Merge pull request #395 from supersjgk/master

Fixed issue #394 - missing _mm256_storeu2_m128i macro
ultravideo · Feb 16, 2024 · d8c9688 · d8c9688
2 parents a909ddd + 203d3be
commit d8c9688
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 0 deletions.
diff --git a/src/strategies/avx2/quant-avx2.c b/src/strategies/avx2/quant-avx2.c
@@ -40,6 +40,8 @@
 #include <immintrin.h>
 #include <stdlib.h>
 
+#include "strategies/missing-intel-intrinsics.h"
+
 #include "avx2_common_functions.h"
 #include "cu.h"
 #include "encoder.h"

diff --git a/src/strategies/missing-intel-intrinsics.h b/src/strategies/missing-intel-intrinsics.h
@@ -30,6 +30,16 @@
   #ifndef _mm256_extract_epi32
     #define _mm256_extract_epi32(a, index) (_mm_extract_epi32(_mm256_extracti128_si256((a), (index) >> 2), (index) & 3))
   #endif
+
+  #ifndef _mm256_storeu2_m128i
+    #define _mm256_storeu2_m128i(/* __m128i* */ hiaddr, /* __m128i* */ loaddr, /* __m256i */ a) \
+    do { __m256i _a = (a); \
+        _mm_storeu_si128((__m128i*)(loaddr), _mm256_castsi256_si128(_a)); \
+        _mm_storeu_si128((__m128i*)(hiaddr), _mm256_extractf128_si256(_a, 0x1)); \
+    } while (0)
+  #endif
+
 #endif
 
 #endif
+