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 static void
  33 emit_vertex_input(struct anv_pipeline *pipeline,
  34                   const VkPipelineVertexInputStateCreateInfo *info)
  35 {
  36    const uint32_t num_dwords = 1 + info->attributeCount * 2;
  37    uint32_t *p;
  38
  39    if (info->attributeCount > 0) {
  40       p = anv_batch_emitn(&pipeline->batch, num_dwords,
  41                           GEN8_3DSTATE_VERTEX_ELEMENTS);
  42    }
  43
  44    for (uint32_t i = 0; i < info->attributeCount; i++) {
  45       const VkVertexInputAttributeDescription *desc =
  46          &info->pVertexAttributeDescriptions[i];
  47       const struct anv_format *format = anv_format_for_vk_format(desc->format);
  48
  49       struct GEN8_VERTEX_ELEMENT_STATE element = {
  50          .VertexBufferIndex = desc->binding,
  51          .Valid = true,
  52          .SourceElementFormat = format->surface_format,
  53          .EdgeFlagEnable = false,
  54          .SourceElementOffset = desc->offsetInBytes,
  55          .Component0Control = VFCOMP_STORE_SRC,
  56          .Component1Control = format->num_channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
  57          .Component2Control = format->num_channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
  58          .Component3Control = format->num_channels >= 4 ? VFCOMP_STORE_SRC : VFCOMP_STORE_1_FP
  59       };
  60       GEN8_VERTEX_ELEMENT_STATE_pack(NULL, &p[1 + i * 2], &element);
  61
  62       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_INSTANCING,
  63                      .InstancingEnable = pipeline->instancing_enable[desc->binding],
  64                      .VertexElementIndex = i,
  65                      /* Vulkan so far doesn't have an instance divisor, so
  66                       * this is always 1 (ignored if not instancing). */
  67                      .InstanceDataStepRate = 1);
  68    }
  69
  70    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_SGVS,
  71                   .VertexIDEnable = pipeline->vs_prog_data.uses_vertexid,
  72                   .VertexIDComponentNumber = 2,
  73                   .VertexIDElementOffset = info->bindingCount,
  74                   .InstanceIDEnable = pipeline->vs_prog_data.uses_instanceid,
  75                   .InstanceIDComponentNumber = 3,
  76                   .InstanceIDElementOffset = info->bindingCount);
  77 }
  78
  79 static void
  80 emit_ia_state(struct anv_pipeline *pipeline,
  81               const VkPipelineInputAssemblyStateCreateInfo *info,
  82               const struct anv_graphics_pipeline_create_info *extra)
  83 {
  84    struct GEN8_3DSTATE_VF vf = {
  85       GEN8_3DSTATE_VF_header,
  86       .IndexedDrawCutIndexEnable = pipeline->primitive_restart
  87    };
  88    GEN8_3DSTATE_VF_pack(NULL, pipeline->gen8.vf, &vf);
  89
  90    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_TOPOLOGY,
  91                   .PrimitiveTopologyType = pipeline->topology);
  92 }
  93
  94 static void
  95 emit_rs_state(struct anv_pipeline *pipeline,
  96               const VkPipelineRasterStateCreateInfo *info,
  97               const struct anv_graphics_pipeline_create_info *extra)
  98 {
  99    static const uint32_t vk_to_gen_cullmode[] = {
 100       [VK_CULL_MODE_NONE]                       = CULLMODE_NONE,
 101       [VK_CULL_MODE_FRONT]                      = CULLMODE_FRONT,
 102       [VK_CULL_MODE_BACK]                       = CULLMODE_BACK,
 103       [VK_CULL_MODE_FRONT_AND_BACK]             = CULLMODE_BOTH
 104    };
 105
 106    static const uint32_t vk_to_gen_fillmode[] = {
 107       [VK_FILL_MODE_POINTS]                     = RASTER_POINT,
 108       [VK_FILL_MODE_WIREFRAME]                  = RASTER_WIREFRAME,
 109       [VK_FILL_MODE_SOLID]                      = RASTER_SOLID
 110    };
 111
 112    static const uint32_t vk_to_gen_front_face[] = {
 113       [VK_FRONT_FACE_CCW]                       = CounterClockwise,
 114       [VK_FRONT_FACE_CW]                        = Clockwise
 115    };
 116
 117    struct GEN8_3DSTATE_SF sf = {
 118       GEN8_3DSTATE_SF_header,
 119       .ViewportTransformEnable = !(extra && extra->disable_viewport),
 120       .TriangleStripListProvokingVertexSelect = 0,
 121       .LineStripListProvokingVertexSelect = 0,
 122       .TriangleFanProvokingVertexSelect = 0,
 123       .PointWidthSource = pipeline->writes_point_size ? Vertex : State,
 124       .PointWidth = 1.0,
 125    };
 126
 127    /* FINISHME: VkBool32 rasterizerDiscardEnable; */
 128
 129    GEN8_3DSTATE_SF_pack(NULL, pipeline->gen8.sf, &sf);
 130
 131    struct GEN8_3DSTATE_RASTER raster = {
 132       GEN8_3DSTATE_RASTER_header,
 133       .FrontWinding = vk_to_gen_front_face[info->frontFace],
 134       .CullMode = vk_to_gen_cullmode[info->cullMode],
 135       .FrontFaceFillMode = vk_to_gen_fillmode[info->fillMode],
 136       .BackFaceFillMode = vk_to_gen_fillmode[info->fillMode],
 137       .ScissorRectangleEnable = !(extra && extra->disable_scissor),
 138       .ViewportZClipTestEnable = info->depthClipEnable
 139    };
 140
 141    GEN8_3DSTATE_RASTER_pack(NULL, pipeline->gen8.raster, &raster);
 142 }
 143
 144 static void
 145 emit_cb_state(struct anv_pipeline *pipeline,
 146               const VkPipelineColorBlendStateCreateInfo *info)
 147 {
 148    struct anv_device *device = pipeline->device;
 149
 150    static const uint32_t vk_to_gen_logic_op[] = {
 151       [VK_LOGIC_OP_COPY]                        = LOGICOP_COPY,
 152       [VK_LOGIC_OP_CLEAR]                       = LOGICOP_CLEAR,
 153       [VK_LOGIC_OP_AND]                         = LOGICOP_AND,
 154       [VK_LOGIC_OP_AND_REVERSE]                 = LOGICOP_AND_REVERSE,
 155       [VK_LOGIC_OP_AND_INVERTED]                = LOGICOP_AND_INVERTED,
 156       [VK_LOGIC_OP_NOOP]                        = LOGICOP_NOOP,
 157       [VK_LOGIC_OP_XOR]                         = LOGICOP_XOR,
 158       [VK_LOGIC_OP_OR]                          = LOGICOP_OR,
 159       [VK_LOGIC_OP_NOR]                         = LOGICOP_NOR,
 160       [VK_LOGIC_OP_EQUIV]                       = LOGICOP_EQUIV,
 161       [VK_LOGIC_OP_INVERT]                      = LOGICOP_INVERT,
 162       [VK_LOGIC_OP_OR_REVERSE]                  = LOGICOP_OR_REVERSE,
 163       [VK_LOGIC_OP_COPY_INVERTED]               = LOGICOP_COPY_INVERTED,
 164       [VK_LOGIC_OP_OR_INVERTED]                 = LOGICOP_OR_INVERTED,
 165       [VK_LOGIC_OP_NAND]                        = LOGICOP_NAND,
 166       [VK_LOGIC_OP_SET]                         = LOGICOP_SET,
 167    };
 168
 169    static const uint32_t vk_to_gen_blend[] = {
 170       [VK_BLEND_ZERO]                           = BLENDFACTOR_ZERO,
 171       [VK_BLEND_ONE]                            = BLENDFACTOR_ONE,
 172       [VK_BLEND_SRC_COLOR]                      = BLENDFACTOR_SRC_COLOR,
 173       [VK_BLEND_ONE_MINUS_SRC_COLOR]            = BLENDFACTOR_INV_SRC_COLOR,
 174       [VK_BLEND_DEST_COLOR]                     = BLENDFACTOR_DST_COLOR,
 175       [VK_BLEND_ONE_MINUS_DEST_COLOR]           = BLENDFACTOR_INV_DST_COLOR,
 176       [VK_BLEND_SRC_ALPHA]                      = BLENDFACTOR_SRC_ALPHA,
 177       [VK_BLEND_ONE_MINUS_SRC_ALPHA]            = BLENDFACTOR_INV_SRC_ALPHA,
 178       [VK_BLEND_DEST_ALPHA]                     = BLENDFACTOR_DST_ALPHA,
 179       [VK_BLEND_ONE_MINUS_DEST_ALPHA]           = BLENDFACTOR_INV_DST_ALPHA,
 180       [VK_BLEND_CONSTANT_COLOR]                 = BLENDFACTOR_CONST_COLOR,
 181       [VK_BLEND_ONE_MINUS_CONSTANT_COLOR]       = BLENDFACTOR_INV_CONST_COLOR,
 182       [VK_BLEND_CONSTANT_ALPHA]                 = BLENDFACTOR_CONST_ALPHA,
 183       [VK_BLEND_ONE_MINUS_CONSTANT_ALPHA]       = BLENDFACTOR_INV_CONST_ALPHA,
 184       [VK_BLEND_SRC_ALPHA_SATURATE]             = BLENDFACTOR_SRC_ALPHA_SATURATE,
 185       [VK_BLEND_SRC1_COLOR]                     = BLENDFACTOR_SRC1_COLOR,
 186       [VK_BLEND_ONE_MINUS_SRC1_COLOR]           = BLENDFACTOR_INV_SRC1_COLOR,
 187       [VK_BLEND_SRC1_ALPHA]                     = BLENDFACTOR_SRC1_ALPHA,
 188       [VK_BLEND_ONE_MINUS_SRC1_ALPHA]           = BLENDFACTOR_INV_SRC1_ALPHA,
 189    };
 190
 191    static const uint32_t vk_to_gen_blend_op[] = {
 192       [VK_BLEND_OP_ADD]                         = BLENDFUNCTION_ADD,
 193       [VK_BLEND_OP_SUBTRACT]                    = BLENDFUNCTION_SUBTRACT,
 194       [VK_BLEND_OP_REVERSE_SUBTRACT]            = BLENDFUNCTION_REVERSE_SUBTRACT,
 195       [VK_BLEND_OP_MIN]                         = BLENDFUNCTION_MIN,
 196       [VK_BLEND_OP_MAX]                         = BLENDFUNCTION_MAX,
 197    };
 198
 199    uint32_t num_dwords = GEN8_BLEND_STATE_length;
 200    pipeline->blend_state =
 201       anv_state_pool_alloc(&device->dynamic_state_pool, num_dwords * 4, 64);
 202
 203    struct GEN8_BLEND_STATE blend_state = {
 204       .AlphaToCoverageEnable = info->alphaToCoverageEnable,
 205    };
 206
 207    for (uint32_t i = 0; i < info->attachmentCount; i++) {
 208       const VkPipelineColorBlendAttachmentState *a = &info->pAttachments[i];
 209
 210       blend_state.Entry[i] = (struct GEN8_BLEND_STATE_ENTRY) {
 211          .LogicOpEnable = info->logicOpEnable,
 212          .LogicOpFunction = vk_to_gen_logic_op[info->logicOp],
 213          .ColorBufferBlendEnable = a->blendEnable,
 214          .PreBlendSourceOnlyClampEnable = false,
 215          .PreBlendColorClampEnable = false,
 216          .PostBlendColorClampEnable = false,
 217          .SourceBlendFactor = vk_to_gen_blend[a->srcBlendColor],
 218          .DestinationBlendFactor = vk_to_gen_blend[a->destBlendColor],
 219          .ColorBlendFunction = vk_to_gen_blend_op[a->blendOpColor],
 220          .SourceAlphaBlendFactor = vk_to_gen_blend[a->srcBlendAlpha],
 221          .DestinationAlphaBlendFactor = vk_to_gen_blend[a->destBlendAlpha],
 222          .AlphaBlendFunction = vk_to_gen_blend_op[a->blendOpAlpha],
 223          .WriteDisableAlpha = !(a->channelWriteMask & VK_CHANNEL_A_BIT),
 224          .WriteDisableRed = !(a->channelWriteMask & VK_CHANNEL_R_BIT),
 225          .WriteDisableGreen = !(a->channelWriteMask & VK_CHANNEL_G_BIT),
 226          .WriteDisableBlue = !(a->channelWriteMask & VK_CHANNEL_B_BIT),
 227       };
 228    }
 229
 230    GEN8_BLEND_STATE_pack(NULL, pipeline->blend_state.map, &blend_state);
 231
 232    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_BLEND_STATE_POINTERS,
 233                   .BlendStatePointer = pipeline->blend_state.offset,
 234                   .BlendStatePointerValid = true);
 235 }
 236
 237 static const uint32_t vk_to_gen_compare_op[] = {
 238    [VK_COMPARE_OP_NEVER]                        = COMPAREFUNCTION_NEVER,
 239    [VK_COMPARE_OP_LESS]                         = COMPAREFUNCTION_LESS,
 240    [VK_COMPARE_OP_EQUAL]                        = COMPAREFUNCTION_EQUAL,
 241    [VK_COMPARE_OP_LESS_EQUAL]                   = COMPAREFUNCTION_LEQUAL,
 242    [VK_COMPARE_OP_GREATER]                      = COMPAREFUNCTION_GREATER,
 243    [VK_COMPARE_OP_NOT_EQUAL]                    = COMPAREFUNCTION_NOTEQUAL,
 244    [VK_COMPARE_OP_GREATER_EQUAL]                = COMPAREFUNCTION_GEQUAL,
 245    [VK_COMPARE_OP_ALWAYS]                       = COMPAREFUNCTION_ALWAYS,
 246 };
 247
 248 static const uint32_t vk_to_gen_stencil_op[] = {
 249    [VK_STENCIL_OP_KEEP]                         = STENCILOP_KEEP,
 250    [VK_STENCIL_OP_ZERO]                         = STENCILOP_ZERO,
 251    [VK_STENCIL_OP_REPLACE]                      = STENCILOP_REPLACE,
 252    [VK_STENCIL_OP_INC_CLAMP]                    = STENCILOP_INCRSAT,
 253    [VK_STENCIL_OP_DEC_CLAMP]                    = STENCILOP_DECRSAT,
 254    [VK_STENCIL_OP_INVERT]                       = STENCILOP_INVERT,
 255    [VK_STENCIL_OP_INC_WRAP]                     = STENCILOP_INCR,
 256    [VK_STENCIL_OP_DEC_WRAP]                     = STENCILOP_DECR,
 257 };
 258
 259 static void
 260 emit_ds_state(struct anv_pipeline *pipeline,
 261               const VkPipelineDepthStencilStateCreateInfo *info)
 262 {
 263    if (info == NULL) {
 264       /* We're going to OR this together with the dynamic state.  We need
 265        * to make sure it's initialized to something useful.
 266        */
 267       memset(pipeline->gen8.wm_depth_stencil, 0,
 268              sizeof(pipeline->gen8.wm_depth_stencil));
 269       return;
 270    }
 271
 272    /* VkBool32 depthBoundsTestEnable; // optional (depth_bounds_test) */
 273
 274    struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
 275       .DepthTestEnable = info->depthTestEnable,
 276       .DepthBufferWriteEnable = info->depthWriteEnable,
 277       .DepthTestFunction = vk_to_gen_compare_op[info->depthCompareOp],
 278       .DoubleSidedStencilEnable = true,
 279
 280       .StencilTestEnable = info->stencilTestEnable,
 281       .StencilFailOp = vk_to_gen_stencil_op[info->front.stencilFailOp],
 282       .StencilPassDepthPassOp = vk_to_gen_stencil_op[info->front.stencilPassOp],
 283       .StencilPassDepthFailOp = vk_to_gen_stencil_op[info->front.stencilDepthFailOp],
 284       .StencilTestFunction = vk_to_gen_compare_op[info->front.stencilCompareOp],
 285       .BackfaceStencilFailOp = vk_to_gen_stencil_op[info->back.stencilFailOp],
 286       .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[info->back.stencilPassOp],
 287       .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[info->back.stencilDepthFailOp],
 288       .BackfaceStencilTestFunction = vk_to_gen_compare_op[info->back.stencilCompareOp],
 289    };
 290
 291    GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, pipeline->gen8.wm_depth_stencil, &wm_depth_stencil);
 292 }
 293
 294 VkResult
 295 gen8_graphics_pipeline_create(
 296     VkDevice                                    _device,
 297     const VkGraphicsPipelineCreateInfo*         pCreateInfo,
 298     const struct anv_graphics_pipeline_create_info *extra,
 299     VkPipeline*                                 pPipeline)
 300 {
 301    ANV_FROM_HANDLE(anv_device, device, _device);
 302    struct anv_pipeline *pipeline;
 303    VkResult result;
 304    uint32_t offset, length;
 305
 306    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
 307
 308    pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
 309                                VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
 310    if (pipeline == NULL)
 311       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 312
 313    result = anv_pipeline_init(pipeline, device, pCreateInfo, extra);
 314    if (result != VK_SUCCESS)
 315       return result;
 316
 317    /* FIXME: The compiler dead-codes FS inputs when we don't have a VS, so we
 318     * hard code this to num_attributes - 2. This is because the attributes
 319     * include VUE header and position, which aren't counted as varying
 320     * inputs. */
 321    if (pipeline->vs_simd8 == NO_KERNEL) {
 322       pipeline->wm_prog_data.num_varying_inputs =
 323          pCreateInfo->pVertexInputState->attributeCount - 2;
 324    }
 325
 326    assert(pCreateInfo->pVertexInputState);
 327    emit_vertex_input(pipeline, pCreateInfo->pVertexInputState);
 328    assert(pCreateInfo->pInputAssemblyState);
 329    emit_ia_state(pipeline, pCreateInfo->pInputAssemblyState, extra);
 330    assert(pCreateInfo->pRasterState);
 331    emit_rs_state(pipeline, pCreateInfo->pRasterState, extra);
 332    emit_ds_state(pipeline, pCreateInfo->pDepthStencilState);
 333    emit_cb_state(pipeline, pCreateInfo->pColorBlendState);
 334
 335    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_STATISTICS,
 336                    .StatisticsEnable = true);
 337    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_HS, .Enable = false);
 338    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_TE, .TEEnable = false);
 339    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_DS, .FunctionEnable = false);
 340    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_STREAMOUT, .SOFunctionEnable = false);
 341
 342    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS,
 343                   .ConstantBufferOffset = 0,
 344                   .ConstantBufferSize = 4);
 345    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS,
 346                   .ConstantBufferOffset = 4,
 347                   .ConstantBufferSize = 4);
 348    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS,
 349                   .ConstantBufferOffset = 8,
 350                   .ConstantBufferSize = 4);
 351
 352    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM_CHROMAKEY,
 353                   .ChromaKeyKillEnable = false);
 354    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_AA_LINE_PARAMETERS);
 355
 356    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_CLIP,
 357                   .ClipEnable = true,
 358                   .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
 359                   .MinimumPointWidth = 0.125,
 360                   .MaximumPointWidth = 255.875);
 361
 362    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM,
 363                   .StatisticsEnable = true,
 364                   .LineEndCapAntialiasingRegionWidth = _05pixels,
 365                   .LineAntialiasingRegionWidth = _10pixels,
 366                   .EarlyDepthStencilControl = NORMAL,
 367                   .ForceThreadDispatchEnable = NORMAL,
 368                   .PointRasterizationRule = RASTRULE_UPPER_RIGHT,
 369                   .BarycentricInterpolationMode =
 370                      pipeline->wm_prog_data.barycentric_interp_modes);
 371
 372    uint32_t samples = 1;
 373    uint32_t log2_samples = __builtin_ffs(samples) - 1;
 374    bool enable_sampling = samples > 1 ? true : false;
 375
 376    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_MULTISAMPLE,
 377                   .PixelPositionOffsetEnable = enable_sampling,
 378                   .PixelLocation = CENTER,
 379                   .NumberofMultisamples = log2_samples);
 380
 381    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SAMPLE_MASK,
 382                   .SampleMask = 0xffff);
 383
 384    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_VS,
 385                   .VSURBStartingAddress = pipeline->urb.vs_start,
 386                   .VSURBEntryAllocationSize = pipeline->urb.vs_size - 1,
 387                   .VSNumberofURBEntries = pipeline->urb.nr_vs_entries);
 388
 389    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_GS,
 390                   .GSURBStartingAddress = pipeline->urb.gs_start,
 391                   .GSURBEntryAllocationSize = pipeline->urb.gs_size - 1,
 392                   .GSNumberofURBEntries = pipeline->urb.nr_gs_entries);
 393
 394    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_HS,
 395                   .HSURBStartingAddress = pipeline->urb.vs_start,
 396                   .HSURBEntryAllocationSize = 0,
 397                   .HSNumberofURBEntries = 0);
 398
 399    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_DS,
 400                   .DSURBStartingAddress = pipeline->urb.vs_start,
 401                   .DSURBEntryAllocationSize = 0,
 402                   .DSNumberofURBEntries = 0);
 403
 404    const struct brw_gs_prog_data *gs_prog_data = &pipeline->gs_prog_data;
 405    offset = 1;
 406    length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
 407
 408    if (pipeline->gs_vec4 == NO_KERNEL)
 409       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS, .Enable = false);
 410    else
 411       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS,
 412                      .SingleProgramFlow = false,
 413                      .KernelStartPointer = pipeline->gs_vec4,
 414                      .VectorMaskEnable = Dmask,
 415                      .SamplerCount = 0,
 416                      .BindingTableEntryCount = 0,
 417                      .ExpectedVertexCount = pipeline->gs_vertex_count,
 418
 419                      .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_GEOMETRY],
 420                      .PerThreadScratchSpace = ffs(gs_prog_data->base.base.total_scratch / 2048),
 421
 422                      .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
 423                      .OutputTopology = gs_prog_data->output_topology,
 424                      .VertexURBEntryReadLength = gs_prog_data->base.urb_read_length,
 425                      .DispatchGRFStartRegisterForURBData =
 426                         gs_prog_data->base.base.dispatch_grf_start_reg,
 427
 428                      .MaximumNumberofThreads = device->info.max_gs_threads / 2 - 1,
 429                      .ControlDataHeaderSize = gs_prog_data->control_data_header_size_hwords,
 430                      .DispatchMode = gs_prog_data->base.dispatch_mode,
 431                      .StatisticsEnable = true,
 432                      .IncludePrimitiveID = gs_prog_data->include_primitive_id,
 433                      .ReorderMode = TRAILING,
 434                      .Enable = true,
 435
 436                      .ControlDataFormat = gs_prog_data->control_data_format,
 437
 438                      .StaticOutput = gs_prog_data->static_vertex_count >= 0,
 439                      .StaticOutputVertexCount =
 440                         gs_prog_data->static_vertex_count >= 0 ?
 441                         gs_prog_data->static_vertex_count : 0,
 442
 443                      /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
 444                       * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
 445                       * UserClipDistanceCullTestEnableBitmask(v)
 446                       */
 447
 448                      .VertexURBEntryOutputReadOffset = offset,
 449                      .VertexURBEntryOutputLength = length);
 450
 451    const struct brw_vue_prog_data *vue_prog_data = &pipeline->vs_prog_data.base;
 452    /* Skip the VUE header and position slots */
 453    offset = 1;
 454    length = (vue_prog_data->vue_map.num_slots + 1) / 2 - offset;
 455
 456    if (pipeline->vs_simd8 == NO_KERNEL || (extra && extra->disable_vs))
 457       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
 458                      .FunctionEnable = false,
 459                      /* Even if VS is disabled, SBE still gets the amount of
 460                       * vertex data to read from this field. */
 461                      .VertexURBEntryOutputReadOffset = offset,
 462                      .VertexURBEntryOutputLength = length);
 463    else
 464       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
 465                      .KernelStartPointer = pipeline->vs_simd8,
 466                      .SingleVertexDispatch = Multiple,
 467                      .VectorMaskEnable = Dmask,
 468                      .SamplerCount = 0,
 469                      .BindingTableEntryCount =
 470                      vue_prog_data->base.binding_table.size_bytes / 4,
 471                      .ThreadDispatchPriority = Normal,
 472                      .FloatingPointMode = IEEE754,
 473                      .IllegalOpcodeExceptionEnable = false,
 474                      .AccessesUAV = false,
 475                      .SoftwareExceptionEnable = false,
 476
 477                      .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_VERTEX],
 478                      .PerThreadScratchSpace = ffs(vue_prog_data->base.total_scratch / 2048),
 479
 480                      .DispatchGRFStartRegisterForURBData =
 481                      vue_prog_data->base.dispatch_grf_start_reg,
 482                      .VertexURBEntryReadLength = vue_prog_data->urb_read_length,
 483                      .VertexURBEntryReadOffset = 0,
 484
 485                      .MaximumNumberofThreads = device->info.max_vs_threads - 1,
 486                      .StatisticsEnable = false,
 487                      .SIMD8DispatchEnable = true,
 488                      .VertexCacheDisable = false,
 489                      .FunctionEnable = true,
 490
 491                      .VertexURBEntryOutputReadOffset = offset,
 492                      .VertexURBEntryOutputLength = length,
 493                      .UserClipDistanceClipTestEnableBitmask = 0,
 494                      .UserClipDistanceCullTestEnableBitmask = 0);
 495
 496    const struct brw_wm_prog_data *wm_prog_data = &pipeline->wm_prog_data;
 497
 498    /* TODO: We should clean this up.  Among other things, this is mostly
 499     * shared with other gens.
 500     */
 501    const struct brw_vue_map *fs_input_map;
 502    if (pipeline->gs_vec4 == NO_KERNEL)
 503       fs_input_map = &vue_prog_data->vue_map;
 504    else
 505       fs_input_map = &gs_prog_data->base.vue_map;
 506
 507    struct GEN8_3DSTATE_SBE_SWIZ swiz = {
 508       GEN8_3DSTATE_SBE_SWIZ_header,
 509    };
 510
 511    int max_source_attr = 0;
 512    for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 513       int input_index = wm_prog_data->urb_setup[attr];
 514
 515       if (input_index < 0)
 516          continue;
 517
 518       /* We have to subtract two slots to accout for the URB entry output
 519        * read offset in the VS and GS stages.
 520        */
 521       int source_attr = fs_input_map->varying_to_slot[attr] - 2;
 522       max_source_attr = MAX2(max_source_attr, source_attr);
 523
 524       if (input_index >= 16)
 525          continue;
 526
 527       swiz.Attribute[input_index].SourceAttribute = source_attr;
 528    }
 529
 530    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE,
 531                   .AttributeSwizzleEnable = true,
 532                   .ForceVertexURBEntryReadLength = false,
 533                   .ForceVertexURBEntryReadOffset = false,
 534                   .VertexURBEntryReadLength = DIV_ROUND_UP(max_source_attr + 1, 2),
 535                   .PointSpriteTextureCoordinateOrigin = UPPERLEFT,
 536                   .NumberofSFOutputAttributes =
 537                      wm_prog_data->num_varying_inputs);
 538
 539    uint32_t *dw = anv_batch_emit_dwords(&pipeline->batch,
 540                                         GEN8_3DSTATE_SBE_SWIZ_length);
 541    GEN8_3DSTATE_SBE_SWIZ_pack(&pipeline->batch, dw, &swiz);
 542
 543    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS,
 544                   .KernelStartPointer0 = pipeline->ps_ksp0,
 545
 546                   .SingleProgramFlow = false,
 547                   .VectorMaskEnable = true,
 548                   .SamplerCount = 1,
 549
 550                   .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
 551                   .PerThreadScratchSpace = ffs(wm_prog_data->base.total_scratch / 2048),
 552
 553                   .MaximumNumberofThreadsPerPSD = 64 - 2,
 554                   .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
 555                      POSOFFSET_SAMPLE: POSOFFSET_NONE,
 556                   .PushConstantEnable = wm_prog_data->base.nr_params > 0,
 557                   ._8PixelDispatchEnable = pipeline->ps_simd8 != NO_KERNEL,
 558                   ._16PixelDispatchEnable = pipeline->ps_simd16 != NO_KERNEL,
 559                   ._32PixelDispatchEnable = false,
 560
 561                   .DispatchGRFStartRegisterForConstantSetupData0 = pipeline->ps_grf_start0,
 562                   .DispatchGRFStartRegisterForConstantSetupData1 = 0,
 563                   .DispatchGRFStartRegisterForConstantSetupData2 = pipeline->ps_grf_start2,
 564
 565                   .KernelStartPointer1 = 0,
 566                   .KernelStartPointer2 = pipeline->ps_ksp2);
 567
 568    bool per_sample_ps = false;
 569    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS_EXTRA,
 570                   .PixelShaderValid = true,
 571                   .PixelShaderKillsPixel = wm_prog_data->uses_kill,
 572                   .PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode,
 573                   .AttributeEnable = wm_prog_data->num_varying_inputs > 0,
 574                   .oMaskPresenttoRenderTarget = wm_prog_data->uses_omask,
 575                   .PixelShaderIsPerSample = per_sample_ps);
 576
 577    *pPipeline = anv_pipeline_to_handle(pipeline);
 578
 579    return VK_SUCCESS;
 580 }
 581
 582 VkResult gen8_compute_pipeline_create(
 583     VkDevice                                    _device,
 584     const VkComputePipelineCreateInfo*          pCreateInfo,
 585     VkPipeline*                                 pPipeline)
 586 {
 587    ANV_FROM_HANDLE(anv_device, device, _device);
 588    struct anv_pipeline *pipeline;
 589    VkResult result;
 590
 591    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
 592
 593    pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
 594                                VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
 595    if (pipeline == NULL)
 596       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 597
 598    pipeline->device = device;
 599    pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
 600
 601    pipeline->blend_state.map = NULL;
 602
 603    result = anv_reloc_list_init(&pipeline->batch_relocs, device);
 604    if (result != VK_SUCCESS) {
 605       anv_device_free(device, pipeline);
 606       return result;
 607    }
 608    pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
 609    pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
 610    pipeline->batch.relocs = &pipeline->batch_relocs;
 611
 612    anv_state_stream_init(&pipeline->program_stream,
 613                          &device->instruction_block_pool);
 614
 615    /* When we free the pipeline, we detect stages based on the NULL status
 616     * of various prog_data pointers.  Make them NULL by default.
 617     */
 618    memset(pipeline->prog_data, 0, sizeof(pipeline->prog_data));
 619    memset(pipeline->scratch_start, 0, sizeof(pipeline->scratch_start));
 620
 621    pipeline->vs_simd8 = NO_KERNEL;
 622    pipeline->vs_vec4 = NO_KERNEL;
 623    pipeline->gs_vec4 = NO_KERNEL;
 624
 625    pipeline->active_stages = 0;
 626    pipeline->total_scratch = 0;
 627
 628    assert(pCreateInfo->stage.stage == VK_SHADER_STAGE_COMPUTE);
 629    ANV_FROM_HANDLE(anv_shader, shader,  pCreateInfo->stage.shader);
 630    anv_pipeline_compile_cs(pipeline, pCreateInfo, shader);
 631
 632    pipeline->use_repclear = false;
 633
 634    const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data;
 635
 636    anv_batch_emit(&pipeline->batch, GEN8_MEDIA_VFE_STATE,
 637                   .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_COMPUTE],
 638                   .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
 639                   .ScratchSpaceBasePointerHigh = 0,
 640                   .StackSize = 0,
 641
 642                   .MaximumNumberofThreads = device->info.max_cs_threads - 1,
 643                   .NumberofURBEntries = 2,
 644                   .ResetGatewayTimer = true,
 645                   .BypassGatewayControl = true,
 646                   .URBEntryAllocationSize = 2,
 647                   .CURBEAllocationSize = 0);
 648
 649    struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
 650    uint32_t group_size = prog_data->local_size[0] *
 651       prog_data->local_size[1] * prog_data->local_size[2];
 652    pipeline->cs_thread_width_max = DIV_ROUND_UP(group_size, prog_data->simd_size);
 653    uint32_t remainder = group_size & (prog_data->simd_size - 1);
 654
 655    if (remainder > 0)
 656       pipeline->cs_right_mask = ~0u >> (32 - remainder);
 657    else
 658       pipeline->cs_right_mask = ~0u >> (32 - prog_data->simd_size);
 659
 660
 661    *pPipeline = anv_pipeline_to_handle(pipeline);
 662
 663    return VK_SUCCESS;
 664 }