diff --git a/build-m1.sh b/build-m1.sh
new file mode 100755
index 000000000..502a340ad
--- /dev/null
+++ b/build-m1.sh
@@ -0,0 +1,31 @@
+#!/usr/bin/env bash
+
+release_mode=$1
+
+warnings_to_disable="-std=c++11 -Wno-switch -Wno-pointer-sign -Wno-tautological-constant-out-of-range-compare -Wno-tautological-compare -Wno-macro-redefined"
+libraries="-pthread -ldl -lm -lstdc++"
+other_args="-DLLVM_BACKEND_SUPPORT -DUSE_NEW_LLVM_ABI_SYSTEM"
+compiler="clang"
+
+if [ -z "$release_mode" ]; then release_mode="0"; fi
+
+if [ "$release_mode" -eq "0" ]; then
+	other_args="${other_args} -g"
+fi
+if [ "$release_mode" -eq "1" ]; then
+	other_args="${other_args} -O3 -march=native"
+fi
+
+if [[ "$(uname)" == "Darwin" ]]; then
+
+	# Set compiler to clang on MacOS
+	# MacOS provides a symlink to clang called gcc, but it's nice to be explicit here.
+	compiler="clang"
+
+	other_args="${other_args} -liconv"
+elif [[ "$(uname)" == "FreeBSD" ]]; then
+	compiler="clang"
+fi
+
+${compiler} src/main.cpp ${warnings_to_disable} ${libraries} ${other_args} -o odin \
+	&& ./odin run examples/demo/demo.odin -llvm-api
diff --git a/src/gb/gb.h b/src/gb/gb.h
index f13693000..b2718561b 100644
--- a/src/gb/gb.h
+++ b/src/gb/gb.h
@@ -157,7 +157,7 @@ extern "C" {
 	#endif
 #endif
 
-#if defined(_WIN64) || defined(__x86_64__) || defined(_M_X64) || defined(__64BIT__) || defined(__powerpc64__) || defined(__ppc64__)
+#if defined(_WIN64) || defined(__x86_64__) || defined(_M_X64) || defined(__64BIT__) || defined(__powerpc64__) || defined(__ppc64__) || defined(__aarch64__)
 	#ifndef GB_ARCH_64_BIT
 	#define GB_ARCH_64_BIT 1
 	#endif
@@ -230,7 +230,7 @@ extern "C" {
 	#define GB_CACHE_LINE_SIZE 128
 	#endif
 
-#elif defined(__arm__)
+#elif defined(__arm__) || defined(__aarch64__) || defined(_M_ARM) || defined(_M_ARM64)
 	#ifndef GB_CPU_ARM
 	#define GB_CPU_ARM 1
 	#endif
@@ -3702,6 +3702,12 @@ gb_inline void *gb_memcopy(void *dest, void const *source, isize n) {
 
 	void *dest_copy = dest;
 	__asm__ __volatile__("rep movsb" : "+D"(dest_copy), "+S"(source), "+c"(n) : : "memory");
+#elif defined(GB_CPU_ARM)
+	u8 *s = cast(u8 *)source;
+	u8 *d = cast(u8 *)dest;
+	for (isize i = 0; i < n; i++) {
+		*d++ = *s++;
+	}
 #else
 	u8 *d = cast(u8 *)dest;
 	u8 const *s = cast(u8 const *)source;
@@ -4438,6 +4444,76 @@ gb_inline i64 gb_atomic64_fetch_or(gbAtomic64 volatile *a, i64 operand) {
 #endif
 }
 
+#elif defined(GB_CPU_ARM)
+
+gb_inline i32  gb_atomic32_load (gbAtomic32 const volatile *a) { 
+	return __atomic_load_n(&a->value, __ATOMIC_SEQ_CST);
+}
+gb_inline void gb_atomic32_store(gbAtomic32 volatile *a, i32 value) { 
+	__atomic_store_n(&a->value, value, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i32 gb_atomic32_compare_exchange(gbAtomic32 volatile *a, i32 expected, i32 desired) {
+	i32 expected_copy = expected;
+	auto result = __atomic_compare_exchange_n(&a->value, &expected_copy, desired, true, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST);
+	if (result) {
+		return expected;
+	} else {
+		return expected_copy;
+	}
+}
+
+gb_inline i32 gb_atomic32_exchanged(gbAtomic32 volatile *a, i32 desired) {
+	return __atomic_exchange_n(&a->value, desired, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i32 gb_atomic32_fetch_add(gbAtomic32 volatile *a, i32 operand) {
+	return __atomic_fetch_add(&a->value, operand, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i32 gb_atomic32_fetch_and(gbAtomic32 volatile *a, i32 operand) {
+	return __atomic_fetch_and(&a->value, operand, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i32 gb_atomic32_fetch_or(gbAtomic32 volatile *a, i32 operand) {
+	return __atomic_fetch_or(&a->value, operand, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i64 gb_atomic64_load(gbAtomic64 const volatile *a) {
+	return __atomic_load_n(&a->value, __ATOMIC_SEQ_CST);
+}
+
+gb_inline void gb_atomic64_store(gbAtomic64 volatile *a, i64 value) {
+	__atomic_store_n(&a->value, value, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i64 gb_atomic64_compare_exchange(gbAtomic64 volatile *a, i64 expected, i64 desired) {
+	i64 expected_copy = expected;
+	auto result = __atomic_compare_exchange_n(&a->value, &expected_copy, desired, true, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST);
+	if (result) {
+		return expected;
+	} else {
+		return expected_copy;
+	}
+}
+
+gb_inline i64 gb_atomic64_exchanged(gbAtomic64 volatile *a, i64 desired) {
+	return __atomic_exchange_n(&a->value, desired, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i64 gb_atomic64_fetch_add(gbAtomic64 volatile *a, i64 operand) {
+	return __atomic_fetch_add(&a->value, operand, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i64 gb_atomic64_fetch_and(gbAtomic64 volatile *a, i64 operand) {
+	return __atomic_fetch_and(&a->value, operand, __ATOMIC_SEQ_CST);
+}
+
+gb_inline i64 gb_atomic64_fetch_or(gbAtomic64 volatile *a, i64 operand) {
+	return __atomic_fetch_or(&a->value, operand, __ATOMIC_SEQ_CST);
+}
+
+
 #else
 #error TODO(bill): Implement Atomics for this CPU
 #endif
diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp
index 5bbccc18a..f8c6b44e1 100644
--- a/src/llvm_backend.cpp
+++ b/src/llvm_backend.cpp
@@ -8312,8 +8312,8 @@ lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValue const &tv,
 		{
 			LLVMTypeRef func_type = LLVMFunctionType(LLVMVoidTypeInContext(p->module->ctx), nullptr, 0, false);
 			LLVMValueRef the_asm = LLVMGetInlineAsm(func_type,
-				"pause", 5,
-				"", 0,
+				cast(char *)"pause", 5,
+				cast(char *)"", 0,
 				/*HasSideEffects*/true, /*IsAlignStack*/false,
 				LLVMInlineAsmDialectATT
 			);