Browse Source

convert h264_deblock_q1 to an inline function.

part 1/2 of fixing Altivec-accelerated H264 luma inloop filter

Originally committed as revision 9348 to svn://svn.ffmpeg.org/ffmpeg/trunk
tags/v0.5
Guillaume Poirier 18 years ago
parent
commit
963eca226e
1 changed files with 22 additions and 18 deletions
  1. +22
    -18
      libavcodec/ppc/h264_altivec.c

+ 22
- 18
libavcodec/ppc/h264_altivec.c View File

@@ -741,24 +741,28 @@ static inline vector unsigned char h264_deblock_mask ( register vector unsigned
}

// out: p1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
#define h264_deblock_q1(p0, p1, p2, q0, tc0) { \
\
register vector unsigned char average = vec_avg(p0, q0); \
register vector unsigned char temp; \
register vector unsigned char uncliped; \
register vector unsigned char ones; \
register vector unsigned char max; \
register vector unsigned char min; \
\
temp = vec_xor(average, p2); \
average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */ \
ones = vec_splat_u8(1); \
temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */ \
uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */\
max = vec_adds(p1, tc0); \
min = vec_subs(p1, tc0); \
p1 = vec_max(min, uncliped); \
p1 = vec_min(max, p1); \
static inline void h264_deblock_q1(register vector unsigned char p0,
register vector unsigned char p1,
register vector unsigned char p2,
register vector unsigned char q0,
register vector unsigned char tc0) {

register vector unsigned char average = vec_avg(p0, q0);
register vector unsigned char temp;
register vector unsigned char uncliped;
register vector unsigned char ones;
register vector unsigned char max;
register vector unsigned char min;

temp = vec_xor(average, p2);
average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
ones = vec_splat_u8(1);
temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
max = vec_adds(p1, tc0);
min = vec_subs(p1, tc0);
p1 = vec_max(min, uncliped);
p1 = vec_min(max, p1);
}

#define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \


Loading…
Cancel
Save