[FFmpeg-cvslog] lavr: x86: optimized 6-channel s16 to fltp conversion

Justin Ruggles git at videolan.org
Fri Aug 24 14:44:30 CEST 2012


ffmpeg | branch: master | Justin Ruggles <justin.ruggles at gmail.com> | Wed May  2 18:03:11 2012 -0400| [6092dafb5a542a3014f9ae7f692085f0e1027e57] | committer: Justin Ruggles

lavr: x86: optimized 6-channel s16 to fltp conversion

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=6092dafb5a542a3014f9ae7f692085f0e1027e57
---

 libavresample/x86/audio_convert.asm    |   83 ++++++++++++++++++++++++++++++++
 libavresample/x86/audio_convert_init.c |   15 ++++++
 libavutil/x86/x86util.asm              |   12 +++++
 3 files changed, 110 insertions(+)

diff --git a/libavresample/x86/audio_convert.asm b/libavresample/x86/audio_convert.asm
index 9f8a13c..12bd577 100644
--- a/libavresample/x86/audio_convert.asm
+++ b/libavresample/x86/audio_convert.asm
@@ -962,3 +962,86 @@ CONV_S16_TO_FLTP_2CH
 INIT_XMM avx
 CONV_S16_TO_FLTP_2CH
 %endif
+
+;------------------------------------------------------------------------------
+; void ff_conv_s16_to_fltp_6ch(float *const *dst, int16_t *src, int len,
+;                              int channels);
+;------------------------------------------------------------------------------
+
+%macro CONV_S16_TO_FLTP_6CH 0
+%if ARCH_X86_64
+cglobal conv_s16_to_fltp_6ch, 3,8,7, dst, src, len, dst1, dst2, dst3, dst4, dst5
+%else
+cglobal conv_s16_to_fltp_6ch, 2,7,7, dst, src, dst1, dst2, dst3, dst4, dst5
+%define lend dword r2m
+%endif
+    mov     dst1q, [dstq+  gprsize]
+    mov     dst2q, [dstq+2*gprsize]
+    mov     dst3q, [dstq+3*gprsize]
+    mov     dst4q, [dstq+4*gprsize]
+    mov     dst5q, [dstq+5*gprsize]
+    mov      dstq, [dstq          ]
+    sub     dst1q, dstq
+    sub     dst2q, dstq
+    sub     dst3q, dstq
+    sub     dst4q, dstq
+    sub     dst5q, dstq
+    mova       m6, [pf_s16_inv_scale]
+.loop:
+    mova       m0, [srcq+0*mmsize]  ; m0 =  0,  1,  2,  3,  4,  5,  6,  7
+    mova       m3, [srcq+1*mmsize]  ; m3 =  8,  9, 10, 11, 12, 13, 14, 15
+    mova       m2, [srcq+2*mmsize]  ; m2 = 16, 17, 18, 19, 20, 21, 22, 23
+    PALIGNR    m1, m3, m0, 12, m4   ; m1 =  6,  7,  8,  9, 10, 11,  x,  x
+    shufps     m3, m2, q1032        ; m3 = 12, 13, 14, 15, 16, 17, 18, 19
+    psrldq     m2, 4                ; m2 = 18, 19, 20, 21, 22, 23,  x,  x
+    SBUTTERFLY2 wd, 0, 1, 4         ; m0 =  0,  6,  1,  7,  2,  8,  3,  9
+                                    ; m1 =  4, 10,  5, 11,  x,  x,  x,  x
+    SBUTTERFLY2 wd, 3, 2, 4         ; m3 = 12, 18, 13, 19, 14, 20, 15, 21
+                                    ; m2 = 16, 22, 17, 23,  x,  x,  x,  x
+    SBUTTERFLY2 dq, 0, 3, 4         ; m0 =  0,  6, 12, 18,  1,  7, 13, 19
+                                    ; m3 =  2,  8, 14, 20,  3,  9, 15, 21
+    punpckldq  m1, m2               ; m1 =  4, 10, 16, 22,  5, 11, 17, 23
+    S16_TO_S32_SX 0, 2              ; m0 =      0,      6,     12,     18
+                                    ; m2 =      1,      7,     13,     19
+    S16_TO_S32_SX 3, 4              ; m3 =      2,      8,     14,     20
+                                    ; m4 =      3,      9,     15,     21
+    S16_TO_S32_SX 1, 5              ; m1 =      4,     10,     16,     22
+                                    ; m5 =      5,     11,     17,     23
+    SWAP 1,2,3,4
+    cvtdq2ps   m0, m0
+    cvtdq2ps   m1, m1
+    cvtdq2ps   m2, m2
+    cvtdq2ps   m3, m3
+    cvtdq2ps   m4, m4
+    cvtdq2ps   m5, m5
+    mulps      m0, m6
+    mulps      m1, m6
+    mulps      m2, m6
+    mulps      m3, m6
+    mulps      m4, m6
+    mulps      m5, m6
+    mova  [dstq      ], m0
+    mova  [dstq+dst1q], m1
+    mova  [dstq+dst2q], m2
+    mova  [dstq+dst3q], m3
+    mova  [dstq+dst4q], m4
+    mova  [dstq+dst5q], m5
+    add      srcq, mmsize*3
+    add      dstq, mmsize
+    sub      lend, mmsize/4
+    jg .loop
+    REP_RET
+%endmacro
+
+%define PALIGNR PALIGNR_MMX
+INIT_XMM sse2
+CONV_S16_TO_FLTP_6CH
+%define PALIGNR PALIGNR_SSSE3
+INIT_XMM ssse3
+CONV_S16_TO_FLTP_6CH
+INIT_XMM sse4
+CONV_S16_TO_FLTP_6CH
+%if HAVE_AVX
+INIT_XMM avx
+CONV_S16_TO_FLTP_6CH
+%endif
diff --git a/libavresample/x86/audio_convert_init.c b/libavresample/x86/audio_convert_init.c
index d57d8f3..8d15824 100644
--- a/libavresample/x86/audio_convert_init.c
+++ b/libavresample/x86/audio_convert_init.c
@@ -111,6 +111,15 @@ extern void ff_conv_s16_to_fltp_2ch_sse2(float *const *dst, int16_t *src,
 extern void ff_conv_s16_to_fltp_2ch_avx (float *const *dst, int16_t *src,
                                          int len, int channels);
 
+extern void ff_conv_s16_to_fltp_6ch_sse2 (float *const *dst, int16_t *src,
+                                          int len, int channels);
+extern void ff_conv_s16_to_fltp_6ch_ssse3(float *const *dst, int16_t *src,
+                                          int len, int channels);
+extern void ff_conv_s16_to_fltp_6ch_sse4 (float *const *dst, int16_t *src,
+                                          int len, int channels);
+extern void ff_conv_s16_to_fltp_6ch_avx  (float *const *dst, int16_t *src,
+                                          int len, int channels);
+
 av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
 {
 #if HAVE_YASM
@@ -164,6 +173,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
                                   6, 16, 4, "SSE2", ff_conv_s16_to_s16p_6ch_sse2);
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
                                   2, 16, 8, "SSE2", ff_conv_s16_to_fltp_2ch_sse2);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
+                                  6, 16, 4, "SSE2", ff_conv_s16_to_fltp_6ch_sse2);
     }
     if (mm_flags & AV_CPU_FLAG_SSSE3 && HAVE_SSE) {
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
@@ -174,6 +185,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
                                   2, 16, 8, "SSSE3", ff_conv_s16_to_s16p_2ch_ssse3);
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
                                   6, 16, 4, "SSSE3", ff_conv_s16_to_s16p_6ch_ssse3);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
+                                  6, 16, 4, "SSSE3", ff_conv_s16_to_fltp_6ch_ssse3);
     }
     if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,
@@ -204,6 +217,8 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
                                   6, 16, 4, "AVX", ff_conv_s16_to_s16p_6ch_avx);
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
                                   2, 16, 8, "AVX", ff_conv_s16_to_fltp_2ch_avx);
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
+                                  6, 16, 4, "AVX", ff_conv_s16_to_fltp_6ch_avx);
     }
 #endif
 }
diff --git a/libavutil/x86/x86util.asm b/libavutil/x86/x86util.asm
index 8450acd..a5d89a1 100644
--- a/libavutil/x86/x86util.asm
+++ b/libavutil/x86/x86util.asm
@@ -637,3 +637,15 @@
         %rotate 1
     %endrep
 %endmacro
+
+%macro PMOVSXWD 2; dst, src
+%if cpuflag(sse4)
+    pmovsxwd     %1, %2
+%else
+    %ifnidn %1, %2
+    mova         %1, %2
+    %endif
+    punpcklwd    %1, %1
+    psrad        %1, 16
+%endif
+%endmacro



More information about the ffmpeg-cvslog mailing list