src/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 "gen8_pack.h"
  33 #include "gen9_pack.h"
  34
  35 #include "genX_pipeline_util.h"
  36
  37 static void
  38 emit_vertex_input(struct anv_pipeline *pipeline,
  39                   const VkPipelineVertexInputStateCreateInfo *info,
  40                   const struct anv_graphics_pipeline_create_info *extra)
  41 {
  42    static_assert(ANV_GEN >= 8, "should be compiling this for gen < 8");
  43
  44    uint32_t elements;
  45    if (extra && extra->disable_vs) {
  46       /* If the VS is disabled, just assume the user knows what they're
  47        * doing and apply the layout blindly.  This can only come from
  48        * meta, so this *should* be safe.
  49        */
  50       elements = 0;
  51       for (uint32_t i = 0; i < info->vertexAttributeDescriptionCount; i++)
  52          elements |= (1 << info->pVertexAttributeDescriptions[i].location);
  53    } else {
  54       /* Pull inputs_read out of the VS prog data */
  55       uint64_t inputs_read = pipeline->vs_prog_data.inputs_read;
  56       assert((inputs_read & ((1 << VERT_ATTRIB_GENERIC0) - 1)) == 0);
  57       elements = inputs_read >> VERT_ATTRIB_GENERIC0;
  58    }
  59
  60    const uint32_t num_dwords = 1 + __builtin_popcount(elements) * 2;
  61
  62    uint32_t *p;
  63    if (elements != 0) {
  64       p = anv_batch_emitn(&pipeline->batch, num_dwords,
  65                           GENX(3DSTATE_VERTEX_ELEMENTS));
  66       memset(p + 1, 0, (num_dwords - 1) * 4);
  67    }
  68
  69    for (uint32_t i = 0; i < info->vertexAttributeDescriptionCount; i++) {
  70       const VkVertexInputAttributeDescription *desc =
  71          &info->pVertexAttributeDescriptions[i];
  72       enum isl_format format = anv_get_isl_format(desc->format,
  73                                                   VK_IMAGE_ASPECT_COLOR_BIT,
  74                                                   VK_IMAGE_TILING_LINEAR);
  75
  76       assert(desc->binding < 32);
  77
  78       if ((elements & (1 << desc->location)) == 0)
  79          continue; /* Binding unused */
  80
  81       uint32_t slot = __builtin_popcount(elements & ((1 << desc->location) - 1));
  82
  83       struct GENX(VERTEX_ELEMENT_STATE) element = {
  84          .VertexBufferIndex = desc->binding,
  85          .Valid = true,
  86          .SourceElementFormat = format,
  87          .EdgeFlagEnable = false,
  88          .SourceElementOffset = desc->offset,
  89          .Component0Control = vertex_element_comp_control(format, 0),
  90          .Component1Control = vertex_element_comp_control(format, 1),
  91          .Component2Control = vertex_element_comp_control(format, 2),
  92          .Component3Control = vertex_element_comp_control(format, 3),
  93       };
  94       GENX(VERTEX_ELEMENT_STATE_pack)(NULL, &p[1 + slot * 2], &element);
  95
  96       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VF_INSTANCING),
  97                      .InstancingEnable = pipeline->instancing_enable[desc->binding],
  98                      .VertexElementIndex = slot,
  99                      /* Vulkan so far doesn't have an instance divisor, so
 100                       * this is always 1 (ignored if not instancing). */
 101                      .InstanceDataStepRate = 1);
 102    }
 103
 104    const uint32_t id_slot = __builtin_popcount(elements);
 105    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VF_SGVS),
 106                   .VertexIDEnable = pipeline->vs_prog_data.uses_vertexid,
 107                   .VertexIDComponentNumber = 2,
 108                   .VertexIDElementOffset = id_slot,
 109                   .InstanceIDEnable = pipeline->vs_prog_data.uses_instanceid,
 110                   .InstanceIDComponentNumber = 3,
 111                   .InstanceIDElementOffset = id_slot);
 112 }
 113
 114 static void
 115 emit_ia_state(struct anv_pipeline *pipeline,
 116               const VkPipelineInputAssemblyStateCreateInfo *info,
 117               const struct anv_graphics_pipeline_create_info *extra)
 118 {
 119    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VF_TOPOLOGY),
 120                   .PrimitiveTopologyType = pipeline->topology);
 121 }
 122
 123 static void
 124 emit_rs_state(struct anv_pipeline *pipeline,
 125               const VkPipelineRasterizationStateCreateInfo *info,
 126               const struct anv_graphics_pipeline_create_info *extra)
 127 {
 128    struct GENX(3DSTATE_SF) sf = {
 129       GENX(3DSTATE_SF_header),
 130       .ViewportTransformEnable = !(extra && extra->disable_viewport),
 131       .TriangleStripListProvokingVertexSelect = 0,
 132       .LineStripListProvokingVertexSelect = 0,
 133       .TriangleFanProvokingVertexSelect = 0,
 134       .PointWidthSource = pipeline->writes_point_size ? Vertex : State,
 135       .PointWidth = 1.0,
 136    };
 137
 138    /* FINISHME: VkBool32 rasterizerDiscardEnable; */
 139
 140    GENX(3DSTATE_SF_pack)(NULL, pipeline->gen8.sf, &sf);
 141
 142    struct GENX(3DSTATE_RASTER) raster = {
 143       GENX(3DSTATE_RASTER_header),
 144       .FrontWinding = vk_to_gen_front_face[info->frontFace],
 145       .CullMode = vk_to_gen_cullmode[info->cullMode],
 146       .FrontFaceFillMode = vk_to_gen_fillmode[info->polygonMode],
 147       .BackFaceFillMode = vk_to_gen_fillmode[info->polygonMode],
 148       .ScissorRectangleEnable = !(extra && extra->disable_scissor),
 149 #if ANV_GEN == 8
 150       .ViewportZClipTestEnable = true,
 151 #else
 152       /* GEN9+ splits ViewportZClipTestEnable into near and far enable bits */
 153       .ViewportZFarClipTestEnable = true,
 154       .ViewportZNearClipTestEnable = true,
 155 #endif
 156    };
 157
 158    GENX(3DSTATE_RASTER_pack)(NULL, pipeline->gen8.raster, &raster);
 159 }
 160
 161 static void
 162 emit_cb_state(struct anv_pipeline *pipeline,
 163               const VkPipelineColorBlendStateCreateInfo *info,
 164               const VkPipelineMultisampleStateCreateInfo *ms_info)
 165 {
 166    struct anv_device *device = pipeline->device;
 167
 168    uint32_t num_dwords = GENX(BLEND_STATE_length);
 169    pipeline->blend_state =
 170       anv_state_pool_alloc(&device->dynamic_state_pool, num_dwords * 4, 64);
 171
 172    struct GENX(BLEND_STATE) blend_state = {
 173       .AlphaToCoverageEnable = ms_info && ms_info->alphaToCoverageEnable,
 174       .AlphaToOneEnable = ms_info && ms_info->alphaToOneEnable,
 175    };
 176
 177    for (uint32_t i = 0; i < info->attachmentCount; i++) {
 178       const VkPipelineColorBlendAttachmentState *a = &info->pAttachments[i];
 179
 180       if (a->srcColorBlendFactor != a->srcAlphaBlendFactor ||
 181           a->dstColorBlendFactor != a->dstAlphaBlendFactor ||
 182           a->colorBlendOp != a->alphaBlendOp) {
 183          blend_state.IndependentAlphaBlendEnable = true;
 184       }
 185
 186       blend_state.Entry[i] = (struct GENX(BLEND_STATE_ENTRY)) {
 187          .LogicOpEnable = info->logicOpEnable,
 188          .LogicOpFunction = vk_to_gen_logic_op[info->logicOp],
 189          .ColorBufferBlendEnable = a->blendEnable,
 190          .PreBlendSourceOnlyClampEnable = false,
 191          .ColorClampRange = COLORCLAMP_RTFORMAT,
 192          .PreBlendColorClampEnable = true,
 193          .PostBlendColorClampEnable = true,
 194          .SourceBlendFactor = vk_to_gen_blend[a->srcColorBlendFactor],
 195          .DestinationBlendFactor = vk_to_gen_blend[a->dstColorBlendFactor],
 196          .ColorBlendFunction = vk_to_gen_blend_op[a->colorBlendOp],
 197          .SourceAlphaBlendFactor = vk_to_gen_blend[a->srcAlphaBlendFactor],
 198          .DestinationAlphaBlendFactor = vk_to_gen_blend[a->dstAlphaBlendFactor],
 199          .AlphaBlendFunction = vk_to_gen_blend_op[a->alphaBlendOp],
 200          .WriteDisableAlpha = !(a->colorWriteMask & VK_COLOR_COMPONENT_A_BIT),
 201          .WriteDisableRed = !(a->colorWriteMask & VK_COLOR_COMPONENT_R_BIT),
 202          .WriteDisableGreen = !(a->colorWriteMask & VK_COLOR_COMPONENT_G_BIT),
 203          .WriteDisableBlue = !(a->colorWriteMask & VK_COLOR_COMPONENT_B_BIT),
 204       };
 205
 206       /* Our hardware applies the blend factor prior to the blend function
 207        * regardless of what function is used.  Technically, this means the
 208        * hardware can do MORE than GL or Vulkan specify.  However, it also
 209        * means that, for MIN and MAX, we have to stomp the blend factor to
 210        * ONE to make it a no-op.
 211        */
 212       if (a->colorBlendOp == VK_BLEND_OP_MIN ||
 213           a->colorBlendOp == VK_BLEND_OP_MAX) {
 214          blend_state.Entry[i].SourceBlendFactor = BLENDFACTOR_ONE;
 215          blend_state.Entry[i].DestinationBlendFactor = BLENDFACTOR_ONE;
 216       }
 217       if (a->alphaBlendOp == VK_BLEND_OP_MIN ||
 218           a->alphaBlendOp == VK_BLEND_OP_MAX) {
 219          blend_state.Entry[i].SourceAlphaBlendFactor = BLENDFACTOR_ONE;
 220          blend_state.Entry[i].DestinationAlphaBlendFactor = BLENDFACTOR_ONE;
 221       }
 222    }
 223
 224    GENX(BLEND_STATE_pack)(NULL, pipeline->blend_state.map, &blend_state);
 225    if (!device->info.has_llc)
 226       anv_state_clflush(pipeline->blend_state);
 227
 228    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_BLEND_STATE_POINTERS),
 229                   .BlendStatePointer = pipeline->blend_state.offset,
 230                   .BlendStatePointerValid = true);
 231 }
 232
 233 static void
 234 emit_ds_state(struct anv_pipeline *pipeline,
 235               const VkPipelineDepthStencilStateCreateInfo *info)
 236 {
 237    uint32_t *dw = ANV_GEN == 8 ?
 238       pipeline->gen8.wm_depth_stencil : pipeline->gen9.wm_depth_stencil;
 239
 240    if (info == NULL) {
 241       /* We're going to OR this together with the dynamic state.  We need
 242        * to make sure it's initialized to something useful.
 243        */
 244       memset(pipeline->gen8.wm_depth_stencil, 0,
 245              sizeof(pipeline->gen8.wm_depth_stencil));
 246       memset(pipeline->gen9.wm_depth_stencil, 0,
 247              sizeof(pipeline->gen9.wm_depth_stencil));
 248       return;
 249    }
 250
 251    /* VkBool32 depthBoundsTestEnable; // optional (depth_bounds_test) */
 252
 253    struct GENX(3DSTATE_WM_DEPTH_STENCIL) wm_depth_stencil = {
 254       .DepthTestEnable = info->depthTestEnable,
 255       .DepthBufferWriteEnable = info->depthWriteEnable,
 256       .DepthTestFunction = vk_to_gen_compare_op[info->depthCompareOp],
 257       .DoubleSidedStencilEnable = true,
 258
 259       .StencilTestEnable = info->stencilTestEnable,
 260       .StencilFailOp = vk_to_gen_stencil_op[info->front.failOp],
 261       .StencilPassDepthPassOp = vk_to_gen_stencil_op[info->front.passOp],
 262       .StencilPassDepthFailOp = vk_to_gen_stencil_op[info->front.depthFailOp],
 263       .StencilTestFunction = vk_to_gen_compare_op[info->front.compareOp],
 264       .BackfaceStencilFailOp = vk_to_gen_stencil_op[info->back.failOp],
 265       .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[info->back.passOp],
 266       .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[info->back.depthFailOp],
 267       .BackfaceStencilTestFunction = vk_to_gen_compare_op[info->back.compareOp],
 268    };
 269
 270    GENX(3DSTATE_WM_DEPTH_STENCIL_pack)(NULL, dw, &wm_depth_stencil);
 271 }
 272
 273 VkResult
 274 genX(graphics_pipeline_create)(
 275     VkDevice                                    _device,
 276     const VkGraphicsPipelineCreateInfo*         pCreateInfo,
 277     const struct anv_graphics_pipeline_create_info *extra,
 278     const VkAllocationCallbacks*                pAllocator,
 279     VkPipeline*                                 pPipeline)
 280 {
 281    ANV_FROM_HANDLE(anv_device, device, _device);
 282    struct anv_pipeline *pipeline;
 283    VkResult result;
 284    uint32_t offset, length;
 285
 286    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
 287
 288    pipeline = anv_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
 289                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 290    if (pipeline == NULL)
 291       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 292
 293    result = anv_pipeline_init(pipeline, device, pCreateInfo, extra, pAllocator);
 294    if (result != VK_SUCCESS) {
 295       anv_free2(&device->alloc, pAllocator, pipeline);
 296       return result;
 297    }
 298
 299    assert(pCreateInfo->pVertexInputState);
 300    emit_vertex_input(pipeline, pCreateInfo->pVertexInputState, extra);
 301    assert(pCreateInfo->pInputAssemblyState);
 302    emit_ia_state(pipeline, pCreateInfo->pInputAssemblyState, extra);
 303    assert(pCreateInfo->pRasterizationState);
 304    emit_rs_state(pipeline, pCreateInfo->pRasterizationState, extra);
 305    emit_ds_state(pipeline, pCreateInfo->pDepthStencilState);
 306    emit_cb_state(pipeline, pCreateInfo->pColorBlendState,
 307                            pCreateInfo->pMultisampleState);
 308
 309    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VF_STATISTICS),
 310                    .StatisticsEnable = true);
 311    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_HS), .Enable = false);
 312    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_TE), .TEEnable = false);
 313    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_DS), .FunctionEnable = false);
 314    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_STREAMOUT), .SOFunctionEnable = false);
 315
 316    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PUSH_CONSTANT_ALLOC_VS),
 317                   .ConstantBufferOffset = 0,
 318                   .ConstantBufferSize = 4);
 319    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PUSH_CONSTANT_ALLOC_GS),
 320                   .ConstantBufferOffset = 4,
 321                   .ConstantBufferSize = 4);
 322    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PUSH_CONSTANT_ALLOC_PS),
 323                   .ConstantBufferOffset = 8,
 324                   .ConstantBufferSize = 4);
 325
 326    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_WM_CHROMAKEY),
 327                   .ChromaKeyKillEnable = false);
 328    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_AA_LINE_PARAMETERS));
 329
 330    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_CLIP),
 331                   .ClipEnable = true,
 332                   .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
 333                   .MinimumPointWidth = 0.125,
 334                   .MaximumPointWidth = 255.875);
 335
 336    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_WM),
 337                   .StatisticsEnable = true,
 338                   .LineEndCapAntialiasingRegionWidth = _05pixels,
 339                   .LineAntialiasingRegionWidth = _10pixels,
 340                   .EarlyDepthStencilControl = NORMAL,
 341                   .ForceThreadDispatchEnable = NORMAL,
 342                   .PointRasterizationRule = RASTRULE_UPPER_RIGHT,
 343                   .BarycentricInterpolationMode =
 344                      pipeline->wm_prog_data.barycentric_interp_modes);
 345
 346    uint32_t samples = 1;
 347    uint32_t log2_samples = __builtin_ffs(samples) - 1;
 348    bool enable_sampling = samples > 1 ? true : false;
 349
 350    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_MULTISAMPLE),
 351                   .PixelPositionOffsetEnable = enable_sampling,
 352                   .PixelLocation = CENTER,
 353                   .NumberofMultisamples = log2_samples);
 354
 355    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_SAMPLE_MASK),
 356                   .SampleMask = 0xffff);
 357
 358    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_URB_VS),
 359                   .VSURBStartingAddress = pipeline->urb.vs_start,
 360                   .VSURBEntryAllocationSize = pipeline->urb.vs_size - 1,
 361                   .VSNumberofURBEntries = pipeline->urb.nr_vs_entries);
 362
 363    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_URB_GS),
 364                   .GSURBStartingAddress = pipeline->urb.gs_start,
 365                   .GSURBEntryAllocationSize = pipeline->urb.gs_size - 1,
 366                   .GSNumberofURBEntries = pipeline->urb.nr_gs_entries);
 367
 368    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_URB_HS),
 369                   .HSURBStartingAddress = pipeline->urb.vs_start,
 370                   .HSURBEntryAllocationSize = 0,
 371                   .HSNumberofURBEntries = 0);
 372
 373    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_URB_DS),
 374                   .DSURBStartingAddress = pipeline->urb.vs_start,
 375                   .DSURBEntryAllocationSize = 0,
 376                   .DSNumberofURBEntries = 0);
 377
 378    const struct brw_gs_prog_data *gs_prog_data = &pipeline->gs_prog_data;
 379    offset = 1;
 380    length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
 381
 382    if (pipeline->gs_kernel == NO_KERNEL)
 383       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_GS), .Enable = false);
 384    else
 385       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_GS),
 386                      .SingleProgramFlow = false,
 387                      .KernelStartPointer = pipeline->gs_kernel,
 388                      .VectorMaskEnable = Dmask,
 389                      .SamplerCount = 0,
 390                      .BindingTableEntryCount = 0,
 391                      .ExpectedVertexCount = pipeline->gs_vertex_count,
 392
 393                      .ScratchSpaceBasePointer = pipeline->scratch_start[MESA_SHADER_GEOMETRY],
 394                      .PerThreadScratchSpace = ffs(gs_prog_data->base.base.total_scratch / 2048),
 395
 396                      .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
 397                      .OutputTopology = gs_prog_data->output_topology,
 398                      .VertexURBEntryReadLength = gs_prog_data->base.urb_read_length,
 399                      .IncludeVertexHandles = gs_prog_data->base.include_vue_handles,
 400                      .DispatchGRFStartRegisterForURBData =
 401                         gs_prog_data->base.base.dispatch_grf_start_reg,
 402
 403                      .MaximumNumberofThreads = device->info.max_gs_threads / 2 - 1,
 404                      .ControlDataHeaderSize = gs_prog_data->control_data_header_size_hwords,
 405                      .DispatchMode = gs_prog_data->base.dispatch_mode,
 406                      .StatisticsEnable = true,
 407                      .IncludePrimitiveID = gs_prog_data->include_primitive_id,
 408                      .ReorderMode = TRAILING,
 409                      .Enable = true,
 410
 411                      .ControlDataFormat = gs_prog_data->control_data_format,
 412
 413                      .StaticOutput = gs_prog_data->static_vertex_count >= 0,
 414                      .StaticOutputVertexCount =
 415                         gs_prog_data->static_vertex_count >= 0 ?
 416                         gs_prog_data->static_vertex_count : 0,
 417
 418                      /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
 419                       * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
 420                       * UserClipDistanceCullTestEnableBitmask(v)
 421                       */
 422
 423                      .VertexURBEntryOutputReadOffset = offset,
 424                      .VertexURBEntryOutputLength = length);
 425
 426    const struct brw_vue_prog_data *vue_prog_data = &pipeline->vs_prog_data.base;
 427    /* Skip the VUE header and position slots */
 428    offset = 1;
 429    length = (vue_prog_data->vue_map.num_slots + 1) / 2 - offset;
 430
 431    uint32_t vs_start = pipeline->vs_simd8 != NO_KERNEL ? pipeline->vs_simd8 :
 432                                                          pipeline->vs_vec4;
 433
 434    if (vs_start == NO_KERNEL || (extra && extra->disable_vs))
 435       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VS),
 436                      .FunctionEnable = false,
 437                      /* Even if VS is disabled, SBE still gets the amount of
 438                       * vertex data to read from this field. */
 439                      .VertexURBEntryOutputReadOffset = offset,
 440                      .VertexURBEntryOutputLength = length);
 441    else
 442       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VS),
 443                      .KernelStartPointer = vs_start,
 444                      .SingleVertexDispatch = Multiple,
 445                      .VectorMaskEnable = Dmask,
 446                      .SamplerCount = 0,
 447                      .BindingTableEntryCount =
 448                      vue_prog_data->base.binding_table.size_bytes / 4,
 449                      .ThreadDispatchPriority = Normal,
 450                      .FloatingPointMode = IEEE754,
 451                      .IllegalOpcodeExceptionEnable = false,
 452                      .AccessesUAV = false,
 453                      .SoftwareExceptionEnable = false,
 454
 455                      .ScratchSpaceBasePointer = pipeline->scratch_start[MESA_SHADER_VERTEX],
 456                      .PerThreadScratchSpace = ffs(vue_prog_data->base.total_scratch / 2048),
 457
 458                      .DispatchGRFStartRegisterForURBData =
 459                      vue_prog_data->base.dispatch_grf_start_reg,
 460                      .VertexURBEntryReadLength = vue_prog_data->urb_read_length,
 461                      .VertexURBEntryReadOffset = 0,
 462
 463                      .MaximumNumberofThreads = device->info.max_vs_threads - 1,
 464                      .StatisticsEnable = false,
 465                      .SIMD8DispatchEnable = pipeline->vs_simd8 != NO_KERNEL,
 466                      .VertexCacheDisable = false,
 467                      .FunctionEnable = true,
 468
 469                      .VertexURBEntryOutputReadOffset = offset,
 470                      .VertexURBEntryOutputLength = length,
 471                      .UserClipDistanceClipTestEnableBitmask = 0,
 472                      .UserClipDistanceCullTestEnableBitmask = 0);
 473
 474    const struct brw_wm_prog_data *wm_prog_data = &pipeline->wm_prog_data;
 475
 476    /* TODO: We should clean this up.  Among other things, this is mostly
 477     * shared with other gens.
 478     */
 479    const struct brw_vue_map *fs_input_map;
 480    if (pipeline->gs_kernel == NO_KERNEL)
 481       fs_input_map = &vue_prog_data->vue_map;
 482    else
 483       fs_input_map = &gs_prog_data->base.vue_map;
 484
 485    struct GENX(3DSTATE_SBE_SWIZ) swiz = {
 486       GENX(3DSTATE_SBE_SWIZ_header),
 487    };
 488
 489    int max_source_attr = 0;
 490    for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 491       int input_index = wm_prog_data->urb_setup[attr];
 492
 493       if (input_index < 0)
 494          continue;
 495
 496       /* We have to subtract two slots to accout for the URB entry output
 497        * read offset in the VS and GS stages.
 498        */
 499       int source_attr = fs_input_map->varying_to_slot[attr] - 2;
 500       max_source_attr = MAX2(max_source_attr, source_attr);
 501
 502       if (input_index >= 16)
 503          continue;
 504
 505       swiz.Attribute[input_index].SourceAttribute = source_attr;
 506    }
 507
 508    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_SBE),
 509                   .AttributeSwizzleEnable = true,
 510                   .ForceVertexURBEntryReadLength = false,
 511                   .ForceVertexURBEntryReadOffset = false,
 512                   .VertexURBEntryReadLength = DIV_ROUND_UP(max_source_attr + 1, 2),
 513                   .PointSpriteTextureCoordinateOrigin = UPPERLEFT,
 514                   .NumberofSFOutputAttributes =
 515                      wm_prog_data->num_varying_inputs,
 516
 517 #if ANV_GEN >= 9
 518                   .Attribute0ActiveComponentFormat = ACF_XYZW,
 519                   .Attribute1ActiveComponentFormat = ACF_XYZW,
 520                   .Attribute2ActiveComponentFormat = ACF_XYZW,
 521                   .Attribute3ActiveComponentFormat = ACF_XYZW,
 522                   .Attribute4ActiveComponentFormat = ACF_XYZW,
 523                   .Attribute5ActiveComponentFormat = ACF_XYZW,
 524                   .Attribute6ActiveComponentFormat = ACF_XYZW,
 525                   .Attribute7ActiveComponentFormat = ACF_XYZW,
 526                   .Attribute8ActiveComponentFormat = ACF_XYZW,
 527                   .Attribute9ActiveComponentFormat = ACF_XYZW,
 528                   .Attribute10ActiveComponentFormat = ACF_XYZW,
 529                   .Attribute11ActiveComponentFormat = ACF_XYZW,
 530                   .Attribute12ActiveComponentFormat = ACF_XYZW,
 531                   .Attribute13ActiveComponentFormat = ACF_XYZW,
 532                   .Attribute14ActiveComponentFormat = ACF_XYZW,
 533                   .Attribute15ActiveComponentFormat = ACF_XYZW,
 534                   /* wow, much field, very attribute */
 535                   .Attribute16ActiveComponentFormat = ACF_XYZW,
 536                   .Attribute17ActiveComponentFormat = ACF_XYZW,
 537                   .Attribute18ActiveComponentFormat = ACF_XYZW,
 538                   .Attribute19ActiveComponentFormat = ACF_XYZW,
 539                   .Attribute20ActiveComponentFormat = ACF_XYZW,
 540                   .Attribute21ActiveComponentFormat = ACF_XYZW,
 541                   .Attribute22ActiveComponentFormat = ACF_XYZW,
 542                   .Attribute23ActiveComponentFormat = ACF_XYZW,
 543                   .Attribute24ActiveComponentFormat = ACF_XYZW,
 544                   .Attribute25ActiveComponentFormat = ACF_XYZW,
 545                   .Attribute26ActiveComponentFormat = ACF_XYZW,
 546                   .Attribute27ActiveComponentFormat = ACF_XYZW,
 547                   .Attribute28ActiveComponentFormat = ACF_XYZW,
 548                   .Attribute29ActiveComponentFormat = ACF_XYZW,
 549                   .Attribute28ActiveComponentFormat = ACF_XYZW,
 550                   .Attribute29ActiveComponentFormat = ACF_XYZW,
 551                   .Attribute30ActiveComponentFormat = ACF_XYZW,
 552 #endif
 553       );
 554
 555    uint32_t *dw = anv_batch_emit_dwords(&pipeline->batch,
 556                                         GENX(3DSTATE_SBE_SWIZ_length));
 557    GENX(3DSTATE_SBE_SWIZ_pack)(&pipeline->batch, dw, &swiz);
 558
 559    const int num_thread_bias = ANV_GEN == 8 ? 2 : 1;
 560    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS),
 561                   .KernelStartPointer0 = pipeline->ps_ksp0,
 562
 563                   .SingleProgramFlow = false,
 564                   .VectorMaskEnable = true,
 565                   .SamplerCount = 1,
 566
 567                   .ScratchSpaceBasePointer = pipeline->scratch_start[MESA_SHADER_FRAGMENT],
 568                   .PerThreadScratchSpace = ffs(wm_prog_data->base.total_scratch / 2048),
 569
 570                   .MaximumNumberofThreadsPerPSD = 64 - num_thread_bias,
 571                   .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
 572                      POSOFFSET_SAMPLE: POSOFFSET_NONE,
 573                   .PushConstantEnable = wm_prog_data->base.nr_params > 0,
 574                   ._8PixelDispatchEnable = pipeline->ps_simd8 != NO_KERNEL,
 575                   ._16PixelDispatchEnable = pipeline->ps_simd16 != NO_KERNEL,
 576                   ._32PixelDispatchEnable = false,
 577
 578                   .DispatchGRFStartRegisterForConstantSetupData0 = pipeline->ps_grf_start0,
 579                   .DispatchGRFStartRegisterForConstantSetupData1 = 0,
 580                   .DispatchGRFStartRegisterForConstantSetupData2 = pipeline->ps_grf_start2,
 581
 582                   .KernelStartPointer1 = 0,
 583                   .KernelStartPointer2 = pipeline->ps_ksp2);
 584
 585    bool per_sample_ps = false;
 586    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS_EXTRA),
 587                   .PixelShaderValid = true,
 588                   .PixelShaderKillsPixel = wm_prog_data->uses_kill,
 589                   .PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode,
 590                   .AttributeEnable = wm_prog_data->num_varying_inputs > 0,
 591                   .oMaskPresenttoRenderTarget = wm_prog_data->uses_omask,
 592                   .PixelShaderIsPerSample = per_sample_ps,
 593 #if ANV_GEN >= 9
 594                   .PixelShaderPullsBary = wm_prog_data->pulls_bary,
 595                   .InputCoverageMaskState = ICMS_NONE
 596 #endif
 597       );
 598
 599    *pPipeline = anv_pipeline_to_handle(pipeline);
 600
 601    return VK_SUCCESS;
 602 }
 603
 604 VkResult genX(compute_pipeline_create)(
 605     VkDevice                                    _device,
 606     const VkComputePipelineCreateInfo*          pCreateInfo,
 607     const VkAllocationCallbacks*                pAllocator,
 608     VkPipeline*                                 pPipeline)
 609 {
 610    ANV_FROM_HANDLE(anv_device, device, _device);
 611    struct anv_pipeline *pipeline;
 612    VkResult result;
 613
 614    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
 615
 616    pipeline = anv_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
 617                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 618    if (pipeline == NULL)
 619       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 620
 621    pipeline->device = device;
 622    pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
 623
 624    pipeline->blend_state.map = NULL;
 625
 626    result = anv_reloc_list_init(&pipeline->batch_relocs,
 627                                 pAllocator ? pAllocator : &device->alloc);
 628    if (result != VK_SUCCESS) {
 629       anv_free2(&device->alloc, pAllocator, pipeline);
 630       return result;
 631    }
 632    pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
 633    pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
 634    pipeline->batch.relocs = &pipeline->batch_relocs;
 635
 636    anv_state_stream_init(&pipeline->program_stream,
 637                          &device->instruction_block_pool);
 638
 639    /* When we free the pipeline, we detect stages based on the NULL status
 640     * of various prog_data pointers.  Make them NULL by default.
 641     */
 642    memset(pipeline->prog_data, 0, sizeof(pipeline->prog_data));
 643    memset(pipeline->scratch_start, 0, sizeof(pipeline->scratch_start));
 644
 645    pipeline->vs_simd8 = NO_KERNEL;
 646    pipeline->vs_vec4 = NO_KERNEL;
 647    pipeline->gs_kernel = NO_KERNEL;
 648
 649    pipeline->active_stages = 0;
 650    pipeline->total_scratch = 0;
 651
 652    assert(pCreateInfo->stage.stage == VK_SHADER_STAGE_COMPUTE_BIT);
 653    ANV_FROM_HANDLE(anv_shader_module, module,  pCreateInfo->stage.module);
 654    anv_pipeline_compile_cs(pipeline, pCreateInfo, module,
 655                            pCreateInfo->stage.pName);
 656
 657    pipeline->use_repclear = false;
 658
 659    const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data;
 660
 661    anv_batch_emit(&pipeline->batch, GENX(MEDIA_VFE_STATE),
 662                   .ScratchSpaceBasePointer = pipeline->scratch_start[MESA_SHADER_COMPUTE],
 663                   .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
 664                   .ScratchSpaceBasePointerHigh = 0,
 665                   .StackSize = 0,
 666
 667                   .MaximumNumberofThreads = device->info.max_cs_threads - 1,
 668                   .NumberofURBEntries = 2,
 669                   .ResetGatewayTimer = true,
 670 #if ANV_GEN == 8
 671                   .BypassGatewayControl = true,
 672 #endif
 673                   .URBEntryAllocationSize = 2,
 674                   .CURBEAllocationSize = 0);
 675
 676    struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
 677    uint32_t group_size = prog_data->local_size[0] *
 678       prog_data->local_size[1] * prog_data->local_size[2];
 679    pipeline->cs_thread_width_max = DIV_ROUND_UP(group_size, prog_data->simd_size);
 680    uint32_t remainder = group_size & (prog_data->simd_size - 1);
 681
 682    if (remainder > 0)
 683       pipeline->cs_right_mask = ~0u >> (32 - remainder);
 684    else
 685       pipeline->cs_right_mask = ~0u >> (32 - prog_data->simd_size);
 686
 687
 688    *pPipeline = anv_pipeline_to_handle(pipeline);
 689
 690    return VK_SUCCESS;
 691 }