Fixed EAC failures
The ETC2 decoder spec has a slight difference with the EAC decoder
spec when it comes to handling the 0 multiplier corner case.
For ETC2, we have (OpenGL ES 3.0 spec, section C.1.3):
"An encoder is not allowed to produce a multiplier of zero, but
the decoder should still be able to handle also this case (and
produce 0 x modifier = 0 in that case)."
For EAC, we have (OpenGL ES 3.0 spec, section C.1.5):
"If the multiplier value is zero, we should set the multiplier
to 1.0/8.0"
In order to take this into account, the EAC decoded output can no
longer be represented by an 8 bit value, but must be represented by
a minimum of 11 bits, as the spec requires. For now, the EAC decoder
decodes EAC into a 32 bit integer format, which then gets converted
to a 32 bit float format internally.
Eventually, it would be possible for the EAC decoder to decode the
image to a signed 16 bit integer internal format, if it was supported.
Fixes all failures in:
dEQP-GLES3.functional.texture.wrap*
Change-Id: I32106383ade56e375229231ff230a2574791caa6
Reviewed-on: https://swiftshader-review.googlesource.com/15188
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
diff --git a/src/Renderer/ETC_Decoder.cpp b/src/Renderer/ETC_Decoder.cpp
index 8e109f3..71163ea 100644
--- a/src/Renderer/ETC_Decoder.cpp
+++ b/src/Renderer/ETC_Decoder.cpp
@@ -26,6 +26,13 @@
return (value < -128) ? -128 : ((value > 127) ? 127 : value);
}
+ inline int clampEAC(int value, bool isSigned)
+ {
+ int min = isSigned ? -1023 : 0;
+ int max = isSigned ? 1023 : 2047;
+ return (value < min) ? min : ((value > max) ? max : value);
+ }
+
struct bgra8
{
unsigned char b;
@@ -84,33 +91,51 @@
// Decodes unsigned single or dual channel block to bytes
static void DecodeBlock(const ETC2** sources, unsigned char *dest, int nbChannels, int x, int y, int w, int h, int pitch, bool isSigned)
{
- if(isSigned)
+ if(nbChannels <= 2) // EAC
{
- signed char* sDst = reinterpret_cast<signed char*>(dest);
for(int j = 0; j < 4 && (y + j) < h; j++)
{
+ int* sDst = reinterpret_cast<int*>(dest);
for(int i = 0; i < 4 && (x + i) < w; i++)
{
for(int c = nbChannels - 1; c >= 0; c--)
{
- sDst[i * nbChannels + c] = clampSByte(sources[c]->getSingleChannel(i, j, isSigned));
+ sDst[i * nbChannels + c] = clampEAC(sources[c]->getSingleChannel(i, j, isSigned, true), isSigned);
}
}
- sDst += pitch;
+ dest += pitch;
}
}
else
{
- for(int j = 0; j < 4 && (y + j) < h; j++)
+ if(isSigned)
{
- for(int i = 0; i < 4 && (x + i) < w; i++)
+ signed char* sDst = reinterpret_cast<signed char*>(dest);
+ for(int j = 0; j < 4 && (y + j) < h; j++)
{
- for(int c = nbChannels - 1; c >= 0; c--)
+ for(int i = 0; i < 4 && (x + i) < w; i++)
{
- dest[i * nbChannels + c] = clampByte(sources[c]->getSingleChannel(i, j, isSigned));
+ for(int c = nbChannels - 1; c >= 0; c--)
+ {
+ sDst[i * nbChannels + c] = clampSByte(sources[c]->getSingleChannel(i, j, isSigned, false));
+ }
}
+ sDst += pitch;
}
- dest += pitch;
+ }
+ else
+ {
+ for(int j = 0; j < 4 && (y + j) < h; j++)
+ {
+ for(int i = 0; i < 4 && (x + i) < w; i++)
+ {
+ for(int c = nbChannels - 1; c >= 0; c--)
+ {
+ dest[i * nbChannels + c] = clampByte(sources[c]->getSingleChannel(i, j, isSigned, false));
+ }
+ }
+ dest += pitch;
+ }
}
}
}
@@ -591,10 +616,14 @@
}
// Single channel utility functions
- inline int getSingleChannel(int x, int y, bool isSigned) const
+ inline int getSingleChannel(int x, int y, bool isSigned, bool isEAC) const
{
int codeword = isSigned ? signed_base_codeword : base_codeword;
- return codeword + getSingleChannelModifier(x, y) * multiplier;
+ return isEAC ?
+ ((multiplier == 0) ?
+ (codeword * 8 + 4 + getSingleChannelModifier(x, y)) :
+ (codeword * 8 + 4 + getSingleChannelModifier(x, y) * multiplier * 8)) :
+ codeword + getSingleChannelModifier(x, y) * multiplier;
}
inline int getSingleChannelIndex(int x, int y) const
diff --git a/src/Renderer/Surface.cpp b/src/Renderer/Surface.cpp
index b809148..e732fb8 100644
--- a/src/Renderer/Surface.cpp
+++ b/src/Renderer/Surface.cpp
@@ -2584,36 +2584,29 @@
{
ASSERT(nbChannels == 1 || nbChannels == 2);
- ETC_Decoder::Decode((const byte*)external.lockRect(0, 0, 0, LOCK_READONLY), (byte*)internal.lockRect(0, 0, 0, LOCK_WRITEONLY), external.width, external.height, internal.width, internal.height, internal.pitchB, internal.bytes,
+ byte *src = (byte*)internal.lockRect(0, 0, 0, LOCK_READWRITE);
+ ETC_Decoder::Decode((const byte*)external.lockRect(0, 0, 0, LOCK_READONLY), src, external.width, external.height, internal.width, internal.height, internal.pitchB, internal.bytes,
(nbChannels == 1) ? (isSigned ? ETC_Decoder::ETC_R_SIGNED : ETC_Decoder::ETC_R_UNSIGNED) : (isSigned ? ETC_Decoder::ETC_RG_SIGNED : ETC_Decoder::ETC_RG_UNSIGNED));
external.unlockRect();
- internal.unlockRect();
- // FIXME: We convert signed data to float, until signed integer internal formats are supported
- // This code can be removed if signed ETC2 images are decoded to internal 8 bit signed R/RG formats
- if(isSigned)
+ // FIXME: We convert EAC data to float, until signed short internal formats are supported
+ // This code can be removed if ETC2 images are decoded to internal 16 bit signed R/RG formats
+ const float normalization = isSigned ? (1.0f / (8.0f * 127.875f)) : (1.0f / (8.0f * 255.875f));
+ for(int y = 0; y < internal.height; y++)
{
- sbyte *src = (sbyte*)internal.lockRect(0, 0, 0, LOCK_READWRITE);
-
- for(int y = 0; y < internal.height; y++)
+ byte* srcRow = src + y * internal.pitchB;
+ for(int x = internal.width - 1; x >= 0; x--)
{
- sbyte* srcRow = src + y * internal.pitchB;
- for(int x = internal.width - 1; x >= 0; x--)
+ int* srcPix = reinterpret_cast<int*>(srcRow + x * internal.bytes);
+ float* dstPix = reinterpret_cast<float*>(srcPix);
+ for(int c = nbChannels - 1; c >= 0; c--)
{
- int dx = x & 0xFFFFFFFC;
- int mx = x - dx;
- sbyte* srcPix = srcRow + dx * internal.bytes + mx * nbChannels;
- float* dstPix = (float*)(srcRow + x * internal.bytes);
- for(int c = nbChannels - 1; c >= 0; c--)
- {
- static const float normalization = 1.0f / 127.875f;
- dstPix[c] = clamp(static_cast<float>(srcPix[c]) * normalization, -1.0f, 1.0f);
- }
+ dstPix[c] = clamp(static_cast<float>(srcPix[c]) * normalization, -1.0f, 1.0f);
}
}
-
- internal.unlockRect();
}
+
+ internal.unlockRect();
}
void Surface::decodeASTC(Buffer &internal, Buffer &external, int xBlockSize, int yBlockSize, int zBlockSize, bool isSRGB)
@@ -3887,13 +3880,13 @@
// ASTC supports HDR, so a floating point format is required to represent it properly
return FORMAT_A32B32G32R32F; // FIXME: 16FP is probably sufficient, but it's currently unsupported
case FORMAT_ATI1:
- case FORMAT_R11_EAC:
return FORMAT_R8;
+ case FORMAT_R11_EAC:
case FORMAT_SIGNED_R11_EAC:
return FORMAT_R32F; // FIXME: Signed 8bit format would be sufficient
case FORMAT_ATI2:
- case FORMAT_RG11_EAC:
return FORMAT_G8R8;
+ case FORMAT_RG11_EAC:
case FORMAT_SIGNED_RG11_EAC:
return FORMAT_G32R32F; // FIXME: Signed 8bit format would be sufficient
case FORMAT_ETC1:
