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

Merge pull request #1914 from mstorsjo/asm-cleanup

Browse Source 
Clean up assembly source files
This commit is contained in:
zhilwang 2015-04-27 17:54:38 -07:00
commit 5cf163449f
5 changed files with 1034 additions and 1034 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
codec/common/arm64/mc_aarch64_neon.S
View File

@ -205,7 +205,7 @@ filter_para: .short 0, 1, -5, 20, 0, 0, 0, 0
sqrshrun \arg0\().2s, \arg0\().2d, #10
uqxtn \arg0\().4h, \arg0\().4s
uqxtn \arg0\().8b, \arg0\().8h
// }
// }
.endm
//(const uint8_t* pSrc {x0}, int32_t iSrcStride{x1}, uint8_t* pDst{x2}, int32_t iDstStride{x3}, int32_t iHeight{x4})

10
codec/decoder/core/x86/dct.asm
View File

@ -114,8 +114,8 @@ WELS_EXTERN IdctResAddPred_mmx
emms
ret
;void WelsBlockZero16x16_sse2(int16_t * block, int32_t stride);
WELS_EXTERN WelsBlockZero16x16_sse2
;void WelsBlockZero16x16_sse2(int16_t * block, int32_t stride);
WELS_EXTERN WelsBlockZero16x16_sse2
%assign push_num 0
LOAD_2_PARA
SIGN_EXTENSION r1, r1d
@ -127,9 +127,9 @@ WELS_EXTERN WelsBlockZero16x16_sse2
add r0, r1
%endrep
ret
;void WelsBlockZero8x8_sse2(int16_t * block, int32_t stride);
WELS_EXTERN WelsBlockZero8x8_sse2
;void WelsBlockZero8x8_sse2(int16_t * block, int32_t stride);
WELS_EXTERN WelsBlockZero8x8_sse2
%assign push_num 0
LOAD_2_PARA
SIGN_EXTENSION r1, r1d

18
codec/encoder/core/arm/svc_motion_estimation.S
View File

@ -72,8 +72,8 @@ WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN SumOf8x8BlockOfFrame_neon
//(uint8_t* pRefPicture, const int32_t kiWidth, const int32_t kiHeight,const int32_t kiRefStride,uint16_t* pFeatureOfBlock, uint32_t pTimesOfFeatureValue[])
stmdb sp!, {r4-r12}
ldr r5, [sp, #40] //pTimesOfFeatureValue
ldr r4, [sp, #36] //pFeatureOfBlock
ldr r5, [sp, #40] //pTimesOfFeatureValue
ldr r4, [sp, #36] //pFeatureOfBlock
mov r8, r0
mov r6, r1
@ -158,8 +158,8 @@ WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN SumOf16x16BlockOfFrame_neon
//(uint8_t* pRefPicture, const int32_t kiWidth, const int32_t kiHeight,const int32_t kiRefStride,uint16_t* pFeatureOfBlock, uint32_t pTimesOfFeatureValue[])
stmdb sp!, {r4-r12}
ldr r5, [sp, #40] //pTimesOfFeatureValue
ldr r4, [sp, #36] //pFeatureOfBlock
ldr r5, [sp, #40] //pTimesOfFeatureValue
ldr r4, [sp, #36] //pFeatureOfBlock
mov r8, r0
mov r6, r1
@ -238,7 +238,7 @@ WELS_ASM_FUNC_END
WELS_ASM_FUNC_BEGIN InitializeHashforFeature_neon
// (uint32_t* pTimesOfFeatureValue, uint16_t* pBuf, const int32_t kiListSize, uint16_t** pLocationOfFeature, uint16_t** pFeatureValuePointerList);
stmdb sp!, {r4-r7}
ldr r4, [sp, #16] //pFeatureValuePointerList
ldr r4, [sp, #16] //pFeatureValuePointerList
bic r5, r2, #3
_hash_assign_loop_x4:
vld1.64 {q0}, [r0]!
@ -272,8 +272,8 @@ _hash_assign_with_copy_x4:
vst1.64 {q2}, [r4]!
_assign_next:
subs r5, r5, #4
bne _hash_assign_loop_x4
subs r5, r5, #4
bne _hash_assign_loop_x4
and r5, r2, #3
cmp r5, #0
@ -299,7 +299,7 @@ mx_x_offset_x4: .short 0x00, 0x04, 0x08, 0x0c, 0x00, 0x00, 0x00, 0x00
WELS_ASM_FUNC_BEGIN FillQpelLocationByFeatureValue_neon
// void (uint16_t* pFeatureOfBlock, const int32_t kiWidth, const int32_t kiHeight, uint16_t** pFeatureValuePointerList)
stmdb sp!, {r4-r8}
vpush {q4-q7}
vpush {q4-q7}
adr r7, mv_x_inc_x4
vld1.64 {q7}, [r7]
adr r7, mv_y_inc_x4
@ -360,7 +360,7 @@ _hash_width_loop:
subs r2, #1
bne _hash_height_loop
vpop {q4-q7}
vpop {q4-q7}
ldmia sp!, {r4-r8}
WELS_ASM_FUNC_END
#endif

4
codec/encoder/core/arm64/svc_motion_estimation_aarch64_neon.S
View File

@ -251,8 +251,8 @@ _hash_assign_with_copy_x4:
st1 {v2.16b, v3.16b}, [x4], #32
_assign_next:
subs x5, x5, #4
cbnz x5, _hash_assign_loop_x4
subs x5, x5, #4
cbnz x5, _hash_assign_loop_x4
and x5, x2, x9
cbz x5, _hash_assign_end

2034
codec/encoder/core/x86/sample_sc.asm
View File

File diff suppressed because it is too large Load Diff