1#include <strings.h> 2#include "pipe/p_context.h" 3#include "pipe/p_defines.h" 4#include "pipe/p_state.h" 5#include "util/u_dynarray.h" 6#include "util/u_debug.h" 7 8#include "pipe/p_shader_tokens.h" 9#include "tgsi/tgsi_parse.h" 10#include "tgsi/tgsi_dump.h" 11#include "tgsi/tgsi_util.h" 12#include "tgsi/tgsi_ureg.h" 13 14#include "draw/draw_context.h" 15 16#include "nv_object.xml.h" 17#include "nouveau_debug.h" 18#include "nv30/nv30-40_3d.xml.h" 19#include "nv30/nv30_state.h" 20 21/* TODO (at least...): 22 * 1. Indexed consts + ARL 23 * 3. NV_vp11, NV_vp2, NV_vp3 features 24 * - extra arith opcodes 25 * - branching 26 * - texture sampling 27 * - indexed attribs 28 * - indexed results 29 * 4. bugs 30 */ 31 32#include "nv30/nv30_vertprog.h" 33#include "nv30/nv40_vertprog.h" 34 35struct nvfx_loop_entry { 36 unsigned brk_target; 37 unsigned cont_target; 38}; 39 40struct nvfx_vpc { 41 struct pipe_shader_state pipe; 42 struct nv30_vertprog *vp; 43 struct tgsi_shader_info* info; 44 45 struct nv30_vertprog_exec *vpi; 46 47 unsigned r_temps; 48 unsigned r_temps_discard; 49 struct nvfx_reg r_result[PIPE_MAX_SHADER_OUTPUTS]; 50 struct nvfx_reg *r_address; 51 struct nvfx_reg *r_temp; 52 struct nvfx_reg *r_const; 53 struct nvfx_reg r_0_1; 54 55 struct nvfx_reg *imm; 56 unsigned nr_imm; 57 58 int hpos_idx; 59 int cvtx_idx; 60 61 unsigned is_nv4x; 62 63 struct util_dynarray label_relocs; 64 struct util_dynarray loop_stack; 65}; 66 67static struct nvfx_reg 68temp(struct nvfx_vpc *vpc) 69{ 70 int idx = ffs(~vpc->r_temps) - 1; 71 72 if (idx < 0 || (!vpc->is_nv4x && idx >= 16)) { 73 NOUVEAU_ERR("out of temps!!\n"); 74 return nvfx_reg(NVFXSR_TEMP, 0); 75 } 76 77 vpc->r_temps |= (1 << idx); 78 vpc->r_temps_discard |= (1 << idx); 79 return nvfx_reg(NVFXSR_TEMP, idx); 80} 81 82static inline void 83release_temps(struct nvfx_vpc *vpc) 84{ 85 vpc->r_temps &= ~vpc->r_temps_discard; 86 vpc->r_temps_discard = 0; 87} 88 89static struct nvfx_reg 90constant(struct nvfx_vpc *vpc, int pipe, float x, float y, float z, float w) 91{ 92 struct nv30_vertprog *vp = vpc->vp; 93 struct nv30_vertprog_data *vpd; 94 int idx; 95 96 if (pipe >= 0) { 97 for (idx = 0; idx < vp->nr_consts; idx++) { 98 if (vp->consts[idx].index == pipe) 99 return nvfx_reg(NVFXSR_CONST, idx); 100 } 101 } 102 103 idx = vp->nr_consts++; 104 vp->consts = realloc(vp->consts, sizeof(*vpd) * vp->nr_consts); 105 vpd = &vp->consts[idx]; 106 107 vpd->index = pipe; 108 vpd->value[0] = x; 109 vpd->value[1] = y; 110 vpd->value[2] = z; 111 vpd->value[3] = w; 112 return nvfx_reg(NVFXSR_CONST, idx); 113} 114 115#define arith(s,t,o,d,m,s0,s1,s2) \ 116 nvfx_insn((s), (NVFX_VP_INST_SLOT_##t << 7) | NVFX_VP_INST_##t##_OP_##o, -1, (d), (m), (s0), (s1), (s2)) 117 118static void 119emit_src(struct nvfx_vpc *vpc, uint32_t *hw, 120 int pos, struct nvfx_src src) 121{ 122 struct nv30_vertprog *vp = vpc->vp; 123 uint32_t sr = 0; 124 struct nvfx_relocation reloc; 125 126 switch (src.reg.type) { 127 case NVFXSR_TEMP: 128 sr |= (NVFX_VP(SRC_REG_TYPE_TEMP) << NVFX_VP(SRC_REG_TYPE_SHIFT)); 129 sr |= (src.reg.index << NVFX_VP(SRC_TEMP_SRC_SHIFT)); 130 break; 131 case NVFXSR_INPUT: 132 sr |= (NVFX_VP(SRC_REG_TYPE_INPUT) << 133 NVFX_VP(SRC_REG_TYPE_SHIFT)); 134 vp->ir |= (1 << src.reg.index); 135 hw[1] |= (src.reg.index << NVFX_VP(INST_INPUT_SRC_SHIFT)); 136 break; 137 case NVFXSR_CONST: 138 sr |= (NVFX_VP(SRC_REG_TYPE_CONST) << 139 NVFX_VP(SRC_REG_TYPE_SHIFT)); 140 if (src.reg.index < 256 && src.reg.index >= -256) { 141 reloc.location = vp->nr_insns - 1; 142 reloc.target = src.reg.index; 143 util_dynarray_append(&vp->const_relocs, struct nvfx_relocation, reloc); 144 } else { 145 hw[1] |= (src.reg.index << NVFX_VP(INST_CONST_SRC_SHIFT)) & 146 NVFX_VP(INST_CONST_SRC_MASK); 147 } 148 break; 149 case NVFXSR_NONE: 150 sr |= (NVFX_VP(SRC_REG_TYPE_INPUT) << 151 NVFX_VP(SRC_REG_TYPE_SHIFT)); 152 break; 153 default: 154 assert(0); 155 } 156 157 if (src.negate) 158 sr |= NVFX_VP(SRC_NEGATE); 159 160 if (src.abs) 161 hw[0] |= (1 << (21 + pos)); 162 163 sr |= ((src.swz[0] << NVFX_VP(SRC_SWZ_X_SHIFT)) | 164 (src.swz[1] << NVFX_VP(SRC_SWZ_Y_SHIFT)) | 165 (src.swz[2] << NVFX_VP(SRC_SWZ_Z_SHIFT)) | 166 (src.swz[3] << NVFX_VP(SRC_SWZ_W_SHIFT))); 167 168 if(src.indirect) { 169 if(src.reg.type == NVFXSR_CONST) 170 hw[3] |= NVFX_VP(INST_INDEX_CONST); 171 else if(src.reg.type == NVFXSR_INPUT) 172 hw[0] |= NVFX_VP(INST_INDEX_INPUT); 173 else 174 assert(0); 175 176 if(src.indirect_reg) 177 hw[0] |= NVFX_VP(INST_ADDR_REG_SELECT_1); 178 hw[0] |= src.indirect_swz << NVFX_VP(INST_ADDR_SWZ_SHIFT); 179 } 180 181 switch (pos) { 182 case 0: 183 hw[1] |= ((sr & NVFX_VP(SRC0_HIGH_MASK)) >> 184 NVFX_VP(SRC0_HIGH_SHIFT)) << NVFX_VP(INST_SRC0H_SHIFT); 185 hw[2] |= (sr & NVFX_VP(SRC0_LOW_MASK)) << 186 NVFX_VP(INST_SRC0L_SHIFT); 187 break; 188 case 1: 189 hw[2] |= sr << NVFX_VP(INST_SRC1_SHIFT); 190 break; 191 case 2: 192 hw[2] |= ((sr & NVFX_VP(SRC2_HIGH_MASK)) >> 193 NVFX_VP(SRC2_HIGH_SHIFT)) << NVFX_VP(INST_SRC2H_SHIFT); 194 hw[3] |= (sr & NVFX_VP(SRC2_LOW_MASK)) << 195 NVFX_VP(INST_SRC2L_SHIFT); 196 break; 197 default: 198 assert(0); 199 } 200} 201 202static void 203emit_dst(struct nvfx_vpc *vpc, uint32_t *hw, 204 int slot, struct nvfx_reg dst) 205{ 206 struct nv30_vertprog *vp = vpc->vp; 207 208 switch (dst.type) { 209 case NVFXSR_NONE: 210 if(!vpc->is_nv4x) 211 hw[0] |= NV30_VP_INST_DEST_TEMP_ID_MASK; 212 else { 213 hw[3] |= NV40_VP_INST_DEST_MASK; 214 if (slot == 0) 215 hw[0] |= NV40_VP_INST_VEC_DEST_TEMP_MASK; 216 else 217 hw[3] |= NV40_VP_INST_SCA_DEST_TEMP_MASK; 218 } 219 break; 220 case NVFXSR_TEMP: 221 if(!vpc->is_nv4x) 222 hw[0] |= (dst.index << NV30_VP_INST_DEST_TEMP_ID_SHIFT); 223 else { 224 hw[3] |= NV40_VP_INST_DEST_MASK; 225 if (slot == 0) 226 hw[0] |= (dst.index << NV40_VP_INST_VEC_DEST_TEMP_SHIFT); 227 else 228 hw[3] |= (dst.index << NV40_VP_INST_SCA_DEST_TEMP_SHIFT); 229 } 230 break; 231 case NVFXSR_OUTPUT: 232 /* TODO: this may be wrong because on nv30 COL0 and BFC0 are swapped */ 233 if(vpc->is_nv4x) { 234 switch (dst.index) { 235 case NV30_VP_INST_DEST_CLP(0): 236 dst.index = NVFX_VP(INST_DEST_FOGC); 237 vp->or |= (1 << 6); 238 break; 239 case NV30_VP_INST_DEST_CLP(1): 240 dst.index = NVFX_VP(INST_DEST_FOGC); 241 vp->or |= (1 << 7); 242 break; 243 case NV30_VP_INST_DEST_CLP(2): 244 dst.index = NVFX_VP(INST_DEST_FOGC); 245 vp->or |= (1 << 8); 246 break; 247 case NV30_VP_INST_DEST_CLP(3): 248 dst.index = NVFX_VP(INST_DEST_PSZ); 249 vp->or |= (1 << 9); 250 break; 251 case NV30_VP_INST_DEST_CLP(4): 252 dst.index = NVFX_VP(INST_DEST_PSZ); 253 vp->or |= (1 << 10); 254 break; 255 case NV30_VP_INST_DEST_CLP(5): 256 dst.index = NVFX_VP(INST_DEST_PSZ); 257 vp->or |= (1 << 11); 258 break; 259 case NV40_VP_INST_DEST_COL0: vp->or |= (1 << 0); break; 260 case NV40_VP_INST_DEST_COL1: vp->or |= (1 << 1); break; 261 case NV40_VP_INST_DEST_BFC0: vp->or |= (1 << 2); break; 262 case NV40_VP_INST_DEST_BFC1: vp->or |= (1 << 3); break; 263 case NV40_VP_INST_DEST_FOGC: vp->or |= (1 << 4); break; 264 case NV40_VP_INST_DEST_PSZ : vp->or |= (1 << 5); break; 265 } 266 } 267 268 if(!vpc->is_nv4x) { 269 hw[3] |= (dst.index << NV30_VP_INST_DEST_SHIFT); 270 hw[0] |= NV30_VP_INST_VEC_DEST_TEMP_MASK; 271 272 /*XXX: no way this is entirely correct, someone needs to 273 * figure out what exactly it is. 274 */ 275 hw[3] |= 0x800; 276 } else { 277 hw[3] |= (dst.index << NV40_VP_INST_DEST_SHIFT); 278 if (slot == 0) { 279 hw[0] |= NV40_VP_INST_VEC_RESULT; 280 hw[0] |= NV40_VP_INST_VEC_DEST_TEMP_MASK; 281 } else { 282 hw[3] |= NV40_VP_INST_SCA_RESULT; 283 hw[3] |= NV40_VP_INST_SCA_DEST_TEMP_MASK; 284 } 285 } 286 break; 287 default: 288 assert(0); 289 } 290} 291 292static void 293nvfx_vp_emit(struct nvfx_vpc *vpc, struct nvfx_insn insn) 294{ 295 struct nv30_vertprog *vp = vpc->vp; 296 unsigned slot = insn.op >> 7; 297 unsigned op = insn.op & 0x7f; 298 uint32_t *hw; 299 300 vp->insns = realloc(vp->insns, ++vp->nr_insns * sizeof(*vpc->vpi)); 301 vpc->vpi = &vp->insns[vp->nr_insns - 1]; 302 memset(vpc->vpi, 0, sizeof(*vpc->vpi)); 303 304 hw = vpc->vpi->data; 305 306 if (insn.cc_test != NVFX_COND_TR) 307 hw[0] |= NVFX_VP(INST_COND_TEST_ENABLE); 308 hw[0] |= (insn.cc_test << NVFX_VP(INST_COND_SHIFT)); 309 hw[0] |= ((insn.cc_swz[0] << NVFX_VP(INST_COND_SWZ_X_SHIFT)) | 310 (insn.cc_swz[1] << NVFX_VP(INST_COND_SWZ_Y_SHIFT)) | 311 (insn.cc_swz[2] << NVFX_VP(INST_COND_SWZ_Z_SHIFT)) | 312 (insn.cc_swz[3] << NVFX_VP(INST_COND_SWZ_W_SHIFT))); 313 if(insn.cc_update) 314 hw[0] |= NVFX_VP(INST_COND_UPDATE_ENABLE); 315 316 if(insn.sat) { 317 assert(vpc->is_nv4x); 318 if(vpc->is_nv4x) 319 hw[0] |= NV40_VP_INST_SATURATE; 320 } 321 322 if(!vpc->is_nv4x) { 323 if(slot == 0) 324 hw[1] |= (op << NV30_VP_INST_VEC_OPCODE_SHIFT); 325 else { 326 hw[0] |= ((op >> 4) << NV30_VP_INST_SCA_OPCODEH_SHIFT); 327 hw[1] |= ((op & 0xf) << NV30_VP_INST_SCA_OPCODEL_SHIFT); 328 } 329// hw[3] |= NVFX_VP(INST_SCA_DEST_TEMP_MASK); 330// hw[3] |= (mask << NVFX_VP(INST_VEC_WRITEMASK_SHIFT)); 331 332 if (insn.dst.type == NVFXSR_OUTPUT) { 333 if (slot) 334 hw[3] |= (insn.mask << NV30_VP_INST_SDEST_WRITEMASK_SHIFT); 335 else 336 hw[3] |= (insn.mask << NV30_VP_INST_VDEST_WRITEMASK_SHIFT); 337 } else { 338 if (slot) 339 hw[3] |= (insn.mask << NV30_VP_INST_STEMP_WRITEMASK_SHIFT); 340 else 341 hw[3] |= (insn.mask << NV30_VP_INST_VTEMP_WRITEMASK_SHIFT); 342 } 343 } else { 344 if (slot == 0) { 345 hw[1] |= (op << NV40_VP_INST_VEC_OPCODE_SHIFT); 346 hw[3] |= NV40_VP_INST_SCA_DEST_TEMP_MASK; 347 hw[3] |= (insn.mask << NV40_VP_INST_VEC_WRITEMASK_SHIFT); 348 } else { 349 hw[1] |= (op << NV40_VP_INST_SCA_OPCODE_SHIFT); 350 hw[0] |= NV40_VP_INST_VEC_DEST_TEMP_MASK ; 351 hw[3] |= (insn.mask << NV40_VP_INST_SCA_WRITEMASK_SHIFT); 352 } 353 } 354 355 emit_dst(vpc, hw, slot, insn.dst); 356 emit_src(vpc, hw, 0, insn.src[0]); 357 emit_src(vpc, hw, 1, insn.src[1]); 358 emit_src(vpc, hw, 2, insn.src[2]); 359 360// if(insn.src[0].indirect || op == NVFX_VP_INST_VEC_OP_ARL) 361// hw[3] |= NV40_VP_INST_SCA_RESULT; 362} 363 364static inline struct nvfx_src 365tgsi_src(struct nvfx_vpc *vpc, const struct tgsi_full_src_register *fsrc) { 366 struct nvfx_src src; 367 368 switch (fsrc->Register.File) { 369 case TGSI_FILE_INPUT: 370 src.reg = nvfx_reg(NVFXSR_INPUT, fsrc->Register.Index); 371 break; 372 case TGSI_FILE_CONSTANT: 373 if(fsrc->Register.Indirect) { 374 src.reg = vpc->r_const[0]; 375 src.reg.index = fsrc->Register.Index; 376 } else { 377 src.reg = vpc->r_const[fsrc->Register.Index]; 378 } 379 break; 380 case TGSI_FILE_IMMEDIATE: 381 src.reg = vpc->imm[fsrc->Register.Index]; 382 break; 383 case TGSI_FILE_TEMPORARY: 384 src.reg = vpc->r_temp[fsrc->Register.Index]; 385 break; 386 default: 387 NOUVEAU_ERR("bad src file\n"); 388 src.reg.index = 0; 389 src.reg.type = -1; 390 break; 391 } 392 393 src.abs = fsrc->Register.Absolute; 394 src.negate = fsrc->Register.Negate; 395 src.swz[0] = fsrc->Register.SwizzleX; 396 src.swz[1] = fsrc->Register.SwizzleY; 397 src.swz[2] = fsrc->Register.SwizzleZ; 398 src.swz[3] = fsrc->Register.SwizzleW; 399 src.indirect = 0; 400 src.indirect_reg = 0; 401 src.indirect_swz = 0; 402 403 if(fsrc->Register.Indirect) { 404 if(fsrc->Indirect.File == TGSI_FILE_ADDRESS && 405 (fsrc->Register.File == TGSI_FILE_CONSTANT || 406 fsrc->Register.File == TGSI_FILE_INPUT)) { 407 src.indirect = 1; 408 src.indirect_reg = fsrc->Indirect.Index; 409 src.indirect_swz = fsrc->Indirect.Swizzle; 410 } else { 411 src.reg.index = 0; 412 src.reg.type = -1; 413 } 414 } 415 416 return src; 417} 418 419static inline struct nvfx_reg 420tgsi_dst(struct nvfx_vpc *vpc, const struct tgsi_full_dst_register *fdst) { 421 struct nvfx_reg dst; 422 423 switch (fdst->Register.File) { 424 case TGSI_FILE_NULL: 425 dst = nvfx_reg(NVFXSR_NONE, 0); 426 break; 427 case TGSI_FILE_OUTPUT: 428 dst = vpc->r_result[fdst->Register.Index]; 429 break; 430 case TGSI_FILE_TEMPORARY: 431 dst = vpc->r_temp[fdst->Register.Index]; 432 break; 433 case TGSI_FILE_ADDRESS: 434 dst = vpc->r_address[fdst->Register.Index]; 435 break; 436 default: 437 NOUVEAU_ERR("bad dst file %i\n", fdst->Register.File); 438 dst.index = 0; 439 dst.type = 0; 440 break; 441 } 442 443 return dst; 444} 445 446static inline int 447tgsi_mask(uint tgsi) 448{ 449 int mask = 0; 450 451 if (tgsi & TGSI_WRITEMASK_X) mask |= NVFX_VP_MASK_X; 452 if (tgsi & TGSI_WRITEMASK_Y) mask |= NVFX_VP_MASK_Y; 453 if (tgsi & TGSI_WRITEMASK_Z) mask |= NVFX_VP_MASK_Z; 454 if (tgsi & TGSI_WRITEMASK_W) mask |= NVFX_VP_MASK_W; 455 return mask; 456} 457 458static bool 459nvfx_vertprog_parse_instruction(struct nvfx_vpc *vpc, 460 unsigned idx, const struct tgsi_full_instruction *finst) 461{ 462 struct nvfx_src src[3], tmp; 463 struct nvfx_reg dst; 464 struct nvfx_reg final_dst; 465 struct nvfx_src none = nvfx_src(nvfx_reg(NVFXSR_NONE, 0)); 466 struct nvfx_insn insn; 467 struct nvfx_relocation reloc; 468 struct nvfx_loop_entry loop; 469 bool sat = false; 470 int mask; 471 int ai = -1, ci = -1, ii = -1; 472 int i; 473 unsigned sub_depth = 0; 474 475 for (i = 0; i < finst->Instruction.NumSrcRegs; i++) { 476 const struct tgsi_full_src_register *fsrc; 477 478 fsrc = &finst->Src[i]; 479 if (fsrc->Register.File == TGSI_FILE_TEMPORARY) { 480 src[i] = tgsi_src(vpc, fsrc); 481 } 482 } 483 484 for (i = 0; i < finst->Instruction.NumSrcRegs; i++) { 485 const struct tgsi_full_src_register *fsrc; 486 487 fsrc = &finst->Src[i]; 488 489 switch (fsrc->Register.File) { 490 case TGSI_FILE_INPUT: 491 if (ai == -1 || ai == fsrc->Register.Index) { 492 ai = fsrc->Register.Index; 493 src[i] = tgsi_src(vpc, fsrc); 494 } else { 495 src[i] = nvfx_src(temp(vpc)); 496 nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, 497 tgsi_src(vpc, fsrc), none, none)); 498 } 499 break; 500 case TGSI_FILE_CONSTANT: 501 if ((ci == -1 && ii == -1) || 502 ci == fsrc->Register.Index) { 503 ci = fsrc->Register.Index; 504 src[i] = tgsi_src(vpc, fsrc); 505 } else { 506 src[i] = nvfx_src(temp(vpc)); 507 nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, 508 tgsi_src(vpc, fsrc), none, none)); 509 } 510 break; 511 case TGSI_FILE_IMMEDIATE: 512 if ((ci == -1 && ii == -1) || 513 ii == fsrc->Register.Index) { 514 ii = fsrc->Register.Index; 515 src[i] = tgsi_src(vpc, fsrc); 516 } else { 517 src[i] = nvfx_src(temp(vpc)); 518 nvfx_vp_emit(vpc, arith(0, VEC, MOV, src[i].reg, NVFX_VP_MASK_ALL, 519 tgsi_src(vpc, fsrc), none, none)); 520 } 521 break; 522 case TGSI_FILE_TEMPORARY: 523 /* handled above */ 524 break; 525 default: 526 NOUVEAU_ERR("bad src file\n"); 527 return false; 528 } 529 } 530 531 for (i = 0; i < finst->Instruction.NumSrcRegs; i++) { 532 if(src[i].reg.type < 0) 533 return false; 534 } 535 536 if(finst->Dst[0].Register.File == TGSI_FILE_ADDRESS && 537 finst->Instruction.Opcode != TGSI_OPCODE_ARL) 538 return false; 539 540 final_dst = dst = tgsi_dst(vpc, &finst->Dst[0]); 541 mask = tgsi_mask(finst->Dst[0].Register.WriteMask); 542 if(finst->Instruction.Saturate) { 543 assert(finst->Instruction.Opcode != TGSI_OPCODE_ARL); 544 if (vpc->is_nv4x) 545 sat = true; 546 else 547 if(dst.type != NVFXSR_TEMP) 548 dst = temp(vpc); 549 } 550 551 switch (finst->Instruction.Opcode) { 552 case TGSI_OPCODE_ADD: 553 nvfx_vp_emit(vpc, arith(sat, VEC, ADD, dst, mask, src[0], none, src[1])); 554 break; 555 case TGSI_OPCODE_ARL: 556 nvfx_vp_emit(vpc, arith(0, VEC, ARL, dst, mask, src[0], none, none)); 557 break; 558 case TGSI_OPCODE_CEIL: 559 tmp = nvfx_src(temp(vpc)); 560 nvfx_vp_emit(vpc, arith(0, VEC, FLR, tmp.reg, mask, neg(src[0]), none, none)); 561 nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, neg(tmp), none, none)); 562 break; 563 case TGSI_OPCODE_CMP: 564 insn = arith(0, VEC, MOV, none.reg, mask, src[0], none, none); 565 insn.cc_update = 1; 566 nvfx_vp_emit(vpc, insn); 567 568 insn = arith(sat, VEC, MOV, dst, mask, src[2], none, none); 569 insn.cc_test = NVFX_COND_GE; 570 nvfx_vp_emit(vpc, insn); 571 572 insn = arith(sat, VEC, MOV, dst, mask, src[1], none, none); 573 insn.cc_test = NVFX_COND_LT; 574 nvfx_vp_emit(vpc, insn); 575 break; 576 case TGSI_OPCODE_COS: 577 nvfx_vp_emit(vpc, arith(sat, SCA, COS, dst, mask, none, none, src[0])); 578 break; 579 case TGSI_OPCODE_DP2: 580 tmp = nvfx_src(temp(vpc)); 581 nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, NVFX_VP_MASK_X | NVFX_VP_MASK_Y, src[0], src[1], none)); 582 nvfx_vp_emit(vpc, arith(sat, VEC, ADD, dst, mask, swz(tmp, X, X, X, X), none, swz(tmp, Y, Y, Y, Y))); 583 break; 584 case TGSI_OPCODE_DP3: 585 nvfx_vp_emit(vpc, arith(sat, VEC, DP3, dst, mask, src[0], src[1], none)); 586 break; 587 case TGSI_OPCODE_DP4: 588 nvfx_vp_emit(vpc, arith(sat, VEC, DP4, dst, mask, src[0], src[1], none)); 589 break; 590 case TGSI_OPCODE_DPH: 591 nvfx_vp_emit(vpc, arith(sat, VEC, DPH, dst, mask, src[0], src[1], none)); 592 break; 593 case TGSI_OPCODE_DST: 594 nvfx_vp_emit(vpc, arith(sat, VEC, DST, dst, mask, src[0], src[1], none)); 595 break; 596 case TGSI_OPCODE_EX2: 597 nvfx_vp_emit(vpc, arith(sat, SCA, EX2, dst, mask, none, none, src[0])); 598 break; 599 case TGSI_OPCODE_EXP: 600 nvfx_vp_emit(vpc, arith(sat, SCA, EXP, dst, mask, none, none, src[0])); 601 break; 602 case TGSI_OPCODE_FLR: 603 nvfx_vp_emit(vpc, arith(sat, VEC, FLR, dst, mask, src[0], none, none)); 604 break; 605 case TGSI_OPCODE_FRC: 606 nvfx_vp_emit(vpc, arith(sat, VEC, FRC, dst, mask, src[0], none, none)); 607 break; 608 case TGSI_OPCODE_LG2: 609 nvfx_vp_emit(vpc, arith(sat, SCA, LG2, dst, mask, none, none, src[0])); 610 break; 611 case TGSI_OPCODE_LIT: 612 nvfx_vp_emit(vpc, arith(sat, SCA, LIT, dst, mask, none, none, src[0])); 613 break; 614 case TGSI_OPCODE_LOG: 615 nvfx_vp_emit(vpc, arith(sat, SCA, LOG, dst, mask, none, none, src[0])); 616 break; 617 case TGSI_OPCODE_LRP: 618 tmp = nvfx_src(temp(vpc)); 619 nvfx_vp_emit(vpc, arith(0, VEC, MAD, tmp.reg, mask, neg(src[0]), src[2], src[2])); 620 nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, mask, src[0], src[1], tmp)); 621 break; 622 case TGSI_OPCODE_MAD: 623 nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, mask, src[0], src[1], src[2])); 624 break; 625 case TGSI_OPCODE_MAX: 626 nvfx_vp_emit(vpc, arith(sat, VEC, MAX, dst, mask, src[0], src[1], none)); 627 break; 628 case TGSI_OPCODE_MIN: 629 nvfx_vp_emit(vpc, arith(sat, VEC, MIN, dst, mask, src[0], src[1], none)); 630 break; 631 case TGSI_OPCODE_MOV: 632 nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, src[0], none, none)); 633 break; 634 case TGSI_OPCODE_MUL: 635 nvfx_vp_emit(vpc, arith(sat, VEC, MUL, dst, mask, src[0], src[1], none)); 636 break; 637 case TGSI_OPCODE_NOP: 638 break; 639 case TGSI_OPCODE_POW: 640 tmp = nvfx_src(temp(vpc)); 641 nvfx_vp_emit(vpc, arith(0, SCA, LG2, tmp.reg, NVFX_VP_MASK_X, none, none, swz(src[0], X, X, X, X))); 642 nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, NVFX_VP_MASK_X, swz(tmp, X, X, X, X), swz(src[1], X, X, X, X), none)); 643 nvfx_vp_emit(vpc, arith(sat, SCA, EX2, dst, mask, none, none, swz(tmp, X, X, X, X))); 644 break; 645 case TGSI_OPCODE_RCP: 646 nvfx_vp_emit(vpc, arith(sat, SCA, RCP, dst, mask, none, none, src[0])); 647 break; 648 case TGSI_OPCODE_RSQ: 649 nvfx_vp_emit(vpc, arith(sat, SCA, RSQ, dst, mask, none, none, abs(src[0]))); 650 break; 651 case TGSI_OPCODE_SEQ: 652 nvfx_vp_emit(vpc, arith(sat, VEC, SEQ, dst, mask, src[0], src[1], none)); 653 break; 654 case TGSI_OPCODE_SGE: 655 nvfx_vp_emit(vpc, arith(sat, VEC, SGE, dst, mask, src[0], src[1], none)); 656 break; 657 case TGSI_OPCODE_SGT: 658 nvfx_vp_emit(vpc, arith(sat, VEC, SGT, dst, mask, src[0], src[1], none)); 659 break; 660 case TGSI_OPCODE_SIN: 661 nvfx_vp_emit(vpc, arith(sat, SCA, SIN, dst, mask, none, none, src[0])); 662 break; 663 case TGSI_OPCODE_SLE: 664 nvfx_vp_emit(vpc, arith(sat, VEC, SLE, dst, mask, src[0], src[1], none)); 665 break; 666 case TGSI_OPCODE_SLT: 667 nvfx_vp_emit(vpc, arith(sat, VEC, SLT, dst, mask, src[0], src[1], none)); 668 break; 669 case TGSI_OPCODE_SNE: 670 nvfx_vp_emit(vpc, arith(sat, VEC, SNE, dst, mask, src[0], src[1], none)); 671 break; 672 case TGSI_OPCODE_SSG: 673 nvfx_vp_emit(vpc, arith(sat, VEC, SSG, dst, mask, src[0], none, none)); 674 break; 675 case TGSI_OPCODE_TRUNC: 676 tmp = nvfx_src(temp(vpc)); 677 insn = arith(0, VEC, MOV, none.reg, mask, src[0], none, none); 678 insn.cc_update = 1; 679 nvfx_vp_emit(vpc, insn); 680 681 nvfx_vp_emit(vpc, arith(0, VEC, FLR, tmp.reg, mask, abs(src[0]), none, none)); 682 nvfx_vp_emit(vpc, arith(sat, VEC, MOV, dst, mask, tmp, none, none)); 683 684 insn = arith(sat, VEC, MOV, dst, mask, neg(tmp), none, none); 685 insn.cc_test = NVFX_COND_LT; 686 nvfx_vp_emit(vpc, insn); 687 break; 688 case TGSI_OPCODE_XPD: 689 tmp = nvfx_src(temp(vpc)); 690 nvfx_vp_emit(vpc, arith(0, VEC, MUL, tmp.reg, mask, swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none)); 691 nvfx_vp_emit(vpc, arith(sat, VEC, MAD, dst, (mask & ~NVFX_VP_MASK_W), swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y), neg(tmp))); 692 break; 693 case TGSI_OPCODE_IF: 694 insn = arith(0, VEC, MOV, none.reg, NVFX_VP_MASK_X, src[0], none, none); 695 insn.cc_update = 1; 696 nvfx_vp_emit(vpc, insn); 697 698 reloc.location = vpc->vp->nr_insns; 699 reloc.target = finst->Label.Label + 1; 700 util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc); 701 702 insn = arith(0, SCA, BRA, none.reg, 0, none, none, none); 703 insn.cc_test = NVFX_COND_EQ; 704 insn.cc_swz[0] = insn.cc_swz[1] = insn.cc_swz[2] = insn.cc_swz[3] = 0; 705 nvfx_vp_emit(vpc, insn); 706 break; 707 case TGSI_OPCODE_ELSE: 708 case TGSI_OPCODE_CAL: 709 reloc.location = vpc->vp->nr_insns; 710 reloc.target = finst->Label.Label; 711 util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc); 712 713 if(finst->Instruction.Opcode == TGSI_OPCODE_CAL) 714 insn = arith(0, SCA, CAL, none.reg, 0, none, none, none); 715 else 716 insn = arith(0, SCA, BRA, none.reg, 0, none, none, none); 717 nvfx_vp_emit(vpc, insn); 718 break; 719 case TGSI_OPCODE_RET: 720 if(sub_depth || !vpc->vp->enabled_ucps) { 721 tmp = none; 722 tmp.swz[0] = tmp.swz[1] = tmp.swz[2] = tmp.swz[3] = 0; 723 nvfx_vp_emit(vpc, arith(0, SCA, RET, none.reg, 0, none, none, tmp)); 724 } else { 725 reloc.location = vpc->vp->nr_insns; 726 reloc.target = vpc->info->num_instructions; 727 util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc); 728 nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); 729 } 730 break; 731 case TGSI_OPCODE_BGNSUB: 732 ++sub_depth; 733 break; 734 case TGSI_OPCODE_ENDSUB: 735 --sub_depth; 736 break; 737 case TGSI_OPCODE_ENDIF: 738 /* nothing to do here */ 739 break; 740 case TGSI_OPCODE_BGNLOOP: 741 loop.cont_target = idx; 742 loop.brk_target = finst->Label.Label + 1; 743 util_dynarray_append(&vpc->loop_stack, struct nvfx_loop_entry, loop); 744 break; 745 case TGSI_OPCODE_ENDLOOP: 746 loop = util_dynarray_pop(&vpc->loop_stack, struct nvfx_loop_entry); 747 748 reloc.location = vpc->vp->nr_insns; 749 reloc.target = loop.cont_target; 750 util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc); 751 752 nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); 753 break; 754 case TGSI_OPCODE_CONT: 755 loop = util_dynarray_top(&vpc->loop_stack, struct nvfx_loop_entry); 756 757 reloc.location = vpc->vp->nr_insns; 758 reloc.target = loop.cont_target; 759 util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc); 760 761 nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); 762 break; 763 case TGSI_OPCODE_BRK: 764 loop = util_dynarray_top(&vpc->loop_stack, struct nvfx_loop_entry); 765 766 reloc.location = vpc->vp->nr_insns; 767 reloc.target = loop.brk_target; 768 util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc); 769 770 nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); 771 break; 772 case TGSI_OPCODE_END: 773 assert(!sub_depth); 774 if(vpc->vp->enabled_ucps) { 775 if(idx != (vpc->info->num_instructions - 1)) { 776 reloc.location = vpc->vp->nr_insns; 777 reloc.target = vpc->info->num_instructions; 778 util_dynarray_append(&vpc->label_relocs, struct nvfx_relocation, reloc); 779 nvfx_vp_emit(vpc, arith(0, SCA, BRA, none.reg, 0, none, none, none)); 780 } 781 } else { 782 if(vpc->vp->nr_insns) 783 vpc->vp->insns[vpc->vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST; 784 nvfx_vp_emit(vpc, arith(0, VEC, NOP, none.reg, 0, none, none, none)); 785 vpc->vp->insns[vpc->vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST; 786 } 787 break; 788 default: 789 NOUVEAU_ERR("invalid opcode %d\n", finst->Instruction.Opcode); 790 return false; 791 } 792 793 if(finst->Instruction.Saturate && !vpc->is_nv4x) { 794 if (!vpc->r_0_1.type) 795 vpc->r_0_1 = constant(vpc, -1, 0, 1, 0, 0); 796 nvfx_vp_emit(vpc, arith(0, VEC, MAX, dst, mask, nvfx_src(dst), swz(nvfx_src(vpc->r_0_1), X, X, X, X), none)); 797 nvfx_vp_emit(vpc, arith(0, VEC, MIN, final_dst, mask, nvfx_src(dst), swz(nvfx_src(vpc->r_0_1), Y, Y, Y, Y), none)); 798 } 799 800 release_temps(vpc); 801 return true; 802} 803 804static bool 805nvfx_vertprog_parse_decl_output(struct nvfx_vpc *vpc, 806 const struct tgsi_full_declaration *fdec) 807{ 808 unsigned num_texcoords = vpc->is_nv4x ? 10 : 8; 809 unsigned idx = fdec->Range.First; 810 unsigned semantic_index = fdec->Semantic.Index; 811 int hw = 0, i; 812 813 switch (fdec->Semantic.Name) { 814 case TGSI_SEMANTIC_POSITION: 815 hw = NVFX_VP(INST_DEST_POS); 816 vpc->hpos_idx = idx; 817 break; 818 case TGSI_SEMANTIC_CLIPVERTEX: 819 vpc->r_result[idx] = temp(vpc); 820 vpc->r_temps_discard = 0; 821 vpc->cvtx_idx = idx; 822 return true; 823 case TGSI_SEMANTIC_COLOR: 824 if (fdec->Semantic.Index == 0) { 825 hw = NVFX_VP(INST_DEST_COL0); 826 } else 827 if (fdec->Semantic.Index == 1) { 828 hw = NVFX_VP(INST_DEST_COL1); 829 } else { 830 NOUVEAU_ERR("bad colour semantic index\n"); 831 return false; 832 } 833 break; 834 case TGSI_SEMANTIC_BCOLOR: 835 if (fdec->Semantic.Index == 0) { 836 hw = NVFX_VP(INST_DEST_BFC0); 837 } else 838 if (fdec->Semantic.Index == 1) { 839 hw = NVFX_VP(INST_DEST_BFC1); 840 } else { 841 NOUVEAU_ERR("bad bcolour semantic index\n"); 842 return false; 843 } 844 break; 845 case TGSI_SEMANTIC_FOG: 846 hw = NVFX_VP(INST_DEST_FOGC); 847 break; 848 case TGSI_SEMANTIC_PSIZE: 849 hw = NVFX_VP(INST_DEST_PSZ); 850 break; 851 case TGSI_SEMANTIC_GENERIC: 852 /* this is really an identifier for VP/FP linkage */ 853 semantic_index += 8; 854 /* fall through */ 855 case TGSI_SEMANTIC_TEXCOORD: 856 for (i = 0; i < num_texcoords; i++) { 857 if (vpc->vp->texcoord[i] == semantic_index) { 858 hw = NVFX_VP(INST_DEST_TC(i)); 859 break; 860 } 861 } 862 863 if (i == num_texcoords) { 864 vpc->r_result[idx] = nvfx_reg(NVFXSR_NONE, 0); 865 return true; 866 } 867 break; 868 case TGSI_SEMANTIC_EDGEFLAG: 869 vpc->r_result[idx] = nvfx_reg(NVFXSR_NONE, 0); 870 return true; 871 default: 872 NOUVEAU_ERR("bad output semantic\n"); 873 return false; 874 } 875 876 vpc->r_result[idx] = nvfx_reg(NVFXSR_OUTPUT, hw); 877 return true; 878} 879 880static bool 881nvfx_vertprog_prepare(struct nvfx_vpc *vpc) 882{ 883 struct tgsi_parse_context p; 884 int high_const = -1, high_temp = -1, high_addr = -1, nr_imm = 0, i; 885 886 tgsi_parse_init(&p, vpc->pipe.tokens); 887 while (!tgsi_parse_end_of_tokens(&p)) { 888 const union tgsi_full_token *tok = &p.FullToken; 889 890 tgsi_parse_token(&p); 891 switch(tok->Token.Type) { 892 case TGSI_TOKEN_TYPE_IMMEDIATE: 893 nr_imm++; 894 break; 895 case TGSI_TOKEN_TYPE_DECLARATION: 896 { 897 const struct tgsi_full_declaration *fdec; 898 899 fdec = &p.FullToken.FullDeclaration; 900 switch (fdec->Declaration.File) { 901 case TGSI_FILE_TEMPORARY: 902 if (fdec->Range.Last > high_temp) { 903 high_temp = 904 fdec->Range.Last; 905 } 906 break; 907 case TGSI_FILE_ADDRESS: 908 if (fdec->Range.Last > high_addr) { 909 high_addr = 910 fdec->Range.Last; 911 } 912 break; 913 case TGSI_FILE_CONSTANT: 914 if (fdec->Range.Last > high_const) { 915 high_const = 916 fdec->Range.Last; 917 } 918 break; 919 case TGSI_FILE_OUTPUT: 920 if (!nvfx_vertprog_parse_decl_output(vpc, fdec)) 921 return false; 922 break; 923 default: 924 break; 925 } 926 } 927 break; 928 default: 929 break; 930 } 931 } 932 tgsi_parse_free(&p); 933 934 if (nr_imm) { 935 vpc->imm = CALLOC(nr_imm, sizeof(struct nvfx_reg)); 936 assert(vpc->imm); 937 } 938 939 if (++high_temp) { 940 vpc->r_temp = CALLOC(high_temp, sizeof(struct nvfx_reg)); 941 for (i = 0; i < high_temp; i++) 942 vpc->r_temp[i] = temp(vpc); 943 } 944 945 if (++high_addr) { 946 vpc->r_address = CALLOC(high_addr, sizeof(struct nvfx_reg)); 947 for (i = 0; i < high_addr; i++) 948 vpc->r_address[i] = nvfx_reg(NVFXSR_TEMP, i); 949 } 950 951 if(++high_const) { 952 vpc->r_const = CALLOC(high_const, sizeof(struct nvfx_reg)); 953 for (i = 0; i < high_const; i++) 954 vpc->r_const[i] = constant(vpc, i, 0, 0, 0, 0); 955 } 956 957 vpc->r_temps_discard = 0; 958 return true; 959} 960 961DEBUG_GET_ONCE_BOOL_OPTION(nvfx_dump_vp, "NVFX_DUMP_VP", false) 962 963bool 964_nvfx_vertprog_translate(uint16_t oclass, struct nv30_vertprog *vp) 965{ 966 struct tgsi_parse_context parse; 967 struct nvfx_vpc *vpc = NULL; 968 struct nvfx_src none = nvfx_src(nvfx_reg(NVFXSR_NONE, 0)); 969 struct util_dynarray insns; 970 int i, ucps; 971 972 vp->translated = false; 973 vp->nr_insns = 0; 974 vp->nr_consts = 0; 975 976 vpc = CALLOC_STRUCT(nvfx_vpc); 977 if (!vpc) 978 return false; 979 vpc->is_nv4x = (oclass >= NV40_3D_CLASS) ? ~0 : 0; 980 vpc->vp = vp; 981 vpc->pipe = vp->pipe; 982 vpc->info = &vp->info; 983 vpc->cvtx_idx = -1; 984 985 if (!nvfx_vertprog_prepare(vpc)) { 986 FREE(vpc); 987 return false; 988 } 989 990 /* Redirect post-transform vertex position to a temp if user clip 991 * planes are enabled. We need to append code to the vtxprog 992 * to handle clip planes later. 993 */ 994 if (vp->enabled_ucps && vpc->cvtx_idx < 0) { 995 vpc->r_result[vpc->hpos_idx] = temp(vpc); 996 vpc->r_temps_discard = 0; 997 vpc->cvtx_idx = vpc->hpos_idx; 998 } 999 1000 util_dynarray_init(&insns); 1001 1002 tgsi_parse_init(&parse, vp->pipe.tokens); 1003 while (!tgsi_parse_end_of_tokens(&parse)) { 1004 tgsi_parse_token(&parse); 1005 1006 switch (parse.FullToken.Token.Type) { 1007 case TGSI_TOKEN_TYPE_IMMEDIATE: 1008 { 1009 const struct tgsi_full_immediate *imm; 1010 1011 imm = &parse.FullToken.FullImmediate; 1012 assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32); 1013 assert(imm->Immediate.NrTokens == 4 + 1); 1014 vpc->imm[vpc->nr_imm++] = 1015 constant(vpc, -1, 1016 imm->u[0].Float, 1017 imm->u[1].Float, 1018 imm->u[2].Float, 1019 imm->u[3].Float); 1020 } 1021 break; 1022 case TGSI_TOKEN_TYPE_INSTRUCTION: 1023 { 1024 const struct tgsi_full_instruction *finst; 1025 unsigned idx = insns.size >> 2; 1026 util_dynarray_append(&insns, unsigned, vp->nr_insns); 1027 finst = &parse.FullToken.FullInstruction; 1028 if (!nvfx_vertprog_parse_instruction(vpc, idx, finst)) 1029 goto out; 1030 } 1031 break; 1032 default: 1033 break; 1034 } 1035 } 1036 1037 util_dynarray_append(&insns, unsigned, vp->nr_insns); 1038 1039 for(unsigned i = 0; i < vpc->label_relocs.size; i += sizeof(struct nvfx_relocation)) 1040 { 1041 struct nvfx_relocation* label_reloc = (struct nvfx_relocation*)((char*)vpc->label_relocs.data + i); 1042 struct nvfx_relocation hw_reloc; 1043 1044 hw_reloc.location = label_reloc->location; 1045 hw_reloc.target = ((unsigned*)insns.data)[label_reloc->target]; 1046 1047 //debug_printf("hw %u -> tgsi %u = hw %u\n", hw_reloc.location, label_reloc->target, hw_reloc.target); 1048 1049 util_dynarray_append(&vp->branch_relocs, struct nvfx_relocation, hw_reloc); 1050 } 1051 util_dynarray_fini(&insns); 1052 util_dynarray_trim(&vp->branch_relocs); 1053 1054 /* XXX: what if we add a RET before?! make sure we jump here...*/ 1055 1056 /* Write out HPOS if it was redirected to a temp earlier */ 1057 if (vpc->r_result[vpc->hpos_idx].type != NVFXSR_OUTPUT) { 1058 struct nvfx_reg hpos = nvfx_reg(NVFXSR_OUTPUT, 1059 NVFX_VP(INST_DEST_POS)); 1060 struct nvfx_src htmp = nvfx_src(vpc->r_result[vpc->hpos_idx]); 1061 1062 nvfx_vp_emit(vpc, arith(0, VEC, MOV, hpos, NVFX_VP_MASK_ALL, htmp, none, none)); 1063 } 1064 1065 /* Insert code to handle user clip planes */ 1066 ucps = vp->enabled_ucps; 1067 while (ucps) { 1068 int i = ffs(ucps) - 1; ucps &= ~(1 << i); 1069 struct nvfx_reg cdst = nvfx_reg(NVFXSR_OUTPUT, NV30_VP_INST_DEST_CLP(i)); 1070 struct nvfx_src ceqn = nvfx_src(nvfx_reg(NVFXSR_CONST, 512 + i)); 1071 struct nvfx_src htmp = nvfx_src(vpc->r_result[vpc->cvtx_idx]); 1072 unsigned mask; 1073 1074 if(vpc->is_nv4x) 1075 { 1076 switch (i) { 1077 case 0: case 3: mask = NVFX_VP_MASK_Y; break; 1078 case 1: case 4: mask = NVFX_VP_MASK_Z; break; 1079 case 2: case 5: mask = NVFX_VP_MASK_W; break; 1080 default: 1081 NOUVEAU_ERR("invalid clip dist #%d\n", i); 1082 goto out; 1083 } 1084 } 1085 else 1086 mask = NVFX_VP_MASK_X; 1087 1088 nvfx_vp_emit(vpc, arith(0, VEC, DP4, cdst, mask, htmp, ceqn, none)); 1089 } 1090 1091 if (vpc->vp->nr_insns) 1092 vpc->vp->insns[vpc->vp->nr_insns - 1].data[3] |= NVFX_VP_INST_LAST; 1093 1094 if(debug_get_option_nvfx_dump_vp()) 1095 { 1096 debug_printf("\n"); 1097 tgsi_dump(vpc->pipe.tokens, 0); 1098 1099 debug_printf("\n%s vertex program:\n", vpc->is_nv4x ? "nv4x" : "nv3x"); 1100 for (i = 0; i < vp->nr_insns; i++) 1101 debug_printf("%3u: %08x %08x %08x %08x\n", i, vp->insns[i].data[0], vp->insns[i].data[1], vp->insns[i].data[2], vp->insns[i].data[3]); 1102 debug_printf("\n"); 1103 } 1104 1105 vp->translated = true; 1106 1107out: 1108 tgsi_parse_free(&parse); 1109 if (vpc) { 1110 util_dynarray_fini(&vpc->label_relocs); 1111 util_dynarray_fini(&vpc->loop_stack); 1112 FREE(vpc->r_temp); 1113 FREE(vpc->r_address); 1114 FREE(vpc->r_const); 1115 FREE(vpc->imm); 1116 FREE(vpc); 1117 } 1118 1119 return vp->translated; 1120} 1121