1/* 2 * Copyright © 2010 Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 */ 23 24#include "brw_cfg.h" 25#include "brw_eu.h" 26#include "brw_fs.h" 27#include "brw_nir.h" 28#include "brw_vec4_tes.h" 29#include "dev/gen_debug.h" 30#include "main/uniforms.h" 31#include "util/macros.h" 32 33enum brw_reg_type 34brw_type_for_base_type(const struct glsl_type *type) 35{ 36 switch (type->base_type) { 37 case GLSL_TYPE_FLOAT16: 38 return BRW_REGISTER_TYPE_HF; 39 case GLSL_TYPE_FLOAT: 40 return BRW_REGISTER_TYPE_F; 41 case GLSL_TYPE_INT: 42 case GLSL_TYPE_BOOL: 43 case GLSL_TYPE_SUBROUTINE: 44 return BRW_REGISTER_TYPE_D; 45 case GLSL_TYPE_INT16: 46 return BRW_REGISTER_TYPE_W; 47 case GLSL_TYPE_INT8: 48 return BRW_REGISTER_TYPE_B; 49 case GLSL_TYPE_UINT: 50 return BRW_REGISTER_TYPE_UD; 51 case GLSL_TYPE_UINT16: 52 return BRW_REGISTER_TYPE_UW; 53 case GLSL_TYPE_UINT8: 54 return BRW_REGISTER_TYPE_UB; 55 case GLSL_TYPE_ARRAY: 56 return brw_type_for_base_type(type->fields.array); 57 case GLSL_TYPE_STRUCT: 58 case GLSL_TYPE_INTERFACE: 59 case GLSL_TYPE_SAMPLER: 60 case GLSL_TYPE_ATOMIC_UINT: 61 /* These should be overridden with the type of the member when 62 * dereferenced into. BRW_REGISTER_TYPE_UD seems like a likely 63 * way to trip up if we don't. 64 */ 65 return BRW_REGISTER_TYPE_UD; 66 case GLSL_TYPE_IMAGE: 67 return BRW_REGISTER_TYPE_UD; 68 case GLSL_TYPE_DOUBLE: 69 return BRW_REGISTER_TYPE_DF; 70 case GLSL_TYPE_UINT64: 71 return BRW_REGISTER_TYPE_UQ; 72 case GLSL_TYPE_INT64: 73 return BRW_REGISTER_TYPE_Q; 74 case GLSL_TYPE_VOID: 75 case GLSL_TYPE_ERROR: 76 case GLSL_TYPE_FUNCTION: 77 unreachable("not reached"); 78 } 79 80 return BRW_REGISTER_TYPE_F; 81} 82 83enum brw_conditional_mod 84brw_conditional_for_comparison(unsigned int op) 85{ 86 switch (op) { 87 case ir_binop_less: 88 return BRW_CONDITIONAL_L; 89 case ir_binop_gequal: 90 return BRW_CONDITIONAL_GE; 91 case ir_binop_equal: 92 case ir_binop_all_equal: /* same as equal for scalars */ 93 return BRW_CONDITIONAL_Z; 94 case ir_binop_nequal: 95 case ir_binop_any_nequal: /* same as nequal for scalars */ 96 return BRW_CONDITIONAL_NZ; 97 default: 98 unreachable("not reached: bad operation for comparison"); 99 } 100} 101 102uint32_t 103brw_math_function(enum opcode op) 104{ 105 switch (op) { 106 case SHADER_OPCODE_RCP: 107 return BRW_MATH_FUNCTION_INV; 108 case SHADER_OPCODE_RSQ: 109 return BRW_MATH_FUNCTION_RSQ; 110 case SHADER_OPCODE_SQRT: 111 return BRW_MATH_FUNCTION_SQRT; 112 case SHADER_OPCODE_EXP2: 113 return BRW_MATH_FUNCTION_EXP; 114 case SHADER_OPCODE_LOG2: 115 return BRW_MATH_FUNCTION_LOG; 116 case SHADER_OPCODE_POW: 117 return BRW_MATH_FUNCTION_POW; 118 case SHADER_OPCODE_SIN: 119 return BRW_MATH_FUNCTION_SIN; 120 case SHADER_OPCODE_COS: 121 return BRW_MATH_FUNCTION_COS; 122 case SHADER_OPCODE_INT_QUOTIENT: 123 return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT; 124 case SHADER_OPCODE_INT_REMAINDER: 125 return BRW_MATH_FUNCTION_INT_DIV_REMAINDER; 126 default: 127 unreachable("not reached: unknown math function"); 128 } 129} 130 131bool 132brw_texture_offset(const nir_tex_instr *tex, unsigned src, 133 uint32_t *offset_bits_out) 134{ 135 if (!nir_src_is_const(tex->src[src].src)) 136 return false; 137 138 const unsigned num_components = nir_tex_instr_src_size(tex, src); 139 140 /* Combine all three offsets into a single unsigned dword: 141 * 142 * bits 11:8 - U Offset (X component) 143 * bits 7:4 - V Offset (Y component) 144 * bits 3:0 - R Offset (Z component) 145 */ 146 uint32_t offset_bits = 0; 147 for (unsigned i = 0; i < num_components; i++) { 148 int offset = nir_src_comp_as_int(tex->src[src].src, i); 149 150 /* offset out of bounds; caller will handle it. */ 151 if (offset > 7 || offset < -8) 152 return false; 153 154 const unsigned shift = 4 * (2 - i); 155 offset_bits |= (offset << shift) & (0xF << shift); 156 } 157 158 *offset_bits_out = offset_bits; 159 160 return true; 161} 162 163const char * 164brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op) 165{ 166 switch (op) { 167 case BRW_OPCODE_ILLEGAL ... BRW_OPCODE_NOP: 168 /* The DO instruction doesn't exist on Gen6+, but we use it to mark the 169 * start of a loop in the IR. 170 */ 171 if (devinfo->gen >= 6 && op == BRW_OPCODE_DO) 172 return "do"; 173 174 /* The following conversion opcodes doesn't exist on Gen8+, but we use 175 * then to mark that we want to do the conversion. 176 */ 177 if (devinfo->gen > 7 && op == BRW_OPCODE_F32TO16) 178 return "f32to16"; 179 180 if (devinfo->gen > 7 && op == BRW_OPCODE_F16TO32) 181 return "f16to32"; 182 183 assert(brw_opcode_desc(devinfo, op)->name); 184 return brw_opcode_desc(devinfo, op)->name; 185 case FS_OPCODE_FB_WRITE: 186 return "fb_write"; 187 case FS_OPCODE_FB_WRITE_LOGICAL: 188 return "fb_write_logical"; 189 case FS_OPCODE_REP_FB_WRITE: 190 return "rep_fb_write"; 191 case FS_OPCODE_FB_READ: 192 return "fb_read"; 193 case FS_OPCODE_FB_READ_LOGICAL: 194 return "fb_read_logical"; 195 196 case SHADER_OPCODE_RCP: 197 return "rcp"; 198 case SHADER_OPCODE_RSQ: 199 return "rsq"; 200 case SHADER_OPCODE_SQRT: 201 return "sqrt"; 202 case SHADER_OPCODE_EXP2: 203 return "exp2"; 204 case SHADER_OPCODE_LOG2: 205 return "log2"; 206 case SHADER_OPCODE_POW: 207 return "pow"; 208 case SHADER_OPCODE_INT_QUOTIENT: 209 return "int_quot"; 210 case SHADER_OPCODE_INT_REMAINDER: 211 return "int_rem"; 212 case SHADER_OPCODE_SIN: 213 return "sin"; 214 case SHADER_OPCODE_COS: 215 return "cos"; 216 217 case SHADER_OPCODE_SEND: 218 return "send"; 219 220 case SHADER_OPCODE_TEX: 221 return "tex"; 222 case SHADER_OPCODE_TEX_LOGICAL: 223 return "tex_logical"; 224 case SHADER_OPCODE_TXD: 225 return "txd"; 226 case SHADER_OPCODE_TXD_LOGICAL: 227 return "txd_logical"; 228 case SHADER_OPCODE_TXF: 229 return "txf"; 230 case SHADER_OPCODE_TXF_LOGICAL: 231 return "txf_logical"; 232 case SHADER_OPCODE_TXF_LZ: 233 return "txf_lz"; 234 case SHADER_OPCODE_TXL: 235 return "txl"; 236 case SHADER_OPCODE_TXL_LOGICAL: 237 return "txl_logical"; 238 case SHADER_OPCODE_TXL_LZ: 239 return "txl_lz"; 240 case SHADER_OPCODE_TXS: 241 return "txs"; 242 case SHADER_OPCODE_TXS_LOGICAL: 243 return "txs_logical"; 244 case FS_OPCODE_TXB: 245 return "txb"; 246 case FS_OPCODE_TXB_LOGICAL: 247 return "txb_logical"; 248 case SHADER_OPCODE_TXF_CMS: 249 return "txf_cms"; 250 case SHADER_OPCODE_TXF_CMS_LOGICAL: 251 return "txf_cms_logical"; 252 case SHADER_OPCODE_TXF_CMS_W: 253 return "txf_cms_w"; 254 case SHADER_OPCODE_TXF_CMS_W_LOGICAL: 255 return "txf_cms_w_logical"; 256 case SHADER_OPCODE_TXF_UMS: 257 return "txf_ums"; 258 case SHADER_OPCODE_TXF_UMS_LOGICAL: 259 return "txf_ums_logical"; 260 case SHADER_OPCODE_TXF_MCS: 261 return "txf_mcs"; 262 case SHADER_OPCODE_TXF_MCS_LOGICAL: 263 return "txf_mcs_logical"; 264 case SHADER_OPCODE_LOD: 265 return "lod"; 266 case SHADER_OPCODE_LOD_LOGICAL: 267 return "lod_logical"; 268 case SHADER_OPCODE_TG4: 269 return "tg4"; 270 case SHADER_OPCODE_TG4_LOGICAL: 271 return "tg4_logical"; 272 case SHADER_OPCODE_TG4_OFFSET: 273 return "tg4_offset"; 274 case SHADER_OPCODE_TG4_OFFSET_LOGICAL: 275 return "tg4_offset_logical"; 276 case SHADER_OPCODE_SAMPLEINFO: 277 return "sampleinfo"; 278 case SHADER_OPCODE_SAMPLEINFO_LOGICAL: 279 return "sampleinfo_logical"; 280 281 case SHADER_OPCODE_IMAGE_SIZE_LOGICAL: 282 return "image_size_logical"; 283 284 case SHADER_OPCODE_SHADER_TIME_ADD: 285 return "shader_time_add"; 286 287 case VEC4_OPCODE_UNTYPED_ATOMIC: 288 return "untyped_atomic"; 289 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL: 290 return "untyped_atomic_logical"; 291 case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL: 292 return "untyped_atomic_float_logical"; 293 case VEC4_OPCODE_UNTYPED_SURFACE_READ: 294 return "untyped_surface_read"; 295 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL: 296 return "untyped_surface_read_logical"; 297 case VEC4_OPCODE_UNTYPED_SURFACE_WRITE: 298 return "untyped_surface_write"; 299 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL: 300 return "untyped_surface_write_logical"; 301 case SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL: 302 return "a64_untyped_read_logical"; 303 case SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL: 304 return "a64_untyped_write_logical"; 305 case SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL: 306 return "a64_byte_scattered_read_logical"; 307 case SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL: 308 return "a64_byte_scattered_write_logical"; 309 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL: 310 return "a64_untyped_atomic_logical"; 311 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_INT64_LOGICAL: 312 return "a64_untyped_atomic_int64_logical"; 313 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL: 314 return "a64_untyped_atomic_float_logical"; 315 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL: 316 return "typed_atomic_logical"; 317 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL: 318 return "typed_surface_read_logical"; 319 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL: 320 return "typed_surface_write_logical"; 321 case SHADER_OPCODE_MEMORY_FENCE: 322 return "memory_fence"; 323 case SHADER_OPCODE_INTERLOCK: 324 /* For an interlock we actually issue a memory fence via sendc. */ 325 return "interlock"; 326 327 case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL: 328 return "byte_scattered_read_logical"; 329 case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL: 330 return "byte_scattered_write_logical"; 331 332 case SHADER_OPCODE_LOAD_PAYLOAD: 333 return "load_payload"; 334 case FS_OPCODE_PACK: 335 return "pack"; 336 337 case SHADER_OPCODE_GEN4_SCRATCH_READ: 338 return "gen4_scratch_read"; 339 case SHADER_OPCODE_GEN4_SCRATCH_WRITE: 340 return "gen4_scratch_write"; 341 case SHADER_OPCODE_GEN7_SCRATCH_READ: 342 return "gen7_scratch_read"; 343 case SHADER_OPCODE_URB_WRITE_SIMD8: 344 return "gen8_urb_write_simd8"; 345 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT: 346 return "gen8_urb_write_simd8_per_slot"; 347 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED: 348 return "gen8_urb_write_simd8_masked"; 349 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT: 350 return "gen8_urb_write_simd8_masked_per_slot"; 351 case SHADER_OPCODE_URB_READ_SIMD8: 352 return "urb_read_simd8"; 353 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT: 354 return "urb_read_simd8_per_slot"; 355 356 case SHADER_OPCODE_FIND_LIVE_CHANNEL: 357 return "find_live_channel"; 358 case SHADER_OPCODE_BROADCAST: 359 return "broadcast"; 360 case SHADER_OPCODE_SHUFFLE: 361 return "shuffle"; 362 case SHADER_OPCODE_SEL_EXEC: 363 return "sel_exec"; 364 case SHADER_OPCODE_QUAD_SWIZZLE: 365 return "quad_swizzle"; 366 case SHADER_OPCODE_CLUSTER_BROADCAST: 367 return "cluster_broadcast"; 368 369 case SHADER_OPCODE_GET_BUFFER_SIZE: 370 return "get_buffer_size"; 371 372 case VEC4_OPCODE_MOV_BYTES: 373 return "mov_bytes"; 374 case VEC4_OPCODE_PACK_BYTES: 375 return "pack_bytes"; 376 case VEC4_OPCODE_UNPACK_UNIFORM: 377 return "unpack_uniform"; 378 case VEC4_OPCODE_DOUBLE_TO_F32: 379 return "double_to_f32"; 380 case VEC4_OPCODE_DOUBLE_TO_D32: 381 return "double_to_d32"; 382 case VEC4_OPCODE_DOUBLE_TO_U32: 383 return "double_to_u32"; 384 case VEC4_OPCODE_TO_DOUBLE: 385 return "single_to_double"; 386 case VEC4_OPCODE_PICK_LOW_32BIT: 387 return "pick_low_32bit"; 388 case VEC4_OPCODE_PICK_HIGH_32BIT: 389 return "pick_high_32bit"; 390 case VEC4_OPCODE_SET_LOW_32BIT: 391 return "set_low_32bit"; 392 case VEC4_OPCODE_SET_HIGH_32BIT: 393 return "set_high_32bit"; 394 395 case FS_OPCODE_DDX_COARSE: 396 return "ddx_coarse"; 397 case FS_OPCODE_DDX_FINE: 398 return "ddx_fine"; 399 case FS_OPCODE_DDY_COARSE: 400 return "ddy_coarse"; 401 case FS_OPCODE_DDY_FINE: 402 return "ddy_fine"; 403 404 case FS_OPCODE_LINTERP: 405 return "linterp"; 406 407 case FS_OPCODE_PIXEL_X: 408 return "pixel_x"; 409 case FS_OPCODE_PIXEL_Y: 410 return "pixel_y"; 411 412 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD: 413 return "uniform_pull_const"; 414 case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7: 415 return "uniform_pull_const_gen7"; 416 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4: 417 return "varying_pull_const_gen4"; 418 case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL: 419 return "varying_pull_const_logical"; 420 421 case FS_OPCODE_DISCARD_JUMP: 422 return "discard_jump"; 423 424 case FS_OPCODE_SET_SAMPLE_ID: 425 return "set_sample_id"; 426 427 case FS_OPCODE_PACK_HALF_2x16_SPLIT: 428 return "pack_half_2x16_split"; 429 430 case FS_OPCODE_PLACEHOLDER_HALT: 431 return "placeholder_halt"; 432 433 case FS_OPCODE_INTERPOLATE_AT_SAMPLE: 434 return "interp_sample"; 435 case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET: 436 return "interp_shared_offset"; 437 case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET: 438 return "interp_per_slot_offset"; 439 440 case VS_OPCODE_URB_WRITE: 441 return "vs_urb_write"; 442 case VS_OPCODE_PULL_CONSTANT_LOAD: 443 return "pull_constant_load"; 444 case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7: 445 return "pull_constant_load_gen7"; 446 447 case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9: 448 return "set_simd4x2_header_gen9"; 449 450 case VS_OPCODE_UNPACK_FLAGS_SIMD4X2: 451 return "unpack_flags_simd4x2"; 452 453 case GS_OPCODE_URB_WRITE: 454 return "gs_urb_write"; 455 case GS_OPCODE_URB_WRITE_ALLOCATE: 456 return "gs_urb_write_allocate"; 457 case GS_OPCODE_THREAD_END: 458 return "gs_thread_end"; 459 case GS_OPCODE_SET_WRITE_OFFSET: 460 return "set_write_offset"; 461 case GS_OPCODE_SET_VERTEX_COUNT: 462 return "set_vertex_count"; 463 case GS_OPCODE_SET_DWORD_2: 464 return "set_dword_2"; 465 case GS_OPCODE_PREPARE_CHANNEL_MASKS: 466 return "prepare_channel_masks"; 467 case GS_OPCODE_SET_CHANNEL_MASKS: 468 return "set_channel_masks"; 469 case GS_OPCODE_GET_INSTANCE_ID: 470 return "get_instance_id"; 471 case GS_OPCODE_FF_SYNC: 472 return "ff_sync"; 473 case GS_OPCODE_SET_PRIMITIVE_ID: 474 return "set_primitive_id"; 475 case GS_OPCODE_SVB_WRITE: 476 return "gs_svb_write"; 477 case GS_OPCODE_SVB_SET_DST_INDEX: 478 return "gs_svb_set_dst_index"; 479 case GS_OPCODE_FF_SYNC_SET_PRIMITIVES: 480 return "gs_ff_sync_set_primitives"; 481 case CS_OPCODE_CS_TERMINATE: 482 return "cs_terminate"; 483 case SHADER_OPCODE_BARRIER: 484 return "barrier"; 485 case SHADER_OPCODE_MULH: 486 return "mulh"; 487 case SHADER_OPCODE_MOV_INDIRECT: 488 return "mov_indirect"; 489 490 case VEC4_OPCODE_URB_READ: 491 return "urb_read"; 492 case TCS_OPCODE_GET_INSTANCE_ID: 493 return "tcs_get_instance_id"; 494 case TCS_OPCODE_URB_WRITE: 495 return "tcs_urb_write"; 496 case TCS_OPCODE_SET_INPUT_URB_OFFSETS: 497 return "tcs_set_input_urb_offsets"; 498 case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS: 499 return "tcs_set_output_urb_offsets"; 500 case TCS_OPCODE_GET_PRIMITIVE_ID: 501 return "tcs_get_primitive_id"; 502 case TCS_OPCODE_CREATE_BARRIER_HEADER: 503 return "tcs_create_barrier_header"; 504 case TCS_OPCODE_SRC0_010_IS_ZERO: 505 return "tcs_src0<0,1,0>_is_zero"; 506 case TCS_OPCODE_RELEASE_INPUT: 507 return "tcs_release_input"; 508 case TCS_OPCODE_THREAD_END: 509 return "tcs_thread_end"; 510 case TES_OPCODE_CREATE_INPUT_READ_HEADER: 511 return "tes_create_input_read_header"; 512 case TES_OPCODE_ADD_INDIRECT_URB_OFFSET: 513 return "tes_add_indirect_urb_offset"; 514 case TES_OPCODE_GET_PRIMITIVE_ID: 515 return "tes_get_primitive_id"; 516 517 case SHADER_OPCODE_RND_MODE: 518 return "rnd_mode"; 519 } 520 521 unreachable("not reached"); 522} 523 524bool 525brw_saturate_immediate(enum brw_reg_type type, struct brw_reg *reg) 526{ 527 union { 528 unsigned ud; 529 int d; 530 float f; 531 double df; 532 } imm, sat_imm = { 0 }; 533 534 const unsigned size = type_sz(type); 535 536 /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise 537 * irrelevant, so just check the size of the type and copy from/to an 538 * appropriately sized field. 539 */ 540 if (size < 8) 541 imm.ud = reg->ud; 542 else 543 imm.df = reg->df; 544 545 switch (type) { 546 case BRW_REGISTER_TYPE_UD: 547 case BRW_REGISTER_TYPE_D: 548 case BRW_REGISTER_TYPE_UW: 549 case BRW_REGISTER_TYPE_W: 550 case BRW_REGISTER_TYPE_UQ: 551 case BRW_REGISTER_TYPE_Q: 552 /* Nothing to do. */ 553 return false; 554 case BRW_REGISTER_TYPE_F: 555 sat_imm.f = CLAMP(imm.f, 0.0f, 1.0f); 556 break; 557 case BRW_REGISTER_TYPE_DF: 558 sat_imm.df = CLAMP(imm.df, 0.0, 1.0); 559 break; 560 case BRW_REGISTER_TYPE_UB: 561 case BRW_REGISTER_TYPE_B: 562 unreachable("no UB/B immediates"); 563 case BRW_REGISTER_TYPE_V: 564 case BRW_REGISTER_TYPE_UV: 565 case BRW_REGISTER_TYPE_VF: 566 unreachable("unimplemented: saturate vector immediate"); 567 case BRW_REGISTER_TYPE_HF: 568 unreachable("unimplemented: saturate HF immediate"); 569 case BRW_REGISTER_TYPE_NF: 570 unreachable("no NF immediates"); 571 } 572 573 if (size < 8) { 574 if (imm.ud != sat_imm.ud) { 575 reg->ud = sat_imm.ud; 576 return true; 577 } 578 } else { 579 if (imm.df != sat_imm.df) { 580 reg->df = sat_imm.df; 581 return true; 582 } 583 } 584 return false; 585} 586 587bool 588brw_negate_immediate(enum brw_reg_type type, struct brw_reg *reg) 589{ 590 switch (type) { 591 case BRW_REGISTER_TYPE_D: 592 case BRW_REGISTER_TYPE_UD: 593 reg->d = -reg->d; 594 return true; 595 case BRW_REGISTER_TYPE_W: 596 case BRW_REGISTER_TYPE_UW: { 597 uint16_t value = -(int16_t)reg->ud; 598 reg->ud = value | (uint32_t)value << 16; 599 return true; 600 } 601 case BRW_REGISTER_TYPE_F: 602 reg->f = -reg->f; 603 return true; 604 case BRW_REGISTER_TYPE_VF: 605 reg->ud ^= 0x80808080; 606 return true; 607 case BRW_REGISTER_TYPE_DF: 608 reg->df = -reg->df; 609 return true; 610 case BRW_REGISTER_TYPE_UQ: 611 case BRW_REGISTER_TYPE_Q: 612 reg->d64 = -reg->d64; 613 return true; 614 case BRW_REGISTER_TYPE_UB: 615 case BRW_REGISTER_TYPE_B: 616 unreachable("no UB/B immediates"); 617 case BRW_REGISTER_TYPE_UV: 618 case BRW_REGISTER_TYPE_V: 619 assert(!"unimplemented: negate UV/V immediate"); 620 case BRW_REGISTER_TYPE_HF: 621 reg->ud ^= 0x80008000; 622 return true; 623 case BRW_REGISTER_TYPE_NF: 624 unreachable("no NF immediates"); 625 } 626 627 return false; 628} 629 630bool 631brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg) 632{ 633 switch (type) { 634 case BRW_REGISTER_TYPE_D: 635 reg->d = abs(reg->d); 636 return true; 637 case BRW_REGISTER_TYPE_W: { 638 uint16_t value = abs((int16_t)reg->ud); 639 reg->ud = value | (uint32_t)value << 16; 640 return true; 641 } 642 case BRW_REGISTER_TYPE_F: 643 reg->f = fabsf(reg->f); 644 return true; 645 case BRW_REGISTER_TYPE_DF: 646 reg->df = fabs(reg->df); 647 return true; 648 case BRW_REGISTER_TYPE_VF: 649 reg->ud &= ~0x80808080; 650 return true; 651 case BRW_REGISTER_TYPE_Q: 652 reg->d64 = imaxabs(reg->d64); 653 return true; 654 case BRW_REGISTER_TYPE_UB: 655 case BRW_REGISTER_TYPE_B: 656 unreachable("no UB/B immediates"); 657 case BRW_REGISTER_TYPE_UQ: 658 case BRW_REGISTER_TYPE_UD: 659 case BRW_REGISTER_TYPE_UW: 660 case BRW_REGISTER_TYPE_UV: 661 /* Presumably the absolute value modifier on an unsigned source is a 662 * nop, but it would be nice to confirm. 663 */ 664 assert(!"unimplemented: abs unsigned immediate"); 665 case BRW_REGISTER_TYPE_V: 666 assert(!"unimplemented: abs V immediate"); 667 case BRW_REGISTER_TYPE_HF: 668 reg->ud &= ~0x80008000; 669 return true; 670 case BRW_REGISTER_TYPE_NF: 671 unreachable("no NF immediates"); 672 } 673 674 return false; 675} 676 677backend_shader::backend_shader(const struct brw_compiler *compiler, 678 void *log_data, 679 void *mem_ctx, 680 const nir_shader *shader, 681 struct brw_stage_prog_data *stage_prog_data) 682 : compiler(compiler), 683 log_data(log_data), 684 devinfo(compiler->devinfo), 685 nir(shader), 686 stage_prog_data(stage_prog_data), 687 mem_ctx(mem_ctx), 688 cfg(NULL), 689 stage(shader->info.stage) 690{ 691 debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage); 692 stage_name = _mesa_shader_stage_to_string(stage); 693 stage_abbrev = _mesa_shader_stage_to_abbrev(stage); 694} 695 696backend_shader::~backend_shader() 697{ 698} 699 700bool 701backend_reg::equals(const backend_reg &r) const 702{ 703 return brw_regs_equal(this, &r) && offset == r.offset; 704} 705 706bool 707backend_reg::negative_equals(const backend_reg &r) const 708{ 709 return brw_regs_negative_equal(this, &r) && offset == r.offset; 710} 711 712bool 713backend_reg::is_zero() const 714{ 715 if (file != IMM) 716 return false; 717 718 assert(type_sz(type) > 1); 719 720 switch (type) { 721 case BRW_REGISTER_TYPE_HF: 722 assert((d & 0xffff) == ((d >> 16) & 0xffff)); 723 return (d & 0xffff) == 0 || (d & 0xffff) == 0x8000; 724 case BRW_REGISTER_TYPE_F: 725 return f == 0; 726 case BRW_REGISTER_TYPE_DF: 727 return df == 0; 728 case BRW_REGISTER_TYPE_W: 729 case BRW_REGISTER_TYPE_UW: 730 assert((d & 0xffff) == ((d >> 16) & 0xffff)); 731 return (d & 0xffff) == 0; 732 case BRW_REGISTER_TYPE_D: 733 case BRW_REGISTER_TYPE_UD: 734 return d == 0; 735 case BRW_REGISTER_TYPE_UQ: 736 case BRW_REGISTER_TYPE_Q: 737 return u64 == 0; 738 default: 739 return false; 740 } 741} 742 743bool 744backend_reg::is_one() const 745{ 746 if (file != IMM) 747 return false; 748 749 assert(type_sz(type) > 1); 750 751 switch (type) { 752 case BRW_REGISTER_TYPE_HF: 753 assert((d & 0xffff) == ((d >> 16) & 0xffff)); 754 return (d & 0xffff) == 0x3c00; 755 case BRW_REGISTER_TYPE_F: 756 return f == 1.0f; 757 case BRW_REGISTER_TYPE_DF: 758 return df == 1.0; 759 case BRW_REGISTER_TYPE_W: 760 case BRW_REGISTER_TYPE_UW: 761 assert((d & 0xffff) == ((d >> 16) & 0xffff)); 762 return (d & 0xffff) == 1; 763 case BRW_REGISTER_TYPE_D: 764 case BRW_REGISTER_TYPE_UD: 765 return d == 1; 766 case BRW_REGISTER_TYPE_UQ: 767 case BRW_REGISTER_TYPE_Q: 768 return u64 == 1; 769 default: 770 return false; 771 } 772} 773 774bool 775backend_reg::is_negative_one() const 776{ 777 if (file != IMM) 778 return false; 779 780 assert(type_sz(type) > 1); 781 782 switch (type) { 783 case BRW_REGISTER_TYPE_HF: 784 assert((d & 0xffff) == ((d >> 16) & 0xffff)); 785 return (d & 0xffff) == 0xbc00; 786 case BRW_REGISTER_TYPE_F: 787 return f == -1.0; 788 case BRW_REGISTER_TYPE_DF: 789 return df == -1.0; 790 case BRW_REGISTER_TYPE_W: 791 assert((d & 0xffff) == ((d >> 16) & 0xffff)); 792 return (d & 0xffff) == 0xffff; 793 case BRW_REGISTER_TYPE_D: 794 return d == -1; 795 case BRW_REGISTER_TYPE_Q: 796 return d64 == -1; 797 default: 798 return false; 799 } 800} 801 802bool 803backend_reg::is_null() const 804{ 805 return file == ARF && nr == BRW_ARF_NULL; 806} 807 808 809bool 810backend_reg::is_accumulator() const 811{ 812 return file == ARF && nr == BRW_ARF_ACCUMULATOR; 813} 814 815bool 816backend_instruction::is_commutative() const 817{ 818 switch (opcode) { 819 case BRW_OPCODE_AND: 820 case BRW_OPCODE_OR: 821 case BRW_OPCODE_XOR: 822 case BRW_OPCODE_ADD: 823 case BRW_OPCODE_MUL: 824 case SHADER_OPCODE_MULH: 825 return true; 826 case BRW_OPCODE_SEL: 827 /* MIN and MAX are commutative. */ 828 if (conditional_mod == BRW_CONDITIONAL_GE || 829 conditional_mod == BRW_CONDITIONAL_L) { 830 return true; 831 } 832 /* fallthrough */ 833 default: 834 return false; 835 } 836} 837 838bool 839backend_instruction::is_3src(const struct gen_device_info *devinfo) const 840{ 841 return ::is_3src(devinfo, opcode); 842} 843 844bool 845backend_instruction::is_tex() const 846{ 847 return (opcode == SHADER_OPCODE_TEX || 848 opcode == FS_OPCODE_TXB || 849 opcode == SHADER_OPCODE_TXD || 850 opcode == SHADER_OPCODE_TXF || 851 opcode == SHADER_OPCODE_TXF_LZ || 852 opcode == SHADER_OPCODE_TXF_CMS || 853 opcode == SHADER_OPCODE_TXF_CMS_W || 854 opcode == SHADER_OPCODE_TXF_UMS || 855 opcode == SHADER_OPCODE_TXF_MCS || 856 opcode == SHADER_OPCODE_TXL || 857 opcode == SHADER_OPCODE_TXL_LZ || 858 opcode == SHADER_OPCODE_TXS || 859 opcode == SHADER_OPCODE_LOD || 860 opcode == SHADER_OPCODE_TG4 || 861 opcode == SHADER_OPCODE_TG4_OFFSET || 862 opcode == SHADER_OPCODE_SAMPLEINFO); 863} 864 865bool 866backend_instruction::is_math() const 867{ 868 return (opcode == SHADER_OPCODE_RCP || 869 opcode == SHADER_OPCODE_RSQ || 870 opcode == SHADER_OPCODE_SQRT || 871 opcode == SHADER_OPCODE_EXP2 || 872 opcode == SHADER_OPCODE_LOG2 || 873 opcode == SHADER_OPCODE_SIN || 874 opcode == SHADER_OPCODE_COS || 875 opcode == SHADER_OPCODE_INT_QUOTIENT || 876 opcode == SHADER_OPCODE_INT_REMAINDER || 877 opcode == SHADER_OPCODE_POW); 878} 879 880bool 881backend_instruction::is_control_flow() const 882{ 883 switch (opcode) { 884 case BRW_OPCODE_DO: 885 case BRW_OPCODE_WHILE: 886 case BRW_OPCODE_IF: 887 case BRW_OPCODE_ELSE: 888 case BRW_OPCODE_ENDIF: 889 case BRW_OPCODE_BREAK: 890 case BRW_OPCODE_CONTINUE: 891 return true; 892 default: 893 return false; 894 } 895} 896 897bool 898backend_instruction::can_do_source_mods() const 899{ 900 switch (opcode) { 901 case BRW_OPCODE_ADDC: 902 case BRW_OPCODE_BFE: 903 case BRW_OPCODE_BFI1: 904 case BRW_OPCODE_BFI2: 905 case BRW_OPCODE_BFREV: 906 case BRW_OPCODE_CBIT: 907 case BRW_OPCODE_FBH: 908 case BRW_OPCODE_FBL: 909 case BRW_OPCODE_SUBB: 910 case SHADER_OPCODE_BROADCAST: 911 case SHADER_OPCODE_CLUSTER_BROADCAST: 912 case SHADER_OPCODE_MOV_INDIRECT: 913 return false; 914 default: 915 return true; 916 } 917} 918 919bool 920backend_instruction::can_do_saturate() const 921{ 922 switch (opcode) { 923 case BRW_OPCODE_ADD: 924 case BRW_OPCODE_ASR: 925 case BRW_OPCODE_AVG: 926 case BRW_OPCODE_DP2: 927 case BRW_OPCODE_DP3: 928 case BRW_OPCODE_DP4: 929 case BRW_OPCODE_DPH: 930 case BRW_OPCODE_F16TO32: 931 case BRW_OPCODE_F32TO16: 932 case BRW_OPCODE_LINE: 933 case BRW_OPCODE_LRP: 934 case BRW_OPCODE_MAC: 935 case BRW_OPCODE_MAD: 936 case BRW_OPCODE_MATH: 937 case BRW_OPCODE_MOV: 938 case BRW_OPCODE_MUL: 939 case SHADER_OPCODE_MULH: 940 case BRW_OPCODE_PLN: 941 case BRW_OPCODE_RNDD: 942 case BRW_OPCODE_RNDE: 943 case BRW_OPCODE_RNDU: 944 case BRW_OPCODE_RNDZ: 945 case BRW_OPCODE_SEL: 946 case BRW_OPCODE_SHL: 947 case BRW_OPCODE_SHR: 948 case FS_OPCODE_LINTERP: 949 case SHADER_OPCODE_COS: 950 case SHADER_OPCODE_EXP2: 951 case SHADER_OPCODE_LOG2: 952 case SHADER_OPCODE_POW: 953 case SHADER_OPCODE_RCP: 954 case SHADER_OPCODE_RSQ: 955 case SHADER_OPCODE_SIN: 956 case SHADER_OPCODE_SQRT: 957 return true; 958 default: 959 return false; 960 } 961} 962 963bool 964backend_instruction::can_do_cmod() const 965{ 966 switch (opcode) { 967 case BRW_OPCODE_ADD: 968 case BRW_OPCODE_ADDC: 969 case BRW_OPCODE_AND: 970 case BRW_OPCODE_ASR: 971 case BRW_OPCODE_AVG: 972 case BRW_OPCODE_CMP: 973 case BRW_OPCODE_CMPN: 974 case BRW_OPCODE_DP2: 975 case BRW_OPCODE_DP3: 976 case BRW_OPCODE_DP4: 977 case BRW_OPCODE_DPH: 978 case BRW_OPCODE_F16TO32: 979 case BRW_OPCODE_F32TO16: 980 case BRW_OPCODE_FRC: 981 case BRW_OPCODE_LINE: 982 case BRW_OPCODE_LRP: 983 case BRW_OPCODE_LZD: 984 case BRW_OPCODE_MAC: 985 case BRW_OPCODE_MACH: 986 case BRW_OPCODE_MAD: 987 case BRW_OPCODE_MOV: 988 case BRW_OPCODE_MUL: 989 case BRW_OPCODE_NOT: 990 case BRW_OPCODE_OR: 991 case BRW_OPCODE_PLN: 992 case BRW_OPCODE_RNDD: 993 case BRW_OPCODE_RNDE: 994 case BRW_OPCODE_RNDU: 995 case BRW_OPCODE_RNDZ: 996 case BRW_OPCODE_SAD2: 997 case BRW_OPCODE_SADA2: 998 case BRW_OPCODE_SHL: 999 case BRW_OPCODE_SHR: 1000 case BRW_OPCODE_SUBB: 1001 case BRW_OPCODE_XOR: 1002 case FS_OPCODE_LINTERP: 1003 return true; 1004 default: 1005 return false; 1006 } 1007} 1008 1009bool 1010backend_instruction::reads_accumulator_implicitly() const 1011{ 1012 switch (opcode) { 1013 case BRW_OPCODE_MAC: 1014 case BRW_OPCODE_MACH: 1015 case BRW_OPCODE_SADA2: 1016 return true; 1017 default: 1018 return false; 1019 } 1020} 1021 1022bool 1023backend_instruction::writes_accumulator_implicitly(const struct gen_device_info *devinfo) const 1024{ 1025 return writes_accumulator || 1026 (devinfo->gen < 6 && 1027 ((opcode >= BRW_OPCODE_ADD && opcode < BRW_OPCODE_NOP) || 1028 (opcode >= FS_OPCODE_DDX_COARSE && opcode <= FS_OPCODE_LINTERP))) || 1029 (opcode == FS_OPCODE_LINTERP && 1030 (!devinfo->has_pln || devinfo->gen <= 6)); 1031} 1032 1033bool 1034backend_instruction::has_side_effects() const 1035{ 1036 switch (opcode) { 1037 case SHADER_OPCODE_SEND: 1038 return send_has_side_effects; 1039 1040 case VEC4_OPCODE_UNTYPED_ATOMIC: 1041 case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL: 1042 case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL: 1043 case SHADER_OPCODE_GEN4_SCRATCH_WRITE: 1044 case VEC4_OPCODE_UNTYPED_SURFACE_WRITE: 1045 case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL: 1046 case SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL: 1047 case SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL: 1048 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL: 1049 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_INT64_LOGICAL: 1050 case SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL: 1051 case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL: 1052 case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL: 1053 case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL: 1054 case SHADER_OPCODE_MEMORY_FENCE: 1055 case SHADER_OPCODE_INTERLOCK: 1056 case SHADER_OPCODE_URB_WRITE_SIMD8: 1057 case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT: 1058 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED: 1059 case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT: 1060 case FS_OPCODE_FB_WRITE: 1061 case FS_OPCODE_FB_WRITE_LOGICAL: 1062 case FS_OPCODE_REP_FB_WRITE: 1063 case SHADER_OPCODE_BARRIER: 1064 case TCS_OPCODE_URB_WRITE: 1065 case TCS_OPCODE_RELEASE_INPUT: 1066 case SHADER_OPCODE_RND_MODE: 1067 return true; 1068 default: 1069 return eot; 1070 } 1071} 1072 1073bool 1074backend_instruction::is_volatile() const 1075{ 1076 switch (opcode) { 1077 case SHADER_OPCODE_SEND: 1078 return send_is_volatile; 1079 1080 case VEC4_OPCODE_UNTYPED_SURFACE_READ: 1081 case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL: 1082 case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL: 1083 case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL: 1084 case SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL: 1085 case SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL: 1086 case SHADER_OPCODE_URB_READ_SIMD8: 1087 case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT: 1088 case VEC4_OPCODE_URB_READ: 1089 return true; 1090 default: 1091 return false; 1092 } 1093} 1094 1095#ifndef NDEBUG 1096static bool 1097inst_is_in_block(const bblock_t *block, const backend_instruction *inst) 1098{ 1099 bool found = false; 1100 foreach_inst_in_block (backend_instruction, i, block) { 1101 if (inst == i) { 1102 found = true; 1103 } 1104 } 1105 return found; 1106} 1107#endif 1108 1109static void 1110adjust_later_block_ips(bblock_t *start_block, int ip_adjustment) 1111{ 1112 for (bblock_t *block_iter = start_block->next(); 1113 block_iter; 1114 block_iter = block_iter->next()) { 1115 block_iter->start_ip += ip_adjustment; 1116 block_iter->end_ip += ip_adjustment; 1117 } 1118} 1119 1120void 1121backend_instruction::insert_after(bblock_t *block, backend_instruction *inst) 1122{ 1123 assert(this != inst); 1124 1125 if (!this->is_head_sentinel()) 1126 assert(inst_is_in_block(block, this) || !"Instruction not in block"); 1127 1128 block->end_ip++; 1129 1130 adjust_later_block_ips(block, 1); 1131 1132 exec_node::insert_after(inst); 1133} 1134 1135void 1136backend_instruction::insert_before(bblock_t *block, backend_instruction *inst) 1137{ 1138 assert(this != inst); 1139 1140 if (!this->is_tail_sentinel()) 1141 assert(inst_is_in_block(block, this) || !"Instruction not in block"); 1142 1143 block->end_ip++; 1144 1145 adjust_later_block_ips(block, 1); 1146 1147 exec_node::insert_before(inst); 1148} 1149 1150void 1151backend_instruction::insert_before(bblock_t *block, exec_list *list) 1152{ 1153 assert(inst_is_in_block(block, this) || !"Instruction not in block"); 1154 1155 unsigned num_inst = list->length(); 1156 1157 block->end_ip += num_inst; 1158 1159 adjust_later_block_ips(block, num_inst); 1160 1161 exec_node::insert_before(list); 1162} 1163 1164void 1165backend_instruction::remove(bblock_t *block) 1166{ 1167 assert(inst_is_in_block(block, this) || !"Instruction not in block"); 1168 1169 adjust_later_block_ips(block, -1); 1170 1171 if (block->start_ip == block->end_ip) { 1172 block->cfg->remove_block(block); 1173 } else { 1174 block->end_ip--; 1175 } 1176 1177 exec_node::remove(); 1178} 1179 1180void 1181backend_shader::dump_instructions() 1182{ 1183 dump_instructions(NULL); 1184} 1185 1186void 1187backend_shader::dump_instructions(const char *name) 1188{ 1189 FILE *file = stderr; 1190 if (name && geteuid() != 0) { 1191 file = fopen(name, "w"); 1192 if (!file) 1193 file = stderr; 1194 } 1195 1196 if (cfg) { 1197 int ip = 0; 1198 foreach_block_and_inst(block, backend_instruction, inst, cfg) { 1199 if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER)) 1200 fprintf(file, "%4d: ", ip++); 1201 dump_instruction(inst, file); 1202 } 1203 } else { 1204 int ip = 0; 1205 foreach_in_list(backend_instruction, inst, &instructions) { 1206 if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER)) 1207 fprintf(file, "%4d: ", ip++); 1208 dump_instruction(inst, file); 1209 } 1210 } 1211 1212 if (file != stderr) { 1213 fclose(file); 1214 } 1215} 1216 1217void 1218backend_shader::calculate_cfg() 1219{ 1220 if (this->cfg) 1221 return; 1222 cfg = new(mem_ctx) cfg_t(&this->instructions); 1223} 1224 1225extern "C" const unsigned * 1226brw_compile_tes(const struct brw_compiler *compiler, 1227 void *log_data, 1228 void *mem_ctx, 1229 const struct brw_tes_prog_key *key, 1230 const struct brw_vue_map *input_vue_map, 1231 struct brw_tes_prog_data *prog_data, 1232 nir_shader *nir, 1233 struct gl_program *prog, 1234 int shader_time_index, 1235 char **error_str) 1236{ 1237 const struct gen_device_info *devinfo = compiler->devinfo; 1238 const bool is_scalar = compiler->scalar_stage[MESA_SHADER_TESS_EVAL]; 1239 const unsigned *assembly; 1240 1241 nir->info.inputs_read = key->inputs_read; 1242 nir->info.patch_inputs_read = key->patch_inputs_read; 1243 1244 nir = brw_nir_apply_sampler_key(nir, compiler, &key->tex, is_scalar); 1245 brw_nir_lower_tes_inputs(nir, input_vue_map); 1246 brw_nir_lower_vue_outputs(nir); 1247 nir = brw_postprocess_nir(nir, compiler, is_scalar); 1248 1249 brw_compute_vue_map(devinfo, &prog_data->base.vue_map, 1250 nir->info.outputs_written, 1251 nir->info.separate_shader); 1252 1253 unsigned output_size_bytes = prog_data->base.vue_map.num_slots * 4 * 4; 1254 1255 assert(output_size_bytes >= 1); 1256 if (output_size_bytes > GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES) { 1257 if (error_str) 1258 *error_str = ralloc_strdup(mem_ctx, "DS outputs exceed maximum size"); 1259 return NULL; 1260 } 1261 1262 prog_data->base.clip_distance_mask = 1263 ((1 << nir->info.clip_distance_array_size) - 1); 1264 prog_data->base.cull_distance_mask = 1265 ((1 << nir->info.cull_distance_array_size) - 1) << 1266 nir->info.clip_distance_array_size; 1267 1268 /* URB entry sizes are stored as a multiple of 64 bytes. */ 1269 prog_data->base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64; 1270 1271 /* On Cannonlake software shall not program an allocation size that 1272 * specifies a size that is a multiple of 3 64B (512-bit) cachelines. 1273 */ 1274 if (devinfo->gen == 10 && 1275 prog_data->base.urb_entry_size % 3 == 0) 1276 prog_data->base.urb_entry_size++; 1277 1278 prog_data->base.urb_read_length = 0; 1279 1280 STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER == TESS_SPACING_EQUAL - 1); 1281 STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL == 1282 TESS_SPACING_FRACTIONAL_ODD - 1); 1283 STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL == 1284 TESS_SPACING_FRACTIONAL_EVEN - 1); 1285 1286 prog_data->partitioning = 1287 (enum brw_tess_partitioning) (nir->info.tess.spacing - 1); 1288 1289 switch (nir->info.tess.primitive_mode) { 1290 case GL_QUADS: 1291 prog_data->domain = BRW_TESS_DOMAIN_QUAD; 1292 break; 1293 case GL_TRIANGLES: 1294 prog_data->domain = BRW_TESS_DOMAIN_TRI; 1295 break; 1296 case GL_ISOLINES: 1297 prog_data->domain = BRW_TESS_DOMAIN_ISOLINE; 1298 break; 1299 default: 1300 unreachable("invalid domain shader primitive mode"); 1301 } 1302 1303 if (nir->info.tess.point_mode) { 1304 prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT; 1305 } else if (nir->info.tess.primitive_mode == GL_ISOLINES) { 1306 prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE; 1307 } else { 1308 /* Hardware winding order is backwards from OpenGL */ 1309 prog_data->output_topology = 1310 nir->info.tess.ccw ? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW 1311 : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW; 1312 } 1313 1314 if (unlikely(INTEL_DEBUG & DEBUG_TES)) { 1315 fprintf(stderr, "TES Input "); 1316 brw_print_vue_map(stderr, input_vue_map); 1317 fprintf(stderr, "TES Output "); 1318 brw_print_vue_map(stderr, &prog_data->base.vue_map); 1319 } 1320 1321 if (is_scalar) { 1322 fs_visitor v(compiler, log_data, mem_ctx, (void *) key, 1323 &prog_data->base.base, NULL, nir, 8, 1324 shader_time_index, input_vue_map); 1325 if (!v.run_tes()) { 1326 if (error_str) 1327 *error_str = ralloc_strdup(mem_ctx, v.fail_msg); 1328 return NULL; 1329 } 1330 1331 prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs; 1332 prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8; 1333 1334 fs_generator g(compiler, log_data, mem_ctx, 1335 &prog_data->base.base, v.promoted_constants, false, 1336 MESA_SHADER_TESS_EVAL); 1337 if (unlikely(INTEL_DEBUG & DEBUG_TES)) { 1338 g.enable_debug(ralloc_asprintf(mem_ctx, 1339 "%s tessellation evaluation shader %s", 1340 nir->info.label ? nir->info.label 1341 : "unnamed", 1342 nir->info.name)); 1343 } 1344 1345 g.generate_code(v.cfg, 8); 1346 1347 assembly = g.get_assembly(); 1348 } else { 1349 brw::vec4_tes_visitor v(compiler, log_data, key, prog_data, 1350 nir, mem_ctx, shader_time_index); 1351 if (!v.run()) { 1352 if (error_str) 1353 *error_str = ralloc_strdup(mem_ctx, v.fail_msg); 1354 return NULL; 1355 } 1356 1357 if (unlikely(INTEL_DEBUG & DEBUG_TES)) 1358 v.dump_instructions(); 1359 1360 assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir, 1361 &prog_data->base, v.cfg); 1362 } 1363 1364 return assembly; 1365} 1366