@scrypted/libav 1.0.167 → 1.0.169

package/opus/celt/arm/celt_mdct_ne10.c

@@ -0,0 +1,258 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+   Written by Viswanath Puttagunta */
+/**
+   @file celt_mdct_ne10.c
+   @brief ARM Neon optimizations for mdct using NE10 library
+ */
+
+/*
+   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 SKIP_CONFIG_H
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+#endif
+
+#include "kiss_fft.h"
+#include "_kiss_fft_guts.h"
+#include "mdct.h"
+#include "stack_alloc.h"
+
+void clt_mdct_forward_neon(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;
+
+   SAVE_STACK;
+
+   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++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2);
+         *yp++ = MULT16_32_Q15(*wp1, *xp1)    - MULT16_32_Q15(*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++ =  -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2);
+         *yp++ = MULT16_32_Q15(*wp2, *xp1)     + MULT16_32_Q15(*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(re,t0)  -  S_MUL(im,t1);
+         yi = S_MUL(im,t0)  +  S_MUL(re,t1);
+         yc.r = yr;
+         yc.i = yi;
+         f2[i] = yc;
+      }
+   }
+
+   opus_fft(st, f2, (kiss_fft_cpx *)f, arch);
+
+   /* Post-rotate */
+   {
+      /* Temp pointers to make it really clear to the compiler what we're doing */
+      const kiss_fft_cpx * OPUS_RESTRICT fp = (kiss_fft_cpx *)f;
+      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(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]);
+         yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]);
+         *yp1 = yr;
+         *yp2 = yi;
+         fp++;
+         yp1 += 2*stride;
+         yp2 -= 2*stride;
+      }
+   }
+   RESTORE_STACK;
+}
+
+void clt_mdct_backward_neon(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;
+   VARDECL(kiss_fft_scalar, f);
+   const kiss_twiddle_scalar *trig;
+   const kiss_fft_state *st = l->kfft[shift];
+
+   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);
+
+   /* 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 = f;
+      const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0];
+      for(i=0;i<N4;i++)
+      {
+         kiss_fft_scalar yr, yi;
+         yr = S_MUL(*xp2, t[i]) + S_MUL(*xp1, t[N4+i]);
+         yi = S_MUL(*xp1, t[i]) - S_MUL(*xp2, t[N4+i]);
+         yp[2*i] = yr;
+         yp[2*i+1] = yi;
+         xp1+=2*stride;
+         xp2-=2*stride;
+      }
+   }
+
+   opus_ifft(st, (kiss_fft_cpx *)f, (kiss_fft_cpx*)(out+(overlap>>1)), arch);
+
+   /* Post-rotate and de-shuffle from both ends of the buffer at once to make
+      it in-place. */
+   {
+      kiss_fft_scalar * yp0 = out+(overlap>>1);
+      kiss_fft_scalar * 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;
+         re = yp0[0];
+         im = yp0[1];
+         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(re,t0) + S_MUL(im,t1);
+         yi = S_MUL(re,t1) - S_MUL(im,t0);
+         re = yp1[0];
+         im = yp1[1];
+         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(re,t0) + S_MUL(im,t1);
+         yi = S_MUL(re,t1) - S_MUL(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--;
+      }
+   }
+   RESTORE_STACK;
+}

package/opus/celt/arm/celt_neon_intr.c

@@ -0,0 +1,272 @@
+/* Copyright (c) 2014-2015 Xiph.Org Foundation
+   Written by Viswanath Puttagunta */
+/**
+   @file celt_neon_intr.c
+   @brief ARM Neon Intrinsic optimizations for celt
+ */
+
+/*
+   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.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <arm_neon.h>
+#include "../pitch.h"
+
+#if defined(FIXED_POINT)
+#include <string.h>
+
+void xcorr_kernel_neon_fixed(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len)
+{
+   int j;
+   int32x4_t a = vld1q_s32(sum);
+   /* Load y[0...3] */
+   /* This requires len>0 to always be valid (which we assert in the C code). */
+   int16x4_t y0 = vld1_s16(y);
+   y += 4;
+
+   /* This loop loads one y value more than we actually need.
+      Therefore we have to stop as soon as there are 8 or fewer samples left
+       (instead of 7), to avoid reading past the end of the array. */
+   for (j = 0; j + 8 < len; j += 8)
+   {
+      /* Load x[0...7] */
+      int16x8_t xx = vld1q_s16(x);
+      int16x4_t x0 = vget_low_s16(xx);
+      int16x4_t x4 = vget_high_s16(xx);
+      /* Load y[4...11] */
+      int16x8_t yy = vld1q_s16(y);
+      int16x4_t y4 = vget_low_s16(yy);
+      int16x4_t y8 = vget_high_s16(yy);
+      int32x4_t a0 = vmlal_lane_s16(a, y0, x0, 0);
+      int32x4_t a1 = vmlal_lane_s16(a0, y4, x4, 0);
+
+      int16x4_t y1 = vext_s16(y0, y4, 1);
+      int16x4_t y5 = vext_s16(y4, y8, 1);
+      int32x4_t a2 = vmlal_lane_s16(a1, y1, x0, 1);
+      int32x4_t a3 = vmlal_lane_s16(a2, y5, x4, 1);
+
+      int16x4_t y2 = vext_s16(y0, y4, 2);
+      int16x4_t y6 = vext_s16(y4, y8, 2);
+      int32x4_t a4 = vmlal_lane_s16(a3, y2, x0, 2);
+      int32x4_t a5 = vmlal_lane_s16(a4, y6, x4, 2);
+
+      int16x4_t y3 = vext_s16(y0, y4, 3);
+      int16x4_t y7 = vext_s16(y4, y8, 3);
+      int32x4_t a6 = vmlal_lane_s16(a5, y3, x0, 3);
+      int32x4_t a7 = vmlal_lane_s16(a6, y7, x4, 3);
+
+      y0 = y8;
+      a = a7;
+      x += 8;
+      y += 8;
+   }
+   if (j + 4 < len) {
+      /* Load x[0...3] */
+      int16x4_t x0 = vld1_s16(x);
+      /* Load y[4...7] */
+      int16x4_t y4 = vld1_s16(y);
+      int32x4_t a0 = vmlal_lane_s16(a, y0, x0, 0);
+      int16x4_t y1 = vext_s16(y0, y4, 1);
+      int32x4_t a1 = vmlal_lane_s16(a0, y1, x0, 1);
+      int16x4_t y2 = vext_s16(y0, y4, 2);
+      int32x4_t a2 = vmlal_lane_s16(a1, y2, x0, 2);
+      int16x4_t y3 = vext_s16(y0, y4, 3);
+      int32x4_t a3 = vmlal_lane_s16(a2, y3, x0, 3);
+      y0 = y4;
+      a = a3;
+      x += 4;
+      y += 4;
+      j += 4;
+   }
+   if (j + 2 < len) {
+      /* Load x[0...1] */
+      int16x4x2_t xx = vld2_dup_s16(x);
+      int16x4_t x0 = xx.val[0];
+      int16x4_t x1 = xx.val[1];
+      /* Load y[4...5].
+         We would like to use vld1_dup_s32(), but casting the pointer would
+          break strict aliasing rules and potentially have alignment issues.
+         Fortunately the compiler seems capable of translating this memcpy()
+          and vdup_n_s32() into the equivalent vld1_dup_s32().*/
+      int32_t yy;
+      memcpy(&yy, y, sizeof(yy));
+      int16x4_t y4 = vreinterpret_s16_s32(vdup_n_s32(yy));
+      int32x4_t a0 = vmlal_s16(a, y0, x0);
+      int16x4_t y1 = vext_s16(y0, y4, 1);
+      /* Replace bottom copy of {y[5], y[4]} in y4 with {y[3], y[2]} from y0,
+          using VSRI instead of VEXT, since it's a data-processing
+          instruction. */
+      y0 = vreinterpret_s16_s64(vsri_n_s64(vreinterpret_s64_s16(y4),
+       vreinterpret_s64_s16(y0), 32));
+      int32x4_t a1 = vmlal_s16(a0, y1, x1);
+      a = a1;
+      x += 2;
+      y += 2;
+      j += 2;
+   }
+   if (j + 1 < len) {
+      /* Load next x. */
+      int16x4_t x0 = vld1_dup_s16(x);
+      int32x4_t a0 = vmlal_s16(a, y0, x0);
+      /* Load last y. */
+      int16x4_t y4 = vld1_dup_s16(y);
+      y0 = vreinterpret_s16_s64(vsri_n_s64(vreinterpret_s64_s16(y4),
+       vreinterpret_s64_s16(y0), 16));
+      a = a0;
+      x++;
+   }
+   /* Load last x. */
+   int16x4_t x0 = vld1_dup_s16(x);
+   int32x4_t a0 = vmlal_s16(a, y0, x0);
+   vst1q_s32(sum, a0);
+}
+
+#else
+
+#if defined(__ARM_FEATURE_FMA) && defined(__ARM_ARCH_ISA_A64)
+/* If we can, force the compiler to use an FMA instruction rather than break
+ *    vmlaq_f32() into fmul/fadd. */
+#ifdef vmlaq_lane_f32
+#undef vmlaq_lane_f32
+#endif
+#define vmlaq_lane_f32(a,b,c,lane) vfmaq_lane_f32(a,b,c,lane)
+#endif
+
+
+/*
+ * Function: xcorr_kernel_neon_float
+ * ---------------------------------
+ * Computes 4 correlation values and stores them in sum[4]
+ */
+static void xcorr_kernel_neon_float(const float32_t *x, const float32_t *y,
+      float32_t sum[4], int len) {
+   float32x4_t YY[3];
+   float32x4_t YEXT[3];
+   float32x4_t XX[2];
+   float32x2_t XX_2;
+   float32x4_t SUMM;
+   const float32_t *xi = x;
+   const float32_t *yi = y;
+
+   celt_assert(len>0);
+
+   YY[0] = vld1q_f32(yi);
+   SUMM = vdupq_n_f32(0);
+
+   /* Consume 8 elements in x vector and 12 elements in y
+    * vector. However, the 12'th element never really gets
+    * touched in this loop. So, if len == 8, then we only
+    * must access y[0] to y[10]. y[11] must not be accessed
+    * hence make sure len > 8 and not len >= 8
+    */
+   while (len > 8) {
+      yi += 4;
+      YY[1] = vld1q_f32(yi);
+      yi += 4;
+      YY[2] = vld1q_f32(yi);
+
+      XX[0] = vld1q_f32(xi);
+      xi += 4;
+      XX[1] = vld1q_f32(xi);
+      xi += 4;
+
+      SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);
+      YEXT[0] = vextq_f32(YY[0], YY[1], 1);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1);
+      YEXT[1] = vextq_f32(YY[0], YY[1], 2);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0);
+      YEXT[2] = vextq_f32(YY[0], YY[1], 3);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1);
+
+      SUMM = vmlaq_lane_f32(SUMM, YY[1], vget_low_f32(XX[1]), 0);
+      YEXT[0] = vextq_f32(YY[1], YY[2], 1);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[1]), 1);
+      YEXT[1] = vextq_f32(YY[1], YY[2], 2);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[1]), 0);
+      YEXT[2] = vextq_f32(YY[1], YY[2], 3);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[1]), 1);
+
+      YY[0] = YY[2];
+      len -= 8;
+   }
+
+   /* Consume 4 elements in x vector and 8 elements in y
+    * vector. However, the 8'th element in y never really gets
+    * touched in this loop. So, if len == 4, then we only
+    * must access y[0] to y[6]. y[7] must not be accessed
+    * hence make sure len>4 and not len>=4
+    */
+   if (len > 4) {
+      yi += 4;
+      YY[1] = vld1q_f32(yi);
+
+      XX[0] = vld1q_f32(xi);
+      xi += 4;
+
+      SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);
+      YEXT[0] = vextq_f32(YY[0], YY[1], 1);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1);
+      YEXT[1] = vextq_f32(YY[0], YY[1], 2);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0);
+      YEXT[2] = vextq_f32(YY[0], YY[1], 3);
+      SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1);
+
+      YY[0] = YY[1];
+      len -= 4;
+   }
+
+   while (--len > 0) {
+      XX_2 = vld1_dup_f32(xi++);
+      SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0);
+      YY[0]= vld1q_f32(++yi);
+   }
+
+   XX_2 = vld1_dup_f32(xi);
+   SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0);
+
+   vst1q_f32(sum, SUMM);
+}
+
+void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y,
+                        opus_val32 *xcorr, int len, int max_pitch, int arch) {
+   int i;
+   (void)arch;
+   celt_assert(max_pitch > 0);
+   celt_sig_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0);
+
+   for (i = 0; i < (max_pitch-3); i += 4) {
+      xcorr_kernel_neon_float((const float32_t *)_x, (const float32_t *)_y+i,
+            (float32_t *)xcorr+i, len);
+   }
+
+   /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */
+   for (; i < max_pitch; i++) {
+      xcorr[i] = celt_inner_prod_neon(_x, _y+i, len);
+   }
+}
+#endif