Fixed _mm256_set_m128 is only availble on gcc8+. issue#5072 (#5075)

Tencent · Oct 10, 2023 · b4f8fa6 · b4f8fa6
1 parent c8f92b9
commit b4f8fa6
Showing 1 changed file with 3 additions and 3 deletions.
diff --git a/src/layer/x86/shufflechannel_x86.cpp b/src/layer/x86/shufflechannel_x86.cpp
@@ -343,9 +343,9 @@ int ShuffleChannel_x86::forward(const Mat& bottom_blob, Mat& top_blob, const Opt
                 for (int i = 0; i < size; i++)
                 {
                     __m256 _p0 = _mm256_loadu_ps(ptr0);
-                    // macro `_mm256_loadu2_m128` is declared in Intel® Intrinsics Guide but somehow missed in <immintrin.h>
-                    // __m256 _p1 = _mm256_loadu2_m128(ptr2, ptr1);
-                    __m256 _p1 = _mm256_set_m128(_mm_loadu_ps(ptr2), _mm_loadu_ps(ptr1));
+
+                    __m256 _p1 = _mm256_castps128_ps256(_mm_loadu_ps(ptr1));
+                    _p1 = _mm256_insertf128_ps(_p1, _mm_loadu_ps(ptr2), 1);
 
                     __m256 _lo = _mm256_unpacklo_ps(_p0, _p1);
                     __m256 _hi = _mm256_unpackhi_ps(_p0, _p1);