zdoom/gzdoom
0
0
Fork 0
mirror of https://github.com/ZDoom/gzdoom.git synced 2024-11-14 16:40:56 +00:00
Code Issues Projects Releases Packages Wiki Activity
gzdoom/dumb/src/helpers/resampler.c
Randy Heit 55142078d8 Normalize line endings
2016-03-01 09:47:10 -06:00

1512 lines
44 KiB
C
Raw Blame History

#include <stdlib.h>
#include <string.h>
#define _USE_MATH_DEFINES
#include <math.h>
#if (defined(_M_IX86) || defined(__i386__) || defined(_M_X64) || defined(__amd64__))
#include <xmmintrin.h>
#define RESAMPLER_SSE
#endif
#ifdef __APPLE__
#include <TargetConditionals.h>
#if TARGET_CPU_ARM || TARGET_CPU_ARM64
#include <arm_neon.h>
#define RESAMPLER_NEON
#endif
#endif
#ifdef _MSC_VER
#define ALIGNED _declspec(align(16))
#else
#define ALIGNED __attribute__((aligned(16)))
#endif
#ifndef M_PI
#define M_PI 3.14159265358979323846
#endif
#include "internal/resampler.h"
enum { RESAMPLER_SHIFT = 10 };
enum { RESAMPLER_SHIFT_EXTRA = 8 };
enum { RESAMPLER_RESOLUTION = 1 << RESAMPLER_SHIFT };
enum { RESAMPLER_RESOLUTION_EXTRA = 1 << (RESAMPLER_SHIFT + RESAMPLER_SHIFT_EXTRA) };
enum { SINC_WIDTH = 16 };
enum { SINC_SAMPLES = RESAMPLER_RESOLUTION * SINC_WIDTH };
enum { CUBIC_SAMPLES = RESAMPLER_RESOLUTION * 4 };
typedef union bigint
{
unsigned long long quad;
#ifndef __BIG_ENDIAN__
struct { unsigned int lo, hi; };
#else
struct { unsigned int hi, lo; };
#endif
} bigint;
// What works well on 32-bit can make for extra work on 64-bit
#if defined(_M_X64) || defined(__amd64__) || TARGET_CPU_ARM64
#define CLEAR_HI(p) (p.quad &= 0xffffffffu)
#define ADD_HI(a,p) (a += p.quad >> 32)
#define PHASE_REDUCE(p) (int)(p.quad >> (32 - RESAMPLER_SHIFT))
#else
#define CLEAR_HI(p) p.hi = 0
#define ADD_HI(a,p) a += p.hi
// Should be equivalent to (int)(p.quad >> (32 - RESAMPLER_SHIFT)),
// since the high part should get zeroed after every sample.
#define PHASE_REDUCE(p) (p.lo >> (32 - RESAMPLER_SHIFT))
#endif
static const float RESAMPLER_BLEP_CUTOFF = 0.90f;
static const float RESAMPLER_BLAM_CUTOFF = 0.93f;
static const float RESAMPLER_SINC_CUTOFF = 0.999f;
ALIGNED static float cubic_lut[CUBIC_SAMPLES];
static float sinc_lut[SINC_SAMPLES + 1];
static float window_lut[SINC_SAMPLES + 1];
enum { resampler_buffer_size = SINC_WIDTH * 4 };
static int fEqual(const double b, const double a)
{
return fabs(a - b) < 1.0e-6;
}
static double sinc(double x)
{
return fEqual(x, 0.0) ? 1.0 : sin(x * M_PI) / (x * M_PI);
}
#ifdef RESAMPLER_SSE
#ifdef _MSC_VER
#include <intrin.h>
#elif defined(__clang__) || defined(__GNUC__)
static inline void
__cpuid(int *data, int selector)
{
#if defined(__PIC__) && defined(__i386__)
asm("xchgl %%ebx, %%esi; cpuid; xchgl %%ebx, %%esi"
: "=a" (data[0]),
"=S" (data[1]),
"=c" (data[2]),
"=d" (data[3])
: "0" (selector));
#elif defined(__PIC__) && defined(__amd64__)
asm("xchg{q} {%%}rbx, %q1; cpuid; xchg{q} {%%}rbx, %q1"
: "=a" (data[0]),
"=&r" (data[1]),
"=c" (data[2]),
"=d" (data[3])
: "0" (selector));
#else
asm("cpuid"
: "=a" (data[0]),
"=b" (data[1]),
"=c" (data[2]),
"=d" (data[3])
: "0" (selector));
#endif
}
#else
#define __cpuid(a,b) memset((a), 0, sizeof(int) * 4)
#endif
static int query_cpu_feature_sse() {
int buffer[4];
__cpuid(buffer,1);
if ((buffer[3]&(1<<25)) == 0) return 0;
return 1;
}
static int resampler_has_sse = 0;
#endif
void resampler_init(void)
{
unsigned i;
double dx = (float)(SINC_WIDTH) / SINC_SAMPLES, x = 0.0;
for (i = 0; i < SINC_SAMPLES + 1; ++i, x += dx)
{
double y = x / SINC_WIDTH;
#if 0
// Blackman
float window = 0.42659 - 0.49656 * cos(M_PI + M_PI * y) + 0.076849 * cos(2.0 * M_PI * y);
#elif 1
// Nuttal 3 term
double window = 0.40897 + 0.5 * cos(M_PI * y) + 0.09103 * cos(2.0 * M_PI * y);
#elif 0
// C.R.Helmrich's 2 term window
float window = 0.79445 * cos(0.5 * M_PI * y) + 0.20555 * cos(1.5 * M_PI * y);
#elif 0
// Lanczos
float window = sinc(y);
#endif
sinc_lut[i] = (float)(fabs(x) < SINC_WIDTH ? sinc(x) : 0.0);
window_lut[i] = (float)window;
}
dx = 1.0 / RESAMPLER_RESOLUTION;
x = 0.0;
for (i = 0; i < RESAMPLER_RESOLUTION; ++i, x += dx)
{
cubic_lut[i*4] = (float)(-0.5 * x * x * x + x * x - 0.5 * x);
cubic_lut[i*4+1] = (float)( 1.5 * x * x * x - 2.5 * x * x + 1.0);
cubic_lut[i*4+2] = (float)(-1.5 * x * x * x + 2.0 * x * x + 0.5 * x);
cubic_lut[i*4+3] = (float)( 0.5 * x * x * x - 0.5 * x * x);
}
#ifdef RESAMPLER_SSE
resampler_has_sse = query_cpu_feature_sse();
#endif
}
typedef struct resampler
{
int write_pos, write_filled;
int read_pos, read_filled;
bigint phase;
bigint phase_inc;
bigint inv_phase;
bigint inv_phase_inc;
unsigned char quality;
signed char delay_added;
signed char delay_removed;
double last_amp;
double accumulator;
float buffer_in[resampler_buffer_size * 2];
float buffer_out[resampler_buffer_size + SINC_WIDTH * 2 - 1];
} resampler;
void * resampler_create(void)
{
resampler * r = ( resampler * ) malloc( sizeof(resampler) );
if ( !r ) return 0;
r->write_pos = SINC_WIDTH - 1;
r->write_filled = 0;
r->read_pos = 0;
r->read_filled = 0;
r->phase.quad = 0;
r->phase_inc.quad = 0;
r->inv_phase.quad = 0;
r->inv_phase_inc.quad = 0;
r->quality = RESAMPLER_QUALITY_MAX;
r->delay_added = -1;
r->delay_removed = -1;
r->last_amp = 0;
r->accumulator = 0;
memset( r->buffer_in, 0, sizeof(r->buffer_in) );
memset( r->buffer_out, 0, sizeof(r->buffer_out) );
return r;
}
void resampler_delete(void * _r)
{
free( _r );
}
void * resampler_dup(const void * _r)
{
void * r_out = malloc( sizeof(resampler) );
if ( !r_out ) return 0;
resampler_dup_inplace(r_out, _r);
return r_out;
}
void resampler_dup_inplace(void *_d, const void *_s)
{
const resampler * r_in = ( const resampler * ) _s;
resampler * r_out = ( resampler * ) _d;
r_out->write_pos = r_in->write_pos;
r_out->write_filled = r_in->write_filled;
r_out->read_pos = r_in->read_pos;
r_out->read_filled = r_in->read_filled;
r_out->phase = r_in->phase;
r_out->phase_inc = r_in->phase_inc;
r_out->inv_phase = r_in->inv_phase;
r_out->inv_phase_inc = r_in->inv_phase_inc;
r_out->quality = r_in->quality;
r_out->delay_added = r_in->delay_added;
r_out->delay_removed = r_in->delay_removed;
r_out->last_amp = r_in->last_amp;
r_out->accumulator = r_in->accumulator;
memcpy( r_out->buffer_in, r_in->buffer_in, sizeof(r_in->buffer_in) );
memcpy( r_out->buffer_out, r_in->buffer_out, sizeof(r_in->buffer_out) );
}
void resampler_set_quality(void *_r, int quality)
{
resampler * r = ( resampler * ) _r;
if (quality < RESAMPLER_QUALITY_MIN)
quality = RESAMPLER_QUALITY_MIN;
else if (quality > RESAMPLER_QUALITY_MAX)
quality = RESAMPLER_QUALITY_MAX;
if ( r->quality != quality )
{
if ( quality == RESAMPLER_QUALITY_BLEP || r->quality == RESAMPLER_QUALITY_BLEP ||
quality == RESAMPLER_QUALITY_BLAM || r->quality == RESAMPLER_QUALITY_BLAM )
{
r->read_pos = 0;
r->read_filled = 0;
r->last_amp = 0;
r->accumulator = 0;
memset( r->buffer_out, 0, sizeof(r->buffer_out) );
}
r->delay_added = -1;
r->delay_removed = -1;
}
r->quality = (unsigned char)quality;
}
int resampler_get_free_count(void *_r)
{
resampler * r = ( resampler * ) _r;
return resampler_buffer_size - r->write_filled;
}
static int resampler_min_filled(resampler *r)
{
switch (r->quality)
{
default:
case RESAMPLER_QUALITY_ZOH:
case RESAMPLER_QUALITY_BLEP:
return 1;
case RESAMPLER_QUALITY_LINEAR:
case RESAMPLER_QUALITY_BLAM:
return 2;
case RESAMPLER_QUALITY_CUBIC:
return 4;
case RESAMPLER_QUALITY_SINC:
return SINC_WIDTH * 2;
}
}
static int resampler_input_delay(resampler *r)
{
switch (r->quality)
{
default:
case RESAMPLER_QUALITY_ZOH:
case RESAMPLER_QUALITY_BLEP:
case RESAMPLER_QUALITY_LINEAR:
case RESAMPLER_QUALITY_BLAM:
return 0;
case RESAMPLER_QUALITY_CUBIC:
return 1;
case RESAMPLER_QUALITY_SINC:
return SINC_WIDTH - 1;
}
}
static int resampler_output_delay(resampler *r)
{
switch (r->quality)
{
default:
case RESAMPLER_QUALITY_ZOH:
case RESAMPLER_QUALITY_LINEAR:
case RESAMPLER_QUALITY_CUBIC:
case RESAMPLER_QUALITY_SINC:
return 0;
case RESAMPLER_QUALITY_BLEP:
case RESAMPLER_QUALITY_BLAM:
return SINC_WIDTH - 1;
}
}
int resampler_ready(void *_r)
{
resampler * r = ( resampler * ) _r;
return r->write_filled > resampler_min_filled(r);
}
void resampler_clear(void *_r)
{
resampler * r = ( resampler * ) _r;
r->write_pos = SINC_WIDTH - 1;
r->write_filled = 0;
r->read_pos = 0;
r->read_filled = 0;
r->phase.quad = 0;
r->delay_added = -1;
r->delay_removed = -1;
memset(r->buffer_in, 0, (SINC_WIDTH - 1) * sizeof(r->buffer_in[0]));
memset(r->buffer_in + resampler_buffer_size, 0, (SINC_WIDTH - 1) * sizeof(r->buffer_in[0]));
if (r->quality == RESAMPLER_QUALITY_BLEP || r->quality == RESAMPLER_QUALITY_BLAM)
{
r->inv_phase.quad = 0;
r->last_amp = 0;
r->accumulator = 0;
memset(r->buffer_out, 0, sizeof(r->buffer_out));
}
}
void resampler_set_rate(void *_r, double new_factor)
{
resampler * r = ( resampler * ) _r;
r->phase_inc.quad = (long long)(new_factor * 0x100000000ll);
new_factor = 1.0 / new_factor;
r->inv_phase_inc.quad = (long long)(new_factor * 0x100000000ll);
}
void resampler_write_sample(void *_r, short s)
{
resampler * r = ( resampler * ) _r;
if ( r->delay_added < 0 )
{
r->delay_added = 0;
r->write_filled = resampler_input_delay( r );
}
if ( r->write_filled < resampler_buffer_size )
{
float s32 = s;
s32 *= 256.0;
r->buffer_in[ r->write_pos ] = s32;
r->buffer_in[ r->write_pos + resampler_buffer_size ] = s32;
++r->write_filled;
r->write_pos = ( r->write_pos + 1 ) % resampler_buffer_size;
}
}
void resampler_write_sample_fixed(void *_r, int s, unsigned char depth)
{
resampler * r = ( resampler * ) _r;
if ( r->delay_added < 0 )
{
r->delay_added = 0;
r->write_filled = resampler_input_delay( r );
}
if ( r->write_filled < resampler_buffer_size )
{
double s32 = s;
s32 /= (double)(1 << (depth - 1));
r->buffer_in[ r->write_pos ] = (float)s32;
r->buffer_in[ r->write_pos + resampler_buffer_size ] = (float)s32;
++r->write_filled;
r->write_pos = ( r->write_pos + 1 ) % resampler_buffer_size;
}
}
static int resampler_run_zoh(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 1;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
do
{
if ( out >= out_end )
break;
*out++ = *in;
phase.quad += phase_inc.quad;
ADD_HI(in, phase);
CLEAR_HI(phase);
}
while ( in < in_end );
r->phase = phase;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#ifndef RESAMPLER_NEON
static int resampler_run_blep(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 1;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
double last_amp = r->last_amp;
bigint inv_phase = r->inv_phase;
bigint inv_phase_inc = r->inv_phase_inc;
const int step = (int)(RESAMPLER_BLEP_CUTOFF * RESAMPLER_RESOLUTION);
const int window_step = RESAMPLER_RESOLUTION;
do
{
double sample;
if ( out + SINC_WIDTH * 2 > out_end )
break;
sample = *in++ - last_amp;
if (sample)
{
double kernel[SINC_WIDTH * 2], kernel_sum = 0.0f;
int phase_reduced = PHASE_REDUCE(inv_phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
int i = SINC_WIDTH;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernel[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
last_amp += sample;
sample /= kernel_sum;
for (i = 0; i < SINC_WIDTH * 2; ++i)
out[i] += (float)(sample * kernel[i]);
}
inv_phase.quad += inv_phase_inc.quad;
ADD_HI(out, inv_phase);
CLEAR_HI(inv_phase);
}
while ( in < in_end );
r->inv_phase = inv_phase;
r->last_amp = last_amp;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifdef RESAMPLER_SSE
static int resampler_run_blep_sse(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 1;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
double last_amp = r->last_amp;
bigint inv_phase = r->inv_phase;
bigint inv_phase_inc = r->inv_phase_inc;
const int step = (int)(RESAMPLER_BLEP_CUTOFF * RESAMPLER_RESOLUTION);
const int window_step = RESAMPLER_RESOLUTION;
do
{
double sample;
if ( out + SINC_WIDTH * 2 > out_end )
break;
sample = *in++ - last_amp;
if (sample)
{
float kernel_sum = 0.0f;
__m128 kernel[SINC_WIDTH / 2];
__m128 temp1, temp2;
__m128 samplex;
float *kernelf = (float*)(&kernel);
int phase_reduced = PHASE_REDUCE(inv_phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
int i = SINC_WIDTH;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernelf[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
last_amp += sample;
sample /= kernel_sum;
samplex = _mm_set1_ps( (float)sample );
for (i = 0; i < SINC_WIDTH / 2; ++i)
{
temp1 = _mm_load_ps( (const float *)( kernel + i ) );
temp1 = _mm_mul_ps( temp1, samplex );
temp2 = _mm_loadu_ps( (const float *) out + i * 4 );
temp1 = _mm_add_ps( temp1, temp2 );
_mm_storeu_ps( (float *) out + i * 4, temp1 );
}
}
inv_phase.quad += inv_phase_inc.quad;
ADD_HI(out, inv_phase);
CLEAR_HI(inv_phase);
}
while ( in < in_end );
r->inv_phase = inv_phase;
r->last_amp = last_amp;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifdef RESAMPLER_NEON
static int resampler_run_blep(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 1;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
float last_amp = r->last_amp;
bigint inv_phase = r->inv_phase;
bigint inv_phase_inc = r->inv_phase_inc;
const int step = RESAMPLER_BLEP_CUTOFF * RESAMPLER_RESOLUTION;
const int window_step = RESAMPLER_RESOLUTION;
do
{
float sample;
if ( out + SINC_WIDTH * 2 > out_end )
break;
sample = *in++ - last_amp;
if (sample)
{
float kernel_sum = 0.0f;
float32x4_t kernel[SINC_WIDTH / 2];
float32x4_t temp1, temp2;
float32x4_t samplex;
float *kernelf = (float*)(&kernel);
int phase_reduced = PHASE_REDUCE(inv_phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
int i = SINC_WIDTH;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernelf[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
last_amp += sample;
sample /= kernel_sum;
samplex = vdupq_n_f32(sample);
for (i = 0; i < SINC_WIDTH / 2; ++i)
{
temp1 = vld1q_f32( (const float32_t *)( kernel + i ) );
temp2 = vld1q_f32( (const float32_t *) out + i * 4 );
temp2 = vmlaq_f32( temp2, temp1, samplex );
vst1q_f32( (float32_t *) out + i * 4, temp2 );
}
}
inv_phase.quad += inv_phase_inc.quad;
ADD_HI(out, inv_phase);
CLEAR_HI(inv_phase);
}
while ( in < in_end );
r->inv_phase = inv_phase;
r->last_amp = last_amp;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
static int resampler_run_linear(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 2;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
do
{
if ( out >= out_end )
break;
*out++ = (float)(in[0] + (in[1] - in[0]) * phase.lo * (1.f / 0x100000000ll));
phase.quad += phase_inc.quad;
ADD_HI(in, phase);
CLEAR_HI(phase);
}
while ( in < in_end );
r->phase = phase;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#ifndef RESAMPLER_NEON
static int resampler_run_blam(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 2;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
double last_amp = r->last_amp;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
bigint inv_phase = r->inv_phase;
bigint inv_phase_inc = r->inv_phase_inc;
const int step = (int)(RESAMPLER_BLAM_CUTOFF * RESAMPLER_RESOLUTION);
const int window_step = RESAMPLER_RESOLUTION;
do
{
double sample;
if ( out + SINC_WIDTH * 2 > out_end )
break;
sample = in[0];
if (phase_inc.quad < 0x100000000ll)
sample += (in[1] - in[0]) * phase.quad * (1.f / 0x100000000ll);
sample -= last_amp;
if (sample)
{
double kernel[SINC_WIDTH * 2], kernel_sum = 0.0f;
int phase_reduced = PHASE_REDUCE(inv_phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
int i = SINC_WIDTH;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernel[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
last_amp += sample;
sample /= kernel_sum;
for (i = 0; i < SINC_WIDTH * 2; ++i)
out[i] += (float)(sample * kernel[i]);
}
if (inv_phase_inc.quad < 0x100000000ll)
{
++in;
inv_phase.quad += inv_phase_inc.quad;
ADD_HI(out, inv_phase);
CLEAR_HI(inv_phase);
}
else
{
phase.quad += phase_inc.quad;
++out;
ADD_HI(in, phase);
CLEAR_HI(phase);
}
}
while ( in < in_end );
r->phase = phase;
r->inv_phase = inv_phase;
r->last_amp = last_amp;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifdef RESAMPLER_SSE
static int resampler_run_blam_sse(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 2;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
double last_amp = r->last_amp;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
bigint inv_phase = r->inv_phase;
bigint inv_phase_inc = r->inv_phase_inc;
const int step = (int)(RESAMPLER_BLAM_CUTOFF * RESAMPLER_RESOLUTION);
const int window_step = RESAMPLER_RESOLUTION;
do
{
double sample;
if ( out + SINC_WIDTH * 2 > out_end )
break;
sample = in[0];
if (phase_inc.quad < 0x100000000ll)
{
sample += (in[1] - in[0]) * phase.quad * (1.f / 0x100000000ll);
}
sample -= last_amp;
if (sample)
{
float kernel_sum = 0.0f;
__m128 kernel[SINC_WIDTH / 2];
__m128 temp1, temp2;
__m128 samplex;
float *kernelf = (float*)(&kernel);
int phase_reduced = PHASE_REDUCE(inv_phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
int i = SINC_WIDTH;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernelf[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
last_amp += sample;
sample /= kernel_sum;
samplex = _mm_set1_ps( (float)sample );
for (i = 0; i < SINC_WIDTH / 2; ++i)
{
temp1 = _mm_load_ps( (const float *)( kernel + i ) );
temp1 = _mm_mul_ps( temp1, samplex );
temp2 = _mm_loadu_ps( (const float *) out + i * 4 );
temp1 = _mm_add_ps( temp1, temp2 );
_mm_storeu_ps( (float *) out + i * 4, temp1 );
}
}
if (inv_phase_inc.quad < 0x100000000ll)
{
++in;
inv_phase.quad += inv_phase_inc.quad;
ADD_HI(out, inv_phase);
CLEAR_HI(inv_phase);
}
else
{
phase.quad += phase_inc.quad;
++out;
if (phase.quad >= 0x100000000ll)
{
++in;
CLEAR_HI(phase);
}
}
}
while ( in < in_end );
r->phase = phase;
r->inv_phase = inv_phase;
r->last_amp = last_amp;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifdef RESAMPLER_NEON
static int resampler_run_blam(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 2;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
float last_amp = r->last_amp;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
bigint inv_phase = r->inv_phase;
bigint inv_phase_inc = r->inv_phase_inc;
const int step = RESAMPLER_BLAM_CUTOFF * RESAMPLER_RESOLUTION;
const int window_step = RESAMPLER_RESOLUTION;
do
{
float sample;
if ( out + SINC_WIDTH * 2 > out_end )
break;
sample = in[0];
if (phase_inc.quad < 0x100000000ll)
sample += (in[1] - in[0]) * phase;
sample -= last_amp;
if (sample)
{
float kernel_sum = 0.0;
float32x4_t kernel[SINC_WIDTH / 2];
float32x4_t temp1, temp2;
float32x4_t samplex;
float *kernelf = (float*)(&kernel);
int phase_reduced = PHASE_REDUCE(inv_phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
int i = SINC_WIDTH;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernelf[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
last_amp += sample;
sample /= kernel_sum;
samplex = vdupq_n_f32(sample);
for (i = 0; i < SINC_WIDTH / 2; ++i)
{
temp1 = vld1q_f32( (const float32_t *)( kernel + i ) );
temp2 = vld1q_f32( (const float32_t *) out + i * 4 );
temp2 = vmlaq_f32( temp2, temp1, samplex );
vst1q_f32( (float32_t *) out + i * 4, temp2 );
}
}
if (inv_phase_inc.quad < 0x100000000ll)
{
++in;
inv_phase.quad += inv_phase_inc.quad;
ADD_HI(out, inv_phase);
CLEAR_HI(inv_phase);
}
else
{
phase.quad += phase_inc.quad;
++out;
if (phase.quad >= 0x100000000ll)
{
++in;
CLEAR_HI(phase);
}
}
}
while ( in < in_end );
r->phase = phase;
r->inv_phase = inv_phase;
r->last_amp = last_amp;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifndef RESAMPLER_NEON
static int resampler_run_cubic(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 4;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
do
{
float * kernel;
int i;
float sample;
if ( out >= out_end )
break;
kernel = cubic_lut + PHASE_REDUCE(phase) * 4;
for (sample = 0, i = 0; i < 4; ++i)
sample += in[i] * kernel[i];
*out++ = sample;
phase.quad += phase_inc.quad;
ADD_HI(in, phase);
CLEAR_HI(phase);
}
while ( in < in_end );
r->phase = phase;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifdef RESAMPLER_SSE
static int resampler_run_cubic_sse(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 4;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
do
{
__m128 temp1, temp2;
__m128 samplex = _mm_setzero_ps();
if ( out >= out_end )
break;
temp1 = _mm_loadu_ps( (const float *)( in ) );
temp2 = _mm_load_ps( (const float *)( cubic_lut + PHASE_REDUCE(phase) * 4 ) );
temp1 = _mm_mul_ps( temp1, temp2 );
samplex = _mm_add_ps( samplex, temp1 );
temp1 = _mm_movehl_ps( temp1, samplex );
samplex = _mm_add_ps( samplex, temp1 );
temp1 = samplex;
temp1 = _mm_shuffle_ps( temp1, samplex, _MM_SHUFFLE(0, 0, 0, 1) );
samplex = _mm_add_ps( samplex, temp1 );
_mm_store_ss( out, samplex );
++out;
phase.quad += phase_inc.quad;
ADD_HI(in, phase);
CLEAR_HI(phase);
}
while ( in < in_end );
r->phase = phase;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifdef RESAMPLER_NEON
static int resampler_run_cubic(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= 4;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
do
{
float32x4_t temp1, temp2;
float32x2_t half;
if ( out >= out_end )
break;
temp1 = vld1q_f32( (const float32_t *)( in ) );
temp2 = vld1q_f32( (const float32_t *)( cubic_lut + PHASE_REDUCE(phase) * 4 ) );
temp1 = vmulq_f32( temp1, temp2 );
half = vadd_f32(vget_high_f32(temp1), vget_low_f32(temp1));
*out++ = vget_lane_f32(vpadd_f32(half, half), 0);
phase.quad += phase_inc.quad;
ADD_HI(in, phase)
CLEAR_HI(phase);
}
while ( in < in_end );
r->phase = phase;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifndef RESAMPLER_NEON
static int resampler_run_sinc(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= SINC_WIDTH * 2;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
int step = phase_inc.quad > 0x100000000ll ?
(int)(RESAMPLER_RESOLUTION / (phase_inc.quad * (1.f / 0x100000000ll)) * RESAMPLER_SINC_CUTOFF) :
(int)(RESAMPLER_RESOLUTION * RESAMPLER_SINC_CUTOFF);
int window_step = RESAMPLER_RESOLUTION;
do
{
double kernel[SINC_WIDTH * 2], kernel_sum = 0.0;
int i = SINC_WIDTH;
int phase_reduced = PHASE_REDUCE(phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
float sample;
if ( out >= out_end )
break;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernel[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
for (sample = 0, i = 0; i < SINC_WIDTH * 2; ++i)
sample += (float)(in[i] * kernel[i]);
*out++ = (float)(sample / kernel_sum);
phase.quad += phase_inc.quad;
ADD_HI(in, phase);
CLEAR_HI(phase);
}
while ( in < in_end );
r->phase = phase;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifdef RESAMPLER_SSE
static int resampler_run_sinc_sse(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= SINC_WIDTH * 2;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
int step = phase_inc.quad > 0x100000000ll ?
(int)(RESAMPLER_RESOLUTION / (phase_inc.quad * (1.f / 0x100000000ll)) * RESAMPLER_SINC_CUTOFF) :
(int)(RESAMPLER_RESOLUTION * RESAMPLER_SINC_CUTOFF);
int window_step = RESAMPLER_RESOLUTION;
do
{
// accumulate in extended precision
float kernel_sum = 0.0;
__m128 kernel[SINC_WIDTH / 2];
__m128 temp1, temp2;
__m128 samplex = _mm_setzero_ps();
float *kernelf = (float*)(&kernel);
int i = SINC_WIDTH;
int phase_reduced = PHASE_REDUCE(phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
if ( out >= out_end )
break;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernelf[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
for (i = 0; i < SINC_WIDTH / 2; ++i)
{
temp1 = _mm_loadu_ps( (const float *)( in + i * 4 ) );
temp2 = _mm_load_ps( (const float *)( kernel + i ) );
temp1 = _mm_mul_ps( temp1, temp2 );
samplex = _mm_add_ps( samplex, temp1 );
}
kernel_sum = 1.0f / kernel_sum;
temp1 = _mm_movehl_ps( temp1, samplex );
samplex = _mm_add_ps( samplex, temp1 );
temp1 = samplex;
temp1 = _mm_shuffle_ps( temp1, samplex, _MM_SHUFFLE(0, 0, 0, 1) );
samplex = _mm_add_ps( samplex, temp1 );
temp1 = _mm_set_ss( kernel_sum );
samplex = _mm_mul_ps( samplex, temp1 );
_mm_store_ss( out, samplex );
++out;
phase.quad += phase_inc.quad;
ADD_HI(in, phase);
CLEAR_HI(phase);
}
while ( in < in_end );
r->phase = phase;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
#ifdef RESAMPLER_NEON
static int resampler_run_sinc(resampler * r, float ** out_, float * out_end)
{
int in_size = r->write_filled;
float const* in_ = r->buffer_in + resampler_buffer_size + r->write_pos - r->write_filled;
int used = 0;
in_size -= SINC_WIDTH * 2;
if ( in_size > 0 )
{
float* out = *out_;
float const* in = in_;
float const* const in_end = in + in_size;
bigint phase = r->phase;
bigint phase_inc = r->phase_inc;
int step = phase_inc.quad > 0x100000000ll ?
(int)(RESAMPLER_RESOLUTION / (phase_inc.quad * (1.f / 0x100000000ll)) * RESAMPLER_SINC_CUTOFF) :
(int)(RESAMPLER_RESOLUTION * RESAMPLER_SINC_CUTOFF);
int window_step = RESAMPLER_RESOLUTION;
do
{
// accumulate in extended precision
float kernel_sum = 0.0;
float32x4_t kernel[SINC_WIDTH / 2];
float32x4_t temp1, temp2;
float32x4_t samplex = {0};
float32x2_t half;
float *kernelf = (float*)(&kernel);
int i = SINC_WIDTH;
int phase_reduced = PHASE_REDUCE(phase);
int phase_adj = phase_reduced * step / RESAMPLER_RESOLUTION;
if ( out >= out_end )
break;
for (; i >= -SINC_WIDTH + 1; --i)
{
int pos = i * step;
int window_pos = i * window_step;
kernel_sum += kernelf[i + SINC_WIDTH - 1] = sinc_lut[abs(phase_adj - pos)] * window_lut[abs(phase_reduced - window_pos)];
}
for (i = 0; i < SINC_WIDTH / 2; ++i)
{
temp1 = vld1q_f32( (const float32_t *)( in + i * 4 ) );
temp2 = vld1q_f32( (const float32_t *)( kernel + i ) );
samplex = vmlaq_f32( samplex, temp1, temp2 );
}
kernel_sum = 1.0 / kernel_sum;
samplex = vmulq_f32(samplex, vmovq_n_f32(kernel_sum));
half = vadd_f32(vget_high_f32(samplex), vget_low_f32(samplex));
*out++ = vget_lane_f32(vpadd_f32(half, half), 0);
phase.quad += phase_inc.quad;
ADD_HI(in, phase);
CLEAR_HI(phase);
}
while ( in < in_end );
r->phase = phase;
*out_ = out;
used = (int)(in - in_);
r->write_filled -= used;
}
return used;
}
#endif
static void resampler_fill(resampler * r)
{
int min_filled = resampler_min_filled(r);
int quality = r->quality;
while ( r->write_filled > min_filled &&
r->read_filled < resampler_buffer_size )
{
int write_pos = ( r->read_pos + r->read_filled ) % resampler_buffer_size;
int write_size = resampler_buffer_size - write_pos;
float * out = r->buffer_out + write_pos;
if ( write_size > ( resampler_buffer_size - r->read_filled ) )
write_size = resampler_buffer_size - r->read_filled;
switch (quality)
{
case RESAMPLER_QUALITY_ZOH:
resampler_run_zoh( r, &out, out + write_size );
break;
case RESAMPLER_QUALITY_BLEP:
{
int used;
int write_extra = 0;
if ( write_pos >= r->read_pos )
write_extra = r->read_pos;
if ( write_extra > SINC_WIDTH * 2 - 1 )
write_extra = SINC_WIDTH * 2 - 1;
memcpy( r->buffer_out + resampler_buffer_size, r->buffer_out, write_extra * sizeof(r->buffer_out[0]) );
#ifdef RESAMPLER_SSE
if ( resampler_has_sse )
used = resampler_run_blep_sse( r, &out, out + write_size + write_extra );
else
#endif
used = resampler_run_blep( r, &out, out + write_size + write_extra );
memcpy( r->buffer_out, r->buffer_out + resampler_buffer_size, write_extra * sizeof(r->buffer_out[0]) );
if (!used)
return;
break;
}
case RESAMPLER_QUALITY_LINEAR:
resampler_run_linear( r, &out, out + write_size );
break;
case RESAMPLER_QUALITY_BLAM:
{
float * out_ = out;
int write_extra = 0;
if ( write_pos >= r->read_pos )
write_extra = r->read_pos;
if ( write_extra > SINC_WIDTH * 2 - 1 )
write_extra = SINC_WIDTH * 2 - 1;
memcpy( r->buffer_out + resampler_buffer_size, r->buffer_out, write_extra * sizeof(r->buffer_out[0]) );
#ifdef RESAMPLER_SSE
if ( resampler_has_sse )
resampler_run_blam_sse( r, &out, out + write_size + write_extra );
else
#endif
resampler_run_blam( r, &out, out + write_size + write_extra );
memcpy( r->buffer_out, r->buffer_out + resampler_buffer_size, write_extra * sizeof(r->buffer_out[0]) );
if ( out == out_ )
return;
break;
}
case RESAMPLER_QUALITY_CUBIC:
#ifdef RESAMPLER_SSE
if ( resampler_has_sse )
resampler_run_cubic_sse( r, &out, out + write_size );
else
#endif
resampler_run_cubic( r, &out, out + write_size );
break;
case RESAMPLER_QUALITY_SINC:
#ifdef RESAMPLER_SSE
if ( resampler_has_sse )
resampler_run_sinc_sse( r, &out, out + write_size );
else
#endif
resampler_run_sinc( r, &out, out + write_size );
break;
}
r->read_filled += (int)(out - r->buffer_out - write_pos);
}
}
static void resampler_fill_and_remove_delay(resampler * r)
{
resampler_fill( r );
if ( r->delay_removed < 0 )
{
int delay = resampler_output_delay( r );
r->delay_removed = 0;
while ( delay-- )
resampler_remove_sample( r, 1 );
}
}
int resampler_get_sample_count(void *_r)
{
resampler * r = ( resampler * ) _r;
if ( r->read_filled < 1 && ((r->quality != RESAMPLER_QUALITY_BLEP && r->quality != RESAMPLER_QUALITY_BLAM) || r->inv_phase_inc.quad))
resampler_fill_and_remove_delay( r );
return r->read_filled;
}
int resampler_get_sample(void *_r)
{
resampler * r = ( resampler * ) _r;
if ( r->read_filled < 1 && r->phase_inc.quad)
resampler_fill_and_remove_delay( r );
if ( r->read_filled < 1 )
return 0;
if ( r->quality == RESAMPLER_QUALITY_BLEP || r->quality == RESAMPLER_QUALITY_BLAM )
return (int)(r->buffer_out[ r->read_pos ] + r->accumulator);
else
return (int)r->buffer_out[ r->read_pos ];
}
float resampler_get_sample_float(void *_r)
{
resampler * r = ( resampler * ) _r;
if ( r->read_filled < 1 && r->phase_inc.quad)
resampler_fill_and_remove_delay( r );
if ( r->read_filled < 1 )
return 0;
if ( r->quality == RESAMPLER_QUALITY_BLEP || r->quality == RESAMPLER_QUALITY_BLAM )
return (float)(r->buffer_out[ r->read_pos ] + r->accumulator);
else
return r->buffer_out[ r->read_pos ];
}
void resampler_remove_sample(void *_r, int decay)
{
resampler * r = ( resampler * ) _r;
if ( r->read_filled > 0 )
{
if ( r->quality == RESAMPLER_QUALITY_BLEP || r->quality == RESAMPLER_QUALITY_BLAM )
{
r->accumulator += r->buffer_out[ r->read_pos ];
r->buffer_out[ r->read_pos ] = 0;
if (decay)
{
r->accumulator -= r->accumulator * (1.0f / 8192.0f);
if (fabs(r->accumulator) < 1e-20f)
r->accumulator = 0;
}
}
--r->read_filled;
r->read_pos = ( r->read_pos + 1 ) % resampler_buffer_size;
}
}
Reference in a new issue View git blame Copy permalink