diff --git a/build/premake/premake4.lua b/build/premake/premake4.lua
index cf9e82a917..180e997b87 100644
--- a/build/premake/premake4.lua
+++ b/build/premake/premake4.lua
@@ -26,7 +26,7 @@ rootdir = "../.."
 
 dofile("extern_libs4.lua")
 
--- detect CPU architecture (simplistic, currently only supports x86 and amd64)
+-- detect CPU architecture (simplistic, currently only supports x86, amd64 and ARM)
 arch = "x86"
 if _OPTIONS["android"] then
 	arch = "arm"
@@ -47,6 +47,8 @@ else
 			arch = "amd64"
 		elseif string.find(machine, "i.86") == 1 then
 			arch = "x86"
+		elseif string.find(machine, "arm") == 1 then
+			arch = "arm"
 		else
 			print("WARNING: Cannot determine architecture from GCC, assuming x86")
 		end
@@ -262,9 +264,13 @@ function project_set_build_flags()
 			if arch == "arm" then
 				-- disable warnings about va_list ABI change
 				buildoptions { "-Wno-psabi" }
-
-				-- target Cortex-A9 CPUs with NEON
-				buildoptions { "-mthumb -mcpu=cortex-a9 -mfpu=neon -mfloat-abi=softfp" }
+				if _OPTIONS["android"] then
+					-- target generic arm CPUs with NEON
+					buildoptions { "-mtune=generic-arm -mfpu=neon -mfloat-abi=softfp" }
+				else
+					-- target Cortex-A15 CPUs with NEON
+					buildoptions { "-mtune=cortex-a15 -mfpu=neon-vfpv4 -mfloat-abi=hard" }
+				end
 			end
 
 			if _OPTIONS["coverage"] then
diff --git a/source/lib/sysdep/arch/arm/arm.cpp b/source/lib/sysdep/arch/arm/arm.cpp
index 867e80ce8e..e86b2aa286 100644
--- a/source/lib/sysdep/arch/arm/arm.cpp
+++ b/source/lib/sysdep/arch/arm/arm.cpp
@@ -25,40 +25,25 @@
  */
 
 #include "precompiled.h"
-
-#include "lib/sysdep/cpu.h"
-
-#include <android/log.h>
-intptr_t cpu_AtomicAdd(volatile intptr_t* location, intptr_t increment)
-{
-	return __sync_fetch_and_add(location, increment);
+
+#include "lib/sysdep/cpu.h"
+
+intptr_t cpu_AtomicAdd(volatile intptr_t* location, intptr_t increment)
+{
+	return __sync_fetch_and_add(location, increment);
 }
 
 bool cpu_CAS(volatile intptr_t* location, intptr_t expected, intptr_t newValue)
 {
 	return __sync_bool_compare_and_swap(location, expected, newValue);
 }
-
-bool cpu_CAS64(volatile i64* location, i64 expected, i64 newValue)
-{
-	// Current versions of GCC don't implement this on ARM:
-	//   return __sync_bool_compare_and_swap(location, expected, newValue);
-	
-	// http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=40fb79c8a88625504857d44de1bc89dc0341e618
-	// adds support for
-	//   return __kernel_cmpxchg64(&expected, &newValue, (long long*)location) == 0;
-	// but only for Linux kernel 3.1
-	
-	// Maybe we can do it with user-space assembly assuming a modern-enough CPU?
-	// That sounds non-trivial so let's just cheat
-#warning TODO: atomic cpu_CAS64 on ARM
-	if (*location != expected)
-		return false;
-	*location = newValue;
-	return true;
-}
-
-const char* cpu_IdentifierString()
+
+bool cpu_CAS64(volatile i64* location, i64 expected, i64 newValue)
+{
+	return __sync_bool_compare_and_swap(location, expected, newValue);
+}
+
+const char* cpu_IdentifierString()
 {
 	return "unknown"; // TODO
 }