ShuriZma
/
RetroArch
mirror of https://github.com/libretro/RetroArch.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Revert "Rollback changes for now until we can get this compiling on MSVC 2005 on" 

This reverts commit 08481e2a68.
This commit is contained in:
twinaphex 2021-06-11 12:05:48 +02:00
parent 5d4069cf8f
commit bd9c35d379
1 changed files with 416 additions and 378 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

154
libretro-common/audio/resampler/drivers/sinc_resampler.c
View File

@ -75,16 +75,15 @@ typedef struct rarch_sinc_resampler
float *phase_table; float *phase_table;
float *buffer_l; float *buffer_l;
float *buffer_r; float *buffer_r;
unsigned enable_avx;
unsigned phase_bits; unsigned phase_bits;
unsigned subphase_bits; unsigned subphase_bits;
unsigned subphase_mask; unsigned subphase_mask;
unsigned taps; unsigned taps;
unsigned ptr; unsigned ptr;
unsigned num_channels;
uint32_t time; uint32_t time;
float subphase_mod; float subphase_mod;
float kaiser_beta; float kaiser_beta;
enum sinc_window window_type;
} rarch_sinc_resampler_t; } rarch_sinc_resampler_t;
#if (defined(__ARM_NEON__) && !defined(DONT_WANT_ARM_OPTIMIZATIONS)) || defined(HAVE_NEON) #if (defined(__ARM_NEON__) && !defined(DONT_WANT_ARM_OPTIMIZATIONS)) || defined(HAVE_NEON)
@ -154,7 +153,7 @@ static void resampler_sinc_process_neon(void *re_, struct resampler_data *data)
#endif #endif
#if defined(__AVX__) #if defined(__AVX__)
static void resampler_sinc_process_avx(void *re_, struct resampler_data *data) static void resampler_sinc_process_avx_kaiser(void *re_, struct resampler_data *data)
{ {
rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_; rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_;
unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits); unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits);
@ -165,8 +164,6 @@ static void resampler_sinc_process_avx(void *re_, struct resampler_data *data)
size_t frames = data->input_frames; size_t frames = data->input_frames;
size_t out_frames = 0; size_t out_frames = 0;
if (resamp->window_type == SINC_WINDOW_KAISER)
{
while (frames) while (frames)
{ {
while (frames && resamp->time >= phases) while (frames && resamp->time >= phases)
@ -235,9 +232,21 @@ static void resampler_sinc_process_avx(void *re_, struct resampler_data *data)
} }
} }
} }
data->output_frames = out_frames;
} }
else
static void resampler_sinc_process_avx(void *re_, struct resampler_data *data)
{ {
rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_;
unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits);
uint32_t ratio = phases / data->ratio;
const float *input = data->data_in;
float *output = data->data_out;
size_t frames = data->input_frames;
size_t out_frames = 0;
while (frames) while (frames)
{ {
while (frames && resamp->time >= phases) while (frames && resamp->time >= phases)
@ -301,14 +310,13 @@ static void resampler_sinc_process_avx(void *re_, struct resampler_data *data)
} }
} }
} }
}
data->output_frames = out_frames; data->output_frames = out_frames;
} }
#endif #endif
#if defined(__SSE__) #if defined(__SSE__)
static void resampler_sinc_process_sse(void *re_, struct resampler_data *data) static void resampler_sinc_process_sse_kaiser(void *re_, struct resampler_data *data)
{ {
rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_; rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_;
unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits); unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits);
@ -319,8 +327,6 @@ static void resampler_sinc_process_sse(void *re_, struct resampler_data *data)
size_t frames = data->input_frames; size_t frames = data->input_frames;
size_t out_frames = 0; size_t out_frames = 0;
if (resamp->window_type == SINC_WINDOW_KAISER)
{
while (frames) while (frames)
{ {
while (frames && resamp->time >= phases) while (frames && resamp->time >= phases)
@ -347,7 +353,9 @@ static void resampler_sinc_process_sse(void *re_, struct resampler_data *data)
while (resamp->time < phases) while (resamp->time < phases)
{ {
unsigned i; unsigned i;
#if 0
__m128 sum; __m128 sum;
#endif
unsigned phase = resamp->time >> resamp->subphase_bits; unsigned phase = resamp->time >> resamp->subphase_bits;
float *phase_table = resamp->phase_table + phase * taps * 2; float *phase_table = resamp->phase_table + phase * taps * 2;
float *delta_table = phase_table + taps; float *delta_table = phase_table + taps;
@ -368,40 +376,55 @@ static void resampler_sinc_process_sse(void *re_, struct resampler_data *data)
sum_r = _mm_add_ps(sum_r, _mm_mul_ps(buf_r, _sinc)); sum_r = _mm_add_ps(sum_r, _mm_mul_ps(buf_r, _sinc));
} }
#ifdef HAVE_GRIFFIN
/* Them annoying shuffles. /* Them annoying shuffles.
* sum_l = { l3, l2, l1, l0 } * sum_l = { l3, l2, l1, l0 }
* sum_r = { r3, r2, r1, r0 } * sum_r = { r3, r2, r1, r0 }
*/ */
sum = _mm_add_ps(_mm_shuffle_ps(sum_l, sum_r, sum = _mm_add_ps(_mm_shuffle_ps(sum_l, sum_r,
_MM_SHUFFLE(1, 0, 1, 0)), _MM_SHUFFLE(1, 0, 1, 0)),
_mm_shuffle_ps(sum_l, sum_r, _MM_SHUFFLE(3, 2, 3, 2))); _mm_shuffle_ps(sum_l, sum_r, _MM_SHUFFLE(3, 2, 3, 2)));
/* sum = { r1, r0, l1, l0 } + { r3, r2, l3, l2 } /* sum = { r1, r0, l1, l0 } + { r3, r2, l3, l2 }
* sum = { R1, R0, L1, L0 } * sum = { R1, R0, L1, L0 }
*/ */
sum = _mm_add_ps(_mm_shuffle_ps(sum, sum, _MM_SHUFFLE(3, 3, 1, 1)), sum); sum = _mm_add_ps(_mm_shuffle_ps(sum, sum, _MM_SHUFFLE(3, 3, 1, 1)), sum);
/* sum = {R1, R1, L1, L1 } + { R1, R0, L1, L0 } /* sum = {R1, R1, L1, L1 } + { R1, R0, L1, L0 }
* sum = { X, R, X, L } * sum = { X, R, X, L }
*/ */
/* Store L */ /* Store L */
_mm_store_ss(output + 0, sum); _mm_store_ss(output++, sum);
/* movehl { X, R, X, L } == { X, R, X, R } */ /* movehl { X, R, X, L } == { X, R, X, R } */
_mm_store_ss(output + 1, _mm_movehl_ps(sum, sum)); _mm_store_ss(output++, _mm_movehl_ps(sum, sum));
#else
#ifdef _MSC_VER
*(output++) = _mm_cvtss_f32(sum_l) + sum_l.m128_f32[1] + sum_l.m128_f32[2] + sum_l.m128_f32[3];
*(output++) = _mm_cvtss_f32(sum_r) + sum_r.m128_f32[1] + sum_r.m128_f32[2] + sum_r.m128_f32[3];
#else
*(output++) = _mm_cvtss_f32(sum_l) + sum_l[1] + sum_l[2] + sum_l[3];
*(output++) = _mm_cvtss_f32(sum_r) + sum_r[1] + sum_r[2] + sum_r[3];
#endif
#endif
output += 2;
out_frames++; out_frames++;
resamp->time += ratio; resamp->time += ratio;
} }
} }
} }
data->output_frames = out_frames;
} }
else
static void resampler_sinc_process_sse(void *re_, struct resampler_data *data)
{ {
rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_;
unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits);
uint32_t ratio = phases / data->ratio;
const float *input = data->data_in;
float *output = data->data_out;
size_t frames = data->input_frames;
size_t out_frames = 0;
while (frames) while (frames)
{ {
while (frames && resamp->time >= phases) while (frames && resamp->time >= phases)
@ -428,7 +451,9 @@ static void resampler_sinc_process_sse(void *re_, struct resampler_data *data)
while (resamp->time < phases) while (resamp->time < phases)
{ {
unsigned i; unsigned i;
#if 0
__m128 sum; __m128 sum;
#endif
unsigned phase = resamp->time >> resamp->subphase_bits; unsigned phase = resamp->time >> resamp->subphase_bits;
float *phase_table = resamp->phase_table + phase * taps; float *phase_table = resamp->phase_table + phase * taps;
@ -444,44 +469,46 @@ static void resampler_sinc_process_sse(void *re_, struct resampler_data *data)
sum_r = _mm_add_ps(sum_r, _mm_mul_ps(buf_r, _sinc)); sum_r = _mm_add_ps(sum_r, _mm_mul_ps(buf_r, _sinc));
} }
#if 0
/* Them annoying shuffles. /* Them annoying shuffles.
* sum_l = { l3, l2, l1, l0 } * sum_l = { l3, l2, l1, l0 }
* sum_r = { r3, r2, r1, r0 } * sum_r = { r3, r2, r1, r0 }
*/ */
sum = _mm_add_ps(_mm_shuffle_ps(sum_l, sum_r, sum = _mm_add_ps(_mm_shuffle_ps(sum_l, sum_r,
_MM_SHUFFLE(1, 0, 1, 0)), _MM_SHUFFLE(1, 0, 1, 0)),
_mm_shuffle_ps(sum_l, sum_r, _MM_SHUFFLE(3, 2, 3, 2))); _mm_shuffle_ps(sum_l, sum_r, _MM_SHUFFLE(3, 2, 3, 2)));
/* sum = { r1, r0, l1, l0 } + { r3, r2, l3, l2 } /* sum = { r1, r0, l1, l0 } + { r3, r2, l3, l2 }
* sum = { R1, R0, L1, L0 } * sum = { R1, R0, L1, L0 }
*/ */
sum = _mm_add_ps(_mm_shuffle_ps(sum, sum, _MM_SHUFFLE(3, 3, 1, 1)), sum); sum = _mm_add_ps(_mm_shuffle_ps(sum, sum, _MM_SHUFFLE(3, 3, 1, 1)), sum);
/* sum = {R1, R1, L1, L1 } + { R1, R0, L1, L0 } /* sum = {R1, R1, L1, L1 } + { R1, R0, L1, L0 }
* sum = { X, R, X, L } * sum = { X, R, X, L }
*/ */
/* Store L */ /* Store L */
_mm_store_ss(output + 0, sum); _mm_store_ss(output++, sum);
/* movehl { X, R, X, L } == { X, R, X, R } */ /* movehl { X, R, X, L } == { X, R, X, R } */
_mm_store_ss(output + 1, _mm_movehl_ps(sum, sum)); _mm_store_ss(output++, _mm_movehl_ps(sum, sum));
#else
#ifdef _MSC_VER
*(output++) = _mm_cvtss_f32(sum_l) + sum_l.m128_f32[1] + sum_l.m128_f32[2] + sum_l.m128_f32[3];
*(output++) = _mm_cvtss_f32(sum_r) + sum_r.m128_f32[1] + sum_r.m128_f32[2] + sum_r.m128_f32[3];
#else
*(output++) = _mm_cvtss_f32(sum_l) + sum_l[1] + sum_l[2] + sum_l[3];
*(output++) = _mm_cvtss_f32(sum_r) + sum_r[1] + sum_r[2] + sum_r[3];
#endif
#endif
output += 2;
out_frames++; out_frames++;
resamp->time += ratio; resamp->time += ratio;
} }
} }
} }
}
data->output_frames = out_frames; data->output_frames = out_frames;
} }
#endif #endif
static void resampler_sinc_process_c(void *re_, struct resampler_data *data) static void resampler_sinc_process_c_kaiser(void *re_, struct resampler_data *data)
{ {
rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_; rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_;
unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits); unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits);
@ -492,8 +519,6 @@ static void resampler_sinc_process_c(void *re_, struct resampler_data *data)
size_t frames = data->input_frames; size_t frames = data->input_frames;
size_t out_frames = 0; size_t out_frames = 0;
if (resamp->window_type == SINC_WINDOW_KAISER)
{
while (frames) while (frames)
{ {
while (frames && resamp->time >= phases) while (frames && resamp->time >= phases)
@ -536,19 +561,30 @@ static void resampler_sinc_process_c(void *re_, struct resampler_data *data)
sum_r += buffer_r[i] * sinc_val; sum_r += buffer_r[i] * sinc_val;
} }
output[0] = sum_l; *output++ = sum_l;
output[1] = sum_r; *output++ = sum_r;
output += 2;
out_frames++; out_frames++;
resamp->time += ratio; resamp->time += ratio;
} }
} }
} }
data->output_frames = out_frames;
} }
else
static void resampler_sinc_process_c(void *re_, struct resampler_data *data)
{ {
rarch_sinc_resampler_t *resamp = (rarch_sinc_resampler_t*)re_;
unsigned phases = 1 << (resamp->phase_bits + resamp->subphase_bits);
uint32_t ratio = phases / data->ratio;
const float *input = data->data_in;
float *output = data->data_out;
size_t frames = data->input_frames;
size_t out_frames = 0;
while (frames) while (frames)
{ {
while (frames && resamp->time >= phases) while (frames && resamp->time >= phases)
@ -588,17 +624,15 @@ static void resampler_sinc_process_c(void *re_, struct resampler_data *data)
sum_r += buffer_r[i] * sinc_val; sum_r += buffer_r[i] * sinc_val;
} }
output[0] = sum_l; *output++ = sum_l;
output[1] = sum_r; *output++ = sum_r;
output += 2;
out_frames++; out_frames++;
resamp->time += ratio; resamp->time += ratio;
} }
} }
} }
}
data->output_frames = out_frames; data->output_frames = out_frames;
} }
@ -735,14 +769,14 @@ static void *resampler_sinc_new(const struct resampler_config *config,
size_t phase_elems = 0; size_t phase_elems = 0;
size_t elems = 0; size_t elems = 0;
unsigned sidelobes = 0; unsigned sidelobes = 0;
unsigned enable_avx = 0;
enum sinc_window window_type = SINC_WINDOW_NONE;
rarch_sinc_resampler_t *re = (rarch_sinc_resampler_t*) rarch_sinc_resampler_t *re = (rarch_sinc_resampler_t*)
calloc(1, sizeof(*re)); calloc(1, sizeof(*re));
if (!re) if (!re)
return NULL; return NULL;
re->window_type = SINC_WINDOW_NONE;
switch (quality) switch (quality)
{ {
case RESAMPLER_QUALITY_LOWEST: case RESAMPLER_QUALITY_LOWEST:
@ -750,34 +784,32 @@ static void *resampler_sinc_new(const struct resampler_config *config,
sidelobes = 2; sidelobes = 2;
re->phase_bits = 12; re->phase_bits = 12;
re->subphase_bits = 10; re->subphase_bits = 10;
re->window_type = SINC_WINDOW_LANCZOS; window_type = SINC_WINDOW_LANCZOS;
re->enable_avx = 0;
break; break;
case RESAMPLER_QUALITY_LOWER: case RESAMPLER_QUALITY_LOWER:
cutoff = 0.98; cutoff = 0.98;
sidelobes = 4; sidelobes = 4;
re->phase_bits = 12; re->phase_bits = 12;
re->subphase_bits = 10; re->subphase_bits = 10;
re->window_type = SINC_WINDOW_LANCZOS; window_type = SINC_WINDOW_LANCZOS;
re->enable_avx = 0;
break; break;
case RESAMPLER_QUALITY_HIGHER: case RESAMPLER_QUALITY_HIGHER:
cutoff = 0.90; cutoff = 0.90;
sidelobes = 32; sidelobes = 32;
re->phase_bits = 10; re->phase_bits = 10;
re->subphase_bits = 14; re->subphase_bits = 14;
re->window_type = SINC_WINDOW_KAISER;
re->kaiser_beta = 10.5; re->kaiser_beta = 10.5;
re->enable_avx = 1; enable_avx = 1;
window_type = SINC_WINDOW_KAISER;
break; break;
case RESAMPLER_QUALITY_HIGHEST: case RESAMPLER_QUALITY_HIGHEST:
cutoff = 0.962; cutoff = 0.962;
sidelobes = 128; sidelobes = 128;
re->phase_bits = 10; re->phase_bits = 10;
re->subphase_bits = 14; re->subphase_bits = 14;
re->window_type = SINC_WINDOW_KAISER;
re->kaiser_beta = 14.5; re->kaiser_beta = 14.5;
re->enable_avx = 1; enable_avx = 1;
window_type = SINC_WINDOW_KAISER;
break; break;
case RESAMPLER_QUALITY_NORMAL: case RESAMPLER_QUALITY_NORMAL:
case RESAMPLER_QUALITY_DONTCARE: case RESAMPLER_QUALITY_DONTCARE:
@ -785,14 +817,14 @@ static void *resampler_sinc_new(const struct resampler_config *config,
sidelobes = 8; sidelobes = 8;
re->phase_bits = 8; re->phase_bits = 8;
re->subphase_bits = 16; re->subphase_bits = 16;
re->window_type = SINC_WINDOW_KAISER;
re->kaiser_beta = 5.5; re->kaiser_beta = 5.5;
re->enable_avx = 0; window_type = SINC_WINDOW_KAISER;
break; break;
} }
re->subphase_mask = (1 << re->subphase_bits) - 1; re->subphase_mask = (1 << re->subphase_bits) - 1;
re->subphase_mod = 1.0f / (1 << re->subphase_bits); re->subphase_mod = 1.0f / (1 << re->subphase_bits);
re->num_channels = 2;
re->taps = sidelobes * 2; re->taps = sidelobes * 2;
/* Downsampling, must lower cutoff, and extend number of /* Downsampling, must lower cutoff, and extend number of
@ -805,7 +837,7 @@ static void *resampler_sinc_new(const struct resampler_config *config,
/* Be SIMD-friendly. */ /* Be SIMD-friendly. */
#if defined(__AVX__) #if defined(__AVX__)
if (re->enable_avx) if (enable_avx)
re->taps = (re->taps + 7) & ~7; re->taps = (re->taps + 7) & ~7;
else else
#endif #endif
@ -818,7 +850,7 @@ static void *resampler_sinc_new(const struct resampler_config *config,
} }
phase_elems = ((1 << re->phase_bits) * re->taps); phase_elems = ((1 << re->phase_bits) * re->taps);
if (re->window_type == SINC_WINDOW_KAISER) if (window_type == SINC_WINDOW_KAISER)
phase_elems = phase_elems * 2; phase_elems = phase_elems * 2;
elems = phase_elems + 4 * re->taps; elems = phase_elems + 4 * re->taps;
@ -832,7 +864,7 @@ static void *resampler_sinc_new(const struct resampler_config *config,
re->buffer_l = re->main_buffer + phase_elems; re->buffer_l = re->main_buffer + phase_elems;
re->buffer_r = re->buffer_l + 2 * re->taps; re->buffer_r = re->buffer_l + 2 * re->taps;
switch (re->window_type) switch (window_type)
{ {
case SINC_WINDOW_LANCZOS: case SINC_WINDOW_LANCZOS:
sinc_init_table_lanczos(re, cutoff, re->phase_table, sinc_init_table_lanczos(re, cutoff, re->phase_table,
@ -847,20 +879,26 @@ static void *resampler_sinc_new(const struct resampler_config *config,
} }
sinc_resampler.process = resampler_sinc_process_c; sinc_resampler.process = resampler_sinc_process_c;
if (window_type == SINC_WINDOW_KAISER)
sinc_resampler.process = resampler_sinc_process_c_kaiser;
if (mask & RESAMPLER_SIMD_AVX && re->enable_avx) if (mask & RESAMPLER_SIMD_AVX && enable_avx)
{ {
#if defined(__AVX__) #if defined(__AVX__)
sinc_resampler.process = resampler_sinc_process_avx; sinc_resampler.process = resampler_sinc_process_avx;
if (window_type == SINC_WINDOW_KAISER)
sinc_resampler.process = resampler_sinc_process_avx_kaiser;
#endif #endif
} }
else if (mask & RESAMPLER_SIMD_SSE) else if (mask & RESAMPLER_SIMD_SSE)
{ {
#if defined(__SSE__) #if defined(__SSE__)
sinc_resampler.process = resampler_sinc_process_sse; sinc_resampler.process = resampler_sinc_process_sse;
if (window_type == SINC_WINDOW_KAISER)
sinc_resampler.process = resampler_sinc_process_sse_kaiser;
#endif #endif
} }
else if (mask & RESAMPLER_SIMD_NEON && re->window_type != SINC_WINDOW_KAISER) else if (mask & RESAMPLER_SIMD_NEON && window_type != SINC_WINDOW_KAISER)
{ {
#if defined(WANT_NEON) #if defined(WANT_NEON)
sinc_resampler.process = resampler_sinc_process_neon; sinc_resampler.process = resampler_sinc_process_neon;