Update to opus-HEAD-66611f1.

celt/mips/celt_mipsr1.h

Index: celt/mips/celt_mipsr1.h
diff --git a/celt/celt.c b/celt/mips/celt_mipsr1.h
similarity index 53%
copy from celt/celt.c
copy to celt/mips/celt_mipsr1.h
index 3e0ce6e6a550768ea603e05b6ac740b029fd55c6..03915d8817c6f0bab1ccb31a563d02658d795ebe 100644
--- a/celt/celt.c
+++ b/celt/mips/celt_mipsr1.h
@@ -27,6 +27,9 @@
    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
+#ifndef __CELT_MIPSR1_H__
+#define __CELT_MIPSR1_H__
+
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
@@ -50,75 +53,16 @@
 #include "celt_lpc.h"
 #include "vq.h"
 
-#ifndef PACKAGE_VERSION
-#define PACKAGE_VERSION "unknown"
-#endif
-
-
-int resampling_factor(opus_int32 rate)
-{
-   int ret;
-   switch (rate)
-   {
-   case 48000:
-      ret = 1;
-      break;
-   case 24000:
-      ret = 2;
-      break;
-   case 16000:
-      ret = 3;
-      break;
-   case 12000:
-      ret = 4;
-      break;
-   case 8000:
-      ret = 6;
-      break;
-   default:
-#ifndef CUSTOM_MODES
-      celt_assert(0);
-#endif
-      ret = 0;
-      break;
-   }
-   return ret;
-}
-
-#ifndef OVERRIDE_COMB_FILTER_CONST
-static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
-      opus_val16 g10, opus_val16 g11, opus_val16 g12)
-{
-   opus_val32 x0, x1, x2, x3, x4;
-   int i;
-   x4 = x[-T-2];
-   x3 = x[-T-1];
-   x2 = x[-T];
-   x1 = x[-T+1];
-   for (i=0;i<N;i++)
-   {
-      x0=x[i-T+2];
-      y[i] = x[i]
-               + MULT16_32_Q15(g10,x2)
-               + MULT16_32_Q15(g11,ADD32(x1,x3))
-               + MULT16_32_Q15(g12,ADD32(x0,x4));
-      x4=x3;
-      x3=x2;
-      x2=x1;
-      x1=x0;
-   }
-
-}
-#endif
-
+#define OVERRIDE_comb_filter
 void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
       opus_val16 g0, opus_val16 g1, int tapset0, int tapset1,
       const opus_val16 *window, int overlap)
 {
    int i;
+   opus_val32 x0, x1, x2, x3, x4;
+
    /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */
    opus_val16 g00, g01, g02, g10, g11, g12;
-   opus_val32 x0, x1, x2, x3, x4;
    static const opus_val16 gains[3][3] = {
          {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)},
          {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)},
@@ -131,34 +75,51 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
          OPUS_MOVE(y, x, N);
       return;
    }
-   g00 = MULT16_16_Q15(g0, gains[tapset0][0]);
-   g01 = MULT16_16_Q15(g0, gains[tapset0][1]);
-   g02 = MULT16_16_Q15(g0, gains[tapset0][2]);
-   g10 = MULT16_16_Q15(g1, gains[tapset1][0]);
-   g11 = MULT16_16_Q15(g1, gains[tapset1][1]);
-   g12 = MULT16_16_Q15(g1, gains[tapset1][2]);
+
+   g00 = MULT16_16_P15(g0, gains[tapset0][0]);
+   g01 = MULT16_16_P15(g0, gains[tapset0][1]);
+   g02 = MULT16_16_P15(g0, gains[tapset0][2]);
+   g10 = MULT16_16_P15(g1, gains[tapset1][0]);
+   g11 = MULT16_16_P15(g1, gains[tapset1][1]);
+   g12 = MULT16_16_P15(g1, gains[tapset1][2]);
    x1 = x[-T1+1];
    x2 = x[-T1  ];
    x3 = x[-T1-1];
    x4 = x[-T1-2];
+   /* If the filter didn't change, we don't need the overlap */
+   if (g0==g1 && T0==T1 && tapset0==tapset1)
+      overlap=0;
+
    for (i=0;i<overlap;i++)
    {
       opus_val16 f;
-      x0=x[i-T1+2];
+      opus_val32 res;
       f = MULT16_16_Q15(window[i],window[i]);
-      y[i] = x[i]
-               + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g00),x[i-T0])
-               + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g01),ADD32(x[i-T0+1],x[i-T0-1]))
-               + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g02),ADD32(x[i-T0+2],x[i-T0-2]))
-               + MULT16_32_Q15(MULT16_16_Q15(f,g10),x2)
-               + MULT16_32_Q15(MULT16_16_Q15(f,g11),ADD32(x1,x3))
-               + MULT16_32_Q15(MULT16_16_Q15(f,g12),ADD32(x0,x4));
+      x0= x[i-T1+2];
+
+      asm volatile("MULT $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g00)), "r" ((int)x[i-T0]));
+
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g01)), "r" ((int)ADD32(x[i-T0-1],x[i-T0+1])));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g02)), "r" ((int)ADD32(x[i-T0-2],x[i-T0+2])));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g10)), "r" ((int)x2));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g11)), "r" ((int)ADD32(x3,x1)));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g12)), "r" ((int)ADD32(x4,x0)));
+
+      asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15));
+
+      y[i] = x[i] + res;
+
       x4=x3;
       x3=x2;
       x2=x1;
       x1=x0;
-
    }
+
+   x4 = x[i-T1-2];
+   x3 = x[i-T1-1];
+   x2 = x[i-T1];
+   x1 = x[i-T1+1];
+
    if (g1==0)
    {
       /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */
@@ -167,57 +128,22 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
       return;
    }
 
-   /* Compute the part with the constant filter. */
-   comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12);
-}
-
-const signed char tf_select_table[4][8] = {
-      {0, -1, 0, -1,    0,-1, 0,-1},
-      {0, -1, 0, -2,    1, 0, 1,-1},
-      {0, -2, 0, -3,    2, 0, 1,-1},
-      {0, -2, 0, -3,    3, 0, 1,-1},
-};
-
-
-void init_caps(const CELTMode *m,int *cap,int LM,int C)
-{
-   int i;
-   for (i=0;i<m->nbEBands;i++)
+   for (i=overlap;i<N;i++)
    {
-      int N;
-      N=(m->eBands[i+1]-m->eBands[i])<<LM;
-      cap[i] = (m->cache.caps[m->nbEBands*(2*LM+C-1)+i]+64)*C*N>>2;
-   }
-}
-
+      opus_val32 res;
+      x0=x[i-T1+2];
 
+      asm volatile("MULT $ac1, %0, %1" : : "r" ((int)g10), "r" ((int)x2));
 
-const char *opus_strerror(int error)
-{
-   static const char * const error_strings[8] = {
-      "success",
-      "invalid argument",
-      "buffer too small",
-      "internal error",
-      "corrupted stream",
-      "request not implemented",
-      "invalid state",
-      "memory allocation failed"
-   };
-   if (error > 0 || error < -7)
-      return "unknown error";
-   else
-      return error_strings[-error];
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g11), "r" ((int)ADD32(x3,x1)));
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g12), "r" ((int)ADD32(x4,x0)));
+      asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15));
+      y[i] = x[i] + res;
+      x4=x3;
+      x3=x2;
+      x2=x1;
+      x1=x0;
+   }
 }
 
-const char *opus_get_version_string(void)
-{
-    return "libopus " PACKAGE_VERSION
-#ifdef FIXED_POINT
-          "-fixed"
-#endif
-#ifdef FUZZING
-          "-fuzzing"
-#endif
-          ;
-}
+#endif /* __CELT_MIPSR1_H__ */