src/mesa/drivers/dri/i965/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_context.h"
  25 #include "brw_cfg.h"
  26 #include "brw_eu.h"
  27 #include "brw_fs.h"
  28 #include "brw_nir.h"
  29 #include "brw_vec4_tes.h"
  30 #include "main/uniforms.h"
  31
  32 extern "C" void
  33 brw_mark_surface_used(struct brw_stage_prog_data *prog_data,
  34                       unsigned surf_index)
  35 {
  36    assert(surf_index < BRW_MAX_SURFACES);
  37
  38    prog_data->binding_table.size_bytes =
  39       MAX2(prog_data->binding_table.size_bytes, (surf_index + 1) * 4);
  40 }
  41
  42 enum brw_reg_type
  43 brw_type_for_base_type(const struct glsl_type *type)
  44 {
  45    switch (type->base_type) {
  46    case GLSL_TYPE_FLOAT:
  47       return BRW_REGISTER_TYPE_F;
  48    case GLSL_TYPE_INT:
  49    case GLSL_TYPE_BOOL:
  50    case GLSL_TYPE_SUBROUTINE:
  51       return BRW_REGISTER_TYPE_D;
  52    case GLSL_TYPE_UINT:
  53       return BRW_REGISTER_TYPE_UD;
  54    case GLSL_TYPE_ARRAY:
  55       return brw_type_for_base_type(type->fields.array);
  56    case GLSL_TYPE_STRUCT:
  57    case GLSL_TYPE_SAMPLER:
  58    case GLSL_TYPE_ATOMIC_UINT:
  59       /* These should be overridden with the type of the member when
  60        * dereferenced into.  BRW_REGISTER_TYPE_UD seems like a likely
  61        * way to trip up if we don't.
  62        */
  63       return BRW_REGISTER_TYPE_UD;
  64    case GLSL_TYPE_IMAGE:
  65       return BRW_REGISTER_TYPE_UD;
  66    case GLSL_TYPE_DOUBLE:
  67       return BRW_REGISTER_TYPE_DF;
  68    case GLSL_TYPE_VOID:
  69    case GLSL_TYPE_ERROR:
  70    case GLSL_TYPE_INTERFACE:
  71    case GLSL_TYPE_FUNCTION:
  72       unreachable("not reached");
  73    }
  74
  75    return BRW_REGISTER_TYPE_F;
  76 }
  77
  78 enum brw_conditional_mod
  79 brw_conditional_for_comparison(unsigned int op)
  80 {
  81    switch (op) {
  82    case ir_binop_less:
  83       return BRW_CONDITIONAL_L;
  84    case ir_binop_greater:
  85       return BRW_CONDITIONAL_G;
  86    case ir_binop_lequal:
  87       return BRW_CONDITIONAL_LE;
  88    case ir_binop_gequal:
  89       return BRW_CONDITIONAL_GE;
  90    case ir_binop_equal:
  91    case ir_binop_all_equal: /* same as equal for scalars */
  92       return BRW_CONDITIONAL_Z;
  93    case ir_binop_nequal:
  94    case ir_binop_any_nequal: /* same as nequal for scalars */
  95       return BRW_CONDITIONAL_NZ;
  96    default:
  97       unreachable("not reached: bad operation for comparison");
  98    }
  99 }
 100
 101 uint32_t
 102 brw_math_function(enum opcode op)
 103 {
 104    switch (op) {
 105    case SHADER_OPCODE_RCP:
 106       return BRW_MATH_FUNCTION_INV;
 107    case SHADER_OPCODE_RSQ:
 108       return BRW_MATH_FUNCTION_RSQ;
 109    case SHADER_OPCODE_SQRT:
 110       return BRW_MATH_FUNCTION_SQRT;
 111    case SHADER_OPCODE_EXP2:
 112       return BRW_MATH_FUNCTION_EXP;
 113    case SHADER_OPCODE_LOG2:
 114       return BRW_MATH_FUNCTION_LOG;
 115    case SHADER_OPCODE_POW:
 116       return BRW_MATH_FUNCTION_POW;
 117    case SHADER_OPCODE_SIN:
 118       return BRW_MATH_FUNCTION_SIN;
 119    case SHADER_OPCODE_COS:
 120       return BRW_MATH_FUNCTION_COS;
 121    case SHADER_OPCODE_INT_QUOTIENT:
 122       return BRW_MATH_FUNCTION_INT_DIV_QUOTIENT;
 123    case SHADER_OPCODE_INT_REMAINDER:
 124       return BRW_MATH_FUNCTION_INT_DIV_REMAINDER;
 125    default:
 126       unreachable("not reached: unknown math function");
 127    }
 128 }
 129
 130 uint32_t
 131 brw_texture_offset(int *offsets, unsigned num_components)
 132 {
 133    if (!offsets) return 0;  /* nonconstant offset; caller will handle it. */
 134
 135    /* Combine all three offsets into a single unsigned dword:
 136     *
 137     *    bits 11:8 - U Offset (X component)
 138     *    bits  7:4 - V Offset (Y component)
 139     *    bits  3:0 - R Offset (Z component)
 140     */
 141    unsigned offset_bits = 0;
 142    for (unsigned i = 0; i < num_components; i++) {
 143       const unsigned shift = 4 * (2 - i);
 144       offset_bits |= (offsets[i] << shift) & (0xF << shift);
 145    }
 146    return offset_bits;
 147 }
 148
 149 const char *
 150 brw_instruction_name(const struct gen_device_info *devinfo, enum opcode op)
 151 {
 152    switch (op) {
 153    case BRW_OPCODE_ILLEGAL ... BRW_OPCODE_NOP:
 154       /* The DO instruction doesn't exist on Gen6+, but we use it to mark the
 155        * start of a loop in the IR.
 156        */
 157       if (devinfo->gen >= 6 && op == BRW_OPCODE_DO)
 158          return "do";
 159
 160       assert(brw_opcode_desc(devinfo, op)->name);
 161       return brw_opcode_desc(devinfo, op)->name;
 162    case FS_OPCODE_FB_WRITE:
 163       return "fb_write";
 164    case FS_OPCODE_FB_WRITE_LOGICAL:
 165       return "fb_write_logical";
 166    case FS_OPCODE_REP_FB_WRITE:
 167       return "rep_fb_write";
 168    case FS_OPCODE_FB_READ:
 169       return "fb_read";
 170    case FS_OPCODE_FB_READ_LOGICAL:
 171       return "fb_read_logical";
 172
 173    case SHADER_OPCODE_RCP:
 174       return "rcp";
 175    case SHADER_OPCODE_RSQ:
 176       return "rsq";
 177    case SHADER_OPCODE_SQRT:
 178       return "sqrt";
 179    case SHADER_OPCODE_EXP2:
 180       return "exp2";
 181    case SHADER_OPCODE_LOG2:
 182       return "log2";
 183    case SHADER_OPCODE_POW:
 184       return "pow";
 185    case SHADER_OPCODE_INT_QUOTIENT:
 186       return "int_quot";
 187    case SHADER_OPCODE_INT_REMAINDER:
 188       return "int_rem";
 189    case SHADER_OPCODE_SIN:
 190       return "sin";
 191    case SHADER_OPCODE_COS:
 192       return "cos";
 193
 194    case SHADER_OPCODE_TEX:
 195       return "tex";
 196    case SHADER_OPCODE_TEX_LOGICAL:
 197       return "tex_logical";
 198    case SHADER_OPCODE_TXD:
 199       return "txd";
 200    case SHADER_OPCODE_TXD_LOGICAL:
 201       return "txd_logical";
 202    case SHADER_OPCODE_TXF:
 203       return "txf";
 204    case SHADER_OPCODE_TXF_LOGICAL:
 205       return "txf_logical";
 206    case SHADER_OPCODE_TXF_LZ:
 207       return "txf_lz";
 208    case SHADER_OPCODE_TXL:
 209       return "txl";
 210    case SHADER_OPCODE_TXL_LOGICAL:
 211       return "txl_logical";
 212    case SHADER_OPCODE_TXL_LZ:
 213       return "txl_lz";
 214    case SHADER_OPCODE_TXS:
 215       return "txs";
 216    case SHADER_OPCODE_TXS_LOGICAL:
 217       return "txs_logical";
 218    case FS_OPCODE_TXB:
 219       return "txb";
 220    case FS_OPCODE_TXB_LOGICAL:
 221       return "txb_logical";
 222    case SHADER_OPCODE_TXF_CMS:
 223       return "txf_cms";
 224    case SHADER_OPCODE_TXF_CMS_LOGICAL:
 225       return "txf_cms_logical";
 226    case SHADER_OPCODE_TXF_CMS_W:
 227       return "txf_cms_w";
 228    case SHADER_OPCODE_TXF_CMS_W_LOGICAL:
 229       return "txf_cms_w_logical";
 230    case SHADER_OPCODE_TXF_UMS:
 231       return "txf_ums";
 232    case SHADER_OPCODE_TXF_UMS_LOGICAL:
 233       return "txf_ums_logical";
 234    case SHADER_OPCODE_TXF_MCS:
 235       return "txf_mcs";
 236    case SHADER_OPCODE_TXF_MCS_LOGICAL:
 237       return "txf_mcs_logical";
 238    case SHADER_OPCODE_LOD:
 239       return "lod";
 240    case SHADER_OPCODE_LOD_LOGICAL:
 241       return "lod_logical";
 242    case SHADER_OPCODE_TG4:
 243       return "tg4";
 244    case SHADER_OPCODE_TG4_LOGICAL:
 245       return "tg4_logical";
 246    case SHADER_OPCODE_TG4_OFFSET:
 247       return "tg4_offset";
 248    case SHADER_OPCODE_TG4_OFFSET_LOGICAL:
 249       return "tg4_offset_logical";
 250    case SHADER_OPCODE_SAMPLEINFO:
 251       return "sampleinfo";
 252    case SHADER_OPCODE_SAMPLEINFO_LOGICAL:
 253       return "sampleinfo_logical";
 254
 255    case SHADER_OPCODE_SHADER_TIME_ADD:
 256       return "shader_time_add";
 257
 258    case SHADER_OPCODE_UNTYPED_ATOMIC:
 259       return "untyped_atomic";
 260    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
 261       return "untyped_atomic_logical";
 262    case SHADER_OPCODE_UNTYPED_SURFACE_READ:
 263       return "untyped_surface_read";
 264    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
 265       return "untyped_surface_read_logical";
 266    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
 267       return "untyped_surface_write";
 268    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
 269       return "untyped_surface_write_logical";
 270    case SHADER_OPCODE_TYPED_ATOMIC:
 271       return "typed_atomic";
 272    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
 273       return "typed_atomic_logical";
 274    case SHADER_OPCODE_TYPED_SURFACE_READ:
 275       return "typed_surface_read";
 276    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
 277       return "typed_surface_read_logical";
 278    case SHADER_OPCODE_TYPED_SURFACE_WRITE:
 279       return "typed_surface_write";
 280    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
 281       return "typed_surface_write_logical";
 282    case SHADER_OPCODE_MEMORY_FENCE:
 283       return "memory_fence";
 284
 285    case SHADER_OPCODE_LOAD_PAYLOAD:
 286       return "load_payload";
 287    case FS_OPCODE_PACK:
 288       return "pack";
 289
 290    case SHADER_OPCODE_GEN4_SCRATCH_READ:
 291       return "gen4_scratch_read";
 292    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
 293       return "gen4_scratch_write";
 294    case SHADER_OPCODE_GEN7_SCRATCH_READ:
 295       return "gen7_scratch_read";
 296    case SHADER_OPCODE_URB_WRITE_SIMD8:
 297       return "gen8_urb_write_simd8";
 298    case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
 299       return "gen8_urb_write_simd8_per_slot";
 300    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
 301       return "gen8_urb_write_simd8_masked";
 302    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
 303       return "gen8_urb_write_simd8_masked_per_slot";
 304    case SHADER_OPCODE_URB_READ_SIMD8:
 305       return "urb_read_simd8";
 306    case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
 307       return "urb_read_simd8_per_slot";
 308
 309    case SHADER_OPCODE_FIND_LIVE_CHANNEL:
 310       return "find_live_channel";
 311    case SHADER_OPCODE_BROADCAST:
 312       return "broadcast";
 313
 314    case VEC4_OPCODE_MOV_BYTES:
 315       return "mov_bytes";
 316    case VEC4_OPCODE_PACK_BYTES:
 317       return "pack_bytes";
 318    case VEC4_OPCODE_UNPACK_UNIFORM:
 319       return "unpack_uniform";
 320
 321    case FS_OPCODE_DDX_COARSE:
 322       return "ddx_coarse";
 323    case FS_OPCODE_DDX_FINE:
 324       return "ddx_fine";
 325    case FS_OPCODE_DDY_COARSE:
 326       return "ddy_coarse";
 327    case FS_OPCODE_DDY_FINE:
 328       return "ddy_fine";
 329
 330    case FS_OPCODE_CINTERP:
 331       return "cinterp";
 332    case FS_OPCODE_LINTERP:
 333       return "linterp";
 334
 335    case FS_OPCODE_PIXEL_X:
 336       return "pixel_x";
 337    case FS_OPCODE_PIXEL_Y:
 338       return "pixel_y";
 339
 340    case FS_OPCODE_GET_BUFFER_SIZE:
 341       return "fs_get_buffer_size";
 342
 343    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
 344       return "uniform_pull_const";
 345    case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7:
 346       return "uniform_pull_const_gen7";
 347    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN4:
 348       return "varying_pull_const_gen4";
 349    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7:
 350       return "varying_pull_const_gen7";
 351    case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_LOGICAL:
 352       return "varying_pull_const_logical";
 353
 354    case FS_OPCODE_MOV_DISPATCH_TO_FLAGS:
 355       return "mov_dispatch_to_flags";
 356    case FS_OPCODE_DISCARD_JUMP:
 357       return "discard_jump";
 358
 359    case FS_OPCODE_SET_SAMPLE_ID:
 360       return "set_sample_id";
 361    case FS_OPCODE_SET_SIMD4X2_OFFSET:
 362       return "set_simd4x2_offset";
 363
 364    case FS_OPCODE_PACK_HALF_2x16_SPLIT:
 365       return "pack_half_2x16_split";
 366    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X:
 367       return "unpack_half_2x16_split_x";
 368    case FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y:
 369       return "unpack_half_2x16_split_y";
 370
 371    case FS_OPCODE_PLACEHOLDER_HALT:
 372       return "placeholder_halt";
 373
 374    case FS_OPCODE_INTERPOLATE_AT_SAMPLE:
 375       return "interp_sample";
 376    case FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET:
 377       return "interp_shared_offset";
 378    case FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET:
 379       return "interp_per_slot_offset";
 380
 381    case VS_OPCODE_URB_WRITE:
 382       return "vs_urb_write";
 383    case VS_OPCODE_PULL_CONSTANT_LOAD:
 384       return "pull_constant_load";
 385    case VS_OPCODE_PULL_CONSTANT_LOAD_GEN7:
 386       return "pull_constant_load_gen7";
 387
 388    case VS_OPCODE_SET_SIMD4X2_HEADER_GEN9:
 389       return "set_simd4x2_header_gen9";
 390
 391    case VS_OPCODE_GET_BUFFER_SIZE:
 392       return "vs_get_buffer_size";
 393
 394    case VS_OPCODE_UNPACK_FLAGS_SIMD4X2:
 395       return "unpack_flags_simd4x2";
 396
 397    case GS_OPCODE_URB_WRITE:
 398       return "gs_urb_write";
 399    case GS_OPCODE_URB_WRITE_ALLOCATE:
 400       return "gs_urb_write_allocate";
 401    case GS_OPCODE_THREAD_END:
 402       return "gs_thread_end";
 403    case GS_OPCODE_SET_WRITE_OFFSET:
 404       return "set_write_offset";
 405    case GS_OPCODE_SET_VERTEX_COUNT:
 406       return "set_vertex_count";
 407    case GS_OPCODE_SET_DWORD_2:
 408       return "set_dword_2";
 409    case GS_OPCODE_PREPARE_CHANNEL_MASKS:
 410       return "prepare_channel_masks";
 411    case GS_OPCODE_SET_CHANNEL_MASKS:
 412       return "set_channel_masks";
 413    case GS_OPCODE_GET_INSTANCE_ID:
 414       return "get_instance_id";
 415    case GS_OPCODE_FF_SYNC:
 416       return "ff_sync";
 417    case GS_OPCODE_SET_PRIMITIVE_ID:
 418       return "set_primitive_id";
 419    case GS_OPCODE_SVB_WRITE:
 420       return "gs_svb_write";
 421    case GS_OPCODE_SVB_SET_DST_INDEX:
 422       return "gs_svb_set_dst_index";
 423    case GS_OPCODE_FF_SYNC_SET_PRIMITIVES:
 424       return "gs_ff_sync_set_primitives";
 425    case CS_OPCODE_CS_TERMINATE:
 426       return "cs_terminate";
 427    case SHADER_OPCODE_BARRIER:
 428       return "barrier";
 429    case SHADER_OPCODE_MULH:
 430       return "mulh";
 431    case SHADER_OPCODE_MOV_INDIRECT:
 432       return "mov_indirect";
 433
 434    case VEC4_OPCODE_URB_READ:
 435       return "urb_read";
 436    case TCS_OPCODE_GET_INSTANCE_ID:
 437       return "tcs_get_instance_id";
 438    case TCS_OPCODE_URB_WRITE:
 439       return "tcs_urb_write";
 440    case TCS_OPCODE_SET_INPUT_URB_OFFSETS:
 441       return "tcs_set_input_urb_offsets";
 442    case TCS_OPCODE_SET_OUTPUT_URB_OFFSETS:
 443       return "tcs_set_output_urb_offsets";
 444    case TCS_OPCODE_GET_PRIMITIVE_ID:
 445       return "tcs_get_primitive_id";
 446    case TCS_OPCODE_CREATE_BARRIER_HEADER:
 447       return "tcs_create_barrier_header";
 448    case TCS_OPCODE_SRC0_010_IS_ZERO:
 449       return "tcs_src0<0,1,0>_is_zero";
 450    case TCS_OPCODE_RELEASE_INPUT:
 451       return "tcs_release_input";
 452    case TCS_OPCODE_THREAD_END:
 453       return "tcs_thread_end";
 454    case TES_OPCODE_CREATE_INPUT_READ_HEADER:
 455       return "tes_create_input_read_header";
 456    case TES_OPCODE_ADD_INDIRECT_URB_OFFSET:
 457       return "tes_add_indirect_urb_offset";
 458    case TES_OPCODE_GET_PRIMITIVE_ID:
 459       return "tes_get_primitive_id";
 460    }
 461
 462    unreachable("not reached");
 463 }
 464
 465 bool
 466 brw_saturate_immediate(enum brw_reg_type type, struct brw_reg *reg)
 467 {
 468    union {
 469       unsigned ud;
 470       int d;
 471       float f;
 472       double df;
 473    } imm, sat_imm = { 0 };
 474
 475    const unsigned size = type_sz(type);
 476
 477    /* We want to either do a 32-bit or 64-bit data copy, the type is otherwise
 478     * irrelevant, so just check the size of the type and copy from/to an
 479     * appropriately sized field.
 480     */
 481    if (size < 8)
 482       imm.ud = reg->ud;
 483    else
 484       imm.df = reg->df;
 485
 486    switch (type) {
 487    case BRW_REGISTER_TYPE_UD:
 488    case BRW_REGISTER_TYPE_D:
 489    case BRW_REGISTER_TYPE_UW:
 490    case BRW_REGISTER_TYPE_W:
 491    case BRW_REGISTER_TYPE_UQ:
 492    case BRW_REGISTER_TYPE_Q:
 493       /* Nothing to do. */
 494       return false;
 495    case BRW_REGISTER_TYPE_F:
 496       sat_imm.f = CLAMP(imm.f, 0.0f, 1.0f);
 497       break;
 498    case BRW_REGISTER_TYPE_DF:
 499       sat_imm.df = CLAMP(imm.df, 0.0, 1.0);
 500       break;
 501    case BRW_REGISTER_TYPE_UB:
 502    case BRW_REGISTER_TYPE_B:
 503       unreachable("no UB/B immediates");
 504    case BRW_REGISTER_TYPE_V:
 505    case BRW_REGISTER_TYPE_UV:
 506    case BRW_REGISTER_TYPE_VF:
 507       unreachable("unimplemented: saturate vector immediate");
 508    case BRW_REGISTER_TYPE_HF:
 509       unreachable("unimplemented: saturate HF immediate");
 510    }
 511
 512    if (size < 8) {
 513       if (imm.ud != sat_imm.ud) {
 514          reg->ud = sat_imm.ud;
 515          return true;
 516       }
 517    } else {
 518       if (imm.df != sat_imm.df) {
 519          reg->df = sat_imm.df;
 520          return true;
 521       }
 522    }
 523    return false;
 524 }
 525
 526 bool
 527 brw_negate_immediate(enum brw_reg_type type, struct brw_reg *reg)
 528 {
 529    switch (type) {
 530    case BRW_REGISTER_TYPE_D:
 531    case BRW_REGISTER_TYPE_UD:
 532       reg->d = -reg->d;
 533       return true;
 534    case BRW_REGISTER_TYPE_W:
 535    case BRW_REGISTER_TYPE_UW:
 536       reg->d = -(int16_t)reg->ud;
 537       return true;
 538    case BRW_REGISTER_TYPE_F:
 539       reg->f = -reg->f;
 540       return true;
 541    case BRW_REGISTER_TYPE_VF:
 542       reg->ud ^= 0x80808080;
 543       return true;
 544    case BRW_REGISTER_TYPE_DF:
 545       reg->df = -reg->df;
 546       return true;
 547    case BRW_REGISTER_TYPE_UB:
 548    case BRW_REGISTER_TYPE_B:
 549       unreachable("no UB/B immediates");
 550    case BRW_REGISTER_TYPE_UV:
 551    case BRW_REGISTER_TYPE_V:
 552       assert(!"unimplemented: negate UV/V immediate");
 553    case BRW_REGISTER_TYPE_UQ:
 554    case BRW_REGISTER_TYPE_Q:
 555       assert(!"unimplemented: negate UQ/Q immediate");
 556    case BRW_REGISTER_TYPE_HF:
 557       assert(!"unimplemented: negate HF immediate");
 558    }
 559
 560    return false;
 561 }
 562
 563 bool
 564 brw_abs_immediate(enum brw_reg_type type, struct brw_reg *reg)
 565 {
 566    switch (type) {
 567    case BRW_REGISTER_TYPE_D:
 568       reg->d = abs(reg->d);
 569       return true;
 570    case BRW_REGISTER_TYPE_W:
 571       reg->d = abs((int16_t)reg->ud);
 572       return true;
 573    case BRW_REGISTER_TYPE_F:
 574       reg->f = fabsf(reg->f);
 575       return true;
 576    case BRW_REGISTER_TYPE_DF:
 577       reg->df = fabs(reg->df);
 578       return true;
 579    case BRW_REGISTER_TYPE_VF:
 580       reg->ud &= ~0x80808080;
 581       return true;
 582    case BRW_REGISTER_TYPE_UB:
 583    case BRW_REGISTER_TYPE_B:
 584       unreachable("no UB/B immediates");
 585    case BRW_REGISTER_TYPE_UQ:
 586    case BRW_REGISTER_TYPE_UD:
 587    case BRW_REGISTER_TYPE_UW:
 588    case BRW_REGISTER_TYPE_UV:
 589       /* Presumably the absolute value modifier on an unsigned source is a
 590        * nop, but it would be nice to confirm.
 591        */
 592       assert(!"unimplemented: abs unsigned immediate");
 593    case BRW_REGISTER_TYPE_V:
 594       assert(!"unimplemented: abs V immediate");
 595    case BRW_REGISTER_TYPE_Q:
 596       assert(!"unimplemented: abs Q immediate");
 597    case BRW_REGISTER_TYPE_HF:
 598       assert(!"unimplemented: abs HF immediate");
 599    }
 600
 601    return false;
 602 }
 603
 604 unsigned
 605 tesslevel_outer_components(GLenum tes_primitive_mode)
 606 {
 607    switch (tes_primitive_mode) {
 608    case GL_QUADS:
 609       return 4;
 610    case GL_TRIANGLES:
 611       return 3;
 612    case GL_ISOLINES:
 613       return 2;
 614    default:
 615       unreachable("Bogus tessellation domain");
 616    }
 617    return 0;
 618 }
 619
 620 unsigned
 621 tesslevel_inner_components(GLenum tes_primitive_mode)
 622 {
 623    switch (tes_primitive_mode) {
 624    case GL_QUADS:
 625       return 2;
 626    case GL_TRIANGLES:
 627       return 1;
 628    case GL_ISOLINES:
 629       return 0;
 630    default:
 631       unreachable("Bogus tessellation domain");
 632    }
 633    return 0;
 634 }
 635
 636 /**
 637  * Given a normal .xyzw writemask, convert it to a writemask for a vector
 638  * that's stored backwards, i.e. .wzyx.
 639  */
 640 unsigned
 641 writemask_for_backwards_vector(unsigned mask)
 642 {
 643    unsigned new_mask = 0;
 644
 645    for (int i = 0; i < 4; i++)
 646       new_mask |= ((mask >> i) & 1) << (3 - i);
 647
 648    return new_mask;
 649 }
 650
 651 backend_shader::backend_shader(const struct brw_compiler *compiler,
 652                                void *log_data,
 653                                void *mem_ctx,
 654                                const nir_shader *shader,
 655                                struct brw_stage_prog_data *stage_prog_data)
 656    : compiler(compiler),
 657      log_data(log_data),
 658      devinfo(compiler->devinfo),
 659      nir(shader),
 660      stage_prog_data(stage_prog_data),
 661      mem_ctx(mem_ctx),
 662      cfg(NULL),
 663      stage(shader->stage)
 664 {
 665    debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
 666    stage_name = _mesa_shader_stage_to_string(stage);
 667    stage_abbrev = _mesa_shader_stage_to_abbrev(stage);
 668    is_passthrough_shader =
 669       nir->info.name && strcmp(nir->info.name, "passthrough") == 0;
 670 }
 671
 672 bool
 673 backend_reg::equals(const backend_reg &r) const
 674 {
 675    return brw_regs_equal(this, &r) && offset == r.offset;
 676 }
 677
 678 bool
 679 backend_reg::is_zero() const
 680 {
 681    if (file != IMM)
 682       return false;
 683
 684    switch (type) {
 685    case BRW_REGISTER_TYPE_F:
 686       return f == 0;
 687    case BRW_REGISTER_TYPE_DF:
 688       return df == 0;
 689    case BRW_REGISTER_TYPE_D:
 690    case BRW_REGISTER_TYPE_UD:
 691       return d == 0;
 692    default:
 693       return false;
 694    }
 695 }
 696
 697 bool
 698 backend_reg::is_one() const
 699 {
 700    if (file != IMM)
 701       return false;
 702
 703    switch (type) {
 704    case BRW_REGISTER_TYPE_F:
 705       return f == 1.0f;
 706    case BRW_REGISTER_TYPE_DF:
 707       return df == 1.0;
 708    case BRW_REGISTER_TYPE_D:
 709    case BRW_REGISTER_TYPE_UD:
 710       return d == 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    default:
 730       return false;
 731    }
 732 }
 733
 734 bool
 735 backend_reg::is_null() const
 736 {
 737    return file == ARF && nr == BRW_ARF_NULL;
 738 }
 739
 740
 741 bool
 742 backend_reg::is_accumulator() const
 743 {
 744    return file == ARF && nr == BRW_ARF_ACCUMULATOR;
 745 }
 746
 747 bool
 748 backend_instruction::is_commutative() const
 749 {
 750    switch (opcode) {
 751    case BRW_OPCODE_AND:
 752    case BRW_OPCODE_OR:
 753    case BRW_OPCODE_XOR:
 754    case BRW_OPCODE_ADD:
 755    case BRW_OPCODE_MUL:
 756    case SHADER_OPCODE_MULH:
 757       return true;
 758    case BRW_OPCODE_SEL:
 759       /* MIN and MAX are commutative. */
 760       if (conditional_mod == BRW_CONDITIONAL_GE ||
 761           conditional_mod == BRW_CONDITIONAL_L) {
 762          return true;
 763       }
 764       /* fallthrough */
 765    default:
 766       return false;
 767    }
 768 }
 769
 770 bool
 771 backend_instruction::is_3src(const struct gen_device_info *devinfo) const
 772 {
 773    return ::is_3src(devinfo, opcode);
 774 }
 775
 776 bool
 777 backend_instruction::is_tex() const
 778 {
 779    return (opcode == SHADER_OPCODE_TEX ||
 780            opcode == FS_OPCODE_TXB ||
 781            opcode == SHADER_OPCODE_TXD ||
 782            opcode == SHADER_OPCODE_TXF ||
 783            opcode == SHADER_OPCODE_TXF_LZ ||
 784            opcode == SHADER_OPCODE_TXF_CMS ||
 785            opcode == SHADER_OPCODE_TXF_CMS_W ||
 786            opcode == SHADER_OPCODE_TXF_UMS ||
 787            opcode == SHADER_OPCODE_TXF_MCS ||
 788            opcode == SHADER_OPCODE_TXL ||
 789            opcode == SHADER_OPCODE_TXL_LZ ||
 790            opcode == SHADER_OPCODE_TXS ||
 791            opcode == SHADER_OPCODE_LOD ||
 792            opcode == SHADER_OPCODE_TG4 ||
 793            opcode == SHADER_OPCODE_TG4_OFFSET ||
 794            opcode == SHADER_OPCODE_SAMPLEINFO);
 795 }
 796
 797 bool
 798 backend_instruction::is_math() const
 799 {
 800    return (opcode == SHADER_OPCODE_RCP ||
 801            opcode == SHADER_OPCODE_RSQ ||
 802            opcode == SHADER_OPCODE_SQRT ||
 803            opcode == SHADER_OPCODE_EXP2 ||
 804            opcode == SHADER_OPCODE_LOG2 ||
 805            opcode == SHADER_OPCODE_SIN ||
 806            opcode == SHADER_OPCODE_COS ||
 807            opcode == SHADER_OPCODE_INT_QUOTIENT ||
 808            opcode == SHADER_OPCODE_INT_REMAINDER ||
 809            opcode == SHADER_OPCODE_POW);
 810 }
 811
 812 bool
 813 backend_instruction::is_control_flow() const
 814 {
 815    switch (opcode) {
 816    case BRW_OPCODE_DO:
 817    case BRW_OPCODE_WHILE:
 818    case BRW_OPCODE_IF:
 819    case BRW_OPCODE_ELSE:
 820    case BRW_OPCODE_ENDIF:
 821    case BRW_OPCODE_BREAK:
 822    case BRW_OPCODE_CONTINUE:
 823       return true;
 824    default:
 825       return false;
 826    }
 827 }
 828
 829 bool
 830 backend_instruction::can_do_source_mods() const
 831 {
 832    switch (opcode) {
 833    case BRW_OPCODE_ADDC:
 834    case BRW_OPCODE_BFE:
 835    case BRW_OPCODE_BFI1:
 836    case BRW_OPCODE_BFI2:
 837    case BRW_OPCODE_BFREV:
 838    case BRW_OPCODE_CBIT:
 839    case BRW_OPCODE_FBH:
 840    case BRW_OPCODE_FBL:
 841    case BRW_OPCODE_SUBB:
 842       return false;
 843    default:
 844       return true;
 845    }
 846 }
 847
 848 bool
 849 backend_instruction::can_do_saturate() const
 850 {
 851    switch (opcode) {
 852    case BRW_OPCODE_ADD:
 853    case BRW_OPCODE_ASR:
 854    case BRW_OPCODE_AVG:
 855    case BRW_OPCODE_DP2:
 856    case BRW_OPCODE_DP3:
 857    case BRW_OPCODE_DP4:
 858    case BRW_OPCODE_DPH:
 859    case BRW_OPCODE_F16TO32:
 860    case BRW_OPCODE_F32TO16:
 861    case BRW_OPCODE_LINE:
 862    case BRW_OPCODE_LRP:
 863    case BRW_OPCODE_MAC:
 864    case BRW_OPCODE_MAD:
 865    case BRW_OPCODE_MATH:
 866    case BRW_OPCODE_MOV:
 867    case BRW_OPCODE_MUL:
 868    case SHADER_OPCODE_MULH:
 869    case BRW_OPCODE_PLN:
 870    case BRW_OPCODE_RNDD:
 871    case BRW_OPCODE_RNDE:
 872    case BRW_OPCODE_RNDU:
 873    case BRW_OPCODE_RNDZ:
 874    case BRW_OPCODE_SEL:
 875    case BRW_OPCODE_SHL:
 876    case BRW_OPCODE_SHR:
 877    case FS_OPCODE_LINTERP:
 878    case SHADER_OPCODE_COS:
 879    case SHADER_OPCODE_EXP2:
 880    case SHADER_OPCODE_LOG2:
 881    case SHADER_OPCODE_POW:
 882    case SHADER_OPCODE_RCP:
 883    case SHADER_OPCODE_RSQ:
 884    case SHADER_OPCODE_SIN:
 885    case SHADER_OPCODE_SQRT:
 886       return true;
 887    default:
 888       return false;
 889    }
 890 }
 891
 892 bool
 893 backend_instruction::can_do_cmod() const
 894 {
 895    switch (opcode) {
 896    case BRW_OPCODE_ADD:
 897    case BRW_OPCODE_ADDC:
 898    case BRW_OPCODE_AND:
 899    case BRW_OPCODE_ASR:
 900    case BRW_OPCODE_AVG:
 901    case BRW_OPCODE_CMP:
 902    case BRW_OPCODE_CMPN:
 903    case BRW_OPCODE_DP2:
 904    case BRW_OPCODE_DP3:
 905    case BRW_OPCODE_DP4:
 906    case BRW_OPCODE_DPH:
 907    case BRW_OPCODE_F16TO32:
 908    case BRW_OPCODE_F32TO16:
 909    case BRW_OPCODE_FRC:
 910    case BRW_OPCODE_LINE:
 911    case BRW_OPCODE_LRP:
 912    case BRW_OPCODE_LZD:
 913    case BRW_OPCODE_MAC:
 914    case BRW_OPCODE_MACH:
 915    case BRW_OPCODE_MAD:
 916    case BRW_OPCODE_MOV:
 917    case BRW_OPCODE_MUL:
 918    case BRW_OPCODE_NOT:
 919    case BRW_OPCODE_OR:
 920    case BRW_OPCODE_PLN:
 921    case BRW_OPCODE_RNDD:
 922    case BRW_OPCODE_RNDE:
 923    case BRW_OPCODE_RNDU:
 924    case BRW_OPCODE_RNDZ:
 925    case BRW_OPCODE_SAD2:
 926    case BRW_OPCODE_SADA2:
 927    case BRW_OPCODE_SHL:
 928    case BRW_OPCODE_SHR:
 929    case BRW_OPCODE_SUBB:
 930    case BRW_OPCODE_XOR:
 931    case FS_OPCODE_CINTERP:
 932    case FS_OPCODE_LINTERP:
 933       return true;
 934    default:
 935       return false;
 936    }
 937 }
 938
 939 bool
 940 backend_instruction::reads_accumulator_implicitly() const
 941 {
 942    switch (opcode) {
 943    case BRW_OPCODE_MAC:
 944    case BRW_OPCODE_MACH:
 945    case BRW_OPCODE_SADA2:
 946       return true;
 947    default:
 948       return false;
 949    }
 950 }
 951
 952 bool
 953 backend_instruction::writes_accumulator_implicitly(const struct gen_device_info *devinfo) const
 954 {
 955    return writes_accumulator ||
 956           (devinfo->gen < 6 &&
 957            ((opcode >= BRW_OPCODE_ADD && opcode < BRW_OPCODE_NOP) ||
 958             (opcode >= FS_OPCODE_DDX_COARSE && opcode <= FS_OPCODE_LINTERP &&
 959              opcode != FS_OPCODE_CINTERP)));
 960 }
 961
 962 bool
 963 backend_instruction::has_side_effects() const
 964 {
 965    switch (opcode) {
 966    case SHADER_OPCODE_UNTYPED_ATOMIC:
 967    case SHADER_OPCODE_UNTYPED_ATOMIC_LOGICAL:
 968    case SHADER_OPCODE_GEN4_SCRATCH_WRITE:
 969    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE:
 970    case SHADER_OPCODE_UNTYPED_SURFACE_WRITE_LOGICAL:
 971    case SHADER_OPCODE_TYPED_ATOMIC:
 972    case SHADER_OPCODE_TYPED_ATOMIC_LOGICAL:
 973    case SHADER_OPCODE_TYPED_SURFACE_WRITE:
 974    case SHADER_OPCODE_TYPED_SURFACE_WRITE_LOGICAL:
 975    case SHADER_OPCODE_MEMORY_FENCE:
 976    case SHADER_OPCODE_URB_WRITE_SIMD8:
 977    case SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT:
 978    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED:
 979    case SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT:
 980    case FS_OPCODE_FB_WRITE:
 981    case FS_OPCODE_FB_WRITE_LOGICAL:
 982    case SHADER_OPCODE_BARRIER:
 983    case TCS_OPCODE_URB_WRITE:
 984    case TCS_OPCODE_RELEASE_INPUT:
 985       return true;
 986    default:
 987       return false;
 988    }
 989 }
 990
 991 bool
 992 backend_instruction::is_volatile() const
 993 {
 994    switch (opcode) {
 995    case SHADER_OPCODE_UNTYPED_SURFACE_READ:
 996    case SHADER_OPCODE_UNTYPED_SURFACE_READ_LOGICAL:
 997    case SHADER_OPCODE_TYPED_SURFACE_READ:
 998    case SHADER_OPCODE_TYPED_SURFACE_READ_LOGICAL:
 999    case SHADER_OPCODE_URB_READ_SIMD8:
1000    case SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT:
1001    case VEC4_OPCODE_URB_READ:
1002       return true;
1003    default:
1004       return false;
1005    }
1006 }
1007
1008 #ifndef NDEBUG
1009 static bool
1010 inst_is_in_block(const bblock_t *block, const backend_instruction *inst)
1011 {
1012    bool found = false;
1013    foreach_inst_in_block (backend_instruction, i, block) {
1014       if (inst == i) {
1015          found = true;
1016       }
1017    }
1018    return found;
1019 }
1020 #endif
1021
1022 static void
1023 adjust_later_block_ips(bblock_t *start_block, int ip_adjustment)
1024 {
1025    for (bblock_t *block_iter = start_block->next();
1026         block_iter;
1027         block_iter = block_iter->next()) {
1028       block_iter->start_ip += ip_adjustment;
1029       block_iter->end_ip += ip_adjustment;
1030    }
1031 }
1032
1033 void
1034 backend_instruction::insert_after(bblock_t *block, backend_instruction *inst)
1035 {
1036    assert(this != inst);
1037
1038    if (!this->is_head_sentinel())
1039       assert(inst_is_in_block(block, this) || !"Instruction not in block");
1040
1041    block->end_ip++;
1042
1043    adjust_later_block_ips(block, 1);
1044
1045    exec_node::insert_after(inst);
1046 }
1047
1048 void
1049 backend_instruction::insert_before(bblock_t *block, backend_instruction *inst)
1050 {
1051    assert(this != inst);
1052
1053    if (!this->is_tail_sentinel())
1054       assert(inst_is_in_block(block, this) || !"Instruction not in block");
1055
1056    block->end_ip++;
1057
1058    adjust_later_block_ips(block, 1);
1059
1060    exec_node::insert_before(inst);
1061 }
1062
1063 void
1064 backend_instruction::insert_before(bblock_t *block, exec_list *list)
1065 {
1066    assert(inst_is_in_block(block, this) || !"Instruction not in block");
1067
1068    unsigned num_inst = list->length();
1069
1070    block->end_ip += num_inst;
1071
1072    adjust_later_block_ips(block, num_inst);
1073
1074    exec_node::insert_before(list);
1075 }
1076
1077 void
1078 backend_instruction::remove(bblock_t *block)
1079 {
1080    assert(inst_is_in_block(block, this) || !"Instruction not in block");
1081
1082    adjust_later_block_ips(block, -1);
1083
1084    if (block->start_ip == block->end_ip) {
1085       block->cfg->remove_block(block);
1086    } else {
1087       block->end_ip--;
1088    }
1089
1090    exec_node::remove();
1091 }
1092
1093 void
1094 backend_shader::dump_instructions()
1095 {
1096    dump_instructions(NULL);
1097 }
1098
1099 void
1100 backend_shader::dump_instructions(const char *name)
1101 {
1102    FILE *file = stderr;
1103    if (name && geteuid() != 0) {
1104       file = fopen(name, "w");
1105       if (!file)
1106          file = stderr;
1107    }
1108
1109    if (cfg) {
1110       int ip = 0;
1111       foreach_block_and_inst(block, backend_instruction, inst, cfg) {
1112          if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1113             fprintf(file, "%4d: ", ip++);
1114          dump_instruction(inst, file);
1115       }
1116    } else {
1117       int ip = 0;
1118       foreach_in_list(backend_instruction, inst, &instructions) {
1119          if (!unlikely(INTEL_DEBUG & DEBUG_OPTIMIZER))
1120             fprintf(file, "%4d: ", ip++);
1121          dump_instruction(inst, file);
1122       }
1123    }
1124
1125    if (file != stderr) {
1126       fclose(file);
1127    }
1128 }
1129
1130 void
1131 backend_shader::calculate_cfg()
1132 {
1133    if (this->cfg)
1134       return;
1135    cfg = new(mem_ctx) cfg_t(&this->instructions);
1136 }
1137
1138 /**
1139  * Sets up the starting offsets for the groups of binding table entries
1140  * commong to all pipeline stages.
1141  *
1142  * Unused groups are initialized to 0xd0d0d0d0 to make it obvious that they're
1143  * unused but also make sure that addition of small offsets to them will
1144  * trigger some of our asserts that surface indices are < BRW_MAX_SURFACES.
1145  */
1146 uint32_t
1147 brw_assign_common_binding_table_offsets(gl_shader_stage stage,
1148                                         const struct gen_device_info *devinfo,
1149                                         const struct gl_shader_program *shader_prog,
1150                                         const struct gl_program *prog,
1151                                         struct brw_stage_prog_data *stage_prog_data,
1152                                         uint32_t next_binding_table_offset)
1153 {
1154    const struct gl_linked_shader *shader = NULL;
1155    int num_textures = util_last_bit(prog->SamplersUsed);
1156
1157    if (shader_prog)
1158       shader = shader_prog->_LinkedShaders[stage];
1159
1160    stage_prog_data->binding_table.texture_start = next_binding_table_offset;
1161    next_binding_table_offset += num_textures;
1162
1163    if (shader) {
1164       assert(shader->NumUniformBlocks <= BRW_MAX_UBO);
1165       stage_prog_data->binding_table.ubo_start = next_binding_table_offset;
1166       next_binding_table_offset += shader->NumUniformBlocks;
1167
1168       assert(shader->NumShaderStorageBlocks <= BRW_MAX_SSBO);
1169       stage_prog_data->binding_table.ssbo_start = next_binding_table_offset;
1170       next_binding_table_offset += shader->NumShaderStorageBlocks;
1171    } else {
1172       stage_prog_data->binding_table.ubo_start = 0xd0d0d0d0;
1173       stage_prog_data->binding_table.ssbo_start = 0xd0d0d0d0;
1174    }
1175
1176    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
1177       stage_prog_data->binding_table.shader_time_start = next_binding_table_offset;
1178       next_binding_table_offset++;
1179    } else {
1180       stage_prog_data->binding_table.shader_time_start = 0xd0d0d0d0;
1181    }
1182
1183    if (prog->UsesGather) {
1184       if (devinfo->gen >= 8) {
1185          stage_prog_data->binding_table.gather_texture_start =
1186             stage_prog_data->binding_table.texture_start;
1187       } else {
1188          stage_prog_data->binding_table.gather_texture_start = next_binding_table_offset;
1189          next_binding_table_offset += num_textures;
1190       }
1191    } else {
1192       stage_prog_data->binding_table.gather_texture_start = 0xd0d0d0d0;
1193    }
1194
1195    if (shader && shader->NumAtomicBuffers) {
1196       stage_prog_data->binding_table.abo_start = next_binding_table_offset;
1197       next_binding_table_offset += shader->NumAtomicBuffers;
1198    } else {
1199       stage_prog_data->binding_table.abo_start = 0xd0d0d0d0;
1200    }
1201
1202    if (shader && shader->NumImages) {
1203       stage_prog_data->binding_table.image_start = next_binding_table_offset;
1204       next_binding_table_offset += shader->NumImages;
1205    } else {
1206       stage_prog_data->binding_table.image_start = 0xd0d0d0d0;
1207    }
1208
1209    /* This may or may not be used depending on how the compile goes. */
1210    stage_prog_data->binding_table.pull_constants_start = next_binding_table_offset;
1211    next_binding_table_offset++;
1212
1213    /* Plane 0 is just the regular texture section */
1214    stage_prog_data->binding_table.plane_start[0] = stage_prog_data->binding_table.texture_start;
1215
1216    stage_prog_data->binding_table.plane_start[1] = next_binding_table_offset;
1217    next_binding_table_offset += num_textures;
1218
1219    stage_prog_data->binding_table.plane_start[2] = next_binding_table_offset;
1220    next_binding_table_offset += num_textures;
1221
1222    /* prog_data->base.binding_table.size will be set by brw_mark_surface_used. */
1223
1224    assert(next_binding_table_offset <= BRW_MAX_SURFACES);
1225    return next_binding_table_offset;
1226 }
1227
1228 static void
1229 setup_vec4_uniform_value(const gl_constant_value **params,
1230                          const gl_constant_value *values,
1231                          unsigned n)
1232 {
1233    static const gl_constant_value zero = { 0 };
1234
1235    for (unsigned i = 0; i < n; ++i)
1236       params[i] = &values[i];
1237
1238    for (unsigned i = n; i < 4; ++i)
1239       params[i] = &zero;
1240 }
1241
1242 void
1243 brw_setup_image_uniform_values(gl_shader_stage stage,
1244                                struct brw_stage_prog_data *stage_prog_data,
1245                                unsigned param_start_index,
1246                                const gl_uniform_storage *storage)
1247 {
1248    const gl_constant_value **param =
1249       &stage_prog_data->param[param_start_index];
1250
1251    for (unsigned i = 0; i < MAX2(storage->array_elements, 1); i++) {
1252       const unsigned image_idx = storage->opaque[stage].index + i;
1253       const brw_image_param *image_param =
1254          &stage_prog_data->image_param[image_idx];
1255
1256       /* Upload the brw_image_param structure.  The order is expected to match
1257        * the BRW_IMAGE_PARAM_*_OFFSET defines.
1258        */
1259       setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_SURFACE_IDX_OFFSET,
1260          (const gl_constant_value *)&image_param->surface_idx, 1);
1261       setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_OFFSET_OFFSET,
1262          (const gl_constant_value *)image_param->offset, 2);
1263       setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_SIZE_OFFSET,
1264          (const gl_constant_value *)image_param->size, 3);
1265       setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_STRIDE_OFFSET,
1266          (const gl_constant_value *)image_param->stride, 4);
1267       setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_TILING_OFFSET,
1268          (const gl_constant_value *)image_param->tiling, 3);
1269       setup_vec4_uniform_value(param + BRW_IMAGE_PARAM_SWIZZLING_OFFSET,
1270          (const gl_constant_value *)image_param->swizzling, 2);
1271       param += BRW_IMAGE_PARAM_SIZE;
1272
1273       brw_mark_surface_used(
1274          stage_prog_data,
1275          stage_prog_data->binding_table.image_start + image_idx);
1276    }
1277 }
1278
1279 /**
1280  * Decide which set of clip planes should be used when clipping via
1281  * gl_Position or gl_ClipVertex.
1282  */
1283 gl_clip_plane *brw_select_clip_planes(struct gl_context *ctx)
1284 {
1285    if (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]) {
1286       /* There is currently a GLSL vertex shader, so clip according to GLSL
1287        * rules, which means compare gl_ClipVertex (or gl_Position, if
1288        * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
1289        * that were stored in EyeUserPlane at the time the clip planes were
1290        * specified.
1291        */
1292       return ctx->Transform.EyeUserPlane;
1293    } else {
1294       /* Either we are using fixed function or an ARB vertex program.  In
1295        * either case the clip planes are going to be compared against
1296        * gl_Position (which is in clip coordinates) so we have to clip using
1297        * _ClipUserPlane, which was transformed into clip coordinates by Mesa
1298        * core.
1299        */
1300       return ctx->Transform._ClipUserPlane;
1301    }
1302 }
1303
1304 extern "C" const unsigned *
1305 brw_compile_tes(const struct brw_compiler *compiler,
1306                 void *log_data,
1307                 void *mem_ctx,
1308                 const struct brw_tes_prog_key *key,
1309                 struct brw_tes_prog_data *prog_data,
1310                 const nir_shader *src_shader,
1311                 struct gl_shader_program *shader_prog,
1312                 int shader_time_index,
1313                 unsigned *final_assembly_size,
1314                 char **error_str)
1315 {
1316    const struct gen_device_info *devinfo = compiler->devinfo;
1317    struct gl_linked_shader *shader =
1318       shader_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
1319    const bool is_scalar = compiler->scalar_stage[MESA_SHADER_TESS_EVAL];
1320
1321    nir_shader *nir = nir_shader_clone(mem_ctx, src_shader);
1322    nir->info.inputs_read = key->inputs_read;
1323    nir->info.patch_inputs_read = key->patch_inputs_read;
1324
1325    struct brw_vue_map input_vue_map;
1326    brw_compute_tess_vue_map(&input_vue_map,
1327                             nir->info.inputs_read & ~VARYING_BIT_PRIMITIVE_ID,
1328                             nir->info.patch_inputs_read);
1329
1330    nir = brw_nir_apply_sampler_key(nir, devinfo, &key->tex, is_scalar);
1331    brw_nir_lower_tes_inputs(nir, &input_vue_map);
1332    brw_nir_lower_vue_outputs(nir, is_scalar);
1333    nir = brw_postprocess_nir(nir, compiler->devinfo, is_scalar);
1334
1335    brw_compute_vue_map(devinfo, &prog_data->base.vue_map,
1336                        nir->info.outputs_written,
1337                        nir->info.separate_shader);
1338
1339    unsigned output_size_bytes = prog_data->base.vue_map.num_slots * 4 * 4;
1340
1341    assert(output_size_bytes >= 1);
1342    if (output_size_bytes > GEN7_MAX_DS_URB_ENTRY_SIZE_BYTES) {
1343       if (error_str)
1344          *error_str = ralloc_strdup(mem_ctx, "DS outputs exceed maximum size");
1345       return NULL;
1346    }
1347
1348    /* URB entry sizes are stored as a multiple of 64 bytes. */
1349    prog_data->base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64;
1350
1351    bool need_patch_header = nir->info.system_values_read &
1352       (BITFIELD64_BIT(SYSTEM_VALUE_TESS_LEVEL_OUTER) |
1353        BITFIELD64_BIT(SYSTEM_VALUE_TESS_LEVEL_INNER));
1354
1355    /* The TES will pull most inputs using URB read messages.
1356     *
1357     * However, we push the patch header for TessLevel factors when required,
1358     * as it's a tiny amount of extra data.
1359     */
1360    prog_data->base.urb_read_length = need_patch_header ? 1 : 0;
1361
1362    if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1363       fprintf(stderr, "TES Input ");
1364       brw_print_vue_map(stderr, &input_vue_map);
1365       fprintf(stderr, "TES Output ");
1366       brw_print_vue_map(stderr, &prog_data->base.vue_map);
1367    }
1368
1369    if (is_scalar) {
1370       fs_visitor v(compiler, log_data, mem_ctx, (void *) key,
1371                    &prog_data->base.base, shader->Program, nir, 8,
1372                    shader_time_index, &input_vue_map);
1373       if (!v.run_tes()) {
1374          if (error_str)
1375             *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1376          return NULL;
1377       }
1378
1379       prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
1380       prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8;
1381
1382       fs_generator g(compiler, log_data, mem_ctx, (void *) key,
1383                      &prog_data->base.base, v.promoted_constants, false,
1384                      MESA_SHADER_TESS_EVAL);
1385       if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
1386          g.enable_debug(ralloc_asprintf(mem_ctx,
1387                                         "%s tessellation evaluation shader %s",
1388                                         nir->info.label ? nir->info.label
1389                                                         : "unnamed",
1390                                         nir->info.name));
1391       }
1392
1393       g.generate_code(v.cfg, 8);
1394
1395       return g.get_assembly(final_assembly_size);
1396    } else {
1397       brw::vec4_tes_visitor v(compiler, log_data, key, prog_data,
1398                               nir, mem_ctx, shader_time_index);
1399       if (!v.run()) {
1400          if (error_str)
1401             *error_str = ralloc_strdup(mem_ctx, v.fail_msg);
1402          return NULL;
1403       }
1404
1405       if (unlikely(INTEL_DEBUG & DEBUG_TES))
1406          v.dump_instructions();
1407
1408       return brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir,
1409                                         &prog_data->base, v.cfg,
1410                                         final_assembly_size);
1411    }
1412 }