src.c

au1200 linux2.6.11 硬件解码mae驱动和maiplayer播放器源码

/* <LIC_AMD_STD>
 * Copyright (C) 2003-2005 Advanced Micro Devices, Inc.  All Rights Reserved.
 * 
 * Unless otherwise designated in writing, this software and any related 
 * documentation are the confidential proprietary information of AMD. 
 * THESE MATERIALS ARE PROVIDED "AS IS" WITHOUT ANY
 * UNLESS OTHERWISE NOTED IN WRITING, EXPRESS OR IMPLIED WARRANTY OF ANY 
 * KIND, INCLUDING BUT NOT LIMITED TO WARRANTIES OF MERCHANTABILITY, 
 * NONINFRINGEMENT, TITLE, FITNESS FOR ANY PARTICULAR PURPOSE AND IN NO 
 * EVENT SHALL AMD OR ITS LICENSORS BE LIABLE FOR ANY DAMAGES WHATSOEVER. 
 * 
 * AMD does not assume any responsibility for any errors which may appear 
 * in the Materials nor any responsibility to support or update the
 * Materials.  AMD retains the right to modify the Materials at any time, 
 * without notice, and is not obligated to provide such modified 
 * Materials to you. AMD is not obligated to furnish, support, or make
 * any further information available to you.
 * </LIC_AMD_STD>  */
/* <CTL_AMD_STD>
 * </CTL_AMD_STD>  */
/* <DOC_AMD_STD>
 * Filename: src.c - sampling rate converter
 * </DOC_AMD_STD>  */

#ifdef POSTPROC_SRC

#include <stdio.h>
#include <string.h>
#include "mai_types.h"
#include "../audio_out.h" 
#include "src.h" 

#if defined(WIN32) || defined(UNDER_CE)
  #include <stdlib.h>
  #include <io.h>  
#else /* Linux */
  /* needed under Linux */
  #include <unistd.h>
  #include <sys/soundcard.h>
  #define DEV_AUDIO "/dev/dsp"
  static char audio_device[32] = DEV_AUDIO;
#endif

// #define DAC_MID_RATES_SUPPORTED  /* The I2S DAC supports these rates */
#define DAC_MID_RATES_MIGHT_BE_SUPPORTED  /* The I2S DAC supports these rates */
// #define DAC_HIGH_RATES_SUPPORTED /* A better DAC will also support these rates */
#define USE_CUBIC_SPLINE_FILTER

#define DPRINTF(_args_)  /* printf _args_ */
#define INFO_PRINTF(_args_) /* printf _args_ */


int sampling_rate_conversion_factor;
short *conversion_buffer;

void close_src(void)
{
  if (sampling_rate_conversion_factor)
    free (conversion_buffer);
  conversion_buffer = NULL;
}
/* The assigned values are important, and are used in the calculation of the buffer length. */
enum
{
  CSR_NONE = 0,
  CSR_OCTUPLE = 64,
  CSR_QUADRUPLE = 32,
  CSR_DOUBLE = 16,
  CSR_HALVE = 8,
  CSR_QUARTER = 4,
};

#ifdef USE_CUBIC_SPLINE_FILTER
#define SPLINE_FIR_LENGTH_MAX 4096
typedef struct filter_fir_struct
{ 
  int elements;
  int output[4];
  int *coef;
  int *writeptr;
  int *readptr;
  int *delay_line_end;
  int delay_line[SPLINE_FIR_LENGTH_MAX];
  int delay_line_length;
  int *save;
} FILTER_FIR, *PFILTER_FIR;
static FILTER_FIR fir[MAX_AUDIO_CHANNELS];
#define CS_F_FAC 0x4000
int cubic_spline_4x0FIR_coef[4] =
{
   (int)(-0.0234 * CS_F_FAC),  
   (int)( 0.2266 * CS_F_FAC),  
   (int)( 0.8672 * CS_F_FAC), 
   (int)(-0.0703 * CS_F_FAC),
};
int /* cs */ cubic_spline_2xFIR_coef[4] = // same as 4x1
{
  (int)(-0.0625 * CS_F_FAC), /* 1 step in the future */
  (int)( 0.5625 * CS_F_FAC), /* time now */
  (int)( 0.5625 * CS_F_FAC), /* 1 step in the past */
  (int)(-0.0625 * CS_F_FAC), /* 2 steps in the past */
};
int cubic_spline_4x2FIR_coef[4] =
{
   (int)(-0.0703 * CS_F_FAC),       
   (int)( 0.8672 * CS_F_FAC),       
   (int)( 0.2266 * CS_F_FAC),      
   (int)(-0.0234 * CS_F_FAC),    
};

/*  From Matlab script
  x = 0:20;
  y = [0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 ] ; 
  xx = 0:.25:20;
  yi = interp1(x,y,xx,'cubic')
   -0.0234   -0.0625   -0.0703         0    
    0.2266    0.5625    0.8672    1.0000    
    0.8672    0.5625    0.2266         0   
   -0.0703   -0.0625   -0.0234         0
*/


/*
  The cubic-spline filter does a reasonable job with four taps.
  It is not as good as an 80-tap fir, but it is a lot less expensive.
*/
static void upsample_cubic_spline2x(short *input, short *output, int how_many, int channels)  // assumes interleaved streams
{
  int cc, ii, temp;
  int loop = how_many/channels;
  int *coef;
  int acc;

  if ((input == NULL) || (output == NULL) )
    return;

  for (cc = 0; cc < channels; cc++)     
  {
    fir[cc].elements = 4;
    fir[cc].delay_line_end = fir[cc].delay_line + fir[cc].delay_line_length;
  }

  for (ii = 0; ii < loop; ii++)      // de-interleave and copy to delay lines
  {
    for (cc = 0; cc < channels; cc++)     
    {
      *fir[cc].writeptr++ = *input++;                  // left is first
      if (fir[cc].writeptr >= fir[cc].delay_line_end)  // stay within circular buffer
        fir[cc].writeptr -= fir[cc].delay_line_length;
    }
  }
  while (loop--)
  {
    for (cc = 0; cc < channels; cc++)     
    {
      coef = cubic_spline_2xFIR_coef;
      acc = 0x2000;                    // rounding
      fir[cc].save = fir[cc].readptr;  // Note that this is the output from the input buffer--NOT the output buffer!

      for (ii = 0; ii < fir[cc].elements; ii++) /* do FIR convolution */
      {
        acc += (*fir[cc].readptr--) * (*coef++);
        if (fir[cc].readptr < fir[cc].delay_line)
          fir[cc].readptr += fir[cc].delay_line_length;
      }
      fir[cc].readptr = fir[cc].save;
      temp = (int)(acc >> 14);
      *output++ = saturate16(temp);           // write to output buffer
    }
    for (cc = 0; cc < channels; cc++)     
    {
      fir[cc].readptr--;  /* time now is actually 1 step in the past, for a group delay of 1 sample */
      if (fir[cc].readptr < fir[cc].delay_line)
        fir[cc].readptr += fir[cc].delay_line_length;
      // Now output the next unmodified sample for a 2x upsample
      *output++ = (short)*fir[cc].readptr;
      fir[cc].readptr += 2;
      if (fir[cc].readptr >= fir[cc].delay_line_end)
        fir[cc].readptr -= fir[cc].delay_line_length;
    }
  }
}
static void upsample_cubic_spline4x(short *input, short *output, int how_many, int channels)  // assumes interleaved streams
{
  int cc, ii, temp;
  int loop = how_many/channels;
  int *coef;
  int acc;

  if ((input == NULL) || (output == NULL) )
    return;

  for (cc = 0; cc < channels; cc++)     
  {
    fir[cc].elements = 4;
    fir[cc].delay_line_end = fir[cc].delay_line + fir[cc].delay_line_length;
  }

  for (ii = 0; ii < loop; ii++)      // de-interleave and copy to delay lines
  {
    for (cc = 0; cc < channels; cc++)     
    {
      *fir[cc].writeptr++ = *input++;                  // left is first
      if (fir[cc].writeptr >= fir[cc].delay_line_end)  // stay within circular buffer
        fir[cc].writeptr -= fir[cc].delay_line_length;
    }
  }
  while (loop--)
  {
    for (cc = 0; cc < channels; cc++)     
    {
      coef = cubic_spline_4x0FIR_coef;
      acc = 0x2000;                    // rounding
      fir[cc].save = fir[cc].readptr;  // Note that this is from the input buffer--NOT the output buffer!
      for (ii = 0; ii < fir[cc].elements; ii++) /* do FIR convolution using 4 values */
      {
        acc += (*fir[cc].readptr--) * (*coef++);
        if (fir[cc].readptr < fir[cc].delay_line)
          fir[cc].readptr += fir[cc].delay_line_length;
      }
      fir[cc].readptr = fir[cc].save;
      temp = (int)(acc >> 14);
      fir[cc].output[0] = saturate16(temp); 

      coef = cubic_spline_2xFIR_coef;  // same as 4x1FIR_coef would be
      acc = 0x2000;                    // rounding
      for (ii = 0; ii < fir[cc].elements; ii++) /* do FIR convolution using 4 values */
      {
        acc += (*fir[cc].readptr--) * (*coef++);
        if (fir[cc].readptr < fir[cc].delay_line)
          fir[cc].readptr += fir[cc].delay_line_length;
      }
      fir[cc].readptr = fir[cc].save;
      temp = (int)(acc >> 14);
      fir[cc].output[1] = saturate16(temp); 

      coef = cubic_spline_4x2FIR_coef; 
      acc = 0x2000;                    // rounding
      for (ii = 0; ii < fir[cc].elements; ii++) /* do FIR convolution using 4 values */
      {
        acc += (*fir[cc].readptr--) * (*coef++);
        if (fir[cc].readptr < fir[cc].delay_line)
          fir[cc].readptr += fir[cc].delay_line_length;
      }
      temp = (int)(acc >> 14);
      fir[cc].output[2] = saturate16(temp); 

      fir[cc].readptr = fir[cc].save-1;      /* time now is actually 1 step in the past, for a group delay of 1 sample */
      if (fir[cc].readptr < fir[cc].delay_line)
        fir[cc].readptr += fir[cc].delay_line_length;
      // Now use the next unmodified sample for 4x3 upsample
      fir[cc].output[3] = *fir[cc].readptr;
      fir[cc].readptr += 2; /* move to the next input sample */
      if (fir[cc].readptr >= fir[cc].delay_line_end)
        fir[cc].readptr -= fir[cc].delay_line_length;
    }
    /* interleave results */
    for (ii = 0; ii < fir[0].elements; ii++)     
      for (cc = 0; cc < channels; cc++)     
        *output++ = fir[cc].output[ii];
  }
}
#endif

/* This is, for now, the classic add/drop-sample method without filtering, except where cubic spline is used. */
static int csr_leftover;
static short csr_leftoverbuf[32];
static short *csr_buf;
static short *csr_buf_base;
short *convert_sampling_rate(short *inbuf, int *length /* in bytes */) // assumes 16 bits per sample 
{
  int ii, jj;
  int loop;
  short *buf = inbuf;

  if ((inbuf == NULL) || (*length == 0))
    return NULL;

  /* only do this if we are downsampling */
  if ((sampling_rate_conversion_factor == CSR_HALVE) || (sampling_rate_conversion_factor == CSR_QUARTER))
  {
    if (csr_buf_base == NULL)
      csr_buf_base = (short *)malloc(16384);
    if (csr_buf_base != NULL)
    {
      if (csr_leftover)
      {
        csr_buf = csr_buf_base;
        memcpy(csr_buf, csr_leftoverbuf, csr_leftover);
        memcpy(((unsigned char*)csr_buf) + csr_leftover, buf, *length);
        *length += csr_leftover;
        csr_leftover = 0;
        buf = csr_buf;
      }
      if (*length & 0xf)  // odd lengths are a problem when downsampling
      {
        int slength = *length & 0xfffff0; 
        csr_leftover = *length - slength;
        memcpy(csr_leftoverbuf, ((unsigned char*)buf) + slength, csr_leftover);
        *length = slength;
      }
    }
  }
  else if (csr_buf_base != NULL)
  {
    free(csr_buf_base);
    csr_buf_base = NULL;
  }
  loop = *length / sizeof(short);
  if (number_of_channels == 1)
  {
    switch (sampling_rate_conversion_factor)
    {
      case CSR_OCTUPLE: // FIXIT use a cubic spline for this one too
        for (jj = 0, ii = 0; ii < loop; ii++, jj += 8)
        {
          conversion_buffer[jj] = buf[ii];
          conversion_buffer[jj+1] = buf[ii];
          conversion_buffer[jj+2] = buf[ii];
          conversion_buffer[jj+3] = buf[ii];
          conversion_buffer[jj+4] = buf[ii];
          conversion_buffer[jj+5] = buf[ii];
          conversion_buffer[jj+6] = buf[ii];
          conversion_buffer[jj+7] = buf[ii];
        }