diff --git a/Common/ColorConv.cpp b/Common/ColorConv.cpp index f0cc37898..90dcdb489 100644 --- a/Common/ColorConv.cpp +++ b/Common/ColorConv.cpp @@ -336,4 +336,135 @@ void ConvertBGR565ToRGBA8888(u32 *dst, const u16 *src, const u32 numPixels) { u16 col0 = src[x]; ARGB8From565(col0, &dst[x]); } -} \ No newline at end of file +} + +void ConvertRGBA4444ToABGR4444(u16 *dst, const u16 *src, const u32 numPixels) { +#ifdef _M_SSE + const __m128i maskB = _mm_set1_epi16(0x00F0); + const __m128i maskG = _mm_set1_epi16(0x0F00); + + const __m128i *srcp = (const __m128i *)src; + __m128i *dstp = (__m128i *)dst; + u32 sseChunks = numPixels / 8; + if (((intptr_t)src & 0xF) || ((intptr_t)dst & 0xF)) { + sseChunks = 0; + } + for (u32 i = 0; i < sseChunks; ++i) { + const __m128i c = _mm_load_si128(&srcp[i]); + __m128i v = _mm_srli_epi16(c, 12); + v = _mm_or_si128(v, _mm_and_si128(_mm_srli_epi16(c, 4), maskB)); + v = _mm_or_si128(v, _mm_and_si128(_mm_slli_epi16(c, 4), maskG)); + v = _mm_or_si128(v, _mm_slli_epi16(c, 12)); + _mm_store_si128(&dstp[i], v); + } + // The remainder is done in chunks of 2, SSE was chunks of 8. + u32 i = sseChunks * 8 / 2; +#else + u32 i = 0; +#endif + + const u32 *src32 = (const u32 *)src; + u32 *dst32 = (u32 *)dst; + for (; i < numPixels / 2; i++) { + const u32 c = src32[i]; + dst32[i] = ((c >> 12) & 0x000F000F) | + ((c >> 4) & 0x00F000F0) | + ((c << 4) & 0x0F000F00) | + ((c << 12) & 0xF000F000); + } + + if (numPixels & 1) { + const u32 i = numPixels - 1; + const u16 c = src[i]; + dst[i] = ((c >> 12) & 0x000F) | + ((c >> 4) & 0x00F0) | + ((c << 4) & 0x0F00) | + ((c << 12) & 0xF000); + } +} + +void ConvertRGBA5551ToABGR1555(u16 *dst, const u16 *src, const u32 numPixels) { +#ifdef _M_SSE + const __m128i maskB = _mm_set1_epi16(0x003E); + const __m128i maskG = _mm_set1_epi16(0x07C0); + + const __m128i *srcp = (const __m128i *)src; + __m128i *dstp = (__m128i *)dst; + u32 sseChunks = numPixels / 8; + if (((intptr_t)src & 0xF) || ((intptr_t)dst & 0xF)) { + sseChunks = 0; + } + for (u32 i = 0; i < sseChunks; ++i) { + const __m128i c = _mm_load_si128(&srcp[i]); + __m128i v = _mm_srli_epi16(c, 15); + v = _mm_or_si128(v, _mm_and_si128(_mm_srli_epi16(c, 9), maskB)); + v = _mm_or_si128(v, _mm_and_si128(_mm_slli_epi16(c, 1), maskG)); + v = _mm_or_si128(v, _mm_slli_epi16(c, 11)); + _mm_store_si128(&dstp[i], v); + } + // The remainder is done in chunks of 2, SSE was chunks of 8. + u32 i = sseChunks * 8 / 2; +#else + u32 i = 0; +#endif + + const u32 *src32 = (const u32 *)src; + u32 *dst32 = (u32 *)dst; + for (; i < numPixels / 2; i++) { + const u32 c = src32[i]; + dst32[i] = ((c >> 15) & 0x00010001) | + ((c >> 9) & 0x003E003E) | + ((c << 1) & 0x07C007C0) | + ((c << 11) & 0xF800F800); + } + + if (numPixels & 1) { + const u32 i = numPixels - 1; + const u16 c = src[i]; + dst[i] = ((c >> 15) & 0x0001) | + ((c >> 9) & 0x003E) | + ((c << 1) & 0x07C0) | + ((c << 11) & 0xF800); + } +} + +void ConvertRGB565ToBGR565(u16 *dst, const u16 *src, const u32 numPixels) { +#ifdef _M_SSE + const __m128i maskG = _mm_set1_epi16(0x07E0); + + const __m128i *srcp = (const __m128i *)src; + __m128i *dstp = (__m128i *)dst; + u32 sseChunks = numPixels / 8; + if (((intptr_t)src & 0xF) || ((intptr_t)dst & 0xF)) { + sseChunks = 0; + } + for (u32 i = 0; i < sseChunks; ++i) { + const __m128i c = _mm_load_si128(&srcp[i]); + __m128i v = _mm_srli_epi16(c, 11); + v = _mm_or_si128(v, _mm_and_si128(c, maskG)); + v = _mm_or_si128(v, _mm_slli_epi16(c, 11)); + _mm_store_si128(&dstp[i], v); + } + // The remainder is done in chunks of 2, SSE was chunks of 8. + u32 i = sseChunks * 8 / 2; +#else + u32 i = 0; +#endif + + const u32 *src32 = (const u32 *)src; + u32 *dst32 = (u32 *)dst; + for (; i < numPixels / 2; i++) { + const u32 c = src32[i]; + dst32[i] = ((c >> 11) & 0x001F001F) | + ((c >> 0) & 0x07E007E0) | + ((c << 11) & 0xF800F800); + } + + if (numPixels & 1) { + const u32 i = numPixels - 1; + const u16 c = src[i]; + dst[i] = ((c >> 11) & 0x001F) | + ((c >> 0) & 0x07E0) | + ((c << 11) & 0xF800); + } +} diff --git a/Common/ColorConv.h b/Common/ColorConv.h index 220cc2bf1..9b1d6c56e 100644 --- a/Common/ColorConv.h +++ b/Common/ColorConv.h @@ -105,6 +105,7 @@ void convert5551_dx9(u16* data, u32* out, int width, int l, int u); // "Complete" set of color conversion functions between the usual formats. void ConvertBGRA8888ToRGBA8888(u32 *dst, const u32 *src, const u32 numPixels); +#define ConvertRGBA8888ToBGRA8888 ConvertBGRA8888ToRGBA8888 void ConvertRGBA8888ToRGBA5551(u16 *dst, const u32 *src, const u32 numPixels); void ConvertRGBA8888ToRGB565(u16 *dst, const u32 *src, const u32 numPixels); @@ -121,3 +122,7 @@ void ConvertRGBA4444ToRGBA8888(u32 *dst, const u16 *src, const u32 numPixels); void ConvertBGRA4444ToRGBA8888(u32 *dst, const u16 *src, const u32 numPixels); void ConvertBGRA5551ToRGBA8888(u32 *dst, const u16 *src, const u32 numPixels); void ConvertBGR565ToRGBA8888(u32 *dst, const u16 *src, const u32 numPixels); + +void ConvertRGBA4444ToABGR4444(u16 *dst, const u16 *src, const u32 numPixels); +void ConvertRGBA5551ToABGR1555(u16 *dst, const u16 *src, const u32 numPixels); +void ConvertRGB565ToBGR565(u16 *dst, const u16 *src, const u32 numPixels); diff --git a/GPU/GLES/TextureCache.cpp b/GPU/GLES/TextureCache.cpp index 77ef5cf4e..c711149e3 100644 --- a/GPU/GLES/TextureCache.cpp +++ b/GPU/GLES/TextureCache.cpp @@ -787,128 +787,18 @@ static void ConvertColors(void *dstBuf, const void *srcBuf, GLuint dstFmt, int n u32 *dst = (u32 *)dstBuf; switch (dstFmt) { case GL_UNSIGNED_SHORT_4_4_4_4: - { -#ifdef _M_SSE - const __m128i maskB = _mm_set1_epi16(0x00F0); - const __m128i maskG = _mm_set1_epi16(0x0F00); - - __m128i *srcp = (__m128i *)src; - __m128i *dstp = (__m128i *)dst; - const int sseChunks = numPixels / 8; - for (int i = 0; i < sseChunks; ++i) { - __m128i c = _mm_load_si128(&srcp[i]); - __m128i v = _mm_srli_epi16(c, 12); - v = _mm_or_si128(v, _mm_and_si128(_mm_srli_epi16(c, 4), maskB)); - v = _mm_or_si128(v, _mm_and_si128(_mm_slli_epi16(c, 4), maskG)); - v = _mm_or_si128(v, _mm_slli_epi16(c, 12)); - _mm_store_si128(&dstp[i], v); - } - // The remainder is done in chunks of 2, SSE was chunks of 8. - int i = sseChunks * 8 / 2; -#else - int i = 0; - // TODO: NEON. -#endif - for (; i < (numPixels + 1) / 2; i++) { - u32 c = src[i]; - dst[i] = ((c >> 12) & 0x000F000F) | - ((c >> 4) & 0x00F000F0) | - ((c << 4) & 0x0F000F00) | - ((c << 12) & 0xF000F000); - } - } + ConvertRGBA4444ToABGR4444((u16 *)dst, (const u16 *)src, numPixels); break; // Final Fantasy 2 uses this heavily in animated textures. case GL_UNSIGNED_SHORT_5_5_5_1: - { -#ifdef _M_SSE - const __m128i maskB = _mm_set1_epi16(0x003E); - const __m128i maskG = _mm_set1_epi16(0x07C0); - - __m128i *srcp = (__m128i *)src; - __m128i *dstp = (__m128i *)dst; - const int sseChunks = numPixels / 8; - for (int i = 0; i < sseChunks; ++i) { - __m128i c = _mm_load_si128(&srcp[i]); - __m128i v = _mm_srli_epi16(c, 15); - v = _mm_or_si128(v, _mm_and_si128(_mm_srli_epi16(c, 9), maskB)); - v = _mm_or_si128(v, _mm_and_si128(_mm_slli_epi16(c, 1), maskG)); - v = _mm_or_si128(v, _mm_slli_epi16(c, 11)); - _mm_store_si128(&dstp[i], v); - } - // The remainder is done in chunks of 2, SSE was chunks of 8. - int i = sseChunks * 8 / 2; -#else - int i = 0; - // TODO: NEON. -#endif - for (; i < (numPixels + 1) / 2; i++) { - u32 c = src[i]; - dst[i] = ((c >> 15) & 0x00010001) | - ((c >> 9) & 0x003E003E) | - ((c << 1) & 0x07C007C0) | - ((c << 11) & 0xF800F800); - } - } + ConvertRGBA5551ToABGR1555((u16 *)dst, (const u16 *)src, numPixels); break; case GL_UNSIGNED_SHORT_5_6_5: - { -#ifdef _M_SSE - const __m128i maskG = _mm_set1_epi16(0x07E0); - - __m128i *srcp = (__m128i *)src; - __m128i *dstp = (__m128i *)dst; - const int sseChunks = numPixels / 8; - for (int i = 0; i < sseChunks; ++i) { - __m128i c = _mm_load_si128(&srcp[i]); - __m128i v = _mm_srli_epi16(c, 11); - v = _mm_or_si128(v, _mm_and_si128(c, maskG)); - v = _mm_or_si128(v, _mm_slli_epi16(c, 11)); - _mm_store_si128(&dstp[i], v); - } - // The remainder is done in chunks of 2, SSE was chunks of 8. - int i = sseChunks * 8 / 2; -#else - int i = 0; - // TODO: NEON. -#endif - for (; i < (numPixels + 1) / 2; i++) { - u32 c = src[i]; - dst[i] = ((c >> 11) & 0x001F001F) | - ((c >> 0) & 0x07E007E0) | - ((c << 11) & 0xF800F800); - } - } + ConvertRGB565ToBGR565((u16 *)dst, (const u16 *)src, numPixels); break; default: if (UseBGRA8888()) { -#ifdef _M_SSE - const __m128i maskGA = _mm_set1_epi32(0xFF00FF00); - - __m128i *srcp = (__m128i *)src; - __m128i *dstp = (__m128i *)dst; - const int sseChunks = numPixels / 4; - for (int 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. - int i = sseChunks * 4; -#else - int i = 0; -#endif - for (; i < numPixels; i++) { - u32 c = src[i]; - dst[i] = ((c >> 16) & 0x000000FF) | - ((c >> 0) & 0xFF00FF00) | - ((c << 16) & 0x00FF0000); - } + ConvertRGBA8888ToBGRA8888(dst, src, numPixels); } else { // No need to convert RGBA8888, right order already if (dst != src)