smpandaresampler.hh

// This Source Code Form is licensed MPL-2.0: http://mozilla.org/MPL/2.0
#ifndef __PANDA_RESAMPLER_HH__
#define __PANDA_RESAMPLER_HH__
 
#include <vector>
#include <memory>
#include <cstdio>
#include <cstdlib>
 
/* ------------------------ configuration ---------------------------- */
 
// uncomment this to use header only mode
// #define PANDA_RESAMPLER_HEADER_ONLY
 
/* ------------------------------------------------------------------- */
 
namespace PandaResampler {
 
typedef unsigned int uint;
 
static inline bool
check (bool value, const char *file, int line, const char *func, const char *what)
{
  if (!value)
    fprintf (stderr, "%s:%d:%s: PANDA_RESAMPLER_CHECK FAILED: %s\n", file, line, func, what);
  return value;
}
 
#define PANDA_RESAMPLER_CHECK(expr) (PandaResampler::check (expr, __FILE__, __LINE__, __func__, #expr))
 
template<class T>
class AlignedArray {
  unsigned char *unaligned_mem;
  T *data;
  size_t n_elements;
 
  void
  allocate_aligned_data()
  {
    /* for SSE we need 16-byte alignment, but we also ensure that no false
     * sharing will occur (at begin and end of data)
     */
    const size_t cache_line_size = 64;
 
    unaligned_mem = (unsigned char *) malloc (n_elements * sizeof (T) + 2 * (cache_line_size - 1));
    unsigned char *aligned_mem = unaligned_mem;
    if ((ptrdiff_t) aligned_mem % cache_line_size)
      aligned_mem += cache_line_size - (ptrdiff_t) aligned_mem % cache_line_size;
 
    data = reinterpret_cast<T *> (aligned_mem);
  }
public:
  AlignedArray (size_t n_elements) :
    n_elements (n_elements)
  {
    allocate_aligned_data();
    for (size_t i = 0; i < n_elements; i++)
      new (data + i) T();
  }
  AlignedArray (const std::vector<T>& elements) :
    AlignedArray (elements.size())
  {
    std::copy (elements.begin(), elements.end(), data);
  }
  ~AlignedArray()
  {
    /* C++ destruction order: last allocated element is deleted first */
    while (n_elements)
      data[--n_elements].~T();
    free (unaligned_mem);
  }
  T&
  operator[] (size_t pos)
  {
    return data[pos];
  }
  const T&
  operator[] (size_t pos) const
  {
    return data[pos];
  }
  size_t size () const
  {
    return n_elements;
  }
  T* begin()
  {
    return &data[0];
  }
  T* end()
  {
    return &data[n_elements];
  }
};
 
class Resampler2 {
  class Impl
  {
  public:
    virtual void   process_block (const float *input, uint n_input_samples, float *output) = 0;
    virtual uint   order() const = 0;
    virtual double delay() const = 0;
    virtual void   reset() = 0;
    virtual bool   sse_enabled() const = 0;
    virtual
    ~Impl()
    {
    }
  };
  std::unique_ptr<Impl> impl_x2;
  std::unique_ptr<Impl> impl_x4;
  std::unique_ptr<Impl> impl_x8;
  uint                  ratio_;
 
  template<uint ORDER, bool USE_SSE>
  class Upsampler2;
  template<uint ORDER, bool USE_SSE>
  class Downsampler2;
  template<uint ORDER>
  class IIRUpsampler2;
  template<uint ORDER>
  class IIRDownsampler2;
  template<uint ORDER>
  class IIRUpsampler2SSE;
  template<uint ORDER>
  class IIRDownsampler2SSE;
public:
  enum Mode {
    UP,
    DOWN
  };
  enum Precision {
    PREC_LINEAR = 1,     /* linear interpolation */
    PREC_48DB = 8,
    PREC_72DB = 12,
    PREC_96DB = 16,
    PREC_120DB = 20,
    PREC_144DB = 24
  };
  enum Filter {
    FILTER_IIR,
    FILTER_FIR,
  };
protected:
  Mode      mode_;
  Precision precision_;
  bool      use_sse_if_available_;
  Filter    filter_;
public:
  Resampler2 (Mode      mode,
              uint      ratio,
              Precision precision,
              bool      use_sse_if_available = true,
              Filter    filter = FILTER_FIR);
  static bool        sse_available();
  static bool        test_filter_impl (bool verbose);
  static Precision   find_precision_for_bits (uint bits);
  static const char  *precision_name (Precision precision);
  void
  process_block (const float *input, uint n_input_samples, float *output)
  {
    if (ratio_ == 2)
      {
        impl_x2->process_block (input, n_input_samples, output);
      }
    else if (ratio_ == 1)
      {
        std::copy (input, input + n_input_samples, output);
      }
    else
      {
        while (n_input_samples)
          {
            const uint block_size = 1024;
            const uint n_todo_samples = std::min (block_size, n_input_samples);
 
            float tmp[block_size * 4];
            float tmp2[block_size * 4];
 
            if (mode_ == UP)
              {
                if (ratio_ == 4)
                  {
                    impl_x2->process_block (input, n_todo_samples, tmp);
                    impl_x4->process_block (tmp, n_todo_samples * 2, output);
                  }
                else /* ratio_ == 8 */
                  {
                    impl_x2->process_block (input, n_todo_samples, tmp);
                    impl_x4->process_block (tmp, n_todo_samples * 2, tmp2);
                    impl_x8->process_block (tmp2, n_todo_samples * 4, output);
                  }
                output += n_todo_samples * ratio_;
              }
            else /* (mode_ == DOWN) */
              {
                if (ratio_ == 4)
                  {
                    impl_x4->process_block (input, n_todo_samples, tmp);
                    impl_x2->process_block (tmp, n_todo_samples / 2, output);
                  }
                else /* ratio_ == 8 */
                  {
                    impl_x8->process_block (input, n_todo_samples, tmp);
                    impl_x4->process_block (tmp, n_todo_samples / 2, tmp2);
                    impl_x2->process_block (tmp2, n_todo_samples / 4, output);
                  }
                output += n_todo_samples / ratio_;
              }
            input += n_todo_samples;
            n_input_samples -= n_todo_samples;
          }
      }
  }
  uint
  order() const
  {
    return impl_x2->order(); // FIXME
  }
  double
  delay() const
  {
    double d = 0;
    if (mode_ == UP)
      {
        if (ratio_ >= 2)
          d += impl_x2->delay();
        if (ratio_ >= 4)
          d += d + impl_x4->delay();
        if (ratio_ >= 8)
          d += d + impl_x8->delay();
      }
    else /* mode_ == DOWN */
      {
        if (ratio_ >= 2)
          d += impl_x2->delay();
        if (ratio_ >= 4)
          d += impl_x4->delay() / 2;
        if (ratio_ >= 8)
          d += impl_x8->delay() / 4;
      }
    return d;
  }
  void
  reset()
  {
    if (ratio_ >= 2)
      impl_x2->reset();
    if (ratio_ >= 4)
      impl_x4->reset();
    if (ratio_ >= 8)
      impl_x8->reset();
  }
  bool
  sse_enabled() const
  {
    return impl_x2->sse_enabled();
  }
protected:
  /* Creates implementation from filter coefficients and Filter implementation class
   *
   * Since up- and downsamplers use different (scaled) coefficients, its possible
   * to specify a scaling factor. Usually 2 for upsampling and 1 for downsampling.
   */
  template<class Filter> static inline Impl*
  create_impl_with_coeffs (const double *d,
                           uint          order,
                           double        scaling)
  {
    float taps[order];
    for (uint i = 0; i < order; i++)
      taps[i] = d[i] * scaling;
 
    Resampler2::Impl *filter = new Filter (taps);
    if (!PANDA_RESAMPLER_CHECK (order == filter->order()))
      return nullptr;
 
    return filter;
  }
  /* creates the actual implementation; specifying USE_SSE=true will use
   * SSE instructions, USE_SSE=false will use FPU instructions
   *
   * Don't use this directly - it's only to be used by
   * bseblockutils.cc's anonymous Impl classes.
   */
  template<bool USE_SSE> inline Impl*
  create_impl (uint stage_ratio);
 
  template<bool USE_SSE> inline Impl*
  create_impl_iir (uint stage_ratio);
 
  template<class CArray>
  inline Impl*
  create_impl_iir_with_coeffs (const CArray& carray, double group_delay);
 
  void
  init_stage (std::unique_ptr<Impl>& impl,
              uint                   stage_ratio);
};
 
} /* namespace PandaResampler */
 
// Make sure implementation is included in header-only mode
#if defined(PANDA_RESAMPLER_HEADER_ONLY) && !defined(PANDA_RESAMPLER_SOURCE)
#       define PANDA_RESAMPLER_SOURCE "pandaresampler.cc"
#       include PANDA_RESAMPLER_SOURCE
#endif
 
#endif /* __PANDA_RESAMPLER_HH__ */