diff --git a/spa/plugins/audioconvert/fmt-ops-avx2.c b/spa/plugins/audioconvert/fmt-ops-avx2.c index 69970d6ee..6350867b9 100644 --- a/spa/plugins/audioconvert/fmt-ops-avx2.c +++ b/spa/plugins/audioconvert/fmt-ops-avx2.c @@ -490,9 +490,9 @@ conv_f32d_to_s32_1s_avx2(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(S24_SCALE); - __m128 int_max = _mm_set1_ps(S24_MAX); - __m128 int_min = _mm_set1_ps(S24_MIN); + __m128 scale = _mm_set1_ps(S25_SCALE); + __m128 int_max = _mm_set1_ps(S25_MAX); + __m128 int_min = _mm_set1_ps(S25_MIN); if (SPA_IS_ALIGNED(s0, 16)) unrolled = n_samples & ~3; @@ -503,7 +503,7 @@ conv_f32d_to_s32_1s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R in[0] = _mm_mul_ps(_mm_load_ps(&s0[n]), scale); in[0] = _MM_CLAMP_PS(in[0], int_min, int_max); out[0] = _mm_cvtps_epi32(in[0]); - out[0] = _mm_slli_epi32(out[0], 8); + out[0] = _mm_slli_epi32(out[0], 7); 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)); @@ -518,7 +518,7 @@ conv_f32d_to_s32_1s_avx2(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_CLAMP_SS(in[0], int_min, int_max); - *d = _mm_cvtss_si32(in[0]) << 8; + *d = _mm_cvtss_si32(in[0]) << 7; d += n_channels; } } @@ -538,12 +538,12 @@ conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R uint32_t n, unrolled; __m256 in[2]; __m256i out[2], t[2]; - __m256 scale = _mm256_set1_ps(S24_SCALE); - __m256 int_min = _mm256_set1_ps(S24_MIN); - __m256 int_max = _mm256_set1_ps(S24_MAX); + __m256 scale = _mm256_set1_ps(S25_SCALE); + __m256 int_min = _mm256_set1_ps(S25_MIN); + __m256 int_max = _mm256_set1_ps(S25_MAX); if (SPA_IS_ALIGNED(s0, 32) && - SPA_IS_ALIGNED(s1, 32)) + SPA_IS_ALIGNED(s1, 32)) unrolled = n_samples & ~7; else unrolled = 0; @@ -557,8 +557,8 @@ conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R 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 */ - out[0] = _mm256_slli_epi32(out[0], 8); - out[1] = _mm256_slli_epi32(out[1], 8); + out[0] = _mm256_slli_epi32(out[0], 7); + out[1] = _mm256_slli_epi32(out[1], 7); t[0] = _mm256_unpacklo_epi32(out[0], out[1]); /* a0 b0 a1 b1 a4 b4 a5 b5 */ t[1] = _mm256_unpackhi_epi32(out[0], out[1]); /* a2 b2 a3 b3 a6 b6 a7 b7 */ @@ -587,9 +587,9 @@ conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R for(; n < n_samples; n++) { __m128 in[2]; __m128i out[2]; - __m128 scale = _mm_set1_ps(S24_SCALE); - __m128 int_min = _mm_set1_ps(S24_MIN); - __m128 int_max = _mm_set1_ps(S24_MAX); + __m128 scale = _mm_set1_ps(S25_SCALE); + __m128 int_min = _mm_set1_ps(S25_MIN); + __m128 int_max = _mm_set1_ps(S25_MAX); in[0] = _mm_load_ss(&s0[n]); in[1] = _mm_load_ss(&s1[n]); @@ -599,7 +599,7 @@ conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R in[0] = _mm_mul_ps(in[0], scale); in[0] = _MM_CLAMP_PS(in[0], int_min, int_max); out[0] = _mm_cvtps_epi32(in[0]); - out[0] = _mm_slli_epi32(out[0], 8); + out[0] = _mm_slli_epi32(out[0], 7); _mm_storel_epi64((__m128i*)d, out[0]); d += n_channels; } @@ -614,14 +614,14 @@ conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R uint32_t n, unrolled; __m256 in[4]; __m256i out[4], t[4]; - __m256 scale = _mm256_set1_ps(S24_SCALE); - __m256 int_min = _mm256_set1_ps(S24_MIN); - __m256 int_max = _mm256_set1_ps(S24_MAX); + __m256 scale = _mm256_set1_ps(S25_SCALE); + __m256 int_min = _mm256_set1_ps(S25_MIN); + __m256 int_max = _mm256_set1_ps(S25_MAX); if (SPA_IS_ALIGNED(s0, 32) && - SPA_IS_ALIGNED(s1, 32) && - SPA_IS_ALIGNED(s2, 32) && - SPA_IS_ALIGNED(s3, 32)) + SPA_IS_ALIGNED(s1, 32) && + SPA_IS_ALIGNED(s2, 32) && + SPA_IS_ALIGNED(s3, 32)) unrolled = n_samples & ~7; else unrolled = 0; @@ -641,10 +641,10 @@ conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R out[1] = _mm256_cvtps_epi32(in[1]); /* b0 b1 b2 b3 b4 b5 b6 b7 */ out[2] = _mm256_cvtps_epi32(in[2]); /* c0 c1 c2 c3 c4 c5 c6 c7 */ out[3] = _mm256_cvtps_epi32(in[3]); /* d0 d1 d2 d3 d4 d5 d6 d7 */ - out[0] = _mm256_slli_epi32(out[0], 8); - out[1] = _mm256_slli_epi32(out[1], 8); - out[2] = _mm256_slli_epi32(out[2], 8); - out[3] = _mm256_slli_epi32(out[3], 8); + out[0] = _mm256_slli_epi32(out[0], 7); + out[1] = _mm256_slli_epi32(out[1], 7); + out[2] = _mm256_slli_epi32(out[2], 7); + out[3] = _mm256_slli_epi32(out[3], 7); t[0] = _mm256_unpacklo_epi32(out[0], out[1]); /* a0 b0 a1 b1 a4 b4 a5 b5 */ t[1] = _mm256_unpackhi_epi32(out[0], out[1]); /* a2 b2 a3 b3 a6 b6 a7 b7 */ @@ -669,9 +669,9 @@ conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R for(; n < n_samples; n++) { __m128 in[4]; __m128i out[4]; - __m128 scale = _mm_set1_ps(S24_SCALE); - __m128 int_min = _mm_set1_ps(S24_MIN); - __m128 int_max = _mm_set1_ps(S24_MAX); + __m128 scale = _mm_set1_ps(S25_SCALE); + __m128 int_min = _mm_set1_ps(S25_MIN); + __m128 int_max = _mm_set1_ps(S25_MAX); in[0] = _mm_load_ss(&s0[n]); in[1] = _mm_load_ss(&s1[n]); @@ -685,7 +685,7 @@ conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_R in[0] = _mm_mul_ps(in[0], scale); in[0] = _MM_CLAMP_PS(in[0], int_min, int_max); out[0] = _mm_cvtps_epi32(in[0]); - out[0] = _mm_slli_epi32(out[0], 8); + out[0] = _mm_slli_epi32(out[0], 7); _mm_storeu_si128((__m128i*)d, out[0]); d += n_channels; } diff --git a/spa/plugins/audioconvert/fmt-ops-sse2.c b/spa/plugins/audioconvert/fmt-ops-sse2.c index 4aca0e9cd..7102c9df6 100644 --- a/spa/plugins/audioconvert/fmt-ops-sse2.c +++ b/spa/plugins/audioconvert/fmt-ops-sse2.c @@ -381,9 +381,9 @@ 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(S24_SCALE); - __m128 int_min = _mm_set1_ps(S24_MIN); - __m128 int_max = _mm_set1_ps(S24_MAX); + __m128 scale = _mm_set1_ps(S25_SCALE); + __m128 int_min = _mm_set1_ps(S25_MIN); + __m128 int_max = _mm_set1_ps(S25_MAX); if (SPA_IS_ALIGNED(s0, 16)) unrolled = n_samples & ~3; @@ -394,7 +394,7 @@ conv_f32d_to_s32_1s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R in[0] = _mm_mul_ps(_mm_load_ps(&s0[n]), scale); in[0] = _MM_CLAMP_PS(in[0], int_min, int_max); out[0] = _mm_cvtps_epi32(in[0]); - out[0] = _mm_slli_epi32(out[0], 8); + out[0] = _mm_slli_epi32(out[0], 7); 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)); @@ -409,7 +409,7 @@ 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_CLAMP_SS(in[0], int_min, int_max); - *d = _mm_cvtss_si32(in[0]) << 8; + *d = _mm_cvtss_si32(in[0]) << 7; d += n_channels; } } @@ -423,12 +423,12 @@ 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(S24_SCALE); - __m128 int_min = _mm_set1_ps(S24_MIN); - __m128 int_max = _mm_set1_ps(S24_MAX); + __m128 scale = _mm_set1_ps(S25_SCALE); + __m128 int_min = _mm_set1_ps(S25_MIN); + __m128 int_max = _mm_set1_ps(S25_MAX); if (SPA_IS_ALIGNED(s0, 16) && - SPA_IS_ALIGNED(s1, 16)) + SPA_IS_ALIGNED(s1, 16)) unrolled = n_samples & ~3; else unrolled = 0; @@ -442,8 +442,8 @@ conv_f32d_to_s32_2s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R out[0] = _mm_cvtps_epi32(in[0]); out[1] = _mm_cvtps_epi32(in[1]); - out[0] = _mm_slli_epi32(out[0], 8); - out[1] = _mm_slli_epi32(out[1], 8); + out[0] = _mm_slli_epi32(out[0], 7); + out[1] = _mm_slli_epi32(out[1], 7); t[0] = _mm_unpacklo_epi32(out[0], out[1]); t[1] = _mm_unpackhi_epi32(out[0], out[1]); @@ -463,7 +463,7 @@ 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_CLAMP_PS(in[0], int_min, int_max); out[0] = _mm_cvtps_epi32(in[0]); - out[0] = _mm_slli_epi32(out[0], 8); + out[0] = _mm_slli_epi32(out[0], 7); _mm_storel_epi64((__m128i*)d, out[0]); d += n_channels; } @@ -478,14 +478,14 @@ 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(S24_SCALE); - __m128 int_min = _mm_set1_ps(S24_MIN); - __m128 int_max = _mm_set1_ps(S24_MAX); + __m128 scale = _mm_set1_ps(S25_SCALE); + __m128 int_min = _mm_set1_ps(S25_MIN); + __m128 int_max = _mm_set1_ps(S25_MAX); if (SPA_IS_ALIGNED(s0, 16) && - SPA_IS_ALIGNED(s1, 16) && - SPA_IS_ALIGNED(s2, 16) && - SPA_IS_ALIGNED(s3, 16)) + SPA_IS_ALIGNED(s1, 16) && + SPA_IS_ALIGNED(s2, 16) && + SPA_IS_ALIGNED(s3, 16)) unrolled = n_samples & ~3; else unrolled = 0; @@ -507,10 +507,10 @@ conv_f32d_to_s32_4s_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_R out[1] = _mm_cvtps_epi32(in[1]); out[2] = _mm_cvtps_epi32(in[2]); out[3] = _mm_cvtps_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); + out[0] = _mm_slli_epi32(out[0], 7); + out[1] = _mm_slli_epi32(out[1], 7); + out[2] = _mm_slli_epi32(out[2], 7); + out[3] = _mm_slli_epi32(out[3], 7); _mm_storeu_si128((__m128i*)(d + 0*n_channels), out[0]); _mm_storeu_si128((__m128i*)(d + 1*n_channels), out[1]); @@ -531,7 +531,7 @@ 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_CLAMP_PS(in[0], int_min, int_max); out[0] = _mm_cvtps_epi32(in[0]); - out[0] = _mm_slli_epi32(out[0], 8); + out[0] = _mm_slli_epi32(out[0], 7); _mm_storeu_si128((__m128i*)d, out[0]); d += n_channels; } @@ -620,6 +620,7 @@ void conv_noise_tri_hf_sse2(struct convert *conv, float *noise, uint32_t n_sampl _mm_store_si128((__m128i*)p, old[0]); } +// FIXME: this function is not covered with tests. static void conv_f32d_to_s32_1s_noise_sse2(struct convert *conv, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src, float *noise, uint32_t n_channels, uint32_t n_samples) @@ -629,9 +630,9 @@ conv_f32d_to_s32_1s_noise_sse2(struct convert *conv, void * SPA_RESTRICT dst, co uint32_t n, unrolled; __m128 in[1]; __m128i out[4]; - __m128 scale = _mm_set1_ps(S24_SCALE); - __m128 int_min = _mm_set1_ps(S24_MIN); - __m128 int_max = _mm_set1_ps(S24_MAX); + __m128 scale = _mm_set1_ps(S25_SCALE); + __m128 int_min = _mm_set1_ps(S25_MIN); + __m128 int_max = _mm_set1_ps(S25_MAX); if (SPA_IS_ALIGNED(s, 16)) unrolled = n_samples & ~3; @@ -643,7 +644,7 @@ conv_f32d_to_s32_1s_noise_sse2(struct convert *conv, void * SPA_RESTRICT dst, co in[0] = _mm_add_ps(in[0], _mm_load_ps(&noise[n])); in[0] = _MM_CLAMP_PS(in[0], int_min, int_max); out[0] = _mm_cvtps_epi32(in[0]); - out[0] = _mm_slli_epi32(out[0], 8); + out[0] = _mm_slli_epi32(out[0], 7); 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)); @@ -659,7 +660,7 @@ conv_f32d_to_s32_1s_noise_sse2(struct convert *conv, void * SPA_RESTRICT dst, co in[0] = _mm_mul_ss(in[0], scale); in[0] = _mm_add_ss(in[0], _mm_load_ss(&noise[n])); in[0] = _MM_CLAMP_SS(in[0], int_min, int_max); - *d = _mm_cvtss_si32(in[0]) << 8; + *d = _mm_cvtss_si32(in[0]) << 7; d += n_channels; } } diff --git a/spa/plugins/audioconvert/fmt-ops.h b/spa/plugins/audioconvert/fmt-ops.h index 17445a063..e9152cb31 100644 --- a/spa/plugins/audioconvert/fmt-ops.h +++ b/spa/plugins/audioconvert/fmt-ops.h @@ -123,8 +123,8 @@ #define S32_MAX (S24_MAX * 256) #define S32_TO_F32(v) ITOF(int32_t, S32_TO_S24_32(v), S24_SCALE, 0.0f) #define S32S_TO_F32(v) S32_TO_F32(bswap_32(v)) -#define F32_TO_S32(v) S24_32_TO_S32(F32_TO_S24_32(v)) -#define F32_TO_S32_D(v,d) S24_32_TO_S32(F32_TO_S24_32_D(v,d)) +#define F32_TO_S32_D(v,d) S25_32_TO_S32(F32_TO_S25_32_D(v,d)) +#define F32_TO_S32(v) F32_TO_S32_D(v, 0.0f) #define F32_TO_S32S(v) bswap_32(F32_TO_S32(v)) #define F32_TO_S32S_D(v,d) bswap_32(F32_TO_S32_D(v,d)) diff --git a/spa/plugins/audioconvert/test-fmt-ops.c b/spa/plugins/audioconvert/test-fmt-ops.c index ee5ce988c..e91b28d81 100644 --- a/spa/plugins/audioconvert/test-fmt-ops.c +++ b/spa/plugins/audioconvert/test-fmt-ops.c @@ -298,16 +298,15 @@ static void test_f32_s32(void) 1.0f/0x40000000, -1.0f/0x40000000, 1.0f/0x80000000, -1.0f/0x80000000, 1.0f/0x100000000, -1.0f/0x100000000, 1.0f/0x200000000, -1.0f/0x200000000, }; - static const int32_t out[] = { 0x00000000, 0x7fffff00, 0x80000000, - 0x40000000, 0xc0000000, 0x7fffff00, 0x80000000, 0x00000100, - 0xffffff00, 0x00000100, 0xffffff00, 0x00000000, 0x00000000, + static const int32_t out[] = { 0x00000000, 0x7fffff80, 0x80000000, + 0x40000000, 0xc0000000, 0x7fffff80, 0x80000000, 0x00000100, + 0xffffff00, 0x00000100, 0xffffff00, 0x00000080, 0xffffff80, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, }; - run_test("test_f32_s32", in, sizeof(in[0]), out, sizeof(out[0]), SPA_N_ELEMENTS(out), true, true, conv_f32_to_s32_c); run_test("test_f32d_s32", in, sizeof(in[0]), out, sizeof(out[0]), SPA_N_ELEMENTS(out), @@ -665,21 +664,16 @@ static void test_lossless_s25_32_to_s32_to_f32_to_s32_to_s25_32_XFAIL(void) spa_assert_se(max_abs_err == 1); } -static void test_lossless_s25_32_to_f32_to_s32_to_s25_32_XFAIL(void) +static void test_lossless_s25_32_to_f32_to_s32_to_s25_32(void) { int32_t i; - int all_lossless = 1; - int32_t max_abs_err = -1; fprintf(stderr, "test %s:\n", __func__); for (i = S25_MIN; i <= S25_MAX; i+=1) { float v = S25_32_TO_F32(i); int32_t t = S32_TO_S25_32(F32_TO_S32(v)); - all_lossless &= i == t; - max_abs_err = SPA_MAX(max_abs_err, SPA_ABS(i - t)); + spa_assert_se(i == t); } - spa_assert_se(!all_lossless); - spa_assert_se(max_abs_err == 1); } static void test_lossless_s32(void) @@ -883,7 +877,7 @@ int main(int argc, char *argv[]) test_lossless_s25_32_to_f32_to_s25_32(); test_lossless_s25_32_to_s32_to_f32_to_s25_32_XFAIL(); test_lossless_s25_32_to_s32_to_f32_to_s32_to_s25_32_XFAIL(); - test_lossless_s25_32_to_f32_to_s32_to_s25_32_XFAIL(); + test_lossless_s25_32_to_f32_to_s32_to_s25_32(); test_lossless_s32(); test_lossless_s32_lossless_subset(); test_lossless_u32();