Update to opus-HEAD-66611f1.

celt/x86/pitch_sse.h

-/* Copyright (c) 2013 Jean-Marc Valin and John Ridges */
+/* Copyright (c) 2013 Jean-Marc Valin and John Ridges
+   Copyright (c) 2014, Cisco Systems, INC MingXiang WeiZhou MinPeng YanWang*/
 /**
    @file pitch_sse.h
    @brief Pitch analysis
  */
 
 /*
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions
    are met:
 
@@ skipping 13 matching lines @@
    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
 #ifndef PITCH_SSE_H
 #define PITCH_SSE_H
 
+#if defined(HAVE_CONFIG_H)
+#include "config.h"
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1)
+void xcorr_kernel_sse4_1(
+                    const opus_int16 *x,
+                    const opus_int16 *y,
+                    opus_val32       sum[4],
+                    int              len);
+
+extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
+                    const opus_int16 *x,
+                    const opus_int16 *y,
+                    opus_val32       sum[4],
+                    int              len);
+
+#define OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+    ((*XCORR_KERNEL_IMPL[(arch) & OPUS_ARCHMASK])(x, y, sum, len))
+
+opus_val32 celt_inner_prod_sse4_1(
+    const opus_int16 *x,
+    const opus_int16 *y,
+    int               N);
+#endif
+
+#if defined(OPUS_X86_MAY_HAVE_SSE2)
+opus_val32 celt_inner_prod_sse2(
+    const opus_int16 *x,
+    const opus_int16 *y,
+    int               N);
+#endif
+
+extern opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])(
+                    const opus_int16 *x,
+                    const opus_int16 *y,
+                    int               N);
+
+#define OVERRIDE_CELT_INNER_PROD
+#define celt_inner_prod(x, y, N, arch) \
+    ((*CELT_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y, N))
+#else
+
 #include <xmmintrin.h>
 #include "arch.h"
 
 #define OVERRIDE_XCORR_KERNEL
-static OPUS_INLINE void xcorr_kernel(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len)
+static OPUS_INLINE void xcorr_kernel_sse(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len)
 {
    int j;
    __m128 xsum1, xsum2;
    xsum1 = _mm_loadu_ps(sum);
    xsum2 = _mm_setzero_ps();
 
    for (j = 0; j < len-3; j += 4)
    {
       __m128 x0 = _mm_loadu_ps(x+j);
       __m128 yj = _mm_loadu_ps(y+j);
@@ skipping 14 matching lines @@
          xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
          if (++j < len)
          {
             xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j)));
          }
       }
    }
    _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2));
 }
 
+#define xcorr_kernel(_x, _y, _z, len, arch) \
+    ((void)(arch),xcorr_kernel_sse(_x, _y, _z, len))
+
 #define OVERRIDE_DUAL_INNER_PROD
 static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
       int N, opus_val32 *xy1, opus_val32 *xy2)
 {
    int i;
    __m128 xsum1, xsum2;
    xsum1 = _mm_setzero_ps();
    xsum2 = _mm_setzero_ps();
    for (i=0;i<N-3;i+=4)
    {
@@ skipping 10 matching lines @@
    xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2));
    xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55));
    _mm_store_ss(xy2, xsum2);
    for (;i<N;i++)
    {
       *xy1 = MAC16_16(*xy1, x[i], y01[i]);
       *xy2 = MAC16_16(*xy2, x[i], y02[i]);
    }
 }
 
+#define OVERRIDE_CELT_INNER_PROD
+static OPUS_INLINE opus_val32 celt_inner_prod_sse(const opus_val16 *x, const opus_val16 *y,
+      int N)
+{
+   int i;
+   float xy;
+   __m128 sum;
+   sum = _mm_setzero_ps();
+   /* FIXME: We should probably go 8-way and use 2 sums. */
+   for (i=0;i<N-3;i+=4)
+   {
+      __m128 xi = _mm_loadu_ps(x+i);
+      __m128 yi = _mm_loadu_ps(y+i);
+      sum = _mm_add_ps(sum,_mm_mul_ps(xi, yi));
+   }
+   /* Horizontal sum */
+   sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum));
+   sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55));
+   _mm_store_ss(&xy, sum);
+   for (;i<N;i++)
+   {
+      xy = MAC16_16(xy, x[i], y[i]);
+   }
+   return xy;
+}
+
+#  define celt_inner_prod(_x, _y, len, arch) \
+    ((void)(arch),celt_inner_prod_sse(_x, _y, len))
+
 #define OVERRIDE_COMB_FILTER_CONST
 static OPUS_INLINE void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
       opus_val16 g10, opus_val16 g11, opus_val16 g12)
 {
    int i;
    __m128 x0v;
    __m128 g10v, g11v, g12v;
    g10v = _mm_load1_ps(&g10);
    g11v = _mm_load1_ps(&g11);
    g12v = _mm_load1_ps(&g12);
@@ skipping 33 matching lines @@
    {
       y[i] = x[i]
                + MULT16_32_Q15(g10,x[i-T])
                + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1]))
                + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2]));
    }
 #endif
 }
 
 #endif
+#endif