Sampler allocation fix
Because samplers can be stored within structures, register
allocation of entire structures within the samplers registers
was pushing sampler indices outside of the allowed limit.
In order to solve this, sampler registers now exclusively
contain samplers, and utility functions to compute sampler
only type size were added to make this possible.
Bug chromium:797264
Change-Id: Ic5a6f09665c39661944444cd736547bce4dff2ab
Reviewed-on: https://swiftshader-review.googlesource.com/15728
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
diff --git a/src/OpenGL/compiler/OutputASM.cpp b/src/OpenGL/compiler/OutputASM.cpp
index ad79503..019a47f 100644
--- a/src/OpenGL/compiler/OutputASM.cpp
+++ b/src/OpenGL/compiler/OutputASM.cpp
@@ -1946,9 +1946,9 @@
const TFieldList& fields = type.getStruct() ? type.getStruct()->fields() : type.getInterfaceBlock()->fields();
int elements = 0;
- for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
+ for(const auto &field : fields)
{
- const TType &fieldType = *((*field)->type());
+ const TType &fieldType = *(field->type());
if(fieldType.totalRegisterCount() <= registers)
{
@@ -1998,9 +1998,9 @@
const TFieldList& fields = type.getStruct() ? type.getStruct()->fields() : type.getInterfaceBlock()->fields();
int elements = 0;
- for(TFieldList::const_iterator field = fields.begin(); field != fields.end(); field++)
+ for(const auto &field : fields)
{
- const TType &fieldType = *((*field)->type());
+ const TType &fieldType = *(field->type());
if(fieldType.totalRegisterCount() <= registers)
{
@@ -2800,9 +2800,9 @@
{
const TFieldList &fields = type.getStruct()->fields();
int fieldVar = var;
- for(size_t i = 0; i < fields.size(); i++)
+ for(const auto &field : fields)
{
- const TType& fieldType = *(fields[i]->type());
+ const TType& fieldType = *(field->type());
setPixelShaderInputs(fieldType, fieldVar, flat);
fieldVar += fieldType.totalRegisterCount();
}
@@ -2894,10 +2894,10 @@
int fieldRegisterIndex = registerIndex;
const TFieldList &fields = type.getStruct()->fields();
- for(size_t i = 0; i < fields.size(); i++)
+ for(const auto &field : fields)
{
- const TType& fieldType = *(fields[i]->type());
- declareVarying(fieldType, varyingName + "." + fields[i]->name(), fieldRegisterIndex);
+ const TType& fieldType = *(field->type());
+ declareVarying(fieldType, varyingName + "." + field->name(), fieldRegisterIndex);
if(fieldRegisterIndex >= 0)
{
fieldRegisterIndex += fieldType.totalRegisterCount();
@@ -2951,7 +2951,7 @@
int blockMemberIndex = blockMemberLookup(type, name, index);
if(blockMemberIndex == -1)
{
- declareUniform(type, name, index);
+ declareUniform(type, name, index, false);
}
else
{
@@ -3052,7 +3052,7 @@
{
case EOpIndexDirect:
ASSERT(left->isArray());
- offset = index * leftType.elementRegisterCount();
+ offset = index * leftType.samplerRegisterCount();
break;
case EOpIndexDirectStruct:
ASSERT(leftType.isStruct());
@@ -3061,7 +3061,7 @@
for(int i = 0; i < index; i++)
{
- offset += fields[i]->type()->totalRegisterCount();
+ offset += fields[i]->type()->totalSamplerRegisterCount();
}
}
break;
@@ -3096,12 +3096,12 @@
if(index == -1)
{
- index = allocate(samplers, sampler);
+ index = allocate(samplers, sampler, true);
if(sampler->getQualifier() == EvqUniform)
{
const char *name = sampler->getSymbol().c_str();
- declareUniform(type, name, index);
+ declareUniform(type, name, index, true);
}
}
@@ -3187,13 +3187,13 @@
return -1;
}
- int OutputASM::allocate(VariableArray &list, TIntermTyped *variable)
+ int OutputASM::allocate(VariableArray &list, TIntermTyped *variable, bool samplersOnly)
{
int index = lookup(list, variable);
if(index == -1)
{
- unsigned int registerCount = variable->blockRegisterCount();
+ unsigned int registerCount = variable->blockRegisterCount(samplersOnly);
for(unsigned int i = 0; i < list.size(); i++)
{
@@ -3281,7 +3281,7 @@
return -1;
}
- void OutputASM::declareUniform(const TType &type, const TString &name, int registerIndex, int blockId, BlockLayoutEncoder* encoder)
+ void OutputASM::declareUniform(const TType &type, const TString &name, int registerIndex, bool samplersOnly, int blockId, BlockLayoutEncoder* encoder)
{
const TStructure *structure = type.getStruct();
const TInterfaceBlock *block = (type.isInterfaceBlock() || (blockId == -1)) ? type.getInterfaceBlock() : nullptr;
@@ -3296,15 +3296,19 @@
shaderObject->activeUniformBlocks[blockId].fields.push_back(activeUniforms.size());
}
int fieldRegisterIndex = encoder ? shaderObject->activeUniformBlocks[blockId].registerIndex + BlockLayoutEncoder::getBlockRegister(blockInfo) : registerIndex;
- activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(),
- fieldRegisterIndex, blockId, blockInfo));
- if(IsSampler(type.getBasicType()))
+ bool isSampler = IsSampler(type.getBasicType());
+ if(isSampler && samplersOnly)
{
for(int i = 0; i < type.totalRegisterCount(); i++)
{
shader->declareSampler(fieldRegisterIndex + i);
}
}
+ if(!isSampler || samplersOnly)
+ {
+ activeUniforms.push_back(Uniform(glVariableType(type), glVariablePrecision(type), name.c_str(), type.getArraySize(),
+ fieldRegisterIndex, blockId, blockInfo));
+ }
}
else if(block)
{
@@ -3322,10 +3326,10 @@
Std140BlockEncoder currentBlockEncoder(isRowMajor);
currentBlockEncoder.enterAggregateType();
- for(size_t i = 0; i < fields.size(); i++)
+ for(const auto &field : fields)
{
- const TType &fieldType = *(fields[i]->type());
- const TString &fieldName = fields[i]->name();
+ const TType &fieldType = *(field->type());
+ const TString &fieldName = field->name();
if(isUniformBlockMember && (fieldName == name))
{
registerIndex = fieldRegisterIndex;
@@ -3333,7 +3337,7 @@
const TString uniformName = block->hasInstanceName() ? blockName + "." + fieldName : fieldName;
- declareUniform(fieldType, uniformName, fieldRegisterIndex, blockId, ¤tBlockEncoder);
+ declareUniform(fieldType, uniformName, fieldRegisterIndex, samplersOnly, blockId, ¤tBlockEncoder);
fieldRegisterIndex += fieldType.totalRegisterCount();
}
currentBlockEncoder.exitAggregateType();
@@ -3352,14 +3356,14 @@
{
encoder->enterAggregateType();
}
- for(size_t j = 0; j < fields.size(); j++)
+ for(const auto &field : fields)
{
- const TType &fieldType = *(fields[j]->type());
- const TString &fieldName = fields[j]->name();
+ const TType &fieldType = *(field->type());
+ const TString &fieldName = field->name();
const TString uniformName = name + "[" + str(i) + "]." + fieldName;
- declareUniform(fieldType, uniformName, fieldRegisterIndex, blockId, encoder);
- fieldRegisterIndex += fieldType.totalRegisterCount();
+ declareUniform(fieldType, uniformName, fieldRegisterIndex, samplersOnly, blockId, encoder);
+ fieldRegisterIndex += samplersOnly ? fieldType.totalSamplerRegisterCount() : fieldType.totalRegisterCount();
}
if(encoder)
{
@@ -3373,14 +3377,14 @@
{
encoder->enterAggregateType();
}
- for(size_t i = 0; i < fields.size(); i++)
+ for(const auto &field : fields)
{
- const TType &fieldType = *(fields[i]->type());
- const TString &fieldName = fields[i]->name();
+ const TType &fieldType = *(field->type());
+ const TString &fieldName = field->name();
const TString uniformName = name + "." + fieldName;
- declareUniform(fieldType, uniformName, fieldRegisterIndex, blockId, encoder);
- fieldRegisterIndex += fieldType.totalRegisterCount();
+ declareUniform(fieldType, uniformName, fieldRegisterIndex, samplersOnly, blockId, encoder);
+ fieldRegisterIndex += samplersOnly ? fieldType.totalSamplerRegisterCount() : fieldType.totalRegisterCount();
}
if(encoder)
{
diff --git a/src/OpenGL/compiler/OutputASM.h b/src/OpenGL/compiler/OutputASM.h
index 584e0d9..72bf974 100644
--- a/src/OpenGL/compiler/OutputASM.h
+++ b/src/OpenGL/compiler/OutputASM.h
@@ -305,10 +305,10 @@
int lookup(VariableArray &list, TIntermTyped *variable);
int lookup(VariableArray &list, TInterfaceBlock *block);
int blockMemberLookup(const TType &type, const TString &name, int registerIndex);
- int allocate(VariableArray &list, TIntermTyped *variable);
+ int allocate(VariableArray &list, TIntermTyped *variable, bool samplersOnly = false);
void free(VariableArray &list, TIntermTyped *variable);
- void declareUniform(const TType &type, const TString &name, int registerIndex, int blockId = -1, BlockLayoutEncoder* encoder = nullptr);
+ void declareUniform(const TType &type, const TString &name, int registerIndex, bool samplersOnly, int blockId = -1, BlockLayoutEncoder* encoder = nullptr);
GLenum glVariableType(const TType &type);
GLenum glVariablePrecision(const TType &type);
diff --git a/src/OpenGL/compiler/Types.h b/src/OpenGL/compiler/Types.h
index d4d98d0..960352f 100644
--- a/src/OpenGL/compiler/Types.h
+++ b/src/OpenGL/compiler/Types.h
@@ -318,6 +318,24 @@
}
}
+ int samplerRegisterCount() const
+ {
+ if(structure)
+ {
+ int registerCount = 0;
+
+ const TFieldList& fields = isInterfaceBlock() ? interfaceBlock->fields() : structure->fields();
+ for(size_t i = 0; i < fields.size(); i++)
+ {
+ registerCount += fields[i]->type()->totalSamplerRegisterCount();
+ }
+
+ return registerCount;
+ }
+
+ return IsSampler(getBasicType()) ? 1 : 0;
+ }
+
int elementRegisterCount() const
{
if(structure || isInterfaceBlock())
@@ -360,6 +378,18 @@
return totalRegisterCount();
}
+ int totalSamplerRegisterCount() const
+ {
+ if(array)
+ {
+ return arraySize * samplerRegisterCount();
+ }
+ else
+ {
+ return samplerRegisterCount();
+ }
+ }
+
int totalRegisterCount() const
{
if(array)
diff --git a/src/OpenGL/compiler/intermediate.h b/src/OpenGL/compiler/intermediate.h
index 7757d89..d277255 100644
--- a/src/OpenGL/compiler/intermediate.h
+++ b/src/OpenGL/compiler/intermediate.h
@@ -330,7 +330,7 @@
TString getCompleteString() const { return type.getCompleteString(); }
int totalRegisterCount() const { return type.totalRegisterCount(); }
- int blockRegisterCount() const { return type.blockRegisterCount(); }
+ int blockRegisterCount(bool samplersOnly) const { return samplersOnly ? type.totalSamplerRegisterCount() : type.blockRegisterCount(); }
int elementRegisterCount() const { return type.elementRegisterCount(); }
int registerSize() const { return type.registerSize(); }
int getArraySize() const { return type.getArraySize(); }
diff --git a/src/Shader/Shader.cpp b/src/Shader/Shader.cpp
index f7ce731..549b8bd 100644
--- a/src/Shader/Shader.cpp
+++ b/src/Shader/Shader.cpp
@@ -1474,7 +1474,10 @@
void Shader::declareSampler(int i)
{
- usedSamplers |= 1 << i;
+ if(i >= 0 && i < 16)
+ {
+ usedSamplers |= 1 << i;
+ }
}
const Shader::Instruction *Shader::getInstruction(size_t i) const
diff --git a/src/Shader/VertexProgram.cpp b/src/Shader/VertexProgram.cpp
index 53f3692..02b699b 100644
--- a/src/Shader/VertexProgram.cpp
+++ b/src/Shader/VertexProgram.cpp
@@ -1586,7 +1586,7 @@
void VertexProgram::TEXSIZE(Vector4f &dst, Float4 &lod, const Src &src1)
{
- Pointer<Byte> texture = data + OFFSET(DrawData, mipmap[16]) + src1.index * sizeof(Texture);
+ Pointer<Byte> texture = data + OFFSET(DrawData, mipmap[TEXTURE_IMAGE_UNITS]) + src1.index * sizeof(Texture);
dst = SamplerCore::textureSize(texture, lod);
}
