Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

Add some additional SSE for color conversion.

Browse source
This commit is contained in:
Unknown W. Brackets 2015-05-17 14:56:46 -07:00
parent 1cb2302f5c
commit ccdaf41b53
1 changed files with 87 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

90
Common/ColorConv.cpp
View file

@ -275,8 +275,51 @@ void ConvertRGBA8888ToRGBA4444(u16 *dst, const u32 *src, const u32 numPixels) {
}
void ConvertRGBA565ToRGBA8888(u32 *dst32, const u16 *src, const u32 numPixels) {
#ifdef _M_SSE
const __m128i mask5 = _mm_set1_epi16(0x001f);
const __m128i mask6 = _mm_set1_epi16(0x003f);
const __m128i mask8 = _mm_set1_epi16(0x00ff);
const __m128i *srcp = (const __m128i *)src;
__m128i *dstp = (__m128i *)dst32;
u32 sseChunks = numPixels / 8;
if (((intptr_t)src & 0xF) || ((intptr_t)dst32 & 0xF)) {
sseChunks = 0;
}
for (u32 i = 0; i < sseChunks; ++i) {
const __m128i c = _mm_load_si128(&srcp[i]);
// Swizzle, resulting in RR00 RR00.
__m128i r = _mm_and_si128(c, mask5);
r = _mm_or_si128(_mm_slli_epi16(r, 3), _mm_srli_epi16(r, 2));
r = _mm_and_si128(r, mask8);
// This one becomes 00GG 00GG.
__m128i g = _mm_and_si128(_mm_srli_epi16(c, 5), mask6);
g = _mm_or_si128(_mm_slli_epi16(g, 2), _mm_srli_epi16(g, 4));
g = _mm_slli_epi16(g, 8);
// Almost done, we aim for BB00 BB00 again here.
__m128i b = _mm_and_si128(_mm_srli_epi16(c, 11), mask5);
b = _mm_or_si128(_mm_slli_epi16(b, 3), _mm_srli_epi16(b, 2));
b = _mm_and_si128(b, mask8);
// Always set to 00FF 00FF.
__m128i a = _mm_slli_epi16(mask8, 8);
// Now combine them, RRGG RRGG and BBAA BBAA, and then interleave.
const __m128i rg = _mm_or_si128(r, g);
const __m128i ba = _mm_or_si128(b, a);
_mm_store_si128(&dstp[i * 2 + 0], _mm_unpacklo_epi16(rg, ba));
_mm_store_si128(&dstp[i * 2 + 1], _mm_unpackhi_epi16(rg, ba));
}
u32 i = sseChunks * 8;
#else
u32 i = 0;
#endif
u8 *dst = (u8 *)dst32;
for (u32 x = 0; x < numPixels; x++) {
for (u32 x = i; x < numPixels; x++) {
u16 col = src[x];
dst[x * 4] = Convert5To8((col) & 0x1f);
dst[x * 4 + 1] = Convert6To8((col >> 5) & 0x3f);
@ -290,7 +333,6 @@ void ConvertRGBA5551ToRGBA8888(u32 *dst32, const u16 *src, const u32 numPixels)
const __m128i mask5 = _mm_set1_epi16(0x001f);
const __m128i mask8 = _mm_set1_epi16(0x00ff);
const __m128i one = _mm_set1_epi16(0x0001);
const __m128i zero = _mm_set1_epi16(0x0000);
const __m128i *srcp = (const __m128i *)src;
__m128i *dstp = (__m128i *)dst32;
@ -341,9 +383,50 @@ void ConvertRGBA5551ToRGBA8888(u32 *dst32, const u16 *src, const u32 numPixels)
}
}
// TODO: This seems to be BGRA4444 -> RGBA888?
void ConvertRGBA4444ToRGBA8888(u32 *dst32, const u16 *src, const u32 numPixels) {
#ifdef _M_SSE
const __m128i mask4 = _mm_set1_epi16(0x000f);
const __m128i mask8 = _mm_set1_epi16(0x00ff);
const __m128i one = _mm_set1_epi16(0x0001);
const __m128i *srcp = (const __m128i *)src;
__m128i *dstp = (__m128i *)dst32;
u32 sseChunks = numPixels / 8;
if (((intptr_t)src & 0xF) || ((intptr_t)dst32 & 0xF)) {
sseChunks = 0;
}
for (u32 i = 0; i < sseChunks; ++i) {
const __m128i c = _mm_load_si128(&srcp[i]);
// Let's just grab R000 R000, without swizzling yet.
__m128i r = _mm_and_si128(_mm_srli_epi16(c, 8), mask4);
// And then 00G0 00G0.
__m128i g = _mm_and_si128(_mm_srli_epi16(c, 4), mask4);
g = _mm_slli_epi16(g, 8);
// Now B000 B000.
__m128i b = _mm_and_si128(c, mask4);
// And lastly 00A0 00A0. No mask needed, we have a wall.
__m128i a = _mm_srli_epi16(c, 12);
a = _mm_slli_epi16(g, 8);
// We swizzle after combining - R0G0 R0G0 and B0A0 B0A0 -> RRGG RRGG and BBAA BBAA.
__m128i rg = _mm_or_si128(r, g);
__m128i ba = _mm_or_si128(b, a);
rg = _mm_or_si128(rg, _mm_slli_epi16(rg, 4));
ba = _mm_or_si128(ba, _mm_slli_epi16(ba, 4));
// And then we can store.
_mm_store_si128(&dstp[i * 2 + 0], _mm_unpacklo_epi16(rg, ba));
_mm_store_si128(&dstp[i * 2 + 1], _mm_unpackhi_epi16(rg, ba));
}
u32 i = sseChunks * 8;
#else
u32 i = 0;
#endif
u8 *dst = (u8 *)dst32;
for (u32 x = 0; x < numPixels; x++) {
for (u32 x = i; x < numPixels; x++) {
u16 col = src[x];
dst[x * 4] = Convert4To8((col >> 8) & 0xf);
dst[x * 4 + 1] = Convert4To8((col >> 4) & 0xf);
@ -352,6 +435,7 @@ void ConvertRGBA4444ToRGBA8888(u32 *dst32, const u16 *src, const u32 numPixels)
}
}
// TODO: This seems to be ABGR4444 -> RGBA888?
void ConvertBGRA4444ToRGBA8888(u32 *dst32, const u16 *src, const u32 numPixels) {
u8 *dst = (u8 *)dst32;
for (u32 x = 0; x < numPixels; x++) {