diff options
| author | Janne Grunau <janne-libav@jannau.net> | 2013-12-20 21:02:43 +0100 |
|---|---|---|
| committer | Janne Grunau <janne-libav@jannau.net> | 2014-01-15 12:31:04 +0100 |
| commit | 36e3b1f2fd262028834a9d7b1eb533c1218ee6c2 (patch) | |
| tree | d897dbdd7b40d0b58d8c9d6d383283a448cb54b6 /libavcodec/aarch64/neon.S | |
| parent | c65d67ef50241886b9749e88fd333c52090f70c7 (diff) | |
| download | ffmpeg-36e3b1f2fd262028834a9d7b1eb533c1218ee6c2.tar.gz | |
aarch64: h264 loop filter NEON optimizations
Ported from ARMv7 NEON.
Diffstat (limited to 'libavcodec/aarch64/neon.S')
| -rw-r--r-- | libavcodec/aarch64/neon.S | 24 |
1 files changed, 24 insertions, 0 deletions
diff --git a/libavcodec/aarch64/neon.S b/libavcodec/aarch64/neon.S index c4310d7bc0..f1072b73e5 100644 --- a/libavcodec/aarch64/neon.S +++ b/libavcodec/aarch64/neon.S @@ -80,6 +80,30 @@ trn2 \r7\().4S, \t1\().4S, \r7\().4S .endm +.macro transpose_4x16B r0, r1, r2, r3, t4, t5, t6, t7 + trn1 \t4\().16B, \r0\().16B, \r1\().16B + trn2 \t5\().16B, \r0\().16B, \r1\().16B + trn1 \t6\().16B, \r2\().16B, \r3\().16B + trn2 \t7\().16B, \r2\().16B, \r3\().16B + + trn1 \r0\().8H, \t4\().8H, \t6\().8H + trn2 \r2\().8H, \t4\().8H, \t6\().8H + trn1 \r1\().8H, \t5\().8H, \t7\().8H + trn2 \r3\().8H, \t5\().8H, \t7\().8H +.endm + +.macro transpose_4x8B r0, r1, r2, r3, t4, t5, t6, t7 + trn1 \t4\().8B, \r0\().8B, \r1\().8B + trn2 \t5\().8B, \r0\().8B, \r1\().8B + trn1 \t6\().8B, \r2\().8B, \r3\().8B + trn2 \t7\().8B, \r2\().8B, \r3\().8B + + trn1 \r0\().4H, \t4\().4H, \t6\().4H + trn2 \r2\().4H, \t4\().4H, \t6\().4H + trn1 \r1\().4H, \t5\().4H, \t7\().4H + trn2 \r3\().4H, \t5\().4H, \t7\().4H +.endm + .macro transpose_4x4H r0, r1, r2, r3, r4, r5, r6, r7 trn1 \r4\().4H, \r0\().4H, \r1\().4H trn2 \r5\().4H, \r0\().4H, \r1\().4H |