src/intel/compiler/brw_shader.cpp

   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 "common/gen_debug.h"
  30 #include "main/uniforms.h"
  31 #include "util/macros.h"
  32
  33 enum brw_reg_type
  34 brw_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_UINT:
  48       return BRW_REGISTER_TYPE_UD;
  49    case GLSL_TYPE_UINT16:
  50       return BRW_REGISTER_TYPE_UW;
  51    case GLSL_TYPE_ARRAY:
  52       return brw_type_for_base_type(type->fields.array);
  53    case GLSL_TYPE_STRUCT:
  54    case GLSL_TYPE_SAMPLER:
  55    case GLSL_TYPE_ATOMIC_UINT:
  56       /* These should be overridden with the type of the member when
  57        * dereferenced into.  BRW_REGISTER_TYPE_UD seems like a likely
  58        * way to trip up if we don't.
  59        */
  60       return BRW_REGISTER_TYPE_UD;
  61    case GLSL_TYPE_IMAGE:
  62       return BRW_REGISTER_TYPE_UD;
  63    case GLSL_TYPE_DOUBLE:
  64       return BRW_REGISTER_TYPE_DF;
  65    case GLSL_TYPE_UINT64:
  66       return BRW_REGISTER_TYPE_UQ;
  67    case GLSL_TYPE_INT64:
  68       return BRW_REGISTER_TYPE_Q;
  69    case GLSL_TYPE_VOID:
  70    case GLSL_TYPE_ERROR:
  71    case GLSL_TYPE_INTERFACE:
  72    case GLSL_TYPE_FUNCTION:
  73       unreachable("not reached");
  74    }
  75
  76    return BRW_REGISTER_TYPE_F;
  77 }
  78
  79 enum brw_conditional_mod
  80 brw_conditional_for_comparison(unsigned int op)
  81 {
  82    switch (op) {
  83    case ir_binop_less:
  84       return BRW_CONDITIONAL_L;
  85    case ir_binop_gequal:
  86       return BRW_CONDITIONAL_GE;
  87    case ir_binop_equal:
  88    case ir_binop_all_equal: /* same as equal for scalars */
  89       return BRW_CONDITIONAL_Z;
  90    case ir_binop_nequal:
  91    case ir_binop_any_nequal: /* same as nequal for scalars */
  92       return BRW_CONDITIONAL_NZ;
  93    default:
  94       unreachable("not reached: bad operation for comparison");
  95    }
  96 }
  97
  98 uint32_t
  99 brw_math_function(enum opcode op)
 100 {
 101    switch (op) {
 102    case SHADER_OPCODE_RCP:
 103       return BRW_MATH_FUNCTION_INV;
 104    case SHADER_OPCODE_RSQ:
 105       return BRW_MATH_FUNCTION_RSQ;
 106    case SHADER_OPCODE_SQRT:
 107       return BRW_MATH_FUNCTION_SQRT;
 108    case SHADER_OPCODE_EXP2:
 109       return BRW_MATH_FUNCTION_EXP;
 110    case SHADER_OPCODE_LOG2:
 111       return BRW_MATH_FUNCTION_LOG;
 112    case SHADER_OPCODE_POW:
 113       return BRW_MATH_FUNCTION_POW;
 114    case SHADER_OPCODE_SIN:
 115       return BRW_MATH_FUNCTION_SIN;
 116    case SHADER_OPCODE_COS:
 117       return BRW_MATH_FUNCTION_COS;
 118    case SHADER_OPCODE_INT_QUOTIENT:
 119       return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
 120    case SHADER_OPCODE_INT_REMAINDER:
 121       return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
 122    default:
 123       unreachable("not reached: unknown math function");
 124    }
 125 }
 126
 127 bool
 128 brw_texture_offset(int *offsets, unsigned num_components, uint32_t *offset_bits)
 129 {
 130    if (!offsets) return false;  /* nonconstant offset; caller will handle it. */
 131
 132    /* offset out of bounds; caller will handle it. */
 133    for (unsigned i = 0; i < num_components; i++)
 134       if (offsets[i] > 7 || offsets[i] < -8)
 135          return false;
 136
 137    /* Combine all three offsets into a single unsigned dword:
 138     *
 139     *    bits 11:8 - U Offset (X component)
 140     *    bits  7:4 - V Offset (Y component)
 141     *    bits  3:0 - R Offset (Z component)
 142     */
 143    *offset_bits = 0;
 144    for (unsigned i = 0; i < num_components; i++) {
 145       const unsigned shift = 4 * (2 - i);
 146       *offset_bits |= (offsets[i] << shift) & (0xF << shift);
 147    }
 148    return true;
 149 }
 150
 151 const char *
 152 brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op)
 153 {
 154    switch (op) {
 155    case BRW_OPCODE_ILLEGAL ... BRW_OPCODE_NOP:
 156       /* The DO instruction doesn't exist on Gen6+, but we use it to mark the
 157        * start of a loop in the IR.
 158        */
 159       if (devinfo->gen >= 6 && op == BRW_OPCODE_DO)
 160          return "do";
 161
 162       /* The following conversion opcodes doesn't exist on Gen8+, but we use
 163        * then to mark that we want to do the conversion.
 164        */
 165       if (devinfo->gen > 7 && op == BRW_OPCODE_F32TO16)
 166          return "f32to16";
 167
 168       if (devinfo->gen > 7 && op == BRW_OPCODE_F16TO32)
 169          return "f16to32";
 170
 171       assert(brw_opcode_desc(devinfo, op)->name);
 172       return brw_opcode_desc(devinfo, op)->name;
 173    case FS_OPCODE_FB_WRITE:
 174       return "fb_write";
 175    case FS_OPCODE_FB_WRITE_LOGICAL:
 176       return "fb_write_logical";
 177    case FS_OPCODE_REP_FB_WRITE:
 178       return "rep_fb_write";
 179    case FS_OPCODE_FB_READ:
 180       return "fb_read";
 181    case FS_OPCODE_FB_READ_LOGICAL:
 182       return "fb_read_logical";
 183
 184    case SHADER_OPCODE_RCP:
 185       return "rcp";
 186    case SHADER_OPCODE_RSQ:
 187       return "rsq";
 188    case SHADER_OPCODE_SQRT:
 189       return "sqrt";
 190    case SHADER_OPCODE_EXP2:
 191       return "exp2";
 192    case SHADER_OPCODE_LOG2:
 193       return "log2";
 194    case SHADER_OPCODE_POW:
 195       return "pow";
 196    case SHADER_OPCODE_INT_QUOTIENT:
 197       return "int_quot";
 198    case SHADER_OPCODE_INT_REMAINDER:
 199       return "int_rem";
 200    case SHADER_OPCODE_SIN:
 201       return "sin";
 202    case SHADER_OPCODE_COS:
 203       return "cos";
 204
 205    case SHADER_OPCODE_TEX:
 206       return "tex";
 207    case SHADER_OPCODE_TEX_LOGICAL:
 208       return "tex_logical";
 209    case SHADER_OPCODE_TXD:
 210       return "txd";
 211    case SHADER_OPCODE_TXD_LOGICAL:
 212       return "txd_logical";
 213    case SHADER_OPCODE_TXF:
 214       return "txf";
 215    case SHADER_OPCODE_TXF_LOGICAL:
 216       return "txf_logical";
 217    case SHADER_OPCODE_TXF_LZ:
 218       return "txf_lz";
 219    case SHADER_OPCODE_TXL:
 220       return "txl";
 221    case SHADER_OPCODE_TXL_LOGICAL:
 222       return "txl_logical";
 223    case SHADER_OPCODE_TXL_LZ:
 224       return "txl_lz";
 225    case SHADER_OPCODE_TXS:
 226       return "txs";
 227    case SHADER_OPCODE_TXS_LOGICAL:
 228       return "txs_logical";
 229    case FS_OPCODE_TXB:
 230       return "txb";
 231    case FS_OPCODE_TXB_LOGICAL:
 232       return "txb_logical";
 233    case SHADER_OPCODE_TXF_CMS:
 234       return "txf_cms";
 235    case SHADER_OPCODE_TXF_CMS_LOGICAL:
 236       return "txf_cms_logical";
 237    case SHADER_OPCODE_TXF_CMS_W:
 238       return "txf_cms_w";
 239    case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
 240       return "txf_cms_w_logical";
 241    case SHADER_OPCODE_TXF_UMS:
 242       return "txf_ums";
 243    case SHADER_OPCODE_TXF_UMS_LOGICAL:
 244       return "txf_ums_logical";
 245    case SHADER_OPCODE_TXF_MCS:
 246       return "txf_mcs";
 247    case SHADER_OPCODE_TXF_MCS_LOGICAL:
 248       return "txf_mcs_logical";
 249    case SHADER_OPCODE_LOD:
 250       return "lod";
 251    case SHADER_OPCODE_LOD_LOGICAL:
 252       return "lod_logical";
 253    case SHADER_OPCODE_TG4:
 254       return "tg4";
 255    case SHADER_OPCODE_TG4_LOGICAL:
 256       return "tg4_logical";
 257    case SHADER_OPCODE_TG4_OFFSET:
 258       return "tg4_offset";
 259    case SHADER_OPCODE_TG4_OFFSET_LOGICAL:
 260       return "tg4_offset_logical";
 261    case SHADER_OPCODE_SAMPLEINFO:
 262       return "sampleinfo";
 263    case SHADER_OPCODE_SAMPLEINFO_LOGICAL:
 264       return "sampleinfo_logical";
 265
 266    case SHADER_OPCODE_SHADER_TIME_ADD:
 267       return "shader_time_add";
 268
 269    case SHADER_OPCODE_UNTYPED_ATOMIC:
 270       return "untyped_atomic";
 271    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
 272       return "untyped_atomic_logical";
 273    case SHADER_OPCODE_UNTYPED_SURFACE_READ:
 274       return "untyped_surface_read";
 275    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
 276       return "untyped_surface_read_logical";
 277    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
 278       return "untyped_surface_write";
 279    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
 280       return "untyped_surface_write_logical";
 281    case SHADER_OPCODE_TYPED_ATOMIC:
 282       return "typed_atomic";
 283    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
 284       return "typed_atomic_logical";
 285    case SHADER_OPCODE_TYPED_SURFACE_READ:
 286       return "typed_surface_read";
 287    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
 288       return "typed_surface_read_logical";
 289    case SHADER_OPCODE_TYPED_SURFACE_WRITE:
 290       return "typed_surface_write";
 291    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
 292       return "typed_surface_write_logical";
 293    case SHADER_OPCODE_MEMORY_FENCE:
 294       return "memory_fence";
 295
 296    case SHADER_OPCODE_BYTE_SCATTERED_READ:
 297       return "byte_scattered_read";
 298    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
 299       return "byte_scattered_read_logical";
 300    case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
 301       return "byte_scattered_write";
 302    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
 303       return "byte_scattered_write_logical";
 304
 305    case SHADER_OPCODE_LOAD_PAYLOAD:
 306       return "load_payload";
 307    case FS_OPCODE_PACK:
 308       return "pack";
 309
 310    case SHADER_OPCODE_GEN4_SCRATCH_READ:
 311       return "gen4_scratch_read";
 312    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
 313       return "gen4_scratch_write";
 314    case SHADER_OPCODE_GEN7_SCRATCH_READ:
 315       return "gen7_scratch_read";
 316    case SHADER_OPCODE_URB_WRITE_SIMD8:
 317       return "gen8_urb_write_simd8";
 318    case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
 319       return "gen8_urb_write_simd8_per_slot";
 320    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
 321       return "gen8_urb_write_simd8_masked";
 322    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
 323       return "gen8_urb_write_simd8_masked_per_slot";
 324    case SHADER_OPCODE_URB_READ_SIMD8:
 325       return "urb_read_simd8";
 326    case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
 327       return "urb_read_simd8_per_slot";
 328
 329    case SHADER_OPCODE_FIND_LIVE_CHANNEL:
 330       return "find_live_channel";
 331    case SHADER_OPCODE_BROADCAST:
 332       return "broadcast";
 333
 334    case VEC4_OPCODE_MOV_BYTES:
 335       return "mov_bytes";
 336    case VEC4_OPCODE_PACK_BYTES:
 337       return "pack_bytes";
 338    case VEC4_OPCODE_UNPACK_UNIFORM:
 339       return "unpack_uniform";
 340    case VEC4_OPCODE_DOUBLE_TO_F32:
 341       return "double_to_f32";
 342    case VEC4_OPCODE_DOUBLE_TO_D32:
 343       return "double_to_d32";
 344    case VEC4_OPCODE_DOUBLE_TO_U32:
 345       return "double_to_u32";
 346    case VEC4_OPCODE_TO_DOUBLE:
 347       return "single_to_double";
 348    case VEC4_OPCODE_PICK_LOW_32BIT:
 349       return "pick_low_32bit";
 350    case VEC4_OPCODE_PICK_HIGH_32BIT:
 351       return "pick_high_32bit";
 352    case VEC4_OPCODE_SET_LOW_32BIT:
 353       return "set_low_32bit";
 354    case VEC4_OPCODE_SET_HIGH_32BIT:
 355       return "set_high_32bit";
 356
 357    case FS_OPCODE_DDX_COARSE:
 358       return "ddx_coarse";
 359    case FS_OPCODE_DDX_FINE:
 360       return "ddx_fine";
 361    case FS_OPCODE_DDY_COARSE:
 362       return "ddy_coarse";
 363    case FS_OPCODE_DDY_FINE:
 364       return "ddy_fine";
 365
 366    case FS_OPCODE_CINTERP:
 367       return "cinterp";
 368    case FS_OPCODE_LINTERP:
 369       return "linterp";
 370
 371    case FS_OPCODE_PIXEL_X:
 372       return "pixel_x";
 373    case FS_OPCODE_PIXEL_Y:
 374       return "pixel_y";
 375
 376    case FS_OPCODE_GET_BUFFER_SIZE:
 377       return "fs_get_buffer_size";
 378
 379    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
 380       return "uniform_pull_const";
 381    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
 382       return "uniform_pull_const_gen7";
 383    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4:
 384       return "varying_pull_const_gen4";
 385    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
 386       return "varying_pull_const_gen7";
 387    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
 388       return "varying_pull_const_logical";
 389
 390    case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
 391       return "mov_dispatch_to_flags";
 392    case FS_OPCODE_DISCARD_JUMP:
 393       return "discard_jump";
 394
 395    case FS_OPCODE_SET_SAMPLE_ID:
 396       return "set_sample_id";
 397
 398    case FS_OPCODE_PACK_HALF_2x16_SPLIT:
 399       return "pack_half_2x16_split";
 400    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
 401       return "unpack_half_2x16_split_x";
 402    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
 403       return "unpack_half_2x16_split_y";
 404
 405    case FS_OPCODE_PLACEHOLDER_HALT:
 406       return "placeholder_halt";
 407
 408    case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
 409       return "interp_sample";
 410    case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
 411       return "interp_shared_offset";
 412    case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
 413       return "interp_per_slot_offset";
 414
 415    case VS_OPCODE_URB_WRITE:
 416       return "vs_urb_write";
 417    case VS_OPCODE_PULL_CONSTANT_LOAD:
 418       return "pull_constant_load";
 419    case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
 420       return "pull_constant_load_gen7";
 421
 422    case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9:
 423       return "set_simd4x2_header_gen9";
 424
 425    case VS_OPCODE_GET_BUFFER_SIZE:
 426       return "vs_get_buffer_size";
 427
 428    case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
 429       return "unpack_flags_simd4x2";
 430
 431    case GS_OPCODE_URB_WRITE:
 432       return "gs_urb_write";
 433    case GS_OPCODE_URB_WRITE_ALLOCATE:
 434       return "gs_urb_write_allocate";
 435    case GS_OPCODE_THREAD_END:
 436       return "gs_thread_end";
 437    case GS_OPCODE_SET_WRITE_OFFSET:
 438       return "set_write_offset";
 439    case GS_OPCODE_SET_VERTEX_COUNT:
 440       return "set_vertex_count";
 441    case GS_OPCODE_SET_DWORD_2:
 442       return "set_dword_2";
 443    case GS_OPCODE_PREPARE_CHANNEL_MASKS:
 444       return "prepare_channel_masks";
 445    case GS_OPCODE_SET_CHANNEL_MASKS:
 446       return "set_channel_masks";
 447    case GS_OPCODE_GET_INSTANCE_ID:
 448       return "get_instance_id";
 449    case GS_OPCODE_FF_SYNC:
 450       return "ff_sync";
 451    case GS_OPCODE_SET_PRIMITIVE_ID:
 452       return "set_primitive_id";
 453    case GS_OPCODE_SVB_WRITE:
 454       return "gs_svb_write";
 455    case GS_OPCODE_SVB_SET_DST_INDEX:
 456       return "gs_svb_set_dst_index";
 457    case GS_OPCODE_FF_SYNC_SET_PRIMITIVES:
 458       return "gs_ff_sync_set_primitives";
 459    case CS_OPCODE_CS_TERMINATE:
 460       return "cs_terminate";
 461    case SHADER_OPCODE_BARRIER:
 462       return "barrier";
 463    case SHADER_OPCODE_MULH:
 464       return "mulh";
 465    case SHADER_OPCODE_MOV_INDIRECT:
 466       return "mov_indirect";
 467
 468    case VEC4_OPCODE_URB_READ:
 469       return "urb_read";
 470    case TCS_OPCODE_GET_INSTANCE_ID:
 471       return "tcs_get_instance_id";
 472    case TCS_OPCODE_URB_WRITE:
 473       return "tcs_urb_write";
 474    case TCS_OPCODE_SET_INPUT_URB_OFFSETS:
 475       return "tcs_set_input_urb_offsets";
 476    case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS:
 477       return "tcs_set_output_urb_offsets";
 478    case TCS_OPCODE_GET_PRIMITIVE_ID:
 479       return "tcs_get_primitive_id";
 480    case TCS_OPCODE_CREATE_BARRIER_HEADER:
 481       return "tcs_create_barrier_header";
 482    case TCS_OPCODE_SRC0_010_IS_ZERO:
 483       return "tcs_src0<0,1,0>_is_zero";
 484    case TCS_OPCODE_RELEASE_INPUT:
 485       return "tcs_release_input";
 486    case TCS_OPCODE_THREAD_END:
 487       return "tcs_thread_end";
 488    case TES_OPCODE_CREATE_INPUT_READ_HEADER:
 489       return "tes_create_input_read_header";
 490    case TES_OPCODE_ADD_INDIRECT_URB_OFFSET:
 491       return "tes_add_indirect_urb_offset";
 492    case TES_OPCODE_GET_PRIMITIVE_ID:
 493       return "tes_get_primitive_id";
 494
 495    case SHADER_OPCODE_RND_MODE:
 496       return "rnd_mode";
 497    }
 498
 499    unreachable("not reached");
 500 }
 501
 502 bool
 503 brw_saturate_immediate(enum brw_reg_type type, struct brw_reg *reg)
 504 {
 505    union {
 506       unsigned ud;
 507       int d;
 508       float f;
 509       double df;
 510    } imm, sat_imm = { 0 };
 511
 512    const unsigned size = type_sz(type);
 513
 514    /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
 515     * irrelevant, so just check the size of the type and copy from/to an
 516     * appropriately sized field.
 517     */
 518    if (size < 8)
 519       imm.ud = reg->ud;
 520    else
 521       imm.df = reg->df;
 522
 523    switch (type) {
 524    case BRW_REGISTER_TYPE_UD:
 525    case BRW_REGISTER_TYPE_D:
 526    case BRW_REGISTER_TYPE_UW:
 527    case BRW_REGISTER_TYPE_W:
 528    case BRW_REGISTER_TYPE_UQ:
 529    case BRW_REGISTER_TYPE_Q:
 530       /* Nothing to do. */
 531       return false;
 532    case BRW_REGISTER_TYPE_F:
 533       sat_imm.f = CLAMP(imm.f, 0.0f, 1.0f);
 534       break;
 535    case BRW_REGISTER_TYPE_DF:
 536       sat_imm.df = CLAMP(imm.df, 0.0, 1.0);
 537       break;
 538    case BRW_REGISTER_TYPE_UB:
 539    case BRW_REGISTER_TYPE_B:
 540       unreachable("no UB/B immediates");
 541    case BRW_REGISTER_TYPE_V:
 542    case BRW_REGISTER_TYPE_UV:
 543    case BRW_REGISTER_TYPE_VF:
 544       unreachable("unimplemented: saturate vector immediate");
 545    case BRW_REGISTER_TYPE_HF:
 546       unreachable("unimplemented: saturate HF immediate");
 547    }
 548
 549    if (size < 8) {
 550       if (imm.ud != sat_imm.ud) {
 551          reg->ud = sat_imm.ud;
 552          return true;
 553       }
 554    } else {
 555       if (imm.df != sat_imm.df) {
 556          reg->df = sat_imm.df;
 557          return true;
 558       }
 559    }
 560    return false;
 561 }
 562
 563 bool
 564 brw_negate_immediate(enum brw_reg_type type, struct brw_reg *reg)
 565 {
 566    switch (type) {
 567    case BRW_REGISTER_TYPE_D:
 568    case BRW_REGISTER_TYPE_UD:
 569       reg->d = -reg->d;
 570       return true;
 571    case BRW_REGISTER_TYPE_W:
 572    case BRW_REGISTER_TYPE_UW:
 573       reg->d = -(int16_t)reg->ud;
 574       return true;
 575    case BRW_REGISTER_TYPE_F:
 576       reg->f = -reg->f;
 577       return true;
 578    case BRW_REGISTER_TYPE_VF:
 579       reg->ud ^= 0x80808080;
 580       return true;
 581    case BRW_REGISTER_TYPE_DF:
 582       reg->df = -reg->df;
 583       return true;
 584    case BRW_REGISTER_TYPE_UQ:
 585    case BRW_REGISTER_TYPE_Q:
 586       reg->d64 = -reg->d64;
 587       return true;
 588    case BRW_REGISTER_TYPE_UB:
 589    case BRW_REGISTER_TYPE_B:
 590       unreachable("no UB/B immediates");
 591    case BRW_REGISTER_TYPE_UV:
 592    case BRW_REGISTER_TYPE_V:
 593       assert(!"unimplemented: negate UV/V immediate");
 594    case BRW_REGISTER_TYPE_HF:
 595       assert(!"unimplemented: negate HF immediate");
 596    }
 597
 598    return false;
 599 }
 600
 601 bool
 602 brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg)
 603 {
 604    switch (type) {
 605    case BRW_REGISTER_TYPE_D:
 606       reg->d = abs(reg->d);
 607       return true;
 608    case BRW_REGISTER_TYPE_W:
 609       reg->d = abs((int16_t)reg->ud);
 610       return true;
 611    case BRW_REGISTER_TYPE_F:
 612       reg->f = fabsf(reg->f);
 613       return true;
 614    case BRW_REGISTER_TYPE_DF:
 615       reg->df = fabs(reg->df);
 616       return true;
 617    case BRW_REGISTER_TYPE_VF:
 618       reg->ud &= ~0x80808080;
 619       return true;
 620    case BRW_REGISTER_TYPE_Q:
 621       reg->d64 = imaxabs(reg->d64);
 622       return true;
 623    case BRW_REGISTER_TYPE_UB:
 624    case BRW_REGISTER_TYPE_B:
 625       unreachable("no UB/B immediates");
 626    case BRW_REGISTER_TYPE_UQ:
 627    case BRW_REGISTER_TYPE_UD:
 628    case BRW_REGISTER_TYPE_UW:
 629    case BRW_REGISTER_TYPE_UV:
 630       /* Presumably the absolute value modifier on an unsigned source is a
 631        * nop, but it would be nice to confirm.
 632        */
 633       assert(!"unimplemented: abs unsigned immediate");
 634    case BRW_REGISTER_TYPE_V:
 635       assert(!"unimplemented: abs V immediate");
 636    case BRW_REGISTER_TYPE_HF:
 637       assert(!"unimplemented: abs HF immediate");
 638    }
 639
 640    return false;
 641 }
 642
 643 backend_shader::backend_shader(const struct brw_compiler *compiler,
 644                                void *log_data,
 645                                void *mem_ctx,
 646                                const nir_shader *shader,
 647                                struct brw_stage_prog_data *stage_prog_data)
 648    : compiler(compiler),
 649      log_data(log_data),
 650      devinfo(compiler->devinfo),
 651      nir(shader),
 652      stage_prog_data(stage_prog_data),
 653      mem_ctx(mem_ctx),
 654      cfg(NULL),
 655      stage(shader->info.stage)
 656 {
 657    debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
 658    stage_name = _mesa_shader_stage_to_string(stage);
 659    stage_abbrev = _mesa_shader_stage_to_abbrev(stage);
 660 }
 661
 662 backend_shader::~backend_shader()
 663 {
 664 }
 665
 666 bool
 667 backend_reg::equals(const backend_reg &r) const
 668 {
 669    return brw_regs_equal(this, &r) && offset == r.offset;
 670 }
 671
 672 bool
 673 backend_reg::is_zero() const
 674 {
 675    if (file != IMM)
 676       return false;
 677
 678    switch (type) {
 679    case BRW_REGISTER_TYPE_F:
 680       return f == 0;
 681    case BRW_REGISTER_TYPE_DF:
 682       return df == 0;
 683    case BRW_REGISTER_TYPE_D:
 684    case BRW_REGISTER_TYPE_UD:
 685       return d == 0;
 686    case BRW_REGISTER_TYPE_UQ:
 687    case BRW_REGISTER_TYPE_Q:
 688       return u64 == 0;
 689    default:
 690       return false;
 691    }
 692 }
 693
 694 bool
 695 backend_reg::is_one() const
 696 {
 697    if (file != IMM)
 698       return false;
 699
 700    switch (type) {
 701    case BRW_REGISTER_TYPE_F:
 702       return f == 1.0f;
 703    case BRW_REGISTER_TYPE_DF:
 704       return df == 1.0;
 705    case BRW_REGISTER_TYPE_D:
 706    case BRW_REGISTER_TYPE_UD:
 707       return d == 1;
 708    case BRW_REGISTER_TYPE_UQ:
 709    case BRW_REGISTER_TYPE_Q:
 710       return u64 == 1;
 711    default:
 712       return false;
 713    }
 714 }
 715
 716 bool
 717 backend_reg::is_negative_one() const
 718 {
 719    if (file != IMM)
 720       return false;
 721
 722    switch (type) {
 723    case BRW_REGISTER_TYPE_F:
 724       return f == -1.0;
 725    case BRW_REGISTER_TYPE_DF:
 726       return df == -1.0;
 727    case BRW_REGISTER_TYPE_D:
 728       return d == -1;
 729    case BRW_REGISTER_TYPE_Q:
 730       return d64 == -1;
 731    default:
 732       return false;
 733    }
 734 }
 735
 736 bool
 737 backend_reg::is_null() const
 738 {
 739    return file == ARF && nr == BRW_ARF_NULL;
 740 }
 741
 742
 743 bool
 744 backend_reg::is_accumulator() const
 745 {
 746    return file == ARF && nr == BRW_ARF_ACCUMULATOR;
 747 }
 748
 749 bool
 750 backend_instruction::is_commutative() const
 751 {
 752    switch (opcode) {
 753    case BRW_OPCODE_AND:
 754    case BRW_OPCODE_OR:
 755    case BRW_OPCODE_XOR:
 756    case BRW_OPCODE_ADD:
 757    case BRW_OPCODE_MUL:
 758    case SHADER_OPCODE_MULH:
 759       return true;
 760    case BRW_OPCODE_SEL:
 761       /* MIN and MAX are commutative. */
 762       if (conditional_mod == BRW_CONDITIONAL_GE ||
 763           conditional_mod == BRW_CONDITIONAL_L) {
 764          return true;
 765       }
 766       /* fallthrough */
 767    default:
 768       return false;
 769    }
 770 }
 771
 772 bool
 773 backend_instruction::is_3src(const struct gen_device_info *devinfo) const
 774 {
 775    return ::is_3src(devinfo, opcode);
 776 }
 777
 778 bool
 779 backend_instruction::is_tex() const
 780 {
 781    return (opcode == SHADER_OPCODE_TEX ||
 782            opcode == FS_OPCODE_TXB ||
 783            opcode == SHADER_OPCODE_TXD ||
 784            opcode == SHADER_OPCODE_TXF ||
 785            opcode == SHADER_OPCODE_TXF_LZ ||
 786            opcode == SHADER_OPCODE_TXF_CMS ||
 787            opcode == SHADER_OPCODE_TXF_CMS_W ||
 788            opcode == SHADER_OPCODE_TXF_UMS ||
 789            opcode == SHADER_OPCODE_TXF_MCS ||
 790            opcode == SHADER_OPCODE_TXL ||
 791            opcode == SHADER_OPCODE_TXL_LZ ||
 792            opcode == SHADER_OPCODE_TXS ||
 793            opcode == SHADER_OPCODE_LOD ||
 794            opcode == SHADER_OPCODE_TG4 ||
 795            opcode == SHADER_OPCODE_TG4_OFFSET ||
 796            opcode == SHADER_OPCODE_SAMPLEINFO);
 797 }
 798
 799 bool
 800 backend_instruction::is_math() const
 801 {
 802    return (opcode == SHADER_OPCODE_RCP ||
 803            opcode == SHADER_OPCODE_RSQ ||
 804            opcode == SHADER_OPCODE_SQRT ||
 805            opcode == SHADER_OPCODE_EXP2 ||
 806            opcode == SHADER_OPCODE_LOG2 ||
 807            opcode == SHADER_OPCODE_SIN ||
 808            opcode == SHADER_OPCODE_COS ||
 809            opcode == SHADER_OPCODE_INT_QUOTIENT ||
 810            opcode == SHADER_OPCODE_INT_REMAINDER ||
 811            opcode == SHADER_OPCODE_POW);
 812 }
 813
 814 bool
 815 backend_instruction::is_control_flow() const
 816 {
 817    switch (opcode) {
 818    case BRW_OPCODE_DO:
 819    case BRW_OPCODE_WHILE:
 820    case BRW_OPCODE_IF:
 821    case BRW_OPCODE_ELSE:
 822    case BRW_OPCODE_ENDIF:
 823    case BRW_OPCODE_BREAK:
 824    case BRW_OPCODE_CONTINUE:
 825       return true;
 826    default:
 827       return false;
 828    }
 829 }
 830
 831 bool
 832 backend_instruction::can_do_source_mods() const
 833 {
 834    switch (opcode) {
 835    case BRW_OPCODE_ADDC:
 836    case BRW_OPCODE_BFE:
 837    case BRW_OPCODE_BFI1:
 838    case BRW_OPCODE_BFI2:
 839    case BRW_OPCODE_BFREV:
 840    case BRW_OPCODE_CBIT:
 841    case BRW_OPCODE_FBH:
 842    case BRW_OPCODE_FBL:
 843    case BRW_OPCODE_SUBB:
 844    case SHADER_OPCODE_BROADCAST:
 845    case SHADER_OPCODE_MOV_INDIRECT:
 846       return false;
 847    default:
 848       return true;
 849    }
 850 }
 851
 852 bool
 853 backend_instruction::can_do_saturate() const
 854 {
 855    switch (opcode) {
 856    case BRW_OPCODE_ADD:
 857    case BRW_OPCODE_ASR:
 858    case BRW_OPCODE_AVG:
 859    case BRW_OPCODE_DP2:
 860    case BRW_OPCODE_DP3:
 861    case BRW_OPCODE_DP4:
 862    case BRW_OPCODE_DPH:
 863    case BRW_OPCODE_F16TO32:
 864    case BRW_OPCODE_F32TO16:
 865    case BRW_OPCODE_LINE:
 866    case BRW_OPCODE_LRP:
 867    case BRW_OPCODE_MAC:
 868    case BRW_OPCODE_MAD:
 869    case BRW_OPCODE_MATH:
 870    case BRW_OPCODE_MOV:
 871    case BRW_OPCODE_MUL:
 872    case SHADER_OPCODE_MULH:
 873    case BRW_OPCODE_PLN:
 874    case BRW_OPCODE_RNDD:
 875    case BRW_OPCODE_RNDE:
 876    case BRW_OPCODE_RNDU:
 877    case BRW_OPCODE_RNDZ:
 878    case BRW_OPCODE_SEL:
 879    case BRW_OPCODE_SHL:
 880    case BRW_OPCODE_SHR:
 881    case FS_OPCODE_LINTERP:
 882    case SHADER_OPCODE_COS:
 883    case SHADER_OPCODE_EXP2:
 884    case SHADER_OPCODE_LOG2:
 885    case SHADER_OPCODE_POW:
 886    case SHADER_OPCODE_RCP:
 887    case SHADER_OPCODE_RSQ:
 888    case SHADER_OPCODE_SIN:
 889    case SHADER_OPCODE_SQRT:
 890       return true;
 891    default:
 892       return false;
 893    }
 894 }
 895
 896 bool
 897 backend_instruction::can_do_cmod() const
 898 {
 899    switch (opcode) {
 900    case BRW_OPCODE_ADD:
 901    case BRW_OPCODE_ADDC:
 902    case BRW_OPCODE_AND:
 903    case BRW_OPCODE_ASR:
 904    case BRW_OPCODE_AVG:
 905    case BRW_OPCODE_CMP:
 906    case BRW_OPCODE_CMPN:
 907    case BRW_OPCODE_DP2:
 908    case BRW_OPCODE_DP3:
 909    case BRW_OPCODE_DP4:
 910    case BRW_OPCODE_DPH:
 911    case BRW_OPCODE_F16TO32:
 912    case BRW_OPCODE_F32TO16:
 913    case BRW_OPCODE_FRC:
 914    case BRW_OPCODE_LINE:
 915    case BRW_OPCODE_LRP:
 916    case BRW_OPCODE_LZD:
 917    case BRW_OPCODE_MAC:
 918    case BRW_OPCODE_MACH:
 919    case BRW_OPCODE_MAD:
 920    case BRW_OPCODE_MOV:
 921    case BRW_OPCODE_MUL:
 922    case BRW_OPCODE_NOT:
 923    case BRW_OPCODE_OR:
 924    case BRW_OPCODE_PLN:
 925    case BRW_OPCODE_RNDD:
 926    case BRW_OPCODE_RNDE:
 927    case BRW_OPCODE_RNDU:
 928    case BRW_OPCODE_RNDZ:
 929    case BRW_OPCODE_SAD2:
 930    case BRW_OPCODE_SADA2:
 931    case BRW_OPCODE_SHL:
 932    case BRW_OPCODE_SHR:
 933    case BRW_OPCODE_SUBB:
 934    case BRW_OPCODE_XOR:
 935    case FS_OPCODE_CINTERP:
 936    case FS_OPCODE_LINTERP:
 937       return true;
 938    default:
 939       return false;
 940    }
 941 }
 942
 943 bool
 944 backend_instruction::reads_accumulator_implicitly() const
 945 {
 946    switch (opcode) {
 947    case BRW_OPCODE_MAC:
 948    case BRW_OPCODE_MACH:
 949    case BRW_OPCODE_SADA2:
 950       return true;
 951    default:
 952       return false;
 953    }
 954 }
 955
 956 bool
 957 backend_instruction::writes_accumulator_implicitly(const struct gen_device_info *devinfo) const
 958 {
 959    return writes_accumulator ||
 960           (devinfo->gen < 6 &&
 961            ((opcode >= BRW_OPCODE_ADD && opcode < BRW_OPCODE_NOP) ||
 962             (opcode >= FS_OPCODE_DDX_COARSE && opcode <= FS_OPCODE_LINTERP &&
 963              opcode != FS_OPCODE_CINTERP)));
 964 }
 965
 966 bool
 967 backend_instruction::has_side_effects() const
 968 {
 969    switch (opcode) {
 970    case SHADER_OPCODE_UNTYPED_ATOMIC:
 971    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
 972    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
 973    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
 974    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
 975    case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
 976    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
 977    case SHADER_OPCODE_TYPED_ATOMIC:
 978    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
 979    case SHADER_OPCODE_TYPED_SURFACE_WRITE:
 980    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
 981    case SHADER_OPCODE_MEMORY_FENCE:
 982    case SHADER_OPCODE_URB_WRITE_SIMD8:
 983    case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
 984    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
 985    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
 986    case FS_OPCODE_FB_WRITE:
 987    case FS_OPCODE_FB_WRITE_LOGICAL:
 988    case SHADER_OPCODE_BARRIER:
 989    case TCS_OPCODE_URB_WRITE:
 990    case TCS_OPCODE_RELEASE_INPUT:
 991    case SHADER_OPCODE_RND_MODE:
 992       return true;
 993    default:
 994       return eot;
 995    }
 996 }
 997
 998 bool
 999 backend_instruction::is_volatile() const
1000 {
1001    switch (opcode) {
1002    case SHADER_OPCODE_UNTYPED_SURFACE_READ:
1003    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
1004    case SHADER_OPCODE_TYPED_SURFACE_READ:
1005    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
1006    case SHADER_OPCODE_BYTE_SCATTERED_READ:
1007    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
1008    case SHADER_OPCODE_URB_READ_SIMD8:
1009    case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
1010    case VEC4_OPCODE_URB_READ:
1011       return true;
1012    default:
1013       return false;
1014    }
1015 }
1016
1017 #ifndef NDEBUG
1018 static bool
1019 inst_is_in_block(const bblock_t *block, const backend_instruction *inst)
1020 {
1021    bool found = false;
1022    foreach_inst_in_block (backend_instruction, i, block) {
1023       if (inst == i) {
1024          found = true;
1025       }
1026    }
1027    return found;
1028 }
1029 #endif
1030
1031 static void
1032 adjust_later_block_ips(bblock_t *start_block, int ip_adjustment)
1033 {
1034    for (bblock_t *block_iter = start_block->next();
1035         block_iter;
1036         block_iter = block_iter->next()) {
1037       block_iter->start_ip += ip_adjustment;
1038       block_iter->end_ip += ip_adjustment;
1039    }
1040 }
1041
1042 void
1043 backend_instruction::insert_after(bblock_t *block, backend_instruction *inst)
1044 {
1045    assert(this != inst);
1046
1047    if (!this->is_head_sentinel())
1048       assert(inst_is_in_block(block, this) || !"Instruction not in block");
1049
1050    block->end_ip++;
1051
1052    adjust_later_block_ips(block, 1);
1053
1054    exec_node::insert_after(inst);
1055 }
1056
1057 void
1058 backend_instruction::insert_before(bblock_t *block, backend_instruction *inst)
1059 {
1060    assert(this != inst);
1061
1062    if (!this->is_tail_sentinel())
1063       assert(inst_is_in_block(block, this) || !"Instruction not in block");
1064
1065    block->end_ip++;
1066
1067    adjust_later_block_ips(block, 1);
1068
1069    exec_node::insert_before(inst);
1070 }
1071
1072 void
1073 backend_instruction::insert_before(bblock_t *block, exec_list *list)
1074 {
1075    assert(inst_is_in_block(block, this) || !"Instruction not in block");
1076
1077    unsigned num_inst = list->length();
1078
1079    block->end_ip += num_inst;
1080
1081    adjust_later_block_ips(block, num_inst);
1082
1083    exec_node::insert_before(list);
1084 }
1085
1086 void
1087 backend_instruction::remove(bblock_t *block)
1088 {
1089    assert(inst_is_in_block(block, this) || !"Instruction not in block");
1090
1091    adjust_later_block_ips(block, -1);
1092
1093    if (block->start_ip == block->end_ip) {
1094       block->cfg->remove_block(block);
1095    } else {
1096       block->end_ip--;
1097    }
1098
1099    exec_node::remove();
1100 }
1101
1102 void
1103 backend_shader::dump_instructions()
1104 {
1105    dump_instructions(NULL);
1106 }
1107
1108 void
1109 backend_shader::dump_instructions(const char *name)
1110 {
1111    FILE *file = stderr;
1112    if (name && geteuid() != 0) {
1113       file = fopen(name, "w");
1114       if (!file)
1115          file = stderr;
1116    }
1117
1118    if (cfg) {
1119       int ip = 0;
1120       foreach_block_and_inst(block, backend_instruction, inst, cfg) {
1121          if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1122             fprintf(file, "%4d: ", ip++);
1123          dump_instruction(inst, file);
1124       }
1125    } else {
1126       int ip = 0;
1127       foreach_in_list(backend_instruction, inst, &instructions) {
1128          if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1129             fprintf(file, "%4d: ", ip++);
1130          dump_instruction(inst, file);
1131       }
1132    }
1133
1134    if (file != stderr) {
1135       fclose(file);
1136    }
1137 }
1138
1139 void
1140 backend_shader::calculate_cfg()
1141 {
1142    if (this->cfg)
1143       return;
1144    cfg = new(mem_ctx) cfg_t(&this->instructions);
1145 }
1146
1147 extern "C" const unsigned *
1148 brw_compile_tes(const struct brw_compiler *compiler,
1149                 void *log_data,
1150                 void *mem_ctx,
1151                 const struct brw_tes_prog_key *key,
1152                 const struct brw_vue_map *input_vue_map,
1153                 struct brw_tes_prog_data *prog_data,
1154                 const nir_shader *src_shader,
1155                 struct gl_program *prog,
1156                 int shader_time_index,
1157                 char **error_str)
1158 {
1159    const struct gen_device_info *devinfo = compiler->devinfo;
1160    const bool is_scalar = compiler->scalar_stage[MESA_SHADER_TESS_EVAL];
1161    const unsigned *assembly;
1162
1163    nir_shader *nir = nir_shader_clone(mem_ctx, src_shader);
1164    nir->info.inputs_read = key->inputs_read;
1165    nir->info.patch_inputs_read = key->patch_inputs_read;
1166
1167    nir = brw_nir_apply_sampler_key(nir, compiler, &key->tex, is_scalar);
1168    brw_nir_lower_tes_inputs(nir, input_vue_map);
1169    brw_nir_lower_vue_outputs(nir, is_scalar);
1170    nir = brw_postprocess_nir(nir, compiler, is_scalar);
1171
1172    brw_compute_vue_map(devinfo, &prog_data->base.vue_map,
1173                        nir->info.outputs_written,
1174                        nir->info.separate_shader);
1175
1176    unsigned output_size_bytes = prog_data->base.vue_map.num_slots * 4 * 4;
1177
1178    assert(output_size_bytes >= 1);
1179    if (output_size_bytes > GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES) {
1180       if (error_str)
1181          *error_str = ralloc_strdup(mem_ctx, "DS outputs exceed maximum size");
1182       return NULL;
1183    }
1184
1185    prog_data->base.clip_distance_mask =
1186       ((1 << nir->info.clip_distance_array_size) - 1);
1187    prog_data->base.cull_distance_mask =
1188       ((1 << nir->info.cull_distance_array_size) - 1) <<
1189       nir->info.clip_distance_array_size;
1190
1191    /* URB entry sizes are stored as a multiple of 64 bytes. */
1192    prog_data->base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64;
1193
1194    /* On Cannonlake software shall not program an allocation size that
1195     * specifies a size that is a multiple of 3 64B (512-bit) cachelines.
1196     */
1197    if (devinfo->gen == 10 &&
1198        prog_data->base.urb_entry_size % 3 == 0)
1199       prog_data->base.urb_entry_size++;
1200
1201    prog_data->base.urb_read_length = 0;
1202
1203    STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER == TESS_SPACING_EQUAL - 1);
1204    STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL ==
1205                  TESS_SPACING_FRACTIONAL_ODD - 1);
1206    STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL ==
1207                  TESS_SPACING_FRACTIONAL_EVEN - 1);
1208
1209    prog_data->partitioning =
1210       (enum brw_tess_partitioning) (nir->info.tess.spacing - 1);
1211
1212    switch (nir->info.tess.primitive_mode) {
1213    case GL_QUADS:
1214       prog_data->domain = BRW_TESS_DOMAIN_QUAD;
1215       break;
1216    case GL_TRIANGLES:
1217       prog_data->domain = BRW_TESS_DOMAIN_TRI;
1218       break;
1219    case GL_ISOLINES:
1220       prog_data->domain = BRW_TESS_DOMAIN_ISOLINE;
1221       break;
1222    default:
1223       unreachable("invalid domain shader primitive mode");
1224    }
1225
1226    if (nir->info.tess.point_mode) {
1227       prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT;
1228    } else if (nir->info.tess.primitive_mode == GL_ISOLINES) {
1229       prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE;
1230    } else {
1231       /* Hardware winding order is backwards from OpenGL */
1232       prog_data->output_topology =
1233          nir->info.tess.ccw ? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
1234                              : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW;
1235    }
1236
1237    if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1238       fprintf(stderr, "TES Input ");
1239       brw_print_vue_map(stderr, input_vue_map);
1240       fprintf(stderr, "TES Output ");
1241       brw_print_vue_map(stderr, &prog_data->base.vue_map);
1242    }
1243
1244    if (is_scalar) {
1245       fs_visitor v(compiler, log_data, mem_ctx, (void *) key,
1246                    &prog_data->base.base, NULL, nir, 8,
1247                    shader_time_index, input_vue_map);
1248       if (!v.run_tes()) {
1249          if (error_str)
1250             *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1251          return NULL;
1252       }
1253
1254       prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
1255       prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8;
1256
1257       fs_generator g(compiler, log_data, mem_ctx, (void *) key,
1258                      &prog_data->base.base, v.promoted_constants, false,
1259                      MESA_SHADER_TESS_EVAL);
1260       if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1261          g.enable_debug(ralloc_asprintf(mem_ctx,
1262                                         "%s tessellation evaluation shader %s",
1263                                         nir->info.label ? nir->info.label
1264                                                         : "unnamed",
1265                                         nir->info.name));
1266       }
1267
1268       g.generate_code(v.cfg, 8);
1269
1270       assembly = g.get_assembly(&prog_data->base.base.program_size);
1271    } else {
1272       brw::vec4_tes_visitor v(compiler, log_data, key, prog_data,
1273                               nir, mem_ctx, shader_time_index);
1274       if (!v.run()) {
1275          if (error_str)
1276             *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1277          return NULL;
1278       }
1279
1280       if (unlikely(INTEL_DEBUG & DEBUG_TES))
1281          v.dump_instructions();
1282
1283       assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir,
1284                                             &prog_data->base, v.cfg,
1285                                             &prog_data->base.base.program_size);
1286    }
1287
1288    return assembly;
1289 }