cannam@154: /* Copyright (c) 2007-2008 CSIRO
cannam@154:    Copyright (c) 2007-2009 Xiph.Org Foundation
cannam@154:    Written by Jean-Marc Valin */
cannam@154: /**
cannam@154:    @file pitch.h
cannam@154:    @brief Pitch analysis
cannam@154:  */
cannam@154: 
cannam@154: /*
cannam@154:    Redistribution and use in source and binary forms, with or without
cannam@154:    modification, are permitted provided that the following conditions
cannam@154:    are met:
cannam@154: 
cannam@154:    - Redistributions of source code must retain the above copyright
cannam@154:    notice, this list of conditions and the following disclaimer.
cannam@154: 
cannam@154:    - Redistributions in binary form must reproduce the above copyright
cannam@154:    notice, this list of conditions and the following disclaimer in the
cannam@154:    documentation and/or other materials provided with the distribution.
cannam@154: 
cannam@154:    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
cannam@154:    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
cannam@154:    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
cannam@154:    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
cannam@154:    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
cannam@154:    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
cannam@154:    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
cannam@154:    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
cannam@154:    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
cannam@154:    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
cannam@154:    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
cannam@154: */
cannam@154: 
cannam@154: #ifndef PITCH_MIPSR1_H
cannam@154: #define PITCH_MIPSR1_H
cannam@154: 
cannam@154: #define OVERRIDE_DUAL_INNER_PROD
cannam@154: static inline void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
cannam@154:       int N, opus_val32 *xy1, opus_val32 *xy2, int arch)
cannam@154: {
cannam@154:    int j;
cannam@154:    opus_val32 xy01=0;
cannam@154:    opus_val32 xy02=0;
cannam@154: 
cannam@154:    (void)arch;
cannam@154: 
cannam@154:    asm volatile("MULT $ac1, $0, $0");
cannam@154:    asm volatile("MULT $ac2, $0, $0");
cannam@154:    /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
cannam@154:    for (j=0;j<N;j++)
cannam@154:    {
cannam@154:       asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
cannam@154:       asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
cannam@154:       ++j;
cannam@154:       asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
cannam@154:       asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
cannam@154:    }
cannam@154:    asm volatile ("mflo %0, $ac1": "=r"(xy01));
cannam@154:    asm volatile ("mflo %0, $ac2": "=r"(xy02));
cannam@154:    *xy1 = xy01;
cannam@154:    *xy2 = xy02;
cannam@154: }
cannam@154: 
cannam@154: static inline void xcorr_kernel_mips(const opus_val16 * x,
cannam@154:       const opus_val16 * y, opus_val32 sum[4], int len)
cannam@154: {
cannam@154:    int j;
cannam@154:    opus_val16 y_0, y_1, y_2, y_3;
cannam@154: 
cannam@154:     opus_int64 sum_0, sum_1, sum_2, sum_3;
cannam@154:     sum_0 =  (opus_int64)sum[0];
cannam@154:     sum_1 =  (opus_int64)sum[1];
cannam@154:     sum_2 =  (opus_int64)sum[2];
cannam@154:     sum_3 =  (opus_int64)sum[3];
cannam@154: 
cannam@154:     y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */
cannam@154:     y_0=*y++;
cannam@154:     y_1=*y++;
cannam@154:     y_2=*y++;
cannam@154:     for (j=0;j<len-3;j+=4)
cannam@154:     {
cannam@154:         opus_val16 tmp;
cannam@154:         tmp = *x++;
cannam@154:         y_3=*y++;
cannam@154: 
cannam@154:         sum_0 = __builtin_mips_madd( sum_0, tmp, y_0);
cannam@154:         sum_1 = __builtin_mips_madd( sum_1, tmp, y_1);
cannam@154:         sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
cannam@154:         sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
cannam@154: 
cannam@154:         tmp=*x++;
cannam@154:         y_0=*y++;
cannam@154: 
cannam@154:         sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
cannam@154:         sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
cannam@154:         sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
cannam@154:         sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
cannam@154: 
cannam@154:        tmp=*x++;
cannam@154:        y_1=*y++;
cannam@154: 
cannam@154:        sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
cannam@154:        sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
cannam@154:        sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
cannam@154:        sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
cannam@154: 
cannam@154: 
cannam@154:       tmp=*x++;
cannam@154:       y_2=*y++;
cannam@154: 
cannam@154:        sum_0 = __builtin_mips_madd( sum_0, tmp, y_3 );
cannam@154:        sum_1 = __builtin_mips_madd( sum_1, tmp, y_0 );
cannam@154:        sum_2 = __builtin_mips_madd( sum_2, tmp, y_1);
cannam@154:        sum_3 = __builtin_mips_madd( sum_3, tmp, y_2);
cannam@154: 
cannam@154:    }
cannam@154:    if (j++<len)
cannam@154:    {
cannam@154:       opus_val16 tmp = *x++;
cannam@154:       y_3=*y++;
cannam@154: 
cannam@154:        sum_0 = __builtin_mips_madd( sum_0, tmp, y_0 );
cannam@154:        sum_1 = __builtin_mips_madd( sum_1, tmp, y_1 );
cannam@154:        sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
cannam@154:        sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
cannam@154:    }
cannam@154: 
cannam@154:    if (j++<len)
cannam@154:    {
cannam@154:       opus_val16 tmp=*x++;
cannam@154:       y_0=*y++;
cannam@154: 
cannam@154:       sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
cannam@154:       sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
cannam@154:       sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
cannam@154:       sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
cannam@154:    }
cannam@154: 
cannam@154:    if (j<len)
cannam@154:    {
cannam@154:       opus_val16 tmp=*x++;
cannam@154:       y_1=*y++;
cannam@154: 
cannam@154:        sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
cannam@154:        sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
cannam@154:        sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
cannam@154:        sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
cannam@154: 
cannam@154:    }
cannam@154: 
cannam@154:    sum[0] = (opus_val32)sum_0;
cannam@154:    sum[1] = (opus_val32)sum_1;
cannam@154:    sum[2] = (opus_val32)sum_2;
cannam@154:    sum[3] = (opus_val32)sum_3;
cannam@154: }
cannam@154: 
cannam@154: #define OVERRIDE_XCORR_KERNEL
cannam@154: #define xcorr_kernel(x, y, sum, len, arch) \
cannam@154:     ((void)(arch), xcorr_kernel_mips(x, y, sum, len))
cannam@154: 
cannam@154: #endif /* PITCH_MIPSR1_H */