mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2026-04-14 08:21:47 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

audioconvert: align some buffers

Browse source 
so that we can use aligned read and writes in SSE.
This commit is contained in:
Wim Taymans 2024-10-23 12:54:23 +02:00
parent 662bf68122
commit 2c132be626
3 changed files with 16 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

10
spa/plugins/audioconvert/channelmix-ops-sse.c
View file

@ -266,7 +266,7 @@ static inline void convolver_run(const float *src, float *dst,
sum[0] = _mm_setzero_ps(); sum[0] = _mm_setzero_ps();
for(i = 0; i < n_taps; i+=4) { for(i = 0; i < n_taps; i+=4) {
t[0] = _mm_loadu_ps(&src[i]); t[0] = _mm_loadu_ps(&src[i]);
sum[0] = _mm_add_ps(sum[0], _mm_mul_ps(_mm_loadu_ps(&taps[i]), t[0])); sum[0] = _mm_add_ps(sum[0], _mm_mul_ps(_mm_load_ps(&taps[i]), t[0]));
} }
sum[0] = _mm_add_ps(sum[0], _mm_movehl_ps(sum[0], sum[0])); sum[0] = _mm_add_ps(sum[0], _mm_movehl_ps(sum[0], sum[0]));
sum[0] = _mm_add_ss(sum[0], _mm_shuffle_ps(sum[0], sum[0], 0x55)); sum[0] = _mm_add_ss(sum[0], _mm_shuffle_ps(sum[0], sum[0], 0x55));
@ -303,8 +303,8 @@ static inline void delay_convolve_run_sse(float *buffer, uint32_t *pos,
w += 4; w += 4;
if (w >= n_buffer) { if (w >= n_buffer) {
w -= n_buffer; w -= n_buffer;
t[0] = _mm_loadu_ps(&buffer[n_buffer]); t[0] = _mm_load_ps(&buffer[n_buffer]);
_mm_storeu_ps(&buffer[0], t[0]); _mm_store_ps(&buffer[0], t[0]);
} }
} }
for(; n < n_samples; n++) { for(; n < n_samples; n++) {
@ -326,8 +326,8 @@ static inline void delay_convolve_run_sse(float *buffer, uint32_t *pos,
w += 4; w += 4;
if (w >= n_buffer) { if (w >= n_buffer) {
w -= n_buffer; w -= n_buffer;
t[0] = _mm_loadu_ps(&buffer[n_buffer]); t[0] = _mm_load_ps(&buffer[n_buffer]);
_mm_storeu_ps(&buffer[0], t[0]); _mm_store_ps(&buffer[0], t[0]);
} }
} }
for(; n < n_samples; n++) { for(; n < n_samples; n++) {

6
spa/plugins/audioconvert/channelmix-ops.c
View file

@ -775,7 +775,11 @@ int channelmix_init(struct channelmix *mix)
mix->delay = (uint32_t)(mix->rear_delay * mix->freq / 1000.0f); mix->delay = (uint32_t)(mix->rear_delay * mix->freq / 1000.0f);
mix->func_name = info->name; mix->func_name = info->name;
spa_zero(mix->taps); spa_zero(mix->taps_mem);
mix->taps = SPA_PTR_ALIGN(mix->taps_mem, CHANNELMIX_OPS_MAX_ALIGN, float);
mix->buffer[0] = SPA_PTR_ALIGN(&mix->buffer_mem[0], CHANNELMIX_OPS_MAX_ALIGN, float);
mix->buffer[1] = SPA_PTR_ALIGN(&mix->buffer_mem[2*BUFFER_SIZE], CHANNELMIX_OPS_MAX_ALIGN, float);
if (mix->hilbert_taps > 0) { if (mix->hilbert_taps > 0) {
mix->n_taps = SPA_CLAMP(mix->hilbert_taps, 15u, MAX_TAPS) | 1; mix->n_taps = SPA_CLAMP(mix->hilbert_taps, 15u, MAX_TAPS) | 1;
blackman_window(mix->taps, mix->n_taps); blackman_window(mix->taps, mix->n_taps);

10
spa/plugins/audioconvert/channelmix-ops.h
View file

@ -25,6 +25,8 @@
#define BUFFER_SIZE 4096 #define BUFFER_SIZE 4096
#define MAX_TAPS 255u #define MAX_TAPS 255u
#define CHANNELMIX_OPS_MAX_ALIGN 16
struct channelmix { struct channelmix {
uint32_t src_chan; uint32_t src_chan;
uint32_t dst_chan; uint32_t dst_chan;
@ -59,10 +61,12 @@ struct channelmix {
uint32_t hilbert_taps; /* to phase shift, 0 disabled */ uint32_t hilbert_taps; /* to phase shift, 0 disabled */
struct lr4 lr4[SPA_AUDIO_MAX_CHANNELS]; struct lr4 lr4[SPA_AUDIO_MAX_CHANNELS];
float buffer[2][BUFFER_SIZE*2 + 16]; float buffer_mem[2 * BUFFER_SIZE*2 + CHANNELMIX_OPS_MAX_ALIGN/4];
float *buffer[2];
uint32_t pos[2]; uint32_t pos[2];
uint32_t delay; uint32_t delay;
float taps[MAX_TAPS]; float taps_mem[MAX_TAPS + CHANNELMIX_OPS_MAX_ALIGN/4];
float *taps;
uint32_t n_taps; uint32_t n_taps;
void (*process) (struct channelmix *mix, void * SPA_RESTRICT dst[], void (*process) (struct channelmix *mix, void * SPA_RESTRICT dst[],
@ -104,8 +108,6 @@ void channelmix_##name##_##arch(struct channelmix *mix, \
void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], \ void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], \
uint32_t n_samples); uint32_t n_samples);
#define CHANNELMIX_OPS_MAX_ALIGN 16
DEFINE_FUNCTION(copy, c); DEFINE_FUNCTION(copy, c);
DEFINE_FUNCTION(f32_n_m, c); DEFINE_FUNCTION(f32_n_m, c);
DEFINE_FUNCTION(f32_1_2, c); DEFINE_FUNCTION(f32_1_2, c);