cannam@154: /*Copyright (c) 2013, Xiph.Org Foundation and contributors.
cannam@154: 
cannam@154:   All rights reserved.
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 are met:
cannam@154: 
cannam@154:     * Redistributions of source code must retain the above copyright notice,
cannam@154:        this list of conditions and the following disclaimer.
cannam@154:     * Redistributions in binary form must reproduce the above copyright notice,
cannam@154:        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 "AS IS"
cannam@154:   AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
cannam@154:   IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
cannam@154:   ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
cannam@154:   LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
cannam@154:   CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
cannam@154:   SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
cannam@154:   INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
cannam@154:   CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
cannam@154:   ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
cannam@154:   POSSIBILITY OF SUCH DAMAGE.*/
cannam@154: 
cannam@154: #ifndef KISS_FFT_MIPSR1_H
cannam@154: #define KISS_FFT_MIPSR1_H
cannam@154: 
cannam@154: #if !defined(KISS_FFT_GUTS_H)
cannam@154: #error "This file should only be included from _kiss_fft_guts.h"
cannam@154: #endif
cannam@154: 
cannam@154: #ifdef FIXED_POINT
cannam@154: 
cannam@154: #define S_MUL_ADD(a, b, c, d) (S_MUL(a,b)+S_MUL(c,d))
cannam@154: #define S_MUL_SUB(a, b, c, d) (S_MUL(a,b)-S_MUL(c,d))
cannam@154: 
cannam@154: #undef S_MUL_ADD
cannam@154: static inline int S_MUL_ADD(int a, int b, int c, int d) {
cannam@154:     int m;
cannam@154:     asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
cannam@154:     asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
cannam@154:     asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
cannam@154:     return m;
cannam@154: }
cannam@154: 
cannam@154: #undef S_MUL_SUB
cannam@154: static inline int S_MUL_SUB(int a, int b, int c, int d) {
cannam@154:     int m;
cannam@154:     asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
cannam@154:     asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
cannam@154:     asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
cannam@154:     return m;
cannam@154: }
cannam@154: 
cannam@154: #undef C_MUL
cannam@154: #   define C_MUL(m,a,b) (m=C_MUL_fun(a,b))
cannam@154: static inline kiss_fft_cpx C_MUL_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) {
cannam@154:     kiss_fft_cpx m;
cannam@154: 
cannam@154:     asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r));
cannam@154:     asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i));
cannam@154:     asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15));
cannam@154:     asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i));
cannam@154:     asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r));
cannam@154:     asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15));
cannam@154: 
cannam@154:     return m;
cannam@154: }
cannam@154: #undef C_MULC
cannam@154: #   define C_MULC(m,a,b) (m=C_MULC_fun(a,b))
cannam@154: static inline kiss_fft_cpx C_MULC_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) {
cannam@154:     kiss_fft_cpx m;
cannam@154: 
cannam@154:     asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r));
cannam@154:     asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i));
cannam@154:     asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15));
cannam@154:     asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r));
cannam@154:     asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i));
cannam@154:     asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15));
cannam@154: 
cannam@154:     return m;
cannam@154: }
cannam@154: 
cannam@154: #endif /* FIXED_POINT */
cannam@154: 
cannam@154: #define OVERRIDE_kf_bfly5
cannam@154: static void kf_bfly5(
cannam@154:                      kiss_fft_cpx * Fout,
cannam@154:                      const size_t fstride,
cannam@154:                      const kiss_fft_state *st,
cannam@154:                      int m,
cannam@154:                      int N,
cannam@154:                      int mm
cannam@154:                     )
cannam@154: {
cannam@154:    kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4;
cannam@154:    int i, u;
cannam@154:    kiss_fft_cpx scratch[13];
cannam@154: 
cannam@154:    const kiss_twiddle_cpx *tw;
cannam@154:    kiss_twiddle_cpx ya,yb;
cannam@154:    kiss_fft_cpx * Fout_beg = Fout;
cannam@154: 
cannam@154: #ifdef FIXED_POINT
cannam@154:    ya.r = 10126;
cannam@154:    ya.i = -31164;
cannam@154:    yb.r = -26510;
cannam@154:    yb.i = -19261;
cannam@154: #else
cannam@154:    ya = st->twiddles[fstride*m];
cannam@154:    yb = st->twiddles[fstride*2*m];
cannam@154: #endif
cannam@154: 
cannam@154:    tw=st->twiddles;
cannam@154: 
cannam@154:    for (i=0;i<N;i++)
cannam@154:    {
cannam@154:       Fout = Fout_beg + i*mm;
cannam@154:       Fout0=Fout;
cannam@154:       Fout1=Fout0+m;
cannam@154:       Fout2=Fout0+2*m;
cannam@154:       Fout3=Fout0+3*m;
cannam@154:       Fout4=Fout0+4*m;
cannam@154: 
cannam@154:       /* For non-custom modes, m is guaranteed to be a multiple of 4. */
cannam@154:       for ( u=0; u<m; ++u ) {
cannam@154:          scratch[0] = *Fout0;
cannam@154: 
cannam@154: 
cannam@154:          C_MUL(scratch[1] ,*Fout1, tw[u*fstride]);
cannam@154:          C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]);
cannam@154:          C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]);
cannam@154:          C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]);
cannam@154: 
cannam@154:          C_ADD( scratch[7],scratch[1],scratch[4]);
cannam@154:          C_SUB( scratch[10],scratch[1],scratch[4]);
cannam@154:          C_ADD( scratch[8],scratch[2],scratch[3]);
cannam@154:          C_SUB( scratch[9],scratch[2],scratch[3]);
cannam@154: 
cannam@154:          Fout0->r += scratch[7].r + scratch[8].r;
cannam@154:          Fout0->i += scratch[7].i + scratch[8].i;
cannam@154:          scratch[5].r = scratch[0].r + S_MUL_ADD(scratch[7].r,ya.r,scratch[8].r,yb.r);
cannam@154:          scratch[5].i = scratch[0].i + S_MUL_ADD(scratch[7].i,ya.r,scratch[8].i,yb.r);
cannam@154: 
cannam@154:          scratch[6].r =  S_MUL_ADD(scratch[10].i,ya.i,scratch[9].i,yb.i);
cannam@154:          scratch[6].i =  -S_MUL_ADD(scratch[10].r,ya.i,scratch[9].r,yb.i);
cannam@154: 
cannam@154:          C_SUB(*Fout1,scratch[5],scratch[6]);
cannam@154:          C_ADD(*Fout4,scratch[5],scratch[6]);
cannam@154: 
cannam@154:          scratch[11].r = scratch[0].r + S_MUL_ADD(scratch[7].r,yb.r,scratch[8].r,ya.r);
cannam@154:          scratch[11].i = scratch[0].i + S_MUL_ADD(scratch[7].i,yb.r,scratch[8].i,ya.r);
cannam@154: 
cannam@154:          scratch[12].r =  S_MUL_SUB(scratch[9].i,ya.i,scratch[10].i,yb.i);
cannam@154:          scratch[12].i =  S_MUL_SUB(scratch[10].r,yb.i,scratch[9].r,ya.i);
cannam@154: 
cannam@154:          C_ADD(*Fout2,scratch[11],scratch[12]);
cannam@154:          C_SUB(*Fout3,scratch[11],scratch[12]);
cannam@154: 
cannam@154:          ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4;
cannam@154:       }
cannam@154:    }
cannam@154: }
cannam@154: 
cannam@154: 
cannam@154: #endif /* KISS_FFT_MIPSR1_H */