member: check max SIMD bitwidth
authorCiara Power <ciara.power@intel.com>
Mon, 19 Oct 2020 13:48:54 +0000 (15:48 +0200)
committerDavid Marchand <david.marchand@redhat.com>
Mon, 19 Oct 2020 14:45:02 +0000 (16:45 +0200)
When choosing a vector path to take, an extra condition must be
satisfied to ensure the max SIMD bitwidth allows for the CPU
enabled path.

Signed-off-by: Ciara Power <ciara.power@intel.com>
Acked-by: Yipeng Wang <yipeng1.wang@intel.com>
lib/librte_member/rte_member_ht.c

index 3ea293a..a85561b 100644 (file)
@@ -7,6 +7,7 @@
 #include <rte_prefetch.h>
 #include <rte_random.h>
 #include <rte_log.h>
+#include <rte_vect.h>
 
 #include "rte_member.h"
 #include "rte_member_ht.h"
@@ -113,7 +114,8 @@ rte_member_create_ht(struct rte_member_setsum *ss,
        }
 #if defined(RTE_ARCH_X86)
        if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) &&
-                       RTE_MEMBER_BUCKET_ENTRIES == 16)
+                       RTE_MEMBER_BUCKET_ENTRIES == 16 &&
+                       rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256)
                ss->sig_cmp_fn = RTE_MEMBER_COMPARE_AVX2;
        else
 #endif