[FFmpeg-devel] [PATCH 5/5] x86: hevc_mc: reimplement SSE4 versions for WP
Christophe Gisquet
christophe.gisquet at gmail.com
Sun Aug 24 10:46:34 CEST 2014
*pel values below are decicyle timings, hevc* are object sizes
for Win64.
qpel epel hevc_mc hevcdsp_init
SSSE3: 68702 14469 116624 185388
SSE4: 65907 13811 121312 300996
Note: without 10 and 12bits, the object sizes are respectively:
hevc_mc hevcdsp_init
SSSE3: 39920 59028
SSE4: 41408 97948
---
libavcodec/x86/hevc_mc.asm | 16 ++
libavcodec/x86/hevcdsp.h | 4 +
libavcodec/x86/hevcdsp_init.c | 384 +++++++++++++++++++++++++++++++-----------
3 files changed, 307 insertions(+), 97 deletions(-)
diff --git a/libavcodec/x86/hevc_mc.asm b/libavcodec/x86/hevc_mc.asm
index 52cc66e..fe46d15 100644
--- a/libavcodec/x86/hevc_mc.asm
+++ b/libavcodec/x86/hevc_mc.asm
@@ -1323,6 +1323,22 @@ cglobal hevc_put_hevc_bi_w%1_%2, 5, 8, 10, dst, dststride, src, srcstride, src2,
RET
%endmacro
+INIT_XMM sse4
+WEIGHTING_FUNCS 2, 8
+WEIGHTING_FUNCS 4, 8
+WEIGHTING_FUNCS 6, 8
+WEIGHTING_FUNCS 8, 8
+
+WEIGHTING_FUNCS 2, 10
+WEIGHTING_FUNCS 4, 10
+WEIGHTING_FUNCS 6, 10
+WEIGHTING_FUNCS 8, 10
+
+WEIGHTING_FUNCS 2, 12
+WEIGHTING_FUNCS 4, 12
+WEIGHTING_FUNCS 6, 12
+WEIGHTING_FUNCS 8, 12
+
INIT_XMM ssse3
WEIGHTING_FUNCS 2, 8
WEIGHTING_FUNCS 4, 8
diff --git a/libavcodec/x86/hevcdsp.h b/libavcodec/x86/hevcdsp.h
index a652cbf..37fcb3e 100644
--- a/libavcodec/x86/hevcdsp.h
+++ b/libavcodec/x86/hevcdsp.h
@@ -129,6 +129,10 @@ WEIGHTING_PROTOTYPES(8, ssse3);
WEIGHTING_PROTOTYPES(10, ssse3);
WEIGHTING_PROTOTYPES(12, ssse3);
+WEIGHTING_PROTOTYPES(8, sse4);
+WEIGHTING_PROTOTYPES(10, sse4);
+WEIGHTING_PROTOTYPES(12, sse4);
+
///////////////////////////////////////////////////////////////////////////////
// TRANSFORM_ADD
///////////////////////////////////////////////////////////////////////////////
diff --git a/libavcodec/x86/hevcdsp_init.c b/libavcodec/x86/hevcdsp_init.c
index 95f69e0..7758a27 100644
--- a/libavcodec/x86/hevcdsp_init.c
+++ b/libavcodec/x86/hevcdsp_init.c
@@ -283,6 +283,7 @@ mc_rep_funcs(qpel_hv,12, 8, 32, ssse3);
mc_rep_funcs(qpel_hv,12, 8, 24, ssse3);
mc_rep_funcs(qpel_hv,12, 8, 16, ssse3);
mc_rep_funcs(qpel_hv,12, 4, 12, ssse3);
+#endif
#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,\
@@ -299,6 +300,9 @@ void ff_hevc_put_hevc_uni_w##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t dststri
} \
}
+#if ARCH_X86_64
+
+# if HAVE_SSSE3_EXTERNAL
mc_rep_uni_w(8, 6, 12, ssse3);
mc_rep_uni_w(8, 8, 16, ssse3);
mc_rep_uni_w(8, 8, 24, ssse3);
@@ -319,6 +323,30 @@ mc_rep_uni_w(12, 8, 24, ssse3);
mc_rep_uni_w(12, 8, 32, ssse3);
mc_rep_uni_w(12, 8, 48, ssse3);
mc_rep_uni_w(12, 8, 64, ssse3);
+# endif
+
+# if HAVE_SSE4_EXTERNAL
+mc_rep_uni_w(8, 6, 12, sse4);
+mc_rep_uni_w(8, 8, 16, sse4);
+mc_rep_uni_w(8, 8, 24, sse4);
+mc_rep_uni_w(8, 8, 32, sse4);
+mc_rep_uni_w(8, 8, 48, sse4);
+mc_rep_uni_w(8, 8, 64, sse4);
+
+mc_rep_uni_w(10, 6, 12, sse4);
+mc_rep_uni_w(10, 8, 16, sse4);
+mc_rep_uni_w(10, 8, 24, sse4);
+mc_rep_uni_w(10, 8, 32, sse4);
+mc_rep_uni_w(10, 8, 48, sse4);
+mc_rep_uni_w(10, 8, 64, sse4);
+
+mc_rep_uni_w(12, 6, 12, sse4);
+mc_rep_uni_w(12, 8, 16, sse4);
+mc_rep_uni_w(12, 8, 24, sse4);
+mc_rep_uni_w(12, 8, 32, sse4);
+mc_rep_uni_w(12, 8, 48, sse4);
+mc_rep_uni_w(12, 8, 64, sse4);
+# endif
#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, \
@@ -338,6 +366,7 @@ void ff_hevc_put_hevc_bi_w##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t dststrid
} \
}
+# if HAVE_SSSE3_EXTERNAL
mc_rep_bi_w(8, 6, 12, ssse3);
mc_rep_bi_w(8, 8, 16, ssse3);
mc_rep_bi_w(8, 8, 24, ssse3);
@@ -358,8 +387,32 @@ mc_rep_bi_w(12, 8, 24, ssse3);
mc_rep_bi_w(12, 8, 32, ssse3);
mc_rep_bi_w(12, 8, 48, ssse3);
mc_rep_bi_w(12, 8, 64, ssse3);
-
-#define mc_uni_w_func(name, bitd, W, opt) \
+# endif
+
+# if HAVE_SSE4_EXTERNAL
+mc_rep_bi_w(8, 6, 12, sse4);
+mc_rep_bi_w(8, 8, 16, sse4);
+mc_rep_bi_w(8, 8, 24, sse4);
+mc_rep_bi_w(8, 8, 32, sse4);
+mc_rep_bi_w(8, 8, 48, sse4);
+mc_rep_bi_w(8, 8, 64, sse4);
+
+mc_rep_bi_w(10, 6, 12, sse4);
+mc_rep_bi_w(10, 8, 16, sse4);
+mc_rep_bi_w(10, 8, 24, sse4);
+mc_rep_bi_w(10, 8, 32, sse4);
+mc_rep_bi_w(10, 8, 48, sse4);
+mc_rep_bi_w(10, 8, 64, sse4);
+
+mc_rep_bi_w(12, 6, 12, sse4);
+mc_rep_bi_w(12, 8, 16, sse4);
+mc_rep_bi_w(12, 8, 24, sse4);
+mc_rep_bi_w(12, 8, 32, sse4);
+mc_rep_bi_w(12, 8, 48, sse4);
+mc_rep_bi_w(12, 8, 64, sse4);
+# endif
+
+#define mc_uni_w_func(name, bitd, W, opt, alt) \
static void put_hevc_uni_w_##name##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t _dststride, \
uint8_t *_src, ptrdiff_t _srcstride, \
int height, int denom, \
@@ -367,57 +420,97 @@ static void put_hevc_uni_w_##name##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t _
intptr_t mx, intptr_t my, int width) \
{ \
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_##name##W##_##bitd##_##alt(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);\
}
-#define mc_uni_w_funcs(name, bitd, opt) \
- mc_uni_w_func(name, bitd, 4, opt); \
- mc_uni_w_func(name, bitd, 8, opt); \
- mc_uni_w_func(name, bitd, 12, opt); \
- mc_uni_w_func(name, bitd, 16, opt); \
- mc_uni_w_func(name, bitd, 24, opt); \
- mc_uni_w_func(name, bitd, 32, opt); \
- mc_uni_w_func(name, bitd, 48, opt); \
- mc_uni_w_func(name, bitd, 64, opt)
-
-mc_uni_w_funcs(pel_pixels, 8, ssse3);
-mc_uni_w_func(pel_pixels, 8, 6, ssse3);
-mc_uni_w_funcs(epel_h, 8, ssse3);
-mc_uni_w_func(epel_h, 8, 6, ssse3);
-mc_uni_w_funcs(epel_v, 8, ssse3);
-mc_uni_w_func(epel_v, 8, 6, ssse3);
-mc_uni_w_funcs(epel_hv, 8, ssse3);
-mc_uni_w_func(epel_hv, 8, 6, ssse3);
-mc_uni_w_funcs(qpel_h, 8, ssse3);
-mc_uni_w_funcs(qpel_v, 8, ssse3);
-mc_uni_w_funcs(qpel_hv, 8, ssse3);
-
-mc_uni_w_funcs(pel_pixels, 10, ssse3);
-mc_uni_w_func(pel_pixels, 10, 6, ssse3);
-mc_uni_w_funcs(epel_h, 10, ssse3);
-mc_uni_w_func(epel_h, 10, 6, ssse3);
-mc_uni_w_funcs(epel_v, 10, ssse3);
-mc_uni_w_func(epel_v, 10, 6, ssse3);
-mc_uni_w_funcs(epel_hv, 10, ssse3);
-mc_uni_w_func(epel_hv, 10, 6, ssse3);
-mc_uni_w_funcs(qpel_h, 10, ssse3);
-mc_uni_w_funcs(qpel_v, 10, ssse3);
-mc_uni_w_funcs(qpel_hv, 10, ssse3);
-
-mc_uni_w_funcs(pel_pixels, 12, ssse3);
-mc_uni_w_func(pel_pixels, 12, 6, ssse3);
-mc_uni_w_funcs(epel_h, 12, ssse3);
-mc_uni_w_func(epel_h, 12, 6, ssse3);
-mc_uni_w_funcs(epel_v, 12, ssse3);
-mc_uni_w_func(epel_v, 12, 6, ssse3);
-mc_uni_w_funcs(epel_hv, 12, ssse3);
-mc_uni_w_func(epel_hv, 12, 6, ssse3);
-mc_uni_w_funcs(qpel_h, 12, ssse3);
-mc_uni_w_funcs(qpel_v, 12, ssse3);
-mc_uni_w_funcs(qpel_hv, 12, ssse3);
-
-#define mc_bi_w_func(name, bitd, W, opt) \
+#define mc_uni_w_funcs(name, bitd, opt, alt) \
+ mc_uni_w_func(name, bitd, 4, opt, alt); \
+ mc_uni_w_func(name, bitd, 8, opt, alt); \
+ mc_uni_w_func(name, bitd, 12, opt, alt); \
+ mc_uni_w_func(name, bitd, 16, opt, alt); \
+ mc_uni_w_func(name, bitd, 24, opt, alt); \
+ mc_uni_w_func(name, bitd, 32, opt, alt); \
+ mc_uni_w_func(name, bitd, 48, opt, alt); \
+ mc_uni_w_func(name, bitd, 64, opt, alt)
+
+# if HAVE_SSSE3_EXTERNAL
+mc_uni_w_funcs(pel_pixels, 8, ssse3, ssse3);
+mc_uni_w_func(pel_pixels, 8, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_h, 8, ssse3, ssse3);
+mc_uni_w_func(epel_h, 8, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_v, 8, ssse3, ssse3);
+mc_uni_w_func(epel_v, 8, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_hv, 8, ssse3, ssse3);
+mc_uni_w_func(epel_hv, 8, 6, ssse3, ssse3);
+mc_uni_w_funcs(qpel_h, 8, ssse3, ssse3);
+mc_uni_w_funcs(qpel_v, 8, ssse3, ssse3);
+mc_uni_w_funcs(qpel_hv, 8, ssse3, ssse3);
+
+mc_uni_w_funcs(pel_pixels, 10, ssse3, ssse3);
+mc_uni_w_func(pel_pixels, 10, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_h, 10, ssse3, ssse3);
+mc_uni_w_func(epel_h, 10, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_v, 10, ssse3, ssse3);
+mc_uni_w_func(epel_v, 10, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_hv, 10, ssse3, ssse3);
+mc_uni_w_func(epel_hv, 10, 6, ssse3, ssse3);
+mc_uni_w_funcs(qpel_h, 10, ssse3, ssse3);
+mc_uni_w_funcs(qpel_v, 10, ssse3, ssse3);
+mc_uni_w_funcs(qpel_hv, 10, ssse3, ssse3);
+
+mc_uni_w_funcs(pel_pixels, 12, ssse3, ssse3);
+mc_uni_w_func(pel_pixels, 12, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_h, 12, ssse3, ssse3);
+mc_uni_w_func(epel_h, 12, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_v, 12, ssse3, ssse3);
+mc_uni_w_func(epel_v, 12, 6, ssse3, ssse3);
+mc_uni_w_funcs(epel_hv, 12, ssse3, ssse3);
+mc_uni_w_func(epel_hv, 12, 6, ssse3, ssse3);
+mc_uni_w_funcs(qpel_h, 12, ssse3, ssse3);
+mc_uni_w_funcs(qpel_v, 12, ssse3, ssse3);
+mc_uni_w_funcs(qpel_hv, 12, ssse3, ssse3);
+# endif
+
+# if HAVE_SSE4_EXTERNAL
+mc_uni_w_funcs(pel_pixels, 8, sse4, ssse3);
+mc_uni_w_func(pel_pixels, 8, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_h, 8, sse4, ssse3);
+mc_uni_w_func(epel_h, 8, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_v, 8, sse4, ssse3);
+mc_uni_w_func(epel_v, 8, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_hv, 8, sse4, ssse3);
+mc_uni_w_func(epel_hv, 8, 6, sse4, ssse3);
+mc_uni_w_funcs(qpel_h, 8, sse4, ssse3);
+mc_uni_w_funcs(qpel_v, 8, sse4, ssse3);
+mc_uni_w_funcs(qpel_hv, 8, sse4, ssse3);
+
+mc_uni_w_funcs(pel_pixels, 10, sse4, ssse3);
+mc_uni_w_func(pel_pixels, 10, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_h, 10, sse4, ssse3);
+mc_uni_w_func(epel_h, 10, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_v, 10, sse4, ssse3);
+mc_uni_w_func(epel_v, 10, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_hv, 10, sse4, ssse3);
+mc_uni_w_func(epel_hv, 10, 6, sse4, ssse3);
+mc_uni_w_funcs(qpel_h, 10, sse4, ssse3);
+mc_uni_w_funcs(qpel_v, 10, sse4, ssse3);
+mc_uni_w_funcs(qpel_hv, 10, sse4, ssse3);
+
+mc_uni_w_funcs(pel_pixels, 12, sse4, ssse3);
+mc_uni_w_func(pel_pixels, 12, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_h, 12, sse4, ssse3);
+mc_uni_w_func(epel_h, 12, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_v, 12, sse4, ssse3);
+mc_uni_w_func(epel_v, 12, 6, sse4, ssse3);
+mc_uni_w_funcs(epel_hv, 12, sse4, ssse3);
+mc_uni_w_func(epel_hv, 12, 6, sse4, ssse3);
+mc_uni_w_funcs(qpel_h, 12, sse4, ssse3);
+mc_uni_w_funcs(qpel_v, 12, sse4, ssse3);
+mc_uni_w_funcs(qpel_hv, 12, sse4, ssse3);
+# endif
+
+#define mc_bi_w_func(name, bitd, W, opt, alt) \
static void put_hevc_bi_w_##name##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t _dststride, \
uint8_t *_src, ptrdiff_t _srcstride, \
int16_t *_src2, \
@@ -426,57 +519,98 @@ static void put_hevc_bi_w_##name##W##_##bitd##_##opt(uint8_t *_dst, ptrdiff_t _d
intptr_t mx, intptr_t my, int width) \
{ \
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_##name##W##_##bitd##_##alt(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); \
}
-#define mc_bi_w_funcs(name, bitd, opt) \
- mc_bi_w_func(name, bitd, 4, opt); \
- mc_bi_w_func(name, bitd, 8, opt); \
- mc_bi_w_func(name, bitd, 12, opt); \
- mc_bi_w_func(name, bitd, 16, opt); \
- mc_bi_w_func(name, bitd, 24, opt); \
- mc_bi_w_func(name, bitd, 32, opt); \
- mc_bi_w_func(name, bitd, 48, opt); \
- mc_bi_w_func(name, bitd, 64, opt)
-
-mc_bi_w_funcs(pel_pixels, 8, ssse3);
-mc_bi_w_func(pel_pixels, 8, 6, ssse3);
-mc_bi_w_funcs(epel_h, 8, ssse3);
-mc_bi_w_func(epel_h, 8, 6, ssse3);
-mc_bi_w_funcs(epel_v, 8, ssse3);
-mc_bi_w_func(epel_v, 8, 6, ssse3);
-mc_bi_w_funcs(epel_hv, 8, ssse3);
-mc_bi_w_func(epel_hv, 8, 6, ssse3);
-mc_bi_w_funcs(qpel_h, 8, ssse3);
-mc_bi_w_funcs(qpel_v, 8, ssse3);
-mc_bi_w_funcs(qpel_hv, 8, ssse3);
-
-mc_bi_w_funcs(pel_pixels, 10, ssse3);
-mc_bi_w_func(pel_pixels, 10, 6, ssse3);
-mc_bi_w_funcs(epel_h, 10, ssse3);
-mc_bi_w_func(epel_h, 10, 6, ssse3);
-mc_bi_w_funcs(epel_v, 10, ssse3);
-mc_bi_w_func(epel_v, 10, 6, ssse3);
-mc_bi_w_funcs(epel_hv, 10, ssse3);
-mc_bi_w_func(epel_hv, 10, 6, ssse3);
-mc_bi_w_funcs(qpel_h, 10, ssse3);
-mc_bi_w_funcs(qpel_v, 10, ssse3);
-mc_bi_w_funcs(qpel_hv, 10, ssse3);
-
-mc_bi_w_funcs(pel_pixels, 12, ssse3);
-mc_bi_w_func(pel_pixels, 12, 6, ssse3);
-mc_bi_w_funcs(epel_h, 12, ssse3);
-mc_bi_w_func(epel_h, 12, 6, ssse3);
-mc_bi_w_funcs(epel_v, 12, ssse3);
-mc_bi_w_func(epel_v, 12, 6, ssse3);
-mc_bi_w_funcs(epel_hv, 12, ssse3);
-mc_bi_w_func(epel_hv, 12, 6, ssse3);
-mc_bi_w_funcs(qpel_h, 12, ssse3);
-mc_bi_w_funcs(qpel_v, 12, ssse3);
-mc_bi_w_funcs(qpel_hv, 12, ssse3);
-#endif //ARCH_X86_64 && HAVE_SSSE3_EXTERNAL
+#define mc_bi_w_funcs(name, bitd, opt, alt) \
+ mc_bi_w_func(name, bitd, 4, opt, alt); \
+ mc_bi_w_func(name, bitd, 8, opt, alt); \
+ mc_bi_w_func(name, bitd, 12, opt, alt); \
+ mc_bi_w_func(name, bitd, 16, opt, alt); \
+ mc_bi_w_func(name, bitd, 24, opt, alt); \
+ mc_bi_w_func(name, bitd, 32, opt, alt); \
+ mc_bi_w_func(name, bitd, 48, opt, alt); \
+ mc_bi_w_func(name, bitd, 64, opt, alt)
+
+# if HAVE_SSSE3_EXTERNAL
+mc_bi_w_funcs(pel_pixels, 8, ssse3, ssse3);
+mc_bi_w_func(pel_pixels, 8, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_h, 8, ssse3, ssse3);
+mc_bi_w_func(epel_h, 8, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_v, 8, ssse3, ssse3);
+mc_bi_w_func(epel_v, 8, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_hv, 8, ssse3, ssse3);
+mc_bi_w_func(epel_hv, 8, 6, ssse3, ssse3);
+mc_bi_w_funcs(qpel_h, 8, ssse3, ssse3);
+mc_bi_w_funcs(qpel_v, 8, ssse3, ssse3);
+mc_bi_w_funcs(qpel_hv, 8, ssse3, ssse3);
+
+mc_bi_w_funcs(pel_pixels, 10, ssse3, ssse3);
+mc_bi_w_func(pel_pixels, 10, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_h, 10, ssse3, ssse3);
+mc_bi_w_func(epel_h, 10, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_v, 10, ssse3, ssse3);
+mc_bi_w_func(epel_v, 10, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_hv, 10, ssse3, ssse3);
+mc_bi_w_func(epel_hv, 10, 6, ssse3, ssse3);
+mc_bi_w_funcs(qpel_h, 10, ssse3, ssse3);
+mc_bi_w_funcs(qpel_v, 10, ssse3, ssse3);
+mc_bi_w_funcs(qpel_hv, 10, ssse3, ssse3);
+
+mc_bi_w_funcs(pel_pixels, 12, ssse3, ssse3);
+mc_bi_w_func(pel_pixels, 12, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_h, 12, ssse3, ssse3);
+mc_bi_w_func(epel_h, 12, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_v, 12, ssse3, ssse3);
+mc_bi_w_func(epel_v, 12, 6, ssse3, ssse3);
+mc_bi_w_funcs(epel_hv, 12, ssse3, ssse3);
+mc_bi_w_func(epel_hv, 12, 6, ssse3, ssse3);
+mc_bi_w_funcs(qpel_h, 12, ssse3, ssse3);
+mc_bi_w_funcs(qpel_v, 12, ssse3, ssse3);
+mc_bi_w_funcs(qpel_hv, 12, ssse3, ssse3);
+# endif
+
+# if HAVE_SSE4_EXTERNAL
+mc_bi_w_funcs(pel_pixels, 8, sse4, ssse3);
+mc_bi_w_func(pel_pixels, 8, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_h, 8, sse4, ssse3);
+mc_bi_w_func(epel_h, 8, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_v, 8, sse4, ssse3);
+mc_bi_w_func(epel_v, 8, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_hv, 8, sse4, ssse3);
+mc_bi_w_func(epel_hv, 8, 6, sse4, ssse3);
+mc_bi_w_funcs(qpel_h, 8, sse4, ssse3);
+mc_bi_w_funcs(qpel_v, 8, sse4, ssse3);
+mc_bi_w_funcs(qpel_hv, 8, sse4, ssse3);
+
+mc_bi_w_funcs(pel_pixels, 10, sse4, ssse3);
+mc_bi_w_func(pel_pixels, 10, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_h, 10, sse4, ssse3);
+mc_bi_w_func(epel_h, 10, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_v, 10, sse4, ssse3);
+mc_bi_w_func(epel_v, 10, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_hv, 10, sse4, ssse3);
+mc_bi_w_func(epel_hv, 10, 6, sse4, ssse3);
+mc_bi_w_funcs(qpel_h, 10, sse4, ssse3);
+mc_bi_w_funcs(qpel_v, 10, sse4, ssse3);
+mc_bi_w_funcs(qpel_hv, 10, sse4, ssse3);
+
+mc_bi_w_funcs(pel_pixels, 12, sse4, ssse3);
+mc_bi_w_func(pel_pixels, 12, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_h, 12, sse4, ssse3);
+mc_bi_w_func(epel_h, 12, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_v, 12, sse4, ssse3);
+mc_bi_w_func(epel_v, 12, 6, sse4, ssse3);
+mc_bi_w_funcs(epel_hv, 12, sse4, ssse3);
+mc_bi_w_func(epel_hv, 12, 6, sse4, ssse3);
+mc_bi_w_funcs(qpel_h, 12, sse4, ssse3);
+mc_bi_w_funcs(qpel_v, 12, sse4, ssse3);
+mc_bi_w_funcs(qpel_hv, 12, sse4, ssse3);
+# endif
+
+#endif //ARCH_X86_64
#define EPEL_LINKS(pointer, my, mx, fname, bitd, opt ) \
@@ -499,6 +633,29 @@ mc_bi_w_funcs(qpel_hv, 12, ssse3);
PEL_LINK(pointer, 8, my , mx , fname##48, bitd, opt ); \
PEL_LINK(pointer, 9, my , mx , fname##64, bitd, opt )
+#define PELW_LINK(dst, idx1, idx2, idx3, name, D, opt) \
+dst ## _uni_w[idx1][idx2][idx3] = put_hevc_uni_w_ ## name ## _ ## D ## _##opt; \
+dst ## _bi_w[idx1][idx2][idx3] = put_hevc_bi_w_ ## name ## _ ## D ## _##opt
+
+#define EPELW_LINKS(pointer, my, mx, fname, bitd, opt ) \
+ PELW_LINK(pointer, 1, my , mx , fname##4 , bitd, opt ); \
+ PELW_LINK(pointer, 2, my , mx , fname##6 , bitd, opt ); \
+ PELW_LINK(pointer, 3, my , mx , fname##8 , bitd, opt ); \
+ PELW_LINK(pointer, 4, my , mx , fname##12, bitd, opt ); \
+ PELW_LINK(pointer, 5, my , mx , fname##16, bitd, opt ); \
+ PELW_LINK(pointer, 6, my , mx , fname##24, bitd, opt ); \
+ PELW_LINK(pointer, 7, my , mx , fname##32, bitd, opt ); \
+ PELW_LINK(pointer, 8, my , mx , fname##48, bitd, opt ); \
+ PELW_LINK(pointer, 9, my , mx , fname##64, bitd, opt )
+#define QPELW_LINKS(pointer, my, mx, fname, bitd, opt) \
+ PELW_LINK(pointer, 1, my , mx , fname##4 , bitd, opt ); \
+ PELW_LINK(pointer, 3, my , mx , fname##8 , bitd, opt ); \
+ PELW_LINK(pointer, 4, my , mx , fname##12, bitd, opt ); \
+ PELW_LINK(pointer, 5, my , mx , fname##16, bitd, opt ); \
+ PELW_LINK(pointer, 6, my , mx , fname##24, bitd, opt ); \
+ PELW_LINK(pointer, 7, my , mx , fname##32, bitd, opt ); \
+ PELW_LINK(pointer, 8, my , mx , fname##48, bitd, opt ); \
+ PELW_LINK(pointer, 9, my , mx , fname##64, bitd, opt )
void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
{
@@ -539,6 +696,17 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 8, ssse3);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 8, ssse3);
}
+ if (EXTERNAL_SSE4(cpu_flags) && ARCH_X86_64) {
+ EPELW_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 8, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 0, 1, epel_h, 8, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 1, 0, epel_v, 8, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 1, 1, epel_hv, 8, sse4);
+
+ QPELW_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 8, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 8, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 8, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 8, sse4);
+ }
if (EXTERNAL_AVX(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_avx;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_avx;
@@ -590,6 +758,17 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 10, ssse3);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 10, ssse3);
}
+ if (EXTERNAL_SSE4(cpu_flags) && ARCH_X86_64) {
+ EPELW_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 10, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 0, 1, epel_h, 10, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 1, 0, epel_v, 10, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 1, 1, epel_hv, 10, sse4);
+
+ QPELW_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 10, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 10, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 10, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 10, sse4);
+ }
if (EXTERNAL_AVX(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_avx;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_avx;
@@ -638,6 +817,17 @@ void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 12, ssse3);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 12, ssse3);
}
+ if (EXTERNAL_SSE4(cpu_flags) && ARCH_X86_64) {
+ EPELW_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 12, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 0, 1, epel_h, 12, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 1, 0, epel_v, 12, sse4);
+ EPELW_LINKS(c->put_hevc_epel, 1, 1, epel_hv, 12, sse4);
+
+ QPELW_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 12, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 12, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 12, sse4);
+ QPELW_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 12, sse4);
+ }
if (EXTERNAL_AVX(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_12_avx;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_12_avx;
--
1.9.2.msysgit.0
More information about the ffmpeg-devel
mailing list