Hash :
6136cbcb
Author :
Date :
2020-09-23T21:31:05
Metal: Implement transform feedback
- XFB is currently emulated by writing to storage buffers.
- Metal doesn't allow vertex shader to both write to storage buffers and
to stage output (i.e clip position). So if GL_RASTERIZER_DISCARD is
NOT enabled, the draw with XFB enabled will have 2 passes:
+ First pass: vertex shader writes to XFB buffers + not write to stage
output + disable rasterizer.
+ Second pass: vertex shader writes to stage output (i.e.
[[position]]) + enable rasterizer. If GL_RASTERIZER_DISCARD is
enabled, the second pass is omitted.
+ This effectively executes the same vertex shader twice. TODO:
possible improvement is writing vertex outputs to buffer in first
pass then re-use that buffer as input for second pass which has a
passthrough vertex shader.
- If GL_RASTERIZER_DISCARD is enabled, and XFB is enabled:
+ Only first pass above will be executed, and the render pass will use
an empty 1x1 texture attachment since rasterization is not needed.
- If GL_RASTERIZER_DISCARD is enabled, but XFB is NOT enabled:
+ we still enable Metal rasterizer.
+ but vertex shader must emulate the discard by writing gl_Position =
(-3, -3, -3, 1). This effectively moves the vertex out of clip
space's visible area.
+ This is because GLSL still allows vertex shader to write to stage
output when rasterizer is disabled. However, Metal doesn't allow
that. In Metal, if rasterizer is disabled, then vertex shader must
not write to stage output.
- See src/libANGLE/renderer/metal/doc/TransformFeedback.md for more
details.
Bug: angleproject:2634
Change-Id: I6c700e031052560326b7f660ee7597202d38e6aa
Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2408594
Reviewed-by: Jonah Ryan-Davis <jonahr@google.com>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Jonah Ryan-Davis <jonahr@google.com>
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344
//
// Copyright (c) 2019 The ANGLE Project Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
//
// TranslatorMetal:
// A GLSL-based translator that outputs shaders that fit GL_KHR_vulkan_glsl.
// It takes into account some considerations for Metal backend also.
// The shaders are then fed into glslang to spit out SPIR-V (libANGLE-side).
// See: https://www.khronos.org/registry/vulkan/specs/misc/GL_KHR_vulkan_glsl.txt
//
// The SPIR-V will then be translated to Metal Shading Language later in Metal backend.
//
#include "compiler/translator/TranslatorMetal.h"
#include "angle_gl.h"
#include "common/utilities.h"
#include "compiler/translator/OutputVulkanGLSLForMetal.h"
#include "compiler/translator/StaticType.h"
#include "compiler/translator/tree_ops/InitializeVariables.h"
#include "compiler/translator/tree_util/BuiltIn.h"
#include "compiler/translator/tree_util/FindMain.h"
#include "compiler/translator/tree_util/FindSymbolNode.h"
#include "compiler/translator/tree_util/IntermNode_util.h"
#include "compiler/translator/tree_util/ReplaceArrayOfMatrixVarying.h"
#include "compiler/translator/tree_util/ReplaceVariable.h"
#include "compiler/translator/tree_util/RunAtTheEndOfShader.h"
#include "compiler/translator/util.h"
namespace sh
{
namespace mtl
{
/** extern */
const char kCoverageMaskEnabledConstName[] = "ANGLECoverageMaskEnabled";
const char kRasterizerDiscardEnabledConstName[] = "ANGLERasterizerDisabled";
} // namespace mtl
namespace
{
constexpr ImmutableString kCoverageMaskField = ImmutableString("coverageMask");
constexpr ImmutableString kSampleMaskWriteFuncName = ImmutableString("ANGLEWriteSampleMask");
TIntermBinary *CreateDriverUniformRef(const TVariable *driverUniforms, const char *fieldName)
{
size_t fieldIndex =
FindFieldIndex(driverUniforms->getType().getInterfaceBlock()->fields(), fieldName);
TIntermSymbol *angleUniformsRef = new TIntermSymbol(driverUniforms);
TConstantUnion *uniformIndex = new TConstantUnion;
uniformIndex->setIConst(static_cast<int>(fieldIndex));
TIntermConstantUnion *indexRef =
new TIntermConstantUnion(uniformIndex, *StaticType::GetBasic<EbtInt>());
return new TIntermBinary(EOpIndexDirectInterfaceBlock, angleUniformsRef, indexRef);
}
// Unlike Vulkan having auto viewport flipping extension, in Metal we have to flip gl_Position.y
// manually.
// This operation performs flipping the gl_Position.y using this expression:
// gl_Position.y = gl_Position.y * negViewportScaleY
ANGLE_NO_DISCARD bool AppendVertexShaderPositionYCorrectionToMain(TCompiler *compiler,
TIntermBlock *root,
TSymbolTable *symbolTable,
TIntermSwizzle *negFlipY)
{
// Create a symbol reference to "gl_Position"
const TVariable *position = BuiltInVariable::gl_Position();
TIntermSymbol *positionRef = new TIntermSymbol(position);
// Create a swizzle to "gl_Position.y"
TVector<int> swizzleOffsetY;
swizzleOffsetY.push_back(1);
TIntermSwizzle *positionY = new TIntermSwizzle(positionRef, swizzleOffsetY);
// Create the expression "gl_Position.y * negFlipY"
TIntermBinary *inverseY = new TIntermBinary(EOpMul, positionY->deepCopy(), negFlipY);
// Create the assignment "gl_Position.y = gl_Position.y * negViewportScaleY
TIntermTyped *positionYLHS = positionY->deepCopy();
TIntermBinary *assignment = new TIntermBinary(TOperator::EOpAssign, positionYLHS, inverseY);
// Append the assignment as a statement at the end of the shader.
return RunAtTheEndOfShader(compiler, root, assignment, symbolTable);
}
// Initialize unused varying outputs.
ANGLE_NO_DISCARD bool InitializeUnusedOutputs(TIntermBlock *root,
TSymbolTable *symbolTable,
const InitVariableList &unusedVars)
{
if (unusedVars.empty())
{
return true;
}
TIntermSequence *insertSequence = new TIntermSequence;
for (const sh::ShaderVariable &var : unusedVars)
{
ASSERT(!var.active);
const TIntermSymbol *symbol = FindSymbolNode(root, var.name);
ASSERT(symbol);
TIntermSequence *initCode = CreateInitCode(symbol, false, false, symbolTable);
insertSequence->insert(insertSequence->end(), initCode->begin(), initCode->end());
}
if (insertSequence)
{
TIntermFunctionDefinition *main = FindMain(root);
TIntermSequence *mainSequence = main->getBody()->getSequence();
// Insert init code at the start of main()
mainSequence->insert(mainSequence->begin(), insertSequence->begin(), insertSequence->end());
}
return true;
}
} // anonymous namespace
TranslatorMetal::TranslatorMetal(sh::GLenum type, ShShaderSpec spec) : TranslatorVulkan(type, spec)
{}
bool TranslatorMetal::translate(TIntermBlock *root,
ShCompileOptions compileOptions,
PerformanceDiagnostics *perfDiagnostics)
{
TInfoSinkBase &sink = getInfoSink().obj;
TOutputVulkanGLSL outputGLSL(sink, getArrayIndexClampingStrategy(), getHashFunction(),
getNameMap(), &getSymbolTable(), getShaderType(),
getShaderVersion(), getOutputType(), false, true, compileOptions);
const TVariable *driverUniforms = nullptr;
if (!TranslatorVulkan::translateImpl(root, compileOptions, perfDiagnostics, &driverUniforms,
&outputGLSL))
{
return false;
}
// Replace array of matrix varyings
if (!ReplaceArrayOfMatrixVaryings(this, root, &getSymbolTable()))
{
return false;
}
if (getShaderType() == GL_VERTEX_SHADER)
{
auto negFlipY = getDriverUniformNegFlipYRef(driverUniforms);
// Append gl_Position.y correction to main
if (!AppendVertexShaderPositionYCorrectionToMain(this, root, &getSymbolTable(), negFlipY))
{
return false;
}
// Insert rasterizer discard logic
if (!insertRasterizerDiscardLogic(root))
{
return false;
}
}
else if (getShaderType() == GL_FRAGMENT_SHADER)
{
if (!insertSampleMaskWritingLogic(root, driverUniforms))
{
return false;
}
}
// Initialize unused varying outputs to avoid spirv-cross dead-code removing them in later
// stage. Only do this if SH_INIT_OUTPUT_VARIABLES is not specified.
if ((getShaderType() == GL_VERTEX_SHADER || getShaderType() == GL_GEOMETRY_SHADER_EXT) &&
!(compileOptions & SH_INIT_OUTPUT_VARIABLES))
{
InitVariableList list;
for (const sh::ShaderVariable &var : mOutputVaryings)
{
if (!var.active)
{
list.push_back(var);
}
}
if (!InitializeUnusedOutputs(root, &getSymbolTable(), list))
{
return false;
}
}
// Write translated shader.
root->traverse(&outputGLSL);
return true;
}
// Metal needs to inverse the depth if depthRange is is reverse order, i.e. depth near > depth far
// This is achieved by multiply the depth value with scale value stored in
// driver uniform's depthRange.reserved
bool TranslatorMetal::transformDepthBeforeCorrection(TIntermBlock *root,
const TVariable *driverUniforms)
{
// Create a symbol reference to "gl_Position"
const TVariable *position = BuiltInVariable::gl_Position();
TIntermSymbol *positionRef = new TIntermSymbol(position);
// Create a swizzle to "gl_Position.z"
TVector<int> swizzleOffsetZ = {2};
TIntermSwizzle *positionZ = new TIntermSwizzle(positionRef, swizzleOffsetZ);
// Create a ref to "depthRange.reserved"
TIntermBinary *viewportZScale = getDriverUniformDepthRangeReservedFieldRef(driverUniforms);
// Create the expression "gl_Position.z * depthRange.reserved".
TIntermBinary *zScale = new TIntermBinary(EOpMul, positionZ->deepCopy(), viewportZScale);
// Create the assignment "gl_Position.z = gl_Position.z * depthRange.reserved"
TIntermTyped *positionZLHS = positionZ->deepCopy();
TIntermBinary *assignment = new TIntermBinary(TOperator::EOpAssign, positionZLHS, zScale);
// Append the assignment as a statement at the end of the shader.
return RunAtTheEndOfShader(this, root, assignment, &getSymbolTable());
}
void TranslatorMetal::createAdditionalGraphicsDriverUniformFields(std::vector<TField *> *fieldsOut)
{
// Add coverage mask to driver uniform. Metal doesn't have built-in GL_SAMPLE_COVERAGE_VALUE
// equivalent functionality, needs to emulate it using fragment shader's [[sample_mask]] output
// value.
TField *coverageMaskField =
new TField(new TType(EbtUInt), kCoverageMaskField, TSourceLoc(), SymbolType::AngleInternal);
fieldsOut->push_back(coverageMaskField);
}
// Add sample_mask writing to main, guarded by the specialization constant
// kCoverageMaskEnabledConstName
ANGLE_NO_DISCARD bool TranslatorMetal::insertSampleMaskWritingLogic(TIntermBlock *root,
const TVariable *driverUniforms)
{
TInfoSinkBase &sink = getInfoSink().obj;
TSymbolTable *symbolTable = &getSymbolTable();
// Insert coverageMaskEnabled specialization constant and sample_mask writing function.
sink << "layout (constant_id=0) const bool " << mtl::kCoverageMaskEnabledConstName;
sink << " = false;\n";
sink << "void " << kSampleMaskWriteFuncName << "(uint mask)\n";
sink << "{\n";
sink << " if (" << mtl::kCoverageMaskEnabledConstName << ")\n";
sink << " {\n";
sink << " gl_SampleMask[0] = int(mask);\n";
sink << " }\n";
sink << "}\n";
// Create kCoverageMaskEnabledConstName and kSampleMaskWriteFuncName variable references.
TType *boolType = new TType(EbtBool);
boolType->setQualifier(EvqConst);
TVariable *coverageMaskEnabledVar =
new TVariable(symbolTable, ImmutableString(mtl::kCoverageMaskEnabledConstName), boolType,
SymbolType::AngleInternal);
TFunction *sampleMaskWriteFunc =
new TFunction(symbolTable, kSampleMaskWriteFuncName, SymbolType::AngleInternal,
StaticType::GetBasic<EbtVoid>(), false);
TType *uintType = new TType(EbtUInt);
TVariable *maskArg =
new TVariable(symbolTable, ImmutableString("mask"), uintType, SymbolType::AngleInternal);
sampleMaskWriteFunc->addParameter(maskArg);
// coverageMask
TIntermBinary *coverageMask = CreateDriverUniformRef(driverUniforms, kCoverageMaskField.data());
// Insert this code to the end of main()
// if (ANGLECoverageMaskEnabled)
// {
// ANGLEWriteSampleMask(ANGLEUniforms.coverageMask);
// }
TIntermSequence *args = new TIntermSequence;
args->push_back(coverageMask);
TIntermAggregate *callSampleMaskWriteFunc =
TIntermAggregate::CreateFunctionCall(*sampleMaskWriteFunc, args);
TIntermBlock *callBlock = new TIntermBlock;
callBlock->appendStatement(callSampleMaskWriteFunc);
TIntermSymbol *coverageMaskEnabled = new TIntermSymbol(coverageMaskEnabledVar);
TIntermIfElse *ifCall = new TIntermIfElse(coverageMaskEnabled, callBlock, nullptr);
return RunAtTheEndOfShader(this, root, ifCall, symbolTable);
}
ANGLE_NO_DISCARD bool TranslatorMetal::insertRasterizerDiscardLogic(TIntermBlock *root)
{
TInfoSinkBase &sink = getInfoSink().obj;
TSymbolTable *symbolTable = &getSymbolTable();
// Insert rasterizationDisabled specialization constant.
sink << "layout (constant_id=0) const bool " << mtl::kRasterizerDiscardEnabledConstName;
sink << " = false;\n";
// Create kRasterizerDiscardEnabledConstName variable reference.
TType *boolType = new TType(EbtBool);
boolType->setQualifier(EvqConst);
TVariable *discardEnabledVar =
new TVariable(symbolTable, ImmutableString(mtl::kRasterizerDiscardEnabledConstName),
boolType, SymbolType::AngleInternal);
// Insert this code to the end of main()
// if (ANGLERasterizerDisabled)
// {
// gl_Position = vec4(-3.0, -3.0, -3.0, 1.0);
// }
// Create a symbol reference to "gl_Position"
const TVariable *position = BuiltInVariable::gl_Position();
TIntermSymbol *positionRef = new TIntermSymbol(position);
// Create vec4(-3, -3, -3, 1):
auto vec4Type = new TType(EbtFloat, 4);
TIntermSequence *vec4Args = new TIntermSequence();
vec4Args->push_back(CreateFloatNode(-3.0f));
vec4Args->push_back(CreateFloatNode(-3.0f));
vec4Args->push_back(CreateFloatNode(-3.0f));
vec4Args->push_back(CreateFloatNode(1.0f));
TIntermAggregate *constVarConstructor =
TIntermAggregate::CreateConstructor(*vec4Type, vec4Args);
// Create the assignment "gl_Position = vec4(-3, -3, -3, 1)"
TIntermBinary *assignment =
new TIntermBinary(TOperator::EOpAssign, positionRef->deepCopy(), constVarConstructor);
TIntermBlock *discardBlock = new TIntermBlock;
discardBlock->appendStatement(assignment);
TIntermSymbol *discardEnabled = new TIntermSymbol(discardEnabledVar);
TIntermIfElse *ifCall = new TIntermIfElse(discardEnabled, discardBlock, nullptr);
return RunAtTheEndOfShader(this, root, ifCall, symbolTable);
}
} // namespace sh