[FFmpeg-cvslog] lavr: Add x86-optimized functions for s32 to s16 conversion

[Justin Ruggles]

ffmpeg | branch: master | Justin Ruggles <justin.ruggles at gmail.com> | Fri Apr 20 13:49:53 2012 -0400| [5904f25b9f0bc5d79685b37c9befb8d12e430352] | committer: Justin Ruggles

lavr: Add x86-optimized functions for s32 to s16 conversion

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=5904f25b9f0bc5d79685b37c9befb8d12e430352
---

 libavresample/x86/audio_convert.asm    |   38 ++++++++++++++++++++++++++++++++
 libavresample/x86/audio_convert_init.c |    9 ++++++++
 2 files changed, 47 insertions(+)

diff --git a/libavresample/x86/audio_convert.asm b/libavresample/x86/audio_convert.asm
index 6e14892..8435ede 100644
--- a/libavresample/x86/audio_convert.asm
+++ b/libavresample/x86/audio_convert.asm
@@ -83,6 +83,44 @@ CONV_S16_TO_FLT
 INIT_XMM sse4
 CONV_S16_TO_FLT
 
+;------------------------------------------------------------------------------
+; void ff_conv_s32_to_s16(int16_t *dst, const int32_t *src, int len);
+;------------------------------------------------------------------------------
+
+%macro CONV_S32_TO_S16 0
+cglobal conv_s32_to_s16, 3,3,4, dst, src, len
+    lea     lenq, [2*lend]
+    lea     srcq, [srcq+2*lenq]
+    add     dstq, lenq
+    neg     lenq
+.loop:
+    mova      m0, [srcq+2*lenq         ]
+    mova      m1, [srcq+2*lenq+  mmsize]
+    mova      m2, [srcq+2*lenq+2*mmsize]
+    mova      m3, [srcq+2*lenq+3*mmsize]
+    psrad     m0, 16
+    psrad     m1, 16
+    psrad     m2, 16
+    psrad     m3, 16
+    packssdw  m0, m1
+    packssdw  m2, m3
+    mova  [dstq+lenq       ], m0
+    mova  [dstq+lenq+mmsize], m2
+    add     lenq, mmsize*2
+    jl .loop
+%if mmsize == 8
+    emms
+    RET
+%else
+    REP_RET
+%endif
+%endmacro
+
+INIT_MMX mmx
+CONV_S32_TO_S16
+INIT_XMM sse2
+CONV_S32_TO_S16
+
 ;-----------------------------------------------------------------------------
 ; 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 2488cd3..c4fe3ae 100644
--- a/libavresample/x86/audio_convert_init.c
+++ b/libavresample/x86/audio_convert_init.c
@@ -27,6 +27,9 @@ 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_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_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);
@@ -37,6 +40,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
     int mm_flags = av_get_cpu_flags();
 
     if (mm_flags & AV_CPU_FLAG_MMX && HAVE_MMX) {
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S32,
+                                  0, 1, 8, "MMX", ff_conv_s32_to_s16_mmx);
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
                                   6, 1, 4, "MMX", ff_conv_fltp_to_flt_6ch_mmx);
     }
@@ -49,6 +54,10 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
                                   6, 16, 4, "AVX", ff_conv_fltp_to_flt_6ch_avx);
     }
     if (mm_flags & AV_CPU_FLAG_SSE2 && HAVE_SSE) {
+        if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
+            ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S32,
+                                      0, 16, 16, "SSE2", ff_conv_s32_to_s16_sse2);
+        }
         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,