generic-library/openh264
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Modify code style, remove trailing space.

Browse Source
This commit is contained in:
Licai Guo
2014-03-03 15:42:01 +08:00
parent b7a25df13f
commit 7768cd0a98
15 changed files with 1636 additions and 1972 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
codec/build/iOS/common/common.xcodeproj/project.pbxproj
View File

@@ -71,7 +71,7 @@
4CE4474718BC61650017DF25 /* typedefs.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = typedefs.h; sourceTree = "<group>"; };
4CE4474918BC61650017DF25 /* WelsThreadLib.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = WelsThreadLib.cpp; sourceTree = "<group>"; };
4CE4474A18BC61650017DF25 /* WelsThreadLib.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = WelsThreadLib.h; sourceTree = "<group>"; };
4CE447BC18C085320017DF25 /* deblocking_neon.S */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.asm; path = deblocking_neon.S; sourceTree = "<group>"; };
4CE447BC18C085320017DF25 /* deblocking_neon.S */ = {isa = PBXFileReference; fileEncoding = 4; indentWidth = 4; lastKnownFileType = sourcecode.asm; path = deblocking_neon.S; sourceTree = "<group>"; tabWidth = 4; usesTabs = 0; };
4CE447BE18C085900017DF25 /* arm_arch_common_macro.S */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.asm; path = arm_arch_common_macro.S; sourceTree = "<group>"; };
/* End PBXFileReference section */

18
codec/build/iOS/dec/welsdec/welsdec.xcodeproj/project.pbxproj
View File

@@ -84,9 +84,9 @@
4CE4464E18BC5EAA0017DF25 /* decoder_context.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = decoder_context.h; sourceTree = "<group>"; };
4CE4464F18BC5EAA0017DF25 /* decoder_core.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = decoder_core.h; sourceTree = "<group>"; };
4CE4465018BC5EAA0017DF25 /* error_code.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = error_code.h; sourceTree = "<group>"; };
4CE4465118BC5EAA0017DF25 /* expand_pic.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = expand_pic.h; sourceTree = "<group>"; };
4CE4465218BC5EAA0017DF25 /* fmo.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = fmo.h; sourceTree = "<group>"; };
4CE4465318BC5EAA0017DF25 /* get_intra_predictor.h */ = {isa = PBXFileReference; fileEncoding = 4; indentWidth = 2; lastKnownFileType = sourcecode.c.h; path = get_intra_predictor.h; sourceTree = "<group>"; tabWidth = 2; usesTabs = 1; };
4CE4465118BC5EAA0017DF25 /* expand_pic.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = expand_pic.h; sourceTree = "<group>"; usesTabs = 1; };
4CE4465218BC5EAA0017DF25 /* fmo.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = fmo.h; sourceTree = "<group>"; usesTabs = 1; };
4CE4465318BC5EAA0017DF25 /* get_intra_predictor.h */ = {isa = PBXFileReference; fileEncoding = 4; indentWidth = 4; lastKnownFileType = sourcecode.c.h; path = get_intra_predictor.h; sourceTree = "<group>"; tabWidth = 4; usesTabs = 0; };
4CE4465418BC5EAA0017DF25 /* manage_dec_ref.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = manage_dec_ref.h; sourceTree = "<group>"; };
4CE4465518BC5EAA0017DF25 /* mb_cache.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = mb_cache.h; sourceTree = "<group>"; };
4CE4465618BC5EAA0017DF25 /* mc.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = mc.h; sourceTree = "<group>"; };
@@ -105,19 +105,19 @@
4CE4466318BC5EAA0017DF25 /* vlc_decoder.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = vlc_decoder.h; sourceTree = "<group>"; };
4CE4466418BC5EAA0017DF25 /* wels_common_basis.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = wels_common_basis.h; sourceTree = "<group>"; };
4CE4466518BC5EAA0017DF25 /* wels_const.h */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.c.h; path = wels_const.h; sourceTree = "<group>"; };
4CE4466718BC5EAA0017DF25 /* au_parser.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = au_parser.cpp; sourceTree = "<group>"; usesTabs = 1; };
4CE4466818BC5EAA0017DF25 /* bit_stream.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = bit_stream.cpp; sourceTree = "<group>"; usesTabs = 1; };
4CE4466918BC5EAA0017DF25 /* deblocking.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = deblocking.cpp; sourceTree = "<group>"; };
4CE4466718BC5EAA0017DF25 /* au_parser.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = au_parser.cpp; sourceTree = "<group>"; usesTabs = 0; };
4CE4466818BC5EAA0017DF25 /* bit_stream.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = bit_stream.cpp; sourceTree = "<group>"; usesTabs = 0; };
4CE4466918BC5EAA0017DF25 /* deblocking.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = deblocking.cpp; sourceTree = "<group>"; tabWidth = 2; };
4CE4466A18BC5EAA0017DF25 /* decode_mb_aux.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = decode_mb_aux.cpp; sourceTree = "<group>"; };
4CE4466B18BC5EAA0017DF25 /* decode_slice.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = decode_slice.cpp; sourceTree = "<group>"; usesTabs = 1; };
4CE4466C18BC5EAA0017DF25 /* decoder.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = decoder.cpp; sourceTree = "<group>"; tabWidth = 2; usesTabs = 1; };
4CE4466B18BC5EAA0017DF25 /* decode_slice.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = decode_slice.cpp; sourceTree = "<group>"; usesTabs = 0; };
4CE4466C18BC5EAA0017DF25 /* decoder.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = decoder.cpp; sourceTree = "<group>"; tabWidth = 4; usesTabs = 0; };
4CE4466D18BC5EAA0017DF25 /* decoder_core.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = decoder_core.cpp; sourceTree = "<group>"; };
4CE4466E18BC5EAA0017DF25 /* decoder_data_tables.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = decoder_data_tables.cpp; sourceTree = "<group>"; };
4CE4466F18BC5EAA0017DF25 /* expand_pic.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = expand_pic.cpp; sourceTree = "<group>"; };
4CE4467018BC5EAA0017DF25 /* fmo.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = fmo.cpp; sourceTree = "<group>"; };
4CE4467118BC5EAA0017DF25 /* get_intra_predictor.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = get_intra_predictor.cpp; sourceTree = "<group>"; };
4CE4467218BC5EAA0017DF25 /* manage_dec_ref.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = manage_dec_ref.cpp; sourceTree = "<group>"; };
4CE4467318BC5EAA0017DF25 /* mc.cpp */ = {isa = PBXFileReference; fileEncoding = 4; indentWidth = 4; lastKnownFileType = sourcecode.cpp.cpp; path = mc.cpp; sourceTree = "<group>"; tabWidth = 1; usesTabs = 1; wrapsLines = 1; };
4CE4467318BC5EAA0017DF25 /* mc.cpp */ = {isa = PBXFileReference; fileEncoding = 4; indentWidth = 4; lastKnownFileType = sourcecode.cpp.cpp; path = mc.cpp; sourceTree = "<group>"; tabWidth = 1; usesTabs = 0; wrapsLines = 1; };
4CE4467418BC5EAA0017DF25 /* mem_align.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = mem_align.cpp; sourceTree = "<group>"; };
4CE4467518BC5EAA0017DF25 /* memmgr_nal_unit.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = memmgr_nal_unit.cpp; sourceTree = "<group>"; };
4CE4467618BC5EAA0017DF25 /* mv_pred.cpp */ = {isa = PBXFileReference; fileEncoding = 4; lastKnownFileType = sourcecode.cpp.cpp; path = mv_pred.cpp; sourceTree = "<group>"; };

37
codec/common/cpu.cpp
View File

@@ -218,25 +218,23 @@ void WelsXmmRegEmptyOp(void * pSrc) {
#if defined(ANDROID_NDK)
uint32_t WelsCPUFeatureDetectAndroid()
{
uint32_t uiCPU = 0;
AndroidCpuFamily cpuFamily = ANDROID_CPU_FAMILY_UNKNOWN;
uint64_t uiFeatures = 0;
cpuFamily = android_getCpuFamily();
if (cpuFamily == ANDROID_CPU_FAMILY_ARM)
{
uiFeatures = android_getCpuFeatures();
if (uiFeatures & ANDROID_CPU_ARM_FEATURE_ARMv7){
uiCPU |= WELS_CPU_ARMv7;
}
if (uiFeatures & ANDROID_CPU_ARM_FEATURE_VFPv3){
uiCPU |= WELS_CPU_VFPv3;
}
if (uiFeatures & ANDROID_CPU_ARM_FEATURE_NEON){
uiCPU |= WELS_CPU_NEON;
}
}
return uiCPU;
uint32_t uiCPU = 0;
AndroidCpuFamily cpuFamily = ANDROID_CPU_FAMILY_UNKNOWN;
uint64_t uiFeatures = 0;
cpuFamily = android_getCpuFamily();
if (cpuFamily == ANDROID_CPU_FAMILY_ARM) {
uiFeatures = android_getCpuFeatures();
if (uiFeatures & ANDROID_CPU_ARM_FEATURE_ARMv7){
uiCPU |= WELS_CPU_ARMv7;
}
if (uiFeatures & ANDROID_CPU_ARM_FEATURE_VFPv3){
uiCPU |= WELS_CPU_VFPv3;
}
if (uiFeatures & ANDROID_CPU_ARM_FEATURE_NEON){
uiCPU |= WELS_CPU_NEON;
}
}
return uiCPU;
}
#endif
@@ -246,7 +244,6 @@ uint32_t WelsCPUFeatureDetectIOS() //Need to be updated for the new device of AP
{
uint32_t uiCPU = 0;
struct utsname sSystemInfo;
uname (&sSystemInfo);
if ((0 != strcmp(sSystemInfo.machine, "iPhone1,1")) && //iPhone 2G

1599
codec/common/deblocking_neon.S
View File

File diff suppressed because it is too large Load Diff

28
codec/common/mc_common.h
View File

@@ -40,34 +40,6 @@ extern "C" {
#endif//__cplusplus
#if defined(HAVE_NEON)
/*
void McCopy_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride,
int32_t iWidth, int32_t iHeight);
void McHorVer20_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride,
int32_t iWidth, int32_t iHeight);
void McHorVer02_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride,
int32_t iWidth, int32_t iHeight);
void McHorVer22_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride,
int32_t iWidth, int32_t iHeight);
void McHorVer01_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer03_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer10_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer11_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer12_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer13_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer21_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer23_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer30_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer31_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer32_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McHorVer33_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iWidth, int32_t iHeight);
void McLuma_neon(uint8_t* pSrc, int32_t src_stride, uint8_t* dst, int32_t dst_stride,
int16_t iMvX, int16_t iMvY, int32_t width, int32_t height);
void McChroma_neon(uint8_t* pSrc, int32_t src_stride, uint8_t* dst, int32_t dst_stride,
int16_t iMvX, int16_t iMvY, int32_t width, int32_t height);
*/
void McCopyWidthEq4_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iHeight);
void McCopyWidthEq8_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride, int32_t iHeight);

80
codec/decoder/core/arm/block_add_neon.S
View File

@@ -37,32 +37,32 @@
.macro ROW_TRANSFORM_1_STEP
// { // input: src_d[0]~[3], output: e_q[0]~[3]; working: $8 $9
vaddl.s16 $4, $0, $2 //int32 e[i][0] = src[0] + src[2];
vsubl.s16 $5, $0, $2 //int32 e[i][1] = src[0] - src[2];
vshr.s16 $8, $1, #1
vshr.s16 $9, $3, #1
vsubl.s16 $6, $8, $3 //int32 e[i][2] = (src[1]>>1)-src[3];
vaddl.s16 $7, $1, $9 //int32 e[i][3] = src[1] + (src[3]>>1);
vaddl.s16 $4, $0, $2 //int32 e[i][0] = src[0] + src[2];
vsubl.s16 $5, $0, $2 //int32 e[i][1] = src[0] - src[2];
vshr.s16 $8, $1, #1
vshr.s16 $9, $3, #1
vsubl.s16 $6, $8, $3 //int32 e[i][2] = (src[1]>>1)-src[3];
vaddl.s16 $7, $1, $9 //int32 e[i][3] = src[1] + (src[3]>>1);
// }
.endm
.macro TRANSFORM_4BYTES // both row & col transform used
// { // output: f_q[0]~[3], input: e_q[0]~[3];
vadd.s32 $0, $4, $7 //int16 f[i][0] = e[i][0] + e[i][3];
vadd.s32 $1, $5, $6 //int16 f[i][1] = e[i][1] + e[i][2];
vsub.s32 $2, $5, $6 //int16 f[i][2] = e[i][1] - e[i][2];
vsub.s32 $3, $4, $7 //int16 f[i][3] = e[i][0] - e[i][3];
vadd.s32 $0, $4, $7 //int16 f[i][0] = e[i][0] + e[i][3];
vadd.s32 $1, $5, $6 //int16 f[i][1] = e[i][1] + e[i][2];
vsub.s32 $2, $5, $6 //int16 f[i][2] = e[i][1] - e[i][2];
vsub.s32 $3, $4, $7 //int16 f[i][3] = e[i][0] - e[i][3];
// }
.endm
.macro COL_TRANSFORM_1_STEP
// { // input: src_q[0]~[3], output: e_q[0]~[3];
vadd.s32 $4, $0, $2 //int32 e[0][j] = f[0][j] + f[2][j];
vsub.s32 $5, $0, $2 //int32 e[1][j] = f[0][j] - f[2][j];
vshr.s32 $6, $1, #1
vshr.s32 $7, $3, #1
vsub.s32 $6, $6, $3 //int32 e[2][j] = (f[1][j]>>1) - f[3][j];
vadd.s32 $7, $1, $7 //int32 e[3][j] = f[1][j] + (f[3][j]>>1);
vadd.s32 $4, $0, $2 //int32 e[0][j] = f[0][j] + f[2][j];
vsub.s32 $5, $0, $2 //int32 e[1][j] = f[0][j] - f[2][j];
vshr.s32 $6, $1, #1
vshr.s32 $7, $3, #1
vsub.s32 $6, $6, $3 //int32 e[2][j] = (f[1][j]>>1) - f[3][j];
vadd.s32 $7, $1, $7 //int32 e[3][j] = f[1][j] + (f[3][j]>>1);
// }
.endm
@@ -70,38 +70,38 @@
.macro ROW_TRANSFORM_1_STEP arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8, arg9
// { // input: src_d[0]~[3], output: e_q[0]~[3]; working: $8 $9
vaddl.s16 \arg4, \arg0, \arg2 //int32 e[i][0] = src[0] + src[2];
vsubl.s16 \arg5, \arg0, \arg2 //int32 e[i][1] = src[0] - src[2];
vshr.s16 \arg8, \arg1, #1
vshr.s16 \arg9, \arg3, #1
vsubl.s16 \arg6, \arg8, \arg3 //int32 e[i][2] = (src[1]>>1)-src[3];
vaddl.s16 \arg7, \arg1, \arg9 //int32 e[i][3] = src[1] + (src[3]>>1);
vaddl.s16 \arg4, \arg0, \arg2 //int32 e[i][0] = src[0] + src[2];
vsubl.s16 \arg5, \arg0, \arg2 //int32 e[i][1] = src[0] - src[2];
vshr.s16 \arg8, \arg1, #1
vshr.s16 \arg9, \arg3, #1
vsubl.s16 \arg6, \arg8, \arg3 //int32 e[i][2] = (src[1]>>1)-src[3];
vaddl.s16 \arg7, \arg1, \arg9 //int32 e[i][3] = src[1] + (src[3]>>1);
// }
.endm
.macro TRANSFORM_4BYTES arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7 // both row & col transform used
// { // output: f_q[0]~[3], input: e_q[0]~[3];
vadd.s32 \arg0, \arg4, \arg7 //int16 f[i][0] = e[i][0] + e[i][3];
vadd.s32 \arg1, \arg5, \arg6 //int16 f[i][1] = e[i][1] + e[i][2];
vsub.s32 \arg2, \arg5, \arg6 //int16 f[i][2] = e[i][1] - e[i][2];
vsub.s32 \arg3, \arg4, \arg7 //int16 f[i][3] = e[i][0] - e[i][3];
vadd.s32 \arg0, \arg4, \arg7 //int16 f[i][0] = e[i][0] + e[i][3];
vadd.s32 \arg1, \arg5, \arg6 //int16 f[i][1] = e[i][1] + e[i][2];
vsub.s32 \arg2, \arg5, \arg6 //int16 f[i][2] = e[i][1] - e[i][2];
vsub.s32 \arg3, \arg4, \arg7 //int16 f[i][3] = e[i][0] - e[i][3];
// }
.endm
.macro COL_TRANSFORM_1_STEP arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7
// { // input: src_q[0]~[3], output: e_q[0]~[3];
vadd.s32 \arg4, \arg0, \arg2 //int32 e[0][j] = f[0][j] + f[2][j];
vsub.s32 \arg5, \arg0, \arg2 //int32 e[1][j] = f[0][j] - f[2][j];
vshr.s32 \arg6, \arg1, #1
vshr.s32 \arg7, \arg3, #1
vsub.s32 \arg6, \arg6, \arg3 //int32 e[2][j] = (f[1][j]>>1) - f[3][j];
vadd.s32 \arg7, \arg1, \arg7 //int32 e[3][j] = f[1][j] + (f[3][j]>>1);
vadd.s32 \arg4, \arg0, \arg2 //int32 e[0][j] = f[0][j] + f[2][j];
vsub.s32 \arg5, \arg0, \arg2 //int32 e[1][j] = f[0][j] - f[2][j];
vshr.s32 \arg6, \arg1, #1
vshr.s32 \arg7, \arg3, #1
vsub.s32 \arg6, \arg6, \arg3 //int32 e[2][j] = (f[1][j]>>1) - f[3][j];
vadd.s32 \arg7, \arg1, \arg7 //int32 e[3][j] = f[1][j] + (f[3][j]>>1);
// }
.endm
#endif
// r0 int16_t* block,
// r1 int8_t* non_zero_count,
WELS_ASM_FUNC_BEGIN SetNonZeroCount_neon
WELS_ASM_FUNC_BEGIN SetNonZeroCount_neon
vld1.64 {d0-d2}, [r1]
@@ -113,12 +113,12 @@
vabs.s8 d2, d2
vst1.64 {d0-d2}, [r1]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
// r0 int16_t * block,
// r1 int32_t stride
WELS_ASM_FUNC_BEGIN WelsResBlockZero16x16_neon// can use for 256*sizeof(int16_t)
WELS_ASM_FUNC_BEGIN WelsResBlockZero16x16_neon// can use for 256*sizeof(int16_t)
push {r2}
mov r2, #16
// each row 16 elements, 16*sizeof(int16_t)
@@ -135,9 +135,9 @@ block_zero_16x16_luma_loop:
bne block_zero_16x16_luma_loop
pop {r2}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN WelsResBlockZero8x8_neon// can use for 64*sizeof(int16_t)
WELS_ASM_FUNC_BEGIN WelsResBlockZero8x8_neon// can use for 64*sizeof(int16_t)
push {r2}
mov r2, #8
// each row 8 elements, 8*sizeof(int16_t)
@@ -153,11 +153,11 @@ block_zero_8x8_chma_loop:
bne block_zero_8x8_chma_loop
pop {r2}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
// uint8_t *pred, const int32_t stride, int16_t *rs
WELS_ASM_FUNC_BEGIN IdctResAddPred_neon
WELS_ASM_FUNC_BEGIN IdctResAddPred_neon
vld4.s16 {d0, d1, d2, d3}, [r2] // cost 3 cycles!
@@ -199,5 +199,5 @@ block_zero_8x8_chma_loop:
vst1.32 {d12[1]},[r2],r1
vst1.32 {d14[0]},[r2],r1
vst1.32 {d14[1]},[r2]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
#endif

173
codec/decoder/core/arm/intra_pred_neon.S
View File

@@ -61,14 +61,8 @@
.endm
#endif
/*
* void get_i16x16_luma_pred_v(uint8_t *pred, const int32_t stride)
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI16x16LumaPredV_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI16x16LumaPredV_neon
//Get the top line data to 'q0'
sub r2, r0, r1
vldm r2, {d0, d1}
@@ -84,17 +78,11 @@ loop_0_get_i16x16_luma_pred_v:
subs r3, #1
bne loop_0_get_i16x16_luma_pred_v
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i16x16_luma_pred_h(uint8_t *pred, const int32_t stride)
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI16x16LumaPredH_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI16x16LumaPredH_neon
sub r2, r0, #1
mov r3, #4
loop_0_get_i16x16_luma_pred_h:
@@ -113,17 +101,10 @@ loop_0_get_i16x16_luma_pred_h:
subs r3, #1
bne loop_0_get_i16x16_luma_pred_h
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i16x16_luma_pred_dc_both(uint8_t *pred, const int32_t stride)
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI16x16LumaPredDc_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI16x16LumaPredDc_neon
//stmdb sp!, { r2-r5, lr}
//Get the left vertical line data
sub r2, r0, #1
@@ -156,15 +137,9 @@ loop_0_get_i16x16_luma_pred_dc_both:
subs r2, #1
bne loop_0_get_i16x16_luma_pred_dc_both
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i16x16_luma_pred_plane(uint8_t *pred, const int32_t stride)
* r0 --- pred
* r1 --- stride
* return --- void
*/
//The table for SIMD instruction {(8,7,6,5,4,3,2,1) * 5}
CONST0_GET_I16X16_LUMA_PRED_PLANE: .long 0x191e2328, 0x050a0f14
@@ -173,7 +148,7 @@ CONST0_GET_I16X16_LUMA_PRED_PLANE: .long 0x191e2328, 0x050a0f14
CONST1_GET_I16X16_LUMA_PRED_PLANE: .long 0xfcfbfaf9, 0x00fffefd
WELS_ASM_FUNC_BEGIN WelsDecoderI16x16LumaPredPlane_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI16x16LumaPredPlane_neon
//stmdb sp!, { r2-r5, lr}
//Load the table {(8,7,6,5,4,3,2,1) * 5}
@@ -261,17 +236,9 @@ loop_0_get_i16x16_luma_pred_plane:
subs r2, #1
bne loop_0_get_i16x16_luma_pred_plane
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i4x4_luma_pred_v(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredV_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredV_neon
//stmdb sp!, { r2-r5, lr}
//Load the top row (4 bytes)
sub r2, r0, r1
@@ -283,16 +250,11 @@ loop_0_get_i16x16_luma_pred_plane:
str r2, [r0], r1
str r2, [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i4x4_luma_pred_h(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredH_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredH_neon
//stmdb sp!, { r2-r5, lr}
//Load the left column (4 bytes)
sub r2, r0, #1
@@ -307,15 +269,10 @@ loop_0_get_i16x16_luma_pred_plane:
vst1.32 {d2[0]}, [r0], r1
vst1.32 {d3[0]}, [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i4x4_luma_pred_d_l(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredDDL_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredDDL_neon
//stmdb sp!, { r2-r5, lr}
//Load the top row data(8 bytes)
sub r2, r0, r1
@@ -350,16 +307,10 @@ loop_0_get_i16x16_luma_pred_plane:
//Save "ddl3, ddl4, ddl5, ddl6"
vst1.32 d0[1], [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i4x4_luma_pred_d_r(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredDDR_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredDDR_neon
//stmdb sp!, { r2-r5, lr}
//Load the top row (4 bytes)
sub r2, r0, r1
@@ -396,16 +347,11 @@ loop_0_get_i16x16_luma_pred_plane:
vext.8 d0, d0, d0, #7
vst1.32 d0[1], [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i4x4_luma_pred_v_l(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredVL_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredVL_neon
//stmdb sp!, { r2-r5, lr}
//Load the top row (8 bytes)
sub r2, r0, r1
@@ -432,17 +378,11 @@ loop_0_get_i16x16_luma_pred_plane:
vst1.32 d0[0], [r0], r1
vst1.32 d1[0], [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i4x4_luma_pred_v_r(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredVR_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredVR_neon
//stmdb sp!, { r2-r5, lr}
//Load the top row (4 bytes)
sub r2, r0, r1
@@ -476,17 +416,11 @@ loop_0_get_i16x16_luma_pred_plane:
vst1.8 d1[2], [r2]!
vst1.16 d1[2], [r2]!
vst1.8 d1[6], [r2]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* get_i4x4_luma_pred_h_u(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
//NO TEST
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredHU_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredHU_neon
//stmdb sp!, { r2-r5, lr}
//Load the left column data
sub r2, r0, #1
@@ -516,15 +450,10 @@ loop_0_get_i16x16_luma_pred_plane:
vst1.32 d1[1], [r0], r1
vst1.32 d0[0], [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i4x4_luma_pred_h_d(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredHD_neon
WELS_ASM_FUNC_BEGIN WelsDecoderI4x4LumaPredHD_neon
//stmdb sp!, { r2-r5, lr}
//Load the data
sub r2, r0, r1
@@ -556,17 +485,10 @@ loop_0_get_i16x16_luma_pred_plane:
vst2.16 {d2[2], d3[2]}, [r0], r1
vst2.16 {d0[1], d1[1]}, [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i_chroma_pred_v(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderIChromaPredV_neon
WELS_ASM_FUNC_BEGIN WelsDecoderIChromaPredV_neon
//stmdb sp!, { r2-r5, lr}
//Get the top row (8 byte)
sub r2, r0, r1
@@ -582,17 +504,10 @@ loop_0_get_i16x16_luma_pred_plane:
vst1.8 {d0}, [r0], r1
vst1.8 {d0}, [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i_chroma_pred_h(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderIChromaPredH_neon
WELS_ASM_FUNC_BEGIN WelsDecoderIChromaPredH_neon
//stmdb sp!, { r2-r5, lr}
////Get the left column (8 byte)
sub r2, r0, #1
@@ -615,17 +530,10 @@ loop_0_get_i16x16_luma_pred_plane:
vst1.8 {d6}, [r0], r1
vst1.8 {d7}, [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i_chroma_pred_dc_both(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
WELS_ASM_FUNC_BEGIN WelsDecoderIChromaPredDC_neon
WELS_ASM_FUNC_BEGIN WelsDecoderIChromaPredDC_neon
//stmdb sp!, { r2-r5, lr}
//Load the left column data (8 bytes)
sub r2, r0, #1
@@ -659,20 +567,15 @@ loop_0_get_i16x16_luma_pred_plane:
vst2.32 {d6[0],d7[0]}, [r0], r1
vst2.32 {d6[0],d7[0]}, [r0]
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
/*
* void get_i_chroma_pred_plane(uint8_t *pred, const int32_t stride);
* r0 --- pred
* r1 --- stride
* return --- void
*/
//Table {{1,2,3,4,1,2,3,4}*17}
CONST0_GET_I_CHROMA_PRED_PLANE: .long 0x44332211, 0x44332211//0x140f0a05, 0x28231e19
//Table {-3,-2,-1,0,1,2,3,4}
CONST1_GET_I_CHROMA_PRED_PLANE: .long 0xfffefffd, 0x0000ffff,0x00020001,0x00040003
WELS_ASM_FUNC_BEGIN WelsDecoderIChromaPredPlane_neon
WELS_ASM_FUNC_BEGIN WelsDecoderIChromaPredPlane_neon
//stmdb sp!, { r2-r5, lr}
//Load the top row data
sub r2, r0, #1
@@ -741,6 +644,6 @@ loop_0_get_i_chroma_pred_plane:
subs r2, #1
bne loop_0_get_i_chroma_pred_plane
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
#endif

375
codec/decoder/core/arm/mc_neon.S
View File

@@ -37,172 +37,168 @@
#ifdef APPLE_IOS
.macro AVERAGE_TWO_8BITS
// { // input:dst_d, src_d A and B; working: q13
vaddl.u8 q13, $2, $1
vrshrn.u16 $0, q13, #1
vaddl.u8 q13, $2, $1
vrshrn.u16 $0, q13, #1
// }
.endm
//h_filter(src) = (src[-2] + src[3]) - 5*(src[-1] + src[2]) + 20*(src[ 0] + src[1]);//
//clip((h_filter(src)+16)>>5)
.macro FILTER_6TAG_8BITS
// { // input:src[-2], src[-1], src[0], src[1], src[2], src[3], dst_d, multiplier a/b; working: q12, q13
vaddl.u8 q12, $0, $5 //q12=src[-2]+src[3]
vaddl.u8 q13, $2, $3 //src[0]+src[1]
vmla.u16 q12, q13, $7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, $1, $4 //src[-1]+src[2]
vmls.s16 q12, q13, $8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 $6, q12, #5
vaddl.u8 q12, $0, $5 //q12=src[-2]+src[3]
vaddl.u8 q13, $2, $3 //src[0]+src[1]
vmla.u16 q12, q13, $7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, $1, $4 //src[-1]+src[2]
vmls.s16 q12, q13, $8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 $6, q12, #5
// }
.endm
.macro FILTER_6TAG_8BITS_AVERAGE_WITH_0
// { // input:src[-2], src[-1], src[0], src[1], src[2], src[3], dst_d, multiplier a/b; working: q12, q13
vaddl.u8 q12, $0, $5 //q12=src[-2]+src[3]
vaddl.u8 q13, $2, $3 //src[0]+src[1]
vmla.u16 q12, q13, $7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, $1, $4 //src[-1]+src[2]
vmls.s16 q12, q13, $8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 $6, q12, #5
vaddl.u8 q13, $2, $6
vrshrn.u16 $6, q13, #1
vaddl.u8 q12, $0, $5 //q12=src[-2]+src[3]
vaddl.u8 q13, $2, $3 //src[0]+src[1]
vmla.u16 q12, q13, $7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, $1, $4 //src[-1]+src[2]
vmls.s16 q12, q13, $8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 $6, q12, #5
vaddl.u8 q13, $2, $6
vrshrn.u16 $6, q13, #1
// }
.endm
.macro FILTER_6TAG_8BITS_AVERAGE_WITH_1
// { // input:src[-2], src[-1], src[0], src[1], src[2], src[3], dst_d, multiplier a/b; working: q12, q13
vaddl.u8 q12, $0, $5 //q12=src[-2]+src[3]
vaddl.u8 q13, $2, $3 //src[0]+src[1]
vmla.u16 q12, q13, $7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, $1, $4 //src[-1]+src[2]
vmls.s16 q12, q13, $8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 $6, q12, #5
vaddl.u8 q13, $3, $6
vrshrn.u16 $6, q13, #1
vaddl.u8 q12, $0, $5 //q12=src[-2]+src[3]
vaddl.u8 q13, $2, $3 //src[0]+src[1]
vmla.u16 q12, q13, $7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, $1, $4 //src[-1]+src[2]
vmls.s16 q12, q13, $8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 $6, q12, #5
vaddl.u8 q13, $3, $6
vrshrn.u16 $6, q13, #1
// }
.endm
.macro FILTER_6TAG_8BITS_TO_16BITS
// { // input:d_src[-2], d_src[-1], d_src[0], d_src[1], d_src[2], d_src[3], dst_q, multiplier a/b; working:q13
vaddl.u8 $6, $0, $5 //dst_q=src[-2]+src[3]
vaddl.u8 q13, $2, $3 //src[0]+src[1]
vmla.u16 $6, q13, $7 //dst_q += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, $1, $4 //src[-1]+src[2]
vmls.s16 $6, q13, $8 //dst_q -= 5*(src[-1]+src[2]), 2 cycles
vaddl.u8 $6, $0, $5 //dst_q=src[-2]+src[3]
vaddl.u8 q13, $2, $3 //src[0]+src[1]
vmla.u16 $6, q13, $7 //dst_q += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, $1, $4 //src[-1]+src[2]
vmls.s16 $6, q13, $8 //dst_q -= 5*(src[-1]+src[2]), 2 cycles
// }
.endm
.macro FILTER_3_IN_16BITS_TO_8BITS
// { // input:a, b, c, dst_d;
vsub.s16 $0, $0, $1 //a-b
vshr.s16 $0, $0, #2 //(a-b)/4
vsub.s16 $0, $0, $1 //(a-b)/4-b
vadd.s16 $0, $0, $2 //(a-b)/4-b+c
vshr.s16 $0, $0, #2 //((a-b)/4-b+c)/4
vadd.s16 $0, $0, $2 //((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16
vqrshrun.s16 $3, $0, #6 //(+32)>>6
vsub.s16 $0, $0, $1 //a-b
vshr.s16 $0, $0, #2 //(a-b)/4
vsub.s16 $0, $0, $1 //(a-b)/4-b
vadd.s16 $0, $0, $2 //(a-b)/4-b+c
vshr.s16 $0, $0, #2 //((a-b)/4-b+c)/4
vadd.s16 $0, $0, $2 //((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16
vqrshrun.s16 $3, $0, #6 //(+32)>>6
// }
.endm
.macro UNPACK_2_16BITS_TO_ABC
// { // input:q_src[-2:5], q_src[6:13](avail 8+5)/q_src[6:**](avail 4+5), dst_a, dst_b, dst_c;
vext.16 $4, $0, $1, #2 //src[0]
vext.16 $3, $0, $1, #3 //src[1]
vadd.s16 $4, $3 //c=src[0]+src[1]
vext.16 $4, $0, $1, #2 //src[0]
vext.16 $3, $0, $1, #3 //src[1]
vadd.s16 $4, $3 //c=src[0]+src[1]
vext.16 $3, $0, $1, #1 //src[-1]
vext.16 $2, $0, $1, #4 //src[2]
vadd.s16 $3, $2 //b=src[-1]+src[2]
vext.16 $3, $0, $1, #1 //src[-1]
vext.16 $2, $0, $1, #4 //src[2]
vadd.s16 $3, $2 //b=src[-1]+src[2]
vext.16 $2, $0, $1, #5 //src[3]
vadd.s16 $2, $0 //a=src[-2]+src[3]
vext.16 $2, $0, $1, #5 //src[3]
vadd.s16 $2, $0 //a=src[-2]+src[3]
// }
.endm
#else
.macro AVERAGE_TWO_8BITS arg0, arg1, arg2
// { // input:dst_d, src_d A and B; working: q13
vaddl.u8 q13, \arg2, \arg1
vrshrn.u16 \arg0, q13, #1
vaddl.u8 q13, \arg2, \arg1
vrshrn.u16 \arg0, q13, #1
// }
.endm
//h_filter(src) = (src[-2] + src[3]) - 5*(src[-1] + src[2]) + 20*(src[ 0] + src[1]);//
//clip((h_filter(src)+16)>>5)
.macro FILTER_6TAG_8BITS arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8
// { // input:src[-2], src[-1], src[0], src[1], src[2], src[3], dst_d, multiplier a/b; working: q12, q13
vaddl.u8 q12, \arg0, \arg5 //q12=src[-2]+src[3]
vaddl.u8 q13, \arg2, \arg3 //src[0]+src[1]
vmla.u16 q12, q13, \arg7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, \arg1, \arg4 //src[-1]+src[2]
vmls.s16 q12, q13, \arg8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 \arg6, q12, #5
vaddl.u8 q12, \arg0, \arg5 //q12=src[-2]+src[3]
vaddl.u8 q13, \arg2, \arg3 //src[0]+src[1]
vmla.u16 q12, q13, \arg7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, \arg1, \arg4 //src[-1]+src[2]
vmls.s16 q12, q13, \arg8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 \arg6, q12, #5
// }
.endm
.macro FILTER_6TAG_8BITS_AVERAGE_WITH_0 arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8
// { // input:src[-2], src[-1], src[0], src[1], src[2], src[3], dst_d, multiplier a/b; working: q12, q13
vaddl.u8 q12, \arg0, \arg5 //q12=src[-2]+src[3]
vaddl.u8 q13, \arg2, \arg3 //src[0]+src[1]
vmla.u16 q12, q13, \arg7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, \arg1, \arg4 //src[-1]+src[2]
vmls.s16 q12, q13, \arg8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 \arg6, q12, #5
vaddl.u8 q13, \arg2, \arg6
vrshrn.u16 \arg6, q13, #1
vaddl.u8 q12, \arg0, \arg5 //q12=src[-2]+src[3]
vaddl.u8 q13, \arg2, \arg3 //src[0]+src[1]
vmla.u16 q12, q13, \arg7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, \arg1, \arg4 //src[-1]+src[2]
vmls.s16 q12, q13, \arg8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 \arg6, q12, #5
vaddl.u8 q13, \arg2, \arg6
vrshrn.u16 \arg6, q13, #1
// }
.endm
.macro FILTER_6TAG_8BITS_AVERAGE_WITH_1 arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8
// { // input:src[-2], src[-1], src[0], src[1], src[2], src[3], dst_d, multiplier a/b; working: q12, q13
vaddl.u8 q12, \arg0, \arg5 //q12=src[-2]+src[3]
vaddl.u8 q13, \arg2, \arg3 //src[0]+src[1]
vmla.u16 q12, q13, \arg7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, \arg1, \arg4 //src[-1]+src[2]
vmls.s16 q12, q13, \arg8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 \arg6, q12, #5
vaddl.u8 q13, \arg3, \arg6
vrshrn.u16 \arg6, q13, #1
vaddl.u8 q12, \arg0, \arg5 //q12=src[-2]+src[3]
vaddl.u8 q13, \arg2, \arg3 //src[0]+src[1]
vmla.u16 q12, q13, \arg7 //q12 += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, \arg1, \arg4 //src[-1]+src[2]
vmls.s16 q12, q13, \arg8 //q12 -= 5*(src[-1]+src[2]), 2 cycles
vqrshrun.s16 \arg6, q12, #5
vaddl.u8 q13, \arg3, \arg6
vrshrn.u16 \arg6, q13, #1
// }
.endm
.macro FILTER_6TAG_8BITS_TO_16BITS arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8
// { // input:d_src[-2], d_src[-1], d_src[0], d_src[1], d_src[2], d_src[3], dst_q, multiplier a/b; working:q13
vaddl.u8 \arg6, \arg0, \arg5 //dst_q=src[-2]+src[3]
vaddl.u8 q13, \arg2, \arg3 //src[0]+src[1]
vmla.u16 \arg6, q13, \arg7 //dst_q += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, \arg1, \arg4 //src[-1]+src[2]
vmls.s16 \arg6, q13, \arg8 //dst_q -= 5*(src[-1]+src[2]), 2 cycles
vaddl.u8 \arg6, \arg0, \arg5 //dst_q=src[-2]+src[3]
vaddl.u8 q13, \arg2, \arg3 //src[0]+src[1]
vmla.u16 \arg6, q13, \arg7 //dst_q += 20*(src[0]+src[1]), 2 cycles
vaddl.u8 q13, \arg1, \arg4 //src[-1]+src[2]
vmls.s16 \arg6, q13, \arg8 //dst_q -= 5*(src[-1]+src[2]), 2 cycles
// }
.endm
.macro FILTER_3_IN_16BITS_TO_8BITS arg0, arg1, arg2, arg3
// { // input:a, b, c, dst_d;
vsub.s16 \arg0, \arg0, \arg1 //a-b
vshr.s16 \arg0, \arg0, #2 //(a-b)/4
vsub.s16 \arg0, \arg0, \arg1 //(a-b)/4-b
vadd.s16 \arg0, \arg0, \arg2 //(a-b)/4-b+c
vshr.s16 \arg0, \arg0, #2 //((a-b)/4-b+c)/4
vadd.s16 \arg0, \arg0, \arg2 //((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16
vqrshrun.s16 \arg3, \arg0, #6 //(+32)>>6
vsub.s16 \arg0, \arg0, \arg1 //a-b
vshr.s16 \arg0, \arg0, #2 //(a-b)/4
vsub.s16 \arg0, \arg0, \arg1 //(a-b)/4-b
vadd.s16 \arg0, \arg0, \arg2 //(a-b)/4-b+c
vshr.s16 \arg0, \arg0, #2 //((a-b)/4-b+c)/4
vadd.s16 \arg0, \arg0, \arg2 //((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16
vqrshrun.s16 \arg3, \arg0, #6 //(+32)>>6
// }
.endm
.macro UNPACK_2_16BITS_TO_ABC arg0, arg1, arg2, arg3, arg4
// { // input:q_src[-2:5], q_src[6:13](avail 8+5)/q_src[6:**](avail 4+5), dst_a, dst_b, dst_c;
vext.16 \arg4, \arg0, \arg1, #2 //src[0]
vext.16 \arg3, \arg0, \arg1, #3 //src[1]
vadd.s16 \arg4, \arg3 //c=src[0]+src[1]
vext.16 \arg4, \arg0, \arg1, #2 //src[0]
vext.16 \arg3, \arg0, \arg1, #3 //src[1]
vadd.s16 \arg4, \arg3 //c=src[0]+src[1]
vext.16 \arg3, \arg0, \arg1, #1 //src[-1]
vext.16 \arg2, \arg0, \arg1, #4 //src[2]
vadd.s16 \arg3,\arg2 //b=src[-1]+src[2]
vext.16 \arg3, \arg0, \arg1, #1 //src[-1]
vext.16 \arg2, \arg0, \arg1, #4 //src[2]
vadd.s16 \arg3,\arg2 //b=src[-1]+src[2]
vext.16 \arg2, \arg0, \arg1, #5 //src[3]
vadd.s16 \arg2, \arg0 //a=src[-2]+src[3]
vext.16 \arg2, \arg0, \arg1, #5 //src[3]
vadd.s16 \arg2, \arg0 //a=src[-2]+src[3]
// }
.endm
#endif
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer20WidthEq16_neon
WELS_ASM_FUNC_BEGIN McHorVer20WidthEq16_neon
push {r4}
ldr r4, [sp, #4]
@@ -231,10 +227,10 @@ w16_h_mc_luma_loop:
cmp r4, #0
bne w16_h_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer20WidthEq8_neon
WELS_ASM_FUNC_BEGIN McHorVer20WidthEq8_neon
push {r4}
ldr r4, [sp, #4]
@@ -260,10 +256,10 @@ w8_h_mc_luma_loop:
cmp r4, #0
bne w8_h_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer20WidthEq4_neon
WELS_ASM_FUNC_BEGIN McHorVer20WidthEq4_neon
push {r4, r5, r6}
ldr r6, [sp, #12]
@@ -298,10 +294,10 @@ w4_h_mc_luma_loop:
bne w4_h_mc_luma_loop
pop {r4, r5, r6}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer10WidthEq16_neon
WELS_ASM_FUNC_BEGIN McHorVer10WidthEq16_neon
push {r4}
ldr r4, [sp, #4]
@@ -330,10 +326,10 @@ w16_xy_10_mc_luma_loop:
cmp r4, #0
bne w16_xy_10_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer10WidthEq8_neon
WELS_ASM_FUNC_BEGIN McHorVer10WidthEq8_neon
push {r4}
ldr r4, [sp, #4]
@@ -359,10 +355,10 @@ w8_xy_10_mc_luma_loop:
cmp r4, #0
bne w8_xy_10_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer10WidthEq4_neon
WELS_ASM_FUNC_BEGIN McHorVer10WidthEq4_neon
push {r4, r5, r6}
ldr r6, [sp, #12]
@@ -397,10 +393,10 @@ w4_xy_10_mc_luma_loop:
bne w4_xy_10_mc_luma_loop
pop {r4, r5, r6}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer30WidthEq16_neon
WELS_ASM_FUNC_BEGIN McHorVer30WidthEq16_neon
push {r4}
ldr r4, [sp, #4]
@@ -429,10 +425,10 @@ w16_xy_30_mc_luma_loop:
cmp r4, #0
bne w16_xy_30_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer30WidthEq8_neon
WELS_ASM_FUNC_BEGIN McHorVer30WidthEq8_neon
push {r4}
ldr r4, [sp, #4]
@@ -458,10 +454,10 @@ w8_xy_30_mc_luma_loop:
cmp r4, #0
bne w8_xy_30_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer30WidthEq4_neon
WELS_ASM_FUNC_BEGIN McHorVer30WidthEq4_neon
push {r4, r5, r6}
ldr r6, [sp, #12]
@@ -496,10 +492,10 @@ w4_xy_30_mc_luma_loop:
bne w4_xy_30_mc_luma_loop
pop {r4, r5, r6}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer01WidthEq16_neon
WELS_ASM_FUNC_BEGIN McHorVer01WidthEq16_neon
push {r4}
ldr r4, [sp, #4]
@@ -578,9 +574,10 @@ w16_xy_01_luma_loop:
cmp r4, #0
bne w16_xy_01_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN McHorVer01WidthEq8_neon
WELS_ASM_FUNC_BEGIN McHorVer01WidthEq8_neon
push {r4}
ldr r4, [sp, #4]
@@ -631,9 +628,10 @@ w8_xy_01_mc_luma_loop:
bne w8_xy_01_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN McHorVer01WidthEq4_neon
WELS_ASM_FUNC_BEGIN McHorVer01WidthEq4_neon
push {r4, r5, r6, r7}
sub r0, r1, lsl #1 //src[-2*src_stride]
pld [r0]
@@ -689,9 +687,10 @@ w4_xy_01_mc_luma_loop:
bne w4_xy_01_mc_luma_loop
pop {r4, r5, r6, r7}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN McHorVer03WidthEq16_neon
WELS_ASM_FUNC_BEGIN McHorVer03WidthEq16_neon
push {r4}
ldr r4, [sp, #4]
@@ -770,9 +769,10 @@ w16_xy_03_luma_loop:
cmp r4, #0
bne w16_xy_03_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN McHorVer03WidthEq8_neon
WELS_ASM_FUNC_BEGIN McHorVer03WidthEq8_neon
push {r4}
ldr r4, [sp, #4]
@@ -823,9 +823,10 @@ w8_xy_03_mc_luma_loop:
bne w8_xy_03_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN McHorVer03WidthEq4_neon
WELS_ASM_FUNC_BEGIN McHorVer03WidthEq4_neon
push {r4, r5, r6, r7}
sub r0, r1, lsl #1 //src[-2*src_stride]
pld [r0]
@@ -881,10 +882,10 @@ w4_xy_03_mc_luma_loop:
bne w4_xy_03_mc_luma_loop
pop {r4, r5, r6, r7}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer02WidthEq16_neon
WELS_ASM_FUNC_BEGIN McHorVer02WidthEq16_neon
push {r4}
ldr r4, [sp, #4]
@@ -963,10 +964,10 @@ w16_v_mc_luma_loop:
cmp r4, #0
bne w16_v_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer02WidthEq8_neon
WELS_ASM_FUNC_BEGIN McHorVer02WidthEq8_neon
push {r4}
ldr r4, [sp, #4]
@@ -1017,10 +1018,10 @@ w8_v_mc_luma_loop:
bne w8_v_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer02WidthEq4_neon
WELS_ASM_FUNC_BEGIN McHorVer02WidthEq4_neon
push {r4, r5, r6, r7}
sub r0, r1, lsl #1 //src[-2*src_stride]
pld [r0]
@@ -1076,10 +1077,10 @@ w4_v_mc_luma_loop:
bne w4_v_mc_luma_loop
pop {r4, r5, r6, r7}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer22WidthEq16_neon
WELS_ASM_FUNC_BEGIN McHorVer22WidthEq16_neon
push {r4}
ldr r4, [sp, #4]
@@ -1089,8 +1090,6 @@ w4_v_mc_luma_loop:
pld [r0, r1]
vmov.u16 q14, #0x0014 // 20
// vmov.u32 d30, #0x00140000
// vorr.u32 d30, #0x0005 //0x0014 0005 0014 0005
vld1.u8 {d0-d2}, [r0], r1 //use 21(16+5), =src[-2]
vld1.u8 {d3-d5}, [r0], r1 //use 21(16+5), =src[-1]
@@ -1115,7 +1114,7 @@ w16_hv_mc_luma_loop:
// horizon filtered
UNPACK_2_16BITS_TO_ABC q9, q10, q11, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q11, q12, q13, d0 //output to q0[0]
// vst1.u8 d18, [r2] //write 8Byte
// vertical filtered into q10/q11
FILTER_6TAG_8BITS_TO_16BITS d2, d5, d8,d11, d14, d17,q11, q14, q15 // only 5 avail
// horizon filtered
@@ -1133,13 +1132,13 @@ w16_hv_mc_luma_loop:
// horizon filtered
UNPACK_2_16BITS_TO_ABC q9, q10, q11, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q11, q12, q13, d3 //output to d3
// vst1.u8 d18, [r2]! //write 8Byte
// vertical filtered into q10/q11
FILTER_6TAG_8BITS_TO_16BITS d5, d8,d11, d14, d17, d2,q11, q14, q15 // only 5 avail
// horizon filtered
UNPACK_2_16BITS_TO_ABC q10, q11, q9, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q9, q12, q13, d4 //output to d4
// vst1.u8 d20, [r2]! //write 8Byte
vst1.u8 {d3, d4}, [r2], r3 //write 16Byte
vld1.u8 {d3-d5}, [r0], r1 //read 3rd row
@@ -1151,13 +1150,12 @@ w16_hv_mc_luma_loop:
// horizon filtered
UNPACK_2_16BITS_TO_ABC q9, q10, q11, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q11, q12, q13, d6 //output to d6
// vst1.u8 d18, [r2]! //write 8Byte
// vertical filtered into q10/q11
FILTER_6TAG_8BITS_TO_16BITS d8,d11, d14, d17, d2, d5,q11, q14, q15 // only 5 avail
// horizon filtered
UNPACK_2_16BITS_TO_ABC q10, q11, q9, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q9, q12, q13, d7 //output to d7
// vst1.u8 d20, [r2]! //write 8Byte
vst1.u8 {d6, d7}, [r2], r3 //write 16Byte
vld1.u8 {d6-d8}, [r0], r1 //read 4th row
@@ -1169,13 +1167,11 @@ w16_hv_mc_luma_loop:
// horizon filtered
UNPACK_2_16BITS_TO_ABC q9, q10, q11, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q11, q12, q13, d9 //output to d9
// vst1.u8 d18, [r2]! //write 8Byte
// vertical filtered into q10/q11
FILTER_6TAG_8BITS_TO_16BITS d11, d14, d17, d2, d5, d8,q11, q14, q15 // only 5 avail
// horizon filtered
UNPACK_2_16BITS_TO_ABC q10, q11, q9, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q9, q12, q13, d10 //output to d10
// vst1.u8 d20, [r2]! //write 8Byte
vst1.u8 {d9, d10}, [r2], r3 //write 16Byte
//d12~d17(q6~q8), d0~d8(q0~q3+d8), --> d0~d14
@@ -1193,10 +1189,10 @@ w16_hv_mc_luma_loop:
cmp r4, #0
bne w16_hv_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer22WidthEq8_neon
WELS_ASM_FUNC_BEGIN McHorVer22WidthEq8_neon
push {r4}
ldr r4, [sp, #4]
@@ -1231,7 +1227,6 @@ w8_hv_mc_luma_loop:
UNPACK_2_16BITS_TO_ABC q6, q7, q11, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q11, q12, q13, d12 //output to q6[0]
vst1.u8 d12, [r2], r3 //write 8Byte
// add r2, #8
vld1.u8 {q0}, [r0], r1 //read 2nd row
//the 2nd row
@@ -1243,7 +1238,6 @@ w8_hv_mc_luma_loop:
UNPACK_2_16BITS_TO_ABC q6, q7, q11, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q11, q12, q13, d12 //output to q6[0]
vst1.u8 d12, [r2], r3 //write 8Byte
// add r2, #8
vld1.u8 {q1}, [r0], r1 //read 3rd row
//the 3rd row
@@ -1255,7 +1249,6 @@ w8_hv_mc_luma_loop:
UNPACK_2_16BITS_TO_ABC q6, q7, q11, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q11, q12, q13, d12 //output to q6[0]
vst1.u8 d12, [r2], r3 //write 8Byte
// add r2, #8
vld1.u8 {q2}, [r0], r1 //read 4th row
//the 4th row
@@ -1267,7 +1260,6 @@ w8_hv_mc_luma_loop:
UNPACK_2_16BITS_TO_ABC q6, q7, q11, q12, q13
FILTER_3_IN_16BITS_TO_8BITS q11, q12, q13, d12 //output to q6[0]
vst1.u8 d12, [r2], r3 //write 8Byte
// add r2, #8
//q4~q5, q0~q2, --> q0~q4
vswp q0, q4
@@ -1279,10 +1271,10 @@ w8_hv_mc_luma_loop:
cmp r4, #0
bne w8_hv_mc_luma_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McHorVer22WidthEq4_neon
WELS_ASM_FUNC_BEGIN McHorVer22WidthEq4_neon
push {r4 ,r5, r6}
ldr r6, [sp, #12]
@@ -1331,7 +1323,6 @@ w4_hv_mc_luma_loop:
vmov r4 ,r5, d22
str r4, [r2], r3 //write 4Byte
str r5, [r2], r3 //write 4Byte
// add r2, #32
//the 3rd&4th row
vld1.u8 {q0}, [r0], r1 //use 9(4+5), =src[3]
@@ -1356,7 +1347,6 @@ w4_hv_mc_luma_loop:
vmov r4 ,r5, d22
str r4, [r2], r3 //write 4Byte
str r5, [r2], r3 //write 4Byte
// add r2, #32
//q4~q6, q0~q1, --> q0~q4
vswp q4, q0
@@ -1370,10 +1360,10 @@ w4_hv_mc_luma_loop:
bne w4_hv_mc_luma_loop
pop {r4, r5, r6}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McCopyWidthEq16_neon
WELS_ASM_FUNC_BEGIN McCopyWidthEq16_neon
push {r4}
ldr r4, [sp, #4]
w16_copy_loop:
@@ -1386,10 +1376,10 @@ w16_copy_loop:
bne w16_copy_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McCopyWidthEq8_neon
WELS_ASM_FUNC_BEGIN McCopyWidthEq8_neon
push {r4}
ldr r4, [sp, #4]
w8_copy_loop:
@@ -1402,10 +1392,10 @@ w8_copy_loop:
bne w8_copy_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN McCopyWidthEq4_neon
WELS_ASM_FUNC_BEGIN McCopyWidthEq4_neon
push {r4, r5, r6}
ldr r4, [sp, #12]
w4_copy_loop:
@@ -1419,10 +1409,10 @@ w4_copy_loop:
bne w4_copy_loop
pop {r4, r5, r6}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* dst, int32_t dst_stride, uint8_t* srcA, uint8_t* srcB, int32_t height
WELS_ASM_FUNC_BEGIN PixelAvgWidthEq16_neon
WELS_ASM_FUNC_BEGIN PixelAvgWidthEq16_neon
push {r4}
ldr r4, [sp, #4]
w16_pix_avg_loop:
@@ -1457,9 +1447,10 @@ w16_pix_avg_loop:
bne w16_pix_avg_loop
pop {r4}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN PixelAvgWidthEq8_neon
WELS_ASM_FUNC_BEGIN PixelAvgWidthEq8_neon
push {r4, r5}
ldr r4, [sp, #8]
mov r5, #16
@@ -1469,8 +1460,6 @@ w8_pix_avg_loop:
vld1.u8 {d2}, [r3], r5
vld1.u8 {d1}, [r2], r5
vld1.u8 {d3}, [r3], r5
// add r2, #32
// add r3, #32
AVERAGE_TWO_8BITS d0, d0, d2
AVERAGE_TWO_8BITS d1, d1, d3
@@ -1481,8 +1470,6 @@ w8_pix_avg_loop:
vld1.u8 {d6}, [r3], r5
vld1.u8 {d5}, [r2], r5
vld1.u8 {d7}, [r3], r5
// add r2, #32
// add r3, #32
AVERAGE_TWO_8BITS d4, d4, d6
AVERAGE_TWO_8BITS d5, d5, d7
@@ -1494,10 +1481,10 @@ w8_pix_avg_loop:
bne w8_pix_avg_loop
pop {r4, r5}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t height
WELS_ASM_FUNC_BEGIN PixelAvgWidthEq4_neon
WELS_ASM_FUNC_BEGIN PixelAvgWidthEq4_neon
push {r4-r8}
ldr r4, [sp, #20]
w4_pix_avg_loop:
@@ -1522,15 +1509,9 @@ w4_pix_avg_loop:
bne w4_pix_avg_loop
pop {r4-r8}
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t* weights, int32_t height
//cA = (8 - dx) * (8 - dy);
//cB = dx * (8 - dy);
//cC = (8 - dx) * dy;
//cD = dx * dy
WELS_ASM_FUNC_BEGIN McChromaWidthEq8_neon
WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN McChromaWidthEq8_neon
push {r4, r5}
ldr r4, [sp, #8]
ldr r5, [sp, #12]
@@ -1572,10 +1553,10 @@ w8_mc_chroma_loop: // each two pxl row
bne w8_mc_chroma_loop
pop {r4, r5}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
//uint8_t* src, int32_t src_stride, uint8_t* dst, int32_t dst_stride, int32_t* weights, int32_t height
WELS_ASM_FUNC_BEGIN McChromaWidthEq4_neon
WELS_ASM_FUNC_BEGIN McChromaWidthEq4_neon
push {r4, r5, r6}
ldr r4, [sp, #12]
@@ -1617,5 +1598,5 @@ w4_mc_chroma_loop: // each two pxl row
bne w4_mc_chroma_loop
pop {r4, r5, r6}
WELS_ASM_FUNC_END
WELS_ASM_FUNC_END
#endif

2
codec/decoder/core/src/deblocking.cpp
View File

@@ -720,7 +720,7 @@ void DeblockingInit (SDeblockingFunc* pFunc, int32_t iCpu) {
#endif
#if defined(HAVE_NEON)
if ( iCpu & WELS_CPU_NEON )
if ( iCpu & WELS_CPU_NEON )
{
pFunc->pfLumaDeblockingLT4Ver = DeblockLumaLt4V_neon;
pFunc->pfLumaDeblockingEQ4Ver = DeblockLumaEq4V_neon;

4
codec/decoder/core/src/decode_slice.cpp
View File

@@ -1152,8 +1152,8 @@ void WelsBlockFuncInit (SBlockFunc* pFunc, int32_t iCpu) {
#ifdef HAVE_NEON
if ( iCpu & WELS_CPU_NEON ) {
pFunc->pWelsBlockZero16x16Func = WelsResBlockZero16x16_neon;
pFunc->pWelsBlockZero8x8Func = WelsResBlockZero8x8_neon;
pFunc->pWelsSetNonZeroCountFunc = SetNonZeroCount_neon;
pFunc->pWelsBlockZero8x8Func = WelsResBlockZero8x8_neon;
pFunc->pWelsSetNonZeroCountFunc = SetNonZeroCount_neon;
}
#endif
}

34
codec/decoder/core/src/decoder.cpp
View File

@@ -667,25 +667,25 @@ void AssignFuncPointerForRec (PWelsDecoderContext pCtx) {
if ( pCtx->uiCpuFlag & WELS_CPU_NEON ) {
pCtx->pIdctResAddPredFunc = IdctResAddPred_neon;
pCtx->pGetI16x16LumaPredFunc[I16_PRED_DC] = WelsDecoderI16x16LumaPredDc_neon;
pCtx->pGetI16x16LumaPredFunc[I16_PRED_P] = WelsDecoderI16x16LumaPredPlane_neon;
pCtx->pGetI16x16LumaPredFunc[I16_PRED_H] = WelsDecoderI16x16LumaPredH_neon;
pCtx->pGetI16x16LumaPredFunc[I16_PRED_V] = WelsDecoderI16x16LumaPredV_neon;
pCtx->pGetI16x16LumaPredFunc[I16_PRED_DC] = WelsDecoderI16x16LumaPredDc_neon;
pCtx->pGetI16x16LumaPredFunc[I16_PRED_P] = WelsDecoderI16x16LumaPredPlane_neon;
pCtx->pGetI16x16LumaPredFunc[I16_PRED_H] = WelsDecoderI16x16LumaPredH_neon;
pCtx->pGetI16x16LumaPredFunc[I16_PRED_V] = WelsDecoderI16x16LumaPredV_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_V ] = WelsDecoderI4x4LumaPredV_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_H ] = WelsDecoderI4x4LumaPredH_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_DDL ] = WelsDecoderI4x4LumaPredDDL_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_DDR ] = WelsDecoderI4x4LumaPredDDR_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_VL ] = WelsDecoderI4x4LumaPredVL_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_VR ] = WelsDecoderI4x4LumaPredVR_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_HU ] = WelsDecoderI4x4LumaPredHU_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_HD ] = WelsDecoderI4x4LumaPredHD_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_V ] = WelsDecoderI4x4LumaPredV_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_H ] = WelsDecoderI4x4LumaPredH_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_DDL ] = WelsDecoderI4x4LumaPredDDL_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_DDR ] = WelsDecoderI4x4LumaPredDDR_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_VL ] = WelsDecoderI4x4LumaPredVL_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_VR ] = WelsDecoderI4x4LumaPredVR_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_HU ] = WelsDecoderI4x4LumaPredHU_neon;
pCtx->pGetI4x4LumaPredFunc[I4_PRED_HD ] = WelsDecoderI4x4LumaPredHD_neon;
pCtx->pGetIChromaPredFunc[C_PRED_H] = WelsDecoderIChromaPredH_neon;
pCtx->pGetIChromaPredFunc[C_PRED_V] = WelsDecoderIChromaPredV_neon;
pCtx->pGetIChromaPredFunc[C_PRED_P ] = WelsDecoderIChromaPredPlane_neon;
pCtx->pGetIChromaPredFunc[C_PRED_DC] = WelsDecoderIChromaPredDC_neon;
}
pCtx->pGetIChromaPredFunc[C_PRED_H] = WelsDecoderIChromaPredH_neon;
pCtx->pGetIChromaPredFunc[C_PRED_V] = WelsDecoderIChromaPredV_neon;
pCtx->pGetIChromaPredFunc[C_PRED_P ] = WelsDecoderIChromaPredPlane_neon;
pCtx->pGetIChromaPredFunc[C_PRED_DC] = WelsDecoderIChromaPredDC_neon;
}
#endif//HAVE_NEON

6
codec/decoder/core/src/mc.cpp
View File

@@ -669,8 +669,8 @@ void McCopy_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t
void McHorVer20_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride,
int32_t iWidth, int32_t iHeight)
{
if (iWidth == 16)
McHorVer20WidthEq16_neon(pSrc, iSrcStride, pDst, iDstStride, iHeight);
if (iWidth == 16)
McHorVer20WidthEq16_neon(pSrc, iSrcStride, pDst, iDstStride, iHeight);
else if (iWidth == 8)
McHorVer20WidthEq8_neon(pSrc, iSrcStride, pDst, iDstStride, iHeight);
else if (iWidth == 4)
@@ -690,7 +690,7 @@ void McHorVer22_neon(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int
int32_t iWidth, int32_t iHeight)
{
if (iWidth == 16)
McHorVer22WidthEq16_neon(pSrc, iSrcStride, pDst, iDstStride, iHeight);
McHorVer22WidthEq16_neon(pSrc, iSrcStride, pDst, iDstStride, iHeight);
else if (iWidth == 8)
McHorVer22WidthEq8_neon(pSrc, iSrcStride, pDst, iDstStride, iHeight);
else if (iWidth == 4)