src/intel/vulkan/gen8_pipeline.c

   1 /*
   2  * Copyright © 2015 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 <assert.h>
  25 #include <stdbool.h>
  26 #include <string.h>
  27 #include <unistd.h>
  28 #include <fcntl.h>
  29
  30 #include "anv_private.h"
  31
  32 #include "genxml/gen_macros.h"
  33 #include "genxml/genX_pack.h"
  34
  35 #include "genX_pipeline_util.h"
  36
  37 static void
  38 emit_ia_state(struct anv_pipeline *pipeline,
  39               const VkPipelineInputAssemblyStateCreateInfo *info,
  40               const struct anv_graphics_pipeline_create_info *extra)
  41 {
  42    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VF_TOPOLOGY),
  43                   .PrimitiveTopologyType = pipeline->topology);
  44 }
  45
  46 static void
  47 emit_rs_state(struct anv_pipeline *pipeline,
  48               const VkPipelineRasterizationStateCreateInfo *info,
  49               const VkPipelineMultisampleStateCreateInfo *ms_info,
  50               const struct anv_graphics_pipeline_create_info *extra)
  51 {
  52    uint32_t samples = 1;
  53
  54    if (ms_info)
  55       samples = ms_info->rasterizationSamples;
  56
  57    struct GENX(3DSTATE_SF) sf = {
  58       GENX(3DSTATE_SF_header),
  59       .ViewportTransformEnable = !(extra && extra->disable_viewport),
  60       .TriangleStripListProvokingVertexSelect = 0,
  61       .LineStripListProvokingVertexSelect = 0,
  62       .TriangleFanProvokingVertexSelect = 1,
  63       .PointWidthSource = Vertex,
  64       .PointWidth = 1.0,
  65    };
  66
  67    /* FINISHME: VkBool32 rasterizerDiscardEnable; */
  68
  69    GENX(3DSTATE_SF_pack)(NULL, pipeline->gen8.sf, &sf);
  70
  71    struct GENX(3DSTATE_RASTER) raster = {
  72       GENX(3DSTATE_RASTER_header),
  73
  74       /* For details on 3DSTATE_RASTER multisample state, see the BSpec table
  75        * "Multisample Modes State".
  76        */
  77       .DXMultisampleRasterizationEnable = samples > 1,
  78       .ForcedSampleCount = FSC_NUMRASTSAMPLES_0,
  79       .ForceMultisampling = false,
  80
  81       .FrontWinding = vk_to_gen_front_face[info->frontFace],
  82       .CullMode = vk_to_gen_cullmode[info->cullMode],
  83       .FrontFaceFillMode = vk_to_gen_fillmode[info->polygonMode],
  84       .BackFaceFillMode = vk_to_gen_fillmode[info->polygonMode],
  85       .ScissorRectangleEnable = !(extra && extra->disable_scissor),
  86 #if GEN_GEN == 8
  87       .ViewportZClipTestEnable = true,
  88 #else
  89       /* GEN9+ splits ViewportZClipTestEnable into near and far enable bits */
  90       .ViewportZFarClipTestEnable = true,
  91       .ViewportZNearClipTestEnable = true,
  92 #endif
  93       .GlobalDepthOffsetEnableSolid = info->depthBiasEnable,
  94       .GlobalDepthOffsetEnableWireframe = info->depthBiasEnable,
  95       .GlobalDepthOffsetEnablePoint = info->depthBiasEnable,
  96    };
  97
  98    GENX(3DSTATE_RASTER_pack)(NULL, pipeline->gen8.raster, &raster);
  99 }
 100
 101 static void
 102 emit_cb_state(struct anv_pipeline *pipeline,
 103               const VkPipelineColorBlendStateCreateInfo *info,
 104               const VkPipelineMultisampleStateCreateInfo *ms_info)
 105 {
 106    struct anv_device *device = pipeline->device;
 107
 108    uint32_t num_dwords = GENX(BLEND_STATE_length);
 109    pipeline->blend_state =
 110       anv_state_pool_alloc(&device->dynamic_state_pool, num_dwords * 4, 64);
 111
 112    struct GENX(BLEND_STATE) blend_state = {
 113       .AlphaToCoverageEnable = ms_info && ms_info->alphaToCoverageEnable,
 114       .AlphaToOneEnable = ms_info && ms_info->alphaToOneEnable,
 115    };
 116
 117    bool has_writeable_rt = false;
 118    for (uint32_t i = 0; i < info->attachmentCount; i++) {
 119       const VkPipelineColorBlendAttachmentState *a = &info->pAttachments[i];
 120
 121       if (a->srcColorBlendFactor != a->srcAlphaBlendFactor ||
 122           a->dstColorBlendFactor != a->dstAlphaBlendFactor ||
 123           a->colorBlendOp != a->alphaBlendOp) {
 124          blend_state.IndependentAlphaBlendEnable = true;
 125       }
 126
 127       blend_state.Entry[i] = (struct GENX(BLEND_STATE_ENTRY)) {
 128          .LogicOpEnable = info->logicOpEnable,
 129          .LogicOpFunction = vk_to_gen_logic_op[info->logicOp],
 130          .ColorBufferBlendEnable = a->blendEnable,
 131          .PreBlendSourceOnlyClampEnable = false,
 132          .ColorClampRange = COLORCLAMP_RTFORMAT,
 133          .PreBlendColorClampEnable = true,
 134          .PostBlendColorClampEnable = true,
 135          .SourceBlendFactor = vk_to_gen_blend[a->srcColorBlendFactor],
 136          .DestinationBlendFactor = vk_to_gen_blend[a->dstColorBlendFactor],
 137          .ColorBlendFunction = vk_to_gen_blend_op[a->colorBlendOp],
 138          .SourceAlphaBlendFactor = vk_to_gen_blend[a->srcAlphaBlendFactor],
 139          .DestinationAlphaBlendFactor = vk_to_gen_blend[a->dstAlphaBlendFactor],
 140          .AlphaBlendFunction = vk_to_gen_blend_op[a->alphaBlendOp],
 141          .WriteDisableAlpha = !(a->colorWriteMask & VK_COLOR_COMPONENT_A_BIT),
 142          .WriteDisableRed = !(a->colorWriteMask & VK_COLOR_COMPONENT_R_BIT),
 143          .WriteDisableGreen = !(a->colorWriteMask & VK_COLOR_COMPONENT_G_BIT),
 144          .WriteDisableBlue = !(a->colorWriteMask & VK_COLOR_COMPONENT_B_BIT),
 145       };
 146
 147       if (a->colorWriteMask != 0)
 148          has_writeable_rt = true;
 149
 150       /* Our hardware applies the blend factor prior to the blend function
 151        * regardless of what function is used.  Technically, this means the
 152        * hardware can do MORE than GL or Vulkan specify.  However, it also
 153        * means that, for MIN and MAX, we have to stomp the blend factor to
 154        * ONE to make it a no-op.
 155        */
 156       if (a->colorBlendOp == VK_BLEND_OP_MIN ||
 157           a->colorBlendOp == VK_BLEND_OP_MAX) {
 158          blend_state.Entry[i].SourceBlendFactor = BLENDFACTOR_ONE;
 159          blend_state.Entry[i].DestinationBlendFactor = BLENDFACTOR_ONE;
 160       }
 161       if (a->alphaBlendOp == VK_BLEND_OP_MIN ||
 162           a->alphaBlendOp == VK_BLEND_OP_MAX) {
 163          blend_state.Entry[i].SourceAlphaBlendFactor = BLENDFACTOR_ONE;
 164          blend_state.Entry[i].DestinationAlphaBlendFactor = BLENDFACTOR_ONE;
 165       }
 166    }
 167
 168    for (uint32_t i = info->attachmentCount; i < 8; i++) {
 169       blend_state.Entry[i].WriteDisableAlpha = true;
 170       blend_state.Entry[i].WriteDisableRed = true;
 171       blend_state.Entry[i].WriteDisableGreen = true;
 172       blend_state.Entry[i].WriteDisableBlue = true;
 173    }
 174
 175    if (info->attachmentCount > 0) {
 176       struct GENX(BLEND_STATE_ENTRY) *bs = &blend_state.Entry[0];
 177
 178       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS_BLEND),
 179                      .AlphaToCoverageEnable = blend_state.AlphaToCoverageEnable,
 180                      .HasWriteableRT = has_writeable_rt,
 181                      .ColorBufferBlendEnable = bs->ColorBufferBlendEnable,
 182                      .SourceAlphaBlendFactor = bs->SourceAlphaBlendFactor,
 183                      .DestinationAlphaBlendFactor =
 184                         bs->DestinationAlphaBlendFactor,
 185                      .SourceBlendFactor = bs->SourceBlendFactor,
 186                      .DestinationBlendFactor = bs->DestinationBlendFactor,
 187                      .AlphaTestEnable = false,
 188                      .IndependentAlphaBlendEnable =
 189                         blend_state.IndependentAlphaBlendEnable);
 190    } else {
 191       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS_BLEND));
 192    }
 193
 194    GENX(BLEND_STATE_pack)(NULL, pipeline->blend_state.map, &blend_state);
 195    if (!device->info.has_llc)
 196       anv_state_clflush(pipeline->blend_state);
 197
 198    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_BLEND_STATE_POINTERS),
 199                   .BlendStatePointer = pipeline->blend_state.offset,
 200                   .BlendStatePointerValid = true);
 201 }
 202
 203 static void
 204 emit_ds_state(struct anv_pipeline *pipeline,
 205               const VkPipelineDepthStencilStateCreateInfo *info)
 206 {
 207    uint32_t *dw = GEN_GEN == 8 ?
 208       pipeline->gen8.wm_depth_stencil : pipeline->gen9.wm_depth_stencil;
 209
 210    if (info == NULL) {
 211       /* We're going to OR this together with the dynamic state.  We need
 212        * to make sure it's initialized to something useful.
 213        */
 214       memset(pipeline->gen8.wm_depth_stencil, 0,
 215              sizeof(pipeline->gen8.wm_depth_stencil));
 216       memset(pipeline->gen9.wm_depth_stencil, 0,
 217              sizeof(pipeline->gen9.wm_depth_stencil));
 218       return;
 219    }
 220
 221    /* VkBool32 depthBoundsTestEnable; // optional (depth_bounds_test) */
 222
 223    struct GENX(3DSTATE_WM_DEPTH_STENCIL) wm_depth_stencil = {
 224       .DepthTestEnable = info->depthTestEnable,
 225       .DepthBufferWriteEnable = info->depthWriteEnable,
 226       .DepthTestFunction = vk_to_gen_compare_op[info->depthCompareOp],
 227       .DoubleSidedStencilEnable = true,
 228
 229       .StencilTestEnable = info->stencilTestEnable,
 230       .StencilBufferWriteEnable = info->stencilTestEnable,
 231       .StencilFailOp = vk_to_gen_stencil_op[info->front.failOp],
 232       .StencilPassDepthPassOp = vk_to_gen_stencil_op[info->front.passOp],
 233       .StencilPassDepthFailOp = vk_to_gen_stencil_op[info->front.depthFailOp],
 234       .StencilTestFunction = vk_to_gen_compare_op[info->front.compareOp],
 235       .BackfaceStencilFailOp = vk_to_gen_stencil_op[info->back.failOp],
 236       .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[info->back.passOp],
 237       .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[info->back.depthFailOp],
 238       .BackfaceStencilTestFunction = vk_to_gen_compare_op[info->back.compareOp],
 239    };
 240
 241    /* From the Broadwell PRM:
 242     *
 243     *    "If Depth_Test_Enable = 1 AND Depth_Test_func = EQUAL, the
 244     *    Depth_Write_Enable must be set to 0."
 245     */
 246    if (info->depthTestEnable && info->depthCompareOp == VK_COMPARE_OP_EQUAL)
 247       wm_depth_stencil.DepthBufferWriteEnable = false;
 248
 249    GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, dw, &wm_depth_stencil);
 250 }
 251
 252 static void
 253 emit_ms_state(struct anv_pipeline *pipeline,
 254               const VkPipelineMultisampleStateCreateInfo *info)
 255 {
 256    uint32_t samples = 1;
 257    uint32_t log2_samples = 0;
 258
 259    /* From the Vulkan 1.0 spec:
 260     *    If pSampleMask is NULL, it is treated as if the mask has all bits
 261     *    enabled, i.e. no coverage is removed from fragments.
 262     *
 263     * 3DSTATE_SAMPLE_MASK.SampleMask is 16 bits.
 264     */
 265    uint32_t sample_mask = 0xffff;
 266
 267    if (info) {
 268       samples = info->rasterizationSamples;
 269       log2_samples = __builtin_ffs(samples) - 1;
 270    }
 271
 272    if (info && info->pSampleMask)
 273       sample_mask &= info->pSampleMask[0];
 274
 275    if (info && info->sampleShadingEnable)
 276       anv_finishme("VkPipelineMultisampleStateCreateInfo::sampleShadingEnable");
 277
 278    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_MULTISAMPLE),
 279
 280       /* The PRM says that this bit is valid only for DX9:
 281        *
 282        *    SW can choose to set this bit only for DX9 API. DX10/OGL API's
 283        *    should not have any effect by setting or not setting this bit.
 284        */
 285       .PixelPositionOffsetEnable = false,
 286
 287       .PixelLocation = CENTER,
 288       .NumberofMultisamples = log2_samples);
 289
 290    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_SAMPLE_MASK),
 291       .SampleMask = sample_mask);
 292 }
 293
 294 VkResult
 295 genX(graphics_pipeline_create)(
 296     VkDevice                                    _device,
 297     struct anv_pipeline_cache *                 cache,
 298     const VkGraphicsPipelineCreateInfo*         pCreateInfo,
 299     const struct anv_graphics_pipeline_create_info *extra,
 300     const VkAllocationCallbacks*                pAllocator,
 301     VkPipeline*                                 pPipeline)
 302 {
 303    ANV_FROM_HANDLE(anv_device, device, _device);
 304    struct anv_pipeline *pipeline;
 305    VkResult result;
 306    uint32_t offset, length;
 307
 308    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
 309
 310    pipeline = anv_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
 311                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 312    if (pipeline == NULL)
 313       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 314
 315    result = anv_pipeline_init(pipeline, device, cache,
 316                               pCreateInfo, extra, pAllocator);
 317    if (result != VK_SUCCESS) {
 318       anv_free2(&device->alloc, pAllocator, pipeline);
 319       return result;
 320    }
 321
 322    assert(pCreateInfo->pVertexInputState);
 323    emit_vertex_input(pipeline, pCreateInfo->pVertexInputState, extra);
 324    assert(pCreateInfo->pInputAssemblyState);
 325    emit_ia_state(pipeline, pCreateInfo->pInputAssemblyState, extra);
 326    assert(pCreateInfo->pRasterizationState);
 327    emit_rs_state(pipeline, pCreateInfo->pRasterizationState,
 328                  pCreateInfo->pMultisampleState, extra);
 329    emit_ms_state(pipeline, pCreateInfo->pMultisampleState);
 330    emit_ds_state(pipeline, pCreateInfo->pDepthStencilState);
 331    emit_cb_state(pipeline, pCreateInfo->pColorBlendState,
 332                            pCreateInfo->pMultisampleState);
 333
 334    emit_urb_setup(pipeline);
 335
 336    const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline);
 337    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_CLIP),
 338                   .ClipEnable = true,
 339                   .EarlyCullEnable = true,
 340                   .APIMode = 1, /* D3D */
 341                   .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
 342
 343                   .ClipMode =
 344                      pCreateInfo->pRasterizationState->rasterizerDiscardEnable ?
 345                      REJECT_ALL : NORMAL,
 346
 347                   .NonPerspectiveBarycentricEnable = wm_prog_data ?
 348                      (wm_prog_data->barycentric_interp_modes & 0x38) != 0 : 0,
 349
 350                   .TriangleStripListProvokingVertexSelect = 0,
 351                   .LineStripListProvokingVertexSelect = 0,
 352                   .TriangleFanProvokingVertexSelect = 1,
 353
 354                   .MinimumPointWidth = 0.125,
 355                   .MaximumPointWidth = 255.875,
 356                   .MaximumVPIndex = pCreateInfo->pViewportState->viewportCount - 1);
 357
 358    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_WM),
 359                   .StatisticsEnable = true,
 360                   .LineEndCapAntialiasingRegionWidth = _05pixels,
 361                   .LineAntialiasingRegionWidth = _10pixels,
 362                   .EarlyDepthStencilControl = NORMAL,
 363                   .ForceThreadDispatchEnable = NORMAL,
 364                   .PointRasterizationRule = RASTRULE_UPPER_RIGHT,
 365                   .BarycentricInterpolationMode =
 366                      pipeline->ps_ksp0 == NO_KERNEL ?
 367                      0 : wm_prog_data->barycentric_interp_modes);
 368
 369    if (pipeline->gs_kernel == NO_KERNEL) {
 370       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_GS), .Enable = false);
 371    } else {
 372       const struct brw_gs_prog_data *gs_prog_data = get_gs_prog_data(pipeline);
 373       offset = 1;
 374       length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
 375
 376       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_GS),
 377                      .SingleProgramFlow = false,
 378                      .KernelStartPointer = pipeline->gs_kernel,
 379                      .VectorMaskEnable = false,
 380                      .SamplerCount = 0,
 381                      .BindingTableEntryCount = 0,
 382                      .ExpectedVertexCount = gs_prog_data->vertices_in,
 383
 384                      .ScratchSpaceBasePointer = pipeline->scratch_start[MESA_SHADER_GEOMETRY],
 385                      .PerThreadScratchSpace = scratch_space(&gs_prog_data->base.base),
 386
 387                      .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
 388                      .OutputTopology = gs_prog_data->output_topology,
 389                      .VertexURBEntryReadLength = gs_prog_data->base.urb_read_length,
 390                      .IncludeVertexHandles = gs_prog_data->base.include_vue_handles,
 391                      .DispatchGRFStartRegisterForURBData =
 392                         gs_prog_data->base.base.dispatch_grf_start_reg,
 393
 394                      .MaximumNumberofThreads = device->info.max_gs_threads / 2 - 1,
 395                      .ControlDataHeaderSize = gs_prog_data->control_data_header_size_hwords,
 396                      .DispatchMode = gs_prog_data->base.dispatch_mode,
 397                      .StatisticsEnable = true,
 398                      .IncludePrimitiveID = gs_prog_data->include_primitive_id,
 399                      .ReorderMode = TRAILING,
 400                      .Enable = true,
 401
 402                      .ControlDataFormat = gs_prog_data->control_data_format,
 403
 404                      .StaticOutput = gs_prog_data->static_vertex_count >= 0,
 405                      .StaticOutputVertexCount =
 406                         gs_prog_data->static_vertex_count >= 0 ?
 407                         gs_prog_data->static_vertex_count : 0,
 408
 409                      /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
 410                       * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
 411                       * UserClipDistanceCullTestEnableBitmask(v)
 412                       */
 413
 414                      .VertexURBEntryOutputReadOffset = offset,
 415                      .VertexURBEntryOutputLength = length);
 416    }
 417
 418    const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);
 419    /* Skip the VUE header and position slots */
 420    offset = 1;
 421    length = (vs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
 422
 423    uint32_t vs_start = pipeline->vs_simd8 != NO_KERNEL ? pipeline->vs_simd8 :
 424                                                          pipeline->vs_vec4;
 425
 426    if (vs_start == NO_KERNEL || (extra && extra->disable_vs))
 427       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VS),
 428                      .FunctionEnable = false,
 429                      /* Even if VS is disabled, SBE still gets the amount of
 430                       * vertex data to read from this field. */
 431                      .VertexURBEntryOutputReadOffset = offset,
 432                      .VertexURBEntryOutputLength = length);
 433    else
 434       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VS),
 435                      .KernelStartPointer = vs_start,
 436                      .SingleVertexDispatch = false,
 437                      .VectorMaskEnable = false,
 438                      .SamplerCount = 0,
 439                      .BindingTableEntryCount =
 440                         vs_prog_data->base.base.binding_table.size_bytes / 4,
 441                      .ThreadDispatchPriority = false,
 442                      .FloatingPointMode = IEEE754,
 443                      .IllegalOpcodeExceptionEnable = false,
 444                      .AccessesUAV = false,
 445                      .SoftwareExceptionEnable = false,
 446
 447                      .ScratchSpaceBasePointer = pipeline->scratch_start[MESA_SHADER_VERTEX],
 448                      .PerThreadScratchSpace = scratch_space(&vs_prog_data->base.base),
 449
 450                      .DispatchGRFStartRegisterForURBData =
 451                         vs_prog_data->base.base.dispatch_grf_start_reg,
 452                      .VertexURBEntryReadLength = vs_prog_data->base.urb_read_length,
 453                      .VertexURBEntryReadOffset = 0,
 454
 455                      .MaximumNumberofThreads = device->info.max_vs_threads - 1,
 456                      .StatisticsEnable = false,
 457                      .SIMD8DispatchEnable = pipeline->vs_simd8 != NO_KERNEL,
 458                      .VertexCacheDisable = false,
 459                      .FunctionEnable = true,
 460
 461                      .VertexURBEntryOutputReadOffset = offset,
 462                      .VertexURBEntryOutputLength = length,
 463                      .UserClipDistanceClipTestEnableBitmask = 0,
 464                      .UserClipDistanceCullTestEnableBitmask = 0);
 465
 466    const int num_thread_bias = GEN_GEN == 8 ? 2 : 1;
 467    if (pipeline->ps_ksp0 == NO_KERNEL) {
 468       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS));
 469       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS_EXTRA),
 470                      .PixelShaderValid = false);
 471    } else {
 472       emit_3dstate_sbe(pipeline);
 473
 474       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS),
 475                      .KernelStartPointer0 = pipeline->ps_ksp0,
 476
 477                      .SingleProgramFlow = false,
 478                      .VectorMaskEnable = true,
 479                      .SamplerCount = 1,
 480
 481                      .ScratchSpaceBasePointer = pipeline->scratch_start[MESA_SHADER_FRAGMENT],
 482                      .PerThreadScratchSpace = scratch_space(&wm_prog_data->base),
 483
 484                      .MaximumNumberofThreadsPerPSD = 64 - num_thread_bias,
 485                      .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
 486                         POSOFFSET_SAMPLE: POSOFFSET_NONE,
 487                      .PushConstantEnable = wm_prog_data->base.nr_params > 0,
 488                      ._8PixelDispatchEnable = pipeline->ps_simd8 != NO_KERNEL,
 489                      ._16PixelDispatchEnable = pipeline->ps_simd16 != NO_KERNEL,
 490                      ._32PixelDispatchEnable = false,
 491
 492                      .DispatchGRFStartRegisterForConstantSetupData0 = pipeline->ps_grf_start0,
 493                      .DispatchGRFStartRegisterForConstantSetupData1 = 0,
 494                      .DispatchGRFStartRegisterForConstantSetupData2 = pipeline->ps_grf_start2,
 495
 496                      .KernelStartPointer1 = 0,
 497                      .KernelStartPointer2 = pipeline->ps_ksp2);
 498
 499       bool per_sample_ps = pCreateInfo->pMultisampleState &&
 500                            pCreateInfo->pMultisampleState->sampleShadingEnable;
 501
 502       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS_EXTRA),
 503                      .PixelShaderValid = true,
 504                      .PixelShaderKillsPixel = wm_prog_data->uses_kill,
 505                      .PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode,
 506                      .AttributeEnable = wm_prog_data->num_varying_inputs > 0,
 507                      .oMaskPresenttoRenderTarget = wm_prog_data->uses_omask,
 508                      .PixelShaderIsPerSample = per_sample_ps,
 509                      .PixelShaderUsesSourceDepth = wm_prog_data->uses_src_depth,
 510                      .PixelShaderUsesSourceW = wm_prog_data->uses_src_w,
 511 #if GEN_GEN >= 9
 512                      .PixelShaderPullsBary = wm_prog_data->pulls_bary,
 513                      .InputCoverageMaskState = wm_prog_data->uses_sample_mask ?
 514                         ICMS_INNER_CONSERVATIVE : ICMS_NONE,
 515 #else
 516                      .PixelShaderUsesInputCoverageMask =
 517                         wm_prog_data->uses_sample_mask,
 518 #endif
 519          );
 520    }
 521
 522    *pPipeline = anv_pipeline_to_handle(pipeline);
 523
 524    return VK_SUCCESS;
 525 }