From 5dc73634469dc54e109fe0fd9f3fc0c24409e548 Mon Sep 17 00:00:00 2001 From: Mans Rullgard Date: Mon, 10 Sep 2012 15:47:00 +0100 Subject: [PATCH] ARM: libswresample: NEON optimised stereo fltp to s16 conversion Adaption to swr by: Michael Niedermayer Signed-off-by: Michael Niedermayer --- libswresample/arm/audio_convert_init.c | 8 ++++ libswresample/arm/audio_convert_neon.S | 64 ++++++++++++++++++++++++++ 2 files changed, 72 insertions(+) diff --git a/libswresample/arm/audio_convert_init.c b/libswresample/arm/audio_convert_init.c index f8f80078a7..a52ea5a196 100644 --- a/libswresample/arm/audio_convert_init.c +++ b/libswresample/arm/audio_convert_init.c @@ -27,11 +27,17 @@ #include "libswresample/audioconvert.h" void swri_oldapi_conv_flt_to_s16_neon(int16_t *dst, const float *src, int len); +void swri_oldapi_conv_fltp_to_s16_2ch_neon(int16_t *dst, float *const *src, int len, int channels); static void conv_flt_to_s16_neon(uint8_t **dst, const uint8_t **src, int len){ swri_oldapi_conv_flt_to_s16_neon((int16_t*)*dst, (const float*)*src, len); } +static void conv_fltp_to_s16_2ch_neon(uint8_t **dst, const uint8_t **src, int len){ + swri_oldapi_conv_fltp_to_s16_2ch_neon((int16_t*)*dst, (float *const*)src, len, 2); +} + + av_cold void swri_audio_convert_init_arm(struct AudioConvert *ac, enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt, @@ -44,5 +50,7 @@ av_cold void swri_audio_convert_init_arm(struct AudioConvert *ac, if (have_neon(cpu_flags)) { if(out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLT || out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_FLTP) ac->simd_f = conv_flt_to_s16_neon; + if(out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLTP && channels == 2) + ac->simd_f = conv_fltp_to_s16_2ch_neon; } } diff --git a/libswresample/arm/audio_convert_neon.S b/libswresample/arm/audio_convert_neon.S index e174a7730f..f6099dac93 100644 --- a/libswresample/arm/audio_convert_neon.S +++ b/libswresample/arm/audio_convert_neon.S @@ -64,3 +64,67 @@ function swri_oldapi_conv_flt_to_s16_neon, export=1 vst1.16 {q2}, [r0,:128]! bx lr endfunc + +function swri_oldapi_conv_fltp_to_s16_2ch_neon, export=1 + ldm r1, {r1, r3} + subs r2, r2, #8 + vld1.32 {q0}, [r1,:128]! + vcvt.s32.f32 q8, q0, #31 + vld1.32 {q1}, [r1,:128]! + vcvt.s32.f32 q9, q1, #31 + vld1.32 {q10}, [r3,:128]! + vcvt.s32.f32 q10, q10, #31 + vld1.32 {q11}, [r3,:128]! + vcvt.s32.f32 q11, q11, #31 + beq 3f + bics r12, r2, #15 + beq 2f +1: subs r12, r12, #16 + vld1.32 {q0}, [r1,:128]! + vcvt.s32.f32 q0, q0, #31 + vsri.32 q10, q8, #16 + vld1.32 {q1}, [r1,:128]! + vcvt.s32.f32 q1, q1, #31 + vld1.32 {q12}, [r3,:128]! + vcvt.s32.f32 q12, q12, #31 + vld1.32 {q13}, [r3,:128]! + vsri.32 q11, q9, #16 + vst1.16 {q10}, [r0,:128]! + vcvt.s32.f32 q13, q13, #31 + vst1.16 {q11}, [r0,:128]! + vsri.32 q12, q0, #16 + vld1.32 {q8}, [r1,:128]! + vsri.32 q13, q1, #16 + vst1.16 {q12}, [r0,:128]! + vcvt.s32.f32 q8, q8, #31 + vld1.32 {q9}, [r1,:128]! + vcvt.s32.f32 q9, q9, #31 + vld1.32 {q10}, [r3,:128]! + vcvt.s32.f32 q10, q10, #31 + vld1.32 {q11}, [r3,:128]! + vcvt.s32.f32 q11, q11, #31 + vst1.16 {q13}, [r0,:128]! + bne 1b + ands r2, r2, #15 + beq 3f +2: vsri.32 q10, q8, #16 + vld1.32 {q0}, [r1,:128]! + vcvt.s32.f32 q0, q0, #31 + vld1.32 {q1}, [r1,:128]! + vcvt.s32.f32 q1, q1, #31 + vld1.32 {q12}, [r3,:128]! + vcvt.s32.f32 q12, q12, #31 + vsri.32 q11, q9, #16 + vld1.32 {q13}, [r3,:128]! + vcvt.s32.f32 q13, q13, #31 + vst1.16 {q10}, [r0,:128]! + vsri.32 q12, q0, #16 + vst1.16 {q11}, [r0,:128]! + vsri.32 q13, q1, #16 + vst1.16 {q12-q13},[r0,:128]! + bx lr +3: vsri.32 q10, q8, #16 + vsri.32 q11, q9, #16 + vst1.16 {q10-q11},[r0,:128]! + bx lr +endfunc