From b079e3283ede4ce91ff1dda5aebdf1d5aadedf8c Mon Sep 17 00:00:00 2001
From: Ken Cooke <ken@highfidelity.io>
Date: Tue, 25 Jun 2019 11:48:41 -0700
Subject: [PATCH] BUGFIX: use unaligned loads
---
.../shared/src/avx2/BlendshapePacking_avx2.cpp | 16 ++++++++--------
1 file changed, 8 insertions(+), 8 deletions(-)
diff --git a/libraries/shared/src/avx2/BlendshapePacking_avx2.cpp b/libraries/shared/src/avx2/BlendshapePacking_avx2.cpp
index 5524c355dc..1fea60a315 100644
--- a/libraries/shared/src/avx2/BlendshapePacking_avx2.cpp
+++ b/libraries/shared/src/avx2/BlendshapePacking_avx2.cpp
@@ -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);