src/intel/vulkan/gen8_pipeline.c

   1 /*
   2  * Copyright © 2015 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include <assert.h>
  25 #include <stdbool.h>
  26 #include <string.h>
  27 #include <unistd.h>
  28 #include <fcntl.h>
  29
  30 #include "anv_private.h"
  31
  32 #include "genxml/gen_macros.h"
  33 #include "genxml/genX_pack.h"
  34
  35 #include "genX_pipeline_util.h"
  36
  37 static void
  38 emit_ia_state(struct anv_pipeline *pipeline,
  39               const VkPipelineInputAssemblyStateCreateInfo *info,
  40               const struct anv_graphics_pipeline_create_info *extra)
  41 {
  42    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VF_TOPOLOGY), vft) {
  43       vft.PrimitiveTopologyType = pipeline->topology;
  44    }
  45 }
  46
  47 static void
  48 emit_rs_state(struct anv_pipeline *pipeline,
  49               const VkPipelineRasterizationStateCreateInfo *info,
  50               const VkPipelineMultisampleStateCreateInfo *ms_info,
  51               const struct anv_graphics_pipeline_create_info *extra)
  52 {
  53    uint32_t samples = 1;
  54
  55    if (ms_info)
  56       samples = ms_info->rasterizationSamples;
  57
  58    struct GENX(3DSTATE_SF) sf = {
  59       GENX(3DSTATE_SF_header),
  60       .ViewportTransformEnable = !(extra && extra->use_rectlist),
  61       .TriangleStripListProvokingVertexSelect = 0,
  62       .LineStripListProvokingVertexSelect = 0,
  63       .TriangleFanProvokingVertexSelect = 1,
  64       .PointWidthSource = Vertex,
  65       .PointWidth = 1.0,
  66    };
  67
  68    /* FINISHME: VkBool32 rasterizerDiscardEnable; */
  69
  70    GENX(3DSTATE_SF_pack)(NULL, pipeline->gen8.sf, &sf);
  71
  72    struct GENX(3DSTATE_RASTER) raster = {
  73       GENX(3DSTATE_RASTER_header),
  74
  75       /* For details on 3DSTATE_RASTER multisample state, see the BSpec table
  76        * "Multisample Modes State".
  77        */
  78       .DXMultisampleRasterizationEnable = samples > 1,
  79       .ForcedSampleCount = FSC_NUMRASTSAMPLES_0,
  80       .ForceMultisampling = false,
  81
  82       .FrontWinding = vk_to_gen_front_face[info->frontFace],
  83       .CullMode = vk_to_gen_cullmode[info->cullMode],
  84       .FrontFaceFillMode = vk_to_gen_fillmode[info->polygonMode],
  85       .BackFaceFillMode = vk_to_gen_fillmode[info->polygonMode],
  86       .ScissorRectangleEnable = !(extra && extra->use_rectlist),
  87 #if GEN_GEN == 8
  88       .ViewportZClipTestEnable = !pipeline->depth_clamp_enable,
  89 #else
  90       /* GEN9+ splits ViewportZClipTestEnable into near and far enable bits */
  91       .ViewportZFarClipTestEnable = !pipeline->depth_clamp_enable,
  92       .ViewportZNearClipTestEnable = !pipeline->depth_clamp_enable,
  93 #endif
  94       .GlobalDepthOffsetEnableSolid = info->depthBiasEnable,
  95       .GlobalDepthOffsetEnableWireframe = info->depthBiasEnable,
  96       .GlobalDepthOffsetEnablePoint = info->depthBiasEnable,
  97    };
  98
  99    GENX(3DSTATE_RASTER_pack)(NULL, pipeline->gen8.raster, &raster);
 100 }
 101
 102 static void
 103 emit_ms_state(struct anv_pipeline *pipeline,
 104               const VkPipelineMultisampleStateCreateInfo *info)
 105 {
 106    uint32_t samples = 1;
 107    uint32_t log2_samples = 0;
 108
 109    /* From the Vulkan 1.0 spec:
 110     *    If pSampleMask is NULL, it is treated as if the mask has all bits
 111     *    enabled, i.e. no coverage is removed from fragments.
 112     *
 113     * 3DSTATE_SAMPLE_MASK.SampleMask is 16 bits.
 114     */
 115    uint32_t sample_mask = 0xffff;
 116
 117    if (info) {
 118       samples = info->rasterizationSamples;
 119       log2_samples = __builtin_ffs(samples) - 1;
 120    }
 121
 122    if (info && info->pSampleMask)
 123       sample_mask &= info->pSampleMask[0];
 124
 125    if (info && info->sampleShadingEnable)
 126       anv_finishme("VkPipelineMultisampleStateCreateInfo::sampleShadingEnable");
 127
 128    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_MULTISAMPLE), ms) {
 129       /* The PRM says that this bit is valid only for DX9:
 130        *
 131        *    SW can choose to set this bit only for DX9 API. DX10/OGL API's
 132        *    should not have any effect by setting or not setting this bit.
 133        */
 134       ms.PixelPositionOffsetEnable = false;
 135
 136       ms.PixelLocation = CENTER;
 137       ms.NumberofMultisamples = log2_samples;
 138    }
 139
 140    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_SAMPLE_MASK), sm) {
 141       sm.SampleMask = sample_mask;
 142    }
 143 }
 144
 145 VkResult
 146 genX(graphics_pipeline_create)(
 147     VkDevice                                    _device,
 148     struct anv_pipeline_cache *                 cache,
 149     const VkGraphicsPipelineCreateInfo*         pCreateInfo,
 150     const struct anv_graphics_pipeline_create_info *extra,
 151     const VkAllocationCallbacks*                pAllocator,
 152     VkPipeline*                                 pPipeline)
 153 {
 154    ANV_FROM_HANDLE(anv_device, device, _device);
 155    ANV_FROM_HANDLE(anv_render_pass, pass, pCreateInfo->renderPass);
 156    struct anv_subpass *subpass = &pass->subpasses[pCreateInfo->subpass];
 157    struct anv_pipeline *pipeline;
 158    VkResult result;
 159    uint32_t offset, length;
 160
 161    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
 162
 163    pipeline = anv_alloc2(&device->alloc, pAllocator, sizeof(*pipeline), 8,
 164                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 165    if (pipeline == NULL)
 166       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 167
 168    result = anv_pipeline_init(pipeline, device, cache,
 169                               pCreateInfo, extra, pAllocator);
 170    if (result != VK_SUCCESS) {
 171       anv_free2(&device->alloc, pAllocator, pipeline);
 172       return result;
 173    }
 174
 175    assert(pCreateInfo->pVertexInputState);
 176    emit_vertex_input(pipeline, pCreateInfo->pVertexInputState, extra);
 177    assert(pCreateInfo->pInputAssemblyState);
 178    emit_ia_state(pipeline, pCreateInfo->pInputAssemblyState, extra);
 179    assert(pCreateInfo->pRasterizationState);
 180    emit_rs_state(pipeline, pCreateInfo->pRasterizationState,
 181                  pCreateInfo->pMultisampleState, extra);
 182    emit_ms_state(pipeline, pCreateInfo->pMultisampleState);
 183    emit_ds_state(pipeline, pCreateInfo->pDepthStencilState, pass, subpass);
 184    emit_cb_state(pipeline, pCreateInfo->pColorBlendState,
 185                            pCreateInfo->pMultisampleState);
 186
 187    emit_urb_setup(pipeline);
 188
 189    const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline);
 190    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_CLIP), clip) {
 191       clip.ClipEnable               = !(extra && extra->use_rectlist);
 192       clip.EarlyCullEnable          = true;
 193       clip.APIMode                  = 1; /* D3D */
 194       clip.ViewportXYClipTestEnable = true;
 195
 196       clip.ClipMode =
 197          pCreateInfo->pRasterizationState->rasterizerDiscardEnable ?
 198          CLIPMODE_REJECT_ALL : CLIPMODE_NORMAL;
 199
 200       clip.NonPerspectiveBarycentricEnable = wm_prog_data ?
 201          (wm_prog_data->barycentric_interp_modes & 0x38) != 0 : 0;
 202
 203       clip.TriangleStripListProvokingVertexSelect  = 0;
 204       clip.LineStripListProvokingVertexSelect      = 0;
 205       clip.TriangleFanProvokingVertexSelect        = 1;
 206
 207       clip.MinimumPointWidth  = 0.125;
 208       clip.MaximumPointWidth  = 255.875;
 209       clip.MaximumVPIndex     = pCreateInfo->pViewportState->viewportCount - 1;
 210    }
 211
 212    anv_batch_emit(&pipeline->batch, GENX(3DSTATE_WM), wm) {
 213       wm.StatisticsEnable                    = true;
 214       wm.LineEndCapAntialiasingRegionWidth   = _05pixels;
 215       wm.LineAntialiasingRegionWidth         = _10pixels;
 216       wm.ForceThreadDispatchEnable           = NORMAL;
 217       wm.PointRasterizationRule              = RASTRULE_UPPER_RIGHT;
 218
 219       if (wm_prog_data && wm_prog_data->early_fragment_tests) {
 220          wm.EarlyDepthStencilControl         = PREPS;
 221       } else if (wm_prog_data && wm_prog_data->has_side_effects) {
 222          wm.EarlyDepthStencilControl         = PSEXEC;
 223       } else {
 224          wm.EarlyDepthStencilControl         = NORMAL;
 225       }
 226
 227       wm.BarycentricInterpolationMode = pipeline->ps_ksp0 == NO_KERNEL ?
 228          0 : wm_prog_data->barycentric_interp_modes;
 229    }
 230
 231    if (pipeline->gs_kernel == NO_KERNEL) {
 232       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_GS), gs);
 233    } else {
 234       const struct brw_gs_prog_data *gs_prog_data = get_gs_prog_data(pipeline);
 235       offset = 1;
 236       length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
 237
 238       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_GS), gs) {
 239          gs.SingleProgramFlow       = false;
 240          gs.KernelStartPointer      = pipeline->gs_kernel;
 241          gs.VectorMaskEnable        = false;
 242          gs.SamplerCount            = 0;
 243          gs.BindingTableEntryCount  = 0;
 244          gs.ExpectedVertexCount     = gs_prog_data->vertices_in;
 245
 246          gs.ScratchSpaceBasePointer = (struct anv_address) {
 247             .bo = anv_scratch_pool_alloc(device, &device->scratch_pool,
 248                                          MESA_SHADER_GEOMETRY,
 249                                          gs_prog_data->base.base.total_scratch),
 250             .offset = 0,
 251          };
 252          gs.PerThreadScratchSpace   = scratch_space(&gs_prog_data->base.base);
 253          gs.OutputVertexSize        = gs_prog_data->output_vertex_size_hwords * 2 - 1;
 254          gs.OutputTopology          = gs_prog_data->output_topology;
 255          gs.VertexURBEntryReadLength = gs_prog_data->base.urb_read_length;
 256          gs.IncludeVertexHandles    = gs_prog_data->base.include_vue_handles;
 257
 258          gs.DispatchGRFStartRegisterForURBData =
 259             gs_prog_data->base.base.dispatch_grf_start_reg;
 260
 261          gs.MaximumNumberofThreads  = device->info.max_gs_threads / 2 - 1;
 262          gs.ControlDataHeaderSize   = gs_prog_data->control_data_header_size_hwords;
 263          gs.DispatchMode            = gs_prog_data->base.dispatch_mode;
 264          gs.StatisticsEnable        = true;
 265          gs.IncludePrimitiveID      = gs_prog_data->include_primitive_id;
 266          gs.ReorderMode             = TRAILING;
 267          gs.Enable                  = true;
 268
 269          gs.ControlDataFormat       = gs_prog_data->control_data_format;
 270
 271          gs.StaticOutput            = gs_prog_data->static_vertex_count >= 0;
 272          gs.StaticOutputVertexCount =
 273             gs_prog_data->static_vertex_count >= 0 ?
 274             gs_prog_data->static_vertex_count : 0;
 275
 276          /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
 277           * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
 278           * UserClipDistanceCullTestEnableBitmask(v)
 279           */
 280
 281          gs.VertexURBEntryOutputReadOffset = offset;
 282          gs.VertexURBEntryOutputLength = length;
 283       }
 284    }
 285
 286    const struct brw_vs_prog_data *vs_prog_data = get_vs_prog_data(pipeline);
 287    /* Skip the VUE header and position slots */
 288    offset = 1;
 289    length = (vs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
 290
 291    uint32_t vs_start = pipeline->vs_simd8 != NO_KERNEL ? pipeline->vs_simd8 :
 292                                                          pipeline->vs_vec4;
 293
 294    if (vs_start == NO_KERNEL || (extra && extra->disable_vs)) {
 295       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VS), vs) {
 296          vs.FunctionEnable = false;
 297          /* Even if VS is disabled, SBE still gets the amount of
 298           * vertex data to read from this field. */
 299          vs.VertexURBEntryOutputReadOffset = offset;
 300          vs.VertexURBEntryOutputLength = length;
 301       }
 302    } else {
 303       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_VS), vs) {
 304          vs.KernelStartPointer            = vs_start;
 305          vs.SingleVertexDispatch          = false;
 306          vs.VectorMaskEnable              = false;
 307          vs.SamplerCount                  = 0;
 308
 309          vs.BindingTableEntryCount =
 310             vs_prog_data->base.base.binding_table.size_bytes / 4,
 311
 312          vs.ThreadDispatchPriority        = false;
 313          vs.FloatingPointMode             = IEEE754;
 314          vs.IllegalOpcodeExceptionEnable  = false;
 315          vs.AccessesUAV                   = false;
 316          vs.SoftwareExceptionEnable       = false;
 317
 318          vs.ScratchSpaceBasePointer = (struct anv_address) {
 319             .bo = anv_scratch_pool_alloc(device, &device->scratch_pool,
 320                                          MESA_SHADER_VERTEX,
 321                                          vs_prog_data->base.base.total_scratch),
 322             .offset = 0,
 323          };
 324          vs.PerThreadScratchSpace   = scratch_space(&vs_prog_data->base.base);
 325
 326          vs.DispatchGRFStartRegisterForURBData =
 327             vs_prog_data->base.base.dispatch_grf_start_reg;
 328
 329          vs.VertexURBEntryReadLength      = vs_prog_data->base.urb_read_length;
 330          vs.VertexURBEntryReadOffset      = 0;
 331
 332          vs.MaximumNumberofThreads        = device->info.max_vs_threads - 1;
 333          vs.StatisticsEnable              = false;
 334          vs.SIMD8DispatchEnable           = pipeline->vs_simd8 != NO_KERNEL;
 335          vs.VertexCacheDisable            = false;
 336          vs.FunctionEnable                = true;
 337
 338          vs.VertexURBEntryOutputReadOffset = offset;
 339          vs.VertexURBEntryOutputLength    = length;
 340
 341          /* TODO */
 342          vs.UserClipDistanceClipTestEnableBitmask = 0;
 343          vs.UserClipDistanceCullTestEnableBitmask = 0;
 344       }
 345    }
 346
 347    const int num_thread_bias = GEN_GEN == 8 ? 2 : 1;
 348    if (pipeline->ps_ksp0 == NO_KERNEL) {
 349       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS), ps);
 350       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS_EXTRA), extra) {
 351          extra.PixelShaderValid = false;
 352       }
 353    } else {
 354       emit_3dstate_sbe(pipeline);
 355
 356       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS), ps) {
 357          ps.KernelStartPointer0     = pipeline->ps_ksp0;
 358          ps.KernelStartPointer1     = 0;
 359          ps.KernelStartPointer2     = pipeline->ps_ksp0 + wm_prog_data->prog_offset_2;
 360          ps._8PixelDispatchEnable   = wm_prog_data->dispatch_8;
 361          ps._16PixelDispatchEnable  = wm_prog_data->dispatch_16;
 362          ps._32PixelDispatchEnable  = false;
 363          ps.SingleProgramFlow       = false;
 364          ps.VectorMaskEnable        = true;
 365          ps.SamplerCount            = 1;
 366          ps.PushConstantEnable      = wm_prog_data->base.nr_params > 0;
 367          ps.PositionXYOffsetSelect  = wm_prog_data->uses_pos_offset ?
 368             POSOFFSET_SAMPLE: POSOFFSET_NONE;
 369
 370          ps.MaximumNumberofThreadsPerPSD = 64 - num_thread_bias;
 371
 372          ps.ScratchSpaceBasePointer = (struct anv_address) {
 373             .bo = anv_scratch_pool_alloc(device, &device->scratch_pool,
 374                                          MESA_SHADER_FRAGMENT,
 375                                          wm_prog_data->base.total_scratch),
 376             .offset = 0,
 377          };
 378          ps.PerThreadScratchSpace   = scratch_space(&wm_prog_data->base);
 379
 380          ps.DispatchGRFStartRegisterForConstantSetupData0 =
 381             wm_prog_data->base.dispatch_grf_start_reg;
 382          ps.DispatchGRFStartRegisterForConstantSetupData1 = 0;
 383          ps.DispatchGRFStartRegisterForConstantSetupData2 =
 384             wm_prog_data->dispatch_grf_start_reg_2;
 385       }
 386
 387       bool per_sample_ps = pCreateInfo->pMultisampleState &&
 388                            pCreateInfo->pMultisampleState->sampleShadingEnable;
 389
 390       anv_batch_emit(&pipeline->batch, GENX(3DSTATE_PS_EXTRA), ps) {
 391          ps.PixelShaderValid              = true;
 392          ps.PixelShaderKillsPixel         = wm_prog_data->uses_kill;
 393          ps.PixelShaderComputedDepthMode  = wm_prog_data->computed_depth_mode;
 394          ps.AttributeEnable               = wm_prog_data->num_varying_inputs > 0;
 395          ps.oMaskPresenttoRenderTarget    = wm_prog_data->uses_omask;
 396          ps.PixelShaderIsPerSample        = per_sample_ps;
 397          ps.PixelShaderUsesSourceDepth    = wm_prog_data->uses_src_depth;
 398          ps.PixelShaderUsesSourceW        = wm_prog_data->uses_src_w;
 399 #if GEN_GEN >= 9
 400          ps.PixelShaderPullsBary    = wm_prog_data->pulls_bary;
 401          ps.InputCoverageMaskState  = wm_prog_data->uses_sample_mask ?
 402             ICMS_INNER_CONSERVATIVE : ICMS_NONE;
 403 #else
 404          ps.PixelShaderUsesInputCoverageMask = wm_prog_data->uses_sample_mask;
 405 #endif
 406       }
 407    }
 408
 409    *pPipeline = anv_pipeline_to_handle(pipeline);
 410
 411    return VK_SUCCESS;
 412 }