mirror of
https://github.com/libretro/pcsx2.git
synced 2024-11-29 04:10:48 +00:00
GregMiscellaneous: zzogl-pg:
* Use sse2 instruction to convert pixel from 32bits to 16bits. Note: Seem slower than basic C. I need to opmitize some memory access... git-svn-id: http://pcsx2.googlecode.com/svn/branches/GregMiscellaneous@3811 96395faa-99c1-11dd-bbfe-3dabce05a288
This commit is contained in:
gregory.hainaut@gmail.com
parent
cfb4ba8b2b
commit
189d92b34d
@ -3013,7 +3013,97 @@ __forceinline void update_4pixels(u32* src, Tdst* pPageOffset, u32 basepage, u32
|
||||
*dst_tmp = (dsrc_tmp & mask) | (*dst_tmp & imask);
|
||||
}
|
||||
|
||||
template <u32 size, u32 pageTable[size][64], typename Tdst, Tdst (*convfn)(u32)>
|
||||
static const __aligned16 u32 pixel_mask[4] = {0x80F8F8F8, 0x80F8F8F8, 0x80F8F8F8, 0x80F8F8F8};
|
||||
template <u32 size, u32 pageTable[size][64], typename Tdst, Tdst (*convfn)(u32), u32 INDEX>
|
||||
__forceinline void update_4pixels_sse2(u32* src, Tdst* pPageOffset, u32 basepage, u32 i_msk, u32 j, u32 mask, u32 imask)
|
||||
{
|
||||
Tdst* dst_tmp;
|
||||
__aligned16 u32 dsrc_tmp[4];
|
||||
|
||||
// load 4 pixel into a 128 array
|
||||
dsrc_tmp[0] = src[RW((j<<6)+INDEX)];
|
||||
dsrc_tmp[1] = src[RW((j<<6)+INDEX+1)];
|
||||
dsrc_tmp[2] = src[RW((j<<6)+INDEX+2)];
|
||||
dsrc_tmp[3] = src[RW((j<<6)+INDEX+3)];
|
||||
|
||||
// transform pixel from ARGB:8888 to ARGB:1555
|
||||
__asm__
|
||||
(
|
||||
".intel_syntax noprefix\n"
|
||||
|
||||
"movdqa xmm0, [%[dsrc_tmp]]\n" // load 4 pixel
|
||||
|
||||
// remove useless bits
|
||||
"movdqa xmm6, [%[pixel_mask]]\n"
|
||||
"pand xmm0, xmm6\n"
|
||||
|
||||
// shuffle to obtain ARARARAR GBGBGBGB
|
||||
"pshuflw xmm0, xmm0, 0b11011000\n" // position 3 1 2 0
|
||||
"pshufhw xmm0, xmm0, 0b11011000\n" // position 3 1 2 0
|
||||
"pshufd xmm0, xmm0, 0b11011000\n" // position 3 1 2 0
|
||||
|
||||
// Convert to 16 bits value with unpack
|
||||
"pxor xmm7, xmm7\n"
|
||||
"movdqa xmm2, xmm0\n"
|
||||
"punpcklbw xmm0, xmm7\n" // GBGB GBGB
|
||||
"punpckhbw xmm2, xmm7\n" // ARAR ARAR
|
||||
|
||||
// shuffle to obtain GGGG BBBB
|
||||
"pshuflw xmm0, xmm0, 0b11011000\n" // position 3 1 2 0
|
||||
"pshufhw xmm0, xmm0, 0b11011000\n" // position 3 1 2 0
|
||||
"pshufd xmm0, xmm0, 0b11011000\n" // position 3 1 2 0
|
||||
|
||||
// shuffle to obtain AAAA RRRR
|
||||
"pshuflw xmm2, xmm2, 0b11011000\n" // position 3 1 2 0
|
||||
"pshufhw xmm2, xmm2, 0b11011000\n" // position 3 1 2 0
|
||||
"pshufd xmm2, xmm2, 0b11011000\n" // position 3 1 2 0
|
||||
|
||||
// FIXME try to work with 16 bits data
|
||||
// convert to 32 bits value with unpack
|
||||
"movdqa xmm1, xmm0\n"
|
||||
"punpcklbw xmm0, xmm7\n" // BB BB
|
||||
"punpckhbw xmm1, xmm7\n" // GG GG
|
||||
|
||||
"movdqa xmm3, xmm2\n"
|
||||
"punpcklbw xmm2, xmm7\n" // RR RR
|
||||
"punpckhbw xmm3, xmm7\n" // AA AA
|
||||
|
||||
// Now data are loaded we can to the pixel conversion
|
||||
// shift each component
|
||||
"pslld xmm3, 8\n" // alpha 7 -> 15
|
||||
"pslld xmm2, 7\n" // red 7:3 -> 14:10
|
||||
"pslld xmm1, 2\n" // blue 7:3 -> 9:5
|
||||
"psrld xmm0, 3\n" // green 7:3 -> 4:0
|
||||
|
||||
// Build the 16 bits pixel values
|
||||
"por xmm0, xmm1\n"
|
||||
"por xmm0, xmm2\n"
|
||||
"por xmm0, xmm3\n"
|
||||
|
||||
// save the result
|
||||
"movdqa [%[dsrc_tmp]], xmm0\n" // load 4 pixel
|
||||
|
||||
".att_syntax\n"
|
||||
:
|
||||
: [dsrc_tmp]"r"(dsrc_tmp), [pixel_mask]"m"(*pixel_mask) // note: I think 'm' only work for STATIC memory...
|
||||
: "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7", "memory"
|
||||
);
|
||||
|
||||
// Group 4 pixel to allow futur sse optimization of the convfn function
|
||||
dst_tmp = pPageOffset + basepage + pageTable[i_msk][(INDEX)];
|
||||
*dst_tmp = ((u16)dsrc_tmp[0] & mask) | (*dst_tmp & imask);
|
||||
|
||||
dst_tmp = pPageOffset + basepage + pageTable[i_msk][INDEX+1];
|
||||
*dst_tmp = ((u16)dsrc_tmp[1] & mask) | (*dst_tmp & imask);
|
||||
|
||||
dst_tmp = pPageOffset + basepage + pageTable[i_msk][INDEX+2];
|
||||
*dst_tmp = ((u16)dsrc_tmp[2] & mask) | (*dst_tmp & imask);
|
||||
|
||||
dst_tmp = pPageOffset + basepage + pageTable[i_msk][INDEX+3];
|
||||
*dst_tmp = ((u16)dsrc_tmp[3] & mask) | (*dst_tmp & imask);
|
||||
}
|
||||
|
||||
template <u32 size, u32 pageTable[size][64], typename Tdst, Tdst (*convfn)(u32), bool do_conversion>
|
||||
void Resolve_32b(const void* psrc, int fbp, int fbw, int fbh, u32 fbm)
|
||||
{
|
||||
#ifdef __LINUX__
|
||||
@ -3063,6 +3153,25 @@ void Resolve_32b(const void* psrc, int fbp, int fbw, int fbh, u32 fbm)
|
||||
// for(int j = fbw_div-1; j >= 0; --j) {
|
||||
for(u32 j = 0 ; j < fbw_div; ++j) {
|
||||
u32 basepage = (i_div + j) * 2048;
|
||||
#ifdef TOTO //ZEROGS_SSE2
|
||||
if (do_conversion) {
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 0>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 4>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 8>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 12>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 16>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 20>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 24>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 28>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 32>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 36>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 40>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 44>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 48>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 52>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 56>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels_sse2<size, pageTable, Tdst, convfn, 60>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
} else {
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 0>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 4>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 8>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
@ -3079,13 +3188,30 @@ void Resolve_32b(const void* psrc, int fbp, int fbw, int fbh, u32 fbm)
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 52>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 56>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 60>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
|
||||
}
|
||||
#else
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 0>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 4>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 8>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 12>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 16>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 20>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 24>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 28>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 32>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 36>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 40>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 44>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 48>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 52>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 56>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
update_4pixels<size, pageTable, Tdst, convfn, 60>(src, pPageOffset, basepage, i_msk, j, mask, imask);
|
||||
#endif
|
||||
}
|
||||
src -= raw_size;
|
||||
// src += raw_size;
|
||||
}
|
||||
#ifdef __LINUX__
|
||||
ZZLog::Error_Log("*** 32 bits: execution time %d", timeGetPreciseTime()-startime);
|
||||
ZZLog::Error_Log("*** 32 bits: execution time %d (convert %d)", timeGetPreciseTime()-startime, do_conversion);
|
||||
#endif
|
||||
}
|
||||
|
||||
@ -3112,7 +3238,7 @@ void _Resolve(const void* psrc, int fbp, int fbw, int fbh, int psm, u32 fbm, boo
|
||||
case PSMCT32:
|
||||
case PSMCT24:
|
||||
#ifdef OPTI_RESOLVE_32
|
||||
Resolve_32b<32, g_pageTable32, u32, dummy_return >(psrc, fbp, fbw, fbh, fbm);
|
||||
Resolve_32b<32, g_pageTable32, u32, dummy_return, false >(psrc, fbp, fbw, fbh, fbm);
|
||||
#else
|
||||
Resolve_32_Bit<u32, dummy_return >(psrc, fbp, fbw, fbh, PSMCT32, fbm);
|
||||
#endif
|
||||
@ -3120,7 +3246,7 @@ void _Resolve(const void* psrc, int fbp, int fbw, int fbh, int psm, u32 fbm, boo
|
||||
|
||||
case PSMCT16:
|
||||
#ifdef OPTI_RESOLVE_32
|
||||
Resolve_32b<64, g_pageTable16, u16, RGBA32to16 >(psrc, fbp, fbw, fbh, fbm);
|
||||
Resolve_32b<64, g_pageTable16, u16, RGBA32to16, true >(psrc, fbp, fbw, fbh, fbm);
|
||||
#else
|
||||
Resolve_32_Bit<u16, RGBA32to16 >(psrc, fbp, fbw, fbh, PSMCT16, fbm);
|
||||
#endif
|
||||
@ -3128,7 +3254,7 @@ void _Resolve(const void* psrc, int fbp, int fbw, int fbh, int psm, u32 fbm, boo
|
||||
|
||||
case PSMCT16S:
|
||||
#ifdef OPTI_RESOLVE_32
|
||||
Resolve_32b<64, g_pageTable16S, u16, RGBA32to16 >(psrc, fbp, fbw, fbh, fbm);
|
||||
Resolve_32b<64, g_pageTable16S, u16, RGBA32to16, true >(psrc, fbp, fbw, fbh, fbm);
|
||||
#else
|
||||
Resolve_32_Bit<u16, RGBA32to16 >(psrc, fbp, fbw, fbh, PSMCT16S, fbm);
|
||||
#endif
|
||||
@ -3137,7 +3263,7 @@ void _Resolve(const void* psrc, int fbp, int fbw, int fbh, int psm, u32 fbm, boo
|
||||
case PSMT32Z:
|
||||
case PSMT24Z:
|
||||
#ifdef OPTI_RESOLVE_32
|
||||
Resolve_32b<32, g_pageTable32Z, u32, dummy_return >(psrc, fbp, fbw, fbh, fbm);
|
||||
Resolve_32b<32, g_pageTable32Z, u32, dummy_return, false >(psrc, fbp, fbw, fbh, fbm);
|
||||
#else
|
||||
Resolve_32_Bit<u32, dummy_return >(psrc, fbp, fbw, fbh, PSMT32Z, fbm);
|
||||
#endif
|
||||
@ -3145,7 +3271,7 @@ void _Resolve(const void* psrc, int fbp, int fbw, int fbh, int psm, u32 fbm, boo
|
||||
|
||||
case PSMT16Z:
|
||||
#ifdef OPTI_RESOLVE_32
|
||||
Resolve_32b<64, g_pageTable16Z, u16, dummy_return >(psrc, fbp, fbw, fbh, fbm);
|
||||
Resolve_32b<64, g_pageTable16Z, u16, dummy_return, false >(psrc, fbp, fbw, fbh, fbm);
|
||||
#else
|
||||
Resolve_32_Bit<u16, dummy_return >(psrc, fbp, fbw, fbh, PSMT16Z, fbm);
|
||||
#endif
|
||||
@ -3153,7 +3279,7 @@ void _Resolve(const void* psrc, int fbp, int fbw, int fbh, int psm, u32 fbm, boo
|
||||
|
||||
case PSMT16SZ:
|
||||
#ifdef OPTI_RESOLVE_32
|
||||
Resolve_32b<64, g_pageTable16SZ, u16, dummy_return >(psrc, fbp, fbw, fbh, fbm);
|
||||
Resolve_32b<64, g_pageTable16SZ, u16, dummy_return, false >(psrc, fbp, fbw, fbh, fbm);
|
||||
#else
|
||||
Resolve_32_Bit<u16, dummy_return >(psrc, fbp, fbw, fbh, PSMT16SZ, fbm);
|
||||
#endif
|
||||
|
||||
Loading…
Reference in New Issue
Block a user