Skip to content

S
swiftshader
Commit 12e857a3 by Chris Forbes
Options

Refactor vertex fetch to use VkFormat more aggressively

Bug: b/152218432 Change-Id: I71eb0cbda5ae98ab39d9953fe049368b8341193a Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42748Tested-by: 's avatarChris Forbes <chrisforbes@google.com> Presubmit-Ready: Chris Forbes <chrisforbes@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by: 's avatarNicolas Capens <nicolascapens@google.com>
parent 95b77b44
Showing with 135 additions and 219 deletions
src/Pipeline/VertexRoutine.cpp
......@@ -225,226 +225,137 @@ Vector4f VertexRoutine::readStream(Pointer<Byte> &buffer, UInt &stride, const St
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
}
switch(stream.attribType)
{
case SpirvShader::ATTRIBTYPE_INT:
if(componentCount >= 1) v.x = As<Float4>(Int4(v.x));
if(componentCount >= 2) v.x = As<Float4>(Int4(v.y));
if(componentCount >= 3) v.x = As<Float4>(Int4(v.z));
if(componentCount >= 4) v.x = As<Float4>(Int4(v.w));
break;
case SpirvShader::ATTRIBTYPE_UINT:
if(componentCount >= 1) v.x = As<Float4>(UInt4(v.x));
if(componentCount >= 2) v.x = As<Float4>(UInt4(v.y));
if(componentCount >= 3) v.x = As<Float4>(UInt4(v.z));
if(componentCount >= 4) v.x = As<Float4>(UInt4(v.w));
break;
default:
break;
}
}
}
break;
case VK_FORMAT_B8G8R8A8_UNORM:
bgra = true;
// [[fallthrough]]
case VK_FORMAT_R8_UNORM:
case VK_FORMAT_R8G8_UNORM:
case VK_FORMAT_R8G8B8A8_UNORM:
case VK_FORMAT_A8B8G8R8_UNORM_PACK32:
v.x = Float4(*Pointer<Byte4>(source0));
v.y = Float4(*Pointer<Byte4>(source1));
v.z = Float4(*Pointer<Byte4>(source2));
v.w = Float4(*Pointer<Byte4>(source3));
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleByte));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleByte));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleByte));
if(componentCount >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleByte));
break;
case VK_FORMAT_R8_UINT:
case VK_FORMAT_R8G8_UINT:
case VK_FORMAT_R8G8B8A8_UINT:
case VK_FORMAT_A8B8G8R8_UNORM_PACK32:
case VK_FORMAT_A8B8G8R8_UINT_PACK32:
if(isNativeFloatAttrib) // Stream: UByte, Shader attrib: Float
{
v.x = Float4(*Pointer<Byte4>(source0));
v.y = Float4(*Pointer<Byte4>(source1));
v.z = Float4(*Pointer<Byte4>(source2));
v.w = Float4(*Pointer<Byte4>(source3));
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
if(normalized)
{
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleByte));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleByte));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleByte));
if(componentCount >= 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)));
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, componentCount);
}
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
break;
case VK_FORMAT_R8_SNORM:
case VK_FORMAT_R8_SINT:
case VK_FORMAT_R8G8_SNORM:
case VK_FORMAT_R8G8_SINT:
case VK_FORMAT_R8G8B8A8_SNORM:
case VK_FORMAT_R8G8B8A8_SINT:
case VK_FORMAT_A8B8G8R8_SNORM_PACK32:
case VK_FORMAT_A8B8G8R8_SINT_PACK32:
if(isNativeFloatAttrib) // Stream: SByte, Shader attrib: Float
{
v.x = Float4(*Pointer<SByte4>(source0));
v.y = Float4(*Pointer<SByte4>(source1));
v.z = Float4(*Pointer<SByte4>(source2));
v.w = Float4(*Pointer<SByte4>(source3));
v.x = Float4(*Pointer<SByte4>(source0));
v.y = Float4(*Pointer<SByte4>(source1));
v.z = Float4(*Pointer<SByte4>(source2));
v.w = Float4(*Pointer<SByte4>(source3));
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
if(normalized)
{
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleSByte));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleSByte));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleSByte));
if(componentCount >= 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)));
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleSByte));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleSByte));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleSByte));
if(componentCount >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleSByte));
break;
case VK_FORMAT_R8_SINT:
case VK_FORMAT_R8G8_SINT:
case VK_FORMAT_R8G8B8A8_SINT:
case VK_FORMAT_A8B8G8R8_SINT_PACK32:
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, componentCount);
}
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
break;
case VK_FORMAT_R16_SNORM:
case VK_FORMAT_R16_SINT:
case VK_FORMAT_R16G16_SNORM:
case VK_FORMAT_R16G16_SINT:
case VK_FORMAT_R16G16B16A16_SNORM:
case VK_FORMAT_R16G16B16A16_SINT:
if(isNativeFloatAttrib) // Stream: Int, Shader attrib: Float
{
v.x = Float4(*Pointer<Short4>(source0));
v.y = Float4(*Pointer<Short4>(source1));
v.z = Float4(*Pointer<Short4>(source2));
v.w = Float4(*Pointer<Short4>(source3));
v.x = Float4(*Pointer<Short4>(source0));
v.y = Float4(*Pointer<Short4>(source1));
v.z = Float4(*Pointer<Short4>(source2));
v.w = Float4(*Pointer<Short4>(source3));
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
if(normalized)
{
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleShort));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleShort));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleShort));
if(componentCount >= 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)));
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleShort));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleShort));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleShort));
if(componentCount >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleShort));
break;
case VK_FORMAT_R16_SINT:
case VK_FORMAT_R16G16_SINT:
case VK_FORMAT_R16G16B16A16_SINT:
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, componentCount);
}
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
break;
case VK_FORMAT_R16_UNORM:
case VK_FORMAT_R16_UINT:
case VK_FORMAT_R16G16_UNORM:
case VK_FORMAT_R16G16_UINT:
case VK_FORMAT_R16G16B16A16_UNORM:
case VK_FORMAT_R16G16B16A16_UINT:
if(isNativeFloatAttrib) // Stream: Int, Shader attrib: Float
{
v.x = Float4(*Pointer<UShort4>(source0));
v.y = Float4(*Pointer<UShort4>(source1));
v.z = Float4(*Pointer<UShort4>(source2));
v.w = Float4(*Pointer<UShort4>(source3));
v.x = Float4(*Pointer<UShort4>(source0));
v.y = Float4(*Pointer<UShort4>(source1));
v.z = Float4(*Pointer<UShort4>(source2));
v.w = Float4(*Pointer<UShort4>(source3));
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
if(normalized)
{
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUShort));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUShort));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUShort));
if(componentCount >= 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)));
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUShort));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUShort));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUShort));
if(componentCount >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUShort));
break;
case VK_FORMAT_R16_UINT:
case VK_FORMAT_R16G16_UINT:
case VK_FORMAT_R16G16B16A16_UINT:
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, componentCount);
}
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
break;
case VK_FORMAT_R32_SINT:
case VK_FORMAT_R32G32_SINT:
case VK_FORMAT_R32G32B32_SINT:
case VK_FORMAT_R32G32B32A32_SINT:
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, componentCount);
if(normalized)
{
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleInt));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleInt));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleInt));
if(componentCount >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleInt));
}
}
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);
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, componentCount);
}
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
break;
case VK_FORMAT_R32_UINT:
case VK_FORMAT_R32G32_UINT:
case VK_FORMAT_R32G32B32_UINT:
case VK_FORMAT_R32G32B32A32_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, componentCount);
if(normalized)
{
if(componentCount >= 1) v.x *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUInt));
if(componentCount >= 2) v.y *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUInt));
if(componentCount >= 3) v.z *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUInt));
if(componentCount >= 4) v.w *= *Pointer<Float4>(constants + OFFSET(Constants, unscaleUInt));
}
}
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);
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, componentCount);
}
transpose4xN(v.x, v.y, v.z, v.w, componentCount);
break;
case VK_FORMAT_R16_SFLOAT:
case VK_FORMAT_R16G16_SFLOAT:
......@@ -504,9 +415,29 @@ Vector4f VertexRoutine::readStream(Pointer<Byte> &buffer, UInt &stride, const St
}
break;
case VK_FORMAT_A2R10G10B10_SNORM_PACK32:
case VK_FORMAT_A2R10G10B10_SINT_PACK32:
bgra = true;
// [[fallthrough]]
case VK_FORMAT_A2B10G10R10_SNORM_PACK32:
{
Int4 src;
src = Insert(src, *Pointer<Int>(source0), 0);
src = Insert(src, *Pointer<Int>(source1), 1);
src = Insert(src, *Pointer<Int>(source2), 2);
src = Insert(src, *Pointer<Int>(source3), 3);
v.x = Float4((src << 22) >> 22);
v.y = Float4((src << 12) >> 22);
v.z = Float4((src << 02) >> 22);
v.w = Float4(src >> 30);
v.x = Max(v.x * Float4(1.0f / 0x1FF), Float4(-1.0f));
v.y = Max(v.y * Float4(1.0f / 0x1FF), Float4(-1.0f));
v.z = Max(v.z * Float4(1.0f / 0x1FF), Float4(-1.0f));
v.w = Max(v.w, Float4(-1.0f));
}
break;
case VK_FORMAT_A2R10G10B10_SINT_PACK32:
bgra = true;
// [[fallthrough]]
case VK_FORMAT_A2B10G10R10_SINT_PACK32:
{
Int4 src;
......@@ -514,35 +445,16 @@ Vector4f VertexRoutine::readStream(Pointer<Byte> &buffer, UInt &stride, const St
src = Insert(src, *Pointer<Int>(source1), 1);
src = Insert(src, *Pointer<Int>(source2), 2);
src = Insert(src, *Pointer<Int>(source3), 3);
if(isNativeFloatAttrib) // Stream: Int, Shader attrib: Float
{
v.x = Float4((src << 22) >> 22);
v.y = Float4((src << 12) >> 22);
v.z = Float4((src << 02) >> 22);
v.w = Float4(src >> 30);
if(normalized)
{
v.x = Max(v.x * Float4(1.0f / 0x1FF), Float4(-1.0f));
v.y = Max(v.y * Float4(1.0f / 0x1FF), Float4(-1.0f));
v.z = Max(v.z * Float4(1.0f / 0x1FF), Float4(-1.0f));
v.w = Max(v.w, Float4(-1.0f));
}
}
else // Stream: UInt, Shader attrib: Int/UInt, no type conversion
{
v.x = As<Float4>((src << 22) >> 22);
v.y = As<Float4>((src << 12) >> 22);
v.z = As<Float4>((src << 02) >> 22);
v.w = As<Float4>(src >> 30);
}
v.x = As<Float4>((src << 22) >> 22);
v.y = As<Float4>((src << 12) >> 22);
v.z = As<Float4>((src << 02) >> 22);
v.w = As<Float4>(src >> 30);
}
break;
case VK_FORMAT_A2R10G10B10_UNORM_PACK32:
case VK_FORMAT_A2R10G10B10_UINT_PACK32:
bgra = true;
// [[fallthrough]]
case VK_FORMAT_A2B10G10R10_UNORM_PACK32:
case VK_FORMAT_A2B10G10R10_UINT_PACK32:
{
Int4 src;
src = Insert(src, *Pointer<Int>(source0), 0);
......@@ -550,28 +462,32 @@ Vector4f VertexRoutine::readStream(Pointer<Byte> &buffer, UInt &stride, const St
src = Insert(src, *Pointer<Int>(source2), 2);
src = Insert(src, *Pointer<Int>(source3), 3);
if(isNativeFloatAttrib) // Stream: Int, Shader attrib: Float
{
v.x = Float4(src & Int4(0x3FF));
v.y = Float4((src >> 10) & Int4(0x3FF));
v.z = Float4((src >> 20) & Int4(0x3FF));
v.w = Float4((src >> 30) & Int4(0x3));
v.x = Float4(src & Int4(0x3FF));
v.y = Float4((src >> 10) & Int4(0x3FF));
v.z = Float4((src >> 20) & Int4(0x3FF));
v.w = Float4((src >> 30) & Int4(0x3));
if(normalized)
{
v.x *= Float4(1.0f / 0x3FF);
v.y *= Float4(1.0f / 0x3FF);
v.z *= Float4(1.0f / 0x3FF);
v.w *= Float4(1.0f / 0x3);
}
}
else // Stream: UInt, Shader attrib: Int/UInt, no type conversion
{
v.x = As<Float4>(src & Int4(0x3FF));
v.y = As<Float4>((src >> 10) & Int4(0x3FF));
v.z = As<Float4>((src >> 20) & Int4(0x3FF));
v.w = As<Float4>((src >> 30) & Int4(0x3));
}
v.x *= Float4(1.0f / 0x3FF);
v.y *= Float4(1.0f / 0x3FF);
v.z *= Float4(1.0f / 0x3FF);
v.w *= Float4(1.0f / 0x3);
}
break;
case VK_FORMAT_A2R10G10B10_UINT_PACK32:
bgra = true;
// [[fallthrough]]
case VK_FORMAT_A2B10G10R10_UINT_PACK32:
{
Int4 src;
src = Insert(src, *Pointer<Int>(source0), 0);
src = Insert(src, *Pointer<Int>(source1), 1);
src = Insert(src, *Pointer<Int>(source2), 2);
src = Insert(src, *Pointer<Int>(source3), 3);
v.x = As<Float4>(src & Int4(0x3FF));
v.y = As<Float4>((src >> 10) & Int4(0x3FF));
v.z = As<Float4>((src >> 20) & Int4(0x3FF));
v.w = As<Float4>((src >> 30) & Int4(0x3));
}
break;
default:
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment