• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Copyright (c) 2015 Xiph.Org Foundation
2    Written by Viswanath Puttagunta */
3 /**
4    @file celt_ne10_mdct.c
5    @brief ARM Neon optimizations for mdct using NE10 library
6  */
7 
8 /*
9    Redistribution and use in source and binary forms, with or without
10    modification, are permitted provided that the following conditions
11    are met:
12 
13    - Redistributions of source code must retain the above copyright
14    notice, this list of conditions and the following disclaimer.
15 
16    - Redistributions in binary form must reproduce the above copyright
17    notice, this list of conditions and the following disclaimer in the
18    documentation and/or other materials provided with the distribution.
19 
20    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
24    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
25    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
26    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
27    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
28    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
29    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
30    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 */
32 
33 #ifndef SKIP_CONFIG_H
34 #ifdef HAVE_CONFIG_H
35 #include "config.h"
36 #endif
37 #endif
38 
39 #include "kiss_fft.h"
40 #include "_kiss_fft_guts.h"
41 #include "mdct.h"
42 #include "stack_alloc.h"
43 
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)44 void clt_mdct_forward_neon(const mdct_lookup *l,
45                            kiss_fft_scalar *in,
46                            kiss_fft_scalar * OPUS_RESTRICT out,
47                            const opus_val16 *window,
48                            int overlap, int shift, int stride, int arch)
49 {
50    int i;
51    int N, N2, N4;
52    VARDECL(kiss_fft_scalar, f);
53    VARDECL(kiss_fft_cpx, f2);
54    const kiss_fft_state *st = l->kfft[shift];
55    const kiss_twiddle_scalar *trig;
56 
57    SAVE_STACK;
58 
59    N = l->n;
60    trig = l->trig;
61    for (i=0;i<shift;i++)
62    {
63       N >>= 1;
64       trig += N;
65    }
66    N2 = N>>1;
67    N4 = N>>2;
68 
69    ALLOC(f, N2, kiss_fft_scalar);
70    ALLOC(f2, N4, kiss_fft_cpx);
71 
72    /* Consider the input to be composed of four blocks: [a, b, c, d] */
73    /* Window, shuffle, fold */
74    {
75       /* Temp pointers to make it really clear to the compiler what we're doing */
76       const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1);
77       const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1);
78       kiss_fft_scalar * OPUS_RESTRICT yp = f;
79       const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1);
80       const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1;
81       for(i=0;i<((overlap+3)>>2);i++)
82       {
83          /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/
84          *yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2);
85          *yp++ = MULT16_32_Q15(*wp1, *xp1)    - MULT16_32_Q15(*wp2, xp2[-N2]);
86          xp1+=2;
87          xp2-=2;
88          wp1+=2;
89          wp2-=2;
90       }
91       wp1 = window;
92       wp2 = window+overlap-1;
93       for(;i<N4-((overlap+3)>>2);i++)
94       {
95          /* Real part arranged as a-bR, Imag part arranged as -c-dR */
96          *yp++ = *xp2;
97          *yp++ = *xp1;
98          xp1+=2;
99          xp2-=2;
100       }
101       for(;i<N4;i++)
102       {
103          /* Real part arranged as a-bR, Imag part arranged as -c-dR */
104          *yp++ =  -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2);
105          *yp++ = MULT16_32_Q15(*wp2, *xp1)     + MULT16_32_Q15(*wp1, xp2[N2]);
106          xp1+=2;
107          xp2-=2;
108          wp1+=2;
109          wp2-=2;
110       }
111    }
112    /* Pre-rotation */
113    {
114       kiss_fft_scalar * OPUS_RESTRICT yp = f;
115       const kiss_twiddle_scalar *t = &trig[0];
116       for(i=0;i<N4;i++)
117       {
118          kiss_fft_cpx yc;
119          kiss_twiddle_scalar t0, t1;
120          kiss_fft_scalar re, im, yr, yi;
121          t0 = t[i];
122          t1 = t[N4+i];
123          re = *yp++;
124          im = *yp++;
125          yr = S_MUL(re,t0)  -  S_MUL(im,t1);
126          yi = S_MUL(im,t0)  +  S_MUL(re,t1);
127          yc.r = yr;
128          yc.i = yi;
129          f2[i] = yc;
130       }
131    }
132 
133    opus_fft(st, f2, (kiss_fft_cpx *)f, arch);
134 
135    /* Post-rotate */
136    {
137       /* Temp pointers to make it really clear to the compiler what we're doing */
138       const kiss_fft_cpx * OPUS_RESTRICT fp = (kiss_fft_cpx *)f;
139       kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
140       kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1);
141       const kiss_twiddle_scalar *t = &trig[0];
142       /* Temp pointers to make it really clear to the compiler what we're doing */
143       for(i=0;i<N4;i++)
144       {
145          kiss_fft_scalar yr, yi;
146          yr = S_MUL(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]);
147          yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]);
148          *yp1 = yr;
149          *yp2 = yi;
150          fp++;
151          yp1 += 2*stride;
152          yp2 -= 2*stride;
153       }
154    }
155    RESTORE_STACK;
156 }
157 
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)158 void clt_mdct_backward_neon(const mdct_lookup *l,
159                             kiss_fft_scalar *in,
160                             kiss_fft_scalar * OPUS_RESTRICT out,
161                             const opus_val16 * OPUS_RESTRICT window,
162                             int overlap, int shift, int stride, int arch)
163 {
164    int i;
165    int N, N2, N4;
166    VARDECL(kiss_fft_scalar, f);
167    const kiss_twiddle_scalar *trig;
168    const kiss_fft_state *st = l->kfft[shift];
169 
170    N = l->n;
171    trig = l->trig;
172    for (i=0;i<shift;i++)
173    {
174       N >>= 1;
175       trig += N;
176    }
177    N2 = N>>1;
178    N4 = N>>2;
179 
180    ALLOC(f, N2, kiss_fft_scalar);
181 
182    /* Pre-rotate */
183    {
184       /* Temp pointers to make it really clear to the compiler what we're doing */
185       const kiss_fft_scalar * OPUS_RESTRICT xp1 = in;
186       const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1);
187       kiss_fft_scalar * OPUS_RESTRICT yp = f;
188       const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0];
189       for(i=0;i<N4;i++)
190       {
191          kiss_fft_scalar yr, yi;
192          yr = S_MUL(*xp2, t[i]) + S_MUL(*xp1, t[N4+i]);
193          yi = S_MUL(*xp1, t[i]) - S_MUL(*xp2, t[N4+i]);
194          yp[2*i] = yr;
195          yp[2*i+1] = yi;
196          xp1+=2*stride;
197          xp2-=2*stride;
198       }
199    }
200 
201    opus_ifft(st, (kiss_fft_cpx *)f, (kiss_fft_cpx*)(out+(overlap>>1)), arch);
202 
203    /* Post-rotate and de-shuffle from both ends of the buffer at once to make
204       it in-place. */
205    {
206       kiss_fft_scalar * yp0 = out+(overlap>>1);
207       kiss_fft_scalar * yp1 = out+(overlap>>1)+N2-2;
208       const kiss_twiddle_scalar *t = &trig[0];
209       /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the
210          middle pair will be computed twice. */
211       for(i=0;i<(N4+1)>>1;i++)
212       {
213          kiss_fft_scalar re, im, yr, yi;
214          kiss_twiddle_scalar t0, t1;
215          re = yp0[0];
216          im = yp0[1];
217          t0 = t[i];
218          t1 = t[N4+i];
219          /* We'd scale up by 2 here, but instead it's done when mixing the windows */
220          yr = S_MUL(re,t0) + S_MUL(im,t1);
221          yi = S_MUL(re,t1) - S_MUL(im,t0);
222          re = yp1[0];
223          im = yp1[1];
224          yp0[0] = yr;
225          yp1[1] = yi;
226 
227          t0 = t[(N4-i-1)];
228          t1 = t[(N2-i-1)];
229          /* We'd scale up by 2 here, but instead it's done when mixing the windows */
230          yr = S_MUL(re,t0) + S_MUL(im,t1);
231          yi = S_MUL(re,t1) - S_MUL(im,t0);
232          yp1[0] = yr;
233          yp0[1] = yi;
234          yp0 += 2;
235          yp1 -= 2;
236       }
237    }
238 
239    /* Mirror on both sides for TDAC */
240    {
241       kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1;
242       kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
243       const opus_val16 * OPUS_RESTRICT wp1 = window;
244       const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1;
245 
246       for(i = 0; i < overlap/2; i++)
247       {
248          kiss_fft_scalar x1, x2;
249          x1 = *xp1;
250          x2 = *yp1;
251          *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1);
252          *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1);
253          wp1++;
254          wp2--;
255       }
256    }
257    RESTORE_STACK;
258 }
259