swr: SIMD rematrixing and SSE/AVX mix_1_1 float

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

5 files changed, 120 insertions, 2 deletions

diff --git a/libswresample/rematrix.c b/libswresample/rematrix.c
index a7d000c016..18e89c96ba 100644
--- a/libswresample/rematrix.c
+++ b/libswresample/rematrix.c
@@ -340,6 +340,9 @@ int swri_rematrix_init(SwrContext *s){
         }
         s->matrix_ch[i][0]= ch_in;
     }
+
+    if(HAVE_YASM && HAVE_MMX) swri_rematrix_init_x86(s);
+
     return 0;
 }
 
@@ -351,12 +354,19 @@ void swri_rematrix_free(SwrContext *s){
 
 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy){
     int out_i, in_i, i, j;
+    int len1 = 0;
+    int off = 0;
 
     if(s->mix_any_f) {
         s->mix_any_f(out->ch, in->ch, s->native_matrix, len);
         return 0;
     }
 
+    if(s->mix_2_1_simd || s->mix_1_1_simd){
+        len1= len&~15;
+        off = len1 * out->bps;
+    }
+
     av_assert0(out->ch_count == av_get_channel_layout_nb_channels(s->out_ch_layout));
     av_assert0(in ->ch_count == av_get_channel_layout_nb_channels(s-> in_ch_layout));
 
@@ -369,7 +379,10 @@ int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mus
         case 1:
             in_i= s->matrix_ch[out_i][1];
             if(s->matrix[out_i][in_i]!=1.0){
-                s->mix_1_1_f(out->ch[out_i], in->ch[in_i], s->native_matrix, in->ch_count*out_i + in_i, len);
+                if(s->mix_1_1_simd && len1)
+                    s->mix_1_1_simd(out->ch[out_i]    , in->ch[in_i]    , s->native_matrix, in->ch_count*out_i + in_i, len1);
+                if(len != len1)
+                    s->mix_1_1_f   (out->ch[out_i]+off, in->ch[in_i]+off, s->native_matrix, in->ch_count*out_i + in_i, len-len1);
             }else if(mustcopy){
                 memcpy(out->ch[out_i], in->ch[in_i], len*out->bps);
             }else{
@@ -379,7 +392,12 @@ int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mus
         case 2: {
             int in_i1 = s->matrix_ch[out_i][1];
             int in_i2 = s->matrix_ch[out_i][2];
-            s->mix_2_1_f(out->ch[out_i], in->ch[in_i1], in->ch[in_i2], s->native_matrix, in->ch_count*out_i + in_i1, in->ch_count*out_i + in_i2, len);
+            if(s->mix_2_1_simd && len1)
+                s->mix_2_1_simd(out->ch[out_i]    , in->ch[in_i1]    , in->ch[in_i2]    , s->native_matrix, in->ch_count*out_i + in_i1, in->ch_count*out_i + in_i2, len1);
+            else
+                s->mix_2_1_f   (out->ch[out_i]    , in->ch[in_i1]    , in->ch[in_i2]    , s->native_matrix, in->ch_count*out_i + in_i1, in->ch_count*out_i + in_i2, len1);
+            if(len != len1)
+                s->mix_2_1_f   (out->ch[out_i]+off, in->ch[in_i1]+off, in->ch[in_i2]+off, s->native_matrix, in->ch_count*out_i + in_i1, in->ch_count*out_i + in_i2, len-len1);
             break;}
         default:
             if(s->int_sample_fmt == AV_SAMPLE_FMT_FLTP){
diff --git a/libswresample/swresample_internal.h b/libswresample/swresample_internal.h
index 2d4240fba9..ddcecf1f45 100644
--- a/libswresample/swresample_internal.h
+++ b/libswresample/swresample_internal.h
@@ -121,6 +121,7 @@ int swri_resample_double(struct ResampleContext *c,double  *dst, const double  *
 int swri_rematrix_init(SwrContext *s);
 void swri_rematrix_free(SwrContext *s);
 int swri_rematrix(SwrContext *s, AudioData *out, AudioData *in, int len, int mustcopy);
+void swri_rematrix_init_x86(struct SwrContext *s);
 
 void swri_get_dither(SwrContext *s, void *dst, int len, unsigned seed, enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt);
 
diff --git a/libswresample/x86/Makefile b/libswresample/x86/Makefile
index 1ba971e02d..e8feede6b6 100644
--- a/libswresample/x86/Makefile
+++ b/libswresample/x86/Makefile
@@ -1,2 +1,3 @@
 YASM-OBJS                       += x86/swresample_x86.o\
                                    x86/audio_convert.o\
+                                   x86/rematrix.o\
diff --git a/libswresample/x86/rematrix.asm b/libswresample/x86/rematrix.asm
new file mode 100644
index 0000000000..aacd3167cb
--- /dev/null
+++ b/libswresample/x86/rematrix.asm
@@ -0,0 +1,66 @@
+;******************************************************************************
+;* Copyright (c) 2012 Michael Niedermayer
+;*
+;* This file is part of FFmpeg.
+;*
+;* FFmpeg is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* FFmpeg is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with FFmpeg; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86inc.asm"
+%include "libavutil/x86/x86util.asm"
+
+SECTION .text
+
+%macro MIX1_FLT 1
+cglobal mix_1_1_%1_float, 5, 5, 3, out, in, coeffp, index, len
+%ifidn %1, a
+    test inq, mmsize-1
+        jne mix_1_1_float_u_int %+ SUFFIX
+    test outq, mmsize-1
+        jne mix_1_1_float_u_int %+ SUFFIX
+%else
+mix_1_1_float_u_int %+ SUFFIX
+%endif
+    VBROADCASTSS m2, [coeffpq + 4*indexq]
+    shl lenq    , 2
+    add inq     , lenq
+    add outq    , lenq
+    neg lenq
+.next:
+%ifidn %1, a
+    mulps        m0, m2, [inq + lenq         ]
+    mulps        m1, m2, [inq + lenq + mmsize]
+%else
+    movu         m0, [inq + lenq         ]
+    movu         m1, [inq + lenq + mmsize]
+    mulps        m0, m0, m2
+    mulps        m1, m1, m2
+%endif
+    mov%1  [outq + lenq         ], m0
+    mov%1  [outq + lenq + mmsize], m1
+    add        lenq, mmsize*2
+        jl .next
+    REP_RET
+%endmacro
+
+INIT_XMM sse
+MIX1_FLT u
+MIX1_FLT a
+
+%if HAVE_AVX
+INIT_YMM avx
+MIX1_FLT u
+MIX1_FLT a
+%endif
diff --git a/libswresample/x86/swresample_x86.c b/libswresample/x86/swresample_x86.c
index 6cd6073e9c..f523dd0ad2 100644
--- a/libswresample/x86/swresample_x86.c
+++ b/libswresample/x86/swresample_x86.c
@@ -142,3 +142,35 @@ MULTI_CAPS_FUNC(AV_CPU_FLAG_SSE, sse)
         }
     }
 }
+
+#define D(type, simd) \
+mix_1_1_func_type ff_mix_1_1_a_## type ## _ ## simd;\
+mix_2_1_func_type ff_mix_2_1_a_## type ## _ ## simd;
+
+D(float, sse)
+D(float, avx)
+D(int16, mmx)
+
+
+void swri_rematrix_init_x86(struct SwrContext *s){
+    int mm_flags = av_get_cpu_flags();
+    int nb_in  = av_get_channel_layout_nb_channels(s->in_ch_layout);
+    int nb_out = av_get_channel_layout_nb_channels(s->out_ch_layout);
+    int num    = nb_in * nb_out;
+    int i,j;
+
+    s->mix_1_1_simd = NULL;
+    s->mix_2_1_simd = NULL;
+
+    if (s->midbuf.fmt == AV_SAMPLE_FMT_S16P){
+    } else if(s->midbuf.fmt == AV_SAMPLE_FMT_FLTP){
+        if(mm_flags & AV_CPU_FLAG_SSE) {
+            s->mix_1_1_simd = ff_mix_1_1_a_float_sse;
+        }
+        if(HAVE_AVX && mm_flags & AV_CPU_FLAG_AVX) {
+            s->mix_1_1_simd = ff_mix_1_1_a_float_avx;
+        }
+        s->native_simd_matrix = av_mallocz(num * sizeof(float));
+        memcpy(s->native_simd_matrix, s->native_matrix, num * sizeof(float));
+    }
+}