audiocvt: 5.1 to Stereo conversion utilizing AVX

This commit is contained in:
Joel Linn 2021-06-11 22:14:54 +02:00 committed by Ryan C. Gordon
parent 20eea021c6
commit b2c8d3e9e4
1 changed files with 93 additions and 1 deletions

View File

@ -43,6 +43,10 @@
#define HAVE_SSE3_INTRINSICS 1
#endif
#if defined(HAVE_IMMINTRIN_H) && !defined(SDL_DISABLE_IMMINTRIN_H)
#define HAVE_AVX_INTRINSICS 1
#endif
#if HAVE_SSE3_INTRINSICS
/* Convert from stereo to mono. Average left and right. */
static void SDLCALL
@ -100,6 +104,88 @@ SDL_ConvertStereoToMono(SDL_AudioCVT * cvt, SDL_AudioFormat format)
}
}
#if HAVE_AVX_INTRINSICS
/* MSVC will always accept AVX intrinsics when compiling for x64 */
#if defined(__clang__)
#pragma clang attribute push (__attribute__((target("avx"))), apply_to=function)
#elif defined(__GNUC__)
#pragma GCC push_options
#pragma GCC target("avx")
#endif
/* Convert from 5.1 to stereo. Average left and right, distribute center, discard LFE. */
static void SDLCALL
SDL_Convert51ToStereo_AVX(SDL_AudioCVT * cvt, SDL_AudioFormat format)
{
float *dst = (float *) cvt->buf;
const float *src = dst;
int i = cvt->len_cvt / (sizeof (float) * 6);
const float two_fifths_f = 1.0f / 2.5f;
const __m256 two_fifths_v = _mm256_set1_ps(two_fifths_f);
const __m256 half = _mm256_set1_ps(0.5f);
LOG_DEBUG_CONVERT("5.1", "stereo (using AVX)");
SDL_assert(format == AUDIO_F32SYS);
/* SDL's 5.1 layout: FL+FR+FC+LFE+BL+BR */
/* This implementation is based on SDL_Convert51ToStereo_SSE */
/* Because AVX operates with two 128 bit lanes, the shuffling */
/* here is not very efficient. AVX512F lifts this limitation. */
while (i >= 4) {
__m128 in0 = _mm_loadu_ps(src); /* 0FL 0FR 0FC 0LF */
__m128 in1 = _mm_loadu_ps(src + 4); /* 0BL 0BR 1FL 1FR */
__m128 in2 = _mm_loadu_ps(src + 8); /* 1FC 1LF 1BL 1BR */
__m128 in3 = _mm_loadu_ps(src + 12); /* 2FL 2FR 2FC 2LF */
__m128 in4 = _mm_loadu_ps(src + 16); /* 2BL 2BR 3FL 3FR */
__m128 in5 = _mm_loadu_ps(src + 20); /* 3FC 3LF 3BL 3BR */
/* 0FC 0FC 1FC 1FC */
__m128 fc_distributed_lo = _mm_shuffle_ps(in0, in2, _MM_SHUFFLE(0, 0, 2, 2));
/* 2FC 2FC 3FC 3FC */
__m128 fc_distributed_hi = _mm_shuffle_ps(in3, in5, _MM_SHUFFLE(0, 0, 2, 2));
/* 0FC 0FC 1FC 1FC 2FC 2FC 3FC 3FC */
__m256 fc_distributed = _mm256_mul_ps(half, _mm256_insertf128_ps(_mm256_castps128_ps256(fc_distributed_lo), fc_distributed_hi, 1));
/* 0FL 0FR 1BL 1BR */
__m128 permuted0_lo = _mm_shuffle_ps(in0, in2, _MM_SHUFFLE(3, 2, 1, 0));
/* 2FL 2FR 3BL 3BR */
__m128 permuted0_hi = _mm_shuffle_ps(in3, in5, _MM_SHUFFLE(3, 2, 1, 0));
/* 0FL 0FR 1BL 1BR 2FL 2FR 3BL 3BR */
__m256 permuted0 = _mm256_insertf128_ps(_mm256_castps128_ps256(permuted0_lo), permuted0_hi, 1);
/* 0BL 0BR 1FL 1FR 2BL 2BR 3FL 3FR */
__m256 permuted1 = _mm256_insertf128_ps(_mm256_castps128_ps256(in1), in4, 1);
/* 0FL 0FR 1BL 1BR 2FL 2FR 3BL 3BR */
/* + 0BL 0BR 1FL 1FR 2BL 2BR 3FL 3FR */
/* = 0L 0R 1L 1R 2L 2R 3L 3R */
__m256 out = _mm256_add_ps(permuted0, permuted1);
out = _mm256_add_ps(out, fc_distributed);
out = _mm256_mul_ps(out, two_fifths_v);
_mm256_storeu_ps(dst, out);
i -= 4; src += 24; dst += 8;
}
/* Finish off any leftovers with scalar operations. */
while (i) {
const float front_center_distributed = src[2] * 0.5f;
dst[0] = (src[0] + front_center_distributed + src[4]) * two_fifths_f; /* left */
dst[1] = (src[1] + front_center_distributed + src[5]) * two_fifths_f; /* right */
i--; src += 6; dst+=2;
}
cvt->len_cvt /= 3;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
#if defined(__clang__)
#pragma clang attribute pop
#elif defined(__GNUC__)
#pragma GCC pop_options
#endif
#endif
#if HAVE_SSE_INTRINSICS
/* Convert from 5.1 to stereo. Average left and right, distribute center, discard LFE. */
@ -1086,8 +1172,14 @@ SDL_BuildAudioCVT(SDL_AudioCVT * cvt,
if ((src_channels == 6) && (dst_channels <= 2)) {
SDL_AudioFilter filter = NULL;
#if HAVE_AVX_INTRINSICS
if (SDL_HasAVX()) {
filter = SDL_Convert51ToStereo_AVX;
}
#endif
#if HAVE_SSE_INTRINSICS
if (SDL_HasSSE()) {
if (!filter && SDL_HasSSE()) {
filter = SDL_Convert51ToStereo_SSE;
}
#endif