preserve-emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2026-09-11 07:03:33 +02:00
Code Issues Packages Projects Releases Wiki Activity

Move some more color conversion functions to ColorConv

Browse Source
This commit is contained in:
Henrik Rydgard
2015-04-08 22:52:47 +02:00
parent 1285ae4a61
commit c2cf2bd97e
7 changed files with 224 additions and 217 deletions
Download Patch File Download Diff File Expand all files Collapse all files
GPU/Common/TextureDecoder.cpp
-121
View File
@@ -328,124 +328,3 @@ void DecodeDXT5Block(u32 *dst, const DXT5Block *src, int pitch) {
dst += pitch;
}
}
void ConvertBGRA8888ToRGBA8888(u32 *dst, const u32 *src, const u32 numPixels) {
#ifdef _M_SSE
const __m128i maskGA = _mm_set1_epi32(0xFF00FF00);
const __m128i *srcp = (const __m128i *)src;
__m128i *dstp = (__m128i *)dst;
u32 sseChunks = numPixels / 4;
if (((intptr_t)src & 0xF) || ((intptr_t)dst & 0xF)) {
sseChunks = 0;
}
for (u32 i = 0; i < sseChunks; ++i) {
__m128i c = _mm_load_si128(&srcp[i]);
__m128i rb = _mm_andnot_si128(maskGA, c);
c = _mm_and_si128(c, maskGA);
__m128i b = _mm_srli_epi32(rb, 16);
__m128i r = _mm_slli_epi32(rb, 16);
c = _mm_or_si128(_mm_or_si128(c, r), b);
_mm_store_si128(&dstp[i], c);
}
// The remainder starts right after those done via SSE.
u32 i = sseChunks * 4;
#else
u32 i = 0;
#endif
for (; i < numPixels; i++) {
const u32 c = src[i];
dst[i] = ((c >> 16) & 0x000000FF) |
((c >> 0) & 0xFF00FF00) |
((c << 16) & 0x00FF0000);
}
}
void ConvertRGBA8888ToRGBA5551(u16 *dst, const u32 *src, const u32 numPixels) {
#if _M_SSE >= 0x401
const __m128i maskAG = _mm_set1_epi32(0x8000F800);
const __m128i maskRB = _mm_set1_epi32(0x00F800F8);
const __m128i mask = _mm_set1_epi32(0x0000FFFF);
const __m128i *srcp = (const __m128i *)src;
__m128i *dstp = (__m128i *)dst;
u32 sseChunks = (numPixels / 4) & ~1;
// SSE 4.1 required for _mm_packus_epi32.
if (((intptr_t)src & 0xF) || ((intptr_t)dst & 0xF) || !cpu_info.bSSE4_1) {
sseChunks = 0;
}
for (u32 i = 0; i < sseChunks; i += 2) {
__m128i c1 = _mm_load_si128(&srcp[i + 0]);
__m128i c2 = _mm_load_si128(&srcp[i + 1]);
__m128i ag, rb;
ag = _mm_and_si128(c1, maskAG);
ag = _mm_or_si128(_mm_srli_epi32(ag, 16), _mm_srli_epi32(ag, 6));
rb = _mm_and_si128(c1, maskRB);
rb = _mm_or_si128(_mm_srli_epi32(rb, 3), _mm_srli_epi32(rb, 9));
c1 = _mm_and_si128(_mm_or_si128(ag, rb), mask);
ag = _mm_and_si128(c2, maskAG);
ag = _mm_or_si128(_mm_srli_epi32(ag, 16), _mm_srli_epi32(ag, 6));
rb = _mm_and_si128(c2, maskRB);
rb = _mm_or_si128(_mm_srli_epi32(rb, 3), _mm_srli_epi32(rb, 9));
c2 = _mm_and_si128(_mm_or_si128(ag, rb), mask);
_mm_store_si128(&dstp[i / 2], _mm_packus_epi32(c1, c2));
}
// The remainder starts right after those done via SSE.
u32 i = sseChunks * 4;
#else
u32 i = 0;
#endif
for (; i < numPixels; i++) {
dst[i] = RGBA8888toRGBA5551(src[i]);
}
}
inline u16 BGRA8888toRGBA5551(u32 px) {
return ((px >> 19) & 0x001F) | ((px >> 6) & 0x03E0) | ((px << 7) & 0x7C00) | ((px >> 16) & 0x8000);
}
void ConvertBGRA8888ToRGBA5551(u16 *dst, const u32 *src, const u32 numPixels) {
#if _M_SSE >= 0x401
const __m128i maskAG = _mm_set1_epi32(0x8000F800);
const __m128i maskRB = _mm_set1_epi32(0x00F800F8);
const __m128i mask = _mm_set1_epi32(0x0000FFFF);
const __m128i *srcp = (const __m128i *)src;
__m128i *dstp = (__m128i *)dst;
u32 sseChunks = (numPixels / 4) & ~1;
// SSE 4.1 required for _mm_packus_epi32.
if (((intptr_t)src & 0xF) || ((intptr_t)dst & 0xF) || !cpu_info.bSSE4_1) {
sseChunks = 0;
}
for (u32 i = 0; i < sseChunks; i += 2) {
__m128i c1 = _mm_load_si128(&srcp[i + 0]);
__m128i c2 = _mm_load_si128(&srcp[i + 1]);
__m128i ag, rb;
ag = _mm_and_si128(c1, maskAG);
ag = _mm_or_si128(_mm_srli_epi32(ag, 16), _mm_srli_epi32(ag, 6));
rb = _mm_and_si128(c1, maskRB);
rb = _mm_or_si128(_mm_srli_epi32(rb, 19), _mm_slli_epi32(rb, 7));
c1 = _mm_and_si128(_mm_or_si128(ag, rb), mask);
ag = _mm_and_si128(c2, maskAG);
ag = _mm_or_si128(_mm_srli_epi32(ag, 16), _mm_srli_epi32(ag, 6));
rb = _mm_and_si128(c2, maskRB);
rb = _mm_or_si128(_mm_srli_epi32(rb, 19), _mm_slli_epi32(rb, 7));
c2 = _mm_and_si128(_mm_or_si128(ag, rb), mask);
_mm_store_si128(&dstp[i / 2], _mm_packus_epi32(c1, c2));
}
// The remainder starts right after those done via SSE.
u32 i = sseChunks * 4;
#else
u32 i = 0;
#endif
for (; i < numPixels; i++) {
dst[i] = BGRA8888toRGBA5551(src[i]);
}
}