Add home made _mm256_set_m128i for compilers who are missing it

apache · Sep 14, 2024 · 9134050 · 9134050
1 parent 86569b7
commit 9134050
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 7 deletions.
diff --git a/cpp/src/arrow/acero/swiss_join_avx2.cc b/cpp/src/arrow/acero/swiss_join_avx2.cc
@@ -15,10 +15,9 @@
 // specific language governing permissions and limitations
 // under the License.
 
-#include <immintrin.h>
-
 #include "arrow/acero/swiss_join_internal.h"
 #include "arrow/util/bit_util.h"
+#include "arrow/util/simd.h"
 
 namespace arrow {
 namespace acero {

diff --git a/cpp/src/arrow/compute/row/compare_internal_avx2.cc b/cpp/src/arrow/compute/row/compare_internal_avx2.cc
@@ -15,11 +15,10 @@
 // specific language governing permissions and limitations
 // under the License.
 
-#include <immintrin.h>
-
 #include "arrow/compute/row/compare_internal.h"
 #include "arrow/compute/util.h"
 #include "arrow/util/bit_util.h"
+#include "arrow/util/simd.h"
 
 namespace arrow {
 namespace compute {

diff --git a/cpp/src/arrow/util/simd.h b/cpp/src/arrow/util/simd.h
@@ -27,18 +27,25 @@
 #else
 // gcc/clang (possibly others)
 
-#  if defined(ARROW_HAVE_BMI2)
+#  if defined(ARROW_HAVE_BMI2) || defined(ARROW_HAVE_RUNTIME_BMI2)
 #    include <x86intrin.h>
 #  endif
 
-#  if defined(ARROW_HAVE_AVX2) || defined(ARROW_HAVE_AVX512)
+#  if defined(ARROW_HAVE_AVX2) || defined(ARROW_HAVE_AVX512) || \
+      defined(ARROW_HAVE_RUNTIME_AVX2) || defined(ARROW_HAVE_RUNTIME_AVX512)
 #    include <immintrin.h>
-#  elif defined(ARROW_HAVE_SSE4_2)
+#  elif defined(ARROW_HAVE_SSE4_2) || defined(ARROW_HAVE_RUNTIME_SSE4_2)
 #    include <nmmintrin.h>
 #  endif
 
 #  ifdef ARROW_HAVE_NEON
 #    include <arm_neon.h>
 #  endif
 
+// GH-44098: Workaround for missing _mm256_set_m128i in older versions of GCC.
+#  if defined(__GNUC__) && !defined(__clang__) && __GNUC__ < 8
+#    define _mm256_set_m128i(hi, lo) \
+      _mm256_inserti128_si256(_mm256_castsi128_si256(lo), (hi), 1)
+#  endif
+
 #endif