mirrors/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-04-21 03:55:32 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

d3d12: Take alignment into account in streamBuffer

Browse source 
And use it for texture upload
This commit is contained in:
vlj 2015-06-02 19:18:16 +02:00 • committed by Vincent Lejeune
parent 878a116c4a
commit 51d287d9b0
2 changed files with 20 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

33
rpcs3/Emu/RSX/D3D12/D3D12.h
View file

@ -31,8 +31,9 @@ void streamToBuffer(void* dst, void* src, size_t sizeInBytes)
{
for (unsigned i = 0; i < sizeInBytes / 16; i++)
{
__m128i *srcPtr = (__m128i*) ((char*)src + i * 16);
_mm_stream_si128((__m128i*)((char*)dst + i * 16), *srcPtr);
const __m128i &srcPtr = _mm_loadu_si128((__m128i*) ((char*)src + i * 16));
_mm_stream_si128((__m128i*)((char*)dst + i * 16), srcPtr);
}
}
@ -44,20 +45,24 @@ inline
void streamBuffer(void* dst, void* src, size_t sizeInBytes)
{
// Assume 64 bytes cache line
assert(powerOf2Align(sizeInBytes, 64));
for (unsigned i = 0; i < sizeInBytes / 64; i++)
unsigned offset = 0;
bool isAligned = !((size_t)src & 15);
for (; (offset + 64) < sizeInBytes; offset += 64)
{
char *line = (char*)src + i * 64;
_mm_prefetch(line, _MM_HINT_NTA);
__m128i *srcPtr = (__m128i*) (line);
_mm_stream_si128((__m128i*)((char*)dst + i * 64), *srcPtr);
srcPtr = (__m128i*) (line + 16);
_mm_stream_si128((__m128i*)((char*)dst + i * 64 + 16), *srcPtr);
srcPtr = (__m128i*) (line + 32);
_mm_stream_si128((__m128i*)((char*)dst + i * 64 + 32), *srcPtr);
srcPtr = (__m128i*) (line + 48);
_mm_stream_si128((__m128i*)((char*)dst + i * 64 + 48), *srcPtr);
char *line = (char*)src + offset;
char *dstline = (char*)dst + offset;
// prefetch next line
_mm_prefetch(line + 16, _MM_HINT_NTA);
__m128i srcPtr = isAligned ? _mm_load_si128((__m128i *)line) : _mm_loadu_si128((__m128i *)line);
_mm_stream_si128((__m128i*)dstline, srcPtr);
srcPtr = isAligned ? _mm_load_si128((__m128i *)(line + 16)) : _mm_loadu_si128((__m128i *)(line + 16));
_mm_stream_si128((__m128i*)(dstline + 16), srcPtr);
srcPtr = isAligned ? _mm_load_si128((__m128i *)(line + 32)) : _mm_loadu_si128((__m128i *)(line + 32));
_mm_stream_si128((__m128i*)(dstline + 32), srcPtr);
srcPtr = isAligned ? _mm_load_si128((__m128i *)(line + 48)) : _mm_loadu_si128((__m128i *)(line + 48));
_mm_stream_si128((__m128i*)(dstline + 48), srcPtr);
}
memcpy((char*)dst + offset, (char*)src + offset, sizeInBytes - offset);
}
inline

2
rpcs3/Emu/RSX/D3D12/D3D12Texture.cpp
View file

@ -246,7 +246,7 @@ size_t D3D12GSRender::UploadTextures()
}
}
else
streamToBuffer((char*)textureData + row * rowPitch, (char*)pixels + row * m_texture_pitch, m_texture_pitch);
streamBuffer((char*)textureData + row * rowPitch, (char*)pixels + row * m_texture_pitch, m_texture_pitch);
}
Texture->Unmap(0, nullptr);