ShuriZma
/
desmume
mirror of https://github.com/TASEmulators/desmume.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

GPU: 

- For SSE2 systems, when reading graphics data from main memory, eliminate the extraneous pshufd instruction.
This commit is contained in:
rogerman 2016-08-03 18:14:44 +00:00
parent 2e2d02302b
commit cd7690730b
1 changed files with 3 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
desmume/src/GPU.cpp
View File

@ -5972,8 +5972,8 @@ void GPUEngineA::_RenderLine_DispCapture_FIFOToBuffer(u16 *fifoLineBuffer)
#ifdef ENABLE_SSE2
for (size_t i = 0; i < GPU_FRAMEBUFFER_NATIVE_WIDTH * sizeof(u16) / sizeof(__m128i); i++)
{
__m128i fifoColor = _mm_set_epi32(DISP_FIFOrecv(), DISP_FIFOrecv(), DISP_FIFOrecv(), DISP_FIFOrecv());
_mm_store_si128((__m128i *)fifoLineBuffer + i, _mm_shuffle_epi32(fifoColor, 0x1B)); // We need to shuffle the four FIFO values back into the correct order, since they were originally loaded in reverse order.
const __m128i fifoColor = _mm_setr_epi32(DISP_FIFOrecv(), DISP_FIFOrecv(), DISP_FIFOrecv(), DISP_FIFOrecv());
_mm_store_si128((__m128i *)fifoLineBuffer + i, fifoColor);
}
#else
for (size_t i = 0; i < GPU_FRAMEBUFFER_NATIVE_WIDTH * sizeof(u16) / sizeof(u32); i++)
@ -6592,8 +6592,7 @@ void GPUEngineA::_HandleDisplayModeMainMemory(const size_t l)
const __m128i alphaBit = _mm_set1_epi16(0x8000);
for (size_t i = 0; i < GPU_FRAMEBUFFER_NATIVE_WIDTH * sizeof(u16) / sizeof(__m128i); i++)
{
__m128i fifoColor = _mm_set_epi32(DISP_FIFOrecv(), DISP_FIFOrecv(), DISP_FIFOrecv(), DISP_FIFOrecv());
fifoColor = _mm_shuffle_epi32(fifoColor, 0x1B); // We need to shuffle the four FIFO values back into the correct order, since they were originally loaded in reverse order.
const __m128i fifoColor = _mm_setr_epi32(DISP_FIFOrecv(), DISP_FIFOrecv(), DISP_FIFOrecv(), DISP_FIFOrecv());
_mm_store_si128((__m128i *)dst + i, _mm_or_si128(fifoColor, alphaBit));
}
#else