swr: MMX2 & SSSE3 int16 resample core

about 4 times faster

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

libswresample/resample.c

@@ -356,15 +356,71 @@ int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensatio
 #define OUT(d, v) d = v
 #include "resample_template.c"
 
+#undef RENAME
+#undef FELEM
+#undef FELEM2
+#undef DELEM
+#undef FELEML
+#undef OUT
+#undef FELEM_MIN
+#undef FELEM_MAX
+#undef FILTER_SHIFT
+
+// XXX FIXME the whole C loop should be written in asm so this x86 specific code here isnt needed
+#if ARCH_X86
+#include "x86/resample_mmx.h"
+#define COMMON_CORE COMMON_CORE_INT16_MMX2
+#define RENAME(N) N ## _int16_mmx2
+#define FILTER_SHIFT 15
+#define DELEM  int16_t
+#define FELEM  int16_t
+#define FELEM2 int32_t
+#define FELEML int64_t
+#define FELEM_MAX INT16_MAX
+#define FELEM_MIN INT16_MIN
+#define OUT(d, v) v = (v + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;\
+                  d = (unsigned)(v + 32768) > 65535 ? (v>>31) ^ 32767 : v
+#include "resample_template.c"
+
+#undef COMMON_CORE
+#undef RENAME
+#undef FELEM
+#undef FELEM2
+#undef DELEM
+#undef FELEML
+#undef OUT
+#undef FELEM_MIN
+#undef FELEM_MAX
+#undef FILTER_SHIFT
+
+#define COMMON_CORE COMMON_CORE_INT16_SSSE3
+#define RENAME(N) N ## _int16_ssse3
+#define FILTER_SHIFT 15
+#define DELEM  int16_t
+#define FELEM  int16_t
+#define FELEM2 int32_t
+#define FELEML int64_t
+#define FELEM_MAX INT16_MAX
+#define FELEM_MIN INT16_MIN
+#define OUT(d, v) v = (v + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;\
+                  d = (unsigned)(v + 32768) > 65535 ? (v>>31) ^ 32767 : v
+#include "resample_template.c"
+#endif // ARCH_X86
 
 int swri_multiple_resample(ResampleContext *c, AudioData *dst, int dst_size, AudioData *src, int src_size, int *consumed){
     int i, ret= -1;
+    int mm_flags = av_get_cpu_flags();
 
     for(i=0; i<dst->ch_count; i++){
-        if(c->format == AV_SAMPLE_FMT_S16P) ret= swri_resample_int16(c, (int16_t*)dst->ch[i], (const int16_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
-        if(c->format == AV_SAMPLE_FMT_S32P) ret= swri_resample_int32(c, (int32_t*)dst->ch[i], (const int32_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
-        if(c->format == AV_SAMPLE_FMT_FLTP) ret= swri_resample_float(c, (float  *)dst->ch[i], (const float  *)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
-        if(c->format == AV_SAMPLE_FMT_DBLP) ret= swri_resample_double(c,(double *)dst->ch[i], (const double *)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
+#if ARCH_X86
+             if(c->format == AV_SAMPLE_FMT_S16P && (mm_flags&AV_CPU_FLAG_SSSE3)) ret= swri_resample_int16_ssse3(c, (int16_t*)dst->ch[i], (const int16_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
+        else if(c->format == AV_SAMPLE_FMT_S16P && (mm_flags&AV_CPU_FLAG_MMX2 )) ret= swri_resample_int16_mmx2 (c, (int16_t*)dst->ch[i], (const int16_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
+        else
+#endif
+             if(c->format == AV_SAMPLE_FMT_S16P) ret= swri_resample_int16(c, (int16_t*)dst->ch[i], (const int16_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
+        else if(c->format == AV_SAMPLE_FMT_S32P) ret= swri_resample_int32(c, (int32_t*)dst->ch[i], (const int32_t*)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
+        else if(c->format == AV_SAMPLE_FMT_FLTP) ret= swri_resample_float(c, (float  *)dst->ch[i], (const float  *)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
+        else if(c->format == AV_SAMPLE_FMT_DBLP) ret= swri_resample_double(c,(double *)dst->ch[i], (const double *)src->ch[i], consumed, src_size, dst_size, i+1==dst->ch_count);
     }
 
     return ret;

libswresample/resample_template.c

@@ -56,11 +56,15 @@ int RENAME(swri_resample)(ResampleContext *c, DELEM *dst, const DELEM *src, int
             if(sample_index + c->filter_length > src_size){
                 break;
             }else{
+#ifdef COMMON_CORE
+                COMMON_CORE
+#else
                 FELEM2 val=0;
                 for(i=0; i<c->filter_length; i++){
                     val += src[sample_index + i] * (FELEM2)filter[i];
                 }
                 OUT(dst[dst_index], val);
+#endif
             }
 
             frac += dst_incr_frac;

libswresample/x86/resample_mmx.h

@@ -0,0 +1,69 @@
+/*
+ * Copyright (c) 2012 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * 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_cpu.h"
+#include "libavutil/cpu.h"
+
+int swri_resample_int16_mmx2 (struct ResampleContext *c, int16_t *dst, const int16_t *src, int *consumed, int src_size, int dst_size, int update_ctx);
+int swri_resample_int16_ssse3(struct ResampleContext *c, int16_t *dst, const int16_t *src, int *consumed, int src_size, int dst_size, int update_ctx);
+
+DECLARE_ALIGNED(16, const uint64_t, ff_resample_int16_rounder)[2]    = { 0x0000000000004000ULL, 0x0000000000000000ULL};
+
+#define COMMON_CORE_INT16_MMX2 \
+    x86_reg len= -2*c->filter_length;\
+__asm volatile(\
+    "movq "MANGLE(ff_resample_int16_rounder)", %%mm0 \n\t"\
+    "1:                         \n\t"\
+    "movq    (%1, %0), %%mm1    \n\t"\
+    "pmaddwd (%2, %0), %%mm1    \n\t"\
+    "paddd  %%mm1, %%mm0        \n\t"\
+    "add       $8, %0           \n\t"\
+    " js 1b                     \n\t"\
+    "pshufw $0x0E, %%mm0, %%mm1 \n\t"\
+    "paddd %%mm1, %%mm0 \n\t"\
+    "psrad    $15, %%mm0        \n\t"\
+    "packssdw %%mm0, %%mm0      \n\t"\
+    "movd %%mm0, (%3)           \n\t"\
+    : "+r" (len)\
+    : "r" (((uint8_t*)(src+sample_index))-len),\
+      "r" (((uint8_t*)filter)-len),\
+      "r" (dst+dst_index)\
+);
+
+#define COMMON_CORE_INT16_SSSE3 \
+    x86_reg len= -2*c->filter_length;\
+__asm volatile(\
+    "movdqa "MANGLE(ff_resample_int16_rounder)", %%xmm0 \n\t"\
+    "1:                           \n\t"\
+    "movdqu  (%1, %0), %%xmm1     \n\t"\
+    "pmaddwd (%2, %0), %%xmm1     \n\t"\
+    "paddd  %%xmm1, %%xmm0        \n\t"\
+    "add       $16, %0            \n\t"\
+    " js 1b                       \n\t"\
+    "phaddd %%xmm0, %%xmm0        \n\t"\
+    "phaddd %%xmm0, %%xmm0        \n\t"\
+    "psrad    $15, %%xmm0         \n\t"\
+    "packssdw %%xmm0, %%xmm0      \n\t"\
+    "movd %%xmm0, (%3)            \n\t"\
+    : "+r" (len)\
+    : "r" (((uint8_t*)(src+sample_index))-len),\
+      "r" (((uint8_t*)filter)-len),\
+      "r" (dst+dst_index)\
+);