The second stride is always the internal buffer one, MAX_PB_SIZE (times 2 to
get the value in bytes).
---
libavcodec/x86/hevc_mc.asm | 30 +++++++++++++++---------------
libavcodec/x86/hevcdsp.h | 4 ++--
libavcodec/x86/hevcdsp_init.c | 16 ++++++++--------
3 files changed, 25 insertions(+), 25 deletions(-)
diff --git a/libavcodec/x86/hevc_mc.asm b/libavcodec/x86/hevc_mc.asm
index 07b0f77..4776eed 100644
--- a/libavcodec/x86/hevc_mc.asm
+++ b/libavcodec/x86/hevc_mc.asm
@@ -1416,11 +1416,11 @@ cglobal hevc_put_hevc_bi_qpel_hv%1_%2, 8, 10, 16, dst,
dststride, src, srcstride
%macro WEIGHTING_FUNCS 2
%if WIN64 || ARCH_X86_32
-cglobal hevc_put_hevc_uni_w%1_%2, 4, 5, 7, dst, dststride, src, srcstride,
height, denom, wx, ox
+cglobal hevc_put_hevc_uni_w%1_%2, 4, 5, 7, dst, dststride, src, height, denom,
wx, ox
mov r4d, denomm
%define SHIFT r4d
%else
-cglobal hevc_put_hevc_uni_w%1_%2, 6, 6, 7, dst, dststride, src, srcstride,
height, denom, wx, ox
+cglobal hevc_put_hevc_uni_w%1_%2, 6, 6, 7, dst, dststride, src, height, denom,
wx, ox
%define SHIFT denomd
%endif
lea SHIFT, [SHIFT+14-%2] ; shift = 14 - bitd + denom
@@ -1481,15 +1481,15 @@ cglobal hevc_put_hevc_uni_w%1_%2, 6, 6, 7, dst,
dststride, src, srcstride, heigh
jnz .loop ; height loop
RET
-cglobal hevc_put_hevc_bi_w%1_%2, 5, 7, 10, dst, dststride, src, srcstride,
src2, height, denom, wx0, wx1, ox0, ox1
- mov r6d, denomm
+cglobal hevc_put_hevc_bi_w%1_%2, 4, 5, 10, dst, dststride, src, src2, height,
denom, wx0, wx1, ox0, ox1
+ mov r5d, denomm
%if %1 <= 4
pxor m1, m1
%endif
movd m2, wx0m ; WX0
- lea r6d, [r6d+14-%2] ; shift = 14 - bitd + denom
+ lea r5d, [r5d+14-%2] ; shift = 14 - bitd + denom
movd m3, wx1m ; WX1
- movd m0, r6d ; shift
+ movd m0, r5d ; shift
%if %1 <= 4
punpcklwd m2, m1
punpcklwd m3, m1
@@ -1497,19 +1497,19 @@ cglobal hevc_put_hevc_bi_w%1_%2, 5, 7, 10, dst,
dststride, src, srcstride, src2,
punpcklwd m2, m2
punpcklwd m3, m3
%endif
- inc r6d
- movd m5, r6d ; shift+1
+ inc r5d
+ movd m5, r5d ; shift+1
pshufd m2, m2, 0
- mov r6d, ox0m
+ mov r5d, ox0m
pshufd m3, m3, 0
- add r6d, ox1m
+ add r5d, ox1m
%if %2 != 8
- shl r6d, %2-8 ; ox << (bitd - 8)
+ shl r5d, %2-8 ; ox << (bitd - 8)
%endif
- inc r6d
- movd m4, r6d ; offset
+ inc r5d
+ movd m4, r5d ; offset
pshufd m4, m4, 0
- mov r6d, heightm
+ mov r4d, heightm
pslld m4, m0
.loop
@@ -1549,7 +1549,7 @@ cglobal hevc_put_hevc_bi_w%1_%2, 5, 7, 10, dst,
dststride, src, srcstride, src2,
add dstq, dststrideq ; dst += dststride
add srcq, 2*MAX_PB_SIZE ; src += srcstride
add src2q, 2*MAX_PB_SIZE ; src2 += srcstride
- dec r6d ; cmp height
+ dec r4d ; cmp height
jnz .loop ; height loop
RET
%endmacro
diff --git a/libavcodec/x86/hevcdsp.h b/libavcodec/x86/hevcdsp.h
index 7864163..ad8168f 100644
--- a/libavcodec/x86/hevcdsp.h
+++ b/libavcodec/x86/hevcdsp.h
@@ -74,8 +74,8 @@ void ff_hevc_put_hevc_bi_w_ ## name ## _ ## D ##
_##opt(uint8_t *_dst, ptrdiff_t
PEL_PROTOTYPE(fname##64, bitd, opt)
#define WEIGHTING_PROTOTYPE(width, bitd, opt) \
-void ff_hevc_put_hevc_uni_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t
dststride, int16_t *_src, ptrdiff_t _srcstride, int height, int denom, int
_wx, int _ox); \
-void ff_hevc_put_hevc_bi_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t
dststride, int16_t *_src, ptrdiff_t _srcstride, int16_t *_src2, int height, int
denom, int _wx0, int _wx1, int _ox0, int _ox1)
+void ff_hevc_put_hevc_uni_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t
dststride, int16_t *_src, int height, int denom, int _wx, int _ox); \
+void ff_hevc_put_hevc_bi_w##width##_##bitd##_##opt(uint8_t *dst, ptrdiff_t
dststride, int16_t *_src, int16_t *_src2, int height, int denom, int _wx0,
int _wx1, int _ox0, int _ox1)
#define WEIGHTING_PROTOTYPES(bitd, opt) \
WEIGHTING_PROTOTYPE(2, bitd, opt); \
diff --git a/libavcodec/x86/hevcdsp_init.c b/libavcodec/x86/hevcdsp_init.c
index f7b3d0f..e493033 100644
--- a/libavcodec/x86/hevcdsp_init.c
+++ b/libavcodec/x86/hevcdsp_init.c
@@ -427,7 +427,7 @@ mc_rep_funcs(qpel_hv,12, 8, 16, sse4);
mc_rep_funcs(qpel_hv,12, 4, 12, sse4);
#define mc_rep_uni_w(bitd, step, W, opt) \
-void ff_hevc_put_hevc_uni_w##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t
dststride, int16_t *_src, ptrdiff_t _srcstride,\
+void ff_hevc_put_hevc_uni_w##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t
dststride, int16_t *_src, \
int height, int denom, int
_wx, int _ox) \
{
\
int i;
\
@@ -436,7 +436,7 @@ void ff_hevc_put_hevc_uni_w##W##_##bitd##_##opt(uint8_t
*_dst, ptrdiff_t dststri
for (i = 0; i < W; i += step) {
\
src= _src + i;
\
dst= _dst + (i * ((bitd + 7) / 8));
\
- ff_hevc_put_hevc_uni_w##step##_##bitd##_##opt(dst, dststride, src,
_srcstride, \
+ ff_hevc_put_hevc_uni_w##step##_##bitd##_##opt(dst, dststride, src,
\
height, denom, _wx, _ox);
\
}
\
}
@@ -463,7 +463,7 @@ mc_rep_uni_w(12, 8, 48, sse4);
mc_rep_uni_w(12, 8, 64, sse4);
#define mc_rep_bi_w(bitd, step, W, opt) \
-void ff_hevc_put_hevc_bi_w##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t
dststride, int16_t *_src, ptrdiff_t _srcstride, \
+void ff_hevc_put_hevc_bi_w##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t
dststride, int16_t *_src, \
int16_t *_src2, int height,
\
int denom, int _wx0, int _wx1,
int _ox0, int _ox1) \
{
\
@@ -475,8 +475,8 @@ void ff_hevc_put_hevc_bi_w##W##_##bitd##_##opt(uint8_t
*_dst, ptrdiff_t dststrid
src = _src + i;
\
src2 = _src2 + i;
\
dst = _dst + (i * ((bitd + 7) / 8));
\
- ff_hevc_put_hevc_bi_w##step##_##bitd##_##opt(dst, dststride, src,
_srcstride, src2, \
- height, denom, _wx0, _wx1,
_ox0, _ox1); \
+ ff_hevc_put_hevc_bi_w##step##_##bitd##_##opt(dst, dststride, src,
src2, \
+ height, denom, _wx0,
_wx1, _ox0, _ox1); \
}
\
}
@@ -510,7 +510,7 @@ void
ff_hevc_put_hevc_uni_w_##name##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t
{
\
LOCAL_ALIGNED_16(int16_t, temp, [71 * MAX_PB_SIZE]);
\
ff_hevc_put_hevc_##name##W##_##bitd##_##opt(temp, _src, _srcstride,
height, mx, my, width); \
- ff_hevc_put_hevc_uni_w##W##_##bitd##_##opt(_dst, _dststride, temp,
MAX_PB_SIZE, height, denom, _wx, _ox);\
+ ff_hevc_put_hevc_uni_w##W##_##bitd##_##opt(_dst, _dststride, temp, height,
denom, _wx, _ox);\
}
#define mc_uni_w_funcs(name, bitd, opt) \
@@ -569,8 +569,8 @@ void
ff_hevc_put_hevc_bi_w_##name##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t _
{
\
LOCAL_ALIGNED_16(int16_t, temp, [71 * MAX_PB_SIZE]);
\
ff_hevc_put_hevc_##name##W##_##bitd##_##opt(temp, _src, _srcstride,
height, mx, my, width); \
- ff_hevc_put_hevc_bi_w##W##_##bitd##_##opt(_dst, _dststride, temp,
MAX_PB_SIZE, _src2, \
- height, denom, _wx0, _wx1, _ox0,
_ox1); \
+ ff_hevc_put_hevc_bi_w##W##_##bitd##_##opt(_dst, _dststride, temp, _src2,
\
+ height, denom, _wx0, _wx1, _ox0,
_ox1); \
}
#define mc_bi_w_funcs(name, bitd, opt) \
--
1.9.2.msysgit.0
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
