Fix for the remaining vertex array issues
Fixes all failures in dEQP.functional.vertex_arrays.*
In OpenGL ES 3.0, vertex attributes streams can be of a
different type from the vertex attribute in the shader.
For this reason, some conversion may be required. This
cl solves this issue by:
1) Sending the information about the vertex attribute's
type in the shader to the vertex routine.
2) Handling this information by adding conversion where
appropriate.
Change-Id: I04a5a34aea12684209e584aa5f15a3edfd57f956
Reviewed-on: https://swiftshader-review.googlesource.com/7254
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
diff --git a/src/OpenGL/compiler/OutputASM.cpp b/src/OpenGL/compiler/OutputASM.cpp
index 29da8e4..4cc379a 100644
--- a/src/OpenGL/compiler/OutputASM.cpp
+++ b/src/OpenGL/compiler/OutputASM.cpp
@@ -2890,12 +2890,25 @@
index = allocate(attributes, attribute);
const TType &type = attribute->getType();
int registerCount = attribute->totalRegisterCount();
+ sw::VertexShader::AttribType attribType = sw::VertexShader::ATTRIBTYPE_FLOAT;
+ switch(type.getBasicType())
+ {
+ case EbtInt:
+ attribType = sw::VertexShader::ATTRIBTYPE_INT;
+ break;
+ case EbtUInt:
+ attribType = sw::VertexShader::ATTRIBTYPE_UINT;
+ break;
+ case EbtFloat:
+ default:
+ break;
+ }
if(vertexShader && (index + registerCount) <= sw::MAX_VERTEX_INPUTS)
{
for(int i = 0; i < registerCount; i++)
{
- vertexShader->setInput(index + i, sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i));
+ vertexShader->setInput(index + i, sw::Shader::Semantic(sw::Shader::USAGE_TEXCOORD, index + i, false), attribType);
}
}
diff --git a/src/Renderer/VertexProcessor.cpp b/src/Renderer/VertexProcessor.cpp
index ced0870..a64c010 100644
--- a/src/Renderer/VertexProcessor.cpp
+++ b/src/Renderer/VertexProcessor.cpp
@@ -958,6 +958,7 @@
state.input[i].type = context->input[i].type;
state.input[i].count = context->input[i].count;
state.input[i].normalized = context->input[i].normalized;
+ state.input[i].attribType = context->vertexShader ? context->vertexShader->getAttribType(i) : VertexShader::ATTRIBTYPE_FLOAT;
}
if(!context->vertexShader)
diff --git a/src/Renderer/VertexProcessor.hpp b/src/Renderer/VertexProcessor.hpp
index f0cf124..278c9b1 100644
--- a/src/Renderer/VertexProcessor.hpp
+++ b/src/Renderer/VertexProcessor.hpp
@@ -18,6 +18,7 @@
#include "Matrix.hpp"
#include "Context.hpp"
#include "RoutineCache.hpp"
+#include "Shader/VertexShader.hpp"
namespace sw
{
@@ -102,6 +103,7 @@
StreamType type : BITS(STREAMTYPE_LAST);
unsigned int count : 3;
bool normalized : 1;
+ unsigned int attribType : BITS(VertexShader::ATTRIBTYPE_LAST);
};
struct Output
diff --git a/src/Shader/VertexRoutine.cpp b/src/Shader/VertexRoutine.cpp
index c67a8bc..087e481 100644
--- a/src/Shader/VertexRoutine.cpp
+++ b/src/Shader/VertexRoutine.cpp
@@ -148,6 +148,8 @@
Pointer<Byte> source2 = source1 + (!textureSampling ? stride : 0);
Pointer<Byte> source3 = source2 + (!textureSampling ? stride : 0);
+ bool isNativeFloatAttrib = (stream.attribType == VertexShader::ATTRIBTYPE_FLOAT) || stream.normalized;
+
switch(stream.type)
{
case STREAMTYPE_FLOAT:
@@ -156,25 +158,47 @@
{
// Null stream, all default components
}
- else if(stream.count == 1)
- {
- v.x.x = *Pointer<Float>(source0);
- v.x.y = *Pointer<Float>(source1);
- v.x.z = *Pointer<Float>(source2);
- v.x.w = *Pointer<Float>(source3);
- }
else
{
- v.x = *Pointer<Float4>(source0);
- v.y = *Pointer<Float4>(source1);
- v.z = *Pointer<Float4>(source2);
- v.w = *Pointer<Float4>(source3);
+ if(stream.count == 1)
+ {
+ v.x.x = *Pointer<Float>(source0);
+ v.x.y = *Pointer<Float>(source1);
+ v.x.z = *Pointer<Float>(source2);
+ v.x.w = *Pointer<Float>(source3);
+ }
+ else
+ {
+ v.x = *Pointer<Float4>(source0);
+ v.y = *Pointer<Float4>(source1);
+ v.z = *Pointer<Float4>(source2);
+ v.w = *Pointer<Float4>(source3);
- transpose4xN(v.x, v.y, v.z, v.w, stream.count);
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
+ }
+
+ switch(stream.attribType)
+ {
+ case VertexShader::ATTRIBTYPE_INT:
+ if(stream.count >= 1) v.x = As<Float4>(Int4(v.x));
+ if(stream.count >= 2) v.x = As<Float4>(Int4(v.y));
+ if(stream.count >= 3) v.x = As<Float4>(Int4(v.z));
+ if(stream.count >= 4) v.x = As<Float4>(Int4(v.w));
+ break;
+ case VertexShader::ATTRIBTYPE_UINT:
+ if(stream.count >= 1) v.x = As<Float4>(UInt4(v.x));
+ if(stream.count >= 2) v.x = As<Float4>(UInt4(v.y));
+ if(stream.count >= 3) v.x = As<Float4>(UInt4(v.z));
+ if(stream.count >= 4) v.x = As<Float4>(UInt4(v.w));
+ break;
+ default:
+ break;
+ }
}
}
break;
case STREAMTYPE_BYTE:
+ if(isNativeFloatAttrib) // Stream: UByte, Shader attrib: Float
{
v.x = Float4(*Pointer<Byte4>(source0));
v.y = Float4(*Pointer<Byte4>(source1));
@@ -191,8 +215,18 @@
if(stream.count >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants,unscaleByte));
}
}
+ else // Stream: UByte, Shader attrib: Int / UInt
+ {
+ v.x = As<Float4>(Int4(*Pointer<Byte4>(source0)));
+ v.y = As<Float4>(Int4(*Pointer<Byte4>(source1)));
+ v.z = As<Float4>(Int4(*Pointer<Byte4>(source2)));
+ v.w = As<Float4>(Int4(*Pointer<Byte4>(source3)));
+
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
+ }
break;
case STREAMTYPE_SBYTE:
+ if(isNativeFloatAttrib) // Stream: SByte, Shader attrib: Float
{
v.x = Float4(*Pointer<SByte4>(source0));
v.y = Float4(*Pointer<SByte4>(source1));
@@ -209,6 +243,15 @@
if(stream.count >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants,unscaleSByte));
}
}
+ else // Stream: SByte, Shader attrib: Int / UInt
+ {
+ v.x = As<Float4>(Int4(*Pointer<SByte4>(source0)));
+ v.y = As<Float4>(Int4(*Pointer<SByte4>(source1)));
+ v.z = As<Float4>(Int4(*Pointer<SByte4>(source2)));
+ v.w = As<Float4>(Int4(*Pointer<SByte4>(source3)));
+
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
+ }
break;
case STREAMTYPE_COLOR:
{
@@ -226,6 +269,7 @@
}
break;
case STREAMTYPE_SHORT:
+ if(isNativeFloatAttrib) // Stream: Int, Shader attrib: Float
{
v.x = Float4(*Pointer<Short4>(source0));
v.y = Float4(*Pointer<Short4>(source1));
@@ -242,8 +286,18 @@
if(stream.count >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants,unscaleShort));
}
}
+ else // Stream: Short, Shader attrib: Int/UInt, no type conversion
+ {
+ v.x = As<Float4>(Int4(*Pointer<Short4>(source0)));
+ v.y = As<Float4>(Int4(*Pointer<Short4>(source1)));
+ v.z = As<Float4>(Int4(*Pointer<Short4>(source2)));
+ v.w = As<Float4>(Int4(*Pointer<Short4>(source3)));
+
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
+ }
break;
case STREAMTYPE_USHORT:
+ if(isNativeFloatAttrib) // Stream: Int, Shader attrib: Float
{
v.x = Float4(*Pointer<UShort4>(source0));
v.y = Float4(*Pointer<UShort4>(source1));
@@ -260,59 +314,70 @@
if(stream.count >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants,unscaleUShort));
}
}
+ else // Stream: UShort, Shader attrib: Int/UInt, no type conversion
+ {
+ v.x = As<Float4>(Int4(*Pointer<UShort4>(source0)));
+ v.y = As<Float4>(Int4(*Pointer<UShort4>(source1)));
+ v.z = As<Float4>(Int4(*Pointer<UShort4>(source2)));
+ v.w = As<Float4>(Int4(*Pointer<UShort4>(source3)));
+
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
+ }
break;
case STREAMTYPE_INT:
+ if(isNativeFloatAttrib) // Stream: Int, Shader attrib: Float
{
+ v.x = Float4(*Pointer<Int4>(source0));
+ v.y = Float4(*Pointer<Int4>(source1));
+ v.z = Float4(*Pointer<Int4>(source2));
+ v.w = Float4(*Pointer<Int4>(source3));
+
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
+
if(stream.normalized)
{
- v.x = Float4(*Pointer<Int4>(source0));
- v.y = Float4(*Pointer<Int4>(source1));
- v.z = Float4(*Pointer<Int4>(source2));
- v.w = Float4(*Pointer<Int4>(source3));
-
- transpose4xN(v.x, v.y, v.z, v.w, stream.count);
-
if(stream.count >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleInt));
if(stream.count >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleInt));
if(stream.count >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleInt));
if(stream.count >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleInt));
}
- else
- {
- v.x = As<Float4>(*Pointer<Int4>(source0));
- v.y = As<Float4>(*Pointer<Int4>(source1));
- v.z = As<Float4>(*Pointer<Int4>(source2));
- v.w = As<Float4>(*Pointer<Int4>(source3));
+ }
+ else // Stream: Int, Shader attrib: Int/UInt, no type conversion
+ {
+ v.x = *Pointer<Float4>(source0);
+ v.y = *Pointer<Float4>(source1);
+ v.z = *Pointer<Float4>(source2);
+ v.w = *Pointer<Float4>(source3);
- transpose4xN(v.x, v.y, v.z, v.w, stream.count);
- }
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
}
break;
case STREAMTYPE_UINT:
+ if(isNativeFloatAttrib) // Stream: UInt, Shader attrib: Float
{
+ v.x = Float4(*Pointer<UInt4>(source0));
+ v.y = Float4(*Pointer<UInt4>(source1));
+ v.z = Float4(*Pointer<UInt4>(source2));
+ v.w = Float4(*Pointer<UInt4>(source3));
+
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
+
if(stream.normalized)
{
- v.x = Float4(*Pointer<UInt4>(source0));
- v.y = Float4(*Pointer<UInt4>(source1));
- v.z = Float4(*Pointer<UInt4>(source2));
- v.w = Float4(*Pointer<UInt4>(source3));
-
- transpose4xN(v.x, v.y, v.z, v.w, stream.count);
-
if(stream.count >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUInt));
if(stream.count >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUInt));
if(stream.count >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUInt));
if(stream.count >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUInt));
}
- else
- {
- v.x = As<Float4>(*Pointer<UInt4>(source0));
- v.y = As<Float4>(*Pointer<UInt4>(source1));
- v.z = As<Float4>(*Pointer<UInt4>(source2));
- v.w = As<Float4>(*Pointer<UInt4>(source3));
+ }
+ else // Stream: UInt, Shader attrib: Int/UInt, no type conversion
+ {
+ v.x = *Pointer<Float4>(source0);
+ v.y = *Pointer<Float4>(source1);
+ v.z = *Pointer<Float4>(source2);
+ v.w = *Pointer<Float4>(source3);
- transpose4xN(v.x, v.y, v.z, v.w, stream.count);
- }
+ transpose4xN(v.x, v.y, v.z, v.w, stream.count);
}
break;
case STREAMTYPE_UDEC3:
@@ -538,7 +603,7 @@
if(stream.count < 1) v.x = Float4(0.0f);
if(stream.count < 2) v.y = Float4(0.0f);
if(stream.count < 3) v.z = Float4(0.0f);
- if(stream.count < 4) v.w = Float4(1.0f);
+ if(stream.count < 4) v.w = isNativeFloatAttrib ? Float4(1.0f) : As<Float4>(Int4(0));
return v;
}
diff --git a/src/Shader/VertexShader.cpp b/src/Shader/VertexShader.cpp
index dd62b4b..a98932b 100644
--- a/src/Shader/VertexShader.cpp
+++ b/src/Shader/VertexShader.cpp
@@ -32,6 +32,7 @@
for(int i = 0; i < MAX_VERTEX_INPUTS; i++)
{
input[i] = Semantic();
+ attribType[i] = ATTRIBTYPE_FLOAT;
}
if(vs) // Make a copy
@@ -43,6 +44,7 @@
memcpy(output, vs->output, sizeof(output));
memcpy(input, vs->input, sizeof(input));
+ memcpy(attribType, vs->attribType, sizeof(attribType));
positionRegister = vs->positionRegister;
pointSizeRegister = vs->pointSizeRegister;
instanceIdDeclared = vs->instanceIdDeclared;
@@ -65,6 +67,7 @@
for(int i = 0; i < MAX_VERTEX_INPUTS; i++)
{
input[i] = Semantic();
+ attribType[i] = ATTRIBTYPE_FLOAT;
}
optimize();
@@ -151,9 +154,10 @@
return textureSampling;
}
- void VertexShader::setInput(int inputIdx, const sw::Shader::Semantic& semantic)
+ void VertexShader::setInput(int inputIdx, const sw::Shader::Semantic& semantic, AttribType aType)
{
input[inputIdx] = semantic;
+ attribType[inputIdx] = aType;
}
void VertexShader::setOutput(int outputIdx, int nbComponents, const sw::Shader::Semantic& semantic)
@@ -181,6 +185,11 @@
return input[inputIdx];
}
+ VertexShader::AttribType VertexShader::getAttribType(int inputIdx) const
+ {
+ return attribType[inputIdx];
+ }
+
const sw::Shader::Semantic& VertexShader::getOutput(int outputIdx, int component) const
{
return output[outputIdx][component];
diff --git a/src/Shader/VertexShader.hpp b/src/Shader/VertexShader.hpp
index 4eb6786..0ca7b93 100644
--- a/src/Shader/VertexShader.hpp
+++ b/src/Shader/VertexShader.hpp
@@ -23,6 +23,15 @@
class VertexShader : public Shader
{
public:
+ enum AttribType : unsigned char
+ {
+ ATTRIBTYPE_FLOAT,
+ ATTRIBTYPE_INT,
+ ATTRIBTYPE_UINT,
+
+ ATTRIBTYPE_LAST = ATTRIBTYPE_UINT
+ };
+
explicit VertexShader(const VertexShader *vs = 0);
explicit VertexShader(const unsigned long *token);
@@ -31,7 +40,7 @@
static int validate(const unsigned long *const token); // Returns number of instructions if valid
bool containsTextureSampling() const;
- void setInput(int inputIdx, const Semantic& semantic);
+ void setInput(int inputIdx, const Semantic& semantic, AttribType attribType = ATTRIBTYPE_FLOAT);
void setOutput(int outputIdx, int nbComponents, const Semantic& semantic);
void setPositionRegister(int posReg);
void setPointSizeRegister(int ptSizeReg);
@@ -39,6 +48,7 @@
const Semantic& getInput(int inputIdx) const;
const Semantic& getOutput(int outputIdx, int component) const;
+ AttribType getAttribType(int inputIndex) const;
int getPositionRegister() const { return positionRegister; }
int getPointSizeRegister() const { return pointSizeRegister; }
bool isInstanceIdDeclared() const { return instanceIdDeclared; }
@@ -52,6 +62,8 @@
Semantic input[MAX_VERTEX_INPUTS];
Semantic output[MAX_VERTEX_OUTPUTS][4];
+ AttribType attribType[MAX_VERTEX_INPUTS];
+
int positionRegister;
int pointSizeRegister;
