From 6c63cbfe7aa24f37064e2384f9b39431a91bfb29 Mon Sep 17 00:00:00 2001 From: Justin Ruggles Date: Mon, 23 Apr 2012 22:22:28 -0400 Subject: [PATCH] lavr: Add x86-optimized function for flt to s16 conversion --- libavresample/x86/audio_convert.asm | 33 ++++++++++++++++++++++++++ libavresample/x86/audio_convert_init.c | 4 ++++ 2 files changed, 37 insertions(+) diff --git a/libavresample/x86/audio_convert.asm b/libavresample/x86/audio_convert.asm index 53091acf99..a28198197f 100644 --- a/libavresample/x86/audio_convert.asm +++ b/libavresample/x86/audio_convert.asm @@ -28,6 +28,7 @@ SECTION_RODATA 32 pf_s32_inv_scale: times 8 dd 0x30000000 pf_s16_inv_scale: times 4 dd 0x38000000 +pf_s16_scale: times 4 dd 0x47000000 SECTION_TEXT @@ -158,6 +159,38 @@ INIT_YMM avx CONV_S32_TO_FLT %endif +;------------------------------------------------------------------------------ +; void ff_conv_flt_to_s16(int16_t *dst, const float *src, int len); +;------------------------------------------------------------------------------ + +INIT_XMM sse2 +cglobal conv_flt_to_s16, 3,3,5, dst, src, len + lea lenq, [2*lend] + lea srcq, [srcq+2*lenq] + add dstq, lenq + neg lenq + mova m4, [pf_s16_scale] +.loop: + mova m0, [srcq+2*lenq ] + mova m1, [srcq+2*lenq+1*mmsize] + mova m2, [srcq+2*lenq+2*mmsize] + mova m3, [srcq+2*lenq+3*mmsize] + mulps m0, m4 + mulps m1, m4 + mulps m2, m4 + mulps m3, m4 + cvtps2dq m0, m0 + cvtps2dq m1, m1 + cvtps2dq m2, m2 + cvtps2dq m3, m3 + packssdw m0, m1 + packssdw m2, m3 + mova [dstq+lenq ], m0 + mova [dstq+lenq+mmsize], m2 + add lenq, mmsize*2 + jl .loop + REP_RET + ;----------------------------------------------------------------------------- ; void ff_conv_fltp_to_flt_6ch(float *dst, float *const *src, int len, ; int channels); diff --git a/libavresample/x86/audio_convert_init.c b/libavresample/x86/audio_convert_init.c index f0a1a1a55f..1b2f93c1da 100644 --- a/libavresample/x86/audio_convert_init.c +++ b/libavresample/x86/audio_convert_init.c @@ -33,6 +33,8 @@ extern void ff_conv_s32_to_s16_sse2(int16_t *dst, const int32_t *src, int len); extern void ff_conv_s32_to_flt_sse2(float *dst, const int32_t *src, int len); extern void ff_conv_s32_to_flt_avx (float *dst, const int32_t *src, int len); +extern void ff_conv_flt_to_s16_sse2(int16_t *dst, const float *src, int len); + extern void ff_conv_fltp_to_flt_6ch_mmx (float *dst, float *const *src, int len); extern void ff_conv_fltp_to_flt_6ch_sse4(float *dst, float *const *src, int len); extern void ff_conv_fltp_to_flt_6ch_avx (float *dst, float *const *src, int len); @@ -67,6 +69,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac) 0, 16, 8, "SSE2", ff_conv_s16_to_flt_sse2); ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32, 0, 16, 8, "SSE2", ff_conv_s32_to_flt_sse2); + ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLT, + 0, 16, 16, "SSE2", ff_conv_flt_to_s16_sse2); } if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) { ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,