diff --git a/defs.bzl b/defs.bzl
index 04f3a718e1..d7dcbd30ec 100644
--- a/defs.bzl
+++ b/defs.bzl
@@ -97,11 +97,17 @@ def get_fbgemm_avx2_srcs(msvc = False):
         "src/UtilsAvx2.cc",
     ]
 
-def get_fbgemm_inline_avx2_srcs(msvc = False):
-    return [
-        #FP16 kernels contain inline assembly and inline assembly syntax for MSVC is different.
-        "src/FbgemmFP16UKernelsAvx2.cc" if not msvc else "src/FbgemmFP16UKernelsIntrinsicAvx2.cc",
-    ]
+def get_fbgemm_inline_avx2_srcs(msvc = False, buck = False):
+    intrinsics_srcs = ["src/FbgemmFP16UKernelsIntrinsicAvx2.cc"]
+
+    #FP16 kernels contain inline assembly and inline assembly syntax for MSVC is different.
+    asm_srcs = ["src/FbgemmFP16UKernelsAvx2.cc"]
+    if buck:
+        return select({
+            "DEFAULT": asm_srcs if not msvc else intrinsics_srcs,
+            "ovr_config//cpu:arm64": intrinsics_srcs,
+        })
+    return asm_srcs if not msvc else intrinsics_srcs
 
 def get_fbgemm_avx512_srcs(msvc = False):
     return [
@@ -116,12 +122,21 @@ def get_fbgemm_avx512_srcs(msvc = False):
         "src/UtilsAvx512.cc",
     ]
 
-def get_fbgemm_inline_avx512_srcs(msvc = False):
-    return [
-        #FP16 kernels contain inline assembly and inline assembly syntax for MSVC is different.
-        "src/FbgemmFP16UKernelsAvx512.cc" if not msvc else "src/FbgemmFP16UKernelsIntrinsicAvx512.cc",
-        "src/FbgemmFP16UKernelsAvx512_256.cc" if not msvc else "src/FbgemmFP16UKernelsIntrinsicAvx512_256.cc",
+def get_fbgemm_inline_avx512_srcs(msvc = False, buck = False):
+    intrinsics_srcs = [
+        "src/FbgemmFP16UKernelsIntrinsicAvx512.cc",
+        "src/FbgemmFP16UKernelsIntrinsicAvx512_256.cc",
+    ]
+    asm_srcs = [
+        "src/FbgemmFP16UKernelsAvx512.cc",
+        "src/FbgemmFP16UKernelsAvx512_256.cc",
     ]
+    if buck:
+        return select({
+            "DEFAULT": asm_srcs if not msvc else intrinsics_srcs,
+            "ovr_config//cpu:arm64": intrinsics_srcs,
+        })
+    return asm_srcs if not msvc else intrinsics_srcs
 
 def get_fbgemm_tests(skip_tests = []):
     return native.glob(["test/*Test.cc"], exclude = skip_tests)
diff --git a/src/FbgemmBfloat16Convert.cc b/src/FbgemmBfloat16Convert.cc
index 76eec63ef6..482c97ea62 100644
--- a/src/FbgemmBfloat16Convert.cc
+++ b/src/FbgemmBfloat16Convert.cc
@@ -43,9 +43,12 @@ namespace fbgemm {
 void FloatToBfloat16_simd(const float* src, bfloat16* dst, size_t size) {
   // Run time CPU detection
   if (cpuinfo_initialize()) {
+#ifndef __aarch64__
     if (fbgemmHasAvx512Support()) {
       FloatToBfloat16_avx512(src, dst, size);
-    } else if (fbgemmHasAvx2Support()) {
+    } else
+#endif
+        if (fbgemmHasAvx2Support()) {
       FloatToBfloat16_avx2(src, dst, size);
     } else {
       FloatToBfloat16_ref(src, dst, size);
@@ -59,9 +62,12 @@ void FloatToBfloat16_simd(const float* src, bfloat16* dst, size_t size) {
 void Bfloat16ToFloat_simd(const bfloat16* src, float* dst, size_t size) {
   // Run time CPU detection
   if (cpuinfo_initialize()) {
+#ifndef __aarch64__
     if (fbgemmHasAvx512Support()) {
       Bfloat16ToFloat_avx512(src, dst, size);
-    } else if (fbgemmHasAvx2Support()) {
+    } else
+#endif
+        if (fbgemmHasAvx2Support()) {
       Bfloat16ToFloat_avx2(src, dst, size);
     } else {
       Bfloat16ToFloat_ref(src, dst, size);
diff --git a/src/FbgemmFP16.cc b/src/FbgemmFP16.cc
index d30f6017d6..2bd66a6866 100644
--- a/src/FbgemmFP16.cc
+++ b/src/FbgemmFP16.cc
@@ -49,6 +49,7 @@ constexpr kernel_array_t<float16> kernel_fp16_avx512_256 = {
     gemmkernel_14x2_Avx512_256_fp16_fA0fB0fC0};
 
 constexpr kernel_array_t<float16> kernel_fp16_avx512 = {
+#ifndef __aarch64__
     nullptr,
     gemmkernel_1x2_Avx512_fp16_fA0fB0fC0,
     gemmkernel_2x2_Avx512_fp16_fA0fB0fC0,
@@ -63,7 +64,11 @@ constexpr kernel_array_t<float16> kernel_fp16_avx512 = {
     gemmkernel_11x2_Avx512_fp16_fA0fB0fC0,
     gemmkernel_12x2_Avx512_fp16_fA0fB0fC0,
     gemmkernel_13x2_Avx512_fp16_fA0fB0fC0,
-    gemmkernel_14x2_Avx512_fp16_fA0fB0fC0};
+    gemmkernel_14x2_Avx512_fp16_fA0fB0fC0
+#else
+    nullptr
+#endif
+};
 
 } // namespace
 
diff --git a/src/FbgemmFP16UKernelsIntrinsicAvx2.cc b/src/FbgemmFP16UKernelsIntrinsicAvx2.cc
index 7e64582d93..4ac159eadc 100644
--- a/src/FbgemmFP16UKernelsIntrinsicAvx2.cc
+++ b/src/FbgemmFP16UKernelsIntrinsicAvx2.cc
@@ -5,7 +5,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-#ifdef _MSC_VER
 #if defined(__x86_64__) || defined(__i386__) || \
     (defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)))
 #include <immintrin.h>
@@ -115,4 +114,3 @@ void NOINLINE gemmkernel_6x2_Avx2_fp16_fA0fB0fC0(GemmParamsFP16* gp) {
 }
 
 } // namespace fbgemm
-#endif // _MSC_VER
diff --git a/src/FbgemmFP16UKernelsIntrinsicAvx512.cc b/src/FbgemmFP16UKernelsIntrinsicAvx512.cc
index 5ad9ea6222..ce69a5d969 100644
--- a/src/FbgemmFP16UKernelsIntrinsicAvx512.cc
+++ b/src/FbgemmFP16UKernelsIntrinsicAvx512.cc
@@ -5,7 +5,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-#ifdef _MSC_VER
 #if defined(__x86_64__) || defined(__i386__) || \
     (defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)))
 #include <immintrin.h>
@@ -140,4 +139,3 @@ void NOINLINE gemmkernel_14x2_Avx512_fp16_fA0fB0fC0(GemmParamsFP16* gp) {
 }
 
 } // namespace fbgemm
-#endif // _MSC_VER
diff --git a/src/FbgemmFP16UKernelsIntrinsicAvx512_256.cc b/src/FbgemmFP16UKernelsIntrinsicAvx512_256.cc
index cf9f5f8f18..3445cecc77 100644
--- a/src/FbgemmFP16UKernelsIntrinsicAvx512_256.cc
+++ b/src/FbgemmFP16UKernelsIntrinsicAvx512_256.cc
@@ -5,7 +5,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-#ifdef _MSC_VER
 #if defined(__x86_64__) || defined(__i386__) || \
     (defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86)))
 #include <immintrin.h>
@@ -121,4 +120,3 @@ void NOINLINE gemmkernel_14x2_Avx512_256_fp16_fA0fB0fC0(GemmParamsFP16* gp) {
 }
 
 } // namespace fbgemm
-#endif // _MSC_VER
diff --git a/src/QuantUtilsAvx2.cc b/src/QuantUtilsAvx2.cc
index dd3772d49c..e96bd379fb 100644
--- a/src/QuantUtilsAvx2.cc
+++ b/src/QuantUtilsAvx2.cc
@@ -166,6 +166,9 @@ void NO_SANITIZE("address") FusedQuantizeDequantizeAvx2(
   float inverse_scale = 1.f / qparams.scale;
   constexpr int32_t min_val = std::numeric_limits<T>::min();
   constexpr int32_t max_val = std::numeric_limits<T>::max();
+  (void)inverse_scale; // Suppress unused variable warning
+  (void)min_val; // Suppress unused variable warning
+  (void)max_val; // Suppress unused variable warning
 #if defined(__AVX2__) && (defined(__FMA__) || defined(_MSC_VER))
 
   constexpr int VLEN = 8;