mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2025-11-07 13:30:09 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

audioconvert: improve format conversion

Browse source 
Make dither noise as a value between -0.5 and 0.5 and add this
to the scaled samples.
For this, we first need to do the scaling and then the CLAMP to
the target depth. This optimizes to the same code but allows us
to avoid under and overflows when we add the dither noise.

Add more dithering methods.

Expose a dither.method property on audioconvert. Disable dither when
the target depth > 16.
This commit is contained in:
Wim Taymans 2022-06-29 14:10:15 +02:00
parent d23b96b033
commit 938f2b123e
6 changed files with 408 additions and 171 deletions
Download patch file Download diff file Expand all files Collapse all files

30
spa/plugins/audioconvert/fmt-ops-avx2.c
View file

@ -550,7 +550,7 @@ conv_f32d_to_s32_1s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
__m128 in[1];
__m128i out[4];
__m128 scale = _mm_set1_ps(S32_SCALE);
__m128 int_min = _mm_set1_ps(S32_MIN);
__m128 int_max = _mm_set1_ps(S32_MAX);
if (SPA_IS_ALIGNED(s0, 16))
unrolled = n_samples & ~3;
@ -559,7 +559,7 @@ conv_f32d_to_s32_1s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
for(n = 0; n < unrolled; n += 4) {
in[0] = _mm_mul_ps(_mm_load_ps(&s0[n]), scale);
in[0] = _mm_min_ps(in[0], int_min);
in[0] = _mm_min_ps(in[0], int_max);
out[0] = _mm_cvtps_epi32(in[0]);
out[1] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(0, 3, 2, 1));
out[2] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(1, 0, 3, 2));
@ -574,7 +574,7 @@ conv_f32d_to_s32_1s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
for(; n < n_samples; n++) {
in[0] = _mm_load_ss(&s0[n]);
in[0] = _mm_mul_ss(in[0], scale);
in[0] = _mm_min_ss(in[0], int_min);
in[0] = _mm_min_ss(in[0], int_max);
*d = _mm_cvtss_si32(in[0]);
d += n_channels;
}
@ -590,7 +590,7 @@ conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
__m256 in[2];
__m256i out[2], t[2];
__m256 scale = _mm256_set1_ps(S32_SCALE);
__m256 int_min = _mm256_set1_ps(S32_MIN);
__m256 int_max = _mm256_set1_ps(S32_MAX);
if (SPA_IS_ALIGNED(s0, 32) &&
SPA_IS_ALIGNED(s1, 32))
@ -602,8 +602,8 @@ conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[0] = _mm256_mul_ps(_mm256_load_ps(&s0[n]), scale);
in[1] = _mm256_mul_ps(_mm256_load_ps(&s1[n]), scale);
in[0] = _mm256_min_ps(in[0], int_min);
in[1] = _mm256_min_ps(in[1], int_min);
in[0] = _mm256_min_ps(in[0], int_max);
in[1] = _mm256_min_ps(in[1], int_max);
out[0] = _mm256_cvtps_epi32(in[0]); /* a0 a1 a2 a3 a4 a5 a6 a7 */
out[1] = _mm256_cvtps_epi32(in[1]); /* b0 b1 b2 b3 b4 b5 b6 b7 */
@ -636,7 +636,7 @@ conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
__m128 in[2];
__m128i out[2];
__m128 scale = _mm_set1_ps(S32_SCALE);
__m128 int_min = _mm_set1_ps(S32_MIN);
__m128 int_max = _mm_set1_ps(S32_MAX);
in[0] = _mm_load_ss(&s0[n]);
in[1] = _mm_load_ss(&s1[n]);
@ -644,7 +644,7 @@ conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[0] = _mm_unpacklo_ps(in[0], in[1]);
in[0] = _mm_mul_ps(in[0], scale);
in[0] = _mm_min_ps(in[0], int_min);
in[0] = _mm_min_ps(in[0], int_max);
out[0] = _mm_cvtps_epi32(in[0]);
_mm_storel_epi64((__m128i*)d, out[0]);
d += n_channels;
@ -661,7 +661,7 @@ conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
__m256 in[4];
__m256i out[4], t[4];
__m256 scale = _mm256_set1_ps(S32_SCALE);
__m256 int_min = _mm256_set1_ps(S32_MIN);
__m256 int_max = _mm256_set1_ps(S32_MAX);
if (SPA_IS_ALIGNED(s0, 32) &&
SPA_IS_ALIGNED(s1, 32) &&
@ -677,10 +677,10 @@ conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[2] = _mm256_mul_ps(_mm256_load_ps(&s2[n]), scale);
in[3] = _mm256_mul_ps(_mm256_load_ps(&s3[n]), scale);
in[0] = _mm256_min_ps(in[0], int_min);
in[1] = _mm256_min_ps(in[1], int_min);
in[2] = _mm256_min_ps(in[2], int_min);
in[3] = _mm256_min_ps(in[3], int_min);
in[0] = _mm256_min_ps(in[0], int_max);
in[1] = _mm256_min_ps(in[1], int_max);
in[2] = _mm256_min_ps(in[2], int_max);
in[3] = _mm256_min_ps(in[3], int_max);
out[0] = _mm256_cvtps_epi32(in[0]); /* a0 a1 a2 a3 a4 a5 a6 a7 */
out[1] = _mm256_cvtps_epi32(in[1]); /* b0 b1 b2 b3 b4 b5 b6 b7 */
@ -711,7 +711,7 @@ conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
__m128 in[4];
__m128i out[4];
__m128 scale = _mm_set1_ps(S32_SCALE);
__m128 int_min = _mm_set1_ps(S32_MIN);
__m128 int_max = _mm_set1_ps(S32_MAX);
in[0] = _mm_load_ss(&s0[n]);
in[1] = _mm_load_ss(&s1[n]);
@ -723,7 +723,7 @@ conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[0] = _mm_unpacklo_ps(in[0], in[1]);
in[0] = _mm_mul_ps(in[0], scale);
in[0] = _mm_min_ps(in[0], int_min);
in[0] = _mm_min_ps(in[0], int_max);
out[0] = _mm_cvtps_epi32(in[0]);
_mm_storeu_si128((__m128i*)d, out[0]);
d += n_channels;