[FFmpeg-cvslog] lavr: Add x86-optimized functions for s16 to flt conversion

[Justin Ruggles]

ffmpeg | branch: master | Justin Ruggles <justin.ruggles at gmail.com> | Mon Apr 23 22:10:26 2012 -0400| [d721f67d0a7faff2cc501f186ef78e6b01394d59] | committer: Justin Ruggles

lavr: Add x86-optimized functions for s16 to flt conversion

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=d721f67d0a7faff2cc501f186ef78e6b01394d59
---

 libavresample/x86/audio_convert.asm    |   36 ++++++++++++++++++++++++++++++++
 libavresample/x86/audio_convert_init.c |    9 ++++++++
 2 files changed, 45 insertions(+)

diff --git a/libavresample/x86/audio_convert.asm b/libavresample/x86/audio_convert.asm
index e2cfbf9..6e14892 100644
--- a/libavresample/x86/audio_convert.asm
+++ b/libavresample/x86/audio_convert.asm
@@ -22,6 +22,11 @@
 
 %include "x86inc.asm"
 %include "x86util.asm"
+%include "util.asm"
+
+SECTION_RODATA 32
+
+pf_s16_inv_scale: times 4 dd 0x38000000
 
 SECTION_TEXT
 
@@ -47,6 +52,37 @@ cglobal conv_s16_to_s32, 3,3,3, dst, src, len
     jl .loop
     REP_RET
 
+;------------------------------------------------------------------------------
+; void ff_conv_s16_to_flt(float *dst, const int16_t *src, int len);
+;------------------------------------------------------------------------------
+
+%macro CONV_S16_TO_FLT 0
+cglobal conv_s16_to_flt, 3,3,3, dst, src, len
+    lea      lenq, [2*lend]
+    add      srcq, lenq
+    lea      dstq, [dstq + 2*lenq]
+    neg      lenq
+    mova       m2, [pf_s16_inv_scale]
+    ALIGN 16
+.loop:
+    mova       m0, [srcq+lenq]
+    S16_TO_S32_SX 0, 1
+    cvtdq2ps   m0, m0
+    cvtdq2ps   m1, m1
+    mulps      m0, m2
+    mulps      m1, m2
+    mova  [dstq+2*lenq       ], m0
+    mova  [dstq+2*lenq+mmsize], m1
+    add      lenq, mmsize
+    jl .loop
+    REP_RET
+%endmacro
+
+INIT_XMM sse2
+CONV_S16_TO_FLT
+INIT_XMM sse4
+CONV_S16_TO_FLT
+
 ;-----------------------------------------------------------------------------
 ; 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 a1dac7e..2488cd3 100644
--- a/libavresample/x86/audio_convert_init.c
+++ b/libavresample/x86/audio_convert_init.c
@@ -24,6 +24,9 @@
 
 extern void ff_conv_s16_to_s32_sse2(int16_t *dst, const int32_t *src, int len);
 
+extern void ff_conv_s16_to_flt_sse2(float *dst, const int16_t *src, int len);
+extern void ff_conv_s16_to_flt_sse4(float *dst, const int16_t *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);
@@ -48,6 +51,12 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
     if (mm_flags & AV_CPU_FLAG_SSE2 && HAVE_SSE) {
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S16,
                                   0, 16, 8, "SSE2", ff_conv_s16_to_s32_sse2);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,
+                                  0, 16, 8, "SSE2", ff_conv_s16_to_flt_sse2);
+    }
+    if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,
+                                  0, 16, 8, "SSE4", ff_conv_s16_to_flt_sse4);
     }
 #endif
 }