thirdparty/ppsspp
2
0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Revert "Merge pull request #7361 from hrydgard/color-conv-centralize"

Browse source 
This reverts commit f1b57dabf5, reversing
changes made to 41001637ce.
This commit is contained in:
Henrik Rydgard 2015-01-22 19:53:32 +01:00
parent 07933cad42
commit cae58cafee
33 changed files with 1195 additions and 476 deletions
Download patch file Download diff file Expand all files Collapse all files

6
CMakeLists.txt
View file

@ -326,8 +326,6 @@ add_library(Common STATIC
${CommonExtra}
Common/ChunkFile.cpp
Common/ChunkFile.h
Common/ColorConv.cpp
Common/ColorConv.h
Common/ConsoleListener.cpp
Common/ConsoleListener.h
Common/Crypto/md5.cpp
@ -1380,8 +1378,6 @@ add_library(GPU OBJECT
GPU/Common/IndexGenerator.h
GPU/Common/TextureDecoder.cpp
GPU/Common/TextureDecoder.h
GPU/Common/TextureScaler.cpp
GPU/Common/TextureScaler.h
GPU/Common/TextureCacheCommon.cpp
GPU/Common/TextureCacheCommon.h
${GPU_NEON}
@ -1410,6 +1406,8 @@ add_library(GPU OBJECT
GPU/GLES/StencilBuffer.cpp
GPU/GLES/TextureCache.cpp
GPU/GLES/TextureCache.h
GPU/GLES/TextureScaler.cpp
GPU/GLES/TextureScaler.h
GPU/GLES/TransformPipeline.cpp
GPU/GLES/TransformPipeline.h
GPU/GLES/VertexShaderGenerator.cpp

2
Common/Arm64Emitter.cpp
View file

@ -1512,7 +1512,7 @@ void ARM64XEmitter::ABI_PushRegisters(BitSet32 registers)
{
if (first)
{
STR(INDEX_PRE, (ARM64Reg)(X0 + it), SP, (u32)-(s32)stack_size);
STR(INDEX_PRE, (ARM64Reg)(X0 + it), SP, -stack_size);
first = false;
current_offset += 16;
}

3
Common/CPUDetect.cpp
View file

@ -21,12 +21,11 @@
#endif
#include <memory.h>
#include "base/logging.h"
#include "base/basictypes.h"
#include "CPUDetect.h"
#include "Common.h"
#include "CPUDetect.h"
#include "StringUtils.h"
#ifdef _WIN32

271
Common/ColorConv.cpp
View file

@ -1,271 +0,0 @@
// 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/.
// TODO: Make SSE2 and NEON versions of many of these.
#include "Common.h"
#include "CPUDetect.h"
#include "ColorConv.h"
#include "CommonTypes.h"
#ifdef _M_SSE
#include <xmmintrin.h>
#endif
inline u16 RGBA8888toRGB565(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 5) & 0x07E0) | ((px >> 8) & 0xF800);
}
inline u16 RGBA8888toRGBA4444(u32 px) {
return ((px >> 4) & 0x000F) | ((px >> 8) & 0x00F0) | ((px >> 12) & 0x0F00) | ((px >> 16) & 0xF000);
}
inline u16 RGBA8888toRGBA5551(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 6) & 0x03E0) | ((px >> 9) & 0x7C00) | ((px >> 16) & 0x8000);
}
inline u16 BGRA8888toRGB565(u32 px) {
return ((px >> 19) & 0x001F) | ((px >> 5) & 0x07E0) | ((px << 8) & 0xF800);
}
inline u16 BGRA8888toRGBA4444(u32 px) {
return ((px >> 20) & 0x000F) | ((px >> 8) & 0x00F0) | ((px << 4) & 0x0F00) | ((px >> 16) & 0xF000);
}
inline u32 RGBA2BGRA(u32 src) {
const u32 r = (src & 0x000000FF) << 16;
const u32 ga = src & 0xFF00FF00;
const u32 b = (src & 0x00FF0000) >> 16;
return r | ga | b;
}
// Used heavily in Test Drive Unlimited (for no good reason...)
void ConvertBGRA8888ToRGB565(u16 *dst, const u32 *src, int numPixels) {
#if _M_SSE >= 0x401
const __m128i maskG = _mm_set1_epi32(0x0000FC00);
const __m128i maskRB = _mm_set1_epi32(0x00F800F8);
const __m128i mask = _mm_set1_epi32(0x0000FFFF);
const __m128i *srcp = (const __m128i *)src;
__m128i *dstp = (__m128i *)dst;
int 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 (int i = 0; i < sseChunks; i += 2) {
__m128i c1 = _mm_load_si128(&srcp[i + 0]);
__m128i c2 = _mm_load_si128(&srcp[i + 1]);
__m128i g, rb;
g = _mm_and_si128(c1, maskG);
g = _mm_srli_epi32(g, 5);
rb = _mm_and_si128(c1, maskRB);
rb = _mm_or_si128(_mm_srli_epi32(rb, 3), _mm_srli_epi32(rb, 8));
c1 = _mm_and_si128(_mm_or_si128(g, rb), mask);
g = _mm_and_si128(c2, maskG);
g = _mm_srli_epi32(g, 5);
rb = _mm_and_si128(c2, maskRB);
rb = _mm_or_si128(_mm_srli_epi32(rb, 3), _mm_srli_epi32(rb, 8));
c2 = _mm_and_si128(_mm_or_si128(g, 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 (int x = 0; x < numPixels; ++x) {
dst[x] = BGRA8888toRGB565(src[x]);
}
}
void ConvertRGBA8888ToRGB565(u16 *dst, const u32 *src, int numPixels) {
for (int x = 0; x < numPixels; x++) {
dst[x] = RGBA8888toRGB565(src[x]);
}
}
void ConvertBGRA8888ToRGBA4444(u16 *dst, const u32 *src, int numPixels) {
for (int x = 0; x < numPixels; ++x) {
dst[x] = BGRA8888toRGBA4444(src[x]);
}
}
void ConvertRGBA8888ToRGBA4444(u16 *dst, const u32 *src, int numPixels) {
for (int x = 0; x < numPixels; ++x) {
dst[x] = RGBA8888toRGBA4444(src[x]);
}
}
void ConvertBGRA8888ToRGBA8888(u32 *dst, const u32 *src, int numPixels) {
#ifdef _M_SSE
const __m128i maskGA = _mm_set1_epi32(0xFF00FF00);
const __m128i *srcp = (const __m128i *)src;
__m128i *dstp = (__m128i *)dst;
int sseChunks = numPixels / 4;
if (((intptr_t)src & 0xF) || ((intptr_t)dst & 0xF)) {
sseChunks = 0;
}
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++) {
const u32 c = src[i];
dst[i] = ((c >> 16) & 0x000000FF) |
((c >> 0) & 0xFF00FF00) |
((c << 16) & 0x00FF0000);
}
}
void ConvertRGBA8888ToRGBA5551(u16 *dst, const u32 *src, int 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;
int 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 (int 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.
int i = sseChunks * 4;
#else
int 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, int 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;
int 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 (int 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.
int i = sseChunks * 4;
#else
int i = 0;
#endif
for (; i < numPixels; i++) {
dst[i] = BGRA8888toRGBA5551(src[i]);
}
}
void ConvertRGB565ToRGBA888F(u32 *dst32, const u16 *src, int numPixels) {
u8 *dst = (u8 *)dst32;
for (int x = 0; x < numPixels; x++) {
u16 col = src[x];
dst[x * 4] = Convert5To8((col)& 0x1f);
dst[x * 4 + 1] = Convert6To8((col >> 5) & 0x3f);
dst[x * 4 + 2] = Convert5To8((col >> 11) & 0x1f);
dst[x * 4 + 3] = 255;
}
}
void ConvertRGBA5551ToRGBA8888(u32 *dst32, const u16 *src, int numPixels) {
u8 *dst = (u8 *)dst32;
for (int x = 0; x < numPixels; x++) {
u16 col = src[x];
dst[x * 4] = Convert5To8((col)& 0x1f);
dst[x * 4 + 1] = Convert5To8((col >> 5) & 0x1f);
dst[x * 4 + 2] = Convert5To8((col >> 10) & 0x1f);
dst[x * 4 + 3] = (col >> 15) ? 255 : 0;
}
}
void ConvertRGBA4444ToRGBA8888(u32 *dst32, const u16 *src, int numPixels) {
u8 *dst = (u8 *)dst32;
for (int x = 0; x < numPixels; x++) {
u16 col = src[x];
dst[x * 4] = Convert4To8((col >> 8) & 0xf);
dst[x * 4 + 1] = Convert4To8((col >> 4) & 0xf);
dst[x * 4 + 2] = Convert4To8(col & 0xf);
dst[x * 4 + 3] = Convert4To8(col >> 12);
}
}

2
Common/Common.vcxproj
View file

@ -194,7 +194,6 @@
<ClInclude Include="BitSet.h" />
<ClInclude Include="ChunkFile.h" />
<ClInclude Include="CodeBlock.h" />
<ClInclude Include="ColorConv.h" />
<ClInclude Include="Common.h" />
<ClInclude Include="CommonFuncs.h" />
<ClInclude Include="CommonTypes.h" />
@ -241,7 +240,6 @@
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
</ClCompile>
<ClCompile Include="ChunkFile.cpp" />
<ClCompile Include="ColorConv.cpp" />
<ClCompile Include="ConsoleListener.cpp" />
<ClCompile Include="CPUDetect.cpp" />
<ClCompile Include="Crypto\md5.cpp" />

2
Common/Common.vcxproj.filters
View file

@ -7,7 +7,6 @@
<ClInclude Include="Atomic_GCC.h" />
<ClInclude Include="Atomic_Win32.h" />
<ClInclude Include="ChunkFile.h" />
<ClInclude Include="ColorConv.h" />
<ClInclude Include="Common.h" />
<ClInclude Include="CommonFuncs.h" />
<ClInclude Include="CommonTypes.h" />
@ -50,7 +49,6 @@
<ItemGroup>
<ClCompile Include="stdafx.cpp" />
<ClCompile Include="ABI.cpp" />
<ClCompile Include="ColorConv.cpp" />
<ClCompile Include="ConsoleListener.cpp" />
<ClCompile Include="CPUDetect.cpp" />
<ClCompile Include="FileUtil.cpp" />

18
Core/Screenshot.cpp
View file

@ -23,7 +23,6 @@
#endif
#include "Common/FileUtil.h"
#include "Common/ColorConv.h"
#include "Core/Config.h"
#include "Core/Screenshot.h"
#include "GPU/Common/GPUDebugInterface.h"
@ -49,7 +48,8 @@ public:
}
}
bool put_buf(const void *buf, int len) override {
bool put_buf(const void *buf, int len) override
{
if (fp_) {
if (fwrite(buf, len, 1, fp_) != 1) {
fclose(fp_);
@ -149,13 +149,14 @@ static const u8 *ConvertBufferTo888RGB(const GPUDebugBuffer &buf, u8 *&temp) {
const u16 *buf16 = (const u16 *)buffer;
const u32 *buf32 = (const u32 *)buffer;
for (u32 y = 0; y < buf.GetHeight(); y++) {
u8 *dst;
if (flip) {
dst = &temp[(buf.GetHeight() - y - 1) * buf.GetStride() * 3];
} else {
dst = &temp[y * buf.GetStride() * 3];
}
for (u32 x = 0; x < buf.GetStride(); x++) {
u8 *dst;
if (flip) {
dst = &temp[(buf.GetHeight() - y - 1) * buf.GetStride() * 3 + x * 3];
} else {
dst = &temp[y * buf.GetStride() * 3 + x * 3];
}
u8 &r = brswap ? dst[2] : dst[0];
u8 &g = dst[1];
u8 &b = brswap ? dst[0] : dst[2];
@ -202,7 +203,6 @@ static const u8 *ConvertBufferTo888RGB(const GPUDebugBuffer &buf, u8 *&temp) {
ERROR_LOG(COMMON, "Unsupported framebuffer format for screenshot: %d", buf.GetFormat());
return nullptr;
}
dst += 3;
}
}
buffer = temp;

127
GPU/Common/TextureDecoder.cpp
View file

@ -17,8 +17,6 @@
#include "ext/xxhash.h"
#include "Common/CPUDetect.h"
#include "Common/ColorConv.h"
#include "GPU/Common/TextureDecoder.h"
// NEON is in a separate file so that it can be compiled with a runtime check.
#include "GPU/Common/TextureDecoderNEON.h"
@ -329,3 +327,128 @@ 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);
}
}
inline u16 RGBA8888toRGBA5551(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 6) & 0x03E0) | ((px >> 9) & 0x7C00) | ((px >> 16) & 0x8000);
}
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]);
}
}

4
GPU/Common/TextureDecoder.h
View file

@ -216,3 +216,7 @@ inline void DeIndexTexture4Optimal(ClutT *dest, const u32 texaddr, int length, C
const u8 *indexed = (const u8 *) Memory::GetPointer(texaddr);
DeIndexTexture4Optimal(dest, indexed, length, color);
}
void ConvertBGRA8888ToRGBA8888(u32 *dst, const u32 *src, const u32 numPixels);
void ConvertRGBA8888ToRGBA5551(u16 *dst, const u32 *src, const u32 numPixels);
void ConvertBGRA8888ToRGBA5551(u16 *dst, const u32 *src, const u32 numPixels);

13
GPU/Common/VertexDecoderCommon.cpp
View file

@ -21,7 +21,6 @@
#include "base/logging.h"
#include "Common/CPUDetect.h"
#include "Common/ColorConv.h"
#include "Core/Config.h"
#include "Core/MemMap.h"
#include "Core/HDRemaster.h"
@ -365,7 +364,8 @@ void VertexDecoder::Step_TcFloatPrescale() const {
uv[1] = uvdata[1] * gstate_c.uv.vScale + gstate_c.uv.vOff;
}
void VertexDecoder::Step_Color565() const {
void VertexDecoder::Step_Color565() const
{
u8 *c = decoded_ + decFmt.c0off;
u16 cdata = *(u16_le *)(ptr_ + coloff);
c[0] = Convert5To8(cdata & 0x1f);
@ -375,7 +375,8 @@ void VertexDecoder::Step_Color565() const {
// Always full alpha.
}
void VertexDecoder::Step_Color5551() const {
void VertexDecoder::Step_Color5551() const
{
u8 *c = decoded_ + decFmt.c0off;
u16 cdata = *(u16_le *)(ptr_ + coloff);
c[0] = Convert5To8(cdata & 0x1f);
@ -385,7 +386,8 @@ void VertexDecoder::Step_Color5551() const {
gstate_c.vertexFullAlpha = gstate_c.vertexFullAlpha && c[3] != 0;
}
void VertexDecoder::Step_Color4444() const {
void VertexDecoder::Step_Color4444() const
{
u8 *c = decoded_ + decFmt.c0off;
u16 cdata = *(u16_le *)(ptr_ + coloff);
for (int j = 0; j < 4; j++)
@ -393,7 +395,8 @@ void VertexDecoder::Step_Color4444() const {
gstate_c.vertexFullAlpha = gstate_c.vertexFullAlpha && c[3] == 255;
}
void VertexDecoder::Step_Color8888() const {
void VertexDecoder::Step_Color8888() const
{
u8 *c = decoded_ + decFmt.c0off;
const u8 *cdata = (const u8*)(ptr_ + coloff);
memcpy(c, cdata, sizeof(u8) * 4);

84
GPU/Directx9/FramebufferDX9.cpp
View file

@ -16,8 +16,6 @@
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#include "math/lin/matrix4x4.h"
#include "Common/ColorConv.h"
#include "Common/CommonTypes.h"
#include "Core/Host.h"
#include "Core/MemMap.h"
#include "Core/Config.h"
@ -40,6 +38,26 @@
#include <algorithm>
namespace DX9 {
inline u16 RGBA8888toRGB565(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 5) & 0x07E0) | ((px >> 8) & 0xF800);
}
inline u16 RGBA8888toRGBA4444(u32 px) {
return ((px >> 4) & 0x000F) | ((px >> 8) & 0x00F0) | ((px >> 12) & 0x0F00) | ((px >> 16) & 0xF000);
}
inline u16 RGBA8888toRGBA5551(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 6) & 0x03E0) | ((px >> 9) & 0x7C00) | ((px >> 16) & 0x8000);
}
inline u16 BGRA8888toRGB565(u32 px) {
return ((px >> 19) & 0x001F) | ((px >> 5) & 0x07E0) | ((px << 8) & 0xF800);
}
inline u16 BGRA8888toRGBA4444(u32 px) {
return ((px >> 20) & 0x000F) | ((px >> 8) & 0x00F0) | ((px << 4) & 0x0F00) | ((px >> 16) & 0xF000);
}
static void ConvertFromRGBA8888(u8 *dst, u8 *src, u32 dstStride, u32 srcStride, u32 width, u32 height, GEBufferFormat format);
void CenterRect(float *x, float *y, float *w, float *h,
@ -131,6 +149,24 @@ namespace DX9 {
}
}
static inline void ARGB8From4444(u16 c, u32 * dst) {
*dst = ((c & 0xf) << 4) | (((c >> 4) & 0xf) << 12) | (((c >> 8) & 0xf) << 20) | ((c >> 12) << 28);
}
static inline void ARGB8From565(u16 c, u32 * dst) {
*dst = ((c & 0x001f) << 19) | (((c >> 5) & 0x003f) << 11) | ((((c >> 10) & 0x001f) << 3)) | 0xFF000000;
}
static inline void ARGB8From5551(u16 c, u32 * dst) {
*dst = ((c & 0x001f) << 19) | (((c >> 5) & 0x001f) << 11) | ((((c >> 10) & 0x001f) << 3)) | 0xFF000000;
}
// TODO: Swizzle the texture access instead.
static inline u32 RGBA2BGRA(u32 src) {
const u32 r = (src & 0x000000FF) << 16;
const u32 ga = src & 0xFF00FF00;
const u32 b = (src & 0x00FF0000) >> 16;
return r | ga | b;
}
void FramebufferManagerDX9::MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height) {
u8 *convBuf = NULL;
D3DLOCKED_RECT rect;
@ -171,26 +207,40 @@ namespace DX9 {
if (srcPixelFormat != GE_FORMAT_8888 || srcStride != 512) {
for (int y = 0; y < height; y++) {
switch (srcPixelFormat) {
// not tested
// not tested
case GE_FORMAT_565:
{
const u16_le *src = (const u16_le *)srcPixels + srcStride * y;
u32 *dst = (u32 *)(convBuf + rect.Pitch * y);
ConvertRGB565ToRGBA888F(dst, src, width);
for (int x = 0; x < width; x++) {
u16_le col0 = src[x+0];
ARGB8From565(col0, &dst[x + 0]);
}
}
break;
// faster
case GE_FORMAT_5551:
{
const u16_le *src = (const u16_le *)srcPixels + srcStride * y;
u32 *dst = (u32 *)(convBuf + rect.Pitch * y);
ConvertRGBA5551ToRGBA8888(dst, src, width);
for (int x = 0; x < width; x++) {
u16_le col0 = src[x+0];
ARGB8From5551(col0, &dst[x + 0]);
}
}
break;
case GE_FORMAT_4444:
{
const u16_le *src = (const u16_le *)srcPixels + srcStride * y;
u32 *dst = (u32 *)(convBuf + rect.Pitch * y);
ConvertRGBA4444ToRGBA8888(dst, src, width);
u8 *dst = (u8 *)(convBuf + rect.Pitch * y);
for (int x = 0; x < width; x++)
{
u16_le col = src[x];
dst[x * 4 + 0] = (col >> 12) << 4;
dst[x * 4 + 1] = ((col >> 8) & 0xf) << 4;
dst[x * 4 + 2] = ((col >> 4) & 0xf) << 4;
dst[x * 4 + 3] = (col & 0xf) << 4;
}
}
break;
@ -198,7 +248,10 @@ namespace DX9 {
{
const u32_le *src = (const u32_le *)srcPixels + srcStride * y;
u32 *dst = (u32 *)(convBuf + rect.Pitch * y);
ConvertBGRA8888ToRGBA8888(dst, src, width);
for (int x = 0; x < width; x++)
{
dst[x] = RGBA2BGRA(src[x]);
}
}
break;
}
@ -207,11 +260,15 @@ namespace DX9 {
for (int y = 0; y < height; y++) {
const u32_le *src = (const u32_le *)srcPixels + srcStride * y;
u32 *dst = (u32 *)(convBuf + rect.Pitch * y);
ConvertBGRA8888ToRGBA8888(dst, src, width);
for (int x = 0; x < width; x++)
{
dst[x] = RGBA2BGRA(src[x]);
}
}
}
drawPixelsTex_->UnlockRect(0);
// D3DXSaveTextureToFile("game:\\cc.png", D3DXIFF_PNG, drawPixelsTex_, NULL);
}
void FramebufferManagerDX9::DrawPixels(VirtualFramebuffer *vfb, int dstX, int dstY, const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height) {
@ -607,6 +664,7 @@ namespace DX9 {
}
void FramebufferManagerDX9::CopyDisplayToOutput() {
fbo_unbind();
dxstate.viewport.set(0, 0, PSP_CoreParameter().pixelWidth, PSP_CoreParameter().pixelHeight);
currentRenderVfb_ = 0;
@ -977,7 +1035,9 @@ namespace DX9 {
switch (format) {
case GE_FORMAT_565: // BGR 565
for (u32 y = 0; y < height; ++y) {
ConvertBGRA8888ToRGB565(dst16, src32, width);
for (u32 x = 0; x < width; ++x) {
dst16[x] = BGRA8888toRGB565(src32[x]);
}
src32 += srcStride;
dst16 += dstStride;
}
@ -991,7 +1051,9 @@ namespace DX9 {
break;
case GE_FORMAT_4444: // ABGR 4444
for (u32 y = 0; y < height; ++y) {
ConvertBGRA8888ToRGBA4444(dst16, src32, width);
for (u32 x = 0; x < width; ++x) {
dst16[x] = BGRA8888toRGBA4444(src32[x]);
}
src32 += srcStride;
dst16 += dstStride;
}

1
GPU/Directx9/StencilBufferDX9.cpp
View file

@ -19,7 +19,6 @@
#include "helper/dx_state.h"
#include "helper/fbo.h"
#include "Common/ColorConv.h"
#include "Core/Reporting.h"
#include "GPU/Directx9/FramebufferDX9.h"
#include "GPU/Directx9/PixelShaderGeneratorDX9.h"

21
GPU/Directx9/TextureCacheDX9.cpp
View file

@ -1703,25 +1703,8 @@ void TextureCacheDX9::LoadTextureLevel(TexCacheEntry &entry, int level, int maxL
gpuStats.numTexturesDecoded++;
u32 *pixelData = (u32 *)finalBuf;
if (scaleFactor > 1 && (entry.status & TexCacheEntry::STATUS_CHANGE_FREQUENT) == 0) {
GEBufferFormat dstFormat;
switch (dstFmt) {
case D3DFMT_A8R8G8B8: dstFormat = GE_FORMAT_8888; break;
case D3DFMT_R5G6B5: dstFormat = GE_FORMAT_565; break;
case D3DFMT_A4R4G4B4: dstFormat = GE_FORMAT_4444; break;
case D3DFMT_A1R5G5B5: dstFormat = GE_FORMAT_5551; break;
default: goto dontScale;
}
scaler.Scale(pixelData, dstFormat, w, h, scaleFactor);
switch (dstFormat) {
case GE_FORMAT_8888: dstFmt = D3DFMT_A8R8G8B8; break;
case GE_FORMAT_565: dstFmt = D3DFMT_R5G6B5; break;
case GE_FORMAT_4444: dstFmt = D3DFMT_A4R4G4B4; break;
case GE_FORMAT_5551: dstFmt = D3DFMT_A1R5G5B5; break;
}
dontScale:
;
}
if (scaleFactor > 1 && (entry.status & TexCacheEntry::STATUS_CHANGE_FREQUENT) == 0)
scaler.Scale(pixelData, dstFmt, w, h, scaleFactor);
if ((entry.status & TexCacheEntry::STATUS_CHANGE_FREQUENT) == 0) {
TexCacheEntry::Status alphaStatus = CheckAlpha(pixelData, dstFmt, w, w, h);

4
GPU/Directx9/TextureCacheDX9.h
View file

@ -24,7 +24,7 @@
#include "helper/fbo.h"
#include "GPU/GPUInterface.h"
#include "GPU/GPUState.h"
#include "GPU/Common/TextureScaler.h"
#include "GPU/Directx9/TextureScalerDX9.h"
#include "GPU/Common/TextureCacheCommon.h"
struct VirtualFramebuffer;
@ -197,7 +197,7 @@ private:
bool clearCacheNextFrame_;
bool lowMemoryMode_;
TextureScaler scaler;
TextureScalerDX9 scaler;
SimpleBuf<u32> tmpTexBuf32;
SimpleBuf<u16> tmpTexBuf16;

692
GPU/Directx9/TextureScalerDX9.cpp Normal file
View file

@ -0,0 +1,692 @@
// Copyright (c) 2012- 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/.
// On Visual Studio 2012, include this before anything else so
// _VARIADIC_MAX gets set to 10, to avoid std::bind compile errors.
// See header file for reasons why.
#if defined(_WIN32) && _MSC_VER == 1700
#include "../native/base/basictypes.h"
#endif
#include <algorithm>
#include "GPU/Directx9/TextureScalerDX9.h"
#include "Core/Config.h"
#include "Common/Common.h"
#include "Common/Log.h"
#include "Common/MsgHandler.h"
#include "Common/CommonFuncs.h"
#include "Common/ThreadPools.h"
#include "Common/CPUDetect.h"
#include "ext/xbrz/xbrz.h"
#include <stdlib.h>
#include <math.h>
#include <D3D9Types.h>
#undef min
#undef max
#if _M_SSE >= 0x402
#include <nmmintrin.h>
#endif
// Report the time and throughput for each larger scaling operation in the log
//#define SCALING_MEASURE_TIME
#ifdef SCALING_MEASURE_TIME
#include "native/base/timeutil.h"
#endif
/////////////////////////////////////// Helper Functions (mostly math for parallelization)
namespace {
//////////////////////////////////////////////////////////////////// Color space conversion
// convert 4444 image to 8888, parallelizable
void convert4444(u16* data, u32* out, int width, int l, int u) {
for(int y = l; y < u; ++y) {
for(int x = 0; x < width; ++x) {
u32 val = data[y*width + x];
u32 r = ((val>> 0) & 0xF) * 17;
u32 g = ((val>> 4) & 0xF) * 17;
u32 b = ((val>> 8) & 0xF) * 17;
u32 a = ((val>>12) & 0xF) * 17;
out[y*width + x] = (a << 24) | (b << 16) | (g << 8) | r;
}
}
}
// convert 565 image to 8888, parallelizable
void convert565(u16* data, u32* out, int width, int l, int u) {
for(int y = l; y < u; ++y) {
for(int x = 0; x < width; ++x) {
u32 val = data[y*width + x];
u32 r = Convert5To8((val ) & 0x1F);
u32 g = Convert6To8((val>> 5) & 0x3F);
u32 b = Convert5To8((val>>11) & 0x1F);
out[y*width + x] = (0xFF << 24) | (b << 16) | (g << 8) | r;
}
}
}
// convert 5551 image to 8888, parallelizable
void convert5551(u16* data, u32* out, int width, int l, int u) {
for(int y = l; y < u; ++y) {
for(int x = 0; x < width; ++x) {
u32 val = data[y*width + x];
u32 r = Convert5To8((val>> 0) & 0x1F);
u32 g = Convert5To8((val>> 5) & 0x1F);
u32 b = Convert5To8((val>>10) & 0x1F);
u32 a = ((val >> 15) & 0x1) * 255;
out[y*width + x] = (a << 24) | (b << 16) | (g << 8) | r;
}
}
}
//////////////////////////////////////////////////////////////////// Various image processing
#define R(_col) ((_col>> 0)&0xFF)
#define G(_col) ((_col>> 8)&0xFF)
#define B(_col) ((_col>>16)&0xFF)
#define A(_col) ((_col>>24)&0xFF)
#define DISTANCE(_p1,_p2) ( abs(static_cast<int>(static_cast<int>(R(_p1))-R(_p2))) + abs(static_cast<int>(static_cast<int>(G(_p1))-G(_p2))) \
+ abs(static_cast<int>(static_cast<int>(B(_p1))-B(_p2))) + abs(static_cast<int>(static_cast<int>(A(_p1))-A(_p2))) )
// this is sadly much faster than an inline function with a loop, at least in VC10
#define MIX_PIXELS(_p0, _p1, _factors) \
( (R(_p0)*(_factors)[0] + R(_p1)*(_factors)[1])/255 << 0 ) | \
( (G(_p0)*(_factors)[0] + G(_p1)*(_factors)[1])/255 << 8 ) | \
( (B(_p0)*(_factors)[0] + B(_p1)*(_factors)[1])/255 << 16 ) | \
( (A(_p0)*(_factors)[0] + A(_p1)*(_factors)[1])/255 << 24 )
#define BLOCK_SIZE 32
// 3x3 convolution with Neumann boundary conditions, parallelizable
// quite slow, could be sped up a lot
// especially handling of separable kernels
void convolve3x3(u32* data, u32* out, const int kernel[3][3], int width, int height, int l, int u) {
for(int yb = 0; yb < (u-l)/BLOCK_SIZE+1; ++yb) {
for(int xb = 0; xb < width/BLOCK_SIZE+1; ++xb) {
for(int y = l+yb*BLOCK_SIZE; y < l+(yb+1)*BLOCK_SIZE && y < u; ++y) {
for(int x = xb*BLOCK_SIZE; x < (xb+1)*BLOCK_SIZE && x < width; ++x) {
int val = 0;
for(int yoff = -1; yoff <= 1; ++yoff) {
int yy = std::max(std::min(y+yoff, height-1), 0);
for(int xoff = -1; xoff <= 1; ++xoff) {
int xx = std::max(std::min(x+xoff, width-1), 0);
val += data[yy*width + xx] * kernel[yoff+1][xoff+1];
}
}
out[y*width + x] = abs(val);
}
}
}
}
}
// deposterization: smoothes posterized gradients from low-color-depth (e.g. 444, 565, compressed) sources
void deposterizeH(u32* data, u32* out, int w, int l, int u) {
static const int T = 8;
for(int y = l; y < u; ++y) {
for(int x = 0; x < w; ++x) {
int inpos = y*w + x;
u32 center = data[inpos];
if(x==0 || x==w-1) {
out[y*w + x] = center;
continue;
}
u32 left = data[inpos - 1];
u32 right = data[inpos + 1];
out[y*w + x] = 0;
for(int c=0; c<4; ++c) {
u8 lc = (( left>>c*8)&0xFF);
u8 cc = ((center>>c*8)&0xFF);
u8 rc = (( right>>c*8)&0xFF);
if((lc != rc) && ((lc == cc && abs((int)((int)rc)-cc) <= T) || (rc == cc && abs((int)((int)lc)-cc) <= T))) {
// blend this component
out[y*w + x] |= ((rc+lc)/2) << (c*8);
} else {
// no change for this component
out[y*w + x] |= cc << (c*8);
}
}
}
}
}
void deposterizeV(u32* data, u32* out, int w, int h, int l, int u) {
static const int T = 8;
for(int xb = 0; xb < w/BLOCK_SIZE+1; ++xb) {
for(int y = l; y < u; ++y) {
for(int x = xb*BLOCK_SIZE; x < (xb+1)*BLOCK_SIZE && x < w; ++x) {
u32 center = data[ y * w + x];
if(y==0 || y==h-1) {
out[y*w + x] = center;
continue;
}
u32 upper = data[(y-1) * w + x];
u32 lower = data[(y+1) * w + x];
out[y*w + x] = 0;
for(int c=0; c<4; ++c) {
u8 uc = (( upper>>c*8)&0xFF);
u8 cc = ((center>>c*8)&0xFF);
u8 lc = (( lower>>c*8)&0xFF);
if((uc != lc) && ((uc == cc && abs((int)((int)lc)-cc) <= T) || (lc == cc && abs((int)((int)uc)-cc) <= T))) {
// blend this component
out[y*w + x] |= ((lc+uc)/2) << (c*8);
} else {
// no change for this component
out[y*w + x] |= cc << (c*8);
}
}
}
}
}
}
// generates a distance mask value for each pixel in data
// higher values -> larger distance to the surrounding pixels
void generateDistanceMask(u32* data, u32* out, int width, int height, int l, int u) {
for(int yb = 0; yb < (u-l)/BLOCK_SIZE+1; ++yb) {
for(int xb = 0; xb < width/BLOCK_SIZE+1; ++xb) {
for(int y = l+yb*BLOCK_SIZE; y < l+(yb+1)*BLOCK_SIZE && y < u; ++y) {
for(int x = xb*BLOCK_SIZE; x < (xb+1)*BLOCK_SIZE && x < width; ++x) {
out[y*width + x] = 0;
u32 center = data[y*width + x];
for(int yoff = -1; yoff <= 1; ++yoff) {
int yy = y+yoff;
if(yy == height || yy == -1) {
out[y*width + x] += 1200; // assume distance at borders, usually makes for better result
continue;
}
for(int xoff = -1; xoff <= 1; ++xoff) {
if(yoff == 0 && xoff == 0) continue;
int xx = x+xoff;
if(xx == width || xx == -1) {
out[y*width + x] += 400; // assume distance at borders, usually makes for better result
continue;
}
out[y*width + x] += DISTANCE(data[yy*width + xx], center);
}
}
}
}
}
}
}
// mix two images based on a mask
void mix(u32* data, u32* source, u32* mask, u32 maskmax, int width, int l, int u) {
for(int y = l; y < u; ++y) {
for(int x = 0; x < width; ++x) {
int pos = y*width + x;
u8 mixFactors[2] = { 0, static_cast<u8>((std::min(mask[pos], maskmax)*255)/maskmax) };
mixFactors[0] = 255-mixFactors[1];
data[pos] = MIX_PIXELS(data[pos], source[pos], mixFactors);
if(A(source[pos]) == 0) data[pos] = data[pos] & 0x00FFFFFF; // xBRZ always does a better job with hard alpha
}
}
}
//////////////////////////////////////////////////////////////////// Bicubic scaling
// generate the value of a Mitchell-Netravali scaling spline at distance d, with parameters A and B
// B=1 C=0 : cubic B spline (very smooth)
// B=C=1/3 : recommended for general upscaling
// B=0 C=1/2 : Catmull-Rom spline (sharp, ringing)
// see Mitchell & Netravali, "Reconstruction Filters in Computer Graphics"
inline float mitchell(float x, float B, float C) {
float ax = fabs(x);
if(ax>=2.0f) return 0.0f;
if(ax>=1.0f) return ((-B-6*C)*(x*x*x) + (6*B+30*C)*(x*x) + (-12*B-48*C)*x + (8*B+24*C))/6.0f;
return ((12-9*B-6*C)*(x*x*x) + (-18+12*B+6*C)*(x*x) + (6-2*B))/6.0f;
}
// arrays for pre-calculating weights and sums (~20KB)
// Dimensions:
// 0: 0 = BSpline, 1 = mitchell
// 2: 2-5x scaling
// 2,3: 5x5 generated pixels
// 4,5: 5x5 pixels sampled from
float bicubicWeights[2][4][5][5][5][5];
float bicubicInvSums[2][4][5][5];
// initialize pre-computed weights array
void initBicubicWeights() {
float B[2] = { 1.0f, 0.334f };
float C[2] = { 0.0f, 0.334f };
for(int type=0; type<2; ++type) {
for(int factor=2; factor<=5; ++factor) {
for(int x=0; x<factor; ++x) {
for(int y=0; y<factor; ++y) {
float sum = 0.0f;
for(int sx = -2; sx <= 2; ++sx) {
for(int sy = -2; sy <= 2; ++sy) {
float dx = (x+0.5f)/factor - (sx+0.5f);
float dy = (y+0.5f)/factor - (sy+0.5f);
float dist = sqrt(dx*dx + dy*dy);
float weight = mitchell(dist, B[type], C[type]);
bicubicWeights[type][factor-2][x][y][sx+2][sy+2] = weight;
sum += weight;
}
}
bicubicInvSums[type][factor-2][x][y] = 1.0f/sum;
}
}
}
}
}
// perform bicubic scaling by factor f, with precomputed spline type T
template<int f, int T>
void scaleBicubicT(u32* data, u32* out, int w, int h, int l, int u) {
int outw = w*f;
for(int yb = 0; yb < (u-l)*f/BLOCK_SIZE+1; ++yb) {
for(int xb = 0; xb < w*f/BLOCK_SIZE+1; ++xb) {
for(int y = l*f+yb*BLOCK_SIZE; y < l*f+(yb+1)*BLOCK_SIZE && y < u*f; ++y) {
for(int x = xb*BLOCK_SIZE; x < (xb+1)*BLOCK_SIZE && x < w*f; ++x) {
float r = 0.0f, g = 0.0f, b = 0.0f, a = 0.0f;
int cx = x/f, cy = y/f;
// sample supporting pixels in original image
for(int sx = -2; sx <= 2; ++sx) {
for(int sy = -2; sy <= 2; ++sy) {
float weight = bicubicWeights[T][f-2][x%f][y%f][sx+2][sy+2];
if(weight != 0.0f) {
// clamp pixel locations
int csy = std::max(std::min(sy+cy,h-1),0);
int csx = std::max(std::min(sx+cx,w-1),0);
// sample & add weighted components
u32 sample = data[csy*w+csx];
r += weight*R(sample);
g += weight*G(sample);
b += weight*B(sample);
a += weight*A(sample);
}
}
}
// generate and write result
float invSum = bicubicInvSums[T][f-2][x%f][y%f];
int ri = std::min(std::max(static_cast<int>(ceilf(r*invSum)),0),255);
int gi = std::min(std::max(static_cast<int>(ceilf(g*invSum)),0),255);
int bi = std::min(std::max(static_cast<int>(ceilf(b*invSum)),0),255);
int ai = std::min(std::max(static_cast<int>(ceilf(a*invSum)),0),255);
out[y*outw + x] = (ai << 24) | (bi << 16) | (gi << 8) | ri;
}
}
}
}
}
#if _M_SSE >= 0x401
template<int f, int T>
void scaleBicubicTSSE41(u32* data, u32* out, int w, int h, int l, int u) {
int outw = w*f;
for(int yb = 0; yb < (u-l)*f/BLOCK_SIZE+1; ++yb) {
for(int xb = 0; xb < w*f/BLOCK_SIZE+1; ++xb) {
for(int y = l*f+yb*BLOCK_SIZE; y < l*f+(yb+1)*BLOCK_SIZE && y < u*f; ++y) {
for(int x = xb*BLOCK_SIZE; x < (xb+1)*BLOCK_SIZE && x < w*f; ++x) {
__m128 result = _mm_set1_ps(0.0f);
int cx = x/f, cy = y/f;
// sample supporting pixels in original image
for(int sx = -2; sx <= 2; ++sx) {
for(int sy = -2; sy <= 2; ++sy) {
float weight = bicubicWeights[T][f-2][x%f][y%f][sx+2][sy+2];
if(weight != 0.0f) {
// clamp pixel locations
int csy = std::max(std::min(sy+cy,h-1),0);
int csx = std::max(std::min(sx+cx,w-1),0);
// sample & add weighted components
__m128i sample = _mm_cvtsi32_si128(data[csy*w+csx]);
sample = _mm_cvtepu8_epi32(sample);
__m128 col = _mm_cvtepi32_ps(sample);
col = _mm_mul_ps(col, _mm_set1_ps(weight));
result = _mm_add_ps(result, col);
}
}
}
// generate and write result
__m128i pixel = _mm_cvtps_epi32(_mm_mul_ps(result, _mm_set1_ps(bicubicInvSums[T][f-2][x%f][y%f])));
pixel = _mm_packs_epi32(pixel, pixel);
pixel = _mm_packus_epi16(pixel, pixel);
out[y*outw + x] = _mm_cvtsi128_si32(pixel);
}
}
}
}
}
#endif
void scaleBicubicBSpline(int factor, u32* data, u32* out, int w, int h, int l, int u) {
#if _M_SSE >= 0x401
if(cpu_info.bSSE4_1) {
switch(factor) {
case 2: scaleBicubicTSSE41<2, 0>(data, out, w, h, l, u); break; // when I first tested this,
case 3: scaleBicubicTSSE41<3, 0>(data, out, w, h, l, u); break; // it was even slower than I had expected
case 4: scaleBicubicTSSE41<4, 0>(data, out, w, h, l, u); break; // turns out I had not included
case 5: scaleBicubicTSSE41<5, 0>(data, out, w, h, l, u); break; // any of these break statements
default: ERROR_LOG(G3D, "Bicubic upsampling only implemented for factors 2 to 5");
}
} else {
#endif
switch(factor) {
case 2: scaleBicubicT<2, 0>(data, out, w, h, l, u); break; // when I first tested this,
case 3: scaleBicubicT<3, 0>(data, out, w, h, l, u); break; // it was even slower than I had expected
case 4: scaleBicubicT<4, 0>(data, out, w, h, l, u); break; // turns out I had not included
case 5: scaleBicubicT<5, 0>(data, out, w, h, l, u); break; // any of these break statements
default: ERROR_LOG(G3D, "Bicubic upsampling only implemented for factors 2 to 5");
}
#if _M_SSE >= 0x401
}
#endif
}
void scaleBicubicMitchell(int factor, u32* data, u32* out, int w, int h, int l, int u) {
#if _M_SSE >= 0x401
if(cpu_info.bSSE4_1) {
switch(factor) {
case 2: scaleBicubicTSSE41<2, 1>(data, out, w, h, l, u); break;
case 3: scaleBicubicTSSE41<3, 1>(data, out, w, h, l, u); break;
case 4: scaleBicubicTSSE41<4, 1>(data, out, w, h, l, u); break;
case 5: scaleBicubicTSSE41<5, 1>(data, out, w, h, l, u); break;
default: ERROR_LOG(G3D, "Bicubic upsampling only implemented for factors 2 to 5");
}
} else {
#endif
switch(factor) {
case 2: scaleBicubicT<2, 1>(data, out, w, h, l, u); break;
case 3: scaleBicubicT<3, 1>(data, out, w, h, l, u); break;
case 4: scaleBicubicT<4, 1>(data, out, w, h, l, u); break;
case 5: scaleBicubicT<5, 1>(data, out, w, h, l, u); break;
default: ERROR_LOG(G3D, "Bicubic upsampling only implemented for factors 2 to 5");
}
#if _M_SSE >= 0x401
}
#endif
}
//////////////////////////////////////////////////////////////////// Bilinear scaling
const static u8 BILINEAR_FACTORS[4][3][2] = {
{ { 44,211}, { 0, 0}, { 0, 0} }, // x2
{ { 64,191}, { 0,255}, { 0, 0} }, // x3
{ { 77,178}, { 26,229}, { 0, 0} }, // x4
{ {102,153}, { 51,204}, { 0,255} }, // x5
};
// integral bilinear upscaling by factor f, horizontal part
template<int f>
void bilinearHt(u32* data, u32* out, int w, int l, int u) {
static_assert(f>1 && f<=5, "Bilinear scaling only implemented for factors 2 to 5");
int outw = w*f;
for(int y = l; y < u; ++y) {
for(int x = 0; x < w; ++x) {
int inpos = y*w + x;
u32 left = data[inpos - (x==0 ?0:1)];
u32 center = data[inpos];
u32 right = data[inpos + (x==w-1?0:1)];
int i=0;
for(; i<f/2+f%2; ++i) { // first half of the new pixels + center, hope the compiler unrolls this
out[y*outw + x*f + i] = MIX_PIXELS(left, center, BILINEAR_FACTORS[f-2][i]);
}
for(; i<f ; ++i) { // second half of the new pixels, hope the compiler unrolls this
out[y*outw + x*f + i] = MIX_PIXELS(right, center, BILINEAR_FACTORS[f-2][f-1-i]);
}
}
}
}
void bilinearH(int factor, u32* data, u32* out, int w, int l, int u) {
switch(factor) {
case 2: bilinearHt<2>(data, out, w, l, u); break;
case 3: bilinearHt<3>(data, out, w, l, u); break;
case 4: bilinearHt<4>(data, out, w, l, u); break;
case 5: bilinearHt<5>(data, out, w, l, u); break;
default: ERROR_LOG(G3D, "Bilinear upsampling only implemented for factors 2 to 5");
}
}
// integral bilinear upscaling by factor f, vertical part
// gl/gu == global lower and upper bound
template<int f>
void bilinearVt(u32* data, u32* out, int w, int gl, int gu, int l, int u) {
static_assert(f>1 && f<=5, "Bilinear scaling only implemented for 2x, 3x, 4x, and 5x");
int outw = w*f;
for(int xb = 0; xb < outw/BLOCK_SIZE+1; ++xb) {
for(int y = l; y < u; ++y) {
u32 uy = y - (y==gl ?0:1);
u32 ly = y + (y==gu-1?0:1);
for(int x = xb*BLOCK_SIZE; x < (xb+1)*BLOCK_SIZE && x < outw; ++x) {
u32 upper = data[uy * outw + x];
u32 center = data[y * outw + x];
u32 lower = data[ly * outw + x];
int i=0;
for(; i<f/2+f%2; ++i) { // first half of the new pixels + center, hope the compiler unrolls this
out[(y*f + i)*outw + x] = MIX_PIXELS(upper, center, BILINEAR_FACTORS[f-2][i]);
}
for(; i<f ; ++i) { // second half of the new pixels, hope the compiler unrolls this
out[(y*f + i)*outw + x] = MIX_PIXELS(lower, center, BILINEAR_FACTORS[f-2][f-1-i]);
}
}
}
}
}
void bilinearV(int factor, u32* data, u32* out, int w, int gl, int gu, int l, int u) {
switch(factor) {
case 2: bilinearVt<2>(data, out, w, gl, gu, l, u); break;
case 3: bilinearVt<3>(data, out, w, gl, gu, l, u); break;
case 4: bilinearVt<4>(data, out, w, gl, gu, l, u); break;
case 5: bilinearVt<5>(data, out, w, gl, gu, l, u); break;
default: ERROR_LOG(G3D, "Bilinear upsampling only implemented for factors 2 to 5");
}
}
#undef BLOCK_SIZE
#undef MIX_PIXELS
#undef DISTANCE
#undef R
#undef G
#undef B
#undef A
// used for debugging texture scaling (writing textures to files)
static int g_imgCount = 0;
void dbgPPM(int w, int h, u8* pixels, const char* prefix = "dbg") { // 3 component RGB
char fn[32];
snprintf(fn, 32, "%s%04d.ppm", prefix, g_imgCount++);
FILE *fp = fopen(fn, "wb");
fprintf(fp, "P6\n%d %d\n255\n", w, h);
for(int j = 0; j < h; ++j) {
for(int i = 0; i < w; ++i) {
static unsigned char color[3];
color[0] = pixels[(j*w+i)*4+0]; /* red */
color[1] = pixels[(j*w+i)*4+1]; /* green */
color[2] = pixels[(j*w+i)*4+2]; /* blue */
fwrite(color, 1, 3, fp);
}
}
fclose(fp);
}
void dbgPGM(int w, int h, u32* pixels, const char* prefix = "dbg") { // 1 component
char fn[32];
snprintf(fn, 32, "%s%04d.pgm", prefix, g_imgCount++);
FILE *fp = fopen(fn, "wb");
fprintf(fp, "P5\n%d %d\n65536\n", w, h);
for(int j = 0; j < h; ++j) {
for(int i = 0; i < w; ++i) {
fwrite((pixels+(j*w+i)), 1, 2, fp);
}
}
fclose(fp);
}
}
/////////////////////////////////////// Texture Scaler
namespace DX9 {
TextureScalerDX9::TextureScalerDX9() {
initBicubicWeights();
}
bool TextureScalerDX9::IsEmptyOrFlat(u32* data, int pixels, u32 fmt) {
int pixelsPerWord = (fmt == D3DFMT_A8R8G8B8) ? 1 : 2;
u32 ref = data[0];
for(int i=0; i<pixels/pixelsPerWord; ++i) {
if(data[i]!=ref) return false;
}
return true;
}
void TextureScalerDX9::Scale(u32* &data, u32 &dstFmt, int &width, int &height, int factor) {
// prevent processing empty or flat textures (this happens a lot in some games)
// doesn't hurt the standard case, will be very quick for textures with actual texture
if(IsEmptyOrFlat(data, width*height, dstFmt)) {
INFO_LOG(G3D, "TextureScaler: early exit -- empty/flat texture");
return;
}
#ifdef SCALING_MEASURE_TIME
double t_start = real_time_now();
#endif
bufInput.resize(width*height); // used to store the input image image if it needs to be reformatted
bufOutput.resize(width*height*factor*factor); // used to store the upscaled image
u32 *inputBuf = bufInput.data();
u32 *outputBuf = bufOutput.data();
// convert texture to correct format for scaling
ConvertTo8888(dstFmt, data, inputBuf, width, height);
// deposterize
if(g_Config.bTexDeposterize) {
bufDeposter.resize(width*height);
DePosterize(inputBuf, bufDeposter.data(), width, height);
inputBuf = bufDeposter.data();
}
// scale
switch(g_Config.iTexScalingType) {
case XBRZ:
ScaleXBRZ(factor, inputBuf, outputBuf, width, height);
break;
case HYBRID:
ScaleHybrid(factor, inputBuf, outputBuf, width, height);
break;
case BICUBIC:
ScaleBicubicMitchell(factor, inputBuf, outputBuf, width, height);
break;
case HYBRID_BICUBIC:
ScaleHybrid(factor, inputBuf, outputBuf, width, height, true);
break;
default:
ERROR_LOG(G3D, "Unknown scaling type: %d", g_Config.iTexScalingType);
}
// update values accordingly
data = outputBuf;
dstFmt = D3DFMT_A8R8G8B8;
width *= factor;
height *= factor;
#ifdef SCALING_MEASURE_TIME
if(width*height > 64*64*factor*factor) {
double t = real_time_now() - t_start;
NOTICE_LOG(MASTER_LOG, "TextureScaler: processed %9d pixels in %6.5lf seconds. (%9.2lf Mpixels/second)",
width*height, t, (width*height)/(t*1000*1000));
}
#endif
}
void TextureScalerDX9::ScaleXBRZ(int factor, u32* source, u32* dest, int width, int height) {
xbrz::ScalerCfg cfg;
GlobalThreadPool::Loop(std::bind(&xbrz::scale, factor, source, dest, width, height, xbrz::ColorFormat::ARGB, cfg, placeholder::_1, placeholder::_2), 0, height);
}
void TextureScalerDX9::ScaleBilinear(int factor, u32* source, u32* dest, int width, int height) {
bufTmp1.resize(width*height*factor);
u32 *tmpBuf = bufTmp1.data();
GlobalThreadPool::Loop(std::bind(&bilinearH, factor, source, tmpBuf, width, placeholder::_1, placeholder::_2), 0, height);
GlobalThreadPool::Loop(std::bind(&bilinearV, factor, tmpBuf, dest, width, 0, height, placeholder::_1, placeholder::_2), 0, height);
}
void TextureScalerDX9::ScaleBicubicBSpline(int factor, u32* source, u32* dest, int width, int height) {
GlobalThreadPool::Loop(std::bind(&scaleBicubicBSpline, factor, source, dest, width, height, placeholder::_1, placeholder::_2), 0, height);
}
void TextureScalerDX9::ScaleBicubicMitchell(int factor, u32* source, u32* dest, int width, int height) {
GlobalThreadPool::Loop(std::bind(&scaleBicubicMitchell, factor, source, dest, width, height, placeholder::_1, placeholder::_2), 0, height);
}
void TextureScalerDX9::ScaleHybrid(int factor, u32* source, u32* dest, int width, int height, bool bicubic) {
// Basic algorithm:
// 1) determine a feature mask C based on a sobel-ish filter + splatting, and upscale that mask bilinearly
// 2) generate 2 scaled images: A - using Bilinear filtering, B - using xBRZ
// 3) output = A*C + B*(1-C)
const static int KERNEL_SPLAT[3][3] = {
{ 1, 1, 1 }, { 1, 1, 1 }, { 1, 1, 1 }
};
bufTmp1.resize(width*height);
bufTmp2.resize(width*height*factor*factor);
bufTmp3.resize(width*height*factor*factor);
GlobalThreadPool::Loop(std::bind(&generateDistanceMask, source, bufTmp1.data(), width, height, placeholder::_1, placeholder::_2), 0, height);
GlobalThreadPool::Loop(std::bind(&convolve3x3, bufTmp1.data(), bufTmp2.data(), KERNEL_SPLAT, width, height, placeholder::_1, placeholder::_2), 0, height);
ScaleBilinear(factor, bufTmp2.data(), bufTmp3.data(), width, height);
// mask C is now in bufTmp3
ScaleXBRZ(factor, source, bufTmp2.data(), width, height);
// xBRZ upscaled source is in bufTmp2
if(bicubic) ScaleBicubicBSpline(factor, source, dest, width, height);
else ScaleBilinear(factor, source, dest, width, height);
// Upscaled source is in dest
// Now we can mix it all together
// The factor 8192 was found through practical testing on a variety of textures
GlobalThreadPool::Loop(std::bind(&mix, dest, bufTmp2.data(), bufTmp3.data(), 8192, width*factor, placeholder::_1, placeholder::_2), 0, height*factor);
}
void TextureScalerDX9::DePosterize(u32* source, u32* dest, int width, int height) {
bufTmp3.resize(width*height);
GlobalThreadPool::Loop(std::bind(&deposterizeH, source, bufTmp3.data(), width, placeholder::_1, placeholder::_2), 0, height);
GlobalThreadPool::Loop(std::bind(&deposterizeV, bufTmp3.data(), dest, width, height, placeholder::_1, placeholder::_2), 0, height);
GlobalThreadPool::Loop(std::bind(&deposterizeH, dest, bufTmp3.data(), width, placeholder::_1, placeholder::_2), 0, height);
GlobalThreadPool::Loop(std::bind(&deposterizeV, bufTmp3.data(), dest, width, height, placeholder::_1, placeholder::_2), 0, height);
}
void TextureScalerDX9::ConvertTo8888(u32 format, u32* source, u32* &dest, int width, int height) {
switch(format) {
case D3DFMT_A8R8G8B8:
dest = source; // already fine
break;
case D3DFMT_A4R4G4B4:
GlobalThreadPool::Loop(std::bind(&convert4444, (u16*)source, dest, width, placeholder::_1, placeholder::_2), 0, height);
break;
case D3DFMT_R5G6B5:
GlobalThreadPool::Loop(std::bind(&convert565, (u16*)source, dest, width, placeholder::_1, placeholder::_2), 0, height);
break;
case D3DFMT_A1R5G5B5:
GlobalThreadPool::Loop(std::bind(&convert5551, (u16*)source, dest, width, placeholder::_1, placeholder::_2), 0, height);
break;
default:
dest = source;
ERROR_LOG(G3D, "iXBRZTexScaling: unsupported texture format");
}
}
};

56
GPU/Directx9/TextureScalerDX9.h Normal file
View file

@ -0,0 +1,56 @@
// Copyright (c) 2012- 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 "Common/MemoryUtil.h"
#include "../Globals.h"
#include "helper/global.h"
//#include "gfx/gl_common.h"
#include <vector>
namespace DX9 {
class TextureScalerDX9 {
public:
TextureScalerDX9();
void Scale(u32* &data, u32 &dstfmt, int &width, int &height, int factor);
enum { XBRZ= 0, HYBRID = 1, BICUBIC = 2, HYBRID_BICUBIC = 3 };
private:
void ScaleXBRZ(int factor, u32* source, u32* dest, int width, int height);
void ScaleBilinear(int factor, u32* source, u32* dest, int width, int height);
void ScaleBicubicBSpline(int factor, u32* source, u32* dest, int width, int height);
void ScaleBicubicMitchell(int factor, u32* source, u32* dest, int width, int height);
void ScaleHybrid(int factor, u32* source, u32* dest, int width, int height, bool bicubic = false);
void ConvertTo8888(u32 format, u32* source, u32* &dest, int width, int height);
void DePosterize(u32* source, u32* dest, int width, int height);
bool IsEmptyOrFlat(u32* data, int pixels, u32 fmt);
// depending on the factor and texture sizes, these can get pretty large
// maximum is (100 MB total for a 512 by 512 texture with scaling factor 5 and hybrid scaling)
// of course, scaling factor 5 is totally silly anyway
SimpleBuf<u32> bufInput, bufDeposter, bufOutput, bufTmp1, bufTmp2, bufTmp3;
};
};

61
GPU/GLES/Framebuffer.cpp
View file

@ -31,7 +31,6 @@
#include "Core/System.h"
#include "Core/Reporting.h"
#include "Core/HLE/sceDisplay.h"
#include "Common/ColorConv.h"
#include "GPU/ge_constants.h"
#include "GPU/GPUState.h"
@ -97,6 +96,22 @@ static const char color_vs[] =
" gl_Position = a_position;\n"
"}\n";
inline u16 RGBA8888toRGB565(u32 px) {
return ((px >> 3) & 0x001F) | ((px >> 5) & 0x07E0) | ((px >> 8) & 0xF800);
}
inline u16 RGBA8888toRGBA4444(u32 px) {
return ((px >> 4) & 0x000F) | ((px >> 8) & 0x00F0) | ((px >> 12) & 0x0F00) | ((px >> 16) & 0xF000);
}
inline u16 BGRA8888toRGB565(u32 px) {
return ((px >> 19) & 0x001F) | ((px >> 5) & 0x07E0) | ((px << 8) & 0xF800);
}
inline u16 BGRA8888toRGBA4444(u32 px) {
return ((px >> 20) & 0x000F) | ((px >> 8) & 0x00F0) | ((px << 4) & 0x0F00) | ((px >> 16) & 0xF000);
}
void ConvertFromRGBA8888(u8 *dst, const u8 *src, u32 dstStride, u32 srcStride, u32 width, u32 height, GEBufferFormat format);
void CenterRect(float *x, float *y, float *w, float *h,
@ -406,7 +421,14 @@ void FramebufferManager::MakePixelTexture(const u8 *srcPixels, GEBufferFormat sr
{
const u16 *src = (const u16 *)srcPixels + srcStride * y;
u8 *dst = convBuf_ + 4 * width * y;
ConvertRGB565ToRGBA888F((u32 *)dst, src, width);
for (int x = 0; x < width; x++)
{
u16 col = src[x];
dst[x * 4] = Convert5To8((col) & 0x1f);
dst[x * 4 + 1] = Convert6To8((col >> 5) & 0x3f);
dst[x * 4 + 2] = Convert5To8((col >> 11) & 0x1f);
dst[x * 4 + 3] = 255;
}
}
break;
@ -414,7 +436,14 @@ void FramebufferManager::MakePixelTexture(const u8 *srcPixels, GEBufferFormat sr
{
const u16 *src = (const u16 *)srcPixels + srcStride * y;
u8 *dst = convBuf_ + 4 * width * y;
ConvertRGBA5551ToRGBA8888((u32 *)dst, src, width);
for (int x = 0; x < width; x++)
{
u16 col = src[x];
dst[x * 4] = Convert5To8((col) & 0x1f);
dst[x * 4 + 1] = Convert5To8((col >> 5) & 0x1f);
dst[x * 4 + 2] = Convert5To8((col >> 10) & 0x1f);
dst[x * 4 + 3] = (col >> 15) ? 255 : 0;
}
}
break;
@ -422,7 +451,14 @@ void FramebufferManager::MakePixelTexture(const u8 *srcPixels, GEBufferFormat sr
{
const u16 *src = (const u16 *)srcPixels + srcStride * y;
u8 *dst = convBuf_ + 4 * width * y;
ConvertRGBA4444ToRGBA8888((u32 *)dst, src, width);
for (int x = 0; x < width; x++)
{
u16 col = src[x];
dst[x * 4] = Convert4To8((col >> 8) & 0xf);
dst[x * 4 + 1] = Convert4To8((col >> 4) & 0xf);
dst[x * 4 + 2] = Convert4To8(col & 0xf);
dst[x * 4 + 3] = Convert4To8(col >> 12);
}
}
break;
@ -1325,6 +1361,7 @@ void FramebufferManager::BlitFramebuffer(VirtualFramebuffer *dst, int dstX, int
void ConvertFromRGBA8888(u8 *dst, const u8 *src, u32 dstStride, u32 srcStride, u32 width, u32 height, GEBufferFormat format) {
// Must skip stride in the cases below. Some games pack data into the cracks, like MotoGP.
const u32 *src32 = (const u32 *)src;
if (format == GE_FORMAT_8888) {
u32 *dst32 = (u32 *)dst;
if (src == dst) {
@ -1350,13 +1387,17 @@ void ConvertFromRGBA8888(u8 *dst, const u8 *src, u32 dstStride, u32 srcStride, u
case GE_FORMAT_565: // BGR 565
if (UseBGRA8888()) {
for (u32 y = 0; y < height; ++y) {
ConvertBGRA8888ToRGB565(dst16, src32, width);
for (u32 x = 0; x < width; ++x) {
dst16[x] = BGRA8888toRGB565(src32[x]);
}
src32 += srcStride;
dst16 += dstStride;
}
} else {
for (u32 y = 0; y < height; ++y) {
ConvertRGBA8888ToRGB565(dst16, src32, width);
for (u32 x = 0; x < width; ++x) {
dst16[x] = RGBA8888toRGB565(src32[x]);
}
src32 += srcStride;
dst16 += dstStride;
}
@ -1380,13 +1421,17 @@ void ConvertFromRGBA8888(u8 *dst, const u8 *src, u32 dstStride, u32 srcStride, u
case GE_FORMAT_4444: // ABGR 4444
if (UseBGRA8888()) {
for (u32 y = 0; y < height; ++y) {
ConvertBGRA8888ToRGBA4444(dst16, src32, width);
for (u32 x = 0; x < width; ++x) {
dst16[x] = BGRA8888toRGBA4444(src32[x]);
}
src32 += srcStride;
dst16 += dstStride;
}
} else {
for (u32 y = 0; y < height; ++y) {
ConvertRGBA8888ToRGBA4444(dst16, src32, width);
for (u32 x = 0; x < width; ++x) {
dst16[x] = RGBA8888toRGBA4444(src32[x]);
}
src32 += srcStride;
dst16 += dstStride;
}

1
GPU/GLES/StencilBuffer.cpp
View file

@ -17,7 +17,6 @@
#include "gfx_es2/glsl_program.h"
#include "gfx_es2/gl_state.h"
#include "Common/ColorConv.h"
#include "Core/Reporting.h"
#include "GPU/GLES/Framebuffer.h"
#include "GPU/GLES/ShaderManager.h"

69
GPU/GLES/TextureCache.cpp
View file

@ -18,7 +18,6 @@
#include <algorithm>
#include <cstring>
#include "Common/ColorConv.h"
#include "Core/Host.h"
#include "Core/MemMap.h"
#include "Core/Reporting.h"
@ -2002,22 +2001,8 @@ void TextureCache::LoadTextureLevel(TexCacheEntry &entry, int level, bool replac
bool useBGRA = UseBGRA8888() && dstFmt == GL_UNSIGNED_BYTE;
u32 *pixelData = (u32 *)finalBuf;
if (scaleFactor > 1 && (entry.status & TexCacheEntry::STATUS_CHANGE_FREQUENT) == 0) {
GEBufferFormat dstFormat;
switch (dstFmt) {
case GL_UNSIGNED_BYTE: dstFormat = GE_FORMAT_8888; break;
case GL_UNSIGNED_SHORT_4_4_4_4: dstFormat = GE_FORMAT_4444; break;
case GL_UNSIGNED_SHORT_5_6_5: dstFormat = GE_FORMAT_565; break;
case GL_UNSIGNED_SHORT_5_5_5_1: dstFormat = GE_FORMAT_5551; break;
}
scaler.Scale(pixelData, dstFormat, w, h, scaleFactor);
switch (dstFormat) {
case GE_FORMAT_565: dstFmt = GL_UNSIGNED_SHORT_5_6_5; break;
case GE_FORMAT_5551: dstFmt = GL_UNSIGNED_SHORT_5_5_5_1; break;
case GE_FORMAT_4444: dstFmt = GL_UNSIGNED_SHORT_4_4_4_4; break;
case GE_FORMAT_8888: dstFmt = GL_UNSIGNED_BYTE; break;
}
}
if (scaleFactor > 1 && (entry.status & TexCacheEntry::STATUS_CHANGE_FREQUENT) == 0)
scaler.Scale(pixelData, dstFmt, w, h, scaleFactor);
if ((entry.status & TexCacheEntry::STATUS_CHANGE_FREQUENT) == 0) {
TexCacheEntry::Status alphaStatus = CheckAlpha(pixelData, dstFmt, useUnpack ? bufw : w, w, h);
@ -2080,33 +2065,47 @@ bool TextureCache::DecodeTexture(u8* output, const GPUgstate &state) {
switch (dstFmt) {
case GL_UNSIGNED_SHORT_4_4_4_4:
for (int y = 0; y < h; y++) {
u16 *src = (u16*)finalBuf + y*bufw;
u32 *dst = (u32*)output + y*w;
ConvertRGBA4444ToRGBA8888(dst, src, bufw);
}
for (int y = 0; y < h; y++)
for (int x = 0; x < bufw; x++) {
u32 val = ((u16*)finalBuf)[y*bufw + x];
u32 r = ((val>>12) & 0xF) * 17;
u32 g = ((val>> 8) & 0xF) * 17;
u32 b = ((val>> 4) & 0xF) * 17;
u32 a = ((val>> 0) & 0xF) * 17;
((u32*)output)[y*w + x] = (a << 24) | (r << 16) | (g << 8) | b;
}
break;
case GL_UNSIGNED_SHORT_5_5_5_1:
for (int y = 0; y < h; y++) {
u16 *src = (u16*)finalBuf + y*bufw;
u32 *dst = (u32*)output + y*w;
ConvertRGBA5551ToRGBA8888(dst, src, bufw);
}
for (int y = 0; y < h; y++)
for (int x = 0; x < bufw; x++) {
u32 val = ((u16*)finalBuf)[y*bufw + x];
u32 r = Convert5To8((val>>11) & 0x1F);
u32 g = Convert5To8((val>> 6) & 0x1F);
u32 b = Convert5To8((val>> 1) & 0x1F);
u32 a = (val & 0x1) * 255;
((u32*)output)[y*w + x] = (a << 24) | (r << 16) | (g << 8) | b;
}
break;
case GL_UNSIGNED_SHORT_5_6_5:
for (int y = 0; y < h; y++) {
u16 *src = (u16*)finalBuf + y*bufw;
u32 *dst = (u32*)output + y*w;
ConvertRGB565ToRGBA888F(dst, src, bufw);
}
for (int y = 0; y < h; y++)
for (int x = 0; x < bufw; x++) {
u32 val = ((u16*)finalBuf)[y*bufw + x];
u32 a = 0xFF;
u32 r = Convert5To8((val>>11) & 0x1F);
u32 g = Convert6To8((val>> 5) & 0x3F);
u32 b = Convert5To8((val ) & 0x1F);
((u32*)output)[y*w + x] = (a << 24) | (r << 16) | (g << 8) | b;
}
break;
default:
for (int y = 0; y < h; y++) {
ConvertBGRA8888ToRGBA8888((u32 *)output + y * w, (u32 *)(finalBuf)+y * bufw, bufw);
}
for (int y = 0; y < h; y++)
for (int x = 0; x < bufw; x++) {
u32 val = ((u32*)finalBuf)[y*bufw + x];
((u32*)output)[y*w + x] = ((val & 0xFF000000)) | ((val & 0x00FF0000)>>16) | ((val & 0x0000FF00)) | ((val & 0x000000FF)<<16);
}
break;
}

2
GPU/GLES/TextureCache.h
View file

@ -25,7 +25,7 @@
#include "Globals.h"
#include "GPU/GPUInterface.h"
#include "GPU/GPUState.h"
#include "GPU/Common/TextureScaler.h"
#include "GPU/GLES/TextureScaler.h"
#include "GPU/Common/TextureCacheCommon.h"
struct VirtualFramebuffer;

89
GPU/Common/TextureScaler.cpp → GPU/GLES/TextureScaler.cpp
View file

@ -21,20 +21,18 @@
#endif
#include <algorithm>
#include <cstdlib>
#include <cmath>
#include "GPU/Common/TextureScaler.h"
#include "GPU/GLES/TextureScaler.h"
#include "Core/Config.h"
#include "Common/Common.h"
#include "Common/ColorConv.h"
#include "Common/Log.h"
#include "Common/MsgHandler.h"
#include "Common/CommonFuncs.h"
#include "Common/ThreadPools.h"
#include "Common/CPUDetect.h"
#include "ext/xbrz/xbrz.h"
#include <stdlib.h>
#include <math.h>
#if _M_SSE >= 0x402
#include <nmmintrin.h>
@ -47,11 +45,53 @@
#include "native/base/timeutil.h"
#endif
// Helper Functions (mostly math for parallelization)
/////////////////////////////////////// Helper Functions (mostly math for parallelization)
namespace {
//////////////////////////////////////////////////////////////////// Color space conversion
// Various image processing
// convert 4444 image to 8888, parallelizable
void convert4444(u16* data, u32* out, int width, int l, int u) {
for(int y = l; y < u; ++y) {
for(int x = 0; x < width; ++x) {
u32 val = data[y*width + x];
u32 r = ((val>>12) & 0xF) * 17;
u32 g = ((val>> 8) & 0xF) * 17;
u32 b = ((val>> 4) & 0xF) * 17;
u32 a = ((val>> 0) & 0xF) * 17;
out[y*width + x] = (a << 24) | (b << 16) | (g << 8) | r;
}
}
}
// convert 565 image to 8888, parallelizable
void convert565(u16* data, u32* out, int width, int l, int u) {
for(int y = l; y < u; ++y) {
for(int x = 0; x < width; ++x) {
u32 val = data[y*width + x];
u32 r = Convert5To8((val>>11) & 0x1F);
u32 g = Convert6To8((val>> 5) & 0x3F);
u32 b = Convert5To8((val ) & 0x1F);
out[y*width + x] = (0xFF << 24) | (b << 16) | (g << 8) | r;
}
}
}
// convert 5551 image to 8888, parallelizable
void convert5551(u16* data, u32* out, int width, int l, int u) {
for(int y = l; y < u; ++y) {
for(int x = 0; x < width; ++x) {
u32 val = data[y*width + x];
u32 r = Convert5To8((val>>11) & 0x1F);
u32 g = Convert5To8((val>> 6) & 0x1F);
u32 b = Convert5To8((val>> 1) & 0x1F);
u32 a = (val & 0x1) * 255;
out[y*width + x] = (a << 24) | (b << 16) | (g << 8) | r;
}
}
}
//////////////////////////////////////////////////////////////////// Various image processing
#define R(_col) ((_col>> 0)&0xFF)
#define G(_col) ((_col>> 8)&0xFF)
@ -174,8 +214,7 @@ namespace {
out[y*width + x] += 400; // assume distance at borders, usually makes for better result
continue;
}
u32 d = data[yy*width + xx];
out[y*width + x] += DISTANCE(d, center);
out[y*width + x] += DISTANCE(data[yy*width + xx], center);
}
}
}
@ -500,7 +539,7 @@ bool TextureScaler::IsEmptyOrFlat(u32* data, int pixels, GLenum fmt) {
return true;
}
void TextureScaler::Scale(u32* &data, GEBufferFormat &dstFmt, int &width, int &height, int factor) {
void TextureScaler::Scale(u32* &data, GLenum &dstFmt, int &width, int &height, int factor) {
// prevent processing empty or flat textures (this happens a lot in some games)
// doesn't hurt the standard case, will be very quick for textures with actual texture
if(IsEmptyOrFlat(data, width*height, dstFmt)) {
@ -547,7 +586,7 @@ void TextureScaler::Scale(u32* &data, GEBufferFormat &dstFmt, int &width, int &h
// update values accordingly
data = outputBuf;
dstFmt = GE_FORMAT_8888;
dstFmt = GL_UNSIGNED_BYTE;
width *= factor;
height *= factor;
@ -618,39 +657,21 @@ void TextureScaler::DePosterize(u32* source, u32* dest, int width, int height) {
GlobalThreadPool::Loop(std::bind(&deposterizeV, bufTmp3.data(), dest, width, height, placeholder::_1, placeholder::_2), 0, height);
}
static void convert4444(u16* data, u32* out, int width, int l, int u) {
for (int y = l; y < u; ++y) {
ConvertRGBA4444ToRGBA8888(out + y * width, data + y * width, width);
}
}
static void convert565(u16* data, u32* out, int width, int l, int u) {
for (int y = l; y < u; ++y) {
ConvertRGB565ToRGBA888F(out + y * width, data + y * width, width);
}
}
static void convert5551(u16* data, u32* out, int width, int l, int u) {
for (int y = l; y < u; ++y) {
ConvertRGBA5551ToRGBA8888(out + y * width, data + y * width, width);
}
}
void TextureScaler::ConvertTo8888(GEBufferFormat format, u32* source, u32* &dest, int width, int height) {
void TextureScaler::ConvertTo8888(GLenum format, u32* source, u32* &dest, int width, int height) {
switch(format) {
case GE_FORMAT_8888:
case GL_UNSIGNED_BYTE:
dest = source; // already fine
break;
case GE_FORMAT_4444:
case GL_UNSIGNED_SHORT_4_4_4_4:
GlobalThreadPool::Loop(std::bind(&convert4444, (u16*)source, dest, width, placeholder::_1, placeholder::_2), 0, height);
break;
case GE_FORMAT_565:
case GL_UNSIGNED_SHORT_5_6_5:
GlobalThreadPool::Loop(std::bind(&convert565, (u16*)source, dest, width, placeholder::_1, placeholder::_2), 0, height);
break;
case GE_FORMAT_5551:
case GL_UNSIGNED_SHORT_5_5_5_1:
GlobalThreadPool::Loop(std::bind(&convert5551, (u16*)source, dest, width, placeholder::_1, placeholder::_2), 0, height);
break;

6
GPU/Common/TextureScaler.h → GPU/GLES/TextureScaler.h
View file

@ -20,15 +20,15 @@
#include "Common/MemoryUtil.h"
#include "../Globals.h"
#include "gfx/gl_common.h"
#include "GPU/ge_constants.h"
#include <vector>
class TextureScaler {
public:
TextureScaler();
void Scale(u32* &data, GEBufferFormat &dstfmt, int &width, int &height, int factor);
void Scale(u32* &data, GLenum &dstfmt, int &width, int &height, int factor);
enum { XBRZ= 0, HYBRID = 1, BICUBIC = 2, HYBRID_BICUBIC = 3 };
@ -38,7 +38,7 @@ private:
void ScaleBicubicBSpline(int factor, u32* source, u32* dest, int width, int height);
void ScaleBicubicMitchell(int factor, u32* source, u32* dest, int width, int height);
void ScaleHybrid(int factor, u32* source, u32* dest, int width, int height, bool bicubic = false);
void ConvertTo8888(GEBufferFormat format, u32* source, u32* &dest, int width, int height);
void ConvertTo8888(GLenum format, u32* source, u32* &dest, int width, int height);
void DePosterize(u32* source, u32* dest, int width, int height);

7
GPU/GPU.vcxproj
View file

@ -191,7 +191,6 @@
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
</ClInclude>
<ClInclude Include="Common\TextureCacheCommon.h" />
<ClInclude Include="Common\TextureScaler.h" />
<ClInclude Include="Common\TransformCommon.h" />
<ClInclude Include="Common\VertexDecoderCommon.h" />
<ClInclude Include="Debugger\Breakpoints.h" />
@ -205,6 +204,7 @@
<ClInclude Include="Directx9\ShaderManagerDX9.h" />
<ClInclude Include="Directx9\StateMappingDX9.h" />
<ClInclude Include="Directx9\TextureCacheDX9.h" />
<ClInclude Include="Directx9\TextureScalerDX9.h" />
<ClInclude Include="Directx9\TransformPipelineDX9.h" />
<ClInclude Include="Directx9\VertexShaderGeneratorDX9.h" />
<ClInclude Include="ge_constants.h" />
@ -217,6 +217,7 @@
<ClInclude Include="GLES\ShaderManager.h" />
<ClInclude Include="GLES\StateMapping.h" />
<ClInclude Include="GLES\TextureCache.h" />
<ClInclude Include="GLES\TextureScaler.h" />
<ClInclude Include="GLES\TransformPipeline.h" />
<ClInclude Include="GLES\VertexShaderGenerator.h" />
<ClInclude Include="GPUCommon.h" />
@ -225,6 +226,7 @@
<ClInclude Include="Math3D.h" />
<ClInclude Include="Null\NullGpu.h" />
<ClInclude Include="Software\Clipper.h" />
<ClInclude Include="Software\Colors.h" />
<ClInclude Include="Software\Lighting.h" />
<ClInclude Include="Software\Rasterizer.h" />
<ClInclude Include="Software\SoftGpu.h" />
@ -245,7 +247,6 @@
<ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild>
</ClCompile>
<ClCompile Include="Common\TextureCacheCommon.cpp" />
<ClCompile Include="Common\TextureScaler.cpp" />
<ClCompile Include="Common\TransformCommon.cpp" />
<ClCompile Include="Common\SoftwareTransformCommon.cpp" />
<ClCompile Include="Common\VertexDecoderArm.cpp">
@ -269,6 +270,7 @@
<ClCompile Include="Directx9\StateMappingDX9.cpp" />
<ClCompile Include="Directx9\StencilBufferDX9.cpp" />
<ClCompile Include="Directx9\TextureCacheDX9.cpp" />
<ClCompile Include="Directx9\TextureScalerDX9.cpp" />
<ClCompile Include="Directx9\TransformPipelineDX9.cpp" />
<ClCompile Include="Directx9\VertexShaderGeneratorDX9.cpp" />
<ClCompile Include="GeDisasm.cpp" />
@ -282,6 +284,7 @@
<ClCompile Include="GLES\StateMapping.cpp" />
<ClCompile Include="GLES\StencilBuffer.cpp" />
<ClCompile Include="GLES\TextureCache.cpp" />
<ClCompile Include="GLES\TextureScaler.cpp" />
<ClCompile Include="GLES\TransformPipeline.cpp" />
<ClCompile Include="GLES\VertexShaderGenerator.cpp" />
<ClCompile Include="GPUCommon.cpp" />

17
GPU/GPU.vcxproj.filters
View file

@ -42,8 +42,8 @@
<ClInclude Include="GPUCommon.h">
<Filter>Common</Filter>
</ClInclude>
<ClInclude Include="Common\TextureScaler.h">
<Filter>Common</Filter>
<ClInclude Include="Software\Colors.h">
<Filter>Software</Filter>
</ClInclude>
<ClInclude Include="Software\Clipper.h">
<Filter>Software</Filter>
@ -72,6 +72,9 @@
<ClInclude Include="GLES\TextureCache.h">
<Filter>GLES</Filter>
</ClInclude>
<ClInclude Include="GLES\TextureScaler.h">
<Filter>GLES</Filter>
</ClInclude>
<ClInclude Include="GLES\TransformPipeline.h">
<Filter>GLES</Filter>
</ClInclude>
@ -93,6 +96,9 @@
<ClInclude Include="Directx9\TransformPipelineDX9.h">
<Filter>DirectX9</Filter>
</ClInclude>
<ClInclude Include="Directx9\TextureScalerDX9.h">
<Filter>DirectX9</Filter>
</ClInclude>
<ClInclude Include="Directx9\TextureCacheDX9.h">
<Filter>DirectX9</Filter>
</ClInclude>
@ -251,8 +257,11 @@
<ClCompile Include="Directx9\PixelShaderGeneratorDX9.cpp">
<Filter>DirectX9</Filter>
</ClCompile>
<ClCompile Include="Common\TextureScaler.cpp">
<Filter>Common</Filter>
<ClCompile Include="Directx9\TextureScalerDX9.cpp">
<Filter>DirectX9</Filter>
</ClCompile>
<ClCompile Include="GLES\TextureScaler.cpp">
<Filter>GLES</Filter>
</ClCompile>
<ClCompile Include="Common\IndexGenerator.cpp">
<Filter>Common</Filter>

60
Common/ColorConv.h → GPU/Software/Colors.h
View file

@ -1,4 +1,4 @@
// Copyright (C) 2015 PPSSPP Project.
// Copyright (c) 2013- 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
@ -15,27 +15,12 @@
// Official git repository and contact information can be found at
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#pragma once
#include "CommonTypes.h"
#include "Common/CommonTypes.h"
inline u8 Convert4To8(u8 v) {
// Swizzle bits: 00001234 -> 12341234
return (v << 4) | (v);
}
inline u8 Convert5To8(u8 v) {
// Swizzle bits: 00012345 -> 12345123
return (v << 3) | (v >> 2);
}
inline u8 Convert6To8(u8 v) {
// Swizzle bits: 00123456 -> 12345612
return (v << 2) | (v >> 4);
}
inline u32 DecodeRGBA4444(u16 src) {
static inline u32 DecodeRGBA4444(u16 src)
{
const u32 r = (src & 0x000F) << 0;
const u32 g = (src & 0x00F0) << 4;
const u32 b = (src & 0x0F00) << 8;
@ -45,7 +30,8 @@ inline u32 DecodeRGBA4444(u16 src) {
return c | (c << 4);
}
inline u32 DecodeRGBA5551(u16 src) {
static inline u32 DecodeRGBA5551(u16 src)
{
u8 r = Convert5To8((src >> 0) & 0x1F);
u8 g = Convert5To8((src >> 5) & 0x1F);
u8 b = Convert5To8((src >> 10) & 0x1F);
@ -54,7 +40,8 @@ inline u32 DecodeRGBA5551(u16 src) {
return (a << 24) | (b << 16) | (g << 8) | r;
}
inline u32 DecodeRGB565(u16 src) {
static inline u32 DecodeRGB565(u16 src)
{
u8 r = Convert5To8((src >> 0) & 0x1F);
u8 g = Convert6To8((src >> 5) & 0x3F);
u8 b = Convert5To8((src >> 11) & 0x1F);
@ -62,7 +49,8 @@ inline u32 DecodeRGB565(u16 src) {
return (a << 24) | (b << 16) | (g << 8) | r;
}
inline u32 DecodeRGBA8888(u32 src) {
static inline u32 DecodeRGBA8888(u32 src)
{
#if 1
return src;
#else
@ -75,7 +63,8 @@ inline u32 DecodeRGBA8888(u32 src) {
#endif
}
inline u16 RGBA8888To565(u32 value) {
static inline u16 RGBA8888To565(u32 value)
{
u8 r = value & 0xFF;
u8 g = (value >> 8) & 0xFF;
u8 b = (value >> 16) & 0xFF;
@ -85,7 +74,8 @@ inline u16 RGBA8888To565(u32 value) {
return (u16)r | ((u16)g << 5) | ((u16)b << 11);
}
inline u16 RGBA8888To5551(u32 value) {
static inline u16 RGBA8888To5551(u32 value)
{
u8 r = value & 0xFF;
u8 g = (value >> 8) & 0xFF;
u8 b = (value >> 16) & 0xFF;
@ -97,24 +87,12 @@ inline u16 RGBA8888To5551(u32 value) {
return (u16)r | ((u16)g << 5) | ((u16)b << 10) | ((u16)a << 15);
}
static inline u16 RGBA8888To4444(u32 value) {
static inline u16 RGBA8888To4444(u32 value)
{
const u32 c = value >> 4;
const u16 r = (c >> 0) & 0x000F;
const u16 g = (c >> 4) & 0x00F0;
const u16 b = (c >> 8) & 0x0F00;
const u16 r = (c >> 0) & 0x000F;
const u16 g = (c >> 4) & 0x00F0;
const u16 b = (c >> 8) & 0x0F00;
const u16 a = (c >> 12) & 0xF000;
return r | g | b | a;
}
void ConvertBGRA8888ToRGB565(u16 *dst, const u32 *src, int numPixels);
void ConvertRGBA8888ToRGB565(u16 *dst, const u32 *src, int numPixels);
void ConvertBGRA8888ToRGBA4444(u16 *dst, const u32 *src, int numPixels);
void ConvertRGBA8888ToRGBA4444(u16 *dst, const u32 *src, int numPixels);
void ConvertRGBA8888ToRGBA5551(u16 *dst, const u32 *src, int numPixels);
void ConvertBGRA8888ToRGBA5551(u16 *dst, const u32 *src, int numPixels);
void ConvertRGB565ToRGBA888F(u32 *dst, const u16 *src, int numPixels);
void ConvertRGBA5551ToRGBA8888(u32 *dst, const u16 *src, int numPixels);
void ConvertRGBA4444ToRGBA8888(u32 *dst, const u16 *src, int numPixels);
void ConvertBGRA8888ToRGBA8888(u32 *dst, const u32 *src, int numPixels);

2
GPU/Software/Rasterizer.cpp
View file

@ -17,7 +17,6 @@
#include "base/basictypes.h"
#include "Common/ColorConv.h"
#include "Common/ThreadPools.h"
#include "Core/Config.h"
#include "Core/MemMap.h"
@ -27,6 +26,7 @@
#include "GPU/Common/TextureDecoder.h"
#include "GPU/Software/SoftGpu.h"
#include "GPU/Software/Rasterizer.h"
#include "GPU/Software/Colors.h"
#include <algorithm>

17
GPU/Software/SoftGpu.cpp
View file

@ -16,7 +16,6 @@
// https://github.com/hrydgard/ppsspp and http://www.ppsspp.org/.
#include "Common/ColorConv.h"
#include "GPU/GPUState.h"
#include "GPU/ge_constants.h"
#include "GPU/Common/TextureDecoder.h"
@ -33,6 +32,7 @@
#include "GPU/Software/SoftGpu.h"
#include "GPU/Software/TransformUnit.h"
#include "GPU/Software/Colors.h"
#include "GPU/Software/Rasterizer.h"
static GLuint temp_texture = 0;
@ -181,7 +181,8 @@ void SoftGPU::SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat for
}
// Copies RGBA8 data from RAM to the currently bound render target.
void SoftGPU::CopyToCurrentFboFromDisplayRam(int srcwidth, int srcheight) {
void SoftGPU::CopyToCurrentFboFromDisplayRam(int srcwidth, int srcheight)
{
float dstwidth = (float)PSP_CoreParameter().pixelWidth;
float dstheight = (float)PSP_CoreParameter().pixelHeight;
@ -212,15 +213,21 @@ void SoftGPU::CopyToCurrentFboFromDisplayRam(int srcwidth, int srcheight) {
switch (displayFormat_) {
case GE_FORMAT_565:
ConvertRGB565ToRGBA888F(buf_line, fb_line, srcwidth);
for (int x = 0; x < srcwidth; ++x) {
buf_line[x] = DecodeRGB565(fb_line[x]);
}
break;
case GE_FORMAT_5551:
ConvertRGBA5551ToRGBA8888(buf_line, fb_line, srcwidth);
for (int x = 0; x < srcwidth; ++x) {
buf_line[x] = DecodeRGBA5551(fb_line[x]);
}
break;
case GE_FORMAT_4444:
ConvertRGBA4444ToRGBA8888(buf_line, fb_line, srcwidth);
for (int x = 0; x < srcwidth; ++x) {
buf_line[x] = DecodeRGBA4444(fb_line[x]);
}
break;
default:

18
Globals.h
View file

@ -30,6 +30,24 @@
#define IS_LITTLE_ENDIAN (*(const u16 *)"\0\xff" >= 0x100)
#define IS_BIG_ENDIAN (*(const u16 *)"\0\xff" < 0x100)
inline u8 Convert4To8(u8 v)
{
// Swizzle bits: 00001234 -> 12341234
return (v << 4) | (v);
}
inline u8 Convert5To8(u8 v)
{
// Swizzle bits: 00012345 -> 12345123
return (v << 3) | (v >> 2);
}
inline u8 Convert6To8(u8 v)
{
// Swizzle bits: 00123456 -> 12345612
return (v << 2) | (v >> 4);
}
static inline u8 clamp_u8(int i) {
#ifdef ARM
asm("usat %0, #8, %1" : "=r"(i) : "r"(i));

2
Qt/Common.pro
View file

@ -45,7 +45,6 @@ win32 {
SOURCES += $$P/Common/ChunkFile.cpp \
$$P/Common/ConsoleListener.cpp \
$$P/Common/ColorConv.cpp \
$$P/Common/FileUtil.cpp \
$$P/Common/LogManager.cpp \
$$P/Common/KeyMap.cpp \
@ -58,7 +57,6 @@ SOURCES += $$P/Common/ChunkFile.cpp \
$$P/Common/Crypto/*.cpp
HEADERS += $$P/Common/ChunkFile.h \
$$P/Common/ConsoleListener.h \
$$P/Common/ColorConv.h \
$$P/Common/FileUtil.h \
$$P/Common/LogManager.h \
$$P/Common/KeyMap.h \

2
Qt/GPU.pro
View file

@ -36,6 +36,7 @@ SOURCES += $$P/GPU/GeDisasm.cpp \ # GPU
$$P/GPU/GLES/StateMapping.cpp \
$$P/GPU/GLES/StencilBuffer.cpp \
$$P/GPU/GLES/TextureCache.cpp \
$$P/GPU/GLES/TextureScaler.cpp \
$$P/GPU/GLES/TransformPipeline.cpp \
$$P/GPU/GLES/VertexShaderGenerator.cpp \
$$P/GPU/Software/*.cpp \
@ -43,7 +44,6 @@ SOURCES += $$P/GPU/GeDisasm.cpp \ # GPU
$$P/GPU/Common/IndexGenerator.cpp \
$$P/GPU/Common/TextureDecoder.cpp \
$$P/GPU/Common/VertexDecoderCommon.cpp \
$$P/GPU/Common/TextureScaler.cpp \
$$P/GPU/Common/TextureCacheCommon.cpp \
$$P/GPU/Common/TransformCommon.cpp \
$$P/GPU/Common/SoftwareTransformCommon.cpp \

2
Windows/WndMainWindow.cpp
View file

@ -68,7 +68,7 @@
#include "GPU/GPUInterface.h"
#include "GPU/GPUState.h"
#include "gfx_es2/gpu_features.h"
#include "GPU/Common/TextureScaler.h"
#include "GPU/GLES/TextureScaler.h"
#include "GPU/GLES/TextureCache.h"
#include "GPU/GLES/Framebuffer.h"
#include "ControlMapping.h"

3
android/jni/Android.mk
View file

@ -129,7 +129,6 @@ EXEC_AND_LIB_FILES := \
$(SRC)/ext/udis86/udis86.c \
$(SRC)/ext/xbrz/xbrz.cpp \
$(SRC)/ext/xxhash.c \
$(SRC)/Common/ColorConv.cpp \
$(SRC)/Common/Crypto/md5.cpp \
$(SRC)/Common/Crypto/sha1.cpp \
$(SRC)/Common/Crypto/sha256.cpp \
@ -153,7 +152,6 @@ EXEC_AND_LIB_FILES := \
$(SRC)/GPU/Common/SoftwareTransformCommon.cpp.arm \
$(SRC)/GPU/Common/VertexDecoderCommon.cpp.arm \
$(SRC)/GPU/Common/TextureCacheCommon.cpp.arm \
$(SRC)/GPU/Common/TextureScaler.cpp.arm \
$(SRC)/GPU/Common/SplineCommon.cpp.arm \
$(SRC)/GPU/Common/DrawEngineCommon.cpp.arm \
$(SRC)/GPU/Common/TransformCommon.cpp.arm \
@ -172,6 +170,7 @@ EXEC_AND_LIB_FILES := \
$(SRC)/GPU/GLES/VertexShaderGenerator.cpp.arm \
$(SRC)/GPU/GLES/FragmentShaderGenerator.cpp.arm \
$(SRC)/GPU/GLES/FragmentTestCache.cpp.arm \
$(SRC)/GPU/GLES/TextureScaler.cpp \
$(SRC)/GPU/GLES/Spline.cpp \
$(SRC)/GPU/Null/NullGpu.cpp \
$(SRC)/GPU/Software/Clipper.cpp \

1
headless/Compare.cpp
View file

@ -18,7 +18,6 @@
#include "headless/Compare.h"
#include "file/file_util.h"
#include "Core/Host.h"
#include "Common/ColorConv.h"
#include "GPU/Common/GPUDebugInterface.h"
#include "GPU/Common/TextureDecoder.h"