drivers/staging/echo/fir.h

   1 /*
   2  * SpanDSP - a series of DSP components for telephony
   3  *
   4  * fir.h - General telephony FIR routines
   5  *
   6  * Written by Steve Underwood <steveu@coppice.org>
   7  *
   8  * Copyright (C) 2002 Steve Underwood
   9  *
  10  * All rights reserved.
  11  *
  12  * This program is free software; you can redistribute it and/or modify
  13  * it under the terms of the GNU General Public License version 2, as
  14  * published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  * GNU General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU General Public License
  22  * along with this program; if not, write to the Free Software
  23  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  24  *
  25  * $Id: fir.h,v 1.8 2006/10/24 13:45:28 steveu Exp $
  26  */
  27
  28 /*! \page fir_page FIR filtering
  29 \section fir_page_sec_1 What does it do?
  30 ???.
  31
  32 \section fir_page_sec_2 How does it work?
  33 ???.
  34 */
  35
  36 #if !defined(_FIR_H_)
  37 #define _FIR_H_
  38
  39 /*
  40    Blackfin NOTES & IDEAS:
  41
  42    A simple dot product function is used to implement the filter.  This performs
  43    just one MAC/cycle which is inefficient but was easy to implement as a first
  44    pass.  The current Blackfin code also uses an unrolled form of the filter
  45    history to avoid 0 length hardware loop issues.  This is wasteful of
  46    memory.
  47
  48    Ideas for improvement:
  49
  50    1/ Rewrite filter for dual MAC inner loop.  The issue here is handling
  51    history sample offsets that are 16 bit aligned - the dual MAC needs
  52    32 bit aligmnent.  There are some good examples in libbfdsp.
  53
  54    2/ Use the hardware circular buffer facility tohalve memory usage.
  55
  56    3/ Consider using internal memory.
  57
  58    Using less memory might also improve speed as cache misses will be
  59    reduced. A drop in MIPs and memory approaching 50% should be
  60    possible.
  61
  62    The foreground and background filters currenlty use a total of
  63    about 10 MIPs/ch as measured with speedtest.c on a 256 TAP echo
  64    can.
  65 */
  66
  67 #if defined(USE_MMX)  ||  defined(USE_SSE2)
  68 #include "mmx.h"
  69 #endif
  70
  71 /*!
  72     16 bit integer FIR descriptor. This defines the working state for a single
  73     instance of an FIR filter using 16 bit integer coefficients.
  74 */
  75 typedef struct
  76 {
  77         int taps;
  78         int curr_pos;
  79         const int16_t *coeffs;
  80         int16_t *history;
  81 } fir16_state_t;
  82
  83 /*!
  84     32 bit integer FIR descriptor. This defines the working state for a single
  85     instance of an FIR filter using 32 bit integer coefficients, and filtering
  86     16 bit integer data.
  87 */
  88 typedef struct
  89 {
  90         int taps;
  91         int curr_pos;
  92         const int32_t *coeffs;
  93         int16_t *history;
  94 } fir32_state_t;
  95
  96 /*!
  97     Floating point FIR descriptor. This defines the working state for a single
  98     instance of an FIR filter using floating point coefficients and data.
  99 */
 100 typedef struct
 101 {
 102         int taps;
 103         int curr_pos;
 104         const float *coeffs;
 105         float *history;
 106 } fir_float_state_t;
 107
 108 #ifdef __cplusplus
 109 extern "C" {
 110 #endif
 111
 112 static __inline__ const int16_t *fir16_create(fir16_state_t *fir,
 113                                               const int16_t *coeffs,
 114                                               int taps)
 115 {
 116         fir->taps = taps;
 117         fir->curr_pos = taps - 1;
 118         fir->coeffs = coeffs;
 119 #if defined(USE_MMX)  ||  defined(USE_SSE2) || defined(__bfin__)
 120         fir->history = kcalloc(2*taps, sizeof(int16_t), GFP_KERNEL);
 121 #else
 122         fir->history = kcalloc(taps, sizeof(int16_t), GFP_KERNEL);
 123 #endif
 124         return fir->history;
 125 }
 126 /*- End of function --------------------------------------------------------*/
 127
 128 static __inline__ void fir16_flush(fir16_state_t *fir)
 129 {
 130 #if defined(USE_MMX)  ||  defined(USE_SSE2) || defined(__bfin__)
 131     memset(fir->history, 0, 2*fir->taps*sizeof(int16_t));
 132 #else
 133     memset(fir->history, 0, fir->taps*sizeof(int16_t));
 134 #endif
 135 }
 136 /*- End of function --------------------------------------------------------*/
 137
 138 static __inline__ void fir16_free(fir16_state_t *fir)
 139 {
 140         kfree(fir->history);
 141 }
 142 /*- End of function --------------------------------------------------------*/
 143
 144 #ifdef __bfin__
 145 static inline int32_t dot_asm(short *x, short *y, int len)
 146 {
 147    int dot;
 148
 149    len--;
 150
 151    __asm__
 152    (
 153    "I0 = %1;\n\t"
 154    "I1 = %2;\n\t"
 155    "A0 = 0;\n\t"
 156    "R0.L = W[I0++] || R1.L = W[I1++];\n\t"
 157    "LOOP dot%= LC0 = %3;\n\t"
 158    "LOOP_BEGIN dot%=;\n\t"
 159       "A0 += R0.L * R1.L (IS) || R0.L = W[I0++] || R1.L = W[I1++];\n\t"
 160    "LOOP_END dot%=;\n\t"
 161    "A0 += R0.L*R1.L (IS);\n\t"
 162    "R0 = A0;\n\t"
 163    "%0 = R0;\n\t"
 164    : "=&d" (dot)
 165    : "a" (x), "a" (y), "a" (len)
 166    : "I0", "I1", "A1", "A0", "R0", "R1"
 167    );
 168
 169    return dot;
 170 }
 171 #endif
 172 /*- End of function --------------------------------------------------------*/
 173
 174 static __inline__ int16_t fir16(fir16_state_t *fir, int16_t sample)
 175 {
 176     int32_t y;
 177 #if defined(USE_MMX)
 178     int i;
 179     mmx_t *mmx_coeffs;
 180     mmx_t *mmx_hist;
 181
 182     fir->history[fir->curr_pos] = sample;
 183     fir->history[fir->curr_pos + fir->taps] = sample;
 184
 185     mmx_coeffs = (mmx_t *) fir->coeffs;
 186     mmx_hist = (mmx_t *) &fir->history[fir->curr_pos];
 187     i = fir->taps;
 188     pxor_r2r(mm4, mm4);
 189     /* 8 samples per iteration, so the filter must be a multiple of 8 long. */
 190     while (i > 0)
 191     {
 192         movq_m2r(mmx_coeffs[0], mm0);
 193         movq_m2r(mmx_coeffs[1], mm2);
 194         movq_m2r(mmx_hist[0], mm1);
 195         movq_m2r(mmx_hist[1], mm3);
 196         mmx_coeffs += 2;
 197         mmx_hist += 2;
 198         pmaddwd_r2r(mm1, mm0);
 199         pmaddwd_r2r(mm3, mm2);
 200         paddd_r2r(mm0, mm4);
 201         paddd_r2r(mm2, mm4);
 202         i -= 8;
 203     }
 204     movq_r2r(mm4, mm0);
 205     psrlq_i2r(32, mm0);
 206     paddd_r2r(mm0, mm4);
 207     movd_r2m(mm4, y);
 208     emms();
 209 #elif defined(USE_SSE2)
 210     int i;
 211     xmm_t *xmm_coeffs;
 212     xmm_t *xmm_hist;
 213
 214     fir->history[fir->curr_pos] = sample;
 215     fir->history[fir->curr_pos + fir->taps] = sample;
 216
 217     xmm_coeffs = (xmm_t *) fir->coeffs;
 218     xmm_hist = (xmm_t *) &fir->history[fir->curr_pos];
 219     i = fir->taps;
 220     pxor_r2r(xmm4, xmm4);
 221     /* 16 samples per iteration, so the filter must be a multiple of 16 long. */
 222     while (i > 0)
 223     {
 224         movdqu_m2r(xmm_coeffs[0], xmm0);
 225         movdqu_m2r(xmm_coeffs[1], xmm2);
 226         movdqu_m2r(xmm_hist[0], xmm1);
 227         movdqu_m2r(xmm_hist[1], xmm3);
 228         xmm_coeffs += 2;
 229         xmm_hist += 2;
 230         pmaddwd_r2r(xmm1, xmm0);
 231         pmaddwd_r2r(xmm3, xmm2);
 232         paddd_r2r(xmm0, xmm4);
 233         paddd_r2r(xmm2, xmm4);
 234         i -= 16;
 235     }
 236     movdqa_r2r(xmm4, xmm0);
 237     psrldq_i2r(8, xmm0);
 238     paddd_r2r(xmm0, xmm4);
 239     movdqa_r2r(xmm4, xmm0);
 240     psrldq_i2r(4, xmm0);
 241     paddd_r2r(xmm0, xmm4);
 242     movd_r2m(xmm4, y);
 243 #elif defined(__bfin__)
 244     fir->history[fir->curr_pos] = sample;
 245     fir->history[fir->curr_pos + fir->taps] = sample;
 246     y = dot_asm((int16_t*)fir->coeffs, &fir->history[fir->curr_pos], fir->taps);
 247 #else
 248     int i;
 249     int offset1;
 250     int offset2;
 251
 252     fir->history[fir->curr_pos] = sample;
 253
 254     offset2 = fir->curr_pos;
 255     offset1 = fir->taps - offset2;
 256     y = 0;
 257     for (i = fir->taps - 1;  i >= offset1;  i--)
 258         y += fir->coeffs[i]*fir->history[i - offset1];
 259     for (  ;  i >= 0;  i--)
 260         y += fir->coeffs[i]*fir->history[i + offset2];
 261 #endif
 262     if (fir->curr_pos <= 0)
 263         fir->curr_pos = fir->taps;
 264     fir->curr_pos--;
 265     return (int16_t) (y >> 15);
 266 }
 267 /*- End of function --------------------------------------------------------*/
 268
 269 static __inline__ const int16_t *fir32_create(fir32_state_t *fir,
 270                                               const int32_t *coeffs,
 271                                               int taps)
 272 {
 273     fir->taps = taps;
 274     fir->curr_pos = taps - 1;
 275     fir->coeffs = coeffs;
 276     fir->history = kcalloc(taps, sizeof(int16_t), GFP_KERNEL);
 277     return fir->history;
 278 }
 279 /*- End of function --------------------------------------------------------*/
 280
 281 static __inline__ void fir32_flush(fir32_state_t *fir)
 282 {
 283     memset(fir->history, 0, fir->taps*sizeof(int16_t));
 284 }
 285 /*- End of function --------------------------------------------------------*/
 286
 287 static __inline__ void fir32_free(fir32_state_t *fir)
 288 {
 289     kfree(fir->history);
 290 }
 291 /*- End of function --------------------------------------------------------*/
 292
 293 static __inline__ int16_t fir32(fir32_state_t *fir, int16_t sample)
 294 {
 295     int i;
 296     int32_t y;
 297     int offset1;
 298     int offset2;
 299
 300     fir->history[fir->curr_pos] = sample;
 301     offset2 = fir->curr_pos;
 302     offset1 = fir->taps - offset2;
 303     y = 0;
 304     for (i = fir->taps - 1;  i >= offset1;  i--)
 305         y += fir->coeffs[i]*fir->history[i - offset1];
 306     for (  ;  i >= 0;  i--)
 307         y += fir->coeffs[i]*fir->history[i + offset2];
 308     if (fir->curr_pos <= 0)
 309         fir->curr_pos = fir->taps;
 310     fir->curr_pos--;
 311     return (int16_t) (y >> 15);
 312 }
 313 /*- End of function --------------------------------------------------------*/
 314
 315 #ifdef __cplusplus
 316 }
 317 #endif
 318
 319 #endif
 320 /*- End of file ------------------------------------------------------------*/