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
  34 static void
  35 emit_vertex_input(struct anv_pipeline *pipeline,
  36                   const VkPipelineVertexInputStateCreateInfo *info)
  37 {
  38    const uint32_t num_dwords = 1 + info->attributeCount * 2;
  39    uint32_t *p;
  40
  41    if (info->attributeCount > 0) {
  42       p = anv_batch_emitn(&pipeline->batch, num_dwords,
  43                           GEN8_3DSTATE_VERTEX_ELEMENTS);
  44    }
  45
  46    for (uint32_t i = 0; i < info->attributeCount; i++) {
  47       const VkVertexInputAttributeDescription *desc =
  48          &info->pVertexAttributeDescriptions[i];
  49       const struct anv_format *format = anv_format_for_vk_format(desc->format);
  50
  51       struct GEN8_VERTEX_ELEMENT_STATE element = {
  52          .VertexBufferIndex = desc->binding,
  53          .Valid = true,
  54          .SourceElementFormat = format->surface_format,
  55          .EdgeFlagEnable = false,
  56          .SourceElementOffset = desc->offsetInBytes,
  57          .Component0Control = VFCOMP_STORE_SRC,
  58          .Component1Control = format->num_channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
  59          .Component2Control = format->num_channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
  60          .Component3Control = format->num_channels >= 4 ? VFCOMP_STORE_SRC : VFCOMP_STORE_1_FP
  61       };
  62       GEN8_VERTEX_ELEMENT_STATE_pack(NULL, &p[1 + i * 2], &element);
  63
  64       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_INSTANCING,
  65                      .InstancingEnable = pipeline->instancing_enable[desc->binding],
  66                      .VertexElementIndex = i,
  67                      /* Vulkan so far doesn't have an instance divisor, so
  68                       * this is always 1 (ignored if not instancing). */
  69                      .InstanceDataStepRate = 1);
  70    }
  71
  72    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_SGVS,
  73                   .VertexIDEnable = pipeline->vs_prog_data.uses_vertexid,
  74                   .VertexIDComponentNumber = 2,
  75                   .VertexIDElementOffset = info->bindingCount,
  76                   .InstanceIDEnable = pipeline->vs_prog_data.uses_instanceid,
  77                   .InstanceIDComponentNumber = 3,
  78                   .InstanceIDElementOffset = info->bindingCount);
  79 }
  80
  81 static void
  82 emit_ia_state(struct anv_pipeline *pipeline,
  83               const VkPipelineInputAssemblyStateCreateInfo *info,
  84               const struct anv_graphics_pipeline_create_info *extra)
  85 {
  86    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_TOPOLOGY,
  87                   .PrimitiveTopologyType = pipeline->topology);
  88 }
  89
  90 static void
  91 emit_rs_state(struct anv_pipeline *pipeline,
  92               const VkPipelineRasterStateCreateInfo *info,
  93               const struct anv_graphics_pipeline_create_info *extra)
  94 {
  95    static const uint32_t vk_to_gen_cullmode[] = {
  96       [VK_CULL_MODE_NONE]                       = CULLMODE_NONE,
  97       [VK_CULL_MODE_FRONT]                      = CULLMODE_FRONT,
  98       [VK_CULL_MODE_BACK]                       = CULLMODE_BACK,
  99       [VK_CULL_MODE_FRONT_AND_BACK]             = CULLMODE_BOTH
 100    };
 101
 102    static const uint32_t vk_to_gen_fillmode[] = {
 103       [VK_FILL_MODE_POINTS]                     = RASTER_POINT,
 104       [VK_FILL_MODE_WIREFRAME]                  = RASTER_WIREFRAME,
 105       [VK_FILL_MODE_SOLID]                      = RASTER_SOLID
 106    };
 107
 108    static const uint32_t vk_to_gen_front_face[] = {
 109       [VK_FRONT_FACE_CCW]                       = CounterClockwise,
 110       [VK_FRONT_FACE_CW]                        = Clockwise
 111    };
 112
 113    struct GEN8_3DSTATE_SF sf = {
 114       GEN8_3DSTATE_SF_header,
 115       .ViewportTransformEnable = !(extra && extra->disable_viewport),
 116       .TriangleStripListProvokingVertexSelect = 0,
 117       .LineStripListProvokingVertexSelect = 0,
 118       .TriangleFanProvokingVertexSelect = 0,
 119       .PointWidthSource = pipeline->writes_point_size ? Vertex : State,
 120       .PointWidth = 1.0,
 121    };
 122
 123    /* FINISHME: VkBool32 rasterizerDiscardEnable; */
 124
 125    GEN8_3DSTATE_SF_pack(NULL, pipeline->gen8.sf, &sf);
 126
 127    struct GEN8_3DSTATE_RASTER raster = {
 128       GEN8_3DSTATE_RASTER_header,
 129       .FrontWinding = vk_to_gen_front_face[info->frontFace],
 130       .CullMode = vk_to_gen_cullmode[info->cullMode],
 131       .FrontFaceFillMode = vk_to_gen_fillmode[info->fillMode],
 132       .BackFaceFillMode = vk_to_gen_fillmode[info->fillMode],
 133       .ScissorRectangleEnable = !(extra && extra->disable_scissor),
 134       .ViewportZClipTestEnable = info->depthClipEnable
 135    };
 136
 137    GEN8_3DSTATE_RASTER_pack(NULL, pipeline->gen8.raster, &raster);
 138 }
 139
 140 static void
 141 emit_cb_state(struct anv_pipeline *pipeline,
 142               const VkPipelineColorBlendStateCreateInfo *info)
 143 {
 144    struct anv_device *device = pipeline->device;
 145
 146    static const uint32_t vk_to_gen_logic_op[] = {
 147       [VK_LOGIC_OP_COPY]                        = LOGICOP_COPY,
 148       [VK_LOGIC_OP_CLEAR]                       = LOGICOP_CLEAR,
 149       [VK_LOGIC_OP_AND]                         = LOGICOP_AND,
 150       [VK_LOGIC_OP_AND_REVERSE]                 = LOGICOP_AND_REVERSE,
 151       [VK_LOGIC_OP_AND_INVERTED]                = LOGICOP_AND_INVERTED,
 152       [VK_LOGIC_OP_NOOP]                        = LOGICOP_NOOP,
 153       [VK_LOGIC_OP_XOR]                         = LOGICOP_XOR,
 154       [VK_LOGIC_OP_OR]                          = LOGICOP_OR,
 155       [VK_LOGIC_OP_NOR]                         = LOGICOP_NOR,
 156       [VK_LOGIC_OP_EQUIV]                       = LOGICOP_EQUIV,
 157       [VK_LOGIC_OP_INVERT]                      = LOGICOP_INVERT,
 158       [VK_LOGIC_OP_OR_REVERSE]                  = LOGICOP_OR_REVERSE,
 159       [VK_LOGIC_OP_COPY_INVERTED]               = LOGICOP_COPY_INVERTED,
 160       [VK_LOGIC_OP_OR_INVERTED]                 = LOGICOP_OR_INVERTED,
 161       [VK_LOGIC_OP_NAND]                        = LOGICOP_NAND,
 162       [VK_LOGIC_OP_SET]                         = LOGICOP_SET,
 163    };
 164
 165    static const uint32_t vk_to_gen_blend[] = {
 166       [VK_BLEND_ZERO]                           = BLENDFACTOR_ZERO,
 167       [VK_BLEND_ONE]                            = BLENDFACTOR_ONE,
 168       [VK_BLEND_SRC_COLOR]                      = BLENDFACTOR_SRC_COLOR,
 169       [VK_BLEND_ONE_MINUS_SRC_COLOR]            = BLENDFACTOR_INV_SRC_COLOR,
 170       [VK_BLEND_DEST_COLOR]                     = BLENDFACTOR_DST_COLOR,
 171       [VK_BLEND_ONE_MINUS_DEST_COLOR]           = BLENDFACTOR_INV_DST_COLOR,
 172       [VK_BLEND_SRC_ALPHA]                      = BLENDFACTOR_SRC_ALPHA,
 173       [VK_BLEND_ONE_MINUS_SRC_ALPHA]            = BLENDFACTOR_INV_SRC_ALPHA,
 174       [VK_BLEND_DEST_ALPHA]                     = BLENDFACTOR_DST_ALPHA,
 175       [VK_BLEND_ONE_MINUS_DEST_ALPHA]           = BLENDFACTOR_INV_DST_ALPHA,
 176       [VK_BLEND_CONSTANT_COLOR]                 = BLENDFACTOR_CONST_COLOR,
 177       [VK_BLEND_ONE_MINUS_CONSTANT_COLOR]       = BLENDFACTOR_INV_CONST_COLOR,
 178       [VK_BLEND_CONSTANT_ALPHA]                 = BLENDFACTOR_CONST_ALPHA,
 179       [VK_BLEND_ONE_MINUS_CONSTANT_ALPHA]       = BLENDFACTOR_INV_CONST_ALPHA,
 180       [VK_BLEND_SRC_ALPHA_SATURATE]             = BLENDFACTOR_SRC_ALPHA_SATURATE,
 181       [VK_BLEND_SRC1_COLOR]                     = BLENDFACTOR_SRC1_COLOR,
 182       [VK_BLEND_ONE_MINUS_SRC1_COLOR]           = BLENDFACTOR_INV_SRC1_COLOR,
 183       [VK_BLEND_SRC1_ALPHA]                     = BLENDFACTOR_SRC1_ALPHA,
 184       [VK_BLEND_ONE_MINUS_SRC1_ALPHA]           = BLENDFACTOR_INV_SRC1_ALPHA,
 185    };
 186
 187    static const uint32_t vk_to_gen_blend_op[] = {
 188       [VK_BLEND_OP_ADD]                         = BLENDFUNCTION_ADD,
 189       [VK_BLEND_OP_SUBTRACT]                    = BLENDFUNCTION_SUBTRACT,
 190       [VK_BLEND_OP_REVERSE_SUBTRACT]            = BLENDFUNCTION_REVERSE_SUBTRACT,
 191       [VK_BLEND_OP_MIN]                         = BLENDFUNCTION_MIN,
 192       [VK_BLEND_OP_MAX]                         = BLENDFUNCTION_MAX,
 193    };
 194
 195    uint32_t num_dwords = GEN8_BLEND_STATE_length;
 196    pipeline->blend_state =
 197       anv_state_pool_alloc(&device->dynamic_state_pool, num_dwords * 4, 64);
 198
 199    struct GEN8_BLEND_STATE blend_state = {
 200       .AlphaToCoverageEnable = info->alphaToCoverageEnable,
 201       .AlphaToOneEnable = info->alphaToOneEnable,
 202    };
 203
 204    for (uint32_t i = 0; i < info->attachmentCount; i++) {
 205       const VkPipelineColorBlendAttachmentState *a = &info->pAttachments[i];
 206
 207       if (a->srcBlendColor != a->srcBlendAlpha ||
 208           a->destBlendColor != a->destBlendAlpha ||
 209           a->blendOpColor != a->blendOpAlpha) {
 210          blend_state.IndependentAlphaBlendEnable = true;
 211       }
 212
 213       blend_state.Entry[i] = (struct GEN8_BLEND_STATE_ENTRY) {
 214          .LogicOpEnable = info->logicOpEnable,
 215          .LogicOpFunction = vk_to_gen_logic_op[info->logicOp],
 216          .ColorBufferBlendEnable = a->blendEnable,
 217          .PreBlendSourceOnlyClampEnable = false,
 218          .ColorClampRange = COLORCLAMP_RTFORMAT,
 219          .PreBlendColorClampEnable = true,
 220          .PostBlendColorClampEnable = true,
 221          .SourceBlendFactor = vk_to_gen_blend[a->srcBlendColor],
 222          .DestinationBlendFactor = vk_to_gen_blend[a->destBlendColor],
 223          .ColorBlendFunction = vk_to_gen_blend_op[a->blendOpColor],
 224          .SourceAlphaBlendFactor = vk_to_gen_blend[a->srcBlendAlpha],
 225          .DestinationAlphaBlendFactor = vk_to_gen_blend[a->destBlendAlpha],
 226          .AlphaBlendFunction = vk_to_gen_blend_op[a->blendOpAlpha],
 227          .WriteDisableAlpha = !(a->channelWriteMask & VK_CHANNEL_A_BIT),
 228          .WriteDisableRed = !(a->channelWriteMask & VK_CHANNEL_R_BIT),
 229          .WriteDisableGreen = !(a->channelWriteMask & VK_CHANNEL_G_BIT),
 230          .WriteDisableBlue = !(a->channelWriteMask & VK_CHANNEL_B_BIT),
 231       };
 232
 233       /* Our hardware applies the blend factor prior to the blend function
 234        * regardless of what function is used.  Technically, this means the
 235        * hardware can do MORE than GL or Vulkan specify.  However, it also
 236        * means that, for MIN and MAX, we have to stomp the blend factor to
 237        * ONE to make it a no-op.
 238        */
 239       if (a->blendOpColor == VK_BLEND_OP_MIN ||
 240           a->blendOpColor == VK_BLEND_OP_MAX) {
 241          blend_state.Entry[i].SourceBlendFactor = BLENDFACTOR_ONE;
 242          blend_state.Entry[i].DestinationBlendFactor = BLENDFACTOR_ONE;
 243       }
 244       if (a->blendOpAlpha == VK_BLEND_OP_MIN ||
 245           a->blendOpAlpha == VK_BLEND_OP_MAX) {
 246          blend_state.Entry[i].SourceAlphaBlendFactor = BLENDFACTOR_ONE;
 247          blend_state.Entry[i].DestinationAlphaBlendFactor = BLENDFACTOR_ONE;
 248       }
 249    }
 250
 251    GEN8_BLEND_STATE_pack(NULL, pipeline->blend_state.map, &blend_state);
 252
 253    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_BLEND_STATE_POINTERS,
 254                   .BlendStatePointer = pipeline->blend_state.offset,
 255                   .BlendStatePointerValid = true);
 256 }
 257
 258 static const uint32_t vk_to_gen_compare_op[] = {
 259    [VK_COMPARE_OP_NEVER]                        = COMPAREFUNCTION_NEVER,
 260    [VK_COMPARE_OP_LESS]                         = COMPAREFUNCTION_LESS,
 261    [VK_COMPARE_OP_EQUAL]                        = COMPAREFUNCTION_EQUAL,
 262    [VK_COMPARE_OP_LESS_EQUAL]                   = COMPAREFUNCTION_LEQUAL,
 263    [VK_COMPARE_OP_GREATER]                      = COMPAREFUNCTION_GREATER,
 264    [VK_COMPARE_OP_NOT_EQUAL]                    = COMPAREFUNCTION_NOTEQUAL,
 265    [VK_COMPARE_OP_GREATER_EQUAL]                = COMPAREFUNCTION_GEQUAL,
 266    [VK_COMPARE_OP_ALWAYS]                       = COMPAREFUNCTION_ALWAYS,
 267 };
 268
 269 static const uint32_t vk_to_gen_stencil_op[] = {
 270    [VK_STENCIL_OP_KEEP]                         = STENCILOP_KEEP,
 271    [VK_STENCIL_OP_ZERO]                         = STENCILOP_ZERO,
 272    [VK_STENCIL_OP_REPLACE]                      = STENCILOP_REPLACE,
 273    [VK_STENCIL_OP_INC_CLAMP]                    = STENCILOP_INCRSAT,
 274    [VK_STENCIL_OP_DEC_CLAMP]                    = STENCILOP_DECRSAT,
 275    [VK_STENCIL_OP_INVERT]                       = STENCILOP_INVERT,
 276    [VK_STENCIL_OP_INC_WRAP]                     = STENCILOP_INCR,
 277    [VK_STENCIL_OP_DEC_WRAP]                     = STENCILOP_DECR,
 278 };
 279
 280 static void
 281 emit_ds_state(struct anv_pipeline *pipeline,
 282               const VkPipelineDepthStencilStateCreateInfo *info)
 283 {
 284    if (info == NULL) {
 285       /* We're going to OR this together with the dynamic state.  We need
 286        * to make sure it's initialized to something useful.
 287        */
 288       memset(pipeline->gen8.wm_depth_stencil, 0,
 289              sizeof(pipeline->gen8.wm_depth_stencil));
 290       return;
 291    }
 292
 293    /* VkBool32 depthBoundsTestEnable; // optional (depth_bounds_test) */
 294
 295    struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
 296       .DepthTestEnable = info->depthTestEnable,
 297       .DepthBufferWriteEnable = info->depthWriteEnable,
 298       .DepthTestFunction = vk_to_gen_compare_op[info->depthCompareOp],
 299       .DoubleSidedStencilEnable = true,
 300
 301       .StencilTestEnable = info->stencilTestEnable,
 302       .StencilFailOp = vk_to_gen_stencil_op[info->front.stencilFailOp],
 303       .StencilPassDepthPassOp = vk_to_gen_stencil_op[info->front.stencilPassOp],
 304       .StencilPassDepthFailOp = vk_to_gen_stencil_op[info->front.stencilDepthFailOp],
 305       .StencilTestFunction = vk_to_gen_compare_op[info->front.stencilCompareOp],
 306       .BackfaceStencilFailOp = vk_to_gen_stencil_op[info->back.stencilFailOp],
 307       .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[info->back.stencilPassOp],
 308       .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[info->back.stencilDepthFailOp],
 309       .BackfaceStencilTestFunction = vk_to_gen_compare_op[info->back.stencilCompareOp],
 310    };
 311
 312    GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, pipeline->gen8.wm_depth_stencil, &wm_depth_stencil);
 313 }
 314
 315 VkResult
 316 gen8_graphics_pipeline_create(
 317     VkDevice                                    _device,
 318     const VkGraphicsPipelineCreateInfo*         pCreateInfo,
 319     const struct anv_graphics_pipeline_create_info *extra,
 320     VkPipeline*                                 pPipeline)
 321 {
 322    ANV_FROM_HANDLE(anv_device, device, _device);
 323    struct anv_pipeline *pipeline;
 324    VkResult result;
 325    uint32_t offset, length;
 326
 327    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
 328
 329    pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
 330                                VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
 331    if (pipeline == NULL)
 332       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 333
 334    result = anv_pipeline_init(pipeline, device, pCreateInfo, extra);
 335    if (result != VK_SUCCESS)
 336       return result;
 337
 338    /* FIXME: The compiler dead-codes FS inputs when we don't have a VS, so we
 339     * hard code this to num_attributes - 2. This is because the attributes
 340     * include VUE header and position, which aren't counted as varying
 341     * inputs. */
 342    if (pipeline->vs_simd8 == NO_KERNEL) {
 343       pipeline->wm_prog_data.num_varying_inputs =
 344          pCreateInfo->pVertexInputState->attributeCount - 2;
 345    }
 346
 347    assert(pCreateInfo->pVertexInputState);
 348    emit_vertex_input(pipeline, pCreateInfo->pVertexInputState);
 349    assert(pCreateInfo->pInputAssemblyState);
 350    emit_ia_state(pipeline, pCreateInfo->pInputAssemblyState, extra);
 351    assert(pCreateInfo->pRasterState);
 352    emit_rs_state(pipeline, pCreateInfo->pRasterState, extra);
 353    emit_ds_state(pipeline, pCreateInfo->pDepthStencilState);
 354    emit_cb_state(pipeline, pCreateInfo->pColorBlendState);
 355
 356    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_STATISTICS,
 357                    .StatisticsEnable = true);
 358    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_HS, .Enable = false);
 359    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_TE, .TEEnable = false);
 360    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_DS, .FunctionEnable = false);
 361    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_STREAMOUT, .SOFunctionEnable = false);
 362
 363    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS,
 364                   .ConstantBufferOffset = 0,
 365                   .ConstantBufferSize = 4);
 366    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS,
 367                   .ConstantBufferOffset = 4,
 368                   .ConstantBufferSize = 4);
 369    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS,
 370                   .ConstantBufferOffset = 8,
 371                   .ConstantBufferSize = 4);
 372
 373    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM_CHROMAKEY,
 374                   .ChromaKeyKillEnable = false);
 375    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_AA_LINE_PARAMETERS);
 376
 377    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_CLIP,
 378                   .ClipEnable = true,
 379                   .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
 380                   .MinimumPointWidth = 0.125,
 381                   .MaximumPointWidth = 255.875);
 382
 383    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM,
 384                   .StatisticsEnable = true,
 385                   .LineEndCapAntialiasingRegionWidth = _05pixels,
 386                   .LineAntialiasingRegionWidth = _10pixels,
 387                   .EarlyDepthStencilControl = NORMAL,
 388                   .ForceThreadDispatchEnable = NORMAL,
 389                   .PointRasterizationRule = RASTRULE_UPPER_RIGHT,
 390                   .BarycentricInterpolationMode =
 391                      pipeline->wm_prog_data.barycentric_interp_modes);
 392
 393    uint32_t samples = 1;
 394    uint32_t log2_samples = __builtin_ffs(samples) - 1;
 395    bool enable_sampling = samples > 1 ? true : false;
 396
 397    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_MULTISAMPLE,
 398                   .PixelPositionOffsetEnable = enable_sampling,
 399                   .PixelLocation = CENTER,
 400                   .NumberofMultisamples = log2_samples);
 401
 402    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SAMPLE_MASK,
 403                   .SampleMask = 0xffff);
 404
 405    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_VS,
 406                   .VSURBStartingAddress = pipeline->urb.vs_start,
 407                   .VSURBEntryAllocationSize = pipeline->urb.vs_size - 1,
 408                   .VSNumberofURBEntries = pipeline->urb.nr_vs_entries);
 409
 410    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_GS,
 411                   .GSURBStartingAddress = pipeline->urb.gs_start,
 412                   .GSURBEntryAllocationSize = pipeline->urb.gs_size - 1,
 413                   .GSNumberofURBEntries = pipeline->urb.nr_gs_entries);
 414
 415    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_HS,
 416                   .HSURBStartingAddress = pipeline->urb.vs_start,
 417                   .HSURBEntryAllocationSize = 0,
 418                   .HSNumberofURBEntries = 0);
 419
 420    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_DS,
 421                   .DSURBStartingAddress = pipeline->urb.vs_start,
 422                   .DSURBEntryAllocationSize = 0,
 423                   .DSNumberofURBEntries = 0);
 424
 425    const struct brw_gs_prog_data *gs_prog_data = &pipeline->gs_prog_data;
 426    offset = 1;
 427    length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
 428
 429    if (pipeline->gs_vec4 == NO_KERNEL)
 430       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS, .Enable = false);
 431    else
 432       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS,
 433                      .SingleProgramFlow = false,
 434                      .KernelStartPointer = pipeline->gs_vec4,
 435                      .VectorMaskEnable = Dmask,
 436                      .SamplerCount = 0,
 437                      .BindingTableEntryCount = 0,
 438                      .ExpectedVertexCount = pipeline->gs_vertex_count,
 439
 440                      .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_GEOMETRY],
 441                      .PerThreadScratchSpace = ffs(gs_prog_data->base.base.total_scratch / 2048),
 442
 443                      .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
 444                      .OutputTopology = gs_prog_data->output_topology,
 445                      .VertexURBEntryReadLength = gs_prog_data->base.urb_read_length,
 446                      .DispatchGRFStartRegisterForURBData =
 447                         gs_prog_data->base.base.dispatch_grf_start_reg,
 448
 449                      .MaximumNumberofThreads = device->info.max_gs_threads / 2 - 1,
 450                      .ControlDataHeaderSize = gs_prog_data->control_data_header_size_hwords,
 451                      .DispatchMode = gs_prog_data->base.dispatch_mode,
 452                      .StatisticsEnable = true,
 453                      .IncludePrimitiveID = gs_prog_data->include_primitive_id,
 454                      .ReorderMode = TRAILING,
 455                      .Enable = true,
 456
 457                      .ControlDataFormat = gs_prog_data->control_data_format,
 458
 459                      .StaticOutput = gs_prog_data->static_vertex_count >= 0,
 460                      .StaticOutputVertexCount =
 461                         gs_prog_data->static_vertex_count >= 0 ?
 462                         gs_prog_data->static_vertex_count : 0,
 463
 464                      /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
 465                       * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
 466                       * UserClipDistanceCullTestEnableBitmask(v)
 467                       */
 468
 469                      .VertexURBEntryOutputReadOffset = offset,
 470                      .VertexURBEntryOutputLength = length);
 471
 472    const struct brw_vue_prog_data *vue_prog_data = &pipeline->vs_prog_data.base;
 473    /* Skip the VUE header and position slots */
 474    offset = 1;
 475    length = (vue_prog_data->vue_map.num_slots + 1) / 2 - offset;
 476
 477    if (pipeline->vs_simd8 == NO_KERNEL || (extra && extra->disable_vs))
 478       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
 479                      .FunctionEnable = false,
 480                      /* Even if VS is disabled, SBE still gets the amount of
 481                       * vertex data to read from this field. */
 482                      .VertexURBEntryOutputReadOffset = offset,
 483                      .VertexURBEntryOutputLength = length);
 484    else
 485       anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
 486                      .KernelStartPointer = pipeline->vs_simd8,
 487                      .SingleVertexDispatch = Multiple,
 488                      .VectorMaskEnable = Dmask,
 489                      .SamplerCount = 0,
 490                      .BindingTableEntryCount =
 491                      vue_prog_data->base.binding_table.size_bytes / 4,
 492                      .ThreadDispatchPriority = Normal,
 493                      .FloatingPointMode = IEEE754,
 494                      .IllegalOpcodeExceptionEnable = false,
 495                      .AccessesUAV = false,
 496                      .SoftwareExceptionEnable = false,
 497
 498                      .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_VERTEX],
 499                      .PerThreadScratchSpace = ffs(vue_prog_data->base.total_scratch / 2048),
 500
 501                      .DispatchGRFStartRegisterForURBData =
 502                      vue_prog_data->base.dispatch_grf_start_reg,
 503                      .VertexURBEntryReadLength = vue_prog_data->urb_read_length,
 504                      .VertexURBEntryReadOffset = 0,
 505
 506                      .MaximumNumberofThreads = device->info.max_vs_threads - 1,
 507                      .StatisticsEnable = false,
 508                      .SIMD8DispatchEnable = true,
 509                      .VertexCacheDisable = false,
 510                      .FunctionEnable = true,
 511
 512                      .VertexURBEntryOutputReadOffset = offset,
 513                      .VertexURBEntryOutputLength = length,
 514                      .UserClipDistanceClipTestEnableBitmask = 0,
 515                      .UserClipDistanceCullTestEnableBitmask = 0);
 516
 517    const struct brw_wm_prog_data *wm_prog_data = &pipeline->wm_prog_data;
 518
 519    /* TODO: We should clean this up.  Among other things, this is mostly
 520     * shared with other gens.
 521     */
 522    const struct brw_vue_map *fs_input_map;
 523    if (pipeline->gs_vec4 == NO_KERNEL)
 524       fs_input_map = &vue_prog_data->vue_map;
 525    else
 526       fs_input_map = &gs_prog_data->base.vue_map;
 527
 528    struct GEN8_3DSTATE_SBE_SWIZ swiz = {
 529       GEN8_3DSTATE_SBE_SWIZ_header,
 530    };
 531
 532    int max_source_attr = 0;
 533    for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 534       int input_index = wm_prog_data->urb_setup[attr];
 535
 536       if (input_index < 0)
 537          continue;
 538
 539       /* We have to subtract two slots to accout for the URB entry output
 540        * read offset in the VS and GS stages.
 541        */
 542       int source_attr = fs_input_map->varying_to_slot[attr] - 2;
 543       max_source_attr = MAX2(max_source_attr, source_attr);
 544
 545       if (input_index >= 16)
 546          continue;
 547
 548       swiz.Attribute[input_index].SourceAttribute = source_attr;
 549    }
 550
 551    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE,
 552                   .AttributeSwizzleEnable = true,
 553                   .ForceVertexURBEntryReadLength = false,
 554                   .ForceVertexURBEntryReadOffset = false,
 555                   .VertexURBEntryReadLength = DIV_ROUND_UP(max_source_attr + 1, 2),
 556                   .PointSpriteTextureCoordinateOrigin = UPPERLEFT,
 557                   .NumberofSFOutputAttributes =
 558                      wm_prog_data->num_varying_inputs);
 559
 560    uint32_t *dw = anv_batch_emit_dwords(&pipeline->batch,
 561                                         GEN8_3DSTATE_SBE_SWIZ_length);
 562    GEN8_3DSTATE_SBE_SWIZ_pack(&pipeline->batch, dw, &swiz);
 563
 564    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS,
 565                   .KernelStartPointer0 = pipeline->ps_ksp0,
 566
 567                   .SingleProgramFlow = false,
 568                   .VectorMaskEnable = true,
 569                   .SamplerCount = 1,
 570
 571                   .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
 572                   .PerThreadScratchSpace = ffs(wm_prog_data->base.total_scratch / 2048),
 573
 574                   .MaximumNumberofThreadsPerPSD = 64 - 2,
 575                   .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
 576                      POSOFFSET_SAMPLE: POSOFFSET_NONE,
 577                   .PushConstantEnable = wm_prog_data->base.nr_params > 0,
 578                   ._8PixelDispatchEnable = pipeline->ps_simd8 != NO_KERNEL,
 579                   ._16PixelDispatchEnable = pipeline->ps_simd16 != NO_KERNEL,
 580                   ._32PixelDispatchEnable = false,
 581
 582                   .DispatchGRFStartRegisterForConstantSetupData0 = pipeline->ps_grf_start0,
 583                   .DispatchGRFStartRegisterForConstantSetupData1 = 0,
 584                   .DispatchGRFStartRegisterForConstantSetupData2 = pipeline->ps_grf_start2,
 585
 586                   .KernelStartPointer1 = 0,
 587                   .KernelStartPointer2 = pipeline->ps_ksp2);
 588
 589    bool per_sample_ps = false;
 590    anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS_EXTRA,
 591                   .PixelShaderValid = true,
 592                   .PixelShaderKillsPixel = wm_prog_data->uses_kill,
 593                   .PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode,
 594                   .AttributeEnable = wm_prog_data->num_varying_inputs > 0,
 595                   .oMaskPresenttoRenderTarget = wm_prog_data->uses_omask,
 596                   .PixelShaderIsPerSample = per_sample_ps);
 597
 598    *pPipeline = anv_pipeline_to_handle(pipeline);
 599
 600    return VK_SUCCESS;
 601 }
 602
 603 VkResult gen8_compute_pipeline_create(
 604     VkDevice                                    _device,
 605     const VkComputePipelineCreateInfo*          pCreateInfo,
 606     VkPipeline*                                 pPipeline)
 607 {
 608    ANV_FROM_HANDLE(anv_device, device, _device);
 609    struct anv_pipeline *pipeline;
 610    VkResult result;
 611
 612    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
 613
 614    pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
 615                                VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
 616    if (pipeline == NULL)
 617       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 618
 619    pipeline->device = device;
 620    pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
 621
 622    pipeline->blend_state.map = NULL;
 623
 624    result = anv_reloc_list_init(&pipeline->batch_relocs, device);
 625    if (result != VK_SUCCESS) {
 626       anv_device_free(device, pipeline);
 627       return result;
 628    }
 629    pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
 630    pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
 631    pipeline->batch.relocs = &pipeline->batch_relocs;
 632
 633    anv_state_stream_init(&pipeline->program_stream,
 634                          &device->instruction_block_pool);
 635
 636    /* When we free the pipeline, we detect stages based on the NULL status
 637     * of various prog_data pointers.  Make them NULL by default.
 638     */
 639    memset(pipeline->prog_data, 0, sizeof(pipeline->prog_data));
 640    memset(pipeline->scratch_start, 0, sizeof(pipeline->scratch_start));
 641
 642    pipeline->vs_simd8 = NO_KERNEL;
 643    pipeline->vs_vec4 = NO_KERNEL;
 644    pipeline->gs_vec4 = NO_KERNEL;
 645
 646    pipeline->active_stages = 0;
 647    pipeline->total_scratch = 0;
 648
 649    assert(pCreateInfo->stage.stage == VK_SHADER_STAGE_COMPUTE);
 650    ANV_FROM_HANDLE(anv_shader, shader,  pCreateInfo->stage.shader);
 651    anv_pipeline_compile_cs(pipeline, pCreateInfo, shader);
 652
 653    pipeline->use_repclear = false;
 654
 655    const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data;
 656
 657    anv_batch_emit(&pipeline->batch, GEN8_MEDIA_VFE_STATE,
 658                   .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_COMPUTE],
 659                   .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
 660                   .ScratchSpaceBasePointerHigh = 0,
 661                   .StackSize = 0,
 662
 663                   .MaximumNumberofThreads = device->info.max_cs_threads - 1,
 664                   .NumberofURBEntries = 2,
 665                   .ResetGatewayTimer = true,
 666                   .BypassGatewayControl = true,
 667                   .URBEntryAllocationSize = 2,
 668                   .CURBEAllocationSize = 0);
 669
 670    struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
 671    uint32_t group_size = prog_data->local_size[0] *
 672       prog_data->local_size[1] * prog_data->local_size[2];
 673    pipeline->cs_thread_width_max = DIV_ROUND_UP(group_size, prog_data->simd_size);
 674    uint32_t remainder = group_size & (prog_data->simd_size - 1);
 675
 676    if (remainder > 0)
 677       pipeline->cs_right_mask = ~0u >> (32 - remainder);
 678    else
 679       pipeline->cs_right_mask = ~0u >> (32 - prog_data->simd_size);
 680
 681
 682    *pPipeline = anv_pipeline_to_handle(pipeline);
 683
 684    return VK_SUCCESS;
 685 }