swresample: change COMMON_CORE_INT16 asm from SSSE3 to SSE2

pshuf+paddd is slightly faster than phaddd.
The real gain is in pre-ssse3 processors like AMD K8 and K10, which get
a big boost in performance compared to the mmxext version
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

libswresample/resample.c

@@ -302,10 +302,10 @@ static int set_compensation(ResampleContext *c, int sample_delta, int compensati
 #include "resample_template.c"
 #undef TEMPLATE_RESAMPLE_S16_MMX2
 
-#if HAVE_SSSE3_INLINE
-#define TEMPLATE_RESAMPLE_S16_SSSE3
+#if HAVE_SSE2_INLINE
+#define TEMPLATE_RESAMPLE_S16_SSE2
 #include "resample_template.c"
-#undef TEMPLATE_RESAMPLE_S16_SSSE3
+#undef TEMPLATE_RESAMPLE_S16_SSE2
 #endif
 
 #endif // HAVE_MMXEXT_INLINE
@@ -317,8 +317,8 @@ static int multiple_resample(ResampleContext *c, AudioData *dst, int dst_size, A
 
     for(i=0; i<dst->ch_count; i++){
 #if HAVE_MMXEXT_INLINE
-#if HAVE_SSSE3_INLINE
-             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);
+#if HAVE_SSE2_INLINE
+             if(c->format == AV_SAMPLE_FMT_S16P && (mm_flags&AV_CPU_FLAG_SSE2)) ret= swri_resample_int16_sse2 (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 && (mm_flags&AV_CPU_FLAG_MMX2 )){

libswresample/resample_template.c

@@ -57,7 +57,7 @@
 
 #elif    defined(TEMPLATE_RESAMPLE_S16)      \
       || defined(TEMPLATE_RESAMPLE_S16_MMX2) \
-      || defined(TEMPLATE_RESAMPLE_S16_SSSE3)
+      || defined(TEMPLATE_RESAMPLE_S16_SSE2)
 
 #    define FILTER_SHIFT 15
 #    define DELEM  int16_t
@@ -74,9 +74,9 @@
 #    elif defined(TEMPLATE_RESAMPLE_S16_MMX2)
 #        define COMMON_CORE COMMON_CORE_INT16_MMX2
 #        define RENAME(N) N ## _int16_mmx2
-#    elif defined(TEMPLATE_RESAMPLE_S16_SSSE3)
-#        define COMMON_CORE COMMON_CORE_INT16_SSSE3
-#        define RENAME(N) N ## _int16_ssse3
+#    elif defined(TEMPLATE_RESAMPLE_S16_SSE2)
+#        define COMMON_CORE COMMON_CORE_INT16_SSE2
+#        define RENAME(N) N ## _int16_sse2
 #    endif
 
 #endif

libswresample/x86/resample_mmx.h

@@ -23,7 +23,7 @@
 #include "libswresample/swresample_internal.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);
+int swri_resample_int16_sse2 (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};
 
@@ -48,7 +48,7 @@ __asm__ volatile(\
       "r" (dst+dst_index)\
 );
 
-#define COMMON_CORE_INT16_SSSE3 \
+#define COMMON_CORE_INT16_SSE2 \
     x86_reg len= -2*c->filter_length;\
 __asm__ volatile(\
     "movdqa "MANGLE(ff_resample_int16_rounder)", %%xmm0 \n\t"\
@@ -58,8 +58,10 @@ __asm__ volatile(\
     "paddd  %%xmm1, %%xmm0        \n\t"\
     "add       $16, %0            \n\t"\
     " js 1b                       \n\t"\
-    "phaddd %%xmm0, %%xmm0        \n\t"\
-    "phaddd %%xmm0, %%xmm0        \n\t"\
+    "pshufd $0x0E, %%xmm0, %%xmm1 \n\t"\
+    "paddd %%xmm1, %%xmm0         \n\t"\
+    "pshufd $0x01, %%xmm0, %%xmm1 \n\t"\
+    "paddd %%xmm1, %%xmm0         \n\t"\
     "psrad    $15, %%xmm0         \n\t"\
     "packssdw %%xmm0, %%xmm0      \n\t"\
     "movd %%xmm0, (%3)            \n\t"\