@scrypted/libav 1.0.167 → 1.0.169

package/opus/celt/mips/mdct_mipsr1.h

@@ -0,0 +1,288 @@
+/* Copyright (c) 2007-2008 CSIRO
+   Copyright (c) 2007-2008 Xiph.Org Foundation
+   Written by Jean-Marc Valin */
+/*
+   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
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   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.
+*/
+
+/* This is a simple MDCT implementation that uses a N/4 complex FFT
+   to do most of the work. It should be relatively straightforward to
+   plug in pretty much and FFT here.
+
+   This replaces the Vorbis FFT (and uses the exact same API), which
+   was a bit too messy and that was ending up duplicating code
+   (might as well use the same FFT everywhere).
+
+   The algorithm is similar to (and inspired from) Fabrice Bellard's
+   MDCT implementation in FFMPEG, but has differences in signs, ordering
+   and scaling in many places.
+*/
+#ifndef MDCT_MIPSR1_H__
+#define MDCT_MIPSR1_H__
+
+#ifndef SKIP_CONFIG_H
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+#endif
+
+#include "mdct.h"
+#include "kiss_fft.h"
+#include "_kiss_fft_guts.h"
+#include <math.h>
+#include "os_support.h"
+#include "mathops.h"
+#include "stack_alloc.h"
+
+/* Forward MDCT trashes the input array */
+#define OVERRIDE_clt_mdct_forward
+void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+      const opus_val16 *window, int overlap, int shift, int stride, int arch)
+{
+   int i;
+   int N, N2, N4;
+   VARDECL(kiss_fft_scalar, f);
+   VARDECL(kiss_fft_cpx, f2);
+   const kiss_fft_state *st = l->kfft[shift];
+   const kiss_twiddle_scalar *trig;
+   opus_val16 scale;
+#ifdef FIXED_POINT
+   /* Allows us to scale with MULT16_32_Q16(), which is faster than
+      MULT16_32_Q15() on ARM. */
+   int scale_shift = st->scale_shift-1;
+#endif
+
+    (void)arch;
+
+   SAVE_STACK;
+   scale = st->scale;
+
+   N = l->n;
+   trig = l->trig;
+   for (i=0;i<shift;i++)
+   {
+      N >>= 1;
+      trig += N;
+   }
+   N2 = N>>1;
+   N4 = N>>2;
+
+   ALLOC(f, N2, kiss_fft_scalar);
+   ALLOC(f2, N4, kiss_fft_cpx);
+
+   /* Consider the input to be composed of four blocks: [a, b, c, d] */
+   /* Window, shuffle, fold */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1);
+      const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1);
+      kiss_fft_scalar * OPUS_RESTRICT yp = f;
+      const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1);
+      const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1;
+      for(i=0;i<((overlap+3)>>2);i++)
+      {
+         /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/
+          *yp++ = S_MUL_ADD(*wp2, xp1[N2],*wp1,*xp2);
+          *yp++ = S_MUL_SUB(*wp1, *xp1,*wp2, xp2[-N2]);
+         xp1+=2;
+         xp2-=2;
+         wp1+=2;
+         wp2-=2;
+      }
+      wp1 = window;
+      wp2 = window+overlap-1;
+      for(;i<N4-((overlap+3)>>2);i++)
+      {
+         /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+         *yp++ = *xp2;
+         *yp++ = *xp1;
+         xp1+=2;
+         xp2-=2;
+      }
+      for(;i<N4;i++)
+      {
+         /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+          *yp++ =  S_MUL_SUB(*wp2, *xp2, *wp1, xp1[-N2]);
+          *yp++ = S_MUL_ADD(*wp2, *xp1, *wp1, xp2[N2]);
+         xp1+=2;
+         xp2-=2;
+         wp1+=2;
+         wp2-=2;
+      }
+   }
+   /* Pre-rotation */
+   {
+      kiss_fft_scalar * OPUS_RESTRICT yp = f;
+      const kiss_twiddle_scalar *t = &trig[0];
+      for(i=0;i<N4;i++)
+      {
+         kiss_fft_cpx yc;
+         kiss_twiddle_scalar t0, t1;
+         kiss_fft_scalar re, im, yr, yi;
+         t0 = t[i];
+         t1 = t[N4+i];
+         re = *yp++;
+         im = *yp++;
+
+         yr = S_MUL_SUB(re,t0,im,t1);
+         yi = S_MUL_ADD(im,t0,re,t1);
+
+         yc.r = yr;
+         yc.i = yi;
+         yc.r = PSHR32(MULT16_32_Q16(scale, yc.r), scale_shift);
+         yc.i = PSHR32(MULT16_32_Q16(scale, yc.i), scale_shift);
+         f2[st->bitrev[i]] = yc;
+      }
+   }
+
+   /* N/4 complex FFT, does not downscale anymore */
+   opus_fft_impl(st, f2);
+
+   /* Post-rotate */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_cpx * OPUS_RESTRICT fp = f2;
+      kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
+      kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1);
+      const kiss_twiddle_scalar *t = &trig[0];
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      for(i=0;i<N4;i++)
+      {
+         kiss_fft_scalar yr, yi;
+         yr = S_MUL_SUB(fp->i,t[N4+i] , fp->r,t[i]);
+         yi = S_MUL_ADD(fp->r,t[N4+i] ,fp->i,t[i]);
+         *yp1 = yr;
+         *yp2 = yi;
+         fp++;
+         yp1 += 2*stride;
+         yp2 -= 2*stride;
+      }
+   }
+   RESTORE_STACK;
+}
+
+#define OVERRIDE_clt_mdct_backward
+void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+      const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch)
+{
+   int i;
+   int N, N2, N4;
+   const kiss_twiddle_scalar *trig;
+
+    (void)arch;
+
+   N = l->n;
+   trig = l->trig;
+   for (i=0;i<shift;i++)
+   {
+      N >>= 1;
+      trig += N;
+   }
+   N2 = N>>1;
+   N4 = N>>2;
+
+   /* Pre-rotate */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_scalar * OPUS_RESTRICT xp1 = in;
+      const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1);
+      kiss_fft_scalar * OPUS_RESTRICT yp = out+(overlap>>1);
+      const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0];
+      const opus_int16 * OPUS_RESTRICT bitrev = l->kfft[shift]->bitrev;
+      for(i=0;i<N4;i++)
+      {
+         int rev;
+         kiss_fft_scalar yr, yi;
+         rev = *bitrev++;
+         yr = S_MUL_ADD(*xp2, t[i] , *xp1, t[N4+i]);
+         yi = S_MUL_SUB(*xp1, t[i] , *xp2, t[N4+i]);
+         /* We swap real and imag because we use an FFT instead of an IFFT. */
+         yp[2*rev+1] = yr;
+         yp[2*rev] = yi;
+         /* Storing the pre-rotation directly in the bitrev order. */
+         xp1+=2*stride;
+         xp2-=2*stride;
+      }
+   }
+
+   opus_fft_impl(l->kfft[shift], (kiss_fft_cpx*)(out+(overlap>>1)));
+
+   /* Post-rotate and de-shuffle from both ends of the buffer at once to make
+      it in-place. */
+   {
+      kiss_fft_scalar * OPUS_RESTRICT yp0 = out+(overlap>>1);
+      kiss_fft_scalar * OPUS_RESTRICT yp1 = out+(overlap>>1)+N2-2;
+      const kiss_twiddle_scalar *t = &trig[0];
+      /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the
+         middle pair will be computed twice. */
+      for(i=0;i<(N4+1)>>1;i++)
+      {
+         kiss_fft_scalar re, im, yr, yi;
+         kiss_twiddle_scalar t0, t1;
+         /* We swap real and imag because we're using an FFT instead of an IFFT. */
+         re = yp0[1];
+         im = yp0[0];
+         t0 = t[i];
+         t1 = t[N4+i];
+         /* We'd scale up by 2 here, but instead it's done when mixing the windows */
+         yr = S_MUL_ADD(re,t0 , im,t1);
+         yi = S_MUL_SUB(re,t1 , im,t0);
+         /* We swap real and imag because we're using an FFT instead of an IFFT. */
+         re = yp1[1];
+         im = yp1[0];
+         yp0[0] = yr;
+         yp1[1] = yi;
+
+         t0 = t[(N4-i-1)];
+         t1 = t[(N2-i-1)];
+         /* We'd scale up by 2 here, but instead it's done when mixing the windows */
+         yr = S_MUL_ADD(re,t0,im,t1);
+         yi = S_MUL_SUB(re,t1,im,t0);
+         yp1[0] = yr;
+         yp0[1] = yi;
+         yp0 += 2;
+         yp1 -= 2;
+      }
+   }
+
+   /* Mirror on both sides for TDAC */
+   {
+      kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1;
+      kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
+      const opus_val16 * OPUS_RESTRICT wp1 = window;
+      const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1;
+
+      for(i = 0; i < overlap/2; i++)
+      {
+         kiss_fft_scalar x1, x2;
+         x1 = *xp1;
+         x2 = *yp1;
+         *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1);
+         *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1);
+         wp1++;
+         wp2--;
+      }
+   }
+}
+#endif /* MDCT_MIPSR1_H__ */

package/opus/celt/mips/pitch_mipsr1.h

@@ -0,0 +1,161 @@
+/* Copyright (c) 2007-2008 CSIRO
+   Copyright (c) 2007-2009 Xiph.Org Foundation
+   Written by Jean-Marc Valin */
+/**
+   @file pitch.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:
+
+   - Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   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_MIPSR1_H
+#define PITCH_MIPSR1_H
+
+#define OVERRIDE_DUAL_INNER_PROD
+static 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 arch)
+{
+   int j;
+   opus_val32 xy01=0;
+   opus_val32 xy02=0;
+
+   (void)arch;
+
+   asm volatile("MULT $ac1, $0, $0");
+   asm volatile("MULT $ac2, $0, $0");
+   /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
+   for (j=0;j<N;j++)
+   {
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
+      asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
+      ++j;
+      asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
+      asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
+   }
+   asm volatile ("mflo %0, $ac1": "=r"(xy01));
+   asm volatile ("mflo %0, $ac2": "=r"(xy02));
+   *xy1 = xy01;
+   *xy2 = xy02;
+}
+
+static inline void xcorr_kernel_mips(const opus_val16 * x,
+      const opus_val16 * y, opus_val32 sum[4], int len)
+{
+   int j;
+   opus_val16 y_0, y_1, y_2, y_3;
+
+    opus_int64 sum_0, sum_1, sum_2, sum_3;
+    sum_0 =  (opus_int64)sum[0];
+    sum_1 =  (opus_int64)sum[1];
+    sum_2 =  (opus_int64)sum[2];
+    sum_3 =  (opus_int64)sum[3];
+
+    y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */
+    y_0=*y++;
+    y_1=*y++;
+    y_2=*y++;
+    for (j=0;j<len-3;j+=4)
+    {
+        opus_val16 tmp;
+        tmp = *x++;
+        y_3=*y++;
+
+        sum_0 = __builtin_mips_madd( sum_0, tmp, y_0);
+        sum_1 = __builtin_mips_madd( sum_1, tmp, y_1);
+        sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
+        sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
+
+        tmp=*x++;
+        y_0=*y++;
+
+        sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
+        sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
+        sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
+        sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
+
+       tmp=*x++;
+       y_1=*y++;
+
+       sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
+       sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
+       sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
+       sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
+
+
+      tmp=*x++;
+      y_2=*y++;
+
+       sum_0 = __builtin_mips_madd( sum_0, tmp, y_3 );
+       sum_1 = __builtin_mips_madd( sum_1, tmp, y_0 );
+       sum_2 = __builtin_mips_madd( sum_2, tmp, y_1);
+       sum_3 = __builtin_mips_madd( sum_3, tmp, y_2);
+
+   }
+   if (j++<len)
+   {
+      opus_val16 tmp = *x++;
+      y_3=*y++;
+
+       sum_0 = __builtin_mips_madd( sum_0, tmp, y_0 );
+       sum_1 = __builtin_mips_madd( sum_1, tmp, y_1 );
+       sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
+       sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
+   }
+
+   if (j++<len)
+   {
+      opus_val16 tmp=*x++;
+      y_0=*y++;
+
+      sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
+      sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
+      sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
+      sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
+   }
+
+   if (j<len)
+   {
+      opus_val16 tmp=*x++;
+      y_1=*y++;
+
+       sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
+       sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
+       sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
+       sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
+
+   }
+
+   sum[0] = (opus_val32)sum_0;
+   sum[1] = (opus_val32)sum_1;
+   sum[2] = (opus_val32)sum_2;
+   sum[3] = (opus_val32)sum_3;
+}
+
+#define OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+    ((void)(arch), xcorr_kernel_mips(x, y, sum, len))
+
+#endif /* PITCH_MIPSR1_H */

package/opus/celt/mips/vq_mipsr1.h

@@ -0,0 +1,116 @@
+/* Copyright (c) 2007-2008 CSIRO
+   Copyright (c) 2007-2009 Xiph.Org Foundation
+   Written by Jean-Marc Valin */
+/*
+   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
+   notice, this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
+   OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+   EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+   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 VQ_MIPSR1_H__
+#define VQ_MIPSR1_H__
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include "mathops.h"
+#include "arch.h"
+
+#define OVERRIDE_vq_exp_rotation1
+static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s)
+{
+   int i;
+   opus_val16 ms;
+   celt_norm *Xptr;
+   Xptr = X;
+   ms = NEG16(s);
+   for (i=0;i<len-stride;i++)
+   {
+      celt_norm x1, x2;
+      x1 = Xptr[0];
+      x2 = Xptr[stride];
+      Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2),  s, x1), 15));
+      *Xptr++      = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
+   }
+   Xptr = &X[len-2*stride-1];
+   for (i=len-2*stride-1;i>=0;i--)
+   {
+      celt_norm x1, x2;
+      x1 = Xptr[0];
+      x2 = Xptr[stride];
+      Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2),  s, x1), 15));
+      *Xptr--      = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
+   }
+}
+
+#define OVERRIDE_renormalise_vector
+void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch)
+{
+   int i;
+#ifdef FIXED_POINT
+   int k;
+#endif
+   opus_val32 E = EPSILON;
+   opus_val16 g;
+   opus_val32 t;
+   celt_norm *xptr = X;
+   int X0, X1;
+
+   (void)arch;
+
+   asm volatile("mult $ac1, $0, $0");
+   asm volatile("MTLO %0, $ac1" : :"r" (E));
+   /*if(N %4)
+       printf("error");*/
+   for (i=0;i<N-2;i+=2)
+   {
+      X0 = (int)*xptr++;
+      asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0));
+
+      X1 = (int)*xptr++;
+      asm volatile("MADD $ac1, %0, %1" : : "r" (X1), "r" (X1));
+   }
+
+   for (;i<N;i++)
+   {
+      X0 = (int)*xptr++;
+      asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0));
+   }
+
+   asm volatile("MFLO %0, $ac1" : "=r" (E));
+#ifdef FIXED_POINT
+   k = celt_ilog2(E)>>1;
+#endif
+   t = VSHR32(E, 2*(k-7));
+   g = MULT16_16_P15(celt_rsqrt_norm(t),gain);
+
+   xptr = X;
+   for (i=0;i<N;i++)
+   {
+      *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1));
+      xptr++;
+   }
+   /*return celt_sqrt(E);*/
+}
+
+#endif /* VQ_MIPSR1_H__ */