Clamp unsigned floating-point formats.
GL_R11F_G11F_B10F is a renderable format (in the EXT_color_buffer_float
extension) with unsigned floating-point values, which we implement using
(signed) half- or single-precision floating-point types. So we need to
clamp values to a positive range before writing to them.
Change-Id: Ic21a5b0b33905c0aeab35299fc268158f8c679f9
Reviewed-on: https://swiftshader-review.googlesource.com/15448
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
diff --git a/src/OpenGL/common/Image.cpp b/src/OpenGL/common/Image.cpp
index 6d63b47..0a7a3fe 100644
--- a/src/OpenGL/common/Image.cpp
+++ b/src/OpenGL/common/Image.cpp
@@ -668,7 +668,7 @@
return sw::FORMAT_X32B32G32R32F;
case GL_R11F_G11F_B10F:
case GL_RGB9_E5:
- return sw::FORMAT_X16B16G16R16F;
+ return sw::FORMAT_X16B16G16R16F_UNSIGNED;
case GL_RGBA:
case GL_RGBA32F:
return sw::FORMAT_A32B32G32R32F;
@@ -710,9 +710,10 @@
return sw::FORMAT_A16B16G16R16F;
case GL_RGB:
case GL_RGB16F:
+ return sw::FORMAT_X16B16G16R16F;
case GL_R11F_G11F_B10F:
case GL_RGB9_E5:
- return sw::FORMAT_X16B16G16R16F;
+ return sw::FORMAT_X16B16G16R16F_UNSIGNED;
default:
UNREACHABLE(format);
}
@@ -916,7 +917,7 @@
}
case GL_UNSIGNED_INT_10F_11F_11F_REV:
case GL_UNSIGNED_INT_5_9_9_9_REV: // 5 is the exponent field, not alpha.
- return sw::FORMAT_X16B16G16R16F;
+ return sw::FORMAT_X16B16G16R16F_UNSIGNED;
default:
UNREACHABLE(type);
}
diff --git a/src/OpenGL/libGLESv2/utilities.cpp b/src/OpenGL/libGLESv2/utilities.cpp
index e068b6a..b43d32f 100644
--- a/src/OpenGL/libGLESv2/utilities.cpp
+++ b/src/OpenGL/libGLESv2/utilities.cpp
@@ -2204,8 +2204,8 @@
case GL_RGBA32UI: return sw::FORMAT_A32B32G32R32UI;
case GL_R16F: return sw::FORMAT_R16F;
case GL_RG16F: return sw::FORMAT_G16R16F;
- case GL_R11F_G11F_B10F:
- case GL_RGB16F: return sw::FORMAT_B16G16R16F;
+ case GL_R11F_G11F_B10F: return sw::FORMAT_X16B16G16R16F_UNSIGNED;
+ case GL_RGB16F: return sw::FORMAT_X16B16G16R16F;
case GL_RGBA16F: return sw::FORMAT_A16B16G16R16F;
case GL_R32F: return sw::FORMAT_R32F;
case GL_RG32F: return sw::FORMAT_G32R32F;
diff --git a/src/Renderer/Blitter.cpp b/src/Renderer/Blitter.cpp
index 7045462..1a1f36b 100644
--- a/src/Renderer/Blitter.cpp
+++ b/src/Renderer/Blitter.cpp
@@ -364,6 +364,7 @@
c = *Pointer<Float4>(element);
break;
case FORMAT_X32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_B32G32R32F:
c.z = *Pointer<Float>(element + 8);
case FORMAT_G32R32F:
@@ -514,6 +515,7 @@
}
break;
case FORMAT_X32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
if(writeA) { *Pointer<Float>(element + 12) = 1.0f; }
case FORMAT_B32G32R32F:
if(writeR) { *Pointer<Float>(element) = c.x; }
@@ -1032,6 +1034,7 @@
case FORMAT_A32B32G32R32UI:
case FORMAT_A32B32G32R32F:
case FORMAT_X32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_B32G32R32F:
case FORMAT_G32R32F:
case FORMAT_R32F:
@@ -1125,7 +1128,11 @@
value *= Float4(scale.x / unscale.x, scale.y / unscale.y, scale.z / unscale.z, scale.w / unscale.w);
}
- if(Surface::isFloatFormat(state.sourceFormat) && !Surface::isFloatFormat(state.destFormat))
+ if(state.destFormat == FORMAT_X32B32G32R32F_UNSIGNED)
+ {
+ value = Max(value, Float4(0.0f)); // TODO: Only necessary if source is signed.
+ }
+ else if(Surface::isFloatFormat(state.sourceFormat) && !Surface::isFloatFormat(state.destFormat))
{
value = Min(value, Float4(scale.x, scale.y, scale.z, scale.w));
@@ -1344,10 +1351,10 @@
}
}
- if(!hasConstantColorI && !hasConstantColorF) { x += w; }
+ if(!state.clearOperation) { x += w; }
}
- if(!hasConstantColorI && !hasConstantColorF) { y += h; }
+ if(!state.clearOperation) { y += h; }
}
}
diff --git a/src/Renderer/Surface.cpp b/src/Renderer/Surface.cpp
index 4a4d543..4e2e26b 100644
--- a/src/Renderer/Surface.cpp
+++ b/src/Renderer/Surface.cpp
@@ -304,8 +304,12 @@
((half*)element)[0] = (half)r;
((half*)element)[1] = (half)g;
break;
+ case FORMAT_X16B16G16R16F_UNSIGNED:
+ r = max(r, 0.0f); g = max(g, 0.0f); b = max(b, 0.0f);
+ // Fall through to FORMAT_X16B16G16R16F.
case FORMAT_X16B16G16R16F:
((half*)element)[3] = 1.0f;
+ // Fall through to FORMAT_B16G16R16F.
case FORMAT_B16G16R16F:
((half*)element)[0] = (half)r;
((half*)element)[1] = (half)g;
@@ -327,8 +331,12 @@
((float*)element)[0] = r;
((float*)element)[1] = g;
break;
+ case FORMAT_X32B32G32R32F_UNSIGNED:
+ r = max(r, 0.0f); g = max(g, 0.0f); b = max(b, 0.0f);
+ // Fall through to FORMAT_X32B32G32R32F.
case FORMAT_X32B32G32R32F:
((float*)element)[3] = 1.0f;
+ // Fall through to FORMAT_B32G32R32F.
case FORMAT_B32G32R32F:
((float*)element)[0] = r;
((float*)element)[1] = g;
@@ -965,6 +973,7 @@
g = ((half*)element)[1];
break;
case FORMAT_X16B16G16R16F:
+ case FORMAT_X16B16G16R16F_UNSIGNED:
case FORMAT_B16G16R16F:
r = ((half*)element)[0];
g = ((half*)element)[1];
@@ -987,6 +996,7 @@
g = ((float*)element)[1];
break;
case FORMAT_X32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_B32G32R32F:
r = ((float*)element)[0];
g = ((float*)element)[1];
@@ -1659,12 +1669,14 @@
case FORMAT_B16G16R16F: return 6;
case FORMAT_X16B16G16R16F: return 8;
case FORMAT_A16B16G16R16F: return 8;
+ case FORMAT_X16B16G16R16F_UNSIGNED: return 8;
case FORMAT_A32F: return 4;
case FORMAT_R32F: return 4;
case FORMAT_G32R32F: return 8;
case FORMAT_B32G32R32F: return 12;
case FORMAT_X32B32G32R32F: return 16;
case FORMAT_A32B32G32R32F: return 16;
+ case FORMAT_X32B32G32R32F_UNSIGNED: return 16;
// Depth/stencil formats
case FORMAT_D16: return 2;
case FORMAT_D32: return 4;
@@ -2897,11 +2909,13 @@
case FORMAT_B16G16R16F:
case FORMAT_X16B16G16R16F:
case FORMAT_A16B16G16R16F:
+ case FORMAT_X16B16G16R16F_UNSIGNED:
case FORMAT_R32F:
case FORMAT_G32R32F:
case FORMAT_B32G32R32F:
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_D32F:
case FORMAT_D32FS8:
case FORMAT_D32F_COMPLEMENTARY:
@@ -2949,6 +2963,7 @@
case FORMAT_G32R32UI:
case FORMAT_X32B32G32R32UI:
case FORMAT_A32B32G32R32UI:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_R8UI:
case FORMAT_G8R8UI:
case FORMAT_X8B8G8R8UI:
@@ -3234,6 +3249,7 @@
case FORMAT_G32R32F: return 2;
case FORMAT_X32B32G32R32F: return 3;
case FORMAT_A32B32G32R32F: return 4;
+ case FORMAT_X32B32G32R32F_UNSIGNED: return 3;
case FORMAT_D32F: return 1;
case FORMAT_D32FS8: return 1;
case FORMAT_D32F_LOCKABLE: return 1;
@@ -3954,12 +3970,14 @@
case FORMAT_B16G16R16F: return FORMAT_X32B32G32R32F;
case FORMAT_X16B16G16R16F: return FORMAT_X32B32G32R32F;
case FORMAT_A16B16G16R16F: return FORMAT_A32B32G32R32F;
+ case FORMAT_X16B16G16R16F_UNSIGNED: return FORMAT_X32B32G32R32F_UNSIGNED;
case FORMAT_A32F: return FORMAT_A32B32G32R32F;
case FORMAT_R32F: return FORMAT_R32F;
case FORMAT_G32R32F: return FORMAT_G32R32F;
case FORMAT_B32G32R32F: return FORMAT_X32B32G32R32F;
case FORMAT_X32B32G32R32F: return FORMAT_X32B32G32R32F;
case FORMAT_A32B32G32R32F: return FORMAT_A32B32G32R32F;
+ case FORMAT_X32B32G32R32F_UNSIGNED: return FORMAT_X32B32G32R32F_UNSIGNED;
// Luminance formats
case FORMAT_L8: return FORMAT_L8;
case FORMAT_A4L4: return FORMAT_A8L8;
@@ -5533,7 +5551,9 @@
else ASSERT(false);
}
}
- else if(internal.format == FORMAT_A32B32G32R32F || internal.format == FORMAT_X32B32G32R32F)
+ else if(internal.format == FORMAT_A32B32G32R32F ||
+ internal.format == FORMAT_X32B32G32R32F ||
+ internal.format == FORMAT_X32B32G32R32F_UNSIGNED)
{
#if defined(__i386__) || defined(__x86_64__)
if(CPUID::supportsSSE())
diff --git a/src/Renderer/Surface.hpp b/src/Renderer/Surface.hpp
index 7075434..062b0d7 100644
--- a/src/Renderer/Surface.hpp
+++ b/src/Renderer/Surface.hpp
@@ -172,12 +172,14 @@
FORMAT_B16G16R16F,
FORMAT_X16B16G16R16F,
FORMAT_A16B16G16R16F,
+ FORMAT_X16B16G16R16F_UNSIGNED,
FORMAT_A32F,
FORMAT_R32F,
FORMAT_G32R32F,
FORMAT_B32G32R32F,
FORMAT_X32B32G32R32F,
FORMAT_A32B32G32R32F,
+ FORMAT_X32B32G32R32F_UNSIGNED,
// Bump map formats
FORMAT_V8U8,
FORMAT_L6V5U5,
diff --git a/src/Shader/PixelPipeline.cpp b/src/Shader/PixelPipeline.cpp
index 5d4df97..e623735 100644
--- a/src/Shader/PixelPipeline.cpp
+++ b/src/Shader/PixelPipeline.cpp
@@ -356,6 +356,7 @@
case FORMAT_G32R32F:
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ // case FORMAT_X32B32G32R32F_UNSIGNED: // Not renderable in any fixed-function API.
convertSigned12(oC, current);
PixelRoutine::fogBlend(oC, fog);
diff --git a/src/Shader/PixelProgram.cpp b/src/Shader/PixelProgram.cpp
index b55bf34..3e9082d 100644
--- a/src/Shader/PixelProgram.cpp
+++ b/src/Shader/PixelProgram.cpp
@@ -632,6 +632,7 @@
case FORMAT_G32R32F:
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_R32I:
case FORMAT_G32R32I:
case FORMAT_A32B32G32R32I:
@@ -771,6 +772,12 @@
case FORMAT_G8R8UI:
case FORMAT_A8B8G8R8UI:
break;
+ case FORMAT_X32B32G32R32F_UNSIGNED:
+ oC[index].x = Max(oC[index].x, Float4(0.0f));
+ oC[index].y = Max(oC[index].y, Float4(0.0f));
+ oC[index].z = Max(oC[index].z, Float4(0.0f));
+ oC[index].w = Max(oC[index].w, Float4(0.0f));
+ break;
default:
ASSERT(false);
}
@@ -1135,7 +1142,7 @@
{
dst = sampleTexture(src1, src0, bias, (src0), (src0), (src0), Bias);
}
-
+
void PixelProgram::TEXOFFSETBIAS(Vector4f &dst, Vector4f &src0, const Src &src1, Vector4f &offset, Float4 &bias)
{
dst = sampleTexture(src1, src0, bias, (src0), (src0), offset, {Bias, Offset});
diff --git a/src/Shader/PixelRoutine.cpp b/src/Shader/PixelRoutine.cpp
index a5b6f56..1c300b0 100644
--- a/src/Shader/PixelRoutine.cpp
+++ b/src/Shader/PixelRoutine.cpp
@@ -2080,6 +2080,7 @@
break;
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_A32B32G32R32I:
case FORMAT_A32B32G32R32UI:
buffer = cBuffer;
@@ -2089,7 +2090,8 @@
pixel.z = *Pointer<Float4>(buffer + 16 * x, 16);
pixel.w = *Pointer<Float4>(buffer + 16 * x + 16, 16);
transpose4x4(pixel.x, pixel.y, pixel.z, pixel.w);
- if(state.targetFormat[index] == FORMAT_X32B32G32R32F)
+ if(state.targetFormat[index] == FORMAT_X32B32G32R32F ||
+ state.targetFormat[index] == FORMAT_X32B32G32R32F_UNSIGNED)
{
pixel.w = Float4(1.0f);
}
@@ -2241,6 +2243,7 @@
break;
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_A32B32G32R32I:
case FORMAT_A32B32G32R32UI:
case FORMAT_A16B16G16R16I:
@@ -2494,6 +2497,7 @@
break;
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_A32B32G32R32I:
case FORMAT_A32B32G32R32UI:
buffer = cBuffer + 16 * x;
diff --git a/src/Shader/SamplerCore.cpp b/src/Shader/SamplerCore.cpp
index 850516d..b28d6e3 100644
--- a/src/Shader/SamplerCore.cpp
+++ b/src/Shader/SamplerCore.cpp
@@ -236,6 +236,7 @@
case FORMAT_G32R32F:
c.z = Short4(defaultColorValue);
case FORMAT_X32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
c.w = Short4(0x1000);
case FORMAT_A32B32G32R32F:
break;
@@ -464,6 +465,7 @@
case FORMAT_G32R32F:
c.z = Float4(defaultColorValue);
case FORMAT_X32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
c.w = Float4(1.0f);
case FORMAT_A32B32G32R32F:
break;
@@ -2624,6 +2626,7 @@
case FORMAT_G32R32F:
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_A8:
case FORMAT_R8:
case FORMAT_L8:
@@ -2695,6 +2698,7 @@
case FORMAT_G32R32F:
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_D32F:
case FORMAT_D32FS8:
case FORMAT_D32F_LOCKABLE:
@@ -2773,6 +2777,7 @@
case FORMAT_G32R32F:
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_A8:
case FORMAT_R8:
case FORMAT_L8:
@@ -2854,6 +2859,7 @@
case FORMAT_G32R32F:
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_A8:
case FORMAT_R8:
case FORMAT_L8:
@@ -2927,6 +2933,7 @@
case FORMAT_G32R32F:
case FORMAT_X32B32G32R32F:
case FORMAT_A32B32G32R32F:
+ case FORMAT_X32B32G32R32F_UNSIGNED:
case FORMAT_A8:
case FORMAT_R8:
case FORMAT_L8:
@@ -2998,6 +3005,7 @@
case FORMAT_G32R32F: return component < 2;
case FORMAT_X32B32G32R32F: return component < 3;
case FORMAT_A32B32G32R32F: return component < 3;
+ case FORMAT_X32B32G32R32F_UNSIGNED: return component < 3;
case FORMAT_A8: return false;
case FORMAT_R8: return component < 1;
case FORMAT_L8: return component < 1;
