diff --git a/libavcodec/dcadec.c b/libavcodec/dcadec.c
index 076a225..282a58b 100644
--- a/libavcodec/dcadec.c
+++ b/libavcodec/dcadec.c
@@ -1096,6 +1096,8 @@ static inline void int8x8_fmul_int32(DCADSPContext *dsp, float *dst,
 }
 #endif

+#include "libavutil/timer.h"
+
 static int dca_subsubframe(DCAContext *s, int base_channel, int block_index)
 {
     int k, l;
@@ -1225,10 +1227,11 @@ static int dca_subsubframe(DCAContext *s, int base_channel, int block_index)
                        "Stream with high frequencies VQ coding\n");
                 s->debug_flag |= 0x01;
             }
-
+{START_TIMER
             int8x8_fmul_int32(&s->dcadsp, subband_samples[k][l],
                               &high_freq_vq[hfvq][subsubframe * 8],
                               s->scale_factor[k][l][0]);
+STOP_TIMER("g")}
         }
     }

diff --git a/libavcodec/x86/dca.h b/libavcodec/x86/dca.h
index ab175b3..e2591e3 100644
--- a/libavcodec/x86/dca.h
+++ b/libavcodec/x86/dca.h
@@ -20,15 +20,40 @@

 #include "config.h"

-#if ARCH_X86_64 && HAVE_SSE2_INLINE
-# include "libavutil/x86/asm.h"
+#define HAVE_SSE2_INTRINSICS   0
+
+#if ARCH_X86_64 && (HAVE_SSE2_INTRINSICS || HAVE_SSE2_INLINE)
 # include "libavutil/mem.h"
 #include "libavcodec/dcadsp.h"

+# if HAVE_SSE2_INTRINSICS
+#   include <emmintrin.h>
+# else
+#   include "libavutil/x86/asm.h"
+# endif
+
 # define int8x8_fmul_int32 int8x8_fmul_int32
 static inline void int8x8_fmul_int32(av_unused DCADSPContext *dsp,
                                      float *dst, const int8_t *src, int scale)
 {
+# if HAVE_SSE2_INTRINSICS
+    float tmp = scale * 1.0f/16; // should generate cvtsi2ss+mulss
+    __m128i in2, in1 = _mm_loadl_epi64((const __m128i*)src);
+    __m128 out1, out2, fscale = _mm_load_ss(&tmp);
+    in1    = _mm_unpacklo_epi8(in1, in1);
+    in2    = in1;
+    in1    = _mm_unpacklo_epi16(in1, in1);
+    in2    = _mm_unpackhi_epi16(in2, in2);
+    in1    = _mm_srai_epi32(in1, 24);
+    in2    = _mm_srai_epi32(in2, 24);
+    fscale = _mm_shuffle_ps(fscale, fscale, 0);
+    out1   = _mm_cvtepi32_ps(in1);
+    out2   = _mm_cvtepi32_ps(in2);
+    out1   = _mm_mul_ps(out1, fscale);
+    out2   = _mm_mul_ps(out2, fscale);
+    _mm_store_ps(dst+0, out1);
+    _mm_store_ps(dst+4, out2);
+# elif HAVE_SSE2_INLINE
     DECLARE_ALIGNED(16, static const uint32_t, inverse16) = 0x3D800000;
     __asm__ volatile (
         "cvtsi2ss        %2, %%xmm0 \n\t"
@@ -50,6 +75,7 @@ static inline void int8x8_fmul_int32(av_unused DCADSPContext *dsp,
         :: "r"(dst), "r"(src), "m"(scale), "m"(inverse16)
         XMM_CLOBBERS_ONLY("xmm0", "xmm1", "xmm2")
     );
+# endif
 }

 #endif /* ARCH_X86_64 && HAVE_SSE2_INLINE */