Colorspace Handler: Add support for NEON-A64.
Signed-off-by: rogerman <rogerman@users.noreply.github.com>
This commit is contained in:
rogerman
parent
8a9fec431a
commit
9ccc791e32
|
|
@ -38,6 +38,9 @@
|
|||
#elif defined(ENABLE_SSE2)
|
||||
#define USEVECTORSIZE_128
|
||||
#define VECTORSIZE 16
|
||||
#elif defined(ENABLE_NEON_A64)
|
||||
#define USEVECTORSIZE_128
|
||||
#define VECTORSIZE 16
|
||||
#elif defined(ENABLE_ALTIVEC)
|
||||
#define USEVECTORSIZE_128
|
||||
#define VECTORSIZE 16
|
||||
|
|
|
|||
|
|
@ -3636,6 +3636,8 @@
|
|||
AB96EE861F990E4700B7AA67 /* lzio.c */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.c; path = lzio.c; sourceTree = "<group>"; };
|
||||
AB96EE871F990E4700B7AA67 /* lzio.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = lzio.h; sourceTree = "<group>"; };
|
||||
AB9971CE134EDA0800531BA7 /* cocoa_globals.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = cocoa_globals.h; sourceTree = "<group>"; };
|
||||
ABA48DF527F95C2E00D961FB /* colorspacehandler_NEON.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = colorspacehandler_NEON.h; sourceTree = "<group>"; };
|
||||
ABA48DF627F95C2E00D961FB /* colorspacehandler_NEON.cpp */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.cpp; path = colorspacehandler_NEON.cpp; sourceTree = "<group>"; };
|
||||
ABA6574914511EC90077E5E9 /* cocoa_cheat.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = cocoa_cheat.h; sourceTree = "<group>"; };
|
||||
ABA6574A14511EC90077E5E9 /* cocoa_cheat.mm */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.objcpp; path = cocoa_cheat.mm; sourceTree = "<group>"; };
|
||||
ABA731251BB5104200B26147 /* SIL Open Font License.txt */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = text; path = "SIL Open Font License.txt"; sourceTree = "<group>"; };
|
||||
|
|
@ -4968,11 +4970,13 @@
|
|||
ABBFFF751D5FD2ED003CD598 /* colorspacehandler_SSE2.cpp */,
|
||||
ABBFFF7B1D610457003CD598 /* colorspacehandler_AVX2.cpp */,
|
||||
ABCC19332287879000DFA471 /* colorspacehandler_AVX512.cpp */,
|
||||
ABA48DF627F95C2E00D961FB /* colorspacehandler_NEON.cpp */,
|
||||
ABBFFF811D611A36003CD598 /* colorspacehandler_AltiVec.cpp */,
|
||||
ABBFFF701D5F9C52003CD598 /* colorspacehandler.h */,
|
||||
ABBFFF761D5FD2ED003CD598 /* colorspacehandler_SSE2.h */,
|
||||
ABBFFF7C1D610457003CD598 /* colorspacehandler_AVX2.h */,
|
||||
ABCC19342287879000DFA471 /* colorspacehandler_AVX512.h */,
|
||||
ABA48DF527F95C2E00D961FB /* colorspacehandler_NEON.h */,
|
||||
ABBFFF821D611A36003CD598 /* colorspacehandler_AltiVec.h */,
|
||||
);
|
||||
path = colorspacehandler;
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
/*
|
||||
Copyright (C) 2016-2021 DeSmuME team
|
||||
Copyright (C) 2016-2022 DeSmuME team
|
||||
|
||||
This file is free software: you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
|
|
@ -30,6 +30,10 @@
|
|||
#include "colorspacehandler_SSE2.cpp"
|
||||
#endif
|
||||
|
||||
#if defined(ENABLE_NEON_A64)
|
||||
#include "colorspacehandler_NEON.cpp"
|
||||
#endif
|
||||
|
||||
#if defined(ENABLE_ALTIVEC)
|
||||
#include "colorspacehandler_AltiVec.cpp"
|
||||
#endif
|
||||
|
|
@ -40,7 +44,7 @@
|
|||
#elif defined(ENABLE_AVX2)
|
||||
#define USEVECTORSIZE_256
|
||||
#define VECTORSIZE 32
|
||||
#elif defined(ENABLE_SSE2) || defined(ENABLE_ALTIVEC)
|
||||
#elif defined(ENABLE_SSE2) || defined(ENABLE_NEON_A64) || defined(ENABLE_ALTIVEC)
|
||||
#define USEVECTORSIZE_128
|
||||
#define VECTORSIZE 16
|
||||
#endif
|
||||
|
|
@ -60,6 +64,8 @@
|
|||
static const ColorspaceHandler_AVX2 csh;
|
||||
#elif defined(ENABLE_SSE2)
|
||||
static const ColorspaceHandler_SSE2 csh;
|
||||
#elif defined(ENABLE_NEON_A64)
|
||||
static const ColorspaceHandler_NEON csh;
|
||||
#elif defined(ENABLE_ALTIVEC)
|
||||
static const ColorspaceHandler_AltiVec csh;
|
||||
#else
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load Diff
|
|
@ -0,0 +1,114 @@
|
|||
/*
|
||||
Copyright (C) 2016-2022 DeSmuME team
|
||||
|
||||
This file 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, either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This file 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 for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License
|
||||
along with the this software. If not, see <http://www.gnu.org/licenses/>.
|
||||
*/
|
||||
|
||||
#ifndef COLORSPACEHANDLER_NEON_H
|
||||
#define COLORSPACEHANDLER_NEON_H
|
||||
|
||||
#include "colorspacehandler.h"
|
||||
|
||||
#ifndef ENABLE_NEON_A64
|
||||
#warning This header requires ARM64 NEON support.
|
||||
#else
|
||||
|
||||
template<bool SWAP_RB> void ColorspaceConvert555To8888_NEON(const v128u16 &srcColor, const v128u16 &srcAlphaBits, v128u32 &dstLo, v128u32 &dstHi);
|
||||
template<bool SWAP_RB> void ColorspaceConvert555XTo888X_NEON(const v128u16 &srcColor, v128u32 &dstLo, v128u32 &dstHi);
|
||||
template<bool SWAP_RB> void ColorspaceConvert555To6665_NEON(const v128u16 &srcColor, const v128u16 &srcAlphaBits, v128u32 &dstLo, v128u32 &dstHi);
|
||||
template<bool SWAP_RB> void ColorspaceConvert555XTo666X_NEON(const v128u16 &srcColor, v128u32 &dstLo, v128u32 &dstHi);
|
||||
template<bool SWAP_RB> void ColorspaceConvert555To8888Opaque_NEON(const v128u16 &srcColor, v128u32 &dstLo, v128u32 &dstHi);
|
||||
template<bool SWAP_RB> void ColorspaceConvert555To6665Opaque_NEON(const v128u16 &srcColor, v128u32 &dstLo, v128u32 &dstHi);
|
||||
template<bool SWAP_RB> v128u32 ColorspaceConvert8888To6665_NEON(const v128u32 &src);
|
||||
template<bool SWAP_RB> v128u32 ColorspaceConvert6665To8888_NEON(const v128u32 &src);
|
||||
template<bool SWAP_RB> v128u16 ColorspaceConvert8888To5551_NEON(const v128u32 &srcLo, const v128u32 &srcHi);
|
||||
template<bool SWAP_RB> v128u16 ColorspaceConvert6665To5551_NEON(const v128u32 &srcLo, const v128u32 &srcHi);
|
||||
template<bool SWAP_RB> v128u32 C6olorspaceConvert888XTo8888Opaque_NEON(const v128u32 &src);
|
||||
|
||||
template<bool SWAP_RB> v128u16 ColorspaceCopy16_NEON(const v128u16 &src);
|
||||
template<bool SWAP_RB> v128u32 ColorspaceCopy32_NEON(const v128u32 &src);
|
||||
|
||||
template<bool SWAP_RB> v128u16 ColorspaceApplyIntensity16_NEON(const v128u16 &src, float intensity);
|
||||
template<bool SWAP_RB> v128u32 ColorspaceApplyIntensity32_NEON(const v128u32 &src, float intensity);
|
||||
|
||||
class ColorspaceHandler_NEON : public ColorspaceHandler
|
||||
{
|
||||
public:
|
||||
ColorspaceHandler_NEON() {};
|
||||
|
||||
template<BESwapFlags BE_BYTESWAP> size_t ConvertBuffer555To8888Opaque(const u16 *__restrict src, u32 *__restrict dst, size_t pixCount) const;
|
||||
template<BESwapFlags BE_BYTESWAP> size_t ConvertBuffer555To8888Opaque_SwapRB(const u16 *__restrict src, u32 *__restrict dst, size_t pixCount) const;
|
||||
template<BESwapFlags BE_BYTESWAP> size_t ConvertBuffer555To8888Opaque_IsUnaligned(const u16 *__restrict src, u32 *__restrict dst, size_t pixCount) const;
|
||||
template<BESwapFlags BE_BYTESWAP> size_t ConvertBuffer555To8888Opaque_SwapRB_IsUnaligned(const u16 *__restrict src, u32 *__restrict dst, size_t pixCount) const;
|
||||
|
||||
template<BESwapFlags BE_BYTESWAP> size_t ConvertBuffer555To6665Opaque(const u16 *__restrict src, u32 *__restrict dst, size_t pixCount) const;
|
||||
template<BESwapFlags BE_BYTESWAP> size_t ConvertBuffer555To6665Opaque_SwapRB(const u16 *__restrict src, u32 *__restrict dst, size_t pixCount) const;
|
||||
template<BESwapFlags BE_BYTESWAP> size_t ConvertBuffer555To6665Opaque_IsUnaligned(const u16 *__restrict src, u32 *__restrict dst, size_t pixCount) const;
|
||||
template<BESwapFlags BE_BYTESWAP> size_t ConvertBuffer555To6665Opaque_SwapRB_IsUnaligned(const u16 *__restrict src, u32 *__restrict dst, size_t pixCount) const;
|
||||
|
||||
size_t ConvertBuffer8888To6665(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer8888To6665_SwapRB(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer8888To6665_IsUnaligned(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer8888To6665_SwapRB_IsUnaligned(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
|
||||
size_t ConvertBuffer6665To8888(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer6665To8888_SwapRB(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer6665To8888_IsUnaligned(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer6665To8888_SwapRB_IsUnaligned(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
|
||||
size_t ConvertBuffer8888To5551(const u32 *__restrict src, u16 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer8888To5551_SwapRB(const u32 *__restrict src, u16 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer8888To5551_IsUnaligned(const u32 *__restrict src, u16 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer8888To5551_SwapRB_IsUnaligned(const u32 *__restrict src, u16 *__restrict dst, size_t pixCount) const;
|
||||
|
||||
size_t ConvertBuffer6665To5551(const u32 *__restrict src, u16 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer6665To5551_SwapRB(const u32 *__restrict src, u16 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer6665To5551_IsUnaligned(const u32 *__restrict src, u16 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer6665To5551_SwapRB_IsUnaligned(const u32 *__restrict src, u16 *__restrict dst, size_t pixCount) const;
|
||||
|
||||
size_t ConvertBuffer888XTo8888Opaque(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer888XTo8888Opaque_SwapRB(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer888XTo8888Opaque_IsUnaligned(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer888XTo8888Opaque_SwapRB_IsUnaligned(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
|
||||
size_t ConvertBuffer555XTo888(const u16 *__restrict src, u8 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer555XTo888_SwapRB(const u16 *__restrict src, u8 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer555XTo888_IsUnaligned(const u16 *__restrict src, u8 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer555XTo888_SwapRB_IsUnaligned(const u16 *__restrict src, u8 *__restrict dst, size_t pixCount) const;
|
||||
|
||||
size_t ConvertBuffer888XTo888(const u32 *__restrict src, u8 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer888XTo888_SwapRB(const u32 *__restrict src, u8 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer888XTo888_IsUnaligned(const u32 *__restrict src, u8 *__restrict dst, size_t pixCount) const;
|
||||
size_t ConvertBuffer888XTo888_SwapRB_IsUnaligned(const u32 *__restrict src, u8 *__restrict dst, size_t pixCount) const;
|
||||
|
||||
size_t CopyBuffer16_SwapRB(const u16 *src, u16 *dst, size_t pixCount) const;
|
||||
size_t CopyBuffer16_SwapRB_IsUnaligned(const u16 *src, u16 *dst, size_t pixCount) const;
|
||||
|
||||
size_t CopyBuffer32_SwapRB(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
size_t CopyBuffer32_SwapRB_IsUnaligned(const u32 *src, u32 *dst, size_t pixCount) const;
|
||||
|
||||
size_t ApplyIntensityToBuffer16(u16 *dst, size_t pixCount, float intensity) const;
|
||||
size_t ApplyIntensityToBuffer16_SwapRB(u16 *dst, size_t pixCount, float intensity) const;
|
||||
size_t ApplyIntensityToBuffer16_IsUnaligned(u16 *dst, size_t pixCount, float intensity) const;
|
||||
size_t ApplyIntensityToBuffer16_SwapRB_IsUnaligned(u16 *dst, size_t pixCount, float intensity) const;
|
||||
|
||||
size_t ApplyIntensityToBuffer32(u32 *dst, size_t pixCount, float intensity) const;
|
||||
size_t ApplyIntensityToBuffer32_SwapRB(u32 *dst, size_t pixCount, float intensity) const;
|
||||
size_t ApplyIntensityToBuffer32_IsUnaligned(u32 *dst, size_t pixCount, float intensity) const;
|
||||
size_t ApplyIntensityToBuffer32_SwapRB_IsUnaligned(u32 *dst, size_t pixCount, float intensity) const;
|
||||
};
|
||||
|
||||
#endif // ENABLE_NEON_A64
|
||||
|
||||
#endif // COLORSPACEHANDLER_NEON_H
|
||||
Loading…
Reference in New Issue