lavr: replace the SSE version of ff_conv_fltp_to_flt_6ch() with SSE4 and AVX

The current SSE version is slower than the MMX version on Athlon64 and Sandy
Bridge, but the SSE4 and AVX versions are faster on Sandy Bridge.

libavresample/x86/audio_convert.asm

@@ -54,26 +54,24 @@ cglobal conv_fltp_to_flt_6ch, 2,8,7, dst, src, src1, src2, src3, src4, src5, len
     mova      m3, [srcq+src3q]
     mova      m4, [srcq+src4q]
     mova      m5, [srcq+src5q]
-%if cpuflag(sse)
+%if cpuflag(sse4)
     SBUTTERFLYPS 0, 1, 6
     SBUTTERFLYPS 2, 3, 6
     SBUTTERFLYPS 4, 5, 6
 
-    movaps    m6, m4
-    shufps    m4, m0, q3210
+    blendps   m6, m4, m0, 1100b
     movlhps   m0, m2
-    movhlps   m6, m2
-    movaps [dstq   ], m0
-    movaps [dstq+16], m4
-    movaps [dstq+32], m6
-
-    movaps    m6, m5
-    shufps    m5, m1, q3210
+    movhlps   m4, m2
+    blendps   m2, m5, m1, 1100b
     movlhps   m1, m3
-    movhlps   m6, m3
+    movhlps   m5, m3
+
+    movaps [dstq   ], m0
+    movaps [dstq+16], m6
+    movaps [dstq+32], m4
     movaps [dstq+48], m1
-    movaps [dstq+64], m5
-    movaps [dstq+80], m6
+    movaps [dstq+64], m2
+    movaps [dstq+80], m5
 %else ; mmx
     SBUTTERFLY dq, 0, 1, 6
     SBUTTERFLY dq, 2, 3, 6
@@ -100,5 +98,9 @@ cglobal conv_fltp_to_flt_6ch, 2,8,7, dst, src, src1, src2, src3, src4, src5, len
 
 INIT_MMX mmx
 CONV_FLTP_TO_FLT_6CH
-INIT_XMM sse
+INIT_XMM sse4
+CONV_FLTP_TO_FLT_6CH
+%if HAVE_AVX
+INIT_XMM avx
 CONV_FLTP_TO_FLT_6CH
+%endif

libavresample/x86/audio_convert_init.c

@@ -22,8 +22,9 @@
 #include "libavutil/cpu.h"
 #include "libavresample/audio_convert.h"
 
-extern void ff_conv_fltp_to_flt_6ch_mmx(float *dst, float *const *src, int len);
-extern void ff_conv_fltp_to_flt_6ch_sse(float *dst, float *const *src, int len);
+extern void ff_conv_fltp_to_flt_6ch_mmx (float *dst, float *const *src, int len);
+extern void ff_conv_fltp_to_flt_6ch_sse4(float *dst, float *const *src, int len);
+extern void ff_conv_fltp_to_flt_6ch_avx (float *dst, float *const *src, int len);
 
 av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
 {
@@ -34,9 +35,13 @@ av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
                                   6, 1, 4, "MMX", ff_conv_fltp_to_flt_6ch_mmx);
     }
-    if (mm_flags & AV_CPU_FLAG_SSE && HAVE_SSE) {
+    if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE) {
         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
-                                  6, 16, 4, "SSE", ff_conv_fltp_to_flt_6ch_sse);
+                                  6, 16, 4, "SSE4", ff_conv_fltp_to_flt_6ch_sse4);
+    }
+    if (mm_flags & AV_CPU_FLAG_AVX && HAVE_AVX) {
+        ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
+                                  6, 16, 4, "AVX", ff_conv_fltp_to_flt_6ch_avx);
     }
 #endif
 }

libavutil/x86/x86util.asm

@@ -42,10 +42,9 @@
 %endmacro
 
 %macro SBUTTERFLYPS 3
-    movaps   m%3, m%1
-    unpcklps m%1, m%2
-    unpckhps m%3, m%2
-    SWAP %2, %3
+    unpcklps m%3, m%1, m%2
+    unpckhps m%1, m%1, m%2
+    SWAP %1, %3, %2
 %endmacro
 
 %macro TRANSPOSE4x4B 5