libretro/ppsspp
1
0
Fork 0
mirror of https://github.com/libretro/ppsspp.git synced 2024-12-01 12:20:24 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #7751 from unknownbrackets/colorconv

Browse Source 
A bit more SSE color conversion funcs, stub out NEON space
This commit is contained in:
Henrik Rydgård 2015-05-18 09:30:54 +02:00
commit 5b75d5d33f
10 changed files with 338 additions and 121 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
CMakeLists.txt
View File

@ -290,7 +290,8 @@ set(CommonExtra)
if(ARM)
set(CommonExtra ${CommonExtra}
Common/ArmCPUDetect.cpp
Common/ArmThunk.cpp)
Common/ArmThunk.cpp
Common/ColorConvNEON.cpp)
elseif(X86)
set(CommonExtra ${CommonExtra}
Common/ABI.cpp

278
Common/ColorConv.cpp
View File

@ -15,9 +15,15 @@
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#include "ColorConv.h"
#include "CommonTypes.h"
// NEON is in a separate file so that it can be compiled with a runtime check.
#include "ColorConvNEON.h"
#include "Common.h"
#include "CPUDetect.h"
#ifdef _M_SSE
#include <xmmintrin.h>
#endif
inline u16 RGBA8888toRGB565(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 5) & 0x07E0) | ((px >> 8) & 0xF800);
@ -269,8 +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);
@ -280,8 +329,52 @@ void ConvertRGBA565ToRGBA8888(u32 *dst32, const u16 *src, const u32 numPixels) {
}
void ConvertRGBA5551ToRGBA8888(u32 *dst32, const u16 *src, const u32 numPixels) {
#ifdef _M_SSE
const __m128i mask5 = _mm_set1_epi16(0x001f);
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]);
// 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), mask5);
g = _mm_or_si128(_mm_slli_epi16(g, 3), _mm_srli_epi16(g, 2));
g = _mm_slli_epi16(g, 8);
// Almost done, we aim for BB00 BB00 again here.
__m128i b = _mm_and_si128(_mm_srli_epi16(c, 10), mask5);
b = _mm_or_si128(_mm_slli_epi16(b, 3), _mm_srli_epi16(b, 2));
b = _mm_and_si128(b, mask8);
// 1 bit A to 00AA 00AA.
__m128i a = _mm_srli_epi16(c, 15);
a = _mm_slli_epi16(_mm_cmpeq_epi16(a, one), 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] = Convert5To8((col >> 5) & 0x1f);
@ -290,9 +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);
@ -301,6 +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++) {
@ -332,4 +467,135 @@ void ConvertBGR565ToRGBA8888(u32 *dst, const u16 *src, const u32 numPixels) {
u16 col0 = src[x];
ARGB8From565(col0, &dst[x]);
}
}
}
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);
}
}

5
Common/ColorConv.h
View File

@ -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);

23
Common/ColorConvNEON.cpp Normal file
View File

@ -0,0 +1,23 @@
// Copyright (c) 2015- PPSSPP Project.
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0 or later versions.
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License 2.0 for more details.
// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#include <arm_neon.h>
#include "ColorConvNEON.h"
#include "Common.h"
#include "CPUDetect.h"
// TODO: NEON color conversion funcs.

20
Common/ColorConvNEON.h Normal file
View File

@ -0,0 +1,20 @@
// Copyright (c) 2015- PPSSPP Project.
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0 or later versions.
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License 2.0 for more details.
// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#pragma once
#include "ColorConv.h"

7
Common/Common.vcxproj
View File

@ -194,6 +194,7 @@
<ClInclude Include="Atomic_GCC.h" />
<ClInclude Include="Atomic_Win32.h" />
<ClInclude Include="BitSet.h" />
<ClInclude Include="ColorConvNEON.h" />
<ClInclude Include="ChunkFile.h" />
<ClInclude Include="CodeBlock.h" />
<ClInclude Include="ColorConv.h" />
@ -242,6 +243,12 @@
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
</ClCompile>
<ClCompile Include="ColorConvNEON.cpp">
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild>
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
</ClCompile>
<ClCompile Include="ChunkFile.cpp" />
<ClCompile Include="ColorConv.cpp" />
<ClCompile Include="ConsoleListener.cpp" />

2
Common/Common.vcxproj.filters
View File

@ -46,6 +46,7 @@
<ClInclude Include="BitSet.h" />
<ClInclude Include="CodeBlock.h" />
<ClInclude Include="ColorConv.h" />
<ClInclude Include="ColorConvNEON.h" />
</ItemGroup>
<ItemGroup>
<ClCompile Include="stdafx.cpp" />
@ -81,6 +82,7 @@
<ClCompile Include="MipsEmitter.cpp" />
<ClCompile Include="Arm64Emitter.cpp" />
<ClCompile Include="ColorConv.cpp" />
<ClCompile Include="ColorConvNEON.cpp" />
</ItemGroup>
<ItemGroup>
<None Include="CMakeLists.txt" />

118
GPU/GLES/TextureCache.cpp
View File

@ -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)

2
Qt/Common.pro
View File

@ -43,6 +43,8 @@ win32 {
HEADERS += $$P/Common/MemArena.h
}
armv7: SOURCES += $$P/Common/ColorConvNEON.cpp
SOURCES += $$P/Common/ChunkFile.cpp \
$$P/Common/ColorConv.cpp \
$$P/Common/ConsoleListener.cpp \

1
android/jni/Android.mk
View File

@ -58,6 +58,7 @@ ARCH_FILES := \
$(SRC)/Common/ArmEmitter.cpp \
$(SRC)/Common/ArmCPUDetect.cpp \
$(SRC)/Common/ArmThunk.cpp \
$(SRC)/Common/ColorConvNEON.cpp.neon \
$(SRC)/Core/MIPS/ARM/ArmCompALU.cpp \
$(SRC)/Core/MIPS/ARM/ArmCompBranch.cpp \
$(SRC)/Core/MIPS/ARM/ArmCompFPU.cpp \