d3d12: Clean up texture upload code

Should be easier to read code (and spot bugs).
Fix crash with mipmap and DXTCn texture format.
This commit is contained in:
vlj 2015-06-17 00:06:58 +02:00 committed by Vincent Lejeune
parent eda3c9084e
commit dbcddcf5e2
1 changed files with 220 additions and 80 deletions

View File

@ -144,6 +144,190 @@ struct MipmapLevelInfo
size_t rowPitch;
};
#define MAX2(a, b) ((a) > (b)) ? (a) : (b)
/**
* Write data, assume src pixels are packed but not mipmaplevel
*/
static std::vector<MipmapLevelInfo>
writeTexelsGeneric(const char *src, char *dst, size_t widthInBlock, size_t heightInBlock, size_t blockSize, size_t mipmapCount)
{
std::vector<MipmapLevelInfo> Result;
size_t offsetInDst = 0, offsetInSrc = 0;
size_t currentHeight = heightInBlock, currentWidth = widthInBlock;
for (unsigned mipLevel = 0; mipLevel < mipmapCount; mipLevel++)
{
size_t rowPitch = powerOf2Align(currentWidth * blockSize, 256);
MipmapLevelInfo currentMipmapLevelInfo = {};
currentMipmapLevelInfo.offset = offsetInDst;
currentMipmapLevelInfo.height = currentHeight;
currentMipmapLevelInfo.width = currentWidth;
currentMipmapLevelInfo.rowPitch = rowPitch;
Result.push_back(currentMipmapLevelInfo);
for (unsigned row = 0; row < currentHeight; row++)
memcpy((char*)dst + offsetInDst + row * rowPitch, (char*)src + offsetInSrc + row * widthInBlock * blockSize, currentWidth * blockSize);
offsetInDst += currentHeight * rowPitch;
offsetInSrc += currentHeight * widthInBlock * blockSize;
currentHeight = MAX2(currentHeight / 2, 1);
currentWidth = MAX2(currentWidth / 2, 1);
}
return Result;
}
/**
* Write data, assume src pixels are swizzled and but not mipmaplevel
*/
static std::vector<MipmapLevelInfo>
writeTexelsSwizzled(const char *src, char *dst, size_t widthInBlock, size_t heightInBlock, size_t blockSize, size_t mipmapCount)
{
std::vector<MipmapLevelInfo> Result;
size_t offsetInDst = 0, offsetInSrc = 0;
size_t currentHeight = heightInBlock, currentWidth = widthInBlock;
for (unsigned mipLevel = 0; mipLevel < mipmapCount; mipLevel++)
{
size_t rowPitch = powerOf2Align(currentWidth * blockSize, 256);
MipmapLevelInfo currentMipmapLevelInfo = {};
currentMipmapLevelInfo.offset = offsetInDst;
currentMipmapLevelInfo.height = currentHeight;
currentMipmapLevelInfo.width = currentWidth;
currentMipmapLevelInfo.rowPitch = rowPitch;
Result.push_back(currentMipmapLevelInfo);
u32 *castedSrc, *castedDst;
u32 log2width, log2height;
castedSrc = (u32*)src + offsetInSrc;
castedDst = (u32*)dst + offsetInDst;
log2width = (u32)(logf((float)currentWidth) / logf(2.f));
log2height = (u32)(logf((float)currentHeight) / logf(2.f));
#pragma omp parallel for
for (unsigned row = 0; row < currentHeight; row++)
for (int j = 0; j < currentWidth; j++)
castedDst[(row * rowPitch / 4) + j] = castedSrc[LinearToSwizzleAddress(j, row, 0, log2width, log2height, 0)];
offsetInDst += currentHeight * rowPitch;
offsetInSrc += currentHeight * widthInBlock * blockSize;
currentHeight = MAX2(currentHeight / 2, 1);
currentWidth = MAX2(currentWidth / 2, 1);
}
return Result;
}
/**
* Write data, assume compressed (DXTCn) format
*/
static std::vector<MipmapLevelInfo>
writeCompressedTexel(const char *src, char *dst, size_t widthInBlock, size_t blockWidth, size_t heightInBlock, size_t blockHeight, size_t blockSize, size_t mipmapCount)
{
std::vector<MipmapLevelInfo> Result;
size_t offsetInDst = 0, offsetInSrc = 0;
size_t currentHeight = heightInBlock, currentWidth = widthInBlock;
for (unsigned mipLevel = 0; mipLevel < mipmapCount; mipLevel++)
{
size_t rowPitch = powerOf2Align(currentWidth * blockSize, 256);
MipmapLevelInfo currentMipmapLevelInfo = {};
currentMipmapLevelInfo.offset = offsetInDst;
currentMipmapLevelInfo.height = currentHeight * blockHeight;
currentMipmapLevelInfo.width = currentWidth * blockWidth;
currentMipmapLevelInfo.rowPitch = rowPitch;
Result.push_back(currentMipmapLevelInfo);
for (unsigned row = 0; row < currentHeight; row++)
memcpy((char*)dst + offsetInDst + row * rowPitch, (char*)src + offsetInSrc + row * currentWidth * blockSize, currentWidth * blockSize);
offsetInDst += currentHeight * rowPitch;
offsetInDst = powerOf2Align(offsetInDst, 512);
offsetInSrc += currentHeight * currentWidth * blockSize;
currentHeight = MAX2(currentHeight / 2, 1);
currentWidth = MAX2(currentWidth / 2, 1);
}
return Result;
}
/**
* Write 16 bytes pixel textures, assume src pixels are packed but not mipmaplevel
*/
static std::vector<MipmapLevelInfo>
write16bTexelsGeneric(const char *src, char *dst, size_t widthInBlock, size_t heightInBlock, size_t blockSize, size_t mipmapCount)
{
std::vector<MipmapLevelInfo> Result;
size_t offsetInDst = 0, offsetInSrc = 0;
size_t currentHeight = heightInBlock, currentWidth = widthInBlock;
size_t srcPitch = widthInBlock * blockSize;
for (unsigned mipLevel = 0; mipLevel < mipmapCount; mipLevel++)
{
size_t rowPitch = powerOf2Align(currentWidth * blockSize, 256);
MipmapLevelInfo currentMipmapLevelInfo = {};
currentMipmapLevelInfo.offset = offsetInDst;
currentMipmapLevelInfo.height = currentHeight;
currentMipmapLevelInfo.width = currentWidth;
currentMipmapLevelInfo.rowPitch = rowPitch;
Result.push_back(currentMipmapLevelInfo);
unsigned short *castedDst = (unsigned short *)dst, *castedSrc = (unsigned short *)src;
for (unsigned row = 0; row < heightInBlock; row++)
for (int j = 0; j < currentWidth; j++)
{
u16 tmp = castedSrc[offsetInSrc / 2 + row * srcPitch / 2 + j];
castedDst[offsetInDst / 2 + row * rowPitch / 2 + j] = (tmp >> 8) | (tmp << 8);
}
offsetInDst += currentHeight * rowPitch;
offsetInSrc += currentHeight * widthInBlock * blockSize;
currentHeight = MAX2(currentHeight / 2, 1);
currentWidth = MAX2(currentWidth / 2, 1);
}
return Result;
}
/**
* Write 16 bytes pixel textures, assume src pixels are packed but not mipmaplevel
*/
static std::vector<MipmapLevelInfo>
write16bX4TexelsGeneric(const char *src, char *dst, size_t widthInBlock, size_t heightInBlock, size_t blockSize, size_t mipmapCount)
{
std::vector<MipmapLevelInfo> Result;
size_t offsetInDst = 0, offsetInSrc = 0;
size_t currentHeight = heightInBlock, currentWidth = widthInBlock;
size_t srcPitch = widthInBlock * blockSize;
for (unsigned mipLevel = 0; mipLevel < mipmapCount; mipLevel++)
{
size_t rowPitch = powerOf2Align(currentWidth * blockSize, 256);
MipmapLevelInfo currentMipmapLevelInfo = {};
currentMipmapLevelInfo.offset = offsetInDst;
currentMipmapLevelInfo.height = currentHeight;
currentMipmapLevelInfo.width = currentWidth;
currentMipmapLevelInfo.rowPitch = rowPitch;
Result.push_back(currentMipmapLevelInfo);
unsigned short *castedDst = (unsigned short *)dst, *castedSrc = (unsigned short *)src;
for (unsigned row = 0; row < heightInBlock; row++)
for (int j = 0; j < currentWidth * 4; j++)
{
u16 tmp = castedSrc[offsetInSrc / 2 + row * srcPitch / 2 + j];
castedDst[offsetInDst / 2 + row * rowPitch / 2 + j] = (tmp >> 8) | (tmp << 8);
}
offsetInDst += currentHeight * rowPitch;
offsetInSrc += currentHeight * widthInBlock * blockSize;
currentHeight = MAX2(currentHeight / 2, 1);
currentWidth = MAX2(currentWidth / 2, 1);
}
return Result;
}
/**
* Create a texture residing in default heap and generate uploads commands in commandList,
* using a temporary texture buffer.
@ -310,7 +494,7 @@ ID3D12Resource *uploadSingleTexture(
size_t rowPitch = powerOf2Align(blockSizeInByte * widthInBlocks, 256);
ID3D12Resource *Texture;
size_t textureSize = rowPitch * heightInBlocks * 4; // * 3 for mipmap levels
size_t textureSize = rowPitch * heightInBlocks * 4; // * 4 for mipmap levels
assert(textureBuffersHeap.canAlloc(textureSize));
size_t heapOffset = textureBuffersHeap.alloc(textureSize);
@ -327,85 +511,41 @@ ID3D12Resource *uploadSingleTexture(
auto pixels = vm::get_ptr<const u8>(texaddr);
void *textureData;
check(Texture->Map(0, nullptr, (void**)&textureData));
std::vector<MipmapLevelInfo> mipInfos;
// Upload with correct rowpitch
std::vector<MipmapLevelInfo> mipinfos;
size_t offsetInDst = 0, offsetInSrc = 0;
size_t currentHeight = heightInBlocks, currentWidth = widthInBlocks;
unsigned tmp = texture.GetMipmap();
for (unsigned mipLevel = 0; mipLevel < texture.GetMipmap(); mipLevel++)
switch (format)
{
MipmapLevelInfo currentMipmapLevelInfo = {};
currentMipmapLevelInfo.offset = offsetInDst;
currentMipmapLevelInfo.height = currentHeight;
currentMipmapLevelInfo.width = currentWidth;
for (unsigned row = 0; row < currentHeight; row++)
{
switch (format)
{
case CELL_GCM_TEXTURE_A8R8G8B8:
{
currentMipmapLevelInfo.rowPitch = powerOf2Align(currentWidth * blockSizeInByte, 256);
if (is_swizzled)
{
u32 *src, *dst;
u32 log2width, log2height;
src = (u32*)pixels + offsetInSrc;
dst = (u32*)textureData + offsetInDst;
log2width = (u32)(logf((float)currentWidth) / logf(2.f));
log2height = (u32)(logf((float)currentHeight) / logf(2.f));
#pragma omp parallel for
for (int j = 0; j < w; j++)
dst[(row * currentMipmapLevelInfo.rowPitch / 4) + j] = src[LinearToSwizzleAddress(j, row, 0, log2width, log2height, 0)];
}
else
memcpy((char*)textureData + offsetInDst + row * currentMipmapLevelInfo.rowPitch, (char*)pixels + offsetInSrc + row * widthInBlocks * blockSizeInByte, currentWidth * blockSizeInByte);
break;
}
case CELL_GCM_TEXTURE_A4R4G4B4:
case CELL_GCM_TEXTURE_R5G6B5:
{
currentMipmapLevelInfo.rowPitch = rowPitch;
unsigned short *dst = (unsigned short *)textureData, *src = (unsigned short *)pixels;
for (int j = 0; j < w; j++)
{
u16 tmp = src[offsetInSrc / 2 + row * srcPitch / 2 + j];
dst[offsetInDst / 2 + row * rowPitch / 2 + j] = (tmp >> 8) | (tmp << 8);
}
break;
}
case CELL_GCM_TEXTURE_W16_Z16_Y16_X16_FLOAT:
{
currentMipmapLevelInfo.rowPitch = rowPitch;
unsigned short *dst = (unsigned short *)textureData, *src = (unsigned short *)pixels;
for (int j = 0; j < w * 4; j++)
{
unsigned short tmp = src[offsetInSrc / 2 + row * w * 4 + j];
dst[offsetInDst / 2 + row * w * 4 + j] = (tmp >> 8) | (tmp << 8);
}
break;
}
default:
{
currentMipmapLevelInfo.rowPitch = rowPitch;
streamBuffer((char*)textureData + offsetInDst + row * rowPitch, (char*)pixels + offsetInSrc + row * srcPitch, srcPitch);
break;
}
}
}
offsetInDst += currentHeight * currentMipmapLevelInfo.rowPitch;
offsetInDst = powerOf2Align(offsetInDst, 256);
offsetInSrc += currentHeight * widthInBlocks * blockSizeInByte;
mipinfos.push_back(currentMipmapLevelInfo);
currentHeight /= 2;
currentWidth /= 2;
case CELL_GCM_TEXTURE_A8R8G8B8:
{
if (is_swizzled)
mipInfos = writeTexelsSwizzled((char*)pixels, (char*)textureData, w, h, 4, texture.GetMipmap());
else
mipInfos = writeTexelsGeneric((char*)pixels, (char*)textureData, w, h, 4, texture.GetMipmap());
break;
}
case CELL_GCM_TEXTURE_A4R4G4B4:
case CELL_GCM_TEXTURE_R5G6B5:
{
mipInfos = write16bTexelsGeneric((char*)pixels, (char*)textureData, w, h, 2, texture.GetMipmap());
break;
}
case CELL_GCM_TEXTURE_W16_Z16_Y16_X16_FLOAT:
{
mipInfos = write16bX4TexelsGeneric((char*)pixels, (char*)textureData, w, h, 8, texture.GetMipmap());
break;
}
case CELL_GCM_TEXTURE_COMPRESSED_DXT1:
case CELL_GCM_TEXTURE_COMPRESSED_DXT23:
case CELL_GCM_TEXTURE_COMPRESSED_DXT45:
{
mipInfos = writeCompressedTexel((char*)pixels, (char*)textureData, widthInBlocks, blockWidthInPixel, heightInBlocks, blockHeightInPixel, blockSizeInByte, texture.GetMipmap());
break;
}
default:
{
mipInfos = writeTexelsGeneric((char*)pixels, (char*)textureData, w, h, blockSizeInByte, texture.GetMipmap());
break;
}
}
Texture->Unmap(0, nullptr);
@ -427,11 +567,11 @@ ID3D12Resource *uploadSingleTexture(
size_t miplevel = 0;
for (const MipmapLevelInfo mli : mipinfos)
for (const MipmapLevelInfo mli : mipInfos)
{
D3D12_TEXTURE_COPY_LOCATION dst = {}, src = {};
dst.pResource = vramTexture;
dst.SubresourceIndex = miplevel;
dst.SubresourceIndex = (UINT)miplevel;
dst.Type = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX;
src.PlacedFootprint.Offset = mli.offset;
src.pResource = Texture;