[vlc-commits] yadif: update asm code from libav.git
Ilkka Ollakka
git at videolan.org
Thu Nov 1 08:32:11 CET 2012
vlc | branch: master | Ilkka Ollakka <ileoo at videolan.org> | Sun Sep 23 12:59:24 2012 +0300| [346f385bb658c2731d0ccb0a85f7da72d607ea7a] | committer: Ilkka Ollakka
yadif: update asm code from libav.git
> http://git.videolan.org/gitweb.cgi/vlc.git/?a=commit;h=346f385bb658c2731d0ccb0a85f7da72d607ea7a
---
modules/video_filter/deinterlace/yadif_template.h | 42 +++++++++------------
1 file changed, 17 insertions(+), 25 deletions(-)
diff --git a/modules/video_filter/deinterlace/yadif_template.h b/modules/video_filter/deinterlace/yadif_template.h
index f70d249..5d055d2 100644
--- a/modules/video_filter/deinterlace/yadif_template.h
+++ b/modules/video_filter/deinterlace/yadif_template.h
@@ -18,6 +18,7 @@
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
+#define MANGLE(a) #a "(%%rip)"
#ifdef COMPILE_TEMPLATE_SSE
#define REGMM "xmm"
#define MM "%%"REGMM
@@ -65,7 +66,7 @@
MOVQ" "MM"2, "MM"5 \n\t"\
"pxor "MM"3, "MM"4 \n\t"\
"pavgb "MM"3, "MM"5 \n\t"\
- "pand %[pb_1], "MM"4 \n\t"\
+ "pand "MANGLE(pb_1)", "MM"4 \n\t"\
"psubusb "MM"4, "MM"5 \n\t"\
PSRL1(MM"5") \
"punpcklbw "MM"7, "MM"5 \n\t" /* (cur[x-refs+j] + cur[x+refs-j])>>1 */\
@@ -95,7 +96,7 @@
#define CHECK2 /* pretend not to have checked dir=2 if dir=1 was bad.\
hurts both quality and speed, but matches the C version. */\
- "paddw %[pw_1], "MM"6 \n\t"\
+ "paddw "MANGLE(pw_1)", "MM"6 \n\t"\
"psllw $14, "MM"6 \n\t"\
"paddsw "MM"6, "MM"2 \n\t"\
MOVQ" "MM"0, "MM"3 \n\t"\
@@ -113,10 +114,7 @@ VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
uint8_t *prev, uint8_t *cur, uint8_t *next,
int w, int prefs, int mrefs, int parity, int mode)
{
- DECLARE_ALIGNED(16, uint8_t, tmp0[16]);
- DECLARE_ALIGNED(16, uint8_t, tmp1[16]);
- DECLARE_ALIGNED(16, uint8_t, tmp2[16]);
- DECLARE_ALIGNED(16, uint8_t, tmp3[16]);
+ DECLARE_ALIGNED(16, uint8_t, tmp[16*4]);
int x;
#define FILTER\
@@ -130,9 +128,9 @@ VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
MOVQ" "MM"3, "MM"4 \n\t"\
"paddw "MM"2, "MM"3 \n\t"\
"psraw $1, "MM"3 \n\t" /* d = (prev2[x] + next2[x])>>1 */\
- MOVQ" "MM"0, %[tmp0] \n\t" /* c */\
- MOVQ" "MM"3, %[tmp1] \n\t" /* d */\
- MOVQ" "MM"1, %[tmp2] \n\t" /* e */\
+ MOVQ" "MM"0, (%[tmp]) \n\t" /* c */\
+ MOVQ" "MM"3, 16(%[tmp]) \n\t" /* d */\
+ MOVQ" "MM"1, 32(%[tmp]) \n\t" /* e */\
"psubw "MM"4, "MM"2 \n\t"\
PABS( MM"4", MM"2") /* temporal_diff0 */\
LOAD("(%[prev],%[mrefs])", MM"3") /* prev[x-refs] */\
@@ -154,7 +152,7 @@ VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
"paddw "MM"4, "MM"3 \n\t" /* temporal_diff2 */\
"psrlw $1, "MM"3 \n\t"\
"pmaxsw "MM"3, "MM"2 \n\t"\
- MOVQ" "MM"2, %[tmp3] \n\t" /* diff */\
+ MOVQ" "MM"2, 48(%[tmp]) \n\t" /* diff */\
\
"paddw "MM"0, "MM"1 \n\t"\
"paddw "MM"0, "MM"0 \n\t"\
@@ -173,7 +171,7 @@ VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
"punpcklbw "MM"7, "MM"3 \n\t" /* ABS(cur[x-refs+1] - cur[x+refs+1]) */\
"paddw "MM"2, "MM"0 \n\t"\
"paddw "MM"3, "MM"0 \n\t"\
- "psubw %[pw_1], "MM"0 \n\t" /* spatial_score */\
+ "psubw "MANGLE(pw_1)", "MM"0 \n\t" /* spatial_score */\
\
CHECK(-2,0)\
CHECK1\
@@ -185,7 +183,7 @@ VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
CHECK2\
\
/* if(p->mode<2) ... */\
- MOVQ" %[tmp3], "MM"6 \n\t" /* diff */\
+ MOVQ" 48(%[tmp]), "MM"6 \n\t" /* diff */\
"cmpl $2, %[mode] \n\t"\
"jge 1f \n\t"\
LOAD("(%["prev2"],%[mrefs],2)", MM"2") /* prev2[x-2*refs] */\
@@ -196,9 +194,9 @@ VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
"paddw "MM"5, "MM"3 \n\t"\
"psrlw $1, "MM"2 \n\t" /* b */\
"psrlw $1, "MM"3 \n\t" /* f */\
- MOVQ" %[tmp0], "MM"4 \n\t" /* c */\
- MOVQ" %[tmp1], "MM"5 \n\t" /* d */\
- MOVQ" %[tmp2], "MM"7 \n\t" /* e */\
+ MOVQ" (%[tmp]), "MM"4 \n\t" /* c */\
+ MOVQ" 16(%[tmp]), "MM"5 \n\t" /* d */\
+ MOVQ" 32(%[tmp]), "MM"7 \n\t" /* e */\
"psubw "MM"4, "MM"2 \n\t" /* b-c */\
"psubw "MM"7, "MM"3 \n\t" /* f-e */\
MOVQ" "MM"5, "MM"0 \n\t"\
@@ -217,7 +215,7 @@ VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
"pmaxsw "MM"4, "MM"6 \n\t" /* diff= MAX3(diff, min, -max); */\
"1: \n\t"\
\
- MOVQ" %[tmp1], "MM"2 \n\t" /* d */\
+ MOVQ" 16(%[tmp]), "MM"2 \n\t" /* d */\
MOVQ" "MM"2, "MM"3 \n\t"\
"psubw "MM"6, "MM"2 \n\t" /* d-diff */\
"paddw "MM"6, "MM"3 \n\t" /* d+diff */\
@@ -225,19 +223,13 @@ VLC_TARGET static void RENAME(yadif_filter_line)(uint8_t *dst,
"pminsw "MM"3, "MM"1 \n\t" /* d = clip(spatial_pred, d-diff, d+diff); */\
"packuswb "MM"1, "MM"1 \n\t"\
\
- :[tmp0]"=m"(tmp0),\
- [tmp1]"=m"(tmp1),\
- [tmp2]"=m"(tmp2),\
- [tmp3]"=m"(tmp3)\
- :[prev] "r"(prev),\
+ ::[prev] "r"(prev),\
[cur] "r"(cur),\
[next] "r"(next),\
[prefs]"r"((x86_reg)prefs),\
[mrefs]"r"((x86_reg)mrefs),\
- [pw_1] "m"(pw_1),\
- [pb_1] "m"(pb_1),\
- [mode] "g"(mode)\
- :REGMM"0",REGMM"1",REGMM"2",REGMM"3",REGMM"4",REGMM"5",REGMM"6",REGMM"7"\
+ [mode] "g"(mode),\
+ [tmp] "r"(tmp)\
);\
__asm__ volatile(MOV" "MM"1, %0" :"=m"(*dst));\
dst += STEP;\
More information about the vlc-commits
mailing list