 3127 dst->f[1] = (float)src->i[1]; 3128 dst->f[2] = (float)src->i[2]; 3129 dst->f[3] = (float)src->i[3]; 3130} 3131 3132static void 3133micro_not(union tgsi_exec_channel *dst, 3134 const union tgsi_exec_channel *src) 3135{ 3136 dst->u[0] = ~src->u[0]; 3137 dst->u[1] = ~src->u[1]; 3138 dst->u[2] = ~src->u[2]; 3139 dst->u[3] = ~src->u[3]; 3140} 3141 3142static void 3143micro_shl(union tgsi_exec_channel *dst, 3144 const union tgsi_exec_channel *src0, 3145 const union tgsi_exec_channel *src1) 3146{ 3147 dst->u[0] = src0->u[0] << src1->u[0]; 3148 dst->u[1] = src0->u[1] << src1->u[1]; 3149 dst->u[2] = src0->u[2] << src1->u[2]; 3150 dst->u[3] = src0->u[3] << src1->u[3]; 3151} 3152 3153static void 3154micro_and(union tgsi_exec_channel *dst, 3155 const union tgsi_exec_channel *src0, 3156 const union tgsi_exec_channel *src1) 3157{ 3158 dst->u[0] = src0->u[0] & src1->u[0]; 3159 dst->u[1] = src0->u[1] & src1->u[1]; 3160 dst->u[2] = src0->u[2] & src1->u[2]; 3161 dst->u[3] = src0->u[3] & src1->u[3]; 3162} 3163 3164static void 3165micro_or(union tgsi_exec_channel *dst, 3166 const union tgsi_exec_channel *src0, 3167 const union tgsi_exec_channel *src1) 3168{ 3169 dst->u[0] = src0->u[0] | src1->u[0]; 3170 dst->u[1] = src0->u[1] | src1->u[1]; 3171 dst->u[2] = src0->u[2] | src1->u[2]; 3172 dst->u[3] = src0->u[3] | src1->u[3]; 3173} 3174 3175static void 3176micro_xor(union tgsi_exec_channel *dst, 3177 const union tgsi_exec_channel *src0, 3178 const union tgsi_exec_channel *src1) 3179{ 3180 dst->u[0] = src0->u[0] ^ src1->u[0]; 3181 dst->u[1] = src0->u[1] ^ src1->u[1]; 3182 dst->u[2] = src0->u[2] ^ src1->u[2]; 3183 dst->u[3] = src0->u[3] ^ src1->u[3]; 3184} 3185 3186static void 3187micro_mod(union tgsi_exec_channel *dst, 3188 const union tgsi_exec_channel *src0, 3189 const union tgsi_exec_channel *src1) 3190{ 3191 dst->i[0] = src0->i[0] % src1->i[0]; 3192 dst->i[1] = src0->i[1] % src1->i[1]; 3193 dst->i[2] = src0->i[2] % src1->i[2]; 3194 dst->i[3] = src0->i[3] % src1->i[3]; 3195} 3196 3197static void 3198micro_f2i(union tgsi_exec_channel *dst, 3199 const union tgsi_exec_channel *src) 3200{ 3201 dst->i[0] = (int)src->f[0]; 3202 dst->i[1] = (int)src->f[1]; 3203 dst->i[2] = (int)src->f[2]; 3204 dst->i[3] = (int)src->f[3]; 3205} 3206 3207static void 3208micro_idiv(union tgsi_exec_channel *dst, 3209 const union tgsi_exec_channel *src0, 3210 const union tgsi_exec_channel *src1) 3211{ 3212 dst->i[0] = src0->i[0] / src1->i[0]; 3213 dst->i[1] = src0->i[1] / src1->i[1]; 3214 dst->i[2] = src0->i[2] / src1->i[2]; 3215 dst->i[3] = src0->i[3] / src1->i[3]; 3216} 3217 3218static void 3219micro_imax(union tgsi_exec_channel *dst, 3220 const union tgsi_exec_channel *src0, 3221 const union tgsi_exec_channel *src1) 3222{ 3223 dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0]; 3224 dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1]; 3225 dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2]; 3226 dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3]; 3227} 3228 3229static void 3230micro_imin(union tgsi_exec_channel *dst, 3231 const union tgsi_exec_channel *src0, 3232 const union tgsi_exec_channel *src1) 3233{ 3234 dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0]; 3235 dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1]; 3236 dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2]; 3237 dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3]; 3238} 3239 3240static void 3241micro_isge(union tgsi_exec_channel *dst, 3242 const union tgsi_exec_channel *src0, 3243 const union tgsi_exec_channel *src1) 3244{ 3245 dst->i[0] = src0->i[0] >= src1->i[0] ? -1 : 0; 3246 dst->i[1] = src0->i[1] >= src1->i[1] ? -1 : 0; 3247 dst->i[2] = src0->i[2] >= src1->i[2] ? -1 : 0; 3248 dst->i[3] = src0->i[3] >= src1->i[3] ? -1 : 0; 3249} 3250 3251static void 3252micro_ishr(union tgsi_exec_channel *dst, 3253 const union tgsi_exec_channel *src0, 3254 const union tgsi_exec_channel *src1) 3255{ 3256 dst->i[0] = src0->i[0] >> src1->i[0]; 3257 dst->i[1] = src0->i[1] >> src1->i[1]; 3258 dst->i[2] = src0->i[2] >> src1->i[2]; 3259 dst->i[3] = src0->i[3] >> src1->i[3]; 3260} 3261 3262static void 3263micro_islt(union tgsi_exec_channel *dst, 3264 const union tgsi_exec_channel *src0, 3265 const union tgsi_exec_channel *src1) 3266{ 3267 dst->i[0] = src0->i[0] < src1->i[0] ? -1 : 0; 3268 dst->i[1] = src0->i[1] < src1->i[1] ? -1 : 0; 3269 dst->i[2] = src0->i[2] < src1->i[2] ? -1 : 0; 3270 dst->i[3] = src0->i[3] < src1->i[3] ? -1 : 0; 3271} 3272 3273static void 3274micro_f2u(union tgsi_exec_channel *dst, 3275 const union tgsi_exec_channel *src) 3276{ 3277 dst->u[0] = (uint)src->f[0]; 3278 dst->u[1] = (uint)src->f[1]; 3279 dst->u[2] = (uint)src->f[2]; 3280 dst->u[3] = (uint)src->f[3]; 3281} 3282 3283static void 3284micro_u2f(union tgsi_exec_channel *dst, 3285 const union tgsi_exec_channel *src) 3286{ 3287 dst->f[0] = (float)src->u[0]; 3288 dst->f[1] = (float)src->u[1]; 3289 dst->f[2] = (float)src->u[2]; 3290 dst->f[3] = (float)src->u[3]; 3291} 3292 3293static void 3294micro_uadd(union tgsi_exec_channel *dst, 3295 const union tgsi_exec_channel *src0, 3296 const union tgsi_exec_channel *src1) 3297{ 3298 dst->u[0] = src0->u[0] + src1->u[0]; 3299 dst->u[1] = src0->u[1] + src1->u[1]; 3300 dst->u[2] = src0->u[2] + src1->u[2]; 3301 dst->u[3] = src0->u[3] + src1->u[3]; 3302} 3303 3304static void 3305micro_udiv(union tgsi_exec_channel *dst, 3306 const union tgsi_exec_channel *src0, 3307 const union tgsi_exec_channel *src1) 3308{ 3309 dst->u[0] = src0->u[0] / src1->u[0]; 3310 dst->u[1] = src0->u[1] / src1->u[1]; 3311 dst->u[2] = src0->u[2] / src1->u[2]; 3312 dst->u[3] = src0->u[3] / src1->u[3]; 3313} 3314 3315static void 3316micro_umad(union tgsi_exec_channel *dst, 3317 const union tgsi_exec_channel *src0, 3318 const union tgsi_exec_channel *src1, 3319 const union tgsi_exec_channel *src2) 3320{ 3321 dst->u[0] = src0->u[0] * src1->u[0] + src2->u[0]; 3322 dst->u[1] = src0->u[1] * src1->u[1] + src2->u[1]; 3323 dst->u[2] = src0->u[2] * src1->u[2] + src2->u[2]; 3324 dst->u[3] = src0->u[3] * src1->u[3] + src2->u[3]; 3325} 3326 3327static void 3328micro_umax(union tgsi_exec_channel *dst, 3329 const union tgsi_exec_channel *src0, 3330 const union tgsi_exec_channel *src1) 3331{ 3332 dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0]; 3333 dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1]; 3334 dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2]; 3335 dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3]; 3336} 3337 3338static void 3339micro_umin(union tgsi_exec_channel *dst, 3340 const union tgsi_exec_channel *src0, 3341 const union tgsi_exec_channel *src1) 3342{ 3343 dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0]; 3344 dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1]; 3345 dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2]; 3346 dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3]; 3347} 3348 3349static void 3350micro_umod(union tgsi_exec_channel *dst, 3351 const union tgsi_exec_channel *src0, 3352 const union tgsi_exec_channel *src1) 3353{ 3354 dst->u[0] = src0->u[0] % src1->u[0]; 3355 dst->u[1] = src0->u[1] % src1->u[1]; 3356 dst->u[2] = src0->u[2] % src1->u[2]; 3357 dst->u[3] = src0->u[3] % src1->u[3]; 3358} 3359 3360static void 3361micro_umul(union tgsi_exec_channel *dst, 3362 const union tgsi_exec_channel *src0, 3363 const union tgsi_exec_channel *src1) 3364{ 3365 dst->u[0] = src0->u[0] * src1->u[0]; 3366 dst->u[1] = src0->u[1] * src1->u[1]; 3367 dst->u[2] = src0->u[2] * src1->u[2]; 3368 dst->u[3] = src0->u[3] * src1->u[3]; 3369} 3370 3371static void 3372micro_useq(union tgsi_exec_channel *dst, 3373 const union tgsi_exec_channel *src0, 3374 const union tgsi_exec_channel *src1) 3375{ 3376 dst->u[0] = src0->u[0] == src1->u[0] ? ~0 : 0; 3377 dst->u[1] = src0->u[1] == src1->u[1] ? ~0 : 0; 3378 dst->u[2] = src0->u[2] == src1->u[2] ? ~0 : 0; 3379 dst->u[3] = src0->u[3] == src1->u[3] ? ~0 : 0; 3380} 3381 3382static void 3383micro_usge(union tgsi_exec_channel *dst, 3384 const union tgsi_exec_channel *src0, 3385 const union tgsi_exec_channel *src1) 3386{ 3387 dst->u[0] = src0->u[0] >= src1->u[0] ? ~0 : 0; 3388 dst->u[1] = src0->u[1] >= src1->u[1] ? ~0 : 0; 3389 dst->u[2] = src0->u[2] >= src1->u[2] ? ~0 : 0; 3390 dst->u[3] = src0->u[3] >= src1->u[3] ? ~0 : 0; 3391} 3392 3393static void 3394micro_ushr(union tgsi_exec_channel *dst, 3395 const union tgsi_exec_channel *src0, 3396 const union tgsi_exec_channel *src1) 3397{ 3398 dst->u[0] = src0->u[0] >> src1->u[0]; 3399 dst->u[1] = src0->u[1] >> src1->u[1]; 3400 dst->u[2] = src0->u[2] >> src1->u[2]; 3401 dst->u[3] = src0->u[3] >> src1->u[3]; 3402} 3403 3404static void 3405micro_uslt(union tgsi_exec_channel *dst, 3406 const union tgsi_exec_channel *src0, 3407 const union tgsi_exec_channel *src1) 3408{ 3409 dst->u[0] = src0->u[0] < src1->u[0] ? ~0 : 0; 3410 dst->u[1] = src0->u[1] < src1->u[1] ? ~0 : 0; 3411 dst->u[2] = src0->u[2] < src1->u[2] ? ~0 : 0; 3412 dst->u[3] = src0->u[3] < src1->u[3] ? ~0 : 0; 3413} 3414 3415static void 3416micro_usne(union tgsi_exec_channel *dst, 3417 const union tgsi_exec_channel *src0, 3418 const union tgsi_exec_channel *src1) 3419{ 3420 dst->u[0] = src0->u[0] != src1->u[0] ? ~0 : 0; 3421 dst->u[1] = src0->u[1] != src1->u[1] ? ~0 : 0; 3422 dst->u[2] = src0->u[2] != src1->u[2] ? ~0 : 0; 3423 dst->u[3] = src0->u[3] != src1->u[3] ? ~0 : 0; 3424} 3425 3426static void 3427micro_uarl(union tgsi_exec_channel *dst, 3428 const union tgsi_exec_channel *src) 3429{ 3430 dst->i[0] = src->u[0]; 3431 dst->i[1] = src->u[1]; 3432 dst->i[2] = src->u[2]; 3433 dst->i[3] = src->u[3]; 3434} 3435 3436static void 3437micro_ucmp(union tgsi_exec_channel *dst, 3438 const union tgsi_exec_channel *src0, 3439 const union tgsi_exec_channel *src1, 3440 const union tgsi_exec_channel *src2) 3441{ 3442 dst->u[0] = src0->u[0] ? src1->u[0] : src2->u[0]; 3443 dst->u[1] = src0->u[1] ? src1->u[1] : src2->u[1]; 3444 dst->u[2] = src0->u[2] ? src1->u[2] : src2->u[2]; 3445 dst->u[3] = src0->u[3] ? src1->u[3] : src2->u[3]; 3446} 3447 3448static void 3449exec_instruction( 3450 struct tgsi_exec_machine *mach, 3451 const struct tgsi_full_instruction *inst, 3452 int *pc ) 3453{ 3454 union tgsi_exec_channel r[10]; 3455 3456 (*pc)++; 3457 3458 switch (inst->Instruction.Opcode) { 3459 case TGSI_OPCODE_ARL: 3460 exec_vector_unary(mach, inst, micro_arl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3461 break; 3462 3463 case TGSI_OPCODE_MOV: 3464 exec_vector_unary(mach, inst, micro_mov, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT); 3465 break; 3466 3467 case TGSI_OPCODE_LIT: 3468 exec_lit(mach, inst); 3469 break; 3470 3471 case TGSI_OPCODE_RCP: 3472 exec_scalar_unary(mach, inst, micro_rcp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3473 break; 3474 3475 case TGSI_OPCODE_RSQ: 3476 exec_scalar_unary(mach, inst, micro_rsq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3477 break; 3478 3479 case TGSI_OPCODE_EXP: 3480 exec_exp(mach, inst); 3481 break; 3482 3483 case TGSI_OPCODE_LOG: 3484 exec_log(mach, inst); 3485 break; 3486 3487 case TGSI_OPCODE_MUL: 3488 exec_vector_binary(mach, inst, micro_mul, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3489 break; 3490 3491 case TGSI_OPCODE_ADD: 3492 exec_vector_binary(mach, inst, micro_add, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3493 break; 3494 3495 case TGSI_OPCODE_DP3: 3496 exec_dp3(mach, inst); 3497 break; 3498 3499 case TGSI_OPCODE_DP4: 3500 exec_dp4(mach, inst); 3501 break; 3502 3503 case TGSI_OPCODE_DST: 3504 exec_dst(mach, inst); 3505 break; 3506 3507 case TGSI_OPCODE_MIN: 3508 exec_vector_binary(mach, inst, micro_min, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3509 break; 3510 3511 case TGSI_OPCODE_MAX: 3512 exec_vector_binary(mach, inst, micro_max, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3513 break; 3514 3515 case TGSI_OPCODE_SLT: 3516 exec_vector_binary(mach, inst, micro_slt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3517 break; 3518 3519 case TGSI_OPCODE_SGE: 3520 exec_vector_binary(mach, inst, micro_sge, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3521 break; 3522 3523 case TGSI_OPCODE_MAD: 3524 exec_vector_trinary(mach, inst, micro_mad, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3525 break; 3526 3527 case TGSI_OPCODE_SUB: 3528 exec_vector_binary(mach, inst, micro_sub, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3529 break; 3530 3531 case TGSI_OPCODE_LRP: 3532 exec_vector_trinary(mach, inst, micro_lrp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3533 break; 3534 3535 case TGSI_OPCODE_CND: 3536 exec_vector_trinary(mach, inst, micro_cnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3537 break; 3538 3539 case TGSI_OPCODE_DP2A: 3540 exec_dp2a(mach, inst); 3541 break; 3542 3543 case TGSI_OPCODE_FRC: 3544 exec_vector_unary(mach, inst, micro_frc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3545 break; 3546 3547 case TGSI_OPCODE_CLAMP: 3548 exec_vector_trinary(mach, inst, micro_clamp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3549 break; 3550 3551 case TGSI_OPCODE_FLR: 3552 exec_vector_unary(mach, inst, micro_flr, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3553 break; 3554 3555 case TGSI_OPCODE_ROUND: 3556 exec_vector_unary(mach, inst, micro_rnd, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3557 break; 3558 3559 case TGSI_OPCODE_EX2: 3560 exec_scalar_unary(mach, inst, micro_exp2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3561 break; 3562 3563 case TGSI_OPCODE_LG2: 3564 exec_scalar_unary(mach, inst, micro_lg2, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3565 break; 3566 3567 case TGSI_OPCODE_POW: 3568 exec_scalar_binary(mach, inst, micro_pow, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3569 break; 3570 3571 case TGSI_OPCODE_XPD: 3572 exec_xpd(mach, inst); 3573 break; 3574 3575 case TGSI_OPCODE_ABS: 3576 exec_vector_unary(mach, inst, micro_abs, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3577 break; 3578 3579 case TGSI_OPCODE_RCC: 3580 exec_scalar_unary(mach, inst, micro_rcc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3581 break; 3582 3583 case TGSI_OPCODE_DPH: 3584 exec_dph(mach, inst); 3585 break; 3586 3587 case TGSI_OPCODE_COS: 3588 exec_scalar_unary(mach, inst, micro_cos, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3589 break; 3590 3591 case TGSI_OPCODE_DDX: 3592 exec_vector_unary(mach, inst, micro_ddx, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3593 break; 3594 3595 case TGSI_OPCODE_DDY: 3596 exec_vector_unary(mach, inst, micro_ddy, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3597 break; 3598 3599 case TGSI_OPCODE_KILP: 3600 exec_kilp (mach, inst); 3601 break; 3602 3603 case TGSI_OPCODE_KIL: 3604 exec_kil (mach, inst); 3605 break; 3606 3607 case TGSI_OPCODE_PK2H: 3608 assert (0); 3609 break; 3610 3611 case TGSI_OPCODE_PK2US: 3612 assert (0); 3613 break; 3614 3615 case TGSI_OPCODE_PK4B: 3616 assert (0); 3617 break; 3618 3619 case TGSI_OPCODE_PK4UB: 3620 assert (0); 3621 break; 3622 3623 case TGSI_OPCODE_RFL: 3624 exec_rfl(mach, inst); 3625 break; 3626 3627 case TGSI_OPCODE_SEQ: 3628 exec_vector_binary(mach, inst, micro_seq, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3629 break; 3630 3631 case TGSI_OPCODE_SFL: 3632 exec_vector(mach, inst, micro_sfl, TGSI_EXEC_DATA_FLOAT); 3633 break; 3634 3635 case TGSI_OPCODE_SGT: 3636 exec_vector_binary(mach, inst, micro_sgt, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3637 break; 3638 3639 case TGSI_OPCODE_SIN: 3640 exec_scalar_unary(mach, inst, micro_sin, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3641 break; 3642 3643 case TGSI_OPCODE_SLE: 3644 exec_vector_binary(mach, inst, micro_sle, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3645 break; 3646 3647 case TGSI_OPCODE_SNE: 3648 exec_vector_binary(mach, inst, micro_sne, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3649 break; 3650 3651 case TGSI_OPCODE_STR: 3652 exec_vector(mach, inst, micro_str, TGSI_EXEC_DATA_FLOAT); 3653 break; 3654 3655 case TGSI_OPCODE_TEX: 3656 /* simple texture lookup */ 3657 /* src[0] = texcoord */ 3658 /* src[1] = sampler unit */ 3659 exec_tex(mach, inst, TEX_MODIFIER_NONE); 3660 break; 3661 3662 case TGSI_OPCODE_TXB: 3663 /* Texture lookup with lod bias */ 3664 /* src[0] = texcoord (src[0].w = LOD bias) */ 3665 /* src[1] = sampler unit */ 3666 exec_tex(mach, inst, TEX_MODIFIER_LOD_BIAS); 3667 break; 3668 3669 case TGSI_OPCODE_TXD: 3670 /* Texture lookup with explict partial derivatives */ 3671 /* src[0] = texcoord */ 3672 /* src[1] = d[strq]/dx */ 3673 /* src[2] = d[strq]/dy */ 3674 /* src[3] = sampler unit */ 3675 exec_txd(mach, inst); 3676 break; 3677 3678 case TGSI_OPCODE_TXL: 3679 /* Texture lookup with explit LOD */ 3680 /* src[0] = texcoord (src[0].w = LOD) */ 3681 /* src[1] = sampler unit */ 3682 exec_tex(mach, inst, TEX_MODIFIER_EXPLICIT_LOD); 3683 break; 3684 3685 case TGSI_OPCODE_TXP: 3686 /* Texture lookup with projection */ 3687 /* src[0] = texcoord (src[0].w = projection) */ 3688 /* src[1] = sampler unit */ 3689 exec_tex(mach, inst, TEX_MODIFIER_PROJECTED); 3690 break; 3691 3692 case TGSI_OPCODE_UP2H: 3693 assert (0); 3694 break; 3695 3696 case TGSI_OPCODE_UP2US: 3697 assert (0); 3698 break; 3699 3700 case TGSI_OPCODE_UP4B: 3701 assert (0); 3702 break; 3703 3704 case TGSI_OPCODE_UP4UB: 3705 assert (0); 3706 break; 3707 3708 case TGSI_OPCODE_X2D: 3709 exec_x2d(mach, inst); 3710 break; 3711 3712 case TGSI_OPCODE_ARA: 3713 assert (0); 3714 break; 3715 3716 case TGSI_OPCODE_ARR: 3717 exec_vector_unary(mach, inst, micro_arr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 3718 break; 3719 3720 case TGSI_OPCODE_BRA: 3721 assert (0); 3722 break; 3723 3724 case TGSI_OPCODE_CAL: 3725 /* skip the call if no execution channels are enabled */ 3726 if (mach->ExecMask) { 3727 /* do the call */ 3728 3729 /* First, record the depths of the execution stacks. 3730 * This is important for deeply nested/looped return statements. 3731 * We have to unwind the stacks by the correct amount. For a 3732 * real code generator, we could determine the number of entries 3733 * to pop off each stack with simple static analysis and avoid 3734 * implementing this data structure at run time. 3735 */ 3736 mach->CallStack[mach->CallStackTop].CondStackTop = mach->CondStackTop; 3737 mach->CallStack[mach->CallStackTop].LoopStackTop = mach->LoopStackTop; 3738 mach->CallStack[mach->CallStackTop].ContStackTop = mach->ContStackTop; 3739 mach->CallStack[mach->CallStackTop].SwitchStackTop = mach->SwitchStackTop; 3740 mach->CallStack[mach->CallStackTop].BreakStackTop = mach->BreakStackTop; 3741 /* note that PC was already incremented above */ 3742 mach->CallStack[mach->CallStackTop].ReturnAddr = *pc; 3743 3744 mach->CallStackTop++; 3745 3746 /* Second, push the Cond, Loop, Cont, Func stacks */ 3747 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); 3748 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3749 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3750 assert(mach->SwitchStackTop < TGSI_EXEC_MAX_SWITCH_NESTING); 3751 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3752 assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING); 3753 3754 mach->CondStack[mach->CondStackTop++] = mach->CondMask; 3755 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; 3756 mach->ContStack[mach->ContStackTop++] = mach->ContMask; 3757 mach->SwitchStack[mach->SwitchStackTop++] = mach->Switch; 3758 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3759 mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask; 3760 3761 /* Finally, jump to the subroutine */ 3762 *pc = inst->Label.Label; 3763 } 3764 break; 3765 3766 case TGSI_OPCODE_RET: 3767 mach->FuncMask &= ~mach->ExecMask; 3768 UPDATE_EXEC_MASK(mach); 3769 3770 if (mach->FuncMask == 0x0) { 3771 /* really return now (otherwise, keep executing */ 3772 3773 if (mach->CallStackTop == 0) { 3774 /* returning from main() */ 3775 mach->CondStackTop = 0; 3776 mach->LoopStackTop = 0; 3777 *pc = -1; 3778 return; 3779 } 3780 3781 assert(mach->CallStackTop > 0); 3782 mach->CallStackTop--; 3783 3784 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; 3785 mach->CondMask = mach->CondStack[mach->CondStackTop]; 3786 3787 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; 3788 mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; 3789 3790 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; 3791 mach->ContMask = mach->ContStack[mach->ContStackTop]; 3792 3793 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop; 3794 mach->Switch = mach->SwitchStack[mach->SwitchStackTop]; 3795 3796 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop; 3797 mach->BreakType = mach->BreakStack[mach->BreakStackTop]; 3798 3799 assert(mach->FuncStackTop > 0); 3800 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; 3801 3802 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; 3803 3804 UPDATE_EXEC_MASK(mach); 3805 } 3806 break; 3807 3808 case TGSI_OPCODE_SSG: 3809 exec_vector_unary(mach, inst, micro_sgn, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3810 break; 3811 3812 case TGSI_OPCODE_CMP: 3813 exec_vector_trinary(mach, inst, micro_cmp, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3814 break; 3815 3816 case TGSI_OPCODE_SCS: 3817 exec_scs(mach, inst); 3818 break; 3819 3820 case TGSI_OPCODE_NRM: 3821 exec_nrm3(mach, inst); 3822 break; 3823 3824 case TGSI_OPCODE_NRM4: 3825 exec_nrm4(mach, inst); 3826 break; 3827 3828 case TGSI_OPCODE_DIV: 3829 exec_vector_binary(mach, inst, micro_div, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3830 break; 3831 3832 case TGSI_OPCODE_DP2: 3833 exec_dp2(mach, inst); 3834 break; 3835 3836 case TGSI_OPCODE_IF: 3837 /* push CondMask */ 3838 assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING); 3839 mach->CondStack[mach->CondStackTop++] = mach->CondMask; 3840 FETCH( &r[0], 0, CHAN_X ); 3841 /* update CondMask */ 3842 if( ! r[0].u[0] ) { 3843 mach->CondMask &= ~0x1; 3844 } 3845 if( ! r[0].u[1] ) { 3846 mach->CondMask &= ~0x2; 3847 } 3848 if( ! r[0].u[2] ) { 3849 mach->CondMask &= ~0x4; 3850 } 3851 if( ! r[0].u[3] ) { 3852 mach->CondMask &= ~0x8; 3853 } 3854 UPDATE_EXEC_MASK(mach); 3855 /* Todo: If CondMask==0, jump to ELSE */ 3856 break; 3857 3858 case TGSI_OPCODE_ELSE: 3859 /* invert CondMask wrt previous mask */ 3860 { 3861 uint prevMask; 3862 assert(mach->CondStackTop > 0); 3863 prevMask = mach->CondStack[mach->CondStackTop - 1]; 3864 mach->CondMask = ~mach->CondMask & prevMask; 3865 UPDATE_EXEC_MASK(mach); 3866 /* Todo: If CondMask==0, jump to ENDIF */ 3867 } 3868 break; 3869 3870 case TGSI_OPCODE_ENDIF: 3871 /* pop CondMask */ 3872 assert(mach->CondStackTop > 0); 3873 mach->CondMask = mach->CondStack[--mach->CondStackTop]; 3874 UPDATE_EXEC_MASK(mach); 3875 break; 3876 3877 case TGSI_OPCODE_END: 3878 /* make sure we end primitives which haven't 3879 * been explicitly emitted */ 3880 conditional_emit_primitive(mach); 3881 /* halt execution */ 3882 *pc = -1; 3883 break; 3884 3885 case TGSI_OPCODE_PUSHA: 3886 assert (0); 3887 break; 3888 3889 case TGSI_OPCODE_POPA: 3890 assert (0); 3891 break; 3892 3893 case TGSI_OPCODE_CEIL: 3894 exec_vector_unary(mach, inst, micro_ceil, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3895 break; 3896 3897 case TGSI_OPCODE_I2F: 3898 exec_vector_unary(mach, inst, micro_i2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_INT); 3899 break; 3900 3901 case TGSI_OPCODE_NOT: 3902 exec_vector_unary(mach, inst, micro_not, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3903 break; 3904 3905 case TGSI_OPCODE_TRUNC: 3906 exec_vector_unary(mach, inst, micro_trunc, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_FLOAT); 3907 break; 3908 3909 case TGSI_OPCODE_SHL: 3910 exec_vector_binary(mach, inst, micro_shl, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3911 break; 3912 3913 case TGSI_OPCODE_AND: 3914 exec_vector_binary(mach, inst, micro_and, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3915 break; 3916 3917 case TGSI_OPCODE_OR: 3918 exec_vector_binary(mach, inst, micro_or, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3919 break; 3920 3921 case TGSI_OPCODE_MOD: 3922 exec_vector_binary(mach, inst, micro_mod, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 3923 break; 3924 3925 case TGSI_OPCODE_XOR: 3926 exec_vector_binary(mach, inst, micro_xor, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 3927 break; 3928 3929 case TGSI_OPCODE_SAD: 3930 assert (0); 3931 break; 3932 3933 case TGSI_OPCODE_TXF: 3934 exec_txf(mach, inst); 3935 break; 3936 3937 case TGSI_OPCODE_TXQ: 3938 exec_txq(mach, inst); 3939 break; 3940 3941 case TGSI_OPCODE_EMIT: 3942 emit_vertex(mach); 3943 break; 3944 3945 case TGSI_OPCODE_ENDPRIM: 3946 emit_primitive(mach); 3947 break; 3948 3949 case TGSI_OPCODE_BGNLOOP: 3950 /* push LoopMask and ContMasks */ 3951 assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3952 assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3953 assert(mach->LoopLabelStackTop < TGSI_EXEC_MAX_LOOP_NESTING); 3954 assert(mach->BreakStackTop < TGSI_EXEC_MAX_BREAK_STACK); 3955 3956 mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask; 3957 mach->ContStack[mach->ContStackTop++] = mach->ContMask; 3958 mach->LoopLabelStack[mach->LoopLabelStackTop++] = *pc - 1; 3959 mach->BreakStack[mach->BreakStackTop++] = mach->BreakType; 3960 mach->BreakType = TGSI_EXEC_BREAK_INSIDE_LOOP; 3961 break; 3962 3963 case TGSI_OPCODE_ENDLOOP: 3964 /* Restore ContMask, but don't pop */ 3965 assert(mach->ContStackTop > 0); 3966 mach->ContMask = mach->ContStack[mach->ContStackTop - 1]; 3967 UPDATE_EXEC_MASK(mach); 3968 if (mach->ExecMask) { 3969 /* repeat loop: jump to instruction just past BGNLOOP */ 3970 assert(mach->LoopLabelStackTop > 0); 3971 *pc = mach->LoopLabelStack[mach->LoopLabelStackTop - 1] + 1; 3972 } 3973 else { 3974 /* exit loop: pop LoopMask */ 3975 assert(mach->LoopStackTop > 0); 3976 mach->LoopMask = mach->LoopStack[--mach->LoopStackTop]; 3977 /* pop ContMask */ 3978 assert(mach->ContStackTop > 0); 3979 mach->ContMask = mach->ContStack[--mach->ContStackTop]; 3980 assert(mach->LoopLabelStackTop > 0); 3981 --mach->LoopLabelStackTop; 3982 3983 mach->BreakType = mach->BreakStack[--mach->BreakStackTop]; 3984 } 3985 UPDATE_EXEC_MASK(mach); 3986 break; 3987 3988 case TGSI_OPCODE_BRK: 3989 exec_break(mach); 3990 break; 3991 3992 case TGSI_OPCODE_CONT: 3993 /* turn off cont channels for each enabled exec channel */ 3994 mach->ContMask &= ~mach->ExecMask; 3995 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 3996 UPDATE_EXEC_MASK(mach); 3997 break; 3998 3999 case TGSI_OPCODE_BGNSUB: 4000 /* no-op */ 4001 break; 4002 4003 case TGSI_OPCODE_ENDSUB: 4004 /* 4005 * XXX: This really should be a no-op. We should never reach this opcode. 4006 */ 4007 4008 assert(mach->CallStackTop > 0); 4009 mach->CallStackTop--; 4010 4011 mach->CondStackTop = mach->CallStack[mach->CallStackTop].CondStackTop; 4012 mach->CondMask = mach->CondStack[mach->CondStackTop]; 4013 4014 mach->LoopStackTop = mach->CallStack[mach->CallStackTop].LoopStackTop; 4015 mach->LoopMask = mach->LoopStack[mach->LoopStackTop]; 4016 4017 mach->ContStackTop = mach->CallStack[mach->CallStackTop].ContStackTop; 4018 mach->ContMask = mach->ContStack[mach->ContStackTop]; 4019 4020 mach->SwitchStackTop = mach->CallStack[mach->CallStackTop].SwitchStackTop; 4021 mach->Switch = mach->SwitchStack[mach->SwitchStackTop]; 4022 4023 mach->BreakStackTop = mach->CallStack[mach->CallStackTop].BreakStackTop; 4024 mach->BreakType = mach->BreakStack[mach->BreakStackTop]; 4025 4026 assert(mach->FuncStackTop > 0); 4027 mach->FuncMask = mach->FuncStack[--mach->FuncStackTop]; 4028 4029 *pc = mach->CallStack[mach->CallStackTop].ReturnAddr; 4030 4031 UPDATE_EXEC_MASK(mach); 4032 break; 4033 4034 case TGSI_OPCODE_NOP: 4035 break; 4036 4037 case TGSI_OPCODE_BREAKC: 4038 FETCH(&r[0], 0, CHAN_X); 4039 /* update CondMask */ 4040 if (r[0].u[0] && (mach->ExecMask & 0x1)) { 4041 mach->LoopMask &= ~0x1; 4042 } 4043 if (r[0].u[1] && (mach->ExecMask & 0x2)) { 4044 mach->LoopMask &= ~0x2; 4045 } 4046 if (r[0].u[2] && (mach->ExecMask & 0x4)) { 4047 mach->LoopMask &= ~0x4; 4048 } 4049 if (r[0].u[3] && (mach->ExecMask & 0x8)) { 4050 mach->LoopMask &= ~0x8; 4051 } 4052 /* Todo: if mach->LoopMask == 0, jump to end of loop */ 4053 UPDATE_EXEC_MASK(mach); 4054 break; 4055 4056 case TGSI_OPCODE_F2I: 4057 exec_vector_unary(mach, inst, micro_f2i, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_FLOAT); 4058 break; 4059 4060 case TGSI_OPCODE_IDIV: 4061 exec_vector_binary(mach, inst, micro_idiv, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4062 break; 4063 4064 case TGSI_OPCODE_IMAX: 4065 exec_vector_binary(mach, inst, micro_imax, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4066 break; 4067 4068 case TGSI_OPCODE_IMIN: 4069 exec_vector_binary(mach, inst, micro_imin, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4070 break; 4071 4072 case TGSI_OPCODE_INEG: 4073 exec_vector_unary(mach, inst, micro_ineg, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4074 break; 4075 4076 case TGSI_OPCODE_ISGE: 4077 exec_vector_binary(mach, inst, micro_isge, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4078 break; 4079 4080 case TGSI_OPCODE_ISHR: 4081 exec_vector_binary(mach, inst, micro_ishr, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4082 break; 4083 4084 case TGSI_OPCODE_ISLT: 4085 exec_vector_binary(mach, inst, micro_islt, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4086 break; 4087 4088 case TGSI_OPCODE_F2U: 4089 exec_vector_unary(mach, inst, micro_f2u, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_FLOAT); 4090 break; 4091 4092 case TGSI_OPCODE_U2F: 4093 exec_vector_unary(mach, inst, micro_u2f, TGSI_EXEC_DATA_FLOAT, TGSI_EXEC_DATA_UINT); 4094 break; 4095 4096 case TGSI_OPCODE_UADD: 4097 exec_vector_binary(mach, inst, micro_uadd, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4098 break; 4099 4100 case TGSI_OPCODE_UDIV: 4101 exec_vector_binary(mach, inst, micro_udiv, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4102 break; 4103 4104 case TGSI_OPCODE_UMAD: 4105 exec_vector_trinary(mach, inst, micro_umad, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4106 break; 4107 4108 case TGSI_OPCODE_UMAX: 4109 exec_vector_binary(mach, inst, micro_umax, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4110 break; 4111 4112 case TGSI_OPCODE_UMIN: 4113 exec_vector_binary(mach, inst, micro_umin, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4114 break; 4115 4116 case TGSI_OPCODE_UMOD: 4117 exec_vector_binary(mach, inst, micro_umod, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4118 break; 4119 4120 case TGSI_OPCODE_UMUL: 4121 exec_vector_binary(mach, inst, micro_umul, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4122 break; 4123 4124 case TGSI_OPCODE_USEQ: 4125 exec_vector_binary(mach, inst, micro_useq, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4126 break; 4127 4128 case TGSI_OPCODE_USGE: 4129 exec_vector_binary(mach, inst, micro_usge, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4130 break; 4131 4132 case TGSI_OPCODE_USHR: 4133 exec_vector_binary(mach, inst, micro_ushr, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4134 break; 4135 4136 case TGSI_OPCODE_USLT: 4137 exec_vector_binary(mach, inst, micro_uslt, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4138 break; 4139 4140 case TGSI_OPCODE_USNE: 4141 exec_vector_binary(mach, inst, micro_usne, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4142 break; 4143 4144 case TGSI_OPCODE_SWITCH: 4145 exec_switch(mach, inst); 4146 break; 4147 4148 case TGSI_OPCODE_CASE: 4149 exec_case(mach, inst); 4150 break; 4151 4152 case TGSI_OPCODE_DEFAULT: 4153 exec_default(mach); 4154 break; 4155 4156 case TGSI_OPCODE_ENDSWITCH: 4157 exec_endswitch(mach); 4158 break; 4159 4160 case TGSI_OPCODE_LOAD: 4161 assert(0); 4162 break; 4163 4164 case TGSI_OPCODE_LOAD_MS: 4165 assert(0); 4166 break; 4167 4168 case TGSI_OPCODE_SAMPLE: 4169 exec_sample(mach, inst, TEX_MODIFIER_NONE); 4170 break; 4171 4172 case TGSI_OPCODE_SAMPLE_B: 4173 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS); 4174 break; 4175 4176 case TGSI_OPCODE_SAMPLE_C: 4177 exec_sample(mach, inst, TEX_MODIFIER_NONE); 4178 break; 4179 4180 case TGSI_OPCODE_SAMPLE_C_LZ: 4181 exec_sample(mach, inst, TEX_MODIFIER_LOD_BIAS); 4182 break; 4183 4184 case TGSI_OPCODE_SAMPLE_D: 4185 exec_sample_d(mach, inst); 4186 break; 4187 4188 case TGSI_OPCODE_SAMPLE_L: 4189 exec_sample(mach, inst, TEX_MODIFIER_EXPLICIT_LOD); 4190 break; 4191 4192 case TGSI_OPCODE_GATHER4: 4193 assert(0); 4194 break; 4195 4196 case TGSI_OPCODE_RESINFO: 4197 assert(0); 4198 break; 4199 4200 case TGSI_OPCODE_SAMPLE_POS: 4201 assert(0); 4202 break; 4203 4204 case TGSI_OPCODE_SAMPLE_INFO: 4205 assert(0); 4206 break; 4207 4208 case TGSI_OPCODE_UARL: 4209 exec_vector_unary(mach, inst, micro_uarl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_UINT); 4210 break; 4211 4212 case TGSI_OPCODE_UCMP: 4213 exec_vector_trinary(mach, inst, micro_ucmp, TGSI_EXEC_DATA_UINT, TGSI_EXEC_DATA_UINT); 4214 break; 4215 4216 case TGSI_OPCODE_IABS: 4217 exec_vector_unary(mach, inst, micro_iabs, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4218 break; 4219 4220 case TGSI_OPCODE_ISSG: 4221 exec_vector_unary(mach, inst, micro_isgn, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_INT); 4222 break; 4223 4224 default: 4225 assert( 0 ); 4226 } 4227} 4228 4229 4230#define DEBUG_EXECUTION 0 4231 4232 4233/** 4234 * Run TGSI interpreter. 4235 * \return bitmask of "alive" quad components 4236 */ 4237uint 4238tgsi_exec_machine_run( struct tgsi_exec_machine *mach ) 4239{ 4240 uint i; 4241 int pc = 0; 4242 4243 mach->CondMask = 0xf; 4244 mach->LoopMask = 0xf; 4245 mach->ContMask = 0xf; 4246 mach->FuncMask = 0xf; 4247 mach->ExecMask = 0xf; 4248 4249 mach->Switch.mask = 0xf; 4250 4251 assert(mach->CondStackTop == 0); 4252 assert(mach->LoopStackTop == 0); 4253 assert(mach->ContStackTop == 0); 4254 assert(mach->SwitchStackTop == 0); 4255 assert(mach->BreakStackTop == 0); 4256 assert(mach->CallStackTop == 0); 4257 4258 mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0; 4259 mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0; 4260 4261 if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) { 4262 mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0; 4263 mach->Primitives[0] = 0; 4264 } 4265 4266 /* execute declarations (interpolants) */ 4267 for (i = 0; i < mach->NumDeclarations; i++) { 4268 exec_declaration( mach, mach->Declarations+i ); 4269 } 4270 4271 { 4272#if DEBUG_EXECUTION 4273 struct tgsi_exec_vector temps[TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS]; 4274 struct tgsi_exec_vector outputs[PIPE_MAX_ATTRIBS]; 4275 uint inst = 1; 4276 4277 memcpy(temps, mach->Temps, sizeof(temps)); 4278 memcpy(outputs, mach->Outputs, sizeof(outputs)); 4279#endif 4280 4281 /* execute instructions, until pc is set to -1 */ 4282 while (pc != -1) { 4283 4284#if DEBUG_EXECUTION 4285 uint i; 4286 4287 tgsi_dump_instruction(&mach->Instructions[pc], inst++); 4288#endif 4289 4290 assert(pc < (int) mach->NumInstructions); 4291 exec_instruction(mach, mach->Instructions + pc, &pc); 4292 4293#if DEBUG_EXECUTION 4294 for (i = 0; i < TGSI_EXEC_NUM_TEMPS + TGSI_EXEC_NUM_TEMP_EXTRAS; i++) { 4295 if (memcmp(&temps[i], &mach->Temps[i], sizeof(temps[i]))) { 4296 uint j; 4297 4298 memcpy(&temps[i], &mach->Temps[i], sizeof(temps[i])); 4299 debug_printf("TEMP[%2u] = ", i); 4300 for (j = 0; j < 4; j++) { 4301 if (j > 0) { 4302 debug_printf(" "); 4303 } 4304 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n", 4305 temps[i].xyzw[0].f[j], temps[i].xyzw[0].u[j], 4306 temps[i].xyzw[1].f[j], temps[i].xyzw[1].u[j], 4307 temps[i].xyzw[2].f[j], temps[i].xyzw[2].u[j], 4308 temps[i].xyzw[3].f[j], temps[i].xyzw[3].u[j]); 4309 } 4310 } 4311 } 4312 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) { 4313 if (memcmp(&outputs[i], &mach->Outputs[i], sizeof(outputs[i]))) { 4314 uint j; 4315 4316 memcpy(&outputs[i], &mach->Outputs[i], sizeof(outputs[i])); 4317 debug_printf("OUT[%2u] = ", i); 4318 for (j = 0; j < 4; j++) { 4319 if (j > 0) { 4320 debug_printf(" "); 4321 } 4322 debug_printf("(%6f %u, %6f %u, %6f %u, %6f %u)\n", 4323 outputs[i].xyzw[0].f[j], outputs[i].xyzw[0].u[j], 4324 outputs[i].xyzw[1].f[j], outputs[i].xyzw[1].u[j], 4325 outputs[i].xyzw[2].f[j], outputs[i].xyzw[2].u[j], 4326 outputs[i].xyzw[3].f[j], outputs[i].xyzw[3].u[j]); 4327 } 4328 } 4329 } 4330#endif 4331 } 4332 } 4333 4334#if 0 4335 /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */ 4336 if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) { 4337 /* 4338 * Scale back depth component. 4339 */ 4340 for (i = 0; i < 4; i++) 4341 mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF; 4342 } 4343#endif 4344 4345 /* Strictly speaking, these assertions aren't really needed but they 4346 * can potentially catch some bugs in the control flow code. 4347 */ 4348 assert(mach->CondStackTop == 0); 4349 assert(mach->LoopStackTop == 0); 4350 assert(mach->ContStackTop == 0); 4351 assert(mach->SwitchStackTop == 0); 4352 assert(mach->BreakStackTop == 0); 4353 assert(mach->CallStackTop == 0); 4354 4355 return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0]; 4356} 4357