Skip to content

S
swiftshader
Commit 9b4b63bf by Alexis Hetu Committed by Alexis Hétu
Options

Increase precision of 8 bit textures

Only use the upper 8 bits of the 16 bit fields when reading and outputting 8 bit values in order to get exact representations of x/255 for each possible 8 bit value. Bug: swiftshader:122 Bug: b/129523279 Change-Id: I0cd54eb325c24f2fc5a53f12a0b90696024ca9d1 Tests: dEQP-VK.memory.pipeline_barrier.all.* Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/33028Tested-by: 's avatarAlexis Hétu <sugoi@google.com> Reviewed-by: 's avatarChris Forbes <chrisforbes@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
parent 67180a07
Showing with 31 additions and 21 deletions
src/Pipeline/SamplerCore.cpp
...@@ -178,6 +178,17 @@ namespace sw ...@@ -178,6 +178,17 @@ namespace sw
c.z *= Float4(1.0f / 0x7F00); c.z *= Float4(1.0f / 0x7F00);
c.w *= Float4(1.0f / 0x7F00); c.w *= Float4(1.0f / 0x7F00);
break; break;
case VK_FORMAT_R8_UNORM:
case VK_FORMAT_R8G8_UNORM:
case VK_FORMAT_R8G8B8A8_UNORM:
case VK_FORMAT_A8B8G8R8_UNORM_PACK32:
case VK_FORMAT_B8G8R8A8_SRGB:
case VK_FORMAT_R8G8B8A8_SRGB:
c.x *= Float4(1.0f / 0xFF00u);
c.y *= Float4(1.0f / 0xFF00u);
c.z *= Float4(1.0f / 0xFF00u);
c.w *= Float4(1.0f / 0xFF00u);
break;
default: default:
for (int component = 0; component < textureComponentCount(); component++) for (int component = 0; component < textureComponentCount(); component++)
{ {
...@@ -218,6 +229,17 @@ namespace sw ...@@ -218,6 +229,17 @@ namespace sw
c.z = Float4(cs.z) * Float4(1.0f / 0x7F00); c.z = Float4(cs.z) * Float4(1.0f / 0x7F00);
c.w = Float4(cs.w) * Float4(1.0f / 0x7F00); c.w = Float4(cs.w) * Float4(1.0f / 0x7F00);
break; break;
case VK_FORMAT_R8_UNORM:
case VK_FORMAT_R8G8_UNORM:
case VK_FORMAT_R8G8B8A8_UNORM:
case VK_FORMAT_A8B8G8R8_UNORM_PACK32:
case VK_FORMAT_B8G8R8A8_SRGB:
case VK_FORMAT_R8G8B8A8_SRGB:
c.x = Float4(As<UShort4>(cs.x)) * Float4(1.0f / 0xFF00u);
c.y = Float4(As<UShort4>(cs.y)) * Float4(1.0f / 0xFF00u);
c.z = Float4(As<UShort4>(cs.z)) * Float4(1.0f / 0xFF00u);
c.w = Float4(As<UShort4>(cs.w)) * Float4(1.0f / 0xFF00u);
break;
default: default:
for(int component = 0; component < textureComponentCount(); component++) for(int component = 0; component < textureComponentCount(); component++)
{ {
...@@ -1379,10 +1401,10 @@ namespace sw ...@@ -1379,10 +1401,10 @@ namespace sw
c.x = As<Short4>(UnpackHigh(c.x, c.y)); c.x = As<Short4>(UnpackHigh(c.x, c.y));
c.y = c.z; c.y = c.z;
c.w = c.x; c.w = c.x;
c.z = UnpackLow(As<Byte8>(c.z), As<Byte8>(c.z)); c.z = UnpackLow(As<Byte8>(Short4(0)), As<Byte8>(c.z));
c.y = UnpackHigh(As<Byte8>(c.y), As<Byte8>(c.y)); c.y = UnpackHigh(As<Byte8>(Short4(0)), As<Byte8>(c.y));
c.x = UnpackLow(As<Byte8>(c.x), As<Byte8>(c.x)); c.x = UnpackLow(As<Byte8>(Short4(0)), As<Byte8>(c.x));
c.w = UnpackHigh(As<Byte8>(c.w), As<Byte8>(c.w)); c.w = UnpackHigh(As<Byte8>(Short4(0)), As<Byte8>(c.w));
break; break;
case VK_FORMAT_R8G8B8A8_UNORM: case VK_FORMAT_R8G8B8A8_UNORM:
case VK_FORMAT_R8G8B8A8_SINT: case VK_FORMAT_R8G8B8A8_SINT:
...@@ -1392,10 +1414,10 @@ namespace sw ...@@ -1392,10 +1414,10 @@ namespace sw
c.x = As<Short4>(UnpackLow(c.x, c.y)); c.x = As<Short4>(UnpackLow(c.x, c.y));
c.y = c.x; c.y = c.x;
c.w = c.z; c.w = c.z;
c.x = UnpackLow(As<Byte8>(c.x), As<Byte8>(c.x)); c.x = UnpackLow(As<Byte8>(Short4(0)), As<Byte8>(c.x));
c.y = UnpackHigh(As<Byte8>(c.y), As<Byte8>(c.y)); c.y = UnpackHigh(As<Byte8>(Short4(0)), As<Byte8>(c.y));
c.z = UnpackLow(As<Byte8>(c.z), As<Byte8>(c.z)); c.z = UnpackLow(As<Byte8>(Short4(0)), As<Byte8>(c.z));
c.w = UnpackHigh(As<Byte8>(c.w), As<Byte8>(c.w)); c.w = UnpackHigh(As<Byte8>(Short4(0)), As<Byte8>(c.w));
// Propagate sign bit // Propagate sign bit
if(state.textureFormat == VK_FORMAT_R8G8B8A8_SINT) if(state.textureFormat == VK_FORMAT_R8G8B8A8_SINT)
{ {
...@@ -1404,16 +1426,6 @@ namespace sw ...@@ -1404,16 +1426,6 @@ namespace sw
c.z >>= 8; c.z >>= 8;
c.w >>= 8; c.w >>= 8;
} }
if (state.textureFormat == VK_FORMAT_R8G8B8A8_SNORM)
{
// TODO: avoid populating the low bits at all.
c.x &= Short4(0xFF00u);
c.y &= Short4(0xFF00u);
c.z &= Short4(0xFF00u);
c.w &= Short4(0xFF00u);
}
break; break;
case VK_FORMAT_R8G8B8A8_UINT: case VK_FORMAT_R8G8B8A8_UINT:
c.z = As<Short4>(UnpackHigh(c.x, c.y)); c.z = As<Short4>(UnpackHigh(c.x, c.y));
...@@ -1439,9 +1451,6 @@ namespace sw ...@@ -1439,9 +1451,6 @@ namespace sw
switch(state.textureFormat) switch(state.textureFormat)
{ {
case VK_FORMAT_R8G8_UNORM: case VK_FORMAT_R8G8_UNORM:
c.y = (c.x & Short4(0xFF00u)) | As<Short4>(As<UShort4>(c.x) >> 8);
c.x = (c.x & Short4(0x00FFu)) | (c.x << 8);
break;
case VK_FORMAT_R8G8_SNORM: case VK_FORMAT_R8G8_SNORM:
c.y = (c.x & Short4(0xFF00u)); c.y = (c.x & Short4(0xFF00u));
c.x = (c.x << 8); c.x = (c.x << 8);
...@@ -1481,6 +1490,7 @@ namespace sw ...@@ -1481,6 +1490,7 @@ namespace sw
} }
break; break;
case VK_FORMAT_R8_SNORM: case VK_FORMAT_R8_SNORM:
case VK_FORMAT_R8_UNORM:
// TODO: avoid populating the low bits at all. // TODO: avoid populating the low bits at all.
c.x = Unpack(As<Byte4>(c0)); c.x = Unpack(As<Byte4>(c0));
c.x &= Short4(0xFF00u); c.x &= Short4(0xFF00u);
......
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