mirror/RetroArch
1
0
Fork 0
mirror of https://github.com/libretro/RetroArch synced 2025-01-30 12:32:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

(resamplers) Turn some 1-line inline functions into macros

Browse Source
This commit is contained in:
twinaphex 2015-07-10 19:46:52 +02:00
parent 7bcd4dd04d
commit 33feed8aa7
2 changed files with 48 additions and 60 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
audio/drivers_resampler/cc_resampler.c
View File

@ -419,10 +419,7 @@ static INLINE float cc_int(float x, float b)
return (val > M_PI) ? M_PI : (val < -M_PI) ? -M_PI : val; return (val > M_PI) ? M_PI : (val < -M_PI) ? -M_PI : val;
} }
static INLINE float cc_kernel(float x, float b) #define cc_kernel(x, b) ((cc_int((x) + 0.5, (b)) - cc_int((x) - 0.5, (b))) / (2.0 * M_PI))
{
return (cc_int(x + 0.5, b) - cc_int(x - 0.5, b)) / (2.0 * M_PI);
}
#else #else
static INLINE float cc_int(float x, float b) static INLINE float cc_int(float x, float b)
{ {
@ -433,10 +430,7 @@ static INLINE float cc_int(float x, float b)
return (val > 0.5) ? 0.5 : (val < -0.5) ? -0.5 : val; return (val > 0.5) ? 0.5 : (val < -0.5) ? -0.5 : val;
} }
static INLINE float cc_kernel(float x, float b) #define cc_kernel(x, b) ((cc_int((x) + 0.5, (b)) - cc_int((x) - 0.5, (b))))
{
return (cc_int(x + 0.5, b) - cc_int(x - 0.5, b));
}
#endif #endif
static INLINE void add_to(const audio_frame_float_t *source, static INLINE void add_to(const audio_frame_float_t *source,
@ -503,15 +497,15 @@ static void resampler_CC_upsample(void *re_, struct resampler_data *data)
while (re->distance < 1.0) while (re->distance < 1.0)
{ {
int i; int i;
float temp;
outp->l = 0.0; outp->l = 0.0;
outp->r = 0.0; outp->r = 0.0;
for (i = 0; i < 4; i++) for (i = 0; i < 4; i++)
{ {
temp = cc_kernel(re->distance + 1.0 - i, b); float temp = cc_kernel(re->distance + 1.0 - i, b);
outp->l += re->buffer[i].l * temp; outp->l += re->buffer[i].l * temp;
outp->r += re->buffer[i].r * temp; outp->r += re->buffer[i].r * temp;
} }
re->distance += ratio; re->distance += ratio;

90
audio/drivers_resampler/sinc.c
View File

@ -121,10 +121,7 @@ static INLINE double sinc(double val)
} }
#if defined(SINC_WINDOW_LANCZOS) #if defined(SINC_WINDOW_LANCZOS)
static INLINE double window_function(double idx) #define window_function(idx) (sinc(M_PI * (idx)))
{
return sinc(M_PI * idx);
}
#elif defined(SINC_WINDOW_KAISER) #elif defined(SINC_WINDOW_KAISER)
/* Modified Bessel function of first order. /* Modified Bessel function of first order.
* Check Wiki for mathematical definition ... */ * Check Wiki for mathematical definition ... */
@ -153,10 +150,7 @@ static INLINE double besseli0(double x)
return sum; return sum;
} }
static INLINE double window_function(double idx) #define window_function(idx) (besseli0(SINC_WINDOW_KAISER_BETA * sqrt(1 - (idx) * (idx))))
{
return besseli0(SINC_WINDOW_KAISER_BETA * sqrt(1 - idx * idx));
}
#else #else
#error "No SINC window function defined." #error "No SINC window function defined."
#endif #endif
@ -247,17 +241,17 @@ static INLINE void process_sinc_C(rarch_sinc_resampler_t *resamp,
float *out_buffer) float *out_buffer)
{ {
unsigned i; unsigned i;
float sum_l = 0.0f; float sum_l = 0.0f;
float sum_r = 0.0f; float sum_r = 0.0f;
const float *buffer_l = resamp->buffer_l + resamp->ptr; const float *buffer_l = resamp->buffer_l + resamp->ptr;
const float *buffer_r = resamp->buffer_r + resamp->ptr; const float *buffer_r = resamp->buffer_r + resamp->ptr;
unsigned taps = resamp->taps;
unsigned taps = resamp->taps; unsigned phase = resamp->time >> SUBPHASE_BITS;
unsigned phase = resamp->time >> SUBPHASE_BITS;
#if SINC_COEFF_LERP #if SINC_COEFF_LERP
const float *phase_table = resamp->phase_table + phase * taps * 2; const float *phase_table = resamp->phase_table + phase * taps * 2;
const float *delta_table = phase_table + taps; const float *delta_table = phase_table + taps;
float delta = (float)(resamp->time & SUBPHASE_MASK) * SUBPHASE_MOD; float delta = (float)
(resamp->time & SUBPHASE_MASK) * SUBPHASE_MOD;
#else #else
const float *phase_table = resamp->phase_table + phase * taps; const float *phase_table = resamp->phase_table + phase * taps;
#endif #endif
@ -283,18 +277,18 @@ static INLINE void process_sinc_C(rarch_sinc_resampler_t *resamp,
static void process_sinc(rarch_sinc_resampler_t *resamp, float *out_buffer) static void process_sinc(rarch_sinc_resampler_t *resamp, float *out_buffer)
{ {
unsigned i; unsigned i;
__m256 sum_l = _mm256_setzero_ps(); __m256 sum_l = _mm256_setzero_ps();
__m256 sum_r = _mm256_setzero_ps(); __m256 sum_r = _mm256_setzero_ps();
const float *buffer_l = resamp->buffer_l + resamp->ptr; const float *buffer_l = resamp->buffer_l + resamp->ptr;
const float *buffer_r = resamp->buffer_r + resamp->ptr; const float *buffer_r = resamp->buffer_r + resamp->ptr;
unsigned taps = resamp->taps; unsigned taps = resamp->taps;
unsigned phase = resamp->time >> SUBPHASE_BITS; unsigned phase = resamp->time >> SUBPHASE_BITS;
#if SINC_COEFF_LERP #if SINC_COEFF_LERP
const float *phase_table = resamp->phase_table + phase * taps * 2; const float *phase_table = resamp->phase_table + phase * taps * 2;
const float *delta_table = phase_table + taps; const float *delta_table = phase_table + taps;
__m256 delta = _mm256_set1_ps((float) __m256 delta = _mm256_set1_ps((float)
(resamp->time & SUBPHASE_MASK) * SUBPHASE_MOD); (resamp->time & SUBPHASE_MASK) * SUBPHASE_MOD);
#else #else
const float *phase_table = resamp->phase_table + phase * taps; const float *phase_table = resamp->phase_table + phase * taps;
@ -302,28 +296,28 @@ static void process_sinc(rarch_sinc_resampler_t *resamp, float *out_buffer)
for (i = 0; i < taps; i += 8) for (i = 0; i < taps; i += 8)
{ {
__m256 buf_l = _mm256_loadu_ps(buffer_l + i); __m256 buf_l = _mm256_loadu_ps(buffer_l + i);
__m256 buf_r = _mm256_loadu_ps(buffer_r + i); __m256 buf_r = _mm256_loadu_ps(buffer_r + i);
#if SINC_COEFF_LERP #if SINC_COEFF_LERP
__m256 deltas = _mm256_load_ps(delta_table + i); __m256 deltas = _mm256_load_ps(delta_table + i);
__m256 sinc = _mm256_add_ps(_mm256_load_ps(phase_table + i), __m256 sinc = _mm256_add_ps(_mm256_load_ps(phase_table + i),
_mm256_mul_ps(deltas, delta)); _mm256_mul_ps(deltas, delta));
#else #else
__m256 sinc = _mm256_load_ps(phase_table + i); __m256 sinc = _mm256_load_ps(phase_table + i);
#endif #endif
sum_l = _mm256_add_ps(sum_l, _mm256_mul_ps(buf_l, sinc)); sum_l = _mm256_add_ps(sum_l, _mm256_mul_ps(buf_l, sinc));
sum_r = _mm256_add_ps(sum_r, _mm256_mul_ps(buf_r, sinc)); sum_r = _mm256_add_ps(sum_r, _mm256_mul_ps(buf_r, sinc));
} }
/* hadd on AVX is weird, and acts on low-lanes /* hadd on AVX is weird, and acts on low-lanes
* and high-lanes separately. */ * and high-lanes separately. */
__m256 res_l = _mm256_hadd_ps(sum_l, sum_l); __m256 res_l = _mm256_hadd_ps(sum_l, sum_l);
__m256 res_r = _mm256_hadd_ps(sum_r, sum_r); __m256 res_r = _mm256_hadd_ps(sum_r, sum_r);
res_l = _mm256_hadd_ps(res_l, res_l); res_l = _mm256_hadd_ps(res_l, res_l);
res_r = _mm256_hadd_ps(res_r, res_r); res_r = _mm256_hadd_ps(res_r, res_r);
res_l = _mm256_add_ps(_mm256_permute2f128_ps(res_l, res_l, 1), res_l); res_l = _mm256_add_ps(_mm256_permute2f128_ps(res_l, res_l, 1), res_l);
res_r = _mm256_add_ps(_mm256_permute2f128_ps(res_r, res_r, 1), res_r); res_r = _mm256_add_ps(_mm256_permute2f128_ps(res_r, res_r, 1), res_r);
/* This is optimized to mov %xmmN, [mem]. /* This is optimized to mov %xmmN, [mem].
* There doesn't seem to be any _mm256_store_ss intrinsic. */ * There doesn't seem to be any _mm256_store_ss intrinsic. */
@ -335,23 +329,23 @@ static void process_sinc(rarch_sinc_resampler_t *resamp, float *out_buffer)
static void process_sinc(rarch_sinc_resampler_t *resamp, float *out_buffer) static void process_sinc(rarch_sinc_resampler_t *resamp, float *out_buffer)
{ {
unsigned i; unsigned i;
__m128 sum_l = _mm_setzero_ps(); __m128 sum;
__m128 sum_r = _mm_setzero_ps(); __m128 sum_l = _mm_setzero_ps();
__m128 sum_r = _mm_setzero_ps();
const float *buffer_l = resamp->buffer_l + resamp->ptr; const float *buffer_l = resamp->buffer_l + resamp->ptr;
const float *buffer_r = resamp->buffer_r + resamp->ptr; const float *buffer_r = resamp->buffer_r + resamp->ptr;
unsigned taps = resamp->taps; unsigned taps = resamp->taps;
unsigned phase = resamp->time >> SUBPHASE_BITS; unsigned phase = resamp->time >> SUBPHASE_BITS;
#if SINC_COEFF_LERP #if SINC_COEFF_LERP
const float *phase_table = resamp->phase_table + phase * taps * 2; const float *phase_table = resamp->phase_table + phase * taps * 2;
const float *delta_table = phase_table + taps; const float *delta_table = phase_table + taps;
__m128 delta = _mm_set1_ps((float) __m128 delta = _mm_set1_ps((float)
(resamp->time & SUBPHASE_MASK) * SUBPHASE_MOD); (resamp->time & SUBPHASE_MASK) * SUBPHASE_MOD);
#else #else
const float *phase_table = resamp->phase_table + phase * taps; const float *phase_table = resamp->phase_table + phase * taps;
#endif #endif
__m128 sum;
for (i = 0; i < taps; i += 4) for (i = 0; i < taps; i += 4)
{ {
@ -360,13 +354,13 @@ static void process_sinc(rarch_sinc_resampler_t *resamp, float *out_buffer)
#if SINC_COEFF_LERP #if SINC_COEFF_LERP
__m128 deltas = _mm_load_ps(delta_table + i); __m128 deltas = _mm_load_ps(delta_table + i);
__m128 _sinc = _mm_add_ps(_mm_load_ps(phase_table + i), __m128 _sinc = _mm_add_ps(_mm_load_ps(phase_table + i),
_mm_mul_ps(deltas, delta)); _mm_mul_ps(deltas, delta));
#else #else
__m128 _sinc = _mm_load_ps(phase_table + i); __m128 _sinc = _mm_load_ps(phase_table + i);
#endif #endif
sum_l = _mm_add_ps(sum_l, _mm_mul_ps(buf_l, _sinc)); sum_l = _mm_add_ps(sum_l, _mm_mul_ps(buf_l, _sinc));
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));
} }
/* Them annoying shuffles. /* Them annoying shuffles.
@ -413,11 +407,11 @@ void process_sinc_neon_asm(float *out, const float *left,
static void process_sinc_neon(rarch_sinc_resampler_t *resamp, static void process_sinc_neon(rarch_sinc_resampler_t *resamp,
float *out_buffer) float *out_buffer)
{ {
const float *buffer_l = resamp->buffer_l + resamp->ptr; const float *buffer_l = resamp->buffer_l + resamp->ptr;
const float *buffer_r = resamp->buffer_r + resamp->ptr; const float *buffer_r = resamp->buffer_r + resamp->ptr;
unsigned phase = resamp->time >> SUBPHASE_BITS; unsigned phase = resamp->time >> SUBPHASE_BITS;
unsigned taps = resamp->taps; unsigned taps = resamp->taps;
const float *phase_table = resamp->phase_table + phase * taps; const float *phase_table = resamp->phase_table + phase * taps;
process_sinc_neon_asm(out_buffer, buffer_l, buffer_r, phase_table, taps); process_sinc_neon_asm(out_buffer, buffer_l, buffer_r, phase_table, taps);