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 "dev/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_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
  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(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
 163 const char *
 164 brw_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_FLOAT_LOGICAL:
 312       return "a64_untyped_atomic_float_logical";
 313    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
 314       return "typed_atomic_logical";
 315    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
 316       return "typed_surface_read_logical";
 317    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
 318       return "typed_surface_write_logical";
 319    case SHADER_OPCODE_MEMORY_FENCE:
 320       return "memory_fence";
 321    case SHADER_OPCODE_INTERLOCK:
 322       /* For an interlock we actually issue a memory fence via sendc. */
 323       return "interlock";
 324
 325    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
 326       return "byte_scattered_read_logical";
 327    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
 328       return "byte_scattered_write_logical";
 329
 330    case SHADER_OPCODE_LOAD_PAYLOAD:
 331       return "load_payload";
 332    case FS_OPCODE_PACK:
 333       return "pack";
 334
 335    case SHADER_OPCODE_GEN4_SCRATCH_READ:
 336       return "gen4_scratch_read";
 337    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
 338       return "gen4_scratch_write";
 339    case SHADER_OPCODE_GEN7_SCRATCH_READ:
 340       return "gen7_scratch_read";
 341    case SHADER_OPCODE_URB_WRITE_SIMD8:
 342       return "gen8_urb_write_simd8";
 343    case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
 344       return "gen8_urb_write_simd8_per_slot";
 345    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
 346       return "gen8_urb_write_simd8_masked";
 347    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
 348       return "gen8_urb_write_simd8_masked_per_slot";
 349    case SHADER_OPCODE_URB_READ_SIMD8:
 350       return "urb_read_simd8";
 351    case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
 352       return "urb_read_simd8_per_slot";
 353
 354    case SHADER_OPCODE_FIND_LIVE_CHANNEL:
 355       return "find_live_channel";
 356    case SHADER_OPCODE_BROADCAST:
 357       return "broadcast";
 358    case SHADER_OPCODE_SHUFFLE:
 359       return "shuffle";
 360    case SHADER_OPCODE_SEL_EXEC:
 361       return "sel_exec";
 362    case SHADER_OPCODE_QUAD_SWIZZLE:
 363       return "quad_swizzle";
 364    case SHADER_OPCODE_CLUSTER_BROADCAST:
 365       return "cluster_broadcast";
 366
 367    case SHADER_OPCODE_GET_BUFFER_SIZE:
 368       return "get_buffer_size";
 369
 370    case VEC4_OPCODE_MOV_BYTES:
 371       return "mov_bytes";
 372    case VEC4_OPCODE_PACK_BYTES:
 373       return "pack_bytes";
 374    case VEC4_OPCODE_UNPACK_UNIFORM:
 375       return "unpack_uniform";
 376    case VEC4_OPCODE_DOUBLE_TO_F32:
 377       return "double_to_f32";
 378    case VEC4_OPCODE_DOUBLE_TO_D32:
 379       return "double_to_d32";
 380    case VEC4_OPCODE_DOUBLE_TO_U32:
 381       return "double_to_u32";
 382    case VEC4_OPCODE_TO_DOUBLE:
 383       return "single_to_double";
 384    case VEC4_OPCODE_PICK_LOW_32BIT:
 385       return "pick_low_32bit";
 386    case VEC4_OPCODE_PICK_HIGH_32BIT:
 387       return "pick_high_32bit";
 388    case VEC4_OPCODE_SET_LOW_32BIT:
 389       return "set_low_32bit";
 390    case VEC4_OPCODE_SET_HIGH_32BIT:
 391       return "set_high_32bit";
 392
 393    case FS_OPCODE_DDX_COARSE:
 394       return "ddx_coarse";
 395    case FS_OPCODE_DDX_FINE:
 396       return "ddx_fine";
 397    case FS_OPCODE_DDY_COARSE:
 398       return "ddy_coarse";
 399    case FS_OPCODE_DDY_FINE:
 400       return "ddy_fine";
 401
 402    case FS_OPCODE_LINTERP:
 403       return "linterp";
 404
 405    case FS_OPCODE_PIXEL_X:
 406       return "pixel_x";
 407    case FS_OPCODE_PIXEL_Y:
 408       return "pixel_y";
 409
 410    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
 411       return "uniform_pull_const";
 412    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
 413       return "uniform_pull_const_gen7";
 414    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4:
 415       return "varying_pull_const_gen4";
 416    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
 417       return "varying_pull_const_logical";
 418
 419    case FS_OPCODE_DISCARD_JUMP:
 420       return "discard_jump";
 421
 422    case FS_OPCODE_SET_SAMPLE_ID:
 423       return "set_sample_id";
 424
 425    case FS_OPCODE_PACK_HALF_2x16_SPLIT:
 426       return "pack_half_2x16_split";
 427
 428    case FS_OPCODE_PLACEHOLDER_HALT:
 429       return "placeholder_halt";
 430
 431    case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
 432       return "interp_sample";
 433    case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
 434       return "interp_shared_offset";
 435    case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
 436       return "interp_per_slot_offset";
 437
 438    case VS_OPCODE_URB_WRITE:
 439       return "vs_urb_write";
 440    case VS_OPCODE_PULL_CONSTANT_LOAD:
 441       return "pull_constant_load";
 442    case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
 443       return "pull_constant_load_gen7";
 444
 445    case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9:
 446       return "set_simd4x2_header_gen9";
 447
 448    case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
 449       return "unpack_flags_simd4x2";
 450
 451    case GS_OPCODE_URB_WRITE:
 452       return "gs_urb_write";
 453    case GS_OPCODE_URB_WRITE_ALLOCATE:
 454       return "gs_urb_write_allocate";
 455    case GS_OPCODE_THREAD_END:
 456       return "gs_thread_end";
 457    case GS_OPCODE_SET_WRITE_OFFSET:
 458       return "set_write_offset";
 459    case GS_OPCODE_SET_VERTEX_COUNT:
 460       return "set_vertex_count";
 461    case GS_OPCODE_SET_DWORD_2:
 462       return "set_dword_2";
 463    case GS_OPCODE_PREPARE_CHANNEL_MASKS:
 464       return "prepare_channel_masks";
 465    case GS_OPCODE_SET_CHANNEL_MASKS:
 466       return "set_channel_masks";
 467    case GS_OPCODE_GET_INSTANCE_ID:
 468       return "get_instance_id";
 469    case GS_OPCODE_FF_SYNC:
 470       return "ff_sync";
 471    case GS_OPCODE_SET_PRIMITIVE_ID:
 472       return "set_primitive_id";
 473    case GS_OPCODE_SVB_WRITE:
 474       return "gs_svb_write";
 475    case GS_OPCODE_SVB_SET_DST_INDEX:
 476       return "gs_svb_set_dst_index";
 477    case GS_OPCODE_FF_SYNC_SET_PRIMITIVES:
 478       return "gs_ff_sync_set_primitives";
 479    case CS_OPCODE_CS_TERMINATE:
 480       return "cs_terminate";
 481    case SHADER_OPCODE_BARRIER:
 482       return "barrier";
 483    case SHADER_OPCODE_MULH:
 484       return "mulh";
 485    case SHADER_OPCODE_MOV_INDIRECT:
 486       return "mov_indirect";
 487
 488    case VEC4_OPCODE_URB_READ:
 489       return "urb_read";
 490    case TCS_OPCODE_GET_INSTANCE_ID:
 491       return "tcs_get_instance_id";
 492    case TCS_OPCODE_URB_WRITE:
 493       return "tcs_urb_write";
 494    case TCS_OPCODE_SET_INPUT_URB_OFFSETS:
 495       return "tcs_set_input_urb_offsets";
 496    case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS:
 497       return "tcs_set_output_urb_offsets";
 498    case TCS_OPCODE_GET_PRIMITIVE_ID:
 499       return "tcs_get_primitive_id";
 500    case TCS_OPCODE_CREATE_BARRIER_HEADER:
 501       return "tcs_create_barrier_header";
 502    case TCS_OPCODE_SRC0_010_IS_ZERO:
 503       return "tcs_src0<0,1,0>_is_zero";
 504    case TCS_OPCODE_RELEASE_INPUT:
 505       return "tcs_release_input";
 506    case TCS_OPCODE_THREAD_END:
 507       return "tcs_thread_end";
 508    case TES_OPCODE_CREATE_INPUT_READ_HEADER:
 509       return "tes_create_input_read_header";
 510    case TES_OPCODE_ADD_INDIRECT_URB_OFFSET:
 511       return "tes_add_indirect_urb_offset";
 512    case TES_OPCODE_GET_PRIMITIVE_ID:
 513       return "tes_get_primitive_id";
 514
 515    case SHADER_OPCODE_RND_MODE:
 516       return "rnd_mode";
 517    }
 518
 519    unreachable("not reached");
 520 }
 521
 522 bool
 523 brw_saturate_immediate(enum brw_reg_type type, struct brw_reg *reg)
 524 {
 525    union {
 526       unsigned ud;
 527       int d;
 528       float f;
 529       double df;
 530    } imm, sat_imm = { 0 };
 531
 532    const unsigned size = type_sz(type);
 533
 534    /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
 535     * irrelevant, so just check the size of the type and copy from/to an
 536     * appropriately sized field.
 537     */
 538    if (size < 8)
 539       imm.ud = reg->ud;
 540    else
 541       imm.df = reg->df;
 542
 543    switch (type) {
 544    case BRW_REGISTER_TYPE_UD:
 545    case BRW_REGISTER_TYPE_D:
 546    case BRW_REGISTER_TYPE_UW:
 547    case BRW_REGISTER_TYPE_W:
 548    case BRW_REGISTER_TYPE_UQ:
 549    case BRW_REGISTER_TYPE_Q:
 550       /* Nothing to do. */
 551       return false;
 552    case BRW_REGISTER_TYPE_F:
 553       sat_imm.f = CLAMP(imm.f, 0.0f, 1.0f);
 554       break;
 555    case BRW_REGISTER_TYPE_DF:
 556       sat_imm.df = CLAMP(imm.df, 0.0, 1.0);
 557       break;
 558    case BRW_REGISTER_TYPE_UB:
 559    case BRW_REGISTER_TYPE_B:
 560       unreachable("no UB/B immediates");
 561    case BRW_REGISTER_TYPE_V:
 562    case BRW_REGISTER_TYPE_UV:
 563    case BRW_REGISTER_TYPE_VF:
 564       unreachable("unimplemented: saturate vector immediate");
 565    case BRW_REGISTER_TYPE_HF:
 566       unreachable("unimplemented: saturate HF immediate");
 567    case BRW_REGISTER_TYPE_NF:
 568       unreachable("no NF immediates");
 569    }
 570
 571    if (size < 8) {
 572       if (imm.ud != sat_imm.ud) {
 573          reg->ud = sat_imm.ud;
 574          return true;
 575       }
 576    } else {
 577       if (imm.df != sat_imm.df) {
 578          reg->df = sat_imm.df;
 579          return true;
 580       }
 581    }
 582    return false;
 583 }
 584
 585 bool
 586 brw_negate_immediate(enum brw_reg_type type, struct brw_reg *reg)
 587 {
 588    switch (type) {
 589    case BRW_REGISTER_TYPE_D:
 590    case BRW_REGISTER_TYPE_UD:
 591       reg->d = -reg->d;
 592       return true;
 593    case BRW_REGISTER_TYPE_W:
 594    case BRW_REGISTER_TYPE_UW: {
 595       uint16_t value = -(int16_t)reg->ud;
 596       reg->ud = value | (uint32_t)value << 16;
 597       return true;
 598    }
 599    case BRW_REGISTER_TYPE_F:
 600       reg->f = -reg->f;
 601       return true;
 602    case BRW_REGISTER_TYPE_VF:
 603       reg->ud ^= 0x80808080;
 604       return true;
 605    case BRW_REGISTER_TYPE_DF:
 606       reg->df = -reg->df;
 607       return true;
 608    case BRW_REGISTER_TYPE_UQ:
 609    case BRW_REGISTER_TYPE_Q:
 610       reg->d64 = -reg->d64;
 611       return true;
 612    case BRW_REGISTER_TYPE_UB:
 613    case BRW_REGISTER_TYPE_B:
 614       unreachable("no UB/B immediates");
 615    case BRW_REGISTER_TYPE_UV:
 616    case BRW_REGISTER_TYPE_V:
 617       assert(!"unimplemented: negate UV/V immediate");
 618    case BRW_REGISTER_TYPE_HF:
 619       reg->ud ^= 0x80008000;
 620       return true;
 621    case BRW_REGISTER_TYPE_NF:
 622       unreachable("no NF immediates");
 623    }
 624
 625    return false;
 626 }
 627
 628 bool
 629 brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg)
 630 {
 631    switch (type) {
 632    case BRW_REGISTER_TYPE_D:
 633       reg->d = abs(reg->d);
 634       return true;
 635    case BRW_REGISTER_TYPE_W: {
 636       uint16_t value = abs((int16_t)reg->ud);
 637       reg->ud = value | (uint32_t)value << 16;
 638       return true;
 639    }
 640    case BRW_REGISTER_TYPE_F:
 641       reg->f = fabsf(reg->f);
 642       return true;
 643    case BRW_REGISTER_TYPE_DF:
 644       reg->df = fabs(reg->df);
 645       return true;
 646    case BRW_REGISTER_TYPE_VF:
 647       reg->ud &= ~0x80808080;
 648       return true;
 649    case BRW_REGISTER_TYPE_Q:
 650       reg->d64 = imaxabs(reg->d64);
 651       return true;
 652    case BRW_REGISTER_TYPE_UB:
 653    case BRW_REGISTER_TYPE_B:
 654       unreachable("no UB/B immediates");
 655    case BRW_REGISTER_TYPE_UQ:
 656    case BRW_REGISTER_TYPE_UD:
 657    case BRW_REGISTER_TYPE_UW:
 658    case BRW_REGISTER_TYPE_UV:
 659       /* Presumably the absolute value modifier on an unsigned source is a
 660        * nop, but it would be nice to confirm.
 661        */
 662       assert(!"unimplemented: abs unsigned immediate");
 663    case BRW_REGISTER_TYPE_V:
 664       assert(!"unimplemented: abs V immediate");
 665    case BRW_REGISTER_TYPE_HF:
 666       reg->ud &= ~0x80008000;
 667       return true;
 668    case BRW_REGISTER_TYPE_NF:
 669       unreachable("no NF immediates");
 670    }
 671
 672    return false;
 673 }
 674
 675 backend_shader::backend_shader(const struct brw_compiler *compiler,
 676                                void *log_data,
 677                                void *mem_ctx,
 678                                const nir_shader *shader,
 679                                struct brw_stage_prog_data *stage_prog_data)
 680    : compiler(compiler),
 681      log_data(log_data),
 682      devinfo(compiler->devinfo),
 683      nir(shader),
 684      stage_prog_data(stage_prog_data),
 685      mem_ctx(mem_ctx),
 686      cfg(NULL),
 687      stage(shader->info.stage)
 688 {
 689    debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
 690    stage_name = _mesa_shader_stage_to_string(stage);
 691    stage_abbrev = _mesa_shader_stage_to_abbrev(stage);
 692 }
 693
 694 backend_shader::~backend_shader()
 695 {
 696 }
 697
 698 bool
 699 backend_reg::equals(const backend_reg &r) const
 700 {
 701    return brw_regs_equal(this, &r) && offset == r.offset;
 702 }
 703
 704 bool
 705 backend_reg::negative_equals(const backend_reg &r) const
 706 {
 707    return brw_regs_negative_equal(this, &r) && offset == r.offset;
 708 }
 709
 710 bool
 711 backend_reg::is_zero() const
 712 {
 713    if (file != IMM)
 714       return false;
 715
 716    assert(type_sz(type) > 1);
 717
 718    switch (type) {
 719    case BRW_REGISTER_TYPE_HF:
 720       assert((d & 0xffff) == ((d >> 16) & 0xffff));
 721       return (d & 0xffff) == 0 || (d & 0xffff) == 0x8000;
 722    case BRW_REGISTER_TYPE_F:
 723       return f == 0;
 724    case BRW_REGISTER_TYPE_DF:
 725       return df == 0;
 726    case BRW_REGISTER_TYPE_W:
 727    case BRW_REGISTER_TYPE_UW:
 728       assert((d & 0xffff) == ((d >> 16) & 0xffff));
 729       return (d & 0xffff) == 0;
 730    case BRW_REGISTER_TYPE_D:
 731    case BRW_REGISTER_TYPE_UD:
 732       return d == 0;
 733    case BRW_REGISTER_TYPE_UQ:
 734    case BRW_REGISTER_TYPE_Q:
 735       return u64 == 0;
 736    default:
 737       return false;
 738    }
 739 }
 740
 741 bool
 742 backend_reg::is_one() const
 743 {
 744    if (file != IMM)
 745       return false;
 746
 747    assert(type_sz(type) > 1);
 748
 749    switch (type) {
 750    case BRW_REGISTER_TYPE_HF:
 751       assert((d & 0xffff) == ((d >> 16) & 0xffff));
 752       return (d & 0xffff) == 0x3c00;
 753    case BRW_REGISTER_TYPE_F:
 754       return f == 1.0f;
 755    case BRW_REGISTER_TYPE_DF:
 756       return df == 1.0;
 757    case BRW_REGISTER_TYPE_W:
 758    case BRW_REGISTER_TYPE_UW:
 759       assert((d & 0xffff) == ((d >> 16) & 0xffff));
 760       return (d & 0xffff) == 1;
 761    case BRW_REGISTER_TYPE_D:
 762    case BRW_REGISTER_TYPE_UD:
 763       return d == 1;
 764    case BRW_REGISTER_TYPE_UQ:
 765    case BRW_REGISTER_TYPE_Q:
 766       return u64 == 1;
 767    default:
 768       return false;
 769    }
 770 }
 771
 772 bool
 773 backend_reg::is_negative_one() const
 774 {
 775    if (file != IMM)
 776       return false;
 777
 778    assert(type_sz(type) > 1);
 779
 780    switch (type) {
 781    case BRW_REGISTER_TYPE_HF:
 782       assert((d & 0xffff) == ((d >> 16) & 0xffff));
 783       return (d & 0xffff) == 0xbc00;
 784    case BRW_REGISTER_TYPE_F:
 785       return f == -1.0;
 786    case BRW_REGISTER_TYPE_DF:
 787       return df == -1.0;
 788    case BRW_REGISTER_TYPE_W:
 789       assert((d & 0xffff) == ((d >> 16) & 0xffff));
 790       return (d & 0xffff) == 0xffff;
 791    case BRW_REGISTER_TYPE_D:
 792       return d == -1;
 793    case BRW_REGISTER_TYPE_Q:
 794       return d64 == -1;
 795    default:
 796       return false;
 797    }
 798 }
 799
 800 bool
 801 backend_reg::is_null() const
 802 {
 803    return file == ARF && nr == BRW_ARF_NULL;
 804 }
 805
 806
 807 bool
 808 backend_reg::is_accumulator() const
 809 {
 810    return file == ARF && nr == BRW_ARF_ACCUMULATOR;
 811 }
 812
 813 bool
 814 backend_instruction::is_commutative() const
 815 {
 816    switch (opcode) {
 817    case BRW_OPCODE_AND:
 818    case BRW_OPCODE_OR:
 819    case BRW_OPCODE_XOR:
 820    case BRW_OPCODE_ADD:
 821    case BRW_OPCODE_MUL:
 822    case SHADER_OPCODE_MULH:
 823       return true;
 824    case BRW_OPCODE_SEL:
 825       /* MIN and MAX are commutative. */
 826       if (conditional_mod == BRW_CONDITIONAL_GE ||
 827           conditional_mod == BRW_CONDITIONAL_L) {
 828          return true;
 829       }
 830       /* fallthrough */
 831    default:
 832       return false;
 833    }
 834 }
 835
 836 bool
 837 backend_instruction::is_3src(const struct gen_device_info *devinfo) const
 838 {
 839    return ::is_3src(devinfo, opcode);
 840 }
 841
 842 bool
 843 backend_instruction::is_tex() const
 844 {
 845    return (opcode == SHADER_OPCODE_TEX ||
 846            opcode == FS_OPCODE_TXB ||
 847            opcode == SHADER_OPCODE_TXD ||
 848            opcode == SHADER_OPCODE_TXF ||
 849            opcode == SHADER_OPCODE_TXF_LZ ||
 850            opcode == SHADER_OPCODE_TXF_CMS ||
 851            opcode == SHADER_OPCODE_TXF_CMS_W ||
 852            opcode == SHADER_OPCODE_TXF_UMS ||
 853            opcode == SHADER_OPCODE_TXF_MCS ||
 854            opcode == SHADER_OPCODE_TXL ||
 855            opcode == SHADER_OPCODE_TXL_LZ ||
 856            opcode == SHADER_OPCODE_TXS ||
 857            opcode == SHADER_OPCODE_LOD ||
 858            opcode == SHADER_OPCODE_TG4 ||
 859            opcode == SHADER_OPCODE_TG4_OFFSET ||
 860            opcode == SHADER_OPCODE_SAMPLEINFO);
 861 }
 862
 863 bool
 864 backend_instruction::is_math() const
 865 {
 866    return (opcode == SHADER_OPCODE_RCP ||
 867            opcode == SHADER_OPCODE_RSQ ||
 868            opcode == SHADER_OPCODE_SQRT ||
 869            opcode == SHADER_OPCODE_EXP2 ||
 870            opcode == SHADER_OPCODE_LOG2 ||
 871            opcode == SHADER_OPCODE_SIN ||
 872            opcode == SHADER_OPCODE_COS ||
 873            opcode == SHADER_OPCODE_INT_QUOTIENT ||
 874            opcode == SHADER_OPCODE_INT_REMAINDER ||
 875            opcode == SHADER_OPCODE_POW);
 876 }
 877
 878 bool
 879 backend_instruction::is_control_flow() const
 880 {
 881    switch (opcode) {
 882    case BRW_OPCODE_DO:
 883    case BRW_OPCODE_WHILE:
 884    case BRW_OPCODE_IF:
 885    case BRW_OPCODE_ELSE:
 886    case BRW_OPCODE_ENDIF:
 887    case BRW_OPCODE_BREAK:
 888    case BRW_OPCODE_CONTINUE:
 889       return true;
 890    default:
 891       return false;
 892    }
 893 }
 894
 895 bool
 896 backend_instruction::can_do_source_mods() const
 897 {
 898    switch (opcode) {
 899    case BRW_OPCODE_ADDC:
 900    case BRW_OPCODE_BFE:
 901    case BRW_OPCODE_BFI1:
 902    case BRW_OPCODE_BFI2:
 903    case BRW_OPCODE_BFREV:
 904    case BRW_OPCODE_CBIT:
 905    case BRW_OPCODE_FBH:
 906    case BRW_OPCODE_FBL:
 907    case BRW_OPCODE_SUBB:
 908    case SHADER_OPCODE_BROADCAST:
 909    case SHADER_OPCODE_CLUSTER_BROADCAST:
 910    case SHADER_OPCODE_MOV_INDIRECT:
 911       return false;
 912    default:
 913       return true;
 914    }
 915 }
 916
 917 bool
 918 backend_instruction::can_do_saturate() const
 919 {
 920    switch (opcode) {
 921    case BRW_OPCODE_ADD:
 922    case BRW_OPCODE_ASR:
 923    case BRW_OPCODE_AVG:
 924    case BRW_OPCODE_DP2:
 925    case BRW_OPCODE_DP3:
 926    case BRW_OPCODE_DP4:
 927    case BRW_OPCODE_DPH:
 928    case BRW_OPCODE_F16TO32:
 929    case BRW_OPCODE_F32TO16:
 930    case BRW_OPCODE_LINE:
 931    case BRW_OPCODE_LRP:
 932    case BRW_OPCODE_MAC:
 933    case BRW_OPCODE_MAD:
 934    case BRW_OPCODE_MATH:
 935    case BRW_OPCODE_MOV:
 936    case BRW_OPCODE_MUL:
 937    case SHADER_OPCODE_MULH:
 938    case BRW_OPCODE_PLN:
 939    case BRW_OPCODE_RNDD:
 940    case BRW_OPCODE_RNDE:
 941    case BRW_OPCODE_RNDU:
 942    case BRW_OPCODE_RNDZ:
 943    case BRW_OPCODE_SEL:
 944    case BRW_OPCODE_SHL:
 945    case BRW_OPCODE_SHR:
 946    case FS_OPCODE_LINTERP:
 947    case SHADER_OPCODE_COS:
 948    case SHADER_OPCODE_EXP2:
 949    case SHADER_OPCODE_LOG2:
 950    case SHADER_OPCODE_POW:
 951    case SHADER_OPCODE_RCP:
 952    case SHADER_OPCODE_RSQ:
 953    case SHADER_OPCODE_SIN:
 954    case SHADER_OPCODE_SQRT:
 955       return true;
 956    default:
 957       return false;
 958    }
 959 }
 960
 961 bool
 962 backend_instruction::can_do_cmod() const
 963 {
 964    switch (opcode) {
 965    case BRW_OPCODE_ADD:
 966    case BRW_OPCODE_ADDC:
 967    case BRW_OPCODE_AND:
 968    case BRW_OPCODE_ASR:
 969    case BRW_OPCODE_AVG:
 970    case BRW_OPCODE_CMP:
 971    case BRW_OPCODE_CMPN:
 972    case BRW_OPCODE_DP2:
 973    case BRW_OPCODE_DP3:
 974    case BRW_OPCODE_DP4:
 975    case BRW_OPCODE_DPH:
 976    case BRW_OPCODE_F16TO32:
 977    case BRW_OPCODE_F32TO16:
 978    case BRW_OPCODE_FRC:
 979    case BRW_OPCODE_LINE:
 980    case BRW_OPCODE_LRP:
 981    case BRW_OPCODE_LZD:
 982    case BRW_OPCODE_MAC:
 983    case BRW_OPCODE_MACH:
 984    case BRW_OPCODE_MAD:
 985    case BRW_OPCODE_MOV:
 986    case BRW_OPCODE_MUL:
 987    case BRW_OPCODE_NOT:
 988    case BRW_OPCODE_OR:
 989    case BRW_OPCODE_PLN:
 990    case BRW_OPCODE_RNDD:
 991    case BRW_OPCODE_RNDE:
 992    case BRW_OPCODE_RNDU:
 993    case BRW_OPCODE_RNDZ:
 994    case BRW_OPCODE_SAD2:
 995    case BRW_OPCODE_SADA2:
 996    case BRW_OPCODE_SHL:
 997    case BRW_OPCODE_SHR:
 998    case BRW_OPCODE_SUBB:
 999    case BRW_OPCODE_XOR:
1000    case FS_OPCODE_LINTERP:
1001       return true;
1002    default:
1003       return false;
1004    }
1005 }
1006
1007 bool
1008 backend_instruction::reads_accumulator_implicitly() const
1009 {
1010    switch (opcode) {
1011    case BRW_OPCODE_MAC:
1012    case BRW_OPCODE_MACH:
1013    case BRW_OPCODE_SADA2:
1014       return true;
1015    default:
1016       return false;
1017    }
1018 }
1019
1020 bool
1021 backend_instruction::writes_accumulator_implicitly(const struct gen_device_info *devinfo) const
1022 {
1023    return writes_accumulator ||
1024           (devinfo->gen < 6 &&
1025            ((opcode >= BRW_OPCODE_ADD && opcode < BRW_OPCODE_NOP) ||
1026             (opcode >= FS_OPCODE_DDX_COARSE && opcode <= FS_OPCODE_LINTERP))) ||
1027           (opcode == FS_OPCODE_LINTERP &&
1028            (!devinfo->has_pln || devinfo->gen <= 6));
1029 }
1030
1031 bool
1032 backend_instruction::has_side_effects() const
1033 {
1034    switch (opcode) {
1035    case SHADER_OPCODE_SEND:
1036       return send_has_side_effects;
1037
1038    case VEC4_OPCODE_UNTYPED_ATOMIC:
1039    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
1040    case SHADER_OPCODE_UNTYPED_ATOMIC_FLOAT_LOGICAL:
1041    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
1042    case VEC4_OPCODE_UNTYPED_SURFACE_WRITE:
1043    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
1044    case SHADER_OPCODE_A64_UNTYPED_WRITE_LOGICAL:
1045    case SHADER_OPCODE_A64_BYTE_SCATTERED_WRITE_LOGICAL:
1046    case SHADER_OPCODE_A64_UNTYPED_ATOMIC_LOGICAL:
1047    case SHADER_OPCODE_A64_UNTYPED_ATOMIC_FLOAT_LOGICAL:
1048    case SHADER_OPCODE_BYTE_SCATTERED_WRITE_LOGICAL:
1049    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
1050    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
1051    case SHADER_OPCODE_MEMORY_FENCE:
1052    case SHADER_OPCODE_INTERLOCK:
1053    case SHADER_OPCODE_URB_WRITE_SIMD8:
1054    case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
1055    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
1056    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
1057    case FS_OPCODE_FB_WRITE:
1058    case FS_OPCODE_FB_WRITE_LOGICAL:
1059    case FS_OPCODE_REP_FB_WRITE:
1060    case SHADER_OPCODE_BARRIER:
1061    case TCS_OPCODE_URB_WRITE:
1062    case TCS_OPCODE_RELEASE_INPUT:
1063    case SHADER_OPCODE_RND_MODE:
1064       return true;
1065    default:
1066       return eot;
1067    }
1068 }
1069
1070 bool
1071 backend_instruction::is_volatile() const
1072 {
1073    switch (opcode) {
1074    case SHADER_OPCODE_SEND:
1075       return send_is_volatile;
1076
1077    case VEC4_OPCODE_UNTYPED_SURFACE_READ:
1078    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
1079    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
1080    case SHADER_OPCODE_BYTE_SCATTERED_READ_LOGICAL:
1081    case SHADER_OPCODE_A64_UNTYPED_READ_LOGICAL:
1082    case SHADER_OPCODE_A64_BYTE_SCATTERED_READ_LOGICAL:
1083    case SHADER_OPCODE_URB_READ_SIMD8:
1084    case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
1085    case VEC4_OPCODE_URB_READ:
1086       return true;
1087    default:
1088       return false;
1089    }
1090 }
1091
1092 #ifndef NDEBUG
1093 static bool
1094 inst_is_in_block(const bblock_t *block, const backend_instruction *inst)
1095 {
1096    bool found = false;
1097    foreach_inst_in_block (backend_instruction, i, block) {
1098       if (inst == i) {
1099          found = true;
1100       }
1101    }
1102    return found;
1103 }
1104 #endif
1105
1106 static void
1107 adjust_later_block_ips(bblock_t *start_block, int ip_adjustment)
1108 {
1109    for (bblock_t *block_iter = start_block->next();
1110         block_iter;
1111         block_iter = block_iter->next()) {
1112       block_iter->start_ip += ip_adjustment;
1113       block_iter->end_ip += ip_adjustment;
1114    }
1115 }
1116
1117 void
1118 backend_instruction::insert_after(bblock_t *block, backend_instruction *inst)
1119 {
1120    assert(this != inst);
1121
1122    if (!this->is_head_sentinel())
1123       assert(inst_is_in_block(block, this) || !"Instruction not in block");
1124
1125    block->end_ip++;
1126
1127    adjust_later_block_ips(block, 1);
1128
1129    exec_node::insert_after(inst);
1130 }
1131
1132 void
1133 backend_instruction::insert_before(bblock_t *block, backend_instruction *inst)
1134 {
1135    assert(this != inst);
1136
1137    if (!this->is_tail_sentinel())
1138       assert(inst_is_in_block(block, this) || !"Instruction not in block");
1139
1140    block->end_ip++;
1141
1142    adjust_later_block_ips(block, 1);
1143
1144    exec_node::insert_before(inst);
1145 }
1146
1147 void
1148 backend_instruction::insert_before(bblock_t *block, exec_list *list)
1149 {
1150    assert(inst_is_in_block(block, this) || !"Instruction not in block");
1151
1152    unsigned num_inst = list->length();
1153
1154    block->end_ip += num_inst;
1155
1156    adjust_later_block_ips(block, num_inst);
1157
1158    exec_node::insert_before(list);
1159 }
1160
1161 void
1162 backend_instruction::remove(bblock_t *block)
1163 {
1164    assert(inst_is_in_block(block, this) || !"Instruction not in block");
1165
1166    adjust_later_block_ips(block, -1);
1167
1168    if (block->start_ip == block->end_ip) {
1169       block->cfg->remove_block(block);
1170    } else {
1171       block->end_ip--;
1172    }
1173
1174    exec_node::remove();
1175 }
1176
1177 void
1178 backend_shader::dump_instructions()
1179 {
1180    dump_instructions(NULL);
1181 }
1182
1183 void
1184 backend_shader::dump_instructions(const char *name)
1185 {
1186    FILE *file = stderr;
1187    if (name && geteuid() != 0) {
1188       file = fopen(name, "w");
1189       if (!file)
1190          file = stderr;
1191    }
1192
1193    if (cfg) {
1194       int ip = 0;
1195       foreach_block_and_inst(block, backend_instruction, inst, cfg) {
1196          if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1197             fprintf(file, "%4d: ", ip++);
1198          dump_instruction(inst, file);
1199       }
1200    } else {
1201       int ip = 0;
1202       foreach_in_list(backend_instruction, inst, &instructions) {
1203          if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1204             fprintf(file, "%4d: ", ip++);
1205          dump_instruction(inst, file);
1206       }
1207    }
1208
1209    if (file != stderr) {
1210       fclose(file);
1211    }
1212 }
1213
1214 void
1215 backend_shader::calculate_cfg()
1216 {
1217    if (this->cfg)
1218       return;
1219    cfg = new(mem_ctx) cfg_t(&this->instructions);
1220 }
1221
1222 extern "C" const unsigned *
1223 brw_compile_tes(const struct brw_compiler *compiler,
1224                 void *log_data,
1225                 void *mem_ctx,
1226                 const struct brw_tes_prog_key *key,
1227                 const struct brw_vue_map *input_vue_map,
1228                 struct brw_tes_prog_data *prog_data,
1229                 nir_shader *nir,
1230                 struct gl_program *prog,
1231                 int shader_time_index,
1232                 char **error_str)
1233 {
1234    const struct gen_device_info *devinfo = compiler->devinfo;
1235    const bool is_scalar = compiler->scalar_stage[MESA_SHADER_TESS_EVAL];
1236    const unsigned *assembly;
1237
1238    nir->info.inputs_read = key->inputs_read;
1239    nir->info.patch_inputs_read = key->patch_inputs_read;
1240
1241    nir = brw_nir_apply_sampler_key(nir, compiler, &key->tex, is_scalar);
1242    brw_nir_lower_tes_inputs(nir, input_vue_map);
1243    brw_nir_lower_vue_outputs(nir);
1244    nir = brw_postprocess_nir(nir, compiler, is_scalar);
1245
1246    brw_compute_vue_map(devinfo, &prog_data->base.vue_map,
1247                        nir->info.outputs_written,
1248                        nir->info.separate_shader);
1249
1250    unsigned output_size_bytes = prog_data->base.vue_map.num_slots * 4 * 4;
1251
1252    assert(output_size_bytes >= 1);
1253    if (output_size_bytes > GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES) {
1254       if (error_str)
1255          *error_str = ralloc_strdup(mem_ctx, "DS outputs exceed maximum size");
1256       return NULL;
1257    }
1258
1259    prog_data->base.clip_distance_mask =
1260       ((1 << nir->info.clip_distance_array_size) - 1);
1261    prog_data->base.cull_distance_mask =
1262       ((1 << nir->info.cull_distance_array_size) - 1) <<
1263       nir->info.clip_distance_array_size;
1264
1265    /* URB entry sizes are stored as a multiple of 64 bytes. */
1266    prog_data->base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64;
1267
1268    /* On Cannonlake software shall not program an allocation size that
1269     * specifies a size that is a multiple of 3 64B (512-bit) cachelines.
1270     */
1271    if (devinfo->gen == 10 &&
1272        prog_data->base.urb_entry_size % 3 == 0)
1273       prog_data->base.urb_entry_size++;
1274
1275    prog_data->base.urb_read_length = 0;
1276
1277    STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER == TESS_SPACING_EQUAL - 1);
1278    STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL ==
1279                  TESS_SPACING_FRACTIONAL_ODD - 1);
1280    STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL ==
1281                  TESS_SPACING_FRACTIONAL_EVEN - 1);
1282
1283    prog_data->partitioning =
1284       (enum brw_tess_partitioning) (nir->info.tess.spacing - 1);
1285
1286    switch (nir->info.tess.primitive_mode) {
1287    case GL_QUADS:
1288       prog_data->domain = BRW_TESS_DOMAIN_QUAD;
1289       break;
1290    case GL_TRIANGLES:
1291       prog_data->domain = BRW_TESS_DOMAIN_TRI;
1292       break;
1293    case GL_ISOLINES:
1294       prog_data->domain = BRW_TESS_DOMAIN_ISOLINE;
1295       break;
1296    default:
1297       unreachable("invalid domain shader primitive mode");
1298    }
1299
1300    if (nir->info.tess.point_mode) {
1301       prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT;
1302    } else if (nir->info.tess.primitive_mode == GL_ISOLINES) {
1303       prog_data->output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE;
1304    } else {
1305       /* Hardware winding order is backwards from OpenGL */
1306       prog_data->output_topology =
1307          nir->info.tess.ccw ? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
1308                              : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW;
1309    }
1310
1311    if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1312       fprintf(stderr, "TES Input ");
1313       brw_print_vue_map(stderr, input_vue_map);
1314       fprintf(stderr, "TES Output ");
1315       brw_print_vue_map(stderr, &prog_data->base.vue_map);
1316    }
1317
1318    if (is_scalar) {
1319       fs_visitor v(compiler, log_data, mem_ctx, (void *) key,
1320                    &prog_data->base.base, NULL, nir, 8,
1321                    shader_time_index, input_vue_map);
1322       if (!v.run_tes()) {
1323          if (error_str)
1324             *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1325          return NULL;
1326       }
1327
1328       prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
1329       prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8;
1330
1331       fs_generator g(compiler, log_data, mem_ctx,
1332                      &prog_data->base.base, v.promoted_constants, false,
1333                      MESA_SHADER_TESS_EVAL);
1334       if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1335          g.enable_debug(ralloc_asprintf(mem_ctx,
1336                                         "%s tessellation evaluation shader %s",
1337                                         nir->info.label ? nir->info.label
1338                                                         : "unnamed",
1339                                         nir->info.name));
1340       }
1341
1342       g.generate_code(v.cfg, 8);
1343
1344       assembly = g.get_assembly();
1345    } else {
1346       brw::vec4_tes_visitor v(compiler, log_data, key, prog_data,
1347                               nir, mem_ctx, shader_time_index);
1348       if (!v.run()) {
1349          if (error_str)
1350             *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1351          return NULL;
1352       }
1353
1354       if (unlikely(INTEL_DEBUG & DEBUG_TES))
1355          v.dump_instructions();
1356
1357       assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir,
1358                                             &prog_data->base, v.cfg);
1359    }
1360
1361    return assembly;
1362 }