[SCM] libav/experimental: Use a faster way to compute 255-val: Instead of creating a vector of all 255s, and then doing the subtraction, nor of the vector with itself: saves one instruction and a register. Patch by Graham Booker % gbooker A tamu P edu%
siretart at users.alioth.debian.org
siretart at users.alioth.debian.org
Sun Jun 30 16:02:03 UTC 2013
The following commit has been merged in the experimental branch:
commit f4a02f6e9bc0901fd9b21544f34e36c93ba9cfc1
Author: Graham Booker <gbooker at tamu.edu>
Date: Sat Jun 16 18:59:10 2007 +0000
Use a faster way to compute 255-val: Instead of creating a vector of
all 255s, and then doing the subtraction, nor of the vector with itself: saves
one instruction and a register.
Patch by Graham Booker % gbooker A tamu P edu%
Originally committed as revision 9340 to svn://svn.ffmpeg.org/ffmpeg/trunk
diff --git a/libavcodec/ppc/h264_altivec.c b/libavcodec/ppc/h264_altivec.c
index 9055d73..9eccf28 100644
--- a/libavcodec/ppc/h264_altivec.c
+++ b/libavcodec/ppc/h264_altivec.c
@@ -766,7 +766,6 @@ static inline vector unsigned char h264_deblock_mask ( register vector unsigned
const vector unsigned char A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \
\
register vector unsigned char pq0bit = vec_xor(p0,q0); \
- register vector unsigned char temp; \
register vector unsigned char q1minus; \
register vector unsigned char p0minus; \
register vector unsigned char stage1; \
@@ -775,11 +774,10 @@ static inline vector unsigned char h264_deblock_mask ( register vector unsigned
register vector unsigned char delta; \
register vector unsigned char deltaneg; \
\
- temp = (vector unsigned char)vec_cmpeq(p0, p0); \
- q1minus = vec_xor(temp, q1); /* 255 - q1 */ \
+ q1minus = vec_nor(q1, q1); /* 255 - q1 */ \
stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \
stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \
- p0minus = vec_xor(temp, p0); /* 255 - p0 */ \
+ p0minus = vec_nor(p0, p0); /* 255 - p0 */ \
stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \
pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \
stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \
--
Libav/FFmpeg packaging
More information about the pkg-multimedia-commits
mailing list