mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2025-11-03 09:01:54 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

audioconvert: use 24 bits for 32 formats

Browse source 
The float only preserves 24 bits so use this. Otherwise we get overflows
and errors in clang.
This commit is contained in:
Wim Taymans 2022-07-07 18:38:32 +02:00
parent 43b964ea26
commit b8a4bf880f
4 changed files with 92 additions and 42 deletions
Download patch file Download diff file Expand all files Collapse all files

48
spa/plugins/audioconvert/fmt-ops-sse2.c
View file

@ -338,7 +338,7 @@ conv_s32_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
float *d0 = dst[0];
uint32_t n, unrolled;
__m128i in;
__m128 out, factor = _mm_set1_ps(1.0f / S32_SCALE);
__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
if (SPA_IS_ALIGNED(d0, 16))
unrolled = n_samples & ~3;
@ -350,13 +350,14 @@ conv_s32_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
s[1*n_channels],
s[2*n_channels],
s[3*n_channels]);
in = _mm_srai_epi32(in, 8);
out = _mm_cvtepi32_ps(in);
out = _mm_mul_ps(out, factor);
_mm_store_ps(&d0[n], out);
s += 4*n_channels;
}
for(; n < n_samples; n++) {
out = _mm_cvtsi32_ss(factor, s[0]);
out = _mm_cvtsi32_ss(factor, s[0]>>8);
out = _mm_mul_ss(out, factor);
_mm_store_ss(&d0[n], out);
s += n_channels;
@ -383,8 +384,9 @@ conv_f32d_to_s32_1s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
uint32_t n, unrolled;
__m128 in[1];
__m128i out[4];
__m128 scale = _mm_set1_ps(S32_SCALE);
__m128 int_max = _mm_set1_ps(S32_MAX);
__m128 scale = _mm_set1_ps(S24_SCALE);
__m128 int_min = _mm_set1_ps(S24_MIN);
__m128 int_max = _mm_set1_ps(S24_MAX);
if (SPA_IS_ALIGNED(s0, 16))
unrolled = n_samples & ~3;
@ -394,7 +396,9 @@ conv_f32d_to_s32_1s_sse2(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_max);
in[0] = _mm_max_ps(in[0], int_min);
out[0] = _mm_cvttps_epi32(in[0]);
out[0] = _mm_slli_epi32(out[0], 8);
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));
out[3] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(2, 1, 0, 3));
@ -409,6 +413,7 @@ conv_f32d_to_s32_1s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[0] = _mm_load_ss(&s0[n]);
in[0] = _mm_mul_ss(in[0], scale);
in[0] = _mm_min_ss(in[0], int_max);
in[0] = _mm_max_ss(in[0], int_min);
*d = _mm_cvtss_si32(in[0]);
d += n_channels;
}
@ -423,8 +428,9 @@ conv_f32d_to_s32_2s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
uint32_t n, unrolled;
__m128 in[2];
__m128i out[2], t[2];
__m128 scale = _mm_set1_ps(S32_SCALE);
__m128 int_max = _mm_set1_ps(S32_MAX);
__m128 scale = _mm_set1_ps(S24_SCALE);
__m128 int_min = _mm_set1_ps(S24_MIN);
__m128 int_max = _mm_set1_ps(S24_MAX);
if (SPA_IS_ALIGNED(s0, 16) &&
SPA_IS_ALIGNED(s1, 16))
@ -437,10 +443,14 @@ conv_f32d_to_s32_2s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[1] = _mm_mul_ps(_mm_load_ps(&s1[n]), scale);
in[0] = _mm_min_ps(in[0], int_max);
in[0] = _mm_max_ps(in[0], int_min);
in[1] = _mm_min_ps(in[1], int_max);
in[1] = _mm_max_ps(in[1], int_min);
out[0] = _mm_cvttps_epi32(in[0]);
out[1] = _mm_cvttps_epi32(in[1]);
out[0] = _mm_slli_epi32(out[0], 8);
out[1] = _mm_slli_epi32(out[1], 8);
t[0] = _mm_unpacklo_epi32(out[0], out[1]);
t[1] = _mm_unpackhi_epi32(out[0], out[1]);
@ -460,6 +470,7 @@ conv_f32d_to_s32_2s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[0] = _mm_mul_ps(in[0], scale);
in[0] = _mm_min_ps(in[0], int_max);
out[0] = _mm_cvttps_epi32(in[0]);
out[0] = _mm_slli_epi32(out[0], 8);
_mm_storel_epi64((__m128i*)d, out[0]);
d += n_channels;
}
@ -474,8 +485,9 @@ conv_f32d_to_s32_4s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
uint32_t n, unrolled;
__m128 in[4];
__m128i out[4];
__m128 scale = _mm_set1_ps(S32_SCALE);
__m128 int_max = _mm_set1_ps(S32_MAX);
__m128 scale = _mm_set1_ps(S24_SCALE);
__m128 int_min = _mm_set1_ps(S24_MIN);
__m128 int_max = _mm_set1_ps(S24_MAX);
if (SPA_IS_ALIGNED(s0, 16) &&
SPA_IS_ALIGNED(s1, 16) &&
@ -495,6 +507,10 @@ conv_f32d_to_s32_4s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[1] = _mm_min_ps(in[1], int_max);
in[2] = _mm_min_ps(in[2], int_max);
in[3] = _mm_min_ps(in[3], int_max);
in[0] = _mm_max_ps(in[0], int_min);
in[1] = _mm_max_ps(in[1], int_min);
in[2] = _mm_max_ps(in[2], int_min);
in[3] = _mm_max_ps(in[3], int_min);
_MM_TRANSPOSE4_PS(in[0], in[1], in[2], in[3]);
@ -502,6 +518,10 @@ conv_f32d_to_s32_4s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
out[1] = _mm_cvttps_epi32(in[1]);
out[2] = _mm_cvttps_epi32(in[2]);
out[3] = _mm_cvttps_epi32(in[3]);
out[0] = _mm_slli_epi32(out[0], 8);
out[1] = _mm_slli_epi32(out[1], 8);
out[2] = _mm_slli_epi32(out[2], 8);
out[3] = _mm_slli_epi32(out[3], 8);
_mm_storeu_si128((__m128i*)(d + 0*n_channels), out[0]);
_mm_storeu_si128((__m128i*)(d + 1*n_channels), out[1]);
@ -521,7 +541,9 @@ conv_f32d_to_s32_4s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R
in[0] = _mm_mul_ps(in[0], scale);
in[0] = _mm_min_ps(in[0], int_max);
in[0] = _mm_max_ps(in[0], int_min);
out[0] = _mm_cvttps_epi32(in[0]);
out[0] = _mm_slli_epi32(out[0], 8);
_mm_storeu_si128((__m128i*)d, out[0]);
d += n_channels;
}
@ -577,8 +599,9 @@ conv_f32d_to_s32_1s_dither_sse2(struct convert *conv, void * SPA_RESTRICT dst, c
uint32_t n, unrolled;
__m128 in[1];
__m128i out[4];
__m128 scale = _mm_set1_ps(S32_SCALE);
__m128 int_max = _mm_set1_ps(S32_MAX);
__m128 scale = _mm_set1_ps(S24_SCALE);
__m128 int_min = _mm_set1_ps(S24_MIN);
__m128 int_max = _mm_set1_ps(S24_MAX);
if (SPA_IS_ALIGNED(s, 16))
unrolled = n_samples & ~3;
@ -589,7 +612,9 @@ conv_f32d_to_s32_1s_dither_sse2(struct convert *conv, void * SPA_RESTRICT dst, c
in[0] = _mm_mul_ps(_mm_load_ps(&s[n]), scale);
in[0] = _mm_add_ps(in[0], _mm_load_ps(&dither[n]));
in[0] = _mm_min_ps(in[0], int_max);
in[0] = _mm_max_ps(in[0], int_min);
out[0] = _mm_cvttps_epi32(in[0]);
out[0] = _mm_slli_epi32(out[0], 8);
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));
out[3] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(2, 1, 0, 3));
@ -605,7 +630,8 @@ conv_f32d_to_s32_1s_dither_sse2(struct convert *conv, void * SPA_RESTRICT dst, c
in[0] = _mm_mul_ss(in[0], scale);
in[0] = _mm_add_ss(in[0], _mm_load_ss(&dither[n]));
in[0] = _mm_min_ss(in[0], int_max);
*d = _mm_cvtss_si32(in[0]);
in[0] = _mm_max_ss(in[0], int_min);
*d = _mm_cvtss_si32(in[0]) << 8;
d += n_channels;
}
}