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

Render3D: 

- Fix possible crash due to aligned access.
This commit is contained in:
rogerman 2015-07-15 04:34:23 +00:00
parent 3c9931b1c8
commit 9b370cd602
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
desmume/src/render3D.cpp
View File

@ -576,12 +576,12 @@ Render3DError Render3D_SSE2::ClearFramebuffer(const GFX3D_State &renderState)
{ {
// Copy the colors to the color buffer. Since we can only copy 8 elements at once, // Copy the colors to the color buffer. Since we can only copy 8 elements at once,
// we need to load-store twice. // we need to load-store twice.
_mm_store_si128( (__m128i *)(this->clearImageColor16Buffer + i + 8), _mm_load_si128((__m128i *)(clearColorBuffer + i + 8)) ); _mm_store_si128( (__m128i *)(this->clearImageColor16Buffer + i + 8), _mm_loadu_si128((__m128i *)(clearColorBuffer + i + 8)) );
_mm_store_si128( (__m128i *)(this->clearImageColor16Buffer + i), _mm_load_si128((__m128i *)(clearColorBuffer + i)) ); _mm_store_si128( (__m128i *)(this->clearImageColor16Buffer + i), _mm_loadu_si128((__m128i *)(clearColorBuffer + i)) );
// Write the depth values to the depth buffer. // Write the depth values to the depth buffer.
__m128i clearDepthHi_vec128 = _mm_load_si128((__m128i *)(clearDepthBuffer + i + 8)); __m128i clearDepthHi_vec128 = _mm_loadu_si128((__m128i *)(clearDepthBuffer + i + 8));
__m128i clearDepthLo_vec128 = _mm_load_si128((__m128i *)(clearDepthBuffer + i)); __m128i clearDepthLo_vec128 = _mm_loadu_si128((__m128i *)(clearDepthBuffer + i));
clearDepthHi_vec128 = _mm_and_si128(clearDepthHi_vec128, depthBitMask_vec128); clearDepthHi_vec128 = _mm_and_si128(clearDepthHi_vec128, depthBitMask_vec128);
clearDepthLo_vec128 = _mm_and_si128(clearDepthLo_vec128, depthBitMask_vec128); clearDepthLo_vec128 = _mm_and_si128(clearDepthLo_vec128, depthBitMask_vec128);
@ -603,8 +603,8 @@ Render3DError Render3D_SSE2::ClearFramebuffer(const GFX3D_State &renderState)
this->clearImageDepthBuffer[i+ 0] = dsDepthToD24_LUT[_mm_extract_epi16(clearDepthLo_vec128, 0)]; this->clearImageDepthBuffer[i+ 0] = dsDepthToD24_LUT[_mm_extract_epi16(clearDepthLo_vec128, 0)];
// Write the fog flags to the fog flag buffer. // Write the fog flags to the fog flag buffer.
clearDepthHi_vec128 = _mm_load_si128((__m128i *)(clearDepthBuffer + i + 8)); clearDepthHi_vec128 = _mm_loadu_si128((__m128i *)(clearDepthBuffer + i + 8));
clearDepthLo_vec128 = _mm_load_si128((__m128i *)(clearDepthBuffer + i)); clearDepthLo_vec128 = _mm_loadu_si128((__m128i *)(clearDepthBuffer + i));
clearDepthHi_vec128 = _mm_and_si128(clearDepthHi_vec128, fogBufferBitMask_vec128); clearDepthHi_vec128 = _mm_and_si128(clearDepthHi_vec128, fogBufferBitMask_vec128);
clearDepthLo_vec128 = _mm_and_si128(clearDepthLo_vec128, fogBufferBitMask_vec128); clearDepthLo_vec128 = _mm_and_si128(clearDepthLo_vec128, fogBufferBitMask_vec128);
clearDepthHi_vec128 = _mm_srli_si128(clearDepthHi_vec128, 15); clearDepthHi_vec128 = _mm_srli_si128(clearDepthHi_vec128, 15);