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

Colorspace Handler: Make the ColorspaceConvert555To*_AltiVec() functions behave consistently with their scalar counterparts by byte swapping the result. Fixes SoftRasterizer's toon table colors on AltiVec-enabled systems.

This commit is contained in:
rogerman 2021-09-11 11:49:07 -07:00
parent 3dabb10145
commit bf71da494e
1 changed files with 10 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
desmume/src/utils/colorspacehandler/colorspacehandler_AltiVec.cpp Normal file → Executable file
View File

@ -31,26 +31,18 @@ FORCEINLINE void ColorspaceConvert555To8888_AltiVec(const v128u16 &srcColor, con
dstLo = vec_unpackl((vector pixel)srcColor);
dstLo = vec_or( vec_sl((v128u8)dstLo, ((v128u8){0,3,3,3, 0,3,3,3, 0,3,3,3, 0,3,3,3})), vec_sr((v128u8)dstLo, ((v128u8){0,2,2,2, 0,2,2,2, 0,2,2,2, 0,2,2,2})) );
dstLo = vec_perm(dstLo, srcAlphaBits, (SWAP_RB) ? ((v128u8){0x11,0x03,0x02,0x01, 0x13,0x07,0x06,0x05, 0x15,0x0B,0x0A,0x09, 0x17,0x0F,0x0E,0x0D}) : ((v128u8){0x11,0x01,0x02,0x03, 0x13,0x05,0x06,0x07, 0x15,0x09,0x0A,0x0B, 0x17,0x0D,0x0E,0x0F}));
dstLo = vec_perm(dstLo, srcAlphaBits, (SWAP_RB) ? ((v128u8){0x01,0x02,0x03,0x11, 0x05,0x06,0x07,0x13, 0x09,0x0A,0x0B,0x15, 0x0D,0x0E,0x0F,0x17}) : ((v128u8){0x03,0x02,0x01,0x11, 0x07,0x06,0x05,0x13, 0x0B,0x0A,0x09,0x15, 0x0F,0x0E,0x0D,0x17}));
dstHi = vec_unpackh((vector pixel)srcColor);
dstHi = vec_or( vec_sl((v128u8)dstHi, ((v128u8){0,3,3,3, 0,3,3,3, 0,3,3,3, 0,3,3,3})), vec_sr((v128u8)dstHi, ((v128u8){0,2,2,2, 0,2,2,2, 0,2,2,2, 0,2,2,2})) );
dstHi = vec_perm(dstHi, srcAlphaBits, (SWAP_RB) ? ((v128u8){0x19,0x03,0x02,0x01, 0x1B,0x07,0x06,0x05, 0x1D,0x0B,0x0A,0x09, 0x1F,0x0F,0x0E,0x0D}) : ((v128u8){0x19,0x01,0x02,0x03, 0x1B,0x05,0x06,0x07, 0x1D,0x09,0x0A,0x0B, 0x1F,0x0D,0x0E,0x0F}));
dstHi = vec_perm(dstHi, srcAlphaBits, (SWAP_RB) ? ((v128u8){0x01,0x02,0x03,0x19, 0x05,0x06,0x07,0x1B, 0x09,0x0A,0x0B,0x1D, 0x0D,0x0E,0x0F,0x1F}) : ((v128u8){0x03,0x02,0x01,0x19, 0x07,0x06,0x05,0x1B, 0x0B,0x0A,0x09,0x1D, 0x0F,0x0E,0x0D,0x1F}));
}
template <bool SWAP_RB>
FORCEINLINE void ColorspaceConvert555XTo888X_AltiVec(const v128u16 &srcColor, v128u32 &dstLo, v128u32 &dstHi)
{
// Conversion algorithm:
// RGB 5-bit to 8-bit formula: dstRGB8 = (srcRGB5 << 3) | ((srcRGB5 >> 2) & 0x07)
dstLo = vec_unpackl((vector pixel)srcColor);
dstLo = vec_or( vec_sl((v128u8)dstLo, ((v128u8){0,3,3,3, 0,3,3,3, 0,3,3,3, 0,3,3,3})), vec_sr((v128u8)dstLo, ((v128u8){0,2,2,2, 0,2,2,2, 0,2,2,2, 0,2,2,2})) );
dstLo = vec_perm(dstLo, ((v128u8){0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0}), (SWAP_RB) ? ((v128u8){0x11,0x03,0x02,0x01, 0x13,0x07,0x06,0x05, 0x15,0x0B,0x0A,0x09, 0x17,0x0F,0x0E,0x0D}) : ((v128u8){0x11,0x01,0x02,0x03, 0x13,0x05,0x06,0x07, 0x15,0x09,0x0A,0x0B, 0x17,0x0D,0x0E,0x0F}));
dstHi = vec_unpackh((vector pixel)srcColor);
dstHi = vec_or( vec_sl((v128u8)dstHi, ((v128u8){0,3,3,3, 0,3,3,3, 0,3,3,3, 0,3,3,3})), vec_sr((v128u8)dstHi, ((v128u8){0,2,2,2, 0,2,2,2, 0,2,2,2, 0,2,2,2})) );
dstHi = vec_perm(dstHi, ((v128u8){0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0}), (SWAP_RB) ? ((v128u8){0x19,0x03,0x02,0x01, 0x1B,0x07,0x06,0x05, 0x1D,0x0B,0x0A,0x09, 0x1F,0x0F,0x0E,0x0D}) : ((v128u8){0x19,0x01,0x02,0x03, 0x1B,0x05,0x06,0x07, 0x1D,0x09,0x0A,0x0B, 0x1F,0x0D,0x0E,0x0F}));
const v128u16 srcAlphaBits16 = {0, 0, 0, 0, 0, 0, 0, 0};
ColorspaceConvert555To8888_AltiVec<SWAP_RB>(srcColor, srcAlphaBits16, dstLo, dstHi);
}
template <bool SWAP_RB>
@ -61,26 +53,18 @@ FORCEINLINE void ColorspaceConvert555To6665_AltiVec(const v128u16 &srcColor, con
dstLo = vec_unpackl((vector pixel)srcColor);
dstLo = vec_or( vec_sl((v128u8)dstLo, ((v128u8){0,1,1,1, 0,1,1,1, 0,1,1,1, 0,1,1,1})), vec_sr((v128u8)dstLo, ((v128u8){0,4,4,4, 0,4,4,4, 0,4,4,4, 0,4,4,4})) );
dstLo = vec_perm(dstLo, srcAlphaBits, (SWAP_RB) ? ((v128u8){0x11,0x03,0x02,0x01, 0x13,0x07,0x06,0x05, 0x15,0x0B,0x0A,0x09, 0x17,0x0F,0x0E,0x0D}) : ((v128u8){0x11,0x01,0x02,0x03, 0x13,0x05,0x06,0x07, 0x15,0x09,0x0A,0x0B, 0x17,0x0D,0x0E,0x0F}));
dstLo = vec_perm(dstLo, srcAlphaBits, (SWAP_RB) ? ((v128u8){0x01,0x02,0x03,0x11, 0x05,0x06,0x07,0x13, 0x09,0x0A,0x0B,0x15, 0x0D,0x0E,0x0F,0x17}) : ((v128u8){0x03,0x02,0x01,0x11, 0x07,0x06,0x05,0x13, 0x0B,0x0A,0x09,0x15, 0x0F,0x0E,0x0D,0x17}));
dstHi = vec_unpackh((vector pixel)srcColor);
dstHi = vec_or( vec_sl((v128u8)dstHi, ((v128u8){0,1,1,1, 0,1,1,1, 0,1,1,1, 0,1,1,1})), vec_sr((v128u8)dstHi, ((v128u8){0,4,4,4, 0,4,4,4, 0,4,4,4, 0,4,4,4})) );
dstHi = vec_perm(dstHi, srcAlphaBits, (SWAP_RB) ? ((v128u8){0x19,0x03,0x02,0x01, 0x1B,0x07,0x06,0x05, 0x1D,0x0B,0x0A,0x09, 0x1F,0x0F,0x0E,0x0D}) : ((v128u8){0x19,0x01,0x02,0x03, 0x1B,0x05,0x06,0x07, 0x1D,0x09,0x0A,0x0B, 0x1F,0x0D,0x0E,0x0F}));
dstHi = vec_perm(dstHi, srcAlphaBits, (SWAP_RB) ? ((v128u8){0x01,0x02,0x03,0x19, 0x05,0x06,0x07,0x1B, 0x09,0x0A,0x0B,0x1D, 0x0D,0x0E,0x0F,0x1F}) : ((v128u8){0x03,0x02,0x01,0x19, 0x07,0x06,0x05,0x1B, 0x0B,0x0A,0x09,0x1D, 0x0F,0x0E,0x0D,0x1F}));
}
template <bool SWAP_RB>
FORCEINLINE void ColorspaceConvert555XTo666X_AltiVec(const v128u16 &srcColor, v128u32 &dstLo, v128u32 &dstHi)
{
// Conversion algorithm:
// RGB 5-bit to 6-bit formula: dstRGB6 = (srcRGB5 << 1) | ((srcRGB5 >> 4) & 0x01)
dstLo = vec_unpackl((vector pixel)srcColor);
dstLo = vec_or( vec_sl((v128u8)dstLo, ((v128u8){0,1,1,1, 0,1,1,1, 0,1,1,1, 0,1,1,1})), vec_sr((v128u8)dstLo, ((v128u8){0,4,4,4, 0,4,4,4, 0,4,4,4, 0,4,4,4})) );
dstLo = vec_perm(dstLo, ((v128u8){0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0}), (SWAP_RB) ? ((v128u8){0x11,0x03,0x02,0x01, 0x13,0x07,0x06,0x05, 0x15,0x0B,0x0A,0x09, 0x17,0x0F,0x0E,0x0D}) : ((v128u8){0x11,0x01,0x02,0x03, 0x13,0x05,0x06,0x07, 0x15,0x09,0x0A,0x0B, 0x17,0x0D,0x0E,0x0F}));
dstHi = vec_unpackh((vector pixel)srcColor);
dstHi = vec_or( vec_sl((v128u8)dstHi, ((v128u8){0,1,1,1, 0,1,1,1, 0,1,1,1, 0,1,1,1})), vec_sr((v128u8)dstHi, ((v128u8){0,4,4,4, 0,4,4,4, 0,4,4,4, 0,4,4,4})) );
dstHi = vec_perm(dstHi, ((v128u8){0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0}), (SWAP_RB) ? ((v128u8){0x19,0x03,0x02,0x01, 0x1B,0x07,0x06,0x05, 0x1D,0x0B,0x0A,0x09, 0x1F,0x0F,0x0E,0x0D}) : ((v128u8){0x19,0x01,0x02,0x03, 0x1B,0x05,0x06,0x07, 0x1D,0x09,0x0A,0x0B, 0x1F,0x0D,0x0E,0x0F}));
const v128u16 srcAlphaBits16 = {0, 0, 0, 0, 0, 0, 0, 0};
ColorspaceConvert555To6665_AltiVec<SWAP_RB>(srcColor, srcAlphaBits16, dstLo, dstHi);
}
template <bool SWAP_RB>
@ -231,7 +215,7 @@ static size_t ColorspaceConvertBuffer555To8888Opaque_AltiVec(const u16 *__restri
{
size_t i = 0;
for (; i < pixCountVec128; i+=8)
for (; i < pixCountVec128; i+=sizeof(v128u16)/sizeof(u16))
{
v128u32 dstConvertedLo, dstConvertedHi;
@ -248,7 +232,7 @@ size_t ColorspaceConvertBuffer555To6665Opaque_AltiVec(const u16 *__restrict src,
{
size_t i = 0;
for (; i < pixCountVec128; i+=8)
for (; i < pixCountVec128; i+=sizeof(v128u16)/sizeof(u16))
{
v128u32 dstConvertedLo, dstConvertedHi;