Add 6-bit and 5-bit sRGB conversion.
Bug 20891368
Change-Id: I3b7066f20f0e669dc1d3abe9222654318ae3b011
Reviewed-on: https://swiftshader-review.googlesource.com/3114
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>
diff --git a/src/Shader/Constants.cpp b/src/Shader/Constants.cpp
index 88a6a81..c9965ff 100644
--- a/src/Shader/Constants.cpp
+++ b/src/Shader/Constants.cpp
@@ -22,11 +22,6 @@
Constants::Constants()
{
- for(int i = 0; i < 256; i++)
- {
- sRGBtoLinear8[i] = (unsigned short)(sw::sRGBtoLinear((float)i / 0xFF) * 0x1000 + 0.5f);
- }
-
static const unsigned int transposeBit0[16] =
{
0x00000000,
@@ -266,10 +261,25 @@
invMaskD01X[i][3] = ~-(i >> 1 & 1);
}
+ for(int i = 0; i < 256; i++)
+ {
+ sRGBtoLinear8_12[i] = (unsigned short)(sw::sRGBtoLinear((float)i / 0xFF) * 0x1000 + 0.5f);
+ }
+
+ for(int i = 0; i < 64; i++)
+ {
+ sRGBtoLinear6_12[i] = (unsigned short)(sw::sRGBtoLinear((float)i / 0x3F) * 0x1000 + 0.5f);
+ }
+
+ for(int i = 0; i < 32; i++)
+ {
+ sRGBtoLinear5_12[i] = (unsigned short)(sw::sRGBtoLinear((float)i / 0x1F) * 0x1000 + 0.5f);
+ }
+
for(int i = 0; i < 0x1000; i++)
{
- linToSRGB12_16[i] = (unsigned short)(clamp(sw::linearToSRGB((float)i / 0x0FFF) * 0xFFFF + 0.5f, 0.0f, (float)0xFFFF));
- sRGBtoLin12_16[i] = (unsigned short)(clamp(sw::sRGBtoLinear((float)i / 0x0FFF) * 0xFFFF + 0.5f, 0.0f, (float)0xFFFF));
+ linearToSRGB12_16[i] = (unsigned short)(clamp(sw::linearToSRGB((float)i / 0x0FFF) * 0xFFFF + 0.5f, 0.0f, (float)0xFFFF));
+ sRGBtoLinear12_16[i] = (unsigned short)(clamp(sw::sRGBtoLinear((float)i / 0x0FFF) * 0xFFFF + 0.5f, 0.0f, (float)0xFFFF));
}
for(int q = 0; q < 4; q++)
diff --git a/src/Shader/Constants.hpp b/src/Shader/Constants.hpp
index a4adc3e..c556b5e 100644
--- a/src/Shader/Constants.hpp
+++ b/src/Shader/Constants.hpp
@@ -19,9 +19,7 @@
struct Constants
{
Constants();
-
- unsigned short sRGBtoLinear8[256];
-
+
unsigned int transposeBit0[16];
unsigned int transposeBit1[16];
unsigned int transposeBit2[16];
@@ -67,8 +65,12 @@
dword4 maskD01X[4];
dword4 invMaskD01X[4];
- unsigned short linToSRGB12_16[4096];
- unsigned short sRGBtoLin12_16[4096];
+ unsigned short sRGBtoLinear8_12[256];
+ unsigned short sRGBtoLinear6_12[64];
+ unsigned short sRGBtoLinear5_12[32];
+
+ unsigned short linearToSRGB12_16[4096];
+ unsigned short sRGBtoLinear12_16[4096];
// Centroid parameters
float4 sampleX[4][16];
diff --git a/src/Shader/PixelRoutine.cpp b/src/Shader/PixelRoutine.cpp
index dda011a..773ceaf 100644
--- a/src/Shader/PixelRoutine.cpp
+++ b/src/Shader/PixelRoutine.cpp
@@ -2561,7 +2561,7 @@
if(postBlendSRGB && state.writeSRGB)
{
- sRGBtoLinear16_16(r, pixel);
+ sRGBtoLinear16_12_16(r, pixel);
}
// Final Color = ObjectColor * SourceBlendFactor + PixelColor * DestinationBlendFactor
@@ -2682,7 +2682,7 @@
if(postBlendSRGB && state.writeSRGB)
{
- linearToSRGB16_16(r, current);
+ linearToSRGB16_12_16(r, current);
}
if(exactColorRounding)
@@ -4365,7 +4365,7 @@
return Float4(cs) * Float4(1.0f / 0xFFFF);
}
- void PixelRoutine::sRGBtoLinear16_16(Registers &r, Vector4s &c)
+ void PixelRoutine::sRGBtoLinear16_12_16(Registers &r, Vector4s &c)
{
c.x = As<UShort4>(c.x) >> 4;
c.y = As<UShort4>(c.y) >> 4;
@@ -4376,7 +4376,7 @@
void PixelRoutine::sRGBtoLinear12_16(Registers &r, Vector4s &c)
{
- Pointer<Byte> LUT = r.constants + OFFSET(Constants,sRGBtoLin12_16);
+ Pointer<Byte> LUT = r.constants + OFFSET(Constants,sRGBtoLinear12_16);
c.x = Insert(c.x, *Pointer<Short>(LUT + 2 * Int(Extract(c.x, 0))), 0);
c.x = Insert(c.x, *Pointer<Short>(LUT + 2 * Int(Extract(c.x, 1))), 1);
@@ -4394,7 +4394,7 @@
c.z = Insert(c.z, *Pointer<Short>(LUT + 2 * Int(Extract(c.z, 3))), 3);
}
- void PixelRoutine::linearToSRGB16_16(Registers &r, Vector4s &c)
+ void PixelRoutine::linearToSRGB16_12_16(Registers &r, Vector4s &c)
{
c.x = As<UShort4>(c.x) >> 4;
c.y = As<UShort4>(c.y) >> 4;
@@ -4405,7 +4405,7 @@
void PixelRoutine::linearToSRGB12_16(Registers &r, Vector4s &c)
{
- Pointer<Byte> LUT = r.constants + OFFSET(Constants,linToSRGB12_16);
+ Pointer<Byte> LUT = r.constants + OFFSET(Constants,linearToSRGB12_16);
c.x = Insert(c.x, *Pointer<Short>(LUT + 2 * Int(Extract(c.x, 0))), 0);
c.x = Insert(c.x, *Pointer<Short>(LUT + 2 * Int(Extract(c.x, 1))), 1);
diff --git a/src/Shader/PixelRoutine.hpp b/src/Shader/PixelRoutine.hpp
index 8331e43..2bde76f 100644
--- a/src/Shader/PixelRoutine.hpp
+++ b/src/Shader/PixelRoutine.hpp
@@ -198,9 +198,9 @@
Float4 convertUnsigned16(UShort4 cs);
UShort4 convertFixed16(Float4 &cf, bool saturate = true);
void convertFixed16(Vector4s &cs, Vector4f &cf, bool saturate = true);
- void sRGBtoLinear16_16(Registers &r, Vector4s &c);
+ void sRGBtoLinear16_12_16(Registers &r, Vector4s &c);
void sRGBtoLinear12_16(Registers &r, Vector4s &c);
- void linearToSRGB16_16(Registers &r, Vector4s &c);
+ void linearToSRGB16_12_16(Registers &r, Vector4s &c);
void linearToSRGB12_16(Registers &r, Vector4s &c);
Float4 sRGBtoLinear(const Float4 &x);
Float4 linearToSRGB(const Float4 &x);
diff --git a/src/Shader/SamplerCore.cpp b/src/Shader/SamplerCore.cpp
index b6d5838..d43f775 100644
--- a/src/Shader/SamplerCore.cpp
+++ b/src/Shader/SamplerCore.cpp
@@ -110,7 +110,7 @@
{
if(state.sRGB && isRGBComponent(component))
{
- sRGBtoLinear16_12(c[component]); // FIXME: Perform linearization at surface level for read-only textures
+ sRGBtoLinear16_8_12(c[component]); // FIXME: Perform linearization at surface level for read-only textures
}
else
{
@@ -265,7 +265,7 @@
{
if(state.sRGB && isRGBComponent(component))
{
- sRGBtoLinear16_12(cs[component]); // FIXME: Perform linearization at surface level for read-only textures
+ sRGBtoLinear16_8_12(cs[component]); // FIXME: Perform linearization at surface level for read-only textures
convertSigned12(c[component], cs[component]);
}
else
@@ -1775,11 +1775,35 @@
cf = Float4(As<UShort4>(cs)) * Float4(1.0f / 0xFFFF);
}
- void SamplerCore::sRGBtoLinear16_12(Short4 &c)
+ void SamplerCore::sRGBtoLinear16_8_12(Short4 &c)
{
c = As<UShort4>(c) >> 8;
- Pointer<Byte> LUT = Pointer<Byte>(constants + OFFSET(Constants,sRGBtoLinear8));
+ Pointer<Byte> LUT = Pointer<Byte>(constants + OFFSET(Constants,sRGBtoLinear8_12));
+
+ c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 0))), 0);
+ c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 1))), 1);
+ c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 2))), 2);
+ c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 3))), 3);
+ }
+
+ void SamplerCore::sRGBtoLinear16_6_12(Short4 &c)
+ {
+ c = As<UShort4>(c) >> 10;
+
+ Pointer<Byte> LUT = Pointer<Byte>(constants + OFFSET(Constants,sRGBtoLinear6_12));
+
+ c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 0))), 0);
+ c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 1))), 1);
+ c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 2))), 2);
+ c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 3))), 3);
+ }
+
+ void SamplerCore::sRGBtoLinear16_5_12(Short4 &c)
+ {
+ c = As<UShort4>(c) >> 11;
+
+ Pointer<Byte> LUT = Pointer<Byte>(constants + OFFSET(Constants,sRGBtoLinear5_12));
c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 0))), 0);
c = Insert(c, *Pointer<Short>(LUT + 2 * Int(Extract(c, 1))), 1);
diff --git a/src/Shader/SamplerCore.hpp b/src/Shader/SamplerCore.hpp
index 7adb464..8a4874c 100644
--- a/src/Shader/SamplerCore.hpp
+++ b/src/Shader/SamplerCore.hpp
@@ -53,7 +53,9 @@
void convertSigned12(Float4 &cf, Short4 &ci);
void convertSigned15(Float4 &cf, Short4 &ci);
void convertUnsigned16(Float4 &cf, Short4 &ci);
- void sRGBtoLinear16_12(Short4 &c);
+ void sRGBtoLinear16_8_12(Short4 &c);
+ void sRGBtoLinear16_6_12(Short4 &c);
+ void sRGBtoLinear16_5_12(Short4 &c);
bool hasFloatTexture() const;
bool hasUnsignedTextureComponent(int component) const;
