mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2025-10-31 22:25:38 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

fmt: use faster s24 to f32 conversion

Browse source
This commit is contained in:
Wim Taymans 2018-11-21 15:51:03 +01:00
parent 25c81e8c34
commit 13010d5d2e
2 changed files with 22 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

17
spa/plugins/audioconvert/fmt-ops-sse.c
View file

@ -111,8 +111,12 @@ conv_s24_to_f32d_1_sse(void *data, int n_dst, void *dst[n_dst], const void *src,
const uint8_t *s = src;
float **d = (float **) dst;
float *d0 = d[0];
int n, n_samples, unrolled;
int n = 0, n_samples, unrolled;
__m128i in;
union {
__m128i in;
uint8_t b[16];
} b;
__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
n_samples = n_bytes / (3 * n_dst);
@ -120,11 +124,12 @@ conv_s24_to_f32d_1_sse(void *data, int n_dst, void *dst[n_dst], const void *src,
unrolled = n_samples / 4;
n_samples = n_samples & 3;
for(n = 0; unrolled--; n += 4) {
in = _mm_set_epi32(READ24(&s[9*n_dst]),
READ24(&s[6*n_dst]),
READ24(&s[3*n_dst]),
READ24(s));
for(; unrolled--; n += 4) {
memcpy(&b.b[1], &s[0], 3);
memcpy(&b.b[5], &s[3], 3);
memcpy(&b.b[9], &s[6], 3);
memcpy(&b.b[13], &s[9], 3);
in = _mm_srai_epi32(b.in, 8);
out = _mm_cvtepi32_ps(in);
out = _mm_mul_ps(out, factor);
_mm_storeu_ps(&d0[n], out);

28
spa/plugins/audioconvert/fmt-ops.c
View file

@ -46,27 +46,21 @@
#define S32_MAX 2147483647
#define S32_SCALE 2147483647
union s24_data {
struct {
#if __BYTE_ORDER == __LITTLE_ENDIAN
uint16_t lo;
int16_t hi;
#else
int16_t hi;
uint16_t lo;
#endif
} s24;
int32_t value;
};
static inline int32_t read_s24(const void *src)
{
union s24_data d;
const uint8_t *s = src;
d.s24.hi = ((int8_t*)s)[2];
d.s24.lo = *((uint16_t*)s);
return d.value;
union {
int8_t v[4];
int32_t value;
} d;
#if __BYTE_ORDER == __LITTLE_ENDIAN
memcpy(&d.v[1], src, 3);
#else
memcpy(&d.v[0], src, 3);
#endif
return d.value >> 8;
}
#define READ24(s) read_s24(s)
#if defined (__SSE__)