Update to opus-HEAD-66611f1.

celt/celt.c

 /* Copyright (c) 2007-2008 CSIRO
    Copyright (c) 2007-2010 Xiph.Org Foundation
    Copyright (c) 2008 Gregory Maxwell
    Written by Jean-Marc Valin and Gregory Maxwell */
 /*
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions
    are met:
 
    - Redistributions of source code must retain the above copyright
@@ skipping 36 matching lines @@
 #include "mathops.h"
 #include "float_cast.h"
 #include <stdarg.h>
 #include "celt_lpc.h"
 #include "vq.h"
 
 #ifndef PACKAGE_VERSION
 #define PACKAGE_VERSION "unknown"
 #endif
 
+#if defined(MIPSr1_ASM)
+#include "mips/celt_mipsr1.h"
+#endif
+
 
 int resampling_factor(opus_int32 rate)
 {
    int ret;
    switch (rate)
    {
    case 48000:
       ret = 1;
       break;
    case 24000:
@@ skipping 12 matching lines @@
 #ifndef CUSTOM_MODES
       celt_assert(0);
 #endif
       ret = 0;
       break;
    }
    return ret;
 }
 
 #ifndef OVERRIDE_COMB_FILTER_CONST
+/* This version should be faster on ARM */
+#ifdef OPUS_ARM_ASM
 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 = SHL32(x[-T-2], 1);
+   x3 = SHL32(x[-T-1], 1);
+   x2 = SHL32(x[-T], 1);
+   x1 = SHL32(x[-T+1], 1);
+   for (i=0;i<N-4;i+=5)
+   {
+      opus_val32 t;
+      x0=SHL32(x[i-T+2],1);
+      t = MAC16_32_Q16(x[i], g10, x2);
+      t = MAC16_32_Q16(t, g11, ADD32(x1,x3));
+      t = MAC16_32_Q16(t, g12, ADD32(x0,x4));
+      y[i] = t;
+      x4=SHL32(x[i-T+3],1);
+      t = MAC16_32_Q16(x[i+1], g10, x1);
+      t = MAC16_32_Q16(t, g11, ADD32(x0,x2));
+      t = MAC16_32_Q16(t, g12, ADD32(x4,x3));
+      y[i+1] = t;
+      x3=SHL32(x[i-T+4],1);
+      t = MAC16_32_Q16(x[i+2], g10, x0);
+      t = MAC16_32_Q16(t, g11, ADD32(x4,x1));
+      t = MAC16_32_Q16(t, g12, ADD32(x3,x2));
+      y[i+2] = t;
+      x2=SHL32(x[i-T+5],1);
+      t = MAC16_32_Q16(x[i+3], g10, x4);
+      t = MAC16_32_Q16(t, g11, ADD32(x3,x0));
+      t = MAC16_32_Q16(t, g12, ADD32(x2,x1));
+      y[i+3] = t;
+      x1=SHL32(x[i-T+6],1);
+      t = MAC16_32_Q16(x[i+4], g10, x3);
+      t = MAC16_32_Q16(t, g11, ADD32(x2,x4));
+      t = MAC16_32_Q16(t, g12, ADD32(x1,x0));
+      y[i+4] = t;
+   }
+#ifdef CUSTOM_MODES
+   for (;i<N;i++)
+   {
+      opus_val32 t;
+      x0=SHL32(x[i-T+2],1);
+      t = MAC16_32_Q16(x[i], g10, x2);
+      t = MAC16_32_Q16(t, g11, ADD32(x1,x3));
+      t = MAC16_32_Q16(t, g12, ADD32(x0,x4));
+      y[i] = t;
+      x4=x3;
+      x3=x2;
+      x2=x1;
+      x1=x0;
+   }
+#endif
+}
+#else
+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
+#endif
 
+#ifndef 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;
    /* 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)},
          {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}};
 
    if (g0==0 && g1==0)
    {
       /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */
       if (x!=y)
          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];
       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));
       x4=x3;
       x3=x2;
       x2=x1;
       x1=x0;
 
    }
    if (g1==0)
    {
       /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */
       if (x!=y)
          OPUS_MOVE(y+overlap, x+overlap, N-overlap);
       return;
    }
 
    /* Compute the part with the constant filter. */
    comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12);
 }
+#endif /* OVERRIDE_comb_filter */
 
 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)
@@ skipping 31 matching lines @@
 {
     return "libopus " PACKAGE_VERSION
 #ifdef FIXED_POINT
           "-fixed"
 #endif
 #ifdef FUZZING
           "-fuzzing"
 #endif
           ;
 }