diff options
| author | Michael Niedermayer <michaelni@gmx.at> | 2011-10-22 11:11:08 +0200 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2011-10-22 11:11:08 +0200 |
| commit | d8cae42d72092501be880348ae0c223d2ea220a2 (patch) | |
| tree | 9f5cd163810d5e7babd864637b10968c99814ab8 /libavcodec/arm | |
| parent | d3d97d43d51840086d86728bbfcd072eea7a5b85 (diff) | |
| parent | 9d4bbf6d4f3a0b78c4ec7a1f366adf5c3cc8ea1a (diff) | |
Merge remote-tracking branch 'mans/ac3'
* mans/ac3:
ac3enc: slightly faster quantize_mantissas_blk_ch()
ac3enc: NEON optimised sum_square_butterfly_float
ac3enc: neon optimised sum_square_butterfly_int32
ac3enc: move inner loop of compute_rematrixing_strategy to ac3dsp
Conflicts:
libavcodec/ac3enc_template.c
Merged-by: Michael Niedermayer <michaelni@gmx.at>
Diffstat (limited to 'libavcodec/arm')
| -rw-r--r-- | libavcodec/arm/ac3dsp_init_arm.c | 10 | ||||
| -rw-r--r-- | libavcodec/arm/ac3dsp_neon.S | 44 |
2 files changed, 54 insertions, 0 deletions
diff --git a/libavcodec/arm/ac3dsp_init_arm.c b/libavcodec/arm/ac3dsp_init_arm.c index 5a9fdf0071..8ab685496d 100644 --- a/libavcodec/arm/ac3dsp_init_arm.c +++ b/libavcodec/arm/ac3dsp_init_arm.c @@ -29,6 +29,14 @@ void ff_ac3_lshift_int16_neon(int16_t *src, unsigned len, unsigned shift); void ff_ac3_rshift_int32_neon(int32_t *src, unsigned len, unsigned shift); void ff_float_to_fixed24_neon(int32_t *dst, const float *src, unsigned int len); void ff_ac3_extract_exponents_neon(uint8_t *exp, int32_t *coef, int nb_coefs); +void ff_ac3_sum_square_butterfly_int32_neon(int64_t sum[4], + const int32_t *coef0, + const int32_t *coef1, + int len); +void ff_ac3_sum_square_butterfly_float_neon(float sum[4], + const float *coef0, + const float *coef1, + int len); void ff_ac3_bit_alloc_calc_bap_armv6(int16_t *mask, int16_t *psd, int start, int end, @@ -52,5 +60,7 @@ av_cold void ff_ac3dsp_init_arm(AC3DSPContext *c, int bit_exact) c->ac3_rshift_int32 = ff_ac3_rshift_int32_neon; c->float_to_fixed24 = ff_float_to_fixed24_neon; c->extract_exponents = ff_ac3_extract_exponents_neon; + c->sum_square_butterfly_int32 = ff_ac3_sum_square_butterfly_int32_neon; + c->sum_square_butterfly_float = ff_ac3_sum_square_butterfly_float_neon; } } diff --git a/libavcodec/arm/ac3dsp_neon.S b/libavcodec/arm/ac3dsp_neon.S index e97197c27a..152afbfd3c 100644 --- a/libavcodec/arm/ac3dsp_neon.S +++ b/libavcodec/arm/ac3dsp_neon.S @@ -108,3 +108,47 @@ function ff_ac3_extract_exponents_neon, export=1 bgt 1b bx lr endfunc + +function ff_ac3_sum_square_butterfly_int32_neon, export=1 + vmov.i64 q0, #0 + vmov.i64 q1, #0 + vmov.i64 q2, #0 + vmov.i64 q3, #0 +1: + vld1.32 {d16}, [r1]! + vld1.32 {d17}, [r2]! + vadd.s32 d18, d16, d17 + vsub.s32 d19, d16, d17 + vmlal.s32 q0, d16, d16 + vmlal.s32 q1, d17, d17 + vmlal.s32 q2, d18, d18 + vmlal.s32 q3, d19, d19 + subs r3, r3, #2 + bgt 1b + vadd.s64 d0, d0, d1 + vadd.s64 d1, d2, d3 + vadd.s64 d2, d4, d5 + vadd.s64 d3, d6, d7 + vst1.64 {q0-q1}, [r0] + bx lr +endfunc + +function ff_ac3_sum_square_butterfly_float_neon, export=1 + vmov.f32 q0, #0.0 + vmov.f32 q1, #0.0 +1: + vld1.32 {d16}, [r1]! + vld1.32 {d17}, [r2]! + vadd.f32 d18, d16, d17 + vsub.f32 d19, d16, d17 + vmla.f32 d0, d16, d16 + vmla.f32 d1, d17, d17 + vmla.f32 d2, d18, d18 + vmla.f32 d3, d19, d19 + subs r3, r3, #2 + bgt 1b + vpadd.f32 d0, d0, d1 + vpadd.f32 d1, d2, d3 + vst1.32 {q0}, [r0] + bx lr +endfunc |