diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2011-12-13 23:21:37 +0100 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2011-12-13 23:21:37 +0100 |
| commit | 3ba0bfe71fb18e955ca0110e5a65105d84932fbc (patch) | |
| tree | b308d41d170483c23c31c87d8dcb19fc169e2eb6 /libavcodec/arm | |
| parent | 36be045ed7942e07742c3cf3d3012b1d2a9ec344 (diff) | |
| parent | a99273ebf328658c183c2d267f1c2b8bfac58bb3 (diff) | |
Merge remote-tracking branch 'qatar/master'
* qatar/master:
ulti: Fix invalid reads
lavf: dealloc private options in av_write_trailer
yadif: support 10bit YUV
vc1: mark with ER_MB_ERROR bits overconsumption
lavc: introduce ER_MB_END and ER_MB_ERROR
error_resilience: use the ER_ namespace
build: move inclusion of subdir.mak to main subdir loop
rv34: NEON optimised 4x4 dequant
rv34: move 4x4 dequant to RV34DSPContext
aacdec: Use intfloat.h rather than local punning union.
Conflicts:
libavcodec/h264.c
libavcodec/vc1dec.c
libavfilter/vf_yadif.c
libavformat/Makefile
Merged-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libavcodec/arm')
| -rw-r--r-- | libavcodec/arm/rv34dsp_init_neon.c | 3 | ||||
| -rw-r--r-- | libavcodec/arm/rv34dsp_neon.S | 24 |
2 files changed, 27 insertions, 0 deletions
diff --git a/libavcodec/arm/rv34dsp_init_neon.c b/libavcodec/arm/rv34dsp_init_neon.c index 9a09fde7a9..acf2a7dcd3 100644 --- a/libavcodec/arm/rv34dsp_init_neon.c +++ b/libavcodec/arm/rv34dsp_init_neon.c @@ -25,9 +25,12 @@ void ff_rv34_inv_transform_neon(DCTELEM *block); void ff_rv34_inv_transform_noround_neon(DCTELEM *block); +void ff_rv34_dequant4x4_neon(DCTELEM *block, int Qdc, int Q); void ff_rv34dsp_init_neon(RV34DSPContext *c, DSPContext* dsp) { c->rv34_inv_transform_tab[0] = ff_rv34_inv_transform_neon; c->rv34_inv_transform_tab[1] = ff_rv34_inv_transform_noround_neon; + + c->rv34_dequant4x4 = ff_rv34_dequant4x4_neon; } diff --git a/libavcodec/arm/rv34dsp_neon.S b/libavcodec/arm/rv34dsp_neon.S index f700f5c321..423b537fb9 100644 --- a/libavcodec/arm/rv34dsp_neon.S +++ b/libavcodec/arm/rv34dsp_neon.S @@ -107,3 +107,27 @@ function ff_rv34_inv_transform_noround_neon, export=1 vst4.16 {d0[3], d1[3], d2[3], d3[3]}, [r2,:64], r1 bx lr endfunc + +function ff_rv34_dequant4x4_neon, export=1 + mov r3, r0 + mov r12, #16 + vdup.16 q0, r2 + vmov.16 d0[0], r1 + vld1.16 {d2}, [r0,:64], r12 + vld1.16 {d4}, [r0,:64], r12 + vld1.16 {d6}, [r0,:64], r12 + vld1.16 {d16}, [r0,:64], r12 + vmull.s16 q1, d2, d0 + vmull.s16 q2, d4, d1 + vmull.s16 q3, d6, d1 + vmull.s16 q8, d16, d1 + vqrshrn.s32 d2, q1, #4 + vqrshrn.s32 d4, q2, #4 + vqrshrn.s32 d6, q3, #4 + vqrshrn.s32 d16, q8, #4 + vst1.16 {d2}, [r3,:64], r12 + vst1.16 {d4}, [r3,:64], r12 + vst1.16 {d6}, [r3,:64], r12 + vst1.16 {d16}, [r3,:64], r12 + bx lr +endfunc |