diff --git a/cmake/ConfigureCompiler.cmake b/cmake/ConfigureCompiler.cmake
index 5be685b34f..1c37e8930f 100644
--- a/cmake/ConfigureCompiler.cmake
+++ b/cmake/ConfigureCompiler.cmake
@@ -255,6 +255,7 @@ else()
   if (GCC)
     add_compile_options(-Wno-pragmas)
     add_compile_options(-mavx)
+#    add_compile_options(-fno-builtin-memcpy)
     # Otherwise `state [[maybe_unused]] int x;` will issue a warning.
     # https://stackoverflow.com/questions/50646334/maybe-unused-on-member-variable-gcc-warns-incorrectly-that-attribute-is
     add_compile_options(-Wno-attributes)
diff --git a/flow/CMakeLists.txt b/flow/CMakeLists.txt
index 080d483edd..61ce9ed17b 100644
--- a/flow/CMakeLists.txt
+++ b/flow/CMakeLists.txt
@@ -75,6 +75,7 @@ set(FLOW_SRCS
   flat_buffers.h
   flow.cpp
   flow.h
+  folly_memcpy.S
   genericactors.actor.cpp
   genericactors.actor.h
   network.cpp
diff --git a/flow/folly_memcpy.S b/flow/folly_memcpy.S
index e6e95371eb..66361a774a 100644
--- a/flow/folly_memcpy.S
+++ b/flow/folly_memcpy.S
@@ -79,9 +79,9 @@ _memcpy_short:
  *
  */
         .align    16
-        .globl    memcpy
-        .type     memcpy, @function
-memcpy:
+        .globl    folly_memcpy
+        .type     folly_memcpy, @function
+folly_memcpy:
         .cfi_startproc
 
         mov       %rdx, %rcx
@@ -173,6 +173,6 @@ memcpy:
         ret
 
         .cfi_endproc
-        .size memcpy, .-memcpy
+        .size folly_memcpy, .-folly_memcpy
 
 #endif
diff --git a/flow/test_memcpy_perf.cpp b/flow/test_memcpy_perf.cpp
index b2750b3585..c54f70aacf 100644
--- a/flow/test_memcpy_perf.cpp
+++ b/flow/test_memcpy_perf.cpp
@@ -12,6 +12,10 @@
 #include "flow/IRandom.h"
 #include "flow/UnitTest.h"
 
+extern "C" {
+	void* folly_memcpy(void* dst, void* src, uint32_t length);
+}
+
 /*
  * Set this to the maximum buffer size you want to test. If it is 0, then the
  * values in the buf_sizes[] array below will be used.
@@ -46,13 +50,13 @@ static size_t buf_sizes[TEST_VALUE_RANGE];
 #define TEST_BATCH_SIZE         100
 
 /* Data is aligned on this many bytes (power of 2) */
-#ifdef RTE_MACHINE_CPUFLAG_AVX512F
+// #ifdef RTE_MACHINE_CPUFLAG_AVX512F
 #define ALIGNMENT_UNIT          64
-#elif defined RTE_MACHINE_CPUFLAG_AVX2
-#define ALIGNMENT_UNIT          32
-#else /* RTE_MACHINE_CPUFLAG */
-#define ALIGNMENT_UNIT          16
-#endif /* RTE_MACHINE_CPUFLAG */
+// #elif defined RTE_MACHINE_CPUFLAG_AVX2
+// #define ALIGNMENT_UNIT          32
+// #else /* RTE_MACHINE_CPUFLAG */
+// #define ALIGNMENT_UNIT          16
+// #endif /* RTE_MACHINE_CPUFLAG */
 
 /*
  * Pointers used in performance tests. The two large buffers are for uncached