[FFmpeg-cvslog] lavr: Add x86-optimized functions for s32 to flt conversion

[Justin Ruggles]

ffmpeg | branch: master | Justin Ruggles <justin.ruggles at gmail.com> | Mon Apr 23 18:29:58 2012 -0400| [97ce1ba8673636c96d3cc002bb76221c60324d95] | committer: Justin Ruggles

lavr: Add x86-optimized functions for s32 to flt conversion

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=97ce1ba8673636c96d3cc002bb76221c60324d95
---

 libavresample/x86/audio_convert.asm    |   37 ++++++++++++++++++++++++++++++++
 libavresample/x86/audio_convert_init.c |    9 ++++++++
 2 files changed, 46 insertions(+)

diff --git a/libavresample/x86/audio_convert.asm b/libavresample/x86/audio_convert.asm
index 8435ede..53091ac 100644
--- a/libavresample/x86/audio_convert.asm
+++ b/libavresample/x86/audio_convert.asm
@@ -26,6 +26,7 @@
 
 SECTION_RODATA 32
 
+pf_s32_inv_scale: times 8 dd 0x30000000
 pf_s16_inv_scale: times 4 dd 0x38000000
 
 SECTION_TEXT
@@ -121,6 +122,42 @@ CONV_S32_TO_S16
 INIT_XMM sse2
 CONV_S32_TO_S16
 
+;------------------------------------------------------------------------------
+; void ff_conv_s32_to_flt(float *dst, const int32_t *src, int len);
+;------------------------------------------------------------------------------
+
+%macro CONV_S32_TO_FLT 0
+cglobal conv_s32_to_flt, 3,3,3, dst, src, len
+    lea     lenq, [4*lend]
+    add     srcq, lenq
+    add     dstq, lenq
+    neg     lenq
+    mova      m0, [pf_s32_inv_scale]
+    ALIGN 16
+.loop:
+    cvtdq2ps  m1, [srcq+lenq       ]
+    cvtdq2ps  m2, [srcq+lenq+mmsize]
+    mulps     m1, m1, m0
+    mulps     m2, m2, m0
+    mova  [dstq+lenq       ], m1
+    mova  [dstq+lenq+mmsize], m2
+    add     lenq, mmsize*2
+    jl .loop
+%if mmsize == 32
+    vzeroupper
+    RET
+%else
+    REP_RET
+%endif
+%endmacro
+
+INIT_XMM sse2
+CONV_S32_TO_FLT
+%if HAVE_AVX
+INIT_YMM avx
+CONV_S32_TO_FLT
+%endif
+
 ;-----------------------------------------------------------------------------
 ; 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 c4fe3ae..f0a1a1a 100644
--- a/libavresample/x86/audio_convert_init.c
+++ b/libavresample/x86/audio_convert_init.c
@@ -30,6 +30,9 @@ extern void ff_conv_s16_to_flt_sse4(float *dst, const int16_t *src, int len);
 extern void ff_conv_s32_to_s16_mmx (int16_t *dst, const int32_t *src, int len);
 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_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);
@@ -62,10 +65,16 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
                                   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);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32,
+                                  0, 16, 8, "SSE2", ff_conv_s32_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);
     }
+    if (mm_flags & AV_CPU_FLAG_AVX && HAVE_AVX) {
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32,
+                                  0, 32, 16, "AVX", ff_conv_s32_to_flt_avx);
+    }
 #endif
 }