From c5178658646b3c554062365ffdd992ca66427caf Mon Sep 17 00:00:00 2001 From: Roman Lebedev Date: Fri, 14 Jun 2024 02:00:27 +0300 Subject: [PATCH] audioconvert: somewhat avoid precision loss in S32 to F32 conversion At the very least, we should go through s25_32 intermediate instead of s24_32, to avoid needlessly loosing 1 LSB precision bit. That being said, i suspect it's still not doing the right thing. Why are we silently dropping those 7 LSB bits? Is that really the way to do it? --- spa/plugins/audioconvert/fmt-ops-avx2.c | 42 ++++++++++++------------- spa/plugins/audioconvert/fmt-ops-sse2.c | 6 ++-- spa/plugins/audioconvert/fmt-ops.h | 2 +- spa/plugins/audioconvert/test-fmt-ops.c | 34 ++++++++------------ 4 files changed, 38 insertions(+), 46 deletions(-) diff --git a/spa/plugins/audioconvert/fmt-ops-avx2.c b/spa/plugins/audioconvert/fmt-ops-avx2.c index 6350867b9..8022f8245 100644 --- a/spa/plugins/audioconvert/fmt-ops-avx2.c +++ b/spa/plugins/audioconvert/fmt-ops-avx2.c @@ -316,7 +316,7 @@ conv_s32_to_f32d_4s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA float *d0 = dst[0], *d1 = dst[1], *d2 = dst[2], *d3 = dst[3]; uint32_t n, unrolled; __m256i in[4]; - __m256 out[4], factor = _mm256_set1_ps(1.0f / S24_SCALE); + __m256 out[4], factor = _mm256_set1_ps(1.0f / S25_SCALE); __m256i mask1 = _mm256_setr_epi32(0*n_channels, 1*n_channels, 2*n_channels, 3*n_channels, 4*n_channels, 5*n_channels, 6*n_channels, 7*n_channels); @@ -334,10 +334,10 @@ conv_s32_to_f32d_4s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA in[2] = _mm256_i32gather_epi32((int*)&s[2], mask1, 4); in[3] = _mm256_i32gather_epi32((int*)&s[3], mask1, 4); - in[0] = _mm256_srai_epi32(in[0], 8); - in[1] = _mm256_srai_epi32(in[1], 8); - in[2] = _mm256_srai_epi32(in[2], 8); - in[3] = _mm256_srai_epi32(in[3], 8); + in[0] = _mm256_srai_epi32(in[0], 7); + in[1] = _mm256_srai_epi32(in[1], 7); + in[2] = _mm256_srai_epi32(in[2], 7); + in[3] = _mm256_srai_epi32(in[3], 7); out[0] = _mm256_cvtepi32_ps(in[0]); out[1] = _mm256_cvtepi32_ps(in[1]); @@ -357,11 +357,11 @@ conv_s32_to_f32d_4s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA s += 8*n_channels; } for(; n < n_samples; n++) { - __m128 out[4], factor = _mm_set1_ps(1.0f / S24_SCALE); - out[0] = _mm_cvtsi32_ss(factor, s[0] >> 8); - out[1] = _mm_cvtsi32_ss(factor, s[1] >> 8); - out[2] = _mm_cvtsi32_ss(factor, s[2] >> 8); - out[3] = _mm_cvtsi32_ss(factor, s[3] >> 8); + __m128 out[4], factor = _mm_set1_ps(1.0f / S25_SCALE); + out[0] = _mm_cvtsi32_ss(factor, s[0] >> 7); + out[1] = _mm_cvtsi32_ss(factor, s[1] >> 7); + out[2] = _mm_cvtsi32_ss(factor, s[2] >> 7); + out[3] = _mm_cvtsi32_ss(factor, s[3] >> 7); out[0] = _mm_mul_ss(out[0], factor); out[1] = _mm_mul_ss(out[1], factor); out[2] = _mm_mul_ss(out[2], factor); @@ -382,7 +382,7 @@ conv_s32_to_f32d_2s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA float *d0 = dst[0], *d1 = dst[1]; uint32_t n, unrolled; __m256i in[4]; - __m256 out[4], factor = _mm256_set1_ps(1.0f / S24_SCALE); + __m256 out[4], factor = _mm256_set1_ps(1.0f / S25_SCALE); __m256i mask1 = _mm256_setr_epi32(0*n_channels, 1*n_channels, 2*n_channels, 3*n_channels, 4*n_channels, 5*n_channels, 6*n_channels, 7*n_channels); @@ -396,8 +396,8 @@ conv_s32_to_f32d_2s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA in[0] = _mm256_i32gather_epi32((int*)&s[0], mask1, 4); in[1] = _mm256_i32gather_epi32((int*)&s[1], mask1, 4); - in[0] = _mm256_srai_epi32(in[0], 8); - in[1] = _mm256_srai_epi32(in[1], 8); + in[0] = _mm256_srai_epi32(in[0], 7); + in[1] = _mm256_srai_epi32(in[1], 7); out[0] = _mm256_cvtepi32_ps(in[0]); out[1] = _mm256_cvtepi32_ps(in[1]); @@ -411,9 +411,9 @@ conv_s32_to_f32d_2s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA s += 8*n_channels; } for(; n < n_samples; n++) { - __m128 out[2], factor = _mm_set1_ps(1.0f / S24_SCALE); - out[0] = _mm_cvtsi32_ss(factor, s[0] >> 8); - out[1] = _mm_cvtsi32_ss(factor, s[1] >> 8); + __m128 out[2], factor = _mm_set1_ps(1.0f / S25_SCALE); + out[0] = _mm_cvtsi32_ss(factor, s[0] >> 7); + out[1] = _mm_cvtsi32_ss(factor, s[1] >> 7); out[0] = _mm_mul_ss(out[0], factor); out[1] = _mm_mul_ss(out[1], factor); _mm_store_ss(&d0[n], out[0]); @@ -430,7 +430,7 @@ conv_s32_to_f32d_1s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA float *d0 = dst[0]; uint32_t n, unrolled; __m256i in[2]; - __m256 out[2], factor = _mm256_set1_ps(1.0f / S24_SCALE); + __m256 out[2], factor = _mm256_set1_ps(1.0f / S25_SCALE); __m256i mask1 = _mm256_setr_epi32(0*n_channels, 1*n_channels, 2*n_channels, 3*n_channels, 4*n_channels, 5*n_channels, 6*n_channels, 7*n_channels); @@ -443,8 +443,8 @@ conv_s32_to_f32d_1s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA in[0] = _mm256_i32gather_epi32(&s[0*n_channels], mask1, 4); in[1] = _mm256_i32gather_epi32(&s[8*n_channels], mask1, 4); - in[0] = _mm256_srai_epi32(in[0], 8); - in[1] = _mm256_srai_epi32(in[1], 8); + in[0] = _mm256_srai_epi32(in[0], 7); + in[1] = _mm256_srai_epi32(in[1], 7); out[0] = _mm256_cvtepi32_ps(in[0]); out[1] = _mm256_cvtepi32_ps(in[1]); @@ -458,8 +458,8 @@ conv_s32_to_f32d_1s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA s += 16*n_channels; } for(; n < n_samples; n++) { - __m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE); - out = _mm_cvtsi32_ss(factor, s[0] >> 8); + __m128 out, factor = _mm_set1_ps(1.0f / S25_SCALE); + out = _mm_cvtsi32_ss(factor, s[0] >> 7); out = _mm_mul_ss(out, factor); _mm_store_ss(&d0[n], out); s += n_channels; diff --git a/spa/plugins/audioconvert/fmt-ops-sse2.c b/spa/plugins/audioconvert/fmt-ops-sse2.c index 7102c9df6..3055173f0 100644 --- a/spa/plugins/audioconvert/fmt-ops-sse2.c +++ b/spa/plugins/audioconvert/fmt-ops-sse2.c @@ -335,7 +335,7 @@ 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 / S24_SCALE); + __m128 out, factor = _mm_set1_ps(1.0f / S25_SCALE); if (SPA_IS_ALIGNED(d0, 16)) unrolled = n_samples & ~3; @@ -347,14 +347,14 @@ 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); + in = _mm_srai_epi32(in, 7); 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]>>8); + out = _mm_cvtsi32_ss(factor, s[0]>>7); out = _mm_mul_ss(out, factor); _mm_store_ss(&d0[n], out); s += n_channels; diff --git a/spa/plugins/audioconvert/fmt-ops.h b/spa/plugins/audioconvert/fmt-ops.h index e9152cb31..0a8373f76 100644 --- a/spa/plugins/audioconvert/fmt-ops.h +++ b/spa/plugins/audioconvert/fmt-ops.h @@ -121,7 +121,7 @@ #define S32_MIN (S24_MIN * 256) #define S32_MAX (S24_MAX * 256) -#define S32_TO_F32(v) ITOF(int32_t, S32_TO_S24_32(v), S24_SCALE, 0.0f) +#define S32_TO_F32(v) ITOF(int32_t, S32_TO_S25_32(v), S25_SCALE, 0.0f) #define S32S_TO_F32(v) S32_TO_F32(bswap_32(v)) #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) diff --git a/spa/plugins/audioconvert/test-fmt-ops.c b/spa/plugins/audioconvert/test-fmt-ops.c index e91b28d81..58c38e836 100644 --- a/spa/plugins/audioconvert/test-fmt-ops.c +++ b/spa/plugins/audioconvert/test-fmt-ops.c @@ -335,10 +335,12 @@ static void test_s32_f32(void) 0x80000100, 0x40000000, 0xc0000000, 0x0080, 0xFFFFFF80, 0x0100, 0xFFFFFF00, 0x0200, 0xFFFFFE00 }; - static const float out[] = { 0.e+00f, 9.9999988079071044921875e-01f, -1.e+00f, + + static const float out[] = { 0.e+00f, 9.99999940395355224609375e-01f, -1.e+00f, 9.9999988079071044921875e-01f, -9.9999988079071044921875e-01f, 5.e-01f, - -5.e-01f, 0.e+00f, -1.1920928955078125e-07f, 1.1920928955078125e-07f, - -1.1920928955078125e-07f, 2.384185791015625e-07f, -2.384185791015625e-07f + -5.e-01f, 5.9604644775390625e-08f, -5.9604644775390625e-08f, + 1.1920928955078125e-07f, -1.1920928955078125e-07f, + 2.384185791015625e-07f, -2.384185791015625e-07f }; run_test("test_s32_f32d", in, sizeof(in[0]), out, sizeof(out[0]), SPA_N_ELEMENTS(out), @@ -630,38 +632,28 @@ static void test_lossless_s25_32_to_f32_to_s25_32(void) } } -static void test_lossless_s25_32_to_s32_to_f32_to_s25_32_XFAIL(void) +static void test_lossless_s25_32_to_s32_to_f32_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 = S32_TO_F32(S25_32_TO_S32(i)); int32_t t = F32_TO_S25_32(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_s25_32_to_s32_to_f32_to_s32_to_s25_32_XFAIL(void) +static void test_lossless_s25_32_to_s32_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 = S32_TO_F32(S25_32_TO_S32(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_s25_32_to_f32_to_s32_to_s25_32(void) @@ -681,7 +673,7 @@ static void test_lossless_s32(void) int64_t i; int32_t max_abs_err = -1; - const int32_t expected_max_abs_err = 255; + const int32_t expected_max_abs_err = 127; fprintf(stderr, "test %s:\n", __func__); for (i = S32_MIN; i < S32_MAX; i += (expected_max_abs_err >> 1)) { float v = S32_TO_F32(i); @@ -708,7 +700,7 @@ static void test_lossless_s32_lossless_subset(void) } } spa_assert_se(!all_lossless); - spa_assert_se(max_abs_err == 255); + spa_assert_se(max_abs_err == 127); } static void test_lossless_u32(void) @@ -875,8 +867,8 @@ int main(int argc, char *argv[]) test_lossless_s24(); test_lossless_u24(); 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_s32_to_f32_to_s25_32(); + test_lossless_s25_32_to_s32_to_f32_to_s32_to_s25_32(); test_lossless_s25_32_to_f32_to_s32_to_s25_32(); test_lossless_s32(); test_lossless_s32_lossless_subset();