claunia/flac
1
0
Fork 0
mirror of https://github.com/claunia/flac.git synced 2025-12-16 18:54:26 +00:00
Code Issues Packages Projects Releases Wiki Activity

libFLAC: Refactoring

Browse Source 
No functional changes.

Patch-from: lvqcl <lvqcl.mail@gmail.com>
This commit is contained in:
Erik de Castro Lopo
2015-11-03 18:08:49 +11:00
parent 239843e5c1
commit 86b36d92d5
4 changed files with 50 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
src/libFLAC/stream_encoder_intrin_avx2.c
View File

@@ -55,29 +55,29 @@ void FLAC__precompute_partition_info_sums_intrin_avx2(const FLAC__int32 residual
/* first do max_partition_order */
{
const unsigned threshold = 32 - FLAC__bitmath_ilog2(default_partition_samples);
unsigned partition, residual_sample, end = (unsigned)(-(int)predictor_order);
__m256i res256, sum256;
__m128i res128, sum128;
if(FLAC__bitmath_ilog2(default_partition_samples) + bps + FLAC__MAX_EXTRA_RESIDUAL_BPS < 32) {
if(bps + FLAC__MAX_EXTRA_RESIDUAL_BPS < threshold) {
for(partition = residual_sample = 0; partition < partitions; partition++) {
__m256i sum256 = _mm256_setzero_si256();
__m128i sum128;
end += default_partition_samples;
sum256 = _mm256_setzero_si256();
for( ; (int)residual_sample < (int)end-7; residual_sample+=8) {
res256 = _mm256_abs_epi32(_mm256_loadu_si256((const __m256i*)(residual+residual_sample)));
__m256i res256 = _mm256_abs_epi32(_mm256_loadu_si256((const __m256i*)(residual+residual_sample)));
sum256 = _mm256_add_epi32(sum256, res256);
}
sum128 = _mm_add_epi32(_mm256_extracti128_si256(sum256, 1), _mm256_castsi256_si128(sum256));
for( ; (int)residual_sample < (int)end-3; residual_sample+=4) {
res128 = _mm_abs_epi32(_mm_loadu_si128((const __m128i*)(residual+residual_sample)));
__m128i res128 = _mm_abs_epi32(_mm_loadu_si128((const __m128i*)(residual+residual_sample)));
sum128 = _mm_add_epi32(sum128, res128);
}
for( ; residual_sample < end; residual_sample++) {
res128 = _mm_cvtsi32_si128(residual[residual_sample]);
__m128i res128 = _mm_cvtsi32_si128(residual[residual_sample]);
res128 = _mm_abs_epi32(res128);
sum128 = _mm_add_epi32(sum128, res128);
}
@@ -89,26 +89,27 @@ void FLAC__precompute_partition_info_sums_intrin_avx2(const FLAC__int32 residual
}
else { /* have to pessimistically use 64 bits for accumulator */
for(partition = residual_sample = 0; partition < partitions; partition++) {
__m256i sum256 = _mm256_setzero_si256();
__m128i sum128;
end += default_partition_samples;
sum256 = _mm256_setzero_si256();
for( ; (int)residual_sample < (int)end-3; residual_sample+=4) {
res128 = _mm_abs_epi32(_mm_loadu_si128((const __m128i*)(residual+residual_sample)));
res256 = _mm256_cvtepu32_epi64(res128);
__m128i res128 = _mm_abs_epi32(_mm_loadu_si128((const __m128i*)(residual+residual_sample)));
__m256i res256 = _mm256_cvtepu32_epi64(res128);
sum256 = _mm256_add_epi64(sum256, res256);
}
sum128 = _mm_add_epi64(_mm256_extracti128_si256(sum256, 1), _mm256_castsi256_si128(sum256));
for( ; (int)residual_sample < (int)end-1; residual_sample+=2) {
res128 = _mm_loadl_epi64((const __m128i*)(residual+residual_sample));
__m128i res128 = _mm_loadl_epi64((const __m128i*)(residual+residual_sample));
res128 = _mm_abs_epi32(res128);
res128 = _mm_cvtepu32_epi64(res128);
sum128 = _mm_add_epi64(sum128, res128);
}
for( ; residual_sample < end; residual_sample++) {
res128 = _mm_cvtsi32_si128(residual[residual_sample]);
__m128i res128 = _mm_cvtsi32_si128(residual[residual_sample]);
res128 = _mm_abs_epi32(res128);
sum128 = _mm_add_epi64(sum128, res128);
}