Mirrors/overte
3
0
Fork 0
mirror of https://github.com/overte-org/overte.git synced 2025-04-14 03:07:05 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

Merge pull request #15838 from kencooke/render-blendshape-optim-avx2

Browse source 
Fix Mac crash caused by PR#15825
This commit is contained in:
Andrew Meadows 2019-06-25 13:54:39 -07:00 committed by GitHub
commit adce2a6fd0
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 10 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
libraries/render-utils/src/Model.cpp
View file

@ -1780,10 +1780,10 @@ void Blender::run() {
blendedMeshSizes.reserve(numMeshes);
QVector<BlendshapeOffset> packedBlendshapeOffsets;
packedBlendshapeOffsets.reserve(numBlendshapeOffsets);
packedBlendshapeOffsets.resize(numBlendshapeOffsets);
QVector<BlendshapeOffsetUnpacked> unpackedBlendshapeOffsets;
unpackedBlendshapeOffsets.reserve(maxBlendshapeOffsets); // reuse for all meshes
unpackedBlendshapeOffsets.resize(maxBlendshapeOffsets); // reuse for all meshes
int offset = 0;
for (auto meshIter = _hfmModel->meshes.cbegin(); meshIter != _hfmModel->meshes.cend(); ++meshIter) {

16
libraries/shared/src/avx2/BlendshapePacking_avx2.cpp
View file

@ -21,14 +21,14 @@ void packBlendshapeOffsets_AVX2(float (*unpacked)[9], uint32_t (*packed)[4], int
//
// deinterleave (8x9 to 9x8 matrix transpose)
//
__m256 s0 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_load_ps(&unpacked[i+0][0])), _mm_load_ps(&unpacked[i+4][0]), 1);
__m256 s1 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_load_ps(&unpacked[i+1][0])), _mm_load_ps(&unpacked[i+5][0]), 1);
__m256 s2 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_load_ps(&unpacked[i+2][0])), _mm_load_ps(&unpacked[i+6][0]), 1);
__m256 s3 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_load_ps(&unpacked[i+3][0])), _mm_load_ps(&unpacked[i+7][0]), 1);
__m256 s4 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_load_ps(&unpacked[i+0][4])), _mm_load_ps(&unpacked[i+4][4]), 1);
__m256 s5 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_load_ps(&unpacked[i+1][4])), _mm_load_ps(&unpacked[i+5][4]), 1);
__m256 s6 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_load_ps(&unpacked[i+2][4])), _mm_load_ps(&unpacked[i+6][4]), 1);
__m256 s7 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_load_ps(&unpacked[i+3][4])), _mm_load_ps(&unpacked[i+7][4]), 1);
__m256 s0 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_loadu_ps(&unpacked[i+0][0])), _mm_loadu_ps(&unpacked[i+4][0]), 1);
__m256 s1 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_loadu_ps(&unpacked[i+1][0])), _mm_loadu_ps(&unpacked[i+5][0]), 1);
__m256 s2 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_loadu_ps(&unpacked[i+2][0])), _mm_loadu_ps(&unpacked[i+6][0]), 1);
__m256 s3 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_loadu_ps(&unpacked[i+3][0])), _mm_loadu_ps(&unpacked[i+7][0]), 1);
__m256 s4 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_loadu_ps(&unpacked[i+0][4])), _mm_loadu_ps(&unpacked[i+4][4]), 1);
__m256 s5 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_loadu_ps(&unpacked[i+1][4])), _mm_loadu_ps(&unpacked[i+5][4]), 1);
__m256 s6 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_loadu_ps(&unpacked[i+2][4])), _mm_loadu_ps(&unpacked[i+6][4]), 1);
__m256 s7 = _mm256_insertf128_ps(_mm256_castps128_ps256(_mm_loadu_ps(&unpacked[i+3][4])), _mm_loadu_ps(&unpacked[i+7][4]), 1);
__m256 t0 = _mm256_unpacklo_ps(s0, s1);
__m256 t1 = _mm256_unpackhi_ps(s0, s1);