mirror of https://github.com/PCSX2/pcsx2.git
GregMiscellaneous: zzogl-pg: Use _aligned_malloc in GetMemoryTarget.
git-svn-id: http://pcsx2.googlecode.com/svn/branches/GregMiscellaneous@3800 96395faa-99c1-11dd-bbfe-3dabce05a288
This commit is contained in:
parent
7afdf9e7c7
commit
4d5f4ad81b
|
@ -84,7 +84,6 @@
|
|||
// ZZogl-PG: GetRectMemAddress(0x3f8000, 0x408000, 0x0, 0x0, 0x0, 0x100, 0x40, 0x3f80, 0x100);
|
||||
// ZZogl-PG: GetRectMemAddress(0x3fc000, 0x40c000, 0x0, 0x0, 0x0, 0x100, 0x40, 0x3fc0, 0x100);
|
||||
|
||||
|
||||
void GetRectMemAddress(int& start, int& end, int psm, int x, int y, int w, int h, int bp, int bw)
|
||||
{
|
||||
FUNCLOG
|
||||
|
|
|
@ -2202,13 +2202,12 @@ ZeroGS::CMemoryTarget* ZeroGS::CMemoryTargetMngr::GetMemoryTarget(const tex0Info
|
|||
|
||||
memcpy_amd(targ->ptex->memptr, g_pbyGSMemory + 4 * GPU_TEXWIDTH * targ->realy, 4 * GPU_TEXWIDTH * targ->height);
|
||||
|
||||
vector<u8> texdata;
|
||||
u8* ptexdata = NULL;
|
||||
|
||||
if (PSMT_ISCLUT(tex0.psm))
|
||||
{
|
||||
texdata.resize(((tex0.cpsm <= 1) ? 4 : 2) * texW * texH);
|
||||
ptexdata = &texdata[0];
|
||||
u32 tex_size = ((tex0.cpsm <= 1) ? 4 : 2) * texW * texH;
|
||||
ptexdata = (u8*)_aligned_malloc(tex_size, 16);
|
||||
|
||||
u8* psrc = (u8*)(g_pbyGSMemory + 4 * GPU_TEXWIDTH * targ->realy);
|
||||
|
||||
|
@ -2231,43 +2230,16 @@ ZeroGS::CMemoryTarget* ZeroGS::CMemoryTargetMngr::GetMemoryTarget(const tex0Info
|
|||
{
|
||||
if (tex0.psm == PSMT16Z || tex0.psm == PSMT16SZ)
|
||||
{
|
||||
texdata.resize(4 * texW * texH
|
||||
#if defined(ZEROGS_SSE2)
|
||||
+ 15 // reserve additional elements for alignment if SSE2 used.
|
||||
// better do it now, so less resizing would be needed
|
||||
#endif
|
||||
);
|
||||
|
||||
ptexdata = &texdata[0];
|
||||
ptexdata = (u8*)_aligned_malloc(4 * texW * texH, 16);
|
||||
|
||||
// needs to be 8 bit, use xmm for unpacking
|
||||
u16* dst = (u16*)ptexdata;
|
||||
u16* src = (u16*)(g_pbyGSMemory + 4 * GPU_TEXWIDTH * targ->realy);
|
||||
|
||||
#if defined(ZEROGS_SSE2)
|
||||
|
||||
if (((u32)(uptr)dst) % 16 != 0)
|
||||
{
|
||||
// This is not unusual situation, when vector<u8> does not 16bit alignment, that is destructive for SSE2
|
||||
// instruction movdqa [%eax], xmm0
|
||||
// The idea would be resize vector to 15 elements, that set ptxedata to aligned position.
|
||||
// Later we would move eax by 16, so only we should verify is first element align
|
||||
// FIXME. As I see, texdata used only once here, it does not have any impact on other code.
|
||||
// Probably, usage of _aligned_maloc() would be preferable.
|
||||
|
||||
// Note: this often happens when changing AA.
|
||||
int disalignment = 16 - ((u32)(uptr)dst) % 16; // This is value of shift. It could be 0 < disalignment <= 15
|
||||
ptexdata = &texdata[disalignment]; // Set pointer to aligned element
|
||||
dst = (u16*)ptexdata;
|
||||
ZZLog::GS_Log("Made alignment for texdata, 0x%x", dst);
|
||||
assert(((u32)(uptr)dst) % 16 == 0); // Assert, because at future could be vectors with uncontigious spaces
|
||||
}
|
||||
|
||||
int iters = targ->height * GPU_TEXWIDTH / 16;
|
||||
|
||||
SSE2_UnswizzleZ16Target(dst, src, iters) ;
|
||||
assert(((u32)(uptr)dst) % 16 == 0);
|
||||
SSE2_UnswizzleZ16Target(dst, src, targ->height * GPU_TEXWIDTH / 16);
|
||||
#else // ZEROGS_SSE2
|
||||
|
||||
for (int i = 0; i < targ->height; ++i)
|
||||
{
|
||||
for (int j = 0; j < GPU_TEXWIDTH; ++j)
|
||||
|
|
Loading…
Reference in New Issue