1 /*
2 * Copyright 2020 Google LLC
3 *
4 * Use of this source code is governed by a BSD-style license that can be
5 * found in the LICENSE file.
6 */
7
8 //#include <d3dcompiler.h>
9
10 #include "src/gpu/ganesh/d3d/GrD3DPipelineStateBuilder.h"
11
12 #include "include/gpu/GrDirectContext.h"
13 #include "include/gpu/d3d/GrD3DTypes.h"
14 #include "include/private/SkSLProgramKind.h"
15 #include "src/core/SkReadBuffer.h"
16 #include "src/core/SkTraceEvent.h"
17 #include "src/gpu/ganesh/GrAutoLocaleSetter.h"
18 #include "src/gpu/ganesh/GrDirectContextPriv.h"
19 #include "src/gpu/ganesh/GrPersistentCacheUtils.h"
20 #include "src/gpu/ganesh/GrShaderCaps.h"
21 #include "src/gpu/ganesh/GrStencilSettings.h"
22 #include "src/gpu/ganesh/d3d/GrD3DGpu.h"
23 #include "src/gpu/ganesh/d3d/GrD3DPipeline.h"
24 #include "src/gpu/ganesh/d3d/GrD3DRenderTarget.h"
25 #include "src/gpu/ganesh/d3d/GrD3DRootSignature.h"
26 #include "src/gpu/ganesh/d3d/GrD3DUtil.h"
27 #include "src/sksl/SkSLCompiler.h"
28 #include "src/sksl/SkSLProgramSettings.h"
29 #include "src/utils/SkShaderUtils.h"
30
31 #include <d3dcompiler.h>
32
33 using namespace skia_private;
34
MakePipelineState(GrD3DGpu * gpu,GrD3DRenderTarget * renderTarget,const GrProgramDesc & desc,const GrProgramInfo & programInfo)35 std::unique_ptr<GrD3DPipelineState> GrD3DPipelineStateBuilder::MakePipelineState(
36 GrD3DGpu* gpu,
37 GrD3DRenderTarget* renderTarget,
38 const GrProgramDesc& desc,
39 const GrProgramInfo& programInfo) {
40 // ensure that we use "." as a decimal separator when creating SkSL code
41 GrAutoLocaleSetter als("C");
42
43 // create a builder. This will be handed off to effects so they can use it to add
44 // uniforms, varyings, textures, etc
45 GrD3DPipelineStateBuilder builder(gpu, renderTarget, desc, programInfo);
46
47 if (!builder.emitAndInstallProcs()) {
48 return nullptr;
49 }
50
51 return builder.finalize();
52 }
53
GrD3DPipelineStateBuilder(GrD3DGpu * gpu,GrD3DRenderTarget * renderTarget,const GrProgramDesc & desc,const GrProgramInfo & programInfo)54 GrD3DPipelineStateBuilder::GrD3DPipelineStateBuilder(GrD3DGpu* gpu,
55 GrD3DRenderTarget* renderTarget,
56 const GrProgramDesc& desc,
57 const GrProgramInfo& programInfo)
58 : INHERITED(desc, programInfo)
59 , fGpu(gpu)
60 , fVaryingHandler(this)
61 , fUniformHandler(this)
62 , fRenderTarget(renderTarget) {}
63
caps() const64 const GrCaps* GrD3DPipelineStateBuilder::caps() const {
65 return fGpu->caps();
66 }
67
shaderCompiler() const68 SkSL::Compiler* GrD3DPipelineStateBuilder::shaderCompiler() const {
69 return fGpu->shaderCompiler();
70 }
71
finalizeFragmentSecondaryColor(GrShaderVar & outputColor)72 void GrD3DPipelineStateBuilder::finalizeFragmentSecondaryColor(GrShaderVar& outputColor) {
73 outputColor.addLayoutQualifier("location = 0, index = 1");
74 }
75
76 // Print the source code for all shaders generated.
77 static const bool gPrintSKSL = false;
78 static const bool gPrintHLSL = false;
79
GrCompileHLSLShader(GrD3DGpu * gpu,const std::string & hlsl,SkSL::ProgramKind kind)80 static gr_cp<ID3DBlob> GrCompileHLSLShader(GrD3DGpu* gpu,
81 const std::string& hlsl,
82 SkSL::ProgramKind kind) {
83 TRACE_EVENT0("skia.shaders", "driver_compile_shader");
84 const char* compileTarget = nullptr;
85 switch (kind) {
86 case SkSL::ProgramKind::kVertex:
87 compileTarget = "vs_5_1";
88 break;
89 case SkSL::ProgramKind::kFragment:
90 compileTarget = "ps_5_1";
91 break;
92 default:
93 SkUNREACHABLE;
94 }
95
96 uint32_t compileFlags = 0;
97 #ifdef SK_DEBUG
98 // Enable better shader debugging with the graphics debugging tools.
99 compileFlags |= D3DCOMPILE_DEBUG | D3DCOMPILE_SKIP_OPTIMIZATION;
100 #endif
101 // SPRIV-cross does matrix multiplication expecting row major matrices
102 compileFlags |= D3DCOMPILE_PACK_MATRIX_ROW_MAJOR;
103
104 gr_cp<ID3DBlob> shader;
105 gr_cp<ID3DBlob> errors;
106 HRESULT hr = D3DCompile(hlsl.c_str(), hlsl.length(), nullptr, nullptr, nullptr, "main",
107 compileTarget, compileFlags, 0, &shader, &errors);
108 if (!SUCCEEDED(hr)) {
109 gpu->getContext()->priv().getShaderErrorHandler()->compileError(
110 hlsl.c_str(), reinterpret_cast<char*>(errors->GetBufferPointer()));
111 }
112 return shader;
113 }
114
loadHLSLFromCache(SkReadBuffer * reader,gr_cp<ID3DBlob> shaders[])115 bool GrD3DPipelineStateBuilder::loadHLSLFromCache(SkReadBuffer* reader, gr_cp<ID3DBlob> shaders[]) {
116
117 std::string hlsl[kGrShaderTypeCount];
118 SkSL::Program::Inputs inputs[kGrShaderTypeCount];
119
120 if (!GrPersistentCacheUtils::UnpackCachedShaders(reader, hlsl, inputs, kGrShaderTypeCount)) {
121 return false;
122 }
123
124 auto compile = [&](SkSL::ProgramKind kind, GrShaderType shaderType) {
125 if (inputs[shaderType].fUseFlipRTUniform) {
126 this->addRTFlipUniform(SKSL_RTFLIP_NAME);
127 }
128 shaders[shaderType] = GrCompileHLSLShader(fGpu, hlsl[shaderType], kind);
129 return shaders[shaderType].get();
130 };
131
132 return compile(SkSL::ProgramKind::kVertex, kVertex_GrShaderType) &&
133 compile(SkSL::ProgramKind::kFragment, kFragment_GrShaderType);
134 }
135
compileD3DProgram(SkSL::ProgramKind kind,const std::string & sksl,const SkSL::ProgramSettings & settings,SkSL::Program::Inputs * outInputs,std::string * outHLSL)136 gr_cp<ID3DBlob> GrD3DPipelineStateBuilder::compileD3DProgram(
137 SkSL::ProgramKind kind,
138 const std::string& sksl,
139 const SkSL::ProgramSettings& settings,
140 SkSL::Program::Inputs* outInputs,
141 std::string* outHLSL) {
142 #ifdef SK_DEBUG
143 std::string src = SkShaderUtils::PrettyPrint(sksl);
144 #else
145 const std::string& src = sksl;
146 #endif
147
148 std::unique_ptr<SkSL::Program> program = fGpu->shaderCompiler()->convertProgram(
149 kind, src, settings);
150 if (!program || !fGpu->shaderCompiler()->toHLSL(*program, outHLSL)) {
151 auto errorHandler = fGpu->getContext()->priv().getShaderErrorHandler();
152 errorHandler->compileError(src.c_str(),
153 fGpu->shaderCompiler()->errorText().c_str());
154 return gr_cp<ID3DBlob>();
155 }
156 *outInputs = program->fInputs;
157
158 if (gPrintSKSL || gPrintHLSL) {
159 SkShaderUtils::PrintShaderBanner(kind);
160 if (gPrintSKSL) {
161 SkDebugf("SKSL:\n");
162 SkShaderUtils::PrintLineByLine(SkShaderUtils::PrettyPrint(sksl));
163 }
164 if (gPrintHLSL) {
165 SkDebugf("HLSL:\n");
166 SkShaderUtils::PrintLineByLine(SkShaderUtils::PrettyPrint(*outHLSL));
167 }
168 }
169
170 if (program->fInputs.fUseFlipRTUniform) {
171 this->addRTFlipUniform(SKSL_RTFLIP_NAME);
172 }
173
174 return GrCompileHLSLShader(fGpu, *outHLSL, kind);
175 }
176
attrib_type_to_format(GrVertexAttribType type)177 static DXGI_FORMAT attrib_type_to_format(GrVertexAttribType type) {
178 switch (type) {
179 case kFloat_GrVertexAttribType:
180 return DXGI_FORMAT_R32_FLOAT;
181 case kFloat2_GrVertexAttribType:
182 return DXGI_FORMAT_R32G32_FLOAT;
183 case kFloat3_GrVertexAttribType:
184 return DXGI_FORMAT_R32G32B32_FLOAT;
185 case kFloat4_GrVertexAttribType:
186 return DXGI_FORMAT_R32G32B32A32_FLOAT;
187 case kHalf_GrVertexAttribType:
188 return DXGI_FORMAT_R16_FLOAT;
189 case kHalf2_GrVertexAttribType:
190 return DXGI_FORMAT_R16G16_FLOAT;
191 case kHalf4_GrVertexAttribType:
192 return DXGI_FORMAT_R16G16B16A16_FLOAT;
193 case kInt2_GrVertexAttribType:
194 return DXGI_FORMAT_R32G32_SINT;
195 case kInt3_GrVertexAttribType:
196 return DXGI_FORMAT_R32G32B32_SINT;
197 case kInt4_GrVertexAttribType:
198 return DXGI_FORMAT_R32G32B32A32_SINT;
199 case kByte_GrVertexAttribType:
200 return DXGI_FORMAT_R8_SINT;
201 case kByte2_GrVertexAttribType:
202 return DXGI_FORMAT_R8G8_SINT;
203 case kByte4_GrVertexAttribType:
204 return DXGI_FORMAT_R8G8B8A8_SINT;
205 case kUByte_GrVertexAttribType:
206 return DXGI_FORMAT_R8_UINT;
207 case kUByte2_GrVertexAttribType:
208 return DXGI_FORMAT_R8G8_UINT;
209 case kUByte4_GrVertexAttribType:
210 return DXGI_FORMAT_R8G8B8A8_UINT;
211 case kUByte_norm_GrVertexAttribType:
212 return DXGI_FORMAT_R8_UNORM;
213 case kUByte4_norm_GrVertexAttribType:
214 return DXGI_FORMAT_R8G8B8A8_UNORM;
215 case kShort2_GrVertexAttribType:
216 return DXGI_FORMAT_R16G16_SINT;
217 case kShort4_GrVertexAttribType:
218 return DXGI_FORMAT_R16G16B16A16_SINT;
219 case kUShort2_GrVertexAttribType:
220 return DXGI_FORMAT_R16G16_UINT;
221 case kUShort2_norm_GrVertexAttribType:
222 return DXGI_FORMAT_R16G16_UNORM;
223 case kInt_GrVertexAttribType:
224 return DXGI_FORMAT_R32_SINT;
225 case kUInt_GrVertexAttribType:
226 return DXGI_FORMAT_R32_UINT;
227 case kUShort_norm_GrVertexAttribType:
228 return DXGI_FORMAT_R16_UNORM;
229 case kUShort4_norm_GrVertexAttribType:
230 return DXGI_FORMAT_R16G16B16A16_UNORM;
231 }
232 SK_ABORT("Unknown vertex attrib type");
233 }
234
setup_vertex_input_layout(const GrGeometryProcessor & geomProc,D3D12_INPUT_ELEMENT_DESC * inputElements)235 static void setup_vertex_input_layout(const GrGeometryProcessor& geomProc,
236 D3D12_INPUT_ELEMENT_DESC* inputElements) {
237 unsigned int slotNumber = 0;
238 unsigned int vertexSlot = 0;
239 unsigned int instanceSlot = 0;
240 if (geomProc.hasVertexAttributes()) {
241 vertexSlot = slotNumber++;
242 }
243 if (geomProc.hasInstanceAttributes()) {
244 instanceSlot = slotNumber++;
245 }
246
247 unsigned int currentAttrib = 0;
248
249 for (auto attrib : geomProc.vertexAttributes()) {
250 // When using SPIRV-Cross it converts the location modifier in SPIRV to be
251 // TEXCOORD<N> where N is the location value for eveery vertext attribute
252 inputElements[currentAttrib] = { "TEXCOORD", currentAttrib,
253 attrib_type_to_format(attrib.cpuType()),
254 vertexSlot, SkToU32(*attrib.offset()),
255 D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA, 0 };
256 currentAttrib++;
257 }
258
259 for (auto attrib : geomProc.instanceAttributes()) {
260 // When using SPIRV-Cross it converts the location modifier in SPIRV to be
261 // TEXCOORD<N> where N is the location value for eveery vertext attribute
262 inputElements[currentAttrib] = { "TEXCOORD", currentAttrib,
263 attrib_type_to_format(attrib.cpuType()),
264 instanceSlot, SkToU32(*attrib.offset()),
265 D3D12_INPUT_CLASSIFICATION_PER_INSTANCE_DATA, 1 };
266 currentAttrib++;
267 }
268 }
269
blend_coeff_to_d3d_blend(skgpu::BlendCoeff coeff)270 static D3D12_BLEND blend_coeff_to_d3d_blend(skgpu::BlendCoeff coeff) {
271 switch (coeff) {
272 case skgpu::BlendCoeff::kZero:
273 return D3D12_BLEND_ZERO;
274 case skgpu::BlendCoeff::kOne:
275 return D3D12_BLEND_ONE;
276 case skgpu::BlendCoeff::kSC:
277 return D3D12_BLEND_SRC_COLOR;
278 case skgpu::BlendCoeff::kISC:
279 return D3D12_BLEND_INV_SRC_COLOR;
280 case skgpu::BlendCoeff::kDC:
281 return D3D12_BLEND_DEST_COLOR;
282 case skgpu::BlendCoeff::kIDC:
283 return D3D12_BLEND_INV_DEST_COLOR;
284 case skgpu::BlendCoeff::kSA:
285 return D3D12_BLEND_SRC_ALPHA;
286 case skgpu::BlendCoeff::kISA:
287 return D3D12_BLEND_INV_SRC_ALPHA;
288 case skgpu::BlendCoeff::kDA:
289 return D3D12_BLEND_DEST_ALPHA;
290 case skgpu::BlendCoeff::kIDA:
291 return D3D12_BLEND_INV_DEST_ALPHA;
292 case skgpu::BlendCoeff::kConstC:
293 return D3D12_BLEND_BLEND_FACTOR;
294 case skgpu::BlendCoeff::kIConstC:
295 return D3D12_BLEND_INV_BLEND_FACTOR;
296 case skgpu::BlendCoeff::kS2C:
297 return D3D12_BLEND_SRC1_COLOR;
298 case skgpu::BlendCoeff::kIS2C:
299 return D3D12_BLEND_INV_SRC1_COLOR;
300 case skgpu::BlendCoeff::kS2A:
301 return D3D12_BLEND_SRC1_ALPHA;
302 case skgpu::BlendCoeff::kIS2A:
303 return D3D12_BLEND_INV_SRC1_ALPHA;
304 case skgpu::BlendCoeff::kIllegal:
305 return D3D12_BLEND_ZERO;
306 }
307 SkUNREACHABLE;
308 }
309
blend_coeff_to_d3d_blend_for_alpha(skgpu::BlendCoeff coeff)310 static D3D12_BLEND blend_coeff_to_d3d_blend_for_alpha(skgpu::BlendCoeff coeff) {
311 switch (coeff) {
312 // Force all srcColor used in alpha slot to alpha version.
313 case skgpu::BlendCoeff::kSC:
314 return D3D12_BLEND_SRC_ALPHA;
315 case skgpu::BlendCoeff::kISC:
316 return D3D12_BLEND_INV_SRC_ALPHA;
317 case skgpu::BlendCoeff::kDC:
318 return D3D12_BLEND_DEST_ALPHA;
319 case skgpu::BlendCoeff::kIDC:
320 return D3D12_BLEND_INV_DEST_ALPHA;
321 case skgpu::BlendCoeff::kS2C:
322 return D3D12_BLEND_SRC1_ALPHA;
323 case skgpu::BlendCoeff::kIS2C:
324 return D3D12_BLEND_INV_SRC1_ALPHA;
325
326 default:
327 return blend_coeff_to_d3d_blend(coeff);
328 }
329 }
330
331
blend_equation_to_d3d_op(skgpu::BlendEquation equation)332 static D3D12_BLEND_OP blend_equation_to_d3d_op(skgpu::BlendEquation equation) {
333 switch (equation) {
334 case skgpu::BlendEquation::kAdd:
335 return D3D12_BLEND_OP_ADD;
336 case skgpu::BlendEquation::kSubtract:
337 return D3D12_BLEND_OP_SUBTRACT;
338 case skgpu::BlendEquation::kReverseSubtract:
339 return D3D12_BLEND_OP_REV_SUBTRACT;
340 default:
341 SkUNREACHABLE;
342 }
343 }
344
fill_in_blend_state(const GrPipeline & pipeline,D3D12_BLEND_DESC * blendDesc)345 static void fill_in_blend_state(const GrPipeline& pipeline, D3D12_BLEND_DESC* blendDesc) {
346 blendDesc->AlphaToCoverageEnable = false;
347 blendDesc->IndependentBlendEnable = false;
348
349 const skgpu::BlendInfo& blendInfo = pipeline.getXferProcessor().getBlendInfo();
350
351 skgpu::BlendEquation equation = blendInfo.fEquation;
352 skgpu::BlendCoeff srcCoeff = blendInfo.fSrcBlend;
353 skgpu::BlendCoeff dstCoeff = blendInfo.fDstBlend;
354 bool blendOff = skgpu::BlendShouldDisable(equation, srcCoeff, dstCoeff);
355
356 auto& rtBlend = blendDesc->RenderTarget[0];
357 rtBlend.BlendEnable = !blendOff;
358 if (!blendOff) {
359 rtBlend.SrcBlend = blend_coeff_to_d3d_blend(srcCoeff);
360 rtBlend.DestBlend = blend_coeff_to_d3d_blend(dstCoeff);
361 rtBlend.BlendOp = blend_equation_to_d3d_op(equation);
362 rtBlend.SrcBlendAlpha = blend_coeff_to_d3d_blend_for_alpha(srcCoeff);
363 rtBlend.DestBlendAlpha = blend_coeff_to_d3d_blend_for_alpha(dstCoeff);
364 rtBlend.BlendOpAlpha = blend_equation_to_d3d_op(equation);
365 }
366
367 if (!blendInfo.fWritesColor) {
368 rtBlend.RenderTargetWriteMask = 0;
369 } else {
370 rtBlend.RenderTargetWriteMask = D3D12_COLOR_WRITE_ENABLE_ALL;
371 }
372 }
373
fill_in_rasterizer_state(const GrPipeline & pipeline,bool multisampleEnable,const GrCaps * caps,D3D12_RASTERIZER_DESC * rasterizer)374 static void fill_in_rasterizer_state(const GrPipeline& pipeline,
375 bool multisampleEnable,
376 const GrCaps* caps,
377 D3D12_RASTERIZER_DESC* rasterizer) {
378 rasterizer->FillMode = (caps->wireframeMode() || pipeline.isWireframe()) ?
379 D3D12_FILL_MODE_WIREFRAME : D3D12_FILL_MODE_SOLID;
380 rasterizer->CullMode = D3D12_CULL_MODE_NONE;
381 rasterizer->FrontCounterClockwise = true;
382 rasterizer->DepthBias = 0;
383 rasterizer->DepthBiasClamp = 0.0f;
384 rasterizer->SlopeScaledDepthBias = 0.0f;
385 rasterizer->DepthClipEnable = false;
386 rasterizer->MultisampleEnable = multisampleEnable;
387 rasterizer->AntialiasedLineEnable = false;
388 rasterizer->ForcedSampleCount = 0;
389 rasterizer->ConservativeRaster = D3D12_CONSERVATIVE_RASTERIZATION_MODE_OFF;
390 }
391
stencil_op_to_d3d_op(GrStencilOp op)392 static D3D12_STENCIL_OP stencil_op_to_d3d_op(GrStencilOp op) {
393 switch (op) {
394 case GrStencilOp::kKeep:
395 return D3D12_STENCIL_OP_KEEP;
396 case GrStencilOp::kZero:
397 return D3D12_STENCIL_OP_ZERO;
398 case GrStencilOp::kReplace:
399 return D3D12_STENCIL_OP_REPLACE;
400 case GrStencilOp::kInvert:
401 return D3D12_STENCIL_OP_INVERT;
402 case GrStencilOp::kIncWrap:
403 return D3D12_STENCIL_OP_INCR;
404 case GrStencilOp::kDecWrap:
405 return D3D12_STENCIL_OP_DECR;
406 case GrStencilOp::kIncClamp:
407 return D3D12_STENCIL_OP_INCR_SAT;
408 case GrStencilOp::kDecClamp:
409 return D3D12_STENCIL_OP_DECR_SAT;
410 }
411 SkUNREACHABLE;
412 }
413
stencil_test_to_d3d_func(GrStencilTest test)414 static D3D12_COMPARISON_FUNC stencil_test_to_d3d_func(GrStencilTest test) {
415 switch (test) {
416 case GrStencilTest::kAlways:
417 return D3D12_COMPARISON_FUNC_ALWAYS;
418 case GrStencilTest::kNever:
419 return D3D12_COMPARISON_FUNC_NEVER;
420 case GrStencilTest::kGreater:
421 return D3D12_COMPARISON_FUNC_GREATER;
422 case GrStencilTest::kGEqual:
423 return D3D12_COMPARISON_FUNC_GREATER_EQUAL;
424 case GrStencilTest::kLess:
425 return D3D12_COMPARISON_FUNC_LESS;
426 case GrStencilTest::kLEqual:
427 return D3D12_COMPARISON_FUNC_LESS_EQUAL;
428 case GrStencilTest::kEqual:
429 return D3D12_COMPARISON_FUNC_EQUAL;
430 case GrStencilTest::kNotEqual:
431 return D3D12_COMPARISON_FUNC_NOT_EQUAL;
432 }
433 SkUNREACHABLE;
434 }
435
setup_stencilop_desc(D3D12_DEPTH_STENCILOP_DESC * desc,const GrStencilSettings::Face & stencilFace)436 static void setup_stencilop_desc(D3D12_DEPTH_STENCILOP_DESC* desc,
437 const GrStencilSettings::Face& stencilFace) {
438 desc->StencilFailOp = stencil_op_to_d3d_op(stencilFace.fFailOp);
439 desc->StencilDepthFailOp = desc->StencilFailOp;
440 desc->StencilPassOp = stencil_op_to_d3d_op(stencilFace.fPassOp);
441 desc->StencilFunc = stencil_test_to_d3d_func(stencilFace.fTest);
442 }
443
fill_in_depth_stencil_state(const GrProgramInfo & programInfo,D3D12_DEPTH_STENCIL_DESC * dsDesc)444 static void fill_in_depth_stencil_state(const GrProgramInfo& programInfo,
445 D3D12_DEPTH_STENCIL_DESC* dsDesc) {
446 GrStencilSettings stencilSettings = programInfo.nonGLStencilSettings();
447 GrSurfaceOrigin origin = programInfo.origin();
448
449 dsDesc->DepthEnable = false;
450 dsDesc->DepthWriteMask = D3D12_DEPTH_WRITE_MASK_ZERO;
451 dsDesc->DepthFunc = D3D12_COMPARISON_FUNC_NEVER;
452 dsDesc->StencilEnable = !stencilSettings.isDisabled();
453 if (!stencilSettings.isDisabled()) {
454 if (stencilSettings.isTwoSided()) {
455 const auto& frontFace = stencilSettings.postOriginCCWFace(origin);
456 const auto& backFace = stencilSettings.postOriginCWFace(origin);
457
458 SkASSERT(frontFace.fTestMask == backFace.fTestMask);
459 SkASSERT(frontFace.fWriteMask == backFace.fWriteMask);
460 dsDesc->StencilReadMask = frontFace.fTestMask;
461 dsDesc->StencilWriteMask = frontFace.fWriteMask;
462
463 setup_stencilop_desc(&dsDesc->FrontFace, frontFace);
464 setup_stencilop_desc(&dsDesc->BackFace, backFace);
465 } else {
466 dsDesc->StencilReadMask = stencilSettings.singleSidedFace().fTestMask;
467 dsDesc->StencilWriteMask = stencilSettings.singleSidedFace().fWriteMask;
468 setup_stencilop_desc(&dsDesc->FrontFace, stencilSettings.singleSidedFace());
469 dsDesc->BackFace = dsDesc->FrontFace;
470 }
471 }
472 }
473
gr_primitive_type_to_d3d(GrPrimitiveType primitiveType)474 static D3D12_PRIMITIVE_TOPOLOGY_TYPE gr_primitive_type_to_d3d(GrPrimitiveType primitiveType) {
475 switch (primitiveType) {
476 case GrPrimitiveType::kTriangles:
477 case GrPrimitiveType::kTriangleStrip: //fall through
478 return D3D12_PRIMITIVE_TOPOLOGY_TYPE_TRIANGLE;
479 case GrPrimitiveType::kPoints:
480 return D3D12_PRIMITIVE_TOPOLOGY_TYPE_POINT;
481 case GrPrimitiveType::kLines: // fall through
482 case GrPrimitiveType::kLineStrip:
483 return D3D12_PRIMITIVE_TOPOLOGY_TYPE_LINE;
484 default:
485 SkUNREACHABLE;
486 }
487 }
488
create_pipeline_state(GrD3DGpu * gpu,const GrProgramInfo & programInfo,const sk_sp<GrD3DRootSignature> & rootSig,gr_cp<ID3DBlob> vertexShader,gr_cp<ID3DBlob> pixelShader,DXGI_FORMAT renderTargetFormat,DXGI_FORMAT depthStencilFormat,unsigned int sampleQualityPattern)489 gr_cp<ID3D12PipelineState> create_pipeline_state(
490 GrD3DGpu* gpu, const GrProgramInfo& programInfo, const sk_sp<GrD3DRootSignature>& rootSig,
491 gr_cp<ID3DBlob> vertexShader, gr_cp<ID3DBlob> pixelShader,
492 DXGI_FORMAT renderTargetFormat, DXGI_FORMAT depthStencilFormat,
493 unsigned int sampleQualityPattern) {
494 D3D12_GRAPHICS_PIPELINE_STATE_DESC psoDesc = {};
495
496 psoDesc.pRootSignature = rootSig->rootSignature();
497
498 psoDesc.VS = { reinterpret_cast<UINT8*>(vertexShader->GetBufferPointer()),
499 vertexShader->GetBufferSize() };
500 psoDesc.PS = { reinterpret_cast<UINT8*>(pixelShader->GetBufferPointer()),
501 pixelShader->GetBufferSize() };
502
503 psoDesc.StreamOutput = { nullptr, 0, nullptr, 0, 0 };
504
505 fill_in_blend_state(programInfo.pipeline(), &psoDesc.BlendState);
506 psoDesc.SampleMask = UINT_MAX;
507
508 fill_in_rasterizer_state(programInfo.pipeline(), programInfo.numSamples() > 1, gpu->caps(),
509 &psoDesc.RasterizerState);
510
511 fill_in_depth_stencil_state(programInfo, &psoDesc.DepthStencilState);
512
513 unsigned int totalAttributeCnt = programInfo.geomProc().numVertexAttributes() +
514 programInfo.geomProc().numInstanceAttributes();
515 AutoSTArray<4, D3D12_INPUT_ELEMENT_DESC> inputElements(totalAttributeCnt);
516 setup_vertex_input_layout(programInfo.geomProc(), inputElements.get());
517
518 psoDesc.InputLayout = { inputElements.get(), totalAttributeCnt };
519
520 psoDesc.IBStripCutValue = D3D12_INDEX_BUFFER_STRIP_CUT_VALUE_DISABLED;
521
522 // This is for geometry or hull shader primitives
523 psoDesc.PrimitiveTopologyType = gr_primitive_type_to_d3d(programInfo.primitiveType());
524
525 psoDesc.NumRenderTargets = 1;
526
527 psoDesc.RTVFormats[0] = renderTargetFormat;
528
529 psoDesc.DSVFormat = depthStencilFormat;
530
531 unsigned int numSamples = programInfo.numSamples();
532 psoDesc.SampleDesc = { numSamples, sampleQualityPattern };
533
534 // Only used for multi-adapter systems.
535 psoDesc.NodeMask = 0;
536
537 psoDesc.CachedPSO = { nullptr, 0 };
538 psoDesc.Flags = D3D12_PIPELINE_STATE_FLAG_NONE;
539
540 gr_cp<ID3D12PipelineState> pipelineState;
541 {
542 TRACE_EVENT0("skia.shaders", "CreateGraphicsPipelineState");
543 GR_D3D_CALL_ERRCHECK(
544 gpu->device()->CreateGraphicsPipelineState(&psoDesc, IID_PPV_ARGS(&pipelineState)));
545 }
546
547 return pipelineState;
548 }
549
550 static constexpr SkFourByteTag kHLSL_Tag = SkSetFourByteTag('H', 'L', 'S', 'L');
551 static constexpr SkFourByteTag kSKSL_Tag = SkSetFourByteTag('S', 'K', 'S', 'L');
552
finalize()553 std::unique_ptr<GrD3DPipelineState> GrD3DPipelineStateBuilder::finalize() {
554 TRACE_EVENT0("skia.shaders", TRACE_FUNC);
555
556 this->finalizeShaders();
557
558 SkSL::ProgramSettings settings;
559 settings.fSharpenTextures = true;
560 settings.fRTFlipOffset = fUniformHandler.getRTFlipOffset();
561 settings.fRTFlipBinding = 0;
562 settings.fRTFlipSet = 0;
563
564 sk_sp<SkData> cached;
565 SkReadBuffer reader;
566 SkFourByteTag shaderType = 0;
567 auto persistentCache = fGpu->getContext()->priv().getPersistentCache();
568 if (persistentCache) {
569 // Shear off the D3D-specific portion of the Desc to get the persistent key. We only cache
570 // shader code, not entire pipelines.
571 sk_sp<SkData> key =
572 SkData::MakeWithoutCopy(this->desc().asKey(), this->desc().initialKeyLength());
573 cached = persistentCache->load(*key);
574 if (cached) {
575 reader.setMemory(cached->data(), cached->size());
576 shaderType = GrPersistentCacheUtils::GetType(&reader);
577 }
578 }
579
580 const GrGeometryProcessor& geomProc = this->geometryProcessor();
581 gr_cp<ID3DBlob> shaders[kGrShaderTypeCount];
582
583 if (kHLSL_Tag == shaderType && this->loadHLSLFromCache(&reader, shaders)) {
584 // We successfully loaded and compiled HLSL
585 } else {
586 SkSL::Program::Inputs inputs[kGrShaderTypeCount];
587 std::string* sksl[kGrShaderTypeCount] = {
588 &fVS.fCompilerString,
589 &fFS.fCompilerString,
590 };
591 std::string cached_sksl[kGrShaderTypeCount];
592 std::string hlsl[kGrShaderTypeCount];
593
594 if (kSKSL_Tag == shaderType) {
595 if (GrPersistentCacheUtils::UnpackCachedShaders(&reader, cached_sksl, inputs,
596 kGrShaderTypeCount)) {
597 for (int i = 0; i < kGrShaderTypeCount; ++i) {
598 sksl[i] = &cached_sksl[i];
599 }
600 }
601 }
602
603 auto compile = [&](SkSL::ProgramKind kind, GrShaderType shaderType) {
604 shaders[shaderType] = this->compileD3DProgram(kind, *sksl[shaderType], settings,
605 &inputs[shaderType], &hlsl[shaderType]);
606 return shaders[shaderType].get();
607 };
608
609 if (!compile(SkSL::ProgramKind::kVertex, kVertex_GrShaderType) ||
610 !compile(SkSL::ProgramKind::kFragment, kFragment_GrShaderType)) {
611 return nullptr;
612 }
613
614 if (persistentCache && !cached) {
615 const bool cacheSkSL = fGpu->getContext()->priv().options().fShaderCacheStrategy ==
616 GrContextOptions::ShaderCacheStrategy::kSkSL;
617 if (cacheSkSL) {
618 // Replace the HLSL with formatted SkSL to be cached. This looks odd, but this is
619 // the last time we're going to use these strings, so it's safe.
620 for (int i = 0; i < kGrShaderTypeCount; ++i) {
621 hlsl[i] = SkShaderUtils::PrettyPrint(*sksl[i]);
622 }
623 }
624 sk_sp<SkData> key =
625 SkData::MakeWithoutCopy(this->desc().asKey(), this->desc().initialKeyLength());
626 SkString description = GrProgramDesc::Describe(fProgramInfo, *this->caps());
627 sk_sp<SkData> data = GrPersistentCacheUtils::PackCachedShaders(
628 cacheSkSL ? kSKSL_Tag : kHLSL_Tag, hlsl, inputs, kGrShaderTypeCount);
629 persistentCache->store(*key, *data, description);
630 }
631 }
632
633 sk_sp<GrD3DRootSignature> rootSig =
634 fGpu->resourceProvider().findOrCreateRootSignature(fUniformHandler.fTextures.count());
635 if (!rootSig) {
636 return nullptr;
637 }
638
639 const GrD3DRenderTarget* rt = static_cast<const GrD3DRenderTarget*>(fRenderTarget);
640 gr_cp<ID3D12PipelineState> pipelineState = create_pipeline_state(
641 fGpu, fProgramInfo, rootSig, std::move(shaders[kVertex_GrShaderType]),
642 std::move(shaders[kFragment_GrShaderType]),
643 rt->dxgiFormat(), rt->stencilDxgiFormat(), rt->sampleQualityPattern());
644 sk_sp<GrD3DPipeline> pipeline = GrD3DPipeline::Make(std::move(pipelineState));
645
646 return std::unique_ptr<GrD3DPipelineState>(
647 new GrD3DPipelineState(std::move(pipeline),
648 std::move(rootSig),
649 fUniformHandles,
650 fUniformHandler.fUniforms,
651 fUniformHandler.fCurrentUBOOffset,
652 fUniformHandler.fSamplers.count(),
653 std::move(fGPImpl),
654 std::move(fXPImpl),
655 std::move(fFPImpls),
656 geomProc.vertexStride(),
657 geomProc.instanceStride()));
658 }
659
660
MakeComputePipeline(GrD3DGpu * gpu,GrD3DRootSignature * rootSig,const char * shader)661 sk_sp<GrD3DPipeline> GrD3DPipelineStateBuilder::MakeComputePipeline(GrD3DGpu* gpu,
662 GrD3DRootSignature* rootSig,
663 const char* shader) {
664 D3D12_COMPUTE_PIPELINE_STATE_DESC psoDesc = {};
665 psoDesc.pRootSignature = rootSig->rootSignature();
666
667 // compile shader
668 gr_cp<ID3DBlob> shaderBlob;
669 {
670 TRACE_EVENT0("skia.shaders", "driver_compile_shader");
671 uint32_t compileFlags = 0;
672 #ifdef SK_DEBUG
673 // Enable better shader debugging with the graphics debugging tools.
674 compileFlags |= D3DCOMPILE_DEBUG | D3DCOMPILE_SKIP_OPTIMIZATION;
675 #endif
676
677 gr_cp<ID3DBlob> errors;
678 HRESULT hr = D3DCompile(shader, strlen(shader), nullptr, nullptr, nullptr, "main",
679 "cs_5_1", compileFlags, 0, &shaderBlob, &errors);
680 if (!SUCCEEDED(hr)) {
681 gpu->getContext()->priv().getShaderErrorHandler()->compileError(
682 shader, reinterpret_cast<char*>(errors->GetBufferPointer()));
683 return nullptr;
684 }
685 psoDesc.CS = { reinterpret_cast<UINT8*>(shaderBlob->GetBufferPointer()),
686 shaderBlob->GetBufferSize() };
687 }
688
689 // Only used for multi-adapter systems.
690 psoDesc.NodeMask = 0;
691
692 psoDesc.CachedPSO = { nullptr, 0 };
693 psoDesc.Flags = D3D12_PIPELINE_STATE_FLAG_NONE;
694
695 gr_cp<ID3D12PipelineState> pipelineState;
696 {
697 TRACE_EVENT0("skia.shaders", "CreateComputePipelineState");
698 GR_D3D_CALL_ERRCHECK(
699 gpu->device()->CreateComputePipelineState(&psoDesc, IID_PPV_ARGS(&pipelineState)));
700 }
701
702 return GrD3DPipeline::Make(std::move(pipelineState));
703 }
704