Chris@69: /* Copyright (c) 2007-2008 CSIRO
Chris@69:    Copyright (c) 2007-2009 Xiph.Org Foundation
Chris@69:    Written by Jean-Marc Valin */
Chris@69: /**
Chris@69:    @file pitch.h
Chris@69:    @brief Pitch analysis
Chris@69:  */
Chris@69: 
Chris@69: /*
Chris@69:    Redistribution and use in source and binary forms, with or without
Chris@69:    modification, are permitted provided that the following conditions
Chris@69:    are met:
Chris@69: 
Chris@69:    - Redistributions of source code must retain the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer.
Chris@69: 
Chris@69:    - Redistributions in binary form must reproduce the above copyright
Chris@69:    notice, this list of conditions and the following disclaimer in the
Chris@69:    documentation and/or other materials provided with the distribution.
Chris@69: 
Chris@69:    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
Chris@69:    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
Chris@69:    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
Chris@69:    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
Chris@69:    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
Chris@69:    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
Chris@69:    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
Chris@69:    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
Chris@69:    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
Chris@69:    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
Chris@69:    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Chris@69: */
Chris@69: 
Chris@69: #ifndef PITCH_H
Chris@69: #define PITCH_H
Chris@69: 
Chris@69: #include "modes.h"
Chris@69: #include "cpu_support.h"
Chris@69: 
Chris@69: #if (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)) \
Chris@69:   || ((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT))
Chris@69: #include "x86/pitch_sse.h"
Chris@69: #endif
Chris@69: 
Chris@69: #if defined(MIPSr1_ASM)
Chris@69: #include "mips/pitch_mipsr1.h"
Chris@69: #endif
Chris@69: 
Chris@69: #if (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR))
Chris@69: # include "arm/pitch_arm.h"
Chris@69: #endif
Chris@69: 
Chris@69: void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp,
Chris@69:       int len, int C, int arch);
Chris@69: 
Chris@69: void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y,
Chris@69:                   int len, int max_pitch, int *pitch, int arch);
Chris@69: 
Chris@69: opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
Chris@69:       int N, int *T0, int prev_period, opus_val16 prev_gain, int arch);
Chris@69: 
Chris@69: 
Chris@69: /* OPT: This is the kernel you really want to optimize. It gets used a lot
Chris@69:    by the prefilter and by the PLC. */
Chris@69: static OPUS_INLINE void xcorr_kernel_c(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len)
Chris@69: {
Chris@69:    int j;
Chris@69:    opus_val16 y_0, y_1, y_2, y_3;
Chris@69:    celt_assert(len>=3);
Chris@69:    y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */
Chris@69:    y_0=*y++;
Chris@69:    y_1=*y++;
Chris@69:    y_2=*y++;
Chris@69:    for (j=0;j<len-3;j+=4)
Chris@69:    {
Chris@69:       opus_val16 tmp;
Chris@69:       tmp = *x++;
Chris@69:       y_3=*y++;
Chris@69:       sum[0] = MAC16_16(sum[0],tmp,y_0);
Chris@69:       sum[1] = MAC16_16(sum[1],tmp,y_1);
Chris@69:       sum[2] = MAC16_16(sum[2],tmp,y_2);
Chris@69:       sum[3] = MAC16_16(sum[3],tmp,y_3);
Chris@69:       tmp=*x++;
Chris@69:       y_0=*y++;
Chris@69:       sum[0] = MAC16_16(sum[0],tmp,y_1);
Chris@69:       sum[1] = MAC16_16(sum[1],tmp,y_2);
Chris@69:       sum[2] = MAC16_16(sum[2],tmp,y_3);
Chris@69:       sum[3] = MAC16_16(sum[3],tmp,y_0);
Chris@69:       tmp=*x++;
Chris@69:       y_1=*y++;
Chris@69:       sum[0] = MAC16_16(sum[0],tmp,y_2);
Chris@69:       sum[1] = MAC16_16(sum[1],tmp,y_3);
Chris@69:       sum[2] = MAC16_16(sum[2],tmp,y_0);
Chris@69:       sum[3] = MAC16_16(sum[3],tmp,y_1);
Chris@69:       tmp=*x++;
Chris@69:       y_2=*y++;
Chris@69:       sum[0] = MAC16_16(sum[0],tmp,y_3);
Chris@69:       sum[1] = MAC16_16(sum[1],tmp,y_0);
Chris@69:       sum[2] = MAC16_16(sum[2],tmp,y_1);
Chris@69:       sum[3] = MAC16_16(sum[3],tmp,y_2);
Chris@69:    }
Chris@69:    if (j++<len)
Chris@69:    {
Chris@69:       opus_val16 tmp = *x++;
Chris@69:       y_3=*y++;
Chris@69:       sum[0] = MAC16_16(sum[0],tmp,y_0);
Chris@69:       sum[1] = MAC16_16(sum[1],tmp,y_1);
Chris@69:       sum[2] = MAC16_16(sum[2],tmp,y_2);
Chris@69:       sum[3] = MAC16_16(sum[3],tmp,y_3);
Chris@69:    }
Chris@69:    if (j++<len)
Chris@69:    {
Chris@69:       opus_val16 tmp=*x++;
Chris@69:       y_0=*y++;
Chris@69:       sum[0] = MAC16_16(sum[0],tmp,y_1);
Chris@69:       sum[1] = MAC16_16(sum[1],tmp,y_2);
Chris@69:       sum[2] = MAC16_16(sum[2],tmp,y_3);
Chris@69:       sum[3] = MAC16_16(sum[3],tmp,y_0);
Chris@69:    }
Chris@69:    if (j<len)
Chris@69:    {
Chris@69:       opus_val16 tmp=*x++;
Chris@69:       y_1=*y++;
Chris@69:       sum[0] = MAC16_16(sum[0],tmp,y_2);
Chris@69:       sum[1] = MAC16_16(sum[1],tmp,y_3);
Chris@69:       sum[2] = MAC16_16(sum[2],tmp,y_0);
Chris@69:       sum[3] = MAC16_16(sum[3],tmp,y_1);
Chris@69:    }
Chris@69: }
Chris@69: 
Chris@69: #ifndef OVERRIDE_XCORR_KERNEL
Chris@69: #define xcorr_kernel(x, y, sum, len, arch) \
Chris@69:     ((void)(arch),xcorr_kernel_c(x, y, sum, len))
Chris@69: #endif /* OVERRIDE_XCORR_KERNEL */
Chris@69: 
Chris@69: 
Chris@69: static OPUS_INLINE void dual_inner_prod_c(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
Chris@69:       int N, opus_val32 *xy1, opus_val32 *xy2)
Chris@69: {
Chris@69:    int i;
Chris@69:    opus_val32 xy01=0;
Chris@69:    opus_val32 xy02=0;
Chris@69:    for (i=0;i<N;i++)
Chris@69:    {
Chris@69:       xy01 = MAC16_16(xy01, x[i], y01[i]);
Chris@69:       xy02 = MAC16_16(xy02, x[i], y02[i]);
Chris@69:    }
Chris@69:    *xy1 = xy01;
Chris@69:    *xy2 = xy02;
Chris@69: }
Chris@69: 
Chris@69: #ifndef OVERRIDE_DUAL_INNER_PROD
Chris@69: # define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \
Chris@69:     ((void)(arch),dual_inner_prod_c(x, y01, y02, N, xy1, xy2))
Chris@69: #endif
Chris@69: 
Chris@69: /*We make sure a C version is always available for cases where the overhead of
Chris@69:   vectorization and passing around an arch flag aren't worth it.*/
Chris@69: static OPUS_INLINE opus_val32 celt_inner_prod_c(const opus_val16 *x,
Chris@69:       const opus_val16 *y, int N)
Chris@69: {
Chris@69:    int i;
Chris@69:    opus_val32 xy=0;
Chris@69:    for (i=0;i<N;i++)
Chris@69:       xy = MAC16_16(xy, x[i], y[i]);
Chris@69:    return xy;
Chris@69: }
Chris@69: 
Chris@69: #if !defined(OVERRIDE_CELT_INNER_PROD)
Chris@69: # define celt_inner_prod(x, y, N, arch) \
Chris@69:     ((void)(arch),celt_inner_prod_c(x, y, N))
Chris@69: #endif
Chris@69: 
Chris@69: #ifdef NON_STATIC_COMB_FILTER_CONST_C
Chris@69: void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
Chris@69:      opus_val16 g10, opus_val16 g11, opus_val16 g12);
Chris@69: #endif
Chris@69: 
Chris@69: 
Chris@69: #ifdef FIXED_POINT
Chris@69: opus_val32
Chris@69: #else
Chris@69: void
Chris@69: #endif
Chris@69: celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y,
Chris@69:       opus_val32 *xcorr, int len, int max_pitch, int arch);
Chris@69: 
Chris@69: #ifndef OVERRIDE_PITCH_XCORR
Chris@69: # define celt_pitch_xcorr celt_pitch_xcorr_c
Chris@69: #endif
Chris@69: 
Chris@69: #endif