[FFmpeg-cvslog] swr: float_to_int16_sse2()

[Michael Niedermayer]

ffmpeg | branch: master | Michael Niedermayer <michaelni at gmx.at> | Sun Apr 29 12:18:14 2012 +0200| [73edb58c3c1898bf3b2700c58236736a9e928e6c] | committer: Michael Niedermayer

swr: float_to_int16_sse2()

Signed-off-by: Michael Niedermayer <michaelni at gmx.at>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=73edb58c3c1898bf3b2700c58236736a9e928e6c
---

 libswresample/x86/audio_convert.asm |   33 +++++++++++++++++++++++++++++++++
 libswresample/x86/swresample_x86.c  |    3 +++
 2 files changed, 36 insertions(+), 0 deletions(-)

diff --git a/libswresample/x86/audio_convert.asm b/libswresample/x86/audio_convert.asm
index 99bfdca..1c70f6e 100644
--- a/libswresample/x86/audio_convert.asm
+++ b/libswresample/x86/audio_convert.asm
@@ -25,6 +25,7 @@ SECTION_RODATA
 
 flt2pm31: times 8 dd 4.6566129e-10
 flt2p31 : times 8 dd 2147483648.0
+flt2p15 : times 8 dd 32768.0
 
 SECTION .text
 
@@ -163,6 +164,36 @@ float_to_int32_u_int %+ SUFFIX
     REP_RET
 %endmacro
 
+%macro FLOAT_TO_INT16 1
+cglobal float_to_int16_%1, 3, 3, 3, dst, src, len
+    mov srcq, [srcq]
+    mov dstq, [dstq]
+%ifidn %1, a
+    test dstq, mmsize-1
+        jne float_to_int16_u_int %+ SUFFIX
+    test srcq, mmsize-1
+        jne float_to_int16_u_int %+ SUFFIX
+%else
+float_to_int16_u_int %+ SUFFIX
+%endif
+    lea     srcq, [srcq + 2*lenq]
+    add     dstq, lenq
+    neg     lenq
+    mova      m2, [flt2p15]
+.next:
+    mov%1     m0, [         srcq+2*lenq]
+    mov%1     m1, [mmsize + srcq+2*lenq]
+    mulps m0, m2
+    mulps m1, m2
+    cvtps2dq  m0, m0
+    cvtps2dq  m1, m1
+    packssdw  m0, m1
+    mov%1 [         dstq+lenq], m0
+    add lenq, mmsize
+        jl .next
+    REP_RET
+%endmacro
+
 
 INIT_MMX mmx
 INT16_TO_INT32 u
@@ -179,3 +210,5 @@ INT16_TO_FLOAT u
 INT16_TO_FLOAT a
 FLOAT_TO_INT32 u
 FLOAT_TO_INT32 a
+FLOAT_TO_INT16 u
+FLOAT_TO_INT16 a
diff --git a/libswresample/x86/swresample_x86.c b/libswresample/x86/swresample_x86.c
index 9b5f3cd..c1ffe43 100644
--- a/libswresample/x86/swresample_x86.c
+++ b/libswresample/x86/swresample_x86.c
@@ -29,6 +29,7 @@ MULTI_CAPS_FUNC_DECL(sse)
 void ff_int32_to_float_a_sse2(uint8_t **dst, const uint8_t **src, int len);
 void ff_int16_to_float_a_sse2(uint8_t **dst, const uint8_t **src, int len);
 void ff_float_to_int32_a_sse2(uint8_t **dst, const uint8_t **src, int len);
+void ff_float_to_int16_a_sse2(uint8_t **dst, const uint8_t **src, int len);
 
 void swri_audio_convert_init_x86(struct AudioConvert *ac,
                                  enum AVSampleFormat out_fmt,
@@ -56,5 +57,7 @@ MULTI_CAPS_FUNC(AV_CPU_FLAG_SSE, sse)
             ac->simd_f =  ff_int16_to_float_a_sse2;
         if(   out_fmt == AV_SAMPLE_FMT_S32  && in_fmt == AV_SAMPLE_FMT_FLT || out_fmt == AV_SAMPLE_FMT_S32P && in_fmt == AV_SAMPLE_FMT_FLTP)
             ac->simd_f =  ff_float_to_int32_a_sse2;
+        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 =  ff_float_to_int16_a_sse2;
     }
 }