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_MIPSR1_H
Chris@69: #define PITCH_MIPSR1_H
Chris@69: 
Chris@69: #define OVERRIDE_DUAL_INNER_PROD
Chris@69: static inline void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
Chris@69:       int N, opus_val32 *xy1, opus_val32 *xy2, int arch)
Chris@69: {
Chris@69:    int j;
Chris@69:    opus_val32 xy01=0;
Chris@69:    opus_val32 xy02=0;
Chris@69: 
Chris@69:    (void)arch;
Chris@69: 
Chris@69:    asm volatile("MULT $ac1, $0, $0");
Chris@69:    asm volatile("MULT $ac2, $0, $0");
Chris@69:    /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
Chris@69:    for (j=0;j<N;j++)
Chris@69:    {
Chris@69:       asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
Chris@69:       asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
Chris@69:       ++j;
Chris@69:       asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
Chris@69:       asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
Chris@69:    }
Chris@69:    asm volatile ("mflo %0, $ac1": "=r"(xy01));
Chris@69:    asm volatile ("mflo %0, $ac2": "=r"(xy02));
Chris@69:    *xy1 = xy01;
Chris@69:    *xy2 = xy02;
Chris@69: }
Chris@69: 
Chris@69: static inline void xcorr_kernel_mips(const opus_val16 * x,
Chris@69:       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: 
Chris@69:     opus_int64 sum_0, sum_1, sum_2, sum_3;
Chris@69:     sum_0 =  (opus_int64)sum[0];
Chris@69:     sum_1 =  (opus_int64)sum[1];
Chris@69:     sum_2 =  (opus_int64)sum[2];
Chris@69:     sum_3 =  (opus_int64)sum[3];
Chris@69: 
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: 
Chris@69:         sum_0 = __builtin_mips_madd( sum_0, tmp, y_0);
Chris@69:         sum_1 = __builtin_mips_madd( sum_1, tmp, y_1);
Chris@69:         sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
Chris@69:         sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
Chris@69: 
Chris@69:         tmp=*x++;
Chris@69:         y_0=*y++;
Chris@69: 
Chris@69:         sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
Chris@69:         sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
Chris@69:         sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
Chris@69:         sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
Chris@69: 
Chris@69:        tmp=*x++;
Chris@69:        y_1=*y++;
Chris@69: 
Chris@69:        sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
Chris@69:        sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
Chris@69:        sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
Chris@69:        sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
Chris@69: 
Chris@69: 
Chris@69:       tmp=*x++;
Chris@69:       y_2=*y++;
Chris@69: 
Chris@69:        sum_0 = __builtin_mips_madd( sum_0, tmp, y_3 );
Chris@69:        sum_1 = __builtin_mips_madd( sum_1, tmp, y_0 );
Chris@69:        sum_2 = __builtin_mips_madd( sum_2, tmp, y_1);
Chris@69:        sum_3 = __builtin_mips_madd( sum_3, tmp, y_2);
Chris@69: 
Chris@69:    }
Chris@69:    if (j++<len)
Chris@69:    {
Chris@69:       opus_val16 tmp = *x++;
Chris@69:       y_3=*y++;
Chris@69: 
Chris@69:        sum_0 = __builtin_mips_madd( sum_0, tmp, y_0 );
Chris@69:        sum_1 = __builtin_mips_madd( sum_1, tmp, y_1 );
Chris@69:        sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
Chris@69:        sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
Chris@69:    }
Chris@69: 
Chris@69:    if (j++<len)
Chris@69:    {
Chris@69:       opus_val16 tmp=*x++;
Chris@69:       y_0=*y++;
Chris@69: 
Chris@69:       sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
Chris@69:       sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
Chris@69:       sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
Chris@69:       sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
Chris@69:    }
Chris@69: 
Chris@69:    if (j<len)
Chris@69:    {
Chris@69:       opus_val16 tmp=*x++;
Chris@69:       y_1=*y++;
Chris@69: 
Chris@69:        sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
Chris@69:        sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
Chris@69:        sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
Chris@69:        sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
Chris@69: 
Chris@69:    }
Chris@69: 
Chris@69:    sum[0] = (opus_val32)sum_0;
Chris@69:    sum[1] = (opus_val32)sum_1;
Chris@69:    sum[2] = (opus_val32)sum_2;
Chris@69:    sum[3] = (opus_val32)sum_3;
Chris@69: }
Chris@69: 
Chris@69: #define OVERRIDE_XCORR_KERNEL
Chris@69: #define xcorr_kernel(x, y, sum, len, arch) \
Chris@69:     ((void)(arch), xcorr_kernel_mips(x, y, sum, len))
Chris@69: 
Chris@69: #endif /* PITCH_MIPSR1_H */