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_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_SAMPLER:
  59    case GLSL_TYPE_ATOMIC_UINT:
  60       /* These should be overridden with the type of the member when
  61        * dereferenced into.  BRW_REGISTER_TYPE_UD seems like a likely
  62        * way to trip up if we don't.
  63        */
  64       return BRW_REGISTER_TYPE_UD;
  65    case GLSL_TYPE_IMAGE:
  66       return BRW_REGISTER_TYPE_UD;
  67    case GLSL_TYPE_DOUBLE:
  68       return BRW_REGISTER_TYPE_DF;
  69    case GLSL_TYPE_UINT64:
  70       return BRW_REGISTER_TYPE_UQ;
  71    case GLSL_TYPE_INT64:
  72       return BRW_REGISTER_TYPE_Q;
  73    case GLSL_TYPE_VOID:
  74    case GLSL_TYPE_ERROR:
  75    case GLSL_TYPE_INTERFACE:
  76    case GLSL_TYPE_FUNCTION:
  77       unreachable("not reached");
  78    }
  79
  80    return BRW_REGISTER_TYPE_F;
  81 }
  82
  83 enum brw_conditional_mod
  84 brw_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
 102 uint32_t
 103 brw_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
 131 bool
 132 brw_texture_offset(int *offsets, unsigned num_components, uint32_t *offset_bits)
 133 {
 134    if (!offsets) return false;  /* nonconstant offset; caller will handle it. */
 135
 136    /* offset out of bounds; caller will handle it. */
 137    for (unsigned i = 0; i < num_components; i++)
 138       if (offsets[i] > 7 || offsets[i] < -8)
 139          return false;
 140
 141    /* Combine all three offsets into a single unsigned dword:
 142     *
 143     *    bits 11:8 - U Offset (X component)
 144     *    bits  7:4 - V Offset (Y component)
 145     *    bits  3:0 - R Offset (Z component)
 146     */
 147    *offset_bits = 0;
 148    for (unsigned i = 0; i < num_components; i++) {
 149       const unsigned shift = 4 * (2 - i);
 150       *offset_bits |= (offsets[i] << shift) & (0xF << shift);
 151    }
 152    return true;
 153 }
 154
 155 const char *
 156 brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op)
 157 {
 158    switch (op) {
 159    case BRW_OPCODE_ILLEGAL ... BRW_OPCODE_NOP:
 160       /* The DO instruction doesn't exist on Gen6+, but we use it to mark the
 161        * start of a loop in the IR.
 162        */
 163       if (devinfo->gen >= 6 && op == BRW_OPCODE_DO)
 164          return "do";
 165
 166       /* The following conversion opcodes doesn't exist on Gen8+, but we use
 167        * then to mark that we want to do the conversion.
 168        */
 169       if (devinfo->gen > 7 && op == BRW_OPCODE_F32TO16)
 170          return "f32to16";
 171
 172       if (devinfo->gen > 7 && op == BRW_OPCODE_F16TO32)
 173          return "f16to32";
 174
 175       assert(brw_opcode_desc(devinfo, op)->name);
 176       return brw_opcode_desc(devinfo, op)->name;
 177    case FS_OPCODE_FB_WRITE:
 178       return "fb_write";
 179    case FS_OPCODE_FB_WRITE_LOGICAL:
 180       return "fb_write_logical";
 181    case FS_OPCODE_REP_FB_WRITE:
 182       return "rep_fb_write";
 183    case FS_OPCODE_FB_READ:
 184       return "fb_read";
 185    case FS_OPCODE_FB_READ_LOGICAL:
 186       return "fb_read_logical";
 187
 188    case SHADER_OPCODE_RCP:
 189       return "rcp";
 190    case SHADER_OPCODE_RSQ:
 191       return "rsq";
 192    case SHADER_OPCODE_SQRT:
 193       return "sqrt";
 194    case SHADER_OPCODE_EXP2:
 195       return "exp2";
 196    case SHADER_OPCODE_LOG2:
 197       return "log2";
 198    case SHADER_OPCODE_POW:
 199       return "pow";
 200    case SHADER_OPCODE_INT_QUOTIENT:
 201       return "int_quot";
 202    case SHADER_OPCODE_INT_REMAINDER:
 203       return "int_rem";
 204    case SHADER_OPCODE_SIN:
 205       return "sin";
 206    case SHADER_OPCODE_COS:
 207       return "cos";
 208
 209    case SHADER_OPCODE_TEX:
 210       return "tex";
 211    case SHADER_OPCODE_TEX_LOGICAL:
 212       return "tex_logical";
 213    case SHADER_OPCODE_TXD:
 214       return "txd";
 215    case SHADER_OPCODE_TXD_LOGICAL:
 216       return "txd_logical";
 217    case SHADER_OPCODE_TXF:
 218       return "txf";
 219    case SHADER_OPCODE_TXF_LOGICAL:
 220       return "txf_logical";
 221    case SHADER_OPCODE_TXF_LZ:
 222       return "txf_lz";
 223    case SHADER_OPCODE_TXL:
 224       return "txl";
 225    case SHADER_OPCODE_TXL_LOGICAL:
 226       return "txl_logical";
 227    case SHADER_OPCODE_TXL_LZ:
 228       return "txl_lz";
 229    case SHADER_OPCODE_TXS:
 230       return "txs";
 231    case SHADER_OPCODE_TXS_LOGICAL:
 232       return "txs_logical";
 233    case FS_OPCODE_TXB:
 234       return "txb";
 235    case FS_OPCODE_TXB_LOGICAL:
 236       return "txb_logical";
 237    case SHADER_OPCODE_TXF_CMS:
 238       return "txf_cms";
 239    case SHADER_OPCODE_TXF_CMS_LOGICAL:
 240       return "txf_cms_logical";
 241    case SHADER_OPCODE_TXF_CMS_W:
 242       return "txf_cms_w";
 243    case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
 244       return "txf_cms_w_logical";
 245    case SHADER_OPCODE_TXF_UMS:
 246       return "txf_ums";
 247    case SHADER_OPCODE_TXF_UMS_LOGICAL:
 248       return "txf_ums_logical";
 249    case SHADER_OPCODE_TXF_MCS:
 250       return "txf_mcs";
 251    case SHADER_OPCODE_TXF_MCS_LOGICAL:
 252       return "txf_mcs_logical";
 253    case SHADER_OPCODE_LOD:
 254       return "lod";
 255    case SHADER_OPCODE_LOD_LOGICAL:
 256       return "lod_logical";
 257    case SHADER_OPCODE_TG4:
 258       return "tg4";
 259    case SHADER_OPCODE_TG4_LOGICAL:
 260       return "tg4_logical";
 261    case SHADER_OPCODE_TG4_OFFSET:
 262       return "tg4_offset";
 263    case SHADER_OPCODE_TG4_OFFSET_LOGICAL:
 264       return "tg4_offset_logical";
 265    case SHADER_OPCODE_SAMPLEINFO:
 266       return "sampleinfo";
 267    case SHADER_OPCODE_SAMPLEINFO_LOGICAL:
 268       return "sampleinfo_logical";
 269
 270    case SHADER_OPCODE_SHADER_TIME_ADD:
 271       return "shader_time_add";
 272
 273    case SHADER_OPCODE_UNTYPED_ATOMIC:
 274       return "untyped_atomic";
 275    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
 276       return "untyped_atomic_logical";
 277    case SHADER_OPCODE_UNTYPED_SURFACE_READ:
 278       return "untyped_surface_read";
 279    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
 280       return "untyped_surface_read_logical";
 281    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
 282       return "untyped_surface_write";
 283    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
 284       return "untyped_surface_write_logical";
 285    case SHADER_OPCODE_TYPED_ATOMIC:
 286       return "typed_atomic";
 287    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
 288       return "typed_atomic_logical";
 289    case SHADER_OPCODE_TYPED_SURFACE_READ:
 290       return "typed_surface_read";
 291    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
 292       return "typed_surface_read_logical";
 293    case SHADER_OPCODE_TYPED_SURFACE_WRITE:
 294       return "typed_surface_write";
 295    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
 296       return "typed_surface_write_logical";
 297    case SHADER_OPCODE_MEMORY_FENCE:
 298       return "memory_fence";
 299
 300    case SHADER_OPCODE_BYTE_SCATTERED_READ:
 301       return "byte_scattered_read";
 302    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
 303       return "byte_scattered_read_logical";
 304    case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
 305       return "byte_scattered_write";
 306    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
 307       return "byte_scattered_write_logical";
 308
 309    case SHADER_OPCODE_LOAD_PAYLOAD:
 310       return "load_payload";
 311    case FS_OPCODE_PACK:
 312       return "pack";
 313
 314    case SHADER_OPCODE_GEN4_SCRATCH_READ:
 315       return "gen4_scratch_read";
 316    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
 317       return "gen4_scratch_write";
 318    case SHADER_OPCODE_GEN7_SCRATCH_READ:
 319       return "gen7_scratch_read";
 320    case SHADER_OPCODE_URB_WRITE_SIMD8:
 321       return "gen8_urb_write_simd8";
 322    case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
 323       return "gen8_urb_write_simd8_per_slot";
 324    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
 325       return "gen8_urb_write_simd8_masked";
 326    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
 327       return "gen8_urb_write_simd8_masked_per_slot";
 328    case SHADER_OPCODE_URB_READ_SIMD8:
 329       return "urb_read_simd8";
 330    case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
 331       return "urb_read_simd8_per_slot";
 332
 333    case SHADER_OPCODE_FIND_LIVE_CHANNEL:
 334       return "find_live_channel";
 335    case SHADER_OPCODE_BROADCAST:
 336       return "broadcast";
 337    case SHADER_OPCODE_SHUFFLE:
 338       return "shuffle";
 339    case SHADER_OPCODE_SEL_EXEC:
 340       return "sel_exec";
 341    case SHADER_OPCODE_QUAD_SWIZZLE:
 342       return "quad_swizzle";
 343    case SHADER_OPCODE_CLUSTER_BROADCAST:
 344       return "cluster_broadcast";
 345
 346    case SHADER_OPCODE_GET_BUFFER_SIZE:
 347       return "get_buffer_size";
 348
 349    case VEC4_OPCODE_MOV_BYTES:
 350       return "mov_bytes";
 351    case VEC4_OPCODE_PACK_BYTES:
 352       return "pack_bytes";
 353    case VEC4_OPCODE_UNPACK_UNIFORM:
 354       return "unpack_uniform";
 355    case VEC4_OPCODE_DOUBLE_TO_F32:
 356       return "double_to_f32";
 357    case VEC4_OPCODE_DOUBLE_TO_D32:
 358       return "double_to_d32";
 359    case VEC4_OPCODE_DOUBLE_TO_U32:
 360       return "double_to_u32";
 361    case VEC4_OPCODE_TO_DOUBLE:
 362       return "single_to_double";
 363    case VEC4_OPCODE_PICK_LOW_32BIT:
 364       return "pick_low_32bit";
 365    case VEC4_OPCODE_PICK_HIGH_32BIT:
 366       return "pick_high_32bit";
 367    case VEC4_OPCODE_SET_LOW_32BIT:
 368       return "set_low_32bit";
 369    case VEC4_OPCODE_SET_HIGH_32BIT:
 370       return "set_high_32bit";
 371
 372    case FS_OPCODE_DDX_COARSE:
 373       return "ddx_coarse";
 374    case FS_OPCODE_DDX_FINE:
 375       return "ddx_fine";
 376    case FS_OPCODE_DDY_COARSE:
 377       return "ddy_coarse";
 378    case FS_OPCODE_DDY_FINE:
 379       return "ddy_fine";
 380
 381    case FS_OPCODE_CINTERP:
 382       return "cinterp";
 383    case FS_OPCODE_LINTERP:
 384       return "linterp";
 385
 386    case FS_OPCODE_PIXEL_X:
 387       return "pixel_x";
 388    case FS_OPCODE_PIXEL_Y:
 389       return "pixel_y";
 390
 391    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
 392       return "uniform_pull_const";
 393    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
 394       return "uniform_pull_const_gen7";
 395    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4:
 396       return "varying_pull_const_gen4";
 397    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
 398       return "varying_pull_const_gen7";
 399    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
 400       return "varying_pull_const_logical";
 401
 402    case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
 403       return "mov_dispatch_to_flags";
 404    case FS_OPCODE_DISCARD_JUMP:
 405       return "discard_jump";
 406
 407    case FS_OPCODE_SET_SAMPLE_ID:
 408       return "set_sample_id";
 409
 410    case FS_OPCODE_PACK_HALF_2x16_SPLIT:
 411       return "pack_half_2x16_split";
 412    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
 413       return "unpack_half_2x16_split_x";
 414    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
 415       return "unpack_half_2x16_split_y";
 416
 417    case FS_OPCODE_PLACEHOLDER_HALT:
 418       return "placeholder_halt";
 419
 420    case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
 421       return "interp_sample";
 422    case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
 423       return "interp_shared_offset";
 424    case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
 425       return "interp_per_slot_offset";
 426
 427    case VS_OPCODE_URB_WRITE:
 428       return "vs_urb_write";
 429    case VS_OPCODE_PULL_CONSTANT_LOAD:
 430       return "pull_constant_load";
 431    case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
 432       return "pull_constant_load_gen7";
 433
 434    case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9:
 435       return "set_simd4x2_header_gen9";
 436
 437    case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
 438       return "unpack_flags_simd4x2";
 439
 440    case GS_OPCODE_URB_WRITE:
 441       return "gs_urb_write";
 442    case GS_OPCODE_URB_WRITE_ALLOCATE:
 443       return "gs_urb_write_allocate";
 444    case GS_OPCODE_THREAD_END:
 445       return "gs_thread_end";
 446    case GS_OPCODE_SET_WRITE_OFFSET:
 447       return "set_write_offset";
 448    case GS_OPCODE_SET_VERTEX_COUNT:
 449       return "set_vertex_count";
 450    case GS_OPCODE_SET_DWORD_2:
 451       return "set_dword_2";
 452    case GS_OPCODE_PREPARE_CHANNEL_MASKS:
 453       return "prepare_channel_masks";
 454    case GS_OPCODE_SET_CHANNEL_MASKS:
 455       return "set_channel_masks";
 456    case GS_OPCODE_GET_INSTANCE_ID:
 457       return "get_instance_id";
 458    case GS_OPCODE_FF_SYNC:
 459       return "ff_sync";
 460    case GS_OPCODE_SET_PRIMITIVE_ID:
 461       return "set_primitive_id";
 462    case GS_OPCODE_SVB_WRITE:
 463       return "gs_svb_write";
 464    case GS_OPCODE_SVB_SET_DST_INDEX:
 465       return "gs_svb_set_dst_index";
 466    case GS_OPCODE_FF_SYNC_SET_PRIMITIVES:
 467       return "gs_ff_sync_set_primitives";
 468    case CS_OPCODE_CS_TERMINATE:
 469       return "cs_terminate";
 470    case SHADER_OPCODE_BARRIER:
 471       return "barrier";
 472    case SHADER_OPCODE_MULH:
 473       return "mulh";
 474    case SHADER_OPCODE_MOV_INDIRECT:
 475       return "mov_indirect";
 476
 477    case VEC4_OPCODE_URB_READ:
 478       return "urb_read";
 479    case TCS_OPCODE_GET_INSTANCE_ID:
 480       return "tcs_get_instance_id";
 481    case TCS_OPCODE_URB_WRITE:
 482       return "tcs_urb_write";
 483    case TCS_OPCODE_SET_INPUT_URB_OFFSETS:
 484       return "tcs_set_input_urb_offsets";
 485    case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS:
 486       return "tcs_set_output_urb_offsets";
 487    case TCS_OPCODE_GET_PRIMITIVE_ID:
 488       return "tcs_get_primitive_id";
 489    case TCS_OPCODE_CREATE_BARRIER_HEADER:
 490       return "tcs_create_barrier_header";
 491    case TCS_OPCODE_SRC0_010_IS_ZERO:
 492       return "tcs_src0<0,1,0>_is_zero";
 493    case TCS_OPCODE_RELEASE_INPUT:
 494       return "tcs_release_input";
 495    case TCS_OPCODE_THREAD_END:
 496       return "tcs_thread_end";
 497    case TES_OPCODE_CREATE_INPUT_READ_HEADER:
 498       return "tes_create_input_read_header";
 499    case TES_OPCODE_ADD_INDIRECT_URB_OFFSET:
 500       return "tes_add_indirect_urb_offset";
 501    case TES_OPCODE_GET_PRIMITIVE_ID:
 502       return "tes_get_primitive_id";
 503
 504    case SHADER_OPCODE_RND_MODE:
 505       return "rnd_mode";
 506    }
 507
 508    unreachable("not reached");
 509 }
 510
 511 bool
 512 brw_saturate_immediate(enum brw_reg_type type, struct brw_reg *reg)
 513 {
 514    union {
 515       unsigned ud;
 516       int d;
 517       float f;
 518       double df;
 519    } imm, sat_imm = { 0 };
 520
 521    const unsigned size = type_sz(type);
 522
 523    /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
 524     * irrelevant, so just check the size of the type and copy from/to an
 525     * appropriately sized field.
 526     */
 527    if (size < 8)
 528       imm.ud = reg->ud;
 529    else
 530       imm.df = reg->df;
 531
 532    switch (type) {
 533    case BRW_REGISTER_TYPE_UD:
 534    case BRW_REGISTER_TYPE_D:
 535    case BRW_REGISTER_TYPE_UW:
 536    case BRW_REGISTER_TYPE_W:
 537    case BRW_REGISTER_TYPE_UQ:
 538    case BRW_REGISTER_TYPE_Q:
 539       /* Nothing to do. */
 540       return false;
 541    case BRW_REGISTER_TYPE_F:
 542       sat_imm.f = CLAMP(imm.f, 0.0f, 1.0f);
 543       break;
 544    case BRW_REGISTER_TYPE_DF:
 545       sat_imm.df = CLAMP(imm.df, 0.0, 1.0);
 546       break;
 547    case BRW_REGISTER_TYPE_UB:
 548    case BRW_REGISTER_TYPE_B:
 549       unreachable("no UB/B immediates");
 550    case BRW_REGISTER_TYPE_V:
 551    case BRW_REGISTER_TYPE_UV:
 552    case BRW_REGISTER_TYPE_VF:
 553       unreachable("unimplemented: saturate vector immediate");
 554    case BRW_REGISTER_TYPE_HF:
 555       unreachable("unimplemented: saturate HF immediate");
 556    case BRW_REGISTER_TYPE_NF:
 557       unreachable("no NF immediates");
 558    }
 559
 560    if (size < 8) {
 561       if (imm.ud != sat_imm.ud) {
 562          reg->ud = sat_imm.ud;
 563          return true;
 564       }
 565    } else {
 566       if (imm.df != sat_imm.df) {
 567          reg->df = sat_imm.df;
 568          return true;
 569       }
 570    }
 571    return false;
 572 }
 573
 574 bool
 575 brw_negate_immediate(enum brw_reg_type type, struct brw_reg *reg)
 576 {
 577    switch (type) {
 578    case BRW_REGISTER_TYPE_D:
 579    case BRW_REGISTER_TYPE_UD:
 580       reg->d = -reg->d;
 581       return true;
 582    case BRW_REGISTER_TYPE_W:
 583    case BRW_REGISTER_TYPE_UW:
 584       reg->d = -(int16_t)reg->ud;
 585       return true;
 586    case BRW_REGISTER_TYPE_F:
 587       reg->f = -reg->f;
 588       return true;
 589    case BRW_REGISTER_TYPE_VF:
 590       reg->ud ^= 0x80808080;
 591       return true;
 592    case BRW_REGISTER_TYPE_DF:
 593       reg->df = -reg->df;
 594       return true;
 595    case BRW_REGISTER_TYPE_UQ:
 596    case BRW_REGISTER_TYPE_Q:
 597       reg->d64 = -reg->d64;
 598       return true;
 599    case BRW_REGISTER_TYPE_UB:
 600    case BRW_REGISTER_TYPE_B:
 601       unreachable("no UB/B immediates");
 602    case BRW_REGISTER_TYPE_UV:
 603    case BRW_REGISTER_TYPE_V:
 604       assert(!"unimplemented: negate UV/V immediate");
 605    case BRW_REGISTER_TYPE_HF:
 606       assert(!"unimplemented: negate HF immediate");
 607    case BRW_REGISTER_TYPE_NF:
 608       unreachable("no NF immediates");
 609    }
 610
 611    return false;
 612 }
 613
 614 bool
 615 brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg)
 616 {
 617    switch (type) {
 618    case BRW_REGISTER_TYPE_D:
 619       reg->d = abs(reg->d);
 620       return true;
 621    case BRW_REGISTER_TYPE_W:
 622       reg->d = abs((int16_t)reg->ud);
 623       return true;
 624    case BRW_REGISTER_TYPE_F:
 625       reg->f = fabsf(reg->f);
 626       return true;
 627    case BRW_REGISTER_TYPE_DF:
 628       reg->df = fabs(reg->df);
 629       return true;
 630    case BRW_REGISTER_TYPE_VF:
 631       reg->ud &= ~0x80808080;
 632       return true;
 633    case BRW_REGISTER_TYPE_Q:
 634       reg->d64 = imaxabs(reg->d64);
 635       return true;
 636    case BRW_REGISTER_TYPE_UB:
 637    case BRW_REGISTER_TYPE_B:
 638       unreachable("no UB/B immediates");
 639    case BRW_REGISTER_TYPE_UQ:
 640    case BRW_REGISTER_TYPE_UD:
 641    case BRW_REGISTER_TYPE_UW:
 642    case BRW_REGISTER_TYPE_UV:
 643       /* Presumably the absolute value modifier on an unsigned source is a
 644        * nop, but it would be nice to confirm.
 645        */
 646       assert(!"unimplemented: abs unsigned immediate");
 647    case BRW_REGISTER_TYPE_V:
 648       assert(!"unimplemented: abs V immediate");
 649    case BRW_REGISTER_TYPE_HF:
 650       assert(!"unimplemented: abs HF immediate");
 651    case BRW_REGISTER_TYPE_NF:
 652       unreachable("no NF immediates");
 653    }
 654
 655    return false;
 656 }
 657
 658 backend_shader::backend_shader(const struct brw_compiler *compiler,
 659                                void *log_data,
 660                                void *mem_ctx,
 661                                const nir_shader *shader,
 662                                struct brw_stage_prog_data *stage_prog_data)
 663    : compiler(compiler),
 664      log_data(log_data),
 665      devinfo(compiler->devinfo),
 666      nir(shader),
 667      stage_prog_data(stage_prog_data),
 668      mem_ctx(mem_ctx),
 669      cfg(NULL),
 670      stage(shader->info.stage)
 671 {
 672    debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
 673    stage_name = _mesa_shader_stage_to_string(stage);
 674    stage_abbrev = _mesa_shader_stage_to_abbrev(stage);
 675 }
 676
 677 backend_shader::~backend_shader()
 678 {
 679 }
 680
 681 bool
 682 backend_reg::equals(const backend_reg &r) const
 683 {
 684    return brw_regs_equal(this, &r) && offset == r.offset;
 685 }
 686
 687 bool
 688 backend_reg::negative_equals(const backend_reg &r) const
 689 {
 690    return brw_regs_negative_equal(this, &r) && offset == r.offset;
 691 }
 692
 693 bool
 694 backend_reg::is_zero() const
 695 {
 696    if (file != IMM)
 697       return false;
 698
 699    switch (type) {
 700    case BRW_REGISTER_TYPE_F:
 701       return f == 0;
 702    case BRW_REGISTER_TYPE_DF:
 703       return df == 0;
 704    case BRW_REGISTER_TYPE_D:
 705    case BRW_REGISTER_TYPE_UD:
 706       return d == 0;
 707    case BRW_REGISTER_TYPE_UQ:
 708    case BRW_REGISTER_TYPE_Q:
 709       return u64 == 0;
 710    default:
 711       return false;
 712    }
 713 }
 714
 715 bool
 716 backend_reg::is_one() const
 717 {
 718    if (file != IMM)
 719       return false;
 720
 721    switch (type) {
 722    case BRW_REGISTER_TYPE_F:
 723       return f == 1.0f;
 724    case BRW_REGISTER_TYPE_DF:
 725       return df == 1.0;
 726    case BRW_REGISTER_TYPE_D:
 727    case BRW_REGISTER_TYPE_UD:
 728       return d == 1;
 729    case BRW_REGISTER_TYPE_UQ:
 730    case BRW_REGISTER_TYPE_Q:
 731       return u64 == 1;
 732    default:
 733       return false;
 734    }
 735 }
 736
 737 bool
 738 backend_reg::is_negative_one() const
 739 {
 740    if (file != IMM)
 741       return false;
 742
 743    switch (type) {
 744    case BRW_REGISTER_TYPE_F:
 745       return f == -1.0;
 746    case BRW_REGISTER_TYPE_DF:
 747       return df == -1.0;
 748    case BRW_REGISTER_TYPE_D:
 749       return d == -1;
 750    case BRW_REGISTER_TYPE_Q:
 751       return d64 == -1;
 752    default:
 753       return false;
 754    }
 755 }
 756
 757 bool
 758 backend_reg::is_null() const
 759 {
 760    return file == ARF && nr == BRW_ARF_NULL;
 761 }
 762
 763
 764 bool
 765 backend_reg::is_accumulator() const
 766 {
 767    return file == ARF && nr == BRW_ARF_ACCUMULATOR;
 768 }
 769
 770 bool
 771 backend_instruction::is_commutative() const
 772 {
 773    switch (opcode) {
 774    case BRW_OPCODE_AND:
 775    case BRW_OPCODE_OR:
 776    case BRW_OPCODE_XOR:
 777    case BRW_OPCODE_ADD:
 778    case BRW_OPCODE_MUL:
 779    case SHADER_OPCODE_MULH:
 780       return true;
 781    case BRW_OPCODE_SEL:
 782       /* MIN and MAX are commutative. */
 783       if (conditional_mod == BRW_CONDITIONAL_GE ||
 784           conditional_mod == BRW_CONDITIONAL_L) {
 785          return true;
 786       }
 787       /* fallthrough */
 788    default:
 789       return false;
 790    }
 791 }
 792
 793 bool
 794 backend_instruction::is_3src(const struct gen_device_info *devinfo) const
 795 {
 796    return ::is_3src(devinfo, opcode);
 797 }
 798
 799 bool
 800 backend_instruction::is_tex() const
 801 {
 802    return (opcode == SHADER_OPCODE_TEX ||
 803            opcode == FS_OPCODE_TXB ||
 804            opcode == SHADER_OPCODE_TXD ||
 805            opcode == SHADER_OPCODE_TXF ||
 806            opcode == SHADER_OPCODE_TXF_LZ ||
 807            opcode == SHADER_OPCODE_TXF_CMS ||
 808            opcode == SHADER_OPCODE_TXF_CMS_W ||
 809            opcode == SHADER_OPCODE_TXF_UMS ||
 810            opcode == SHADER_OPCODE_TXF_MCS ||
 811            opcode == SHADER_OPCODE_TXL ||
 812            opcode == SHADER_OPCODE_TXL_LZ ||
 813            opcode == SHADER_OPCODE_TXS ||
 814            opcode == SHADER_OPCODE_LOD ||
 815            opcode == SHADER_OPCODE_TG4 ||
 816            opcode == SHADER_OPCODE_TG4_OFFSET ||
 817            opcode == SHADER_OPCODE_SAMPLEINFO);
 818 }
 819
 820 bool
 821 backend_instruction::is_math() const
 822 {
 823    return (opcode == SHADER_OPCODE_RCP ||
 824            opcode == SHADER_OPCODE_RSQ ||
 825            opcode == SHADER_OPCODE_SQRT ||
 826            opcode == SHADER_OPCODE_EXP2 ||
 827            opcode == SHADER_OPCODE_LOG2 ||
 828            opcode == SHADER_OPCODE_SIN ||
 829            opcode == SHADER_OPCODE_COS ||
 830            opcode == SHADER_OPCODE_INT_QUOTIENT ||
 831            opcode == SHADER_OPCODE_INT_REMAINDER ||
 832            opcode == SHADER_OPCODE_POW);
 833 }
 834
 835 bool
 836 backend_instruction::is_control_flow() const
 837 {
 838    switch (opcode) {
 839    case BRW_OPCODE_DO:
 840    case BRW_OPCODE_WHILE:
 841    case BRW_OPCODE_IF:
 842    case BRW_OPCODE_ELSE:
 843    case BRW_OPCODE_ENDIF:
 844    case BRW_OPCODE_BREAK:
 845    case BRW_OPCODE_CONTINUE:
 846       return true;
 847    default:
 848       return false;
 849    }
 850 }
 851
 852 bool
 853 backend_instruction::can_do_source_mods() const
 854 {
 855    switch (opcode) {
 856    case BRW_OPCODE_ADDC:
 857    case BRW_OPCODE_BFE:
 858    case BRW_OPCODE_BFI1:
 859    case BRW_OPCODE_BFI2:
 860    case BRW_OPCODE_BFREV:
 861    case BRW_OPCODE_CBIT:
 862    case BRW_OPCODE_FBH:
 863    case BRW_OPCODE_FBL:
 864    case BRW_OPCODE_SUBB:
 865    case SHADER_OPCODE_BROADCAST:
 866    case SHADER_OPCODE_CLUSTER_BROADCAST:
 867    case SHADER_OPCODE_MOV_INDIRECT:
 868       return false;
 869    default:
 870       return true;
 871    }
 872 }
 873
 874 bool
 875 backend_instruction::can_do_saturate() const
 876 {
 877    switch (opcode) {
 878    case BRW_OPCODE_ADD:
 879    case BRW_OPCODE_ASR:
 880    case BRW_OPCODE_AVG:
 881    case BRW_OPCODE_DP2:
 882    case BRW_OPCODE_DP3:
 883    case BRW_OPCODE_DP4:
 884    case BRW_OPCODE_DPH:
 885    case BRW_OPCODE_F16TO32:
 886    case BRW_OPCODE_F32TO16:
 887    case BRW_OPCODE_LINE:
 888    case BRW_OPCODE_LRP:
 889    case BRW_OPCODE_MAC:
 890    case BRW_OPCODE_MAD:
 891    case BRW_OPCODE_MATH:
 892    case BRW_OPCODE_MOV:
 893    case BRW_OPCODE_MUL:
 894    case SHADER_OPCODE_MULH:
 895    case BRW_OPCODE_PLN:
 896    case BRW_OPCODE_RNDD:
 897    case BRW_OPCODE_RNDE:
 898    case BRW_OPCODE_RNDU:
 899    case BRW_OPCODE_RNDZ:
 900    case BRW_OPCODE_SEL:
 901    case BRW_OPCODE_SHL:
 902    case BRW_OPCODE_SHR:
 903    case FS_OPCODE_LINTERP:
 904    case SHADER_OPCODE_COS:
 905    case SHADER_OPCODE_EXP2:
 906    case SHADER_OPCODE_LOG2:
 907    case SHADER_OPCODE_POW:
 908    case SHADER_OPCODE_RCP:
 909    case SHADER_OPCODE_RSQ:
 910    case SHADER_OPCODE_SIN:
 911    case SHADER_OPCODE_SQRT:
 912       return true;
 913    default:
 914       return false;
 915    }
 916 }
 917
 918 bool
 919 backend_instruction::can_do_cmod() const
 920 {
 921    switch (opcode) {
 922    case BRW_OPCODE_ADD:
 923    case BRW_OPCODE_ADDC:
 924    case BRW_OPCODE_AND:
 925    case BRW_OPCODE_ASR:
 926    case BRW_OPCODE_AVG:
 927    case BRW_OPCODE_CMP:
 928    case BRW_OPCODE_CMPN:
 929    case BRW_OPCODE_DP2:
 930    case BRW_OPCODE_DP3:
 931    case BRW_OPCODE_DP4:
 932    case BRW_OPCODE_DPH:
 933    case BRW_OPCODE_F16TO32:
 934    case BRW_OPCODE_F32TO16:
 935    case BRW_OPCODE_FRC:
 936    case BRW_OPCODE_LINE:
 937    case BRW_OPCODE_LRP:
 938    case BRW_OPCODE_LZD:
 939    case BRW_OPCODE_MAC:
 940    case BRW_OPCODE_MACH:
 941    case BRW_OPCODE_MAD:
 942    case BRW_OPCODE_MOV:
 943    case BRW_OPCODE_MUL:
 944    case BRW_OPCODE_NOT:
 945    case BRW_OPCODE_OR:
 946    case BRW_OPCODE_PLN:
 947    case BRW_OPCODE_RNDD:
 948    case BRW_OPCODE_RNDE:
 949    case BRW_OPCODE_RNDU:
 950    case BRW_OPCODE_RNDZ:
 951    case BRW_OPCODE_SAD2:
 952    case BRW_OPCODE_SADA2:
 953    case BRW_OPCODE_SHL:
 954    case BRW_OPCODE_SHR:
 955    case BRW_OPCODE_SUBB:
 956    case BRW_OPCODE_XOR:
 957    case FS_OPCODE_CINTERP:
 958    case FS_OPCODE_LINTERP:
 959       return true;
 960    default:
 961       return false;
 962    }
 963 }
 964
 965 bool
 966 backend_instruction::reads_accumulator_implicitly() const
 967 {
 968    switch (opcode) {
 969    case BRW_OPCODE_MAC:
 970    case BRW_OPCODE_MACH:
 971    case BRW_OPCODE_SADA2:
 972       return true;
 973    default:
 974       return false;
 975    }
 976 }
 977
 978 bool
 979 backend_instruction::writes_accumulator_implicitly(const struct gen_device_info *devinfo) const
 980 {
 981    return writes_accumulator ||
 982           (devinfo->gen < 6 &&
 983            ((opcode >= BRW_OPCODE_ADD && opcode < BRW_OPCODE_NOP) ||
 984             (opcode >= FS_OPCODE_DDX_COARSE && opcode <= FS_OPCODE_LINTERP &&
 985              opcode != FS_OPCODE_CINTERP)));
 986 }
 987
 988 bool
 989 backend_instruction::has_side_effects() const
 990 {
 991    switch (opcode) {
 992    case SHADER_OPCODE_UNTYPED_ATOMIC:
 993    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
 994    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
 995    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
 996    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
 997    case SHADER_OPCODE_BYTE_SCATTERED_WRITE:
 998    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
 999    case SHADER_OPCODE_TYPED_ATOMIC:
1000    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
1001    case SHADER_OPCODE_TYPED_SURFACE_WRITE:
1002    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
1003    case SHADER_OPCODE_MEMORY_FENCE:
1004    case SHADER_OPCODE_URB_WRITE_SIMD8:
1005    case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
1006    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
1007    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
1008    case FS_OPCODE_FB_WRITE:
1009    case FS_OPCODE_FB_WRITE_LOGICAL:
1010    case SHADER_OPCODE_BARRIER:
1011    case TCS_OPCODE_URB_WRITE:
1012    case TCS_OPCODE_RELEASE_INPUT:
1013    case SHADER_OPCODE_RND_MODE:
1014       return true;
1015    default:
1016       return eot;
1017    }
1018 }
1019
1020 bool
1021 backend_instruction::is_volatile() const
1022 {
1023    switch (opcode) {
1024    case SHADER_OPCODE_UNTYPED_SURFACE_READ:
1025    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
1026    case SHADER_OPCODE_TYPED_SURFACE_READ:
1027    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
1028    case SHADER_OPCODE_BYTE_SCATTERED_READ:
1029    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
1030    case SHADER_OPCODE_URB_READ_SIMD8:
1031    case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
1032    case VEC4_OPCODE_URB_READ:
1033       return true;
1034    default:
1035       return false;
1036    }
1037 }
1038
1039 #ifndef NDEBUG
1040 static bool
1041 inst_is_in_block(const bblock_t *block, const backend_instruction *inst)
1042 {
1043    bool found = false;
1044    foreach_inst_in_block (backend_instruction, i, block) {
1045       if (inst == i) {
1046          found = true;
1047       }
1048    }
1049    return found;
1050 }
1051 #endif
1052
1053 static void
1054 adjust_later_block_ips(bblock_t *start_block, int ip_adjustment)
1055 {
1056    for (bblock_t *block_iter = start_block->next();
1057         block_iter;
1058         block_iter = block_iter->next()) {
1059       block_iter->start_ip += ip_adjustment;
1060       block_iter->end_ip += ip_adjustment;
1061    }
1062 }
1063
1064 void
1065 backend_instruction::insert_after(bblock_t *block, backend_instruction *inst)
1066 {
1067    assert(this != inst);
1068
1069    if (!this->is_head_sentinel())
1070       assert(inst_is_in_block(block, this) || !"Instruction not in block");
1071
1072    block->end_ip++;
1073
1074    adjust_later_block_ips(block, 1);
1075
1076    exec_node::insert_after(inst);
1077 }
1078
1079 void
1080 backend_instruction::insert_before(bblock_t *block, backend_instruction *inst)
1081 {
1082    assert(this != inst);
1083
1084    if (!this->is_tail_sentinel())
1085       assert(inst_is_in_block(block, this) || !"Instruction not in block");
1086
1087    block->end_ip++;
1088
1089    adjust_later_block_ips(block, 1);
1090
1091    exec_node::insert_before(inst);
1092 }
1093
1094 void
1095 backend_instruction::insert_before(bblock_t *block, exec_list *list)
1096 {
1097    assert(inst_is_in_block(block, this) || !"Instruction not in block");
1098
1099    unsigned num_inst = list->length();
1100
1101    block->end_ip += num_inst;
1102
1103    adjust_later_block_ips(block, num_inst);
1104
1105    exec_node::insert_before(list);
1106 }
1107
1108 void
1109 backend_instruction::remove(bblock_t *block)
1110 {
1111    assert(inst_is_in_block(block, this) || !"Instruction not in block");
1112
1113    adjust_later_block_ips(block, -1);
1114
1115    if (block->start_ip == block->end_ip) {
1116       block->cfg->remove_block(block);
1117    } else {
1118       block->end_ip--;
1119    }
1120
1121    exec_node::remove();
1122 }
1123
1124 void
1125 backend_shader::dump_instructions()
1126 {
1127    dump_instructions(NULL);
1128 }
1129
1130 void
1131 backend_shader::dump_instructions(const char *name)
1132 {
1133    FILE *file = stderr;
1134    if (name && geteuid() != 0) {
1135       file = fopen(name, "w");
1136       if (!file)
1137          file = stderr;
1138    }
1139
1140    if (cfg) {
1141       int ip = 0;
1142       foreach_block_and_inst(block, backend_instruction, inst, cfg) {
1143          if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1144             fprintf(file, "%4d: ", ip++);
1145          dump_instruction(inst, file);
1146       }
1147    } else {
1148       int ip = 0;
1149       foreach_in_list(backend_instruction, inst, &instructions) {
1150          if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1151             fprintf(file, "%4d: ", ip++);
1152          dump_instruction(inst, file);
1153       }
1154    }
1155
1156    if (file != stderr) {
1157       fclose(file);
1158    }
1159 }
1160
1161 void
1162 backend_shader::calculate_cfg()
1163 {
1164    if (this->cfg)
1165       return;
1166    cfg = new(mem_ctx) cfg_t(&this->instructions);
1167 }
1168
1169 extern "C" const unsigned *
1170 brw_compile_tes(const struct brw_compiler *compiler,
1171                 void *log_data,
1172                 void *mem_ctx,
1173                 const struct brw_tes_prog_key *key,
1174                 const struct brw_vue_map *input_vue_map,
1175                 struct brw_tes_prog_data *prog_data,
1176                 const nir_shader *src_shader,
1177                 struct gl_program *prog,
1178                 int shader_time_index,
1179                 char **error_str)
1180 {
1181    const struct gen_device_info *devinfo = compiler->devinfo;
1182    const bool is_scalar = compiler->scalar_stage[MESA_SHADER_TESS_EVAL];
1183    const unsigned *assembly;
1184
1185    nir_shader *nir = nir_shader_clone(mem_ctx, src_shader);
1186    nir->info.inputs_read = key->inputs_read;
1187    nir->info.patch_inputs_read = key->patch_inputs_read;
1188
1189    nir = brw_nir_apply_sampler_key(nir, compiler, &key->tex, is_scalar);
1190    brw_nir_lower_tes_inputs(nir, input_vue_map);
1191    brw_nir_lower_vue_outputs(nir, is_scalar);
1192    nir = brw_postprocess_nir(nir, compiler, is_scalar);
1193
1194    brw_compute_vue_map(devinfo, &prog_data->base.vue_map,
1195                        nir->info.outputs_written,
1196                        nir->info.separate_shader);
1197
1198    unsigned output_size_bytes = prog_data->base.vue_map.num_slots * 4 * 4;
1199
1200    assert(output_size_bytes >= 1);
1201    if (output_size_bytes > GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES) {
1202       if (error_str)
1203          *error_str = ralloc_strdup(mem_ctx, "DS outputs exceed maximum size");
1204       return NULL;
1205    }
1206
1207    prog_data->base.clip_distance_mask =
1208       ((1 << nir->info.clip_distance_array_size) - 1);
1209    prog_data->base.cull_distance_mask =
1210       ((1 << nir->info.cull_distance_array_size) - 1) <<
1211       nir->info.clip_distance_array_size;
1212
1213    /* URB entry sizes are stored as a multiple of 64 bytes. */
1214    prog_data->base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64;
1215
1216    /* On Cannonlake software shall not program an allocation size that
1217     * specifies a size that is a multiple of 3 64B (512-bit) cachelines.
1218     */
1219    if (devinfo->gen == 10 &&
1220        prog_data->base.urb_entry_size % 3 == 0)
1221       prog_data->base.urb_entry_size++;
1222
1223    prog_data->base.urb_read_length = 0;
1224
1225    STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER == TESS_SPACING_EQUAL - 1);
1226    STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL ==
1227                  TESS_SPACING_FRACTIONAL_ODD - 1);
1228    STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL ==
1229                  TESS_SPACING_FRACTIONAL_EVEN - 1);
1230
1231    prog_data->partitioning =
1232       (enum brw_tess_partitioning) (nir->info.tess.spacing - 1);
1233
1234    switch (nir->info.tess.primitive_mode) {
1235    case GL_QUADS:
1236       prog_data->domain = BRW_TESS_DOMAIN_QUAD;
1237       break;
1238    case GL_TRIANGLES:
1239       prog_data->domain = BRW_TESS_DOMAIN_TRI;
1240       break;
1241    case GL_ISOLINES:
1242       prog_data->domain = BRW_TESS_DOMAIN_ISOLINE;
1243       break;
1244    default:
1245       unreachable("invalid domain shader primitive mode");
1246    }
1247
1248    if (nir->info.tess.point_mode) {
1249       prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT;
1250    } else if (nir->info.tess.primitive_mode == GL_ISOLINES) {
1251       prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE;
1252    } else {
1253       /* Hardware winding order is backwards from OpenGL */
1254       prog_data->output_topology =
1255          nir->info.tess.ccw ? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
1256                              : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW;
1257    }
1258
1259    if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1260       fprintf(stderr, "TES Input ");
1261       brw_print_vue_map(stderr, input_vue_map);
1262       fprintf(stderr, "TES Output ");
1263       brw_print_vue_map(stderr, &prog_data->base.vue_map);
1264    }
1265
1266    if (is_scalar) {
1267       fs_visitor v(compiler, log_data, mem_ctx, (void *) key,
1268                    &prog_data->base.base, NULL, nir, 8,
1269                    shader_time_index, input_vue_map);
1270       if (!v.run_tes()) {
1271          if (error_str)
1272             *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1273          return NULL;
1274       }
1275
1276       prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
1277       prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8;
1278
1279       fs_generator g(compiler, log_data, mem_ctx, (void *) key,
1280                      &prog_data->base.base, v.promoted_constants, false,
1281                      MESA_SHADER_TESS_EVAL);
1282       if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1283          g.enable_debug(ralloc_asprintf(mem_ctx,
1284                                         "%s tessellation evaluation shader %s",
1285                                         nir->info.label ? nir->info.label
1286                                                         : "unnamed",
1287                                         nir->info.name));
1288       }
1289
1290       g.generate_code(v.cfg, 8);
1291
1292       assembly = g.get_assembly();
1293    } else {
1294       brw::vec4_tes_visitor v(compiler, log_data, key, prog_data,
1295                               nir, mem_ctx, shader_time_index);
1296       if (!v.run()) {
1297          if (error_str)
1298             *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1299          return NULL;
1300       }
1301
1302       if (unlikely(INTEL_DEBUG & DEBUG_TES))
1303          v.dump_instructions();
1304
1305       assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir,
1306                                             &prog_data->base, v.cfg);
1307    }
1308
1309    return assembly;
1310 }