Index: lib/Driver/Tools.cpp
===================================================================
--- lib/Driver/Tools.cpp
+++ lib/Driver/Tools.cpp
@@ -4882,63 +4882,57 @@
     break;
   }
 
-  const char *result = llvm::StringSwitch<const char *>(MArch)
-    .Cases("armv2", "armv2a","arm2")
-    .Case("armv3", "arm6")
-    .Case("armv3m", "arm7m")
-    .Case("armv4", "strongarm")
-    .Case("armv4t", "arm7tdmi")
-    .Case("thumbv4t", "arm7tdmi")
-    .Cases("armv5", "armv5t", "arm10tdmi")
-    .Cases("thumbv5", "thumbv5t", "arm10tdmi")
-    .Cases("armv5e", "armv5te", "arm1022e")
-    .Cases("thumbv5e", "thumbv5te", "arm1022e")
-    .Case("armv5tej", "arm926ej-s")
-    .Case("thumbv5tej", "arm926ej-s")
-    .Cases("armv6", "armv6k", "arm1136jf-s")
-    .Cases("thumbv6", "thumbv6k", "arm1136jf-s")
-    .Case("armv6j", "arm1136j-s")
-    .Case("thumbv6j", "arm1136j-s")
-    .Cases("armv6z", "armv6zk", "arm1176jzf-s")
-    .Cases("thumbv6z", "thumbv6zk", "arm1176jzf-s")
-    .Case("armv6t2", "arm1156t2-s")
-    .Case("thumbv6t2", "arm1156t2-s")
-    .Cases("armv6m", "armv6-m", "cortex-m0")
-    .Case("thumbv6m", "cortex-m0")
-    .Cases("armv7", "armv7a", "armv7-a", "cortex-a8")
-    .Cases("armebv7", "armebv7a", "armebv7-a", "cortex-a8")
-    .Cases("thumbv7", "thumbv7a", "cortex-a8")
-    .Cases("thumbebv7", "thumbebv7a", "cortex-a8")
-    .Cases("armv7l", "armv7-l", "cortex-a8")
-    .Cases("armebv7l", "armebv7-l", "cortex-a8")
-    .Cases("armv7s", "armv7-s", "swift")
-    .Cases("armebv7s", "armebv7-s", "swift")
-    .Cases("armv7r", "armv7-r", "cortex-r4")
-    .Cases("armebv7r", "armebv7-r", "cortex-r4")
-    .Case("thumbv7r", "cortex-r4")
-    .Case("thumbebv7r", "cortex-r4")
-    .Cases("armv7m", "armv7-m", "cortex-m3")
-    .Cases("armebv7m", "armebv7-m", "cortex-m3")
-    .Case("thumbv7m", "cortex-m3")
-    .Case("thumbebv7m", "cortex-m3")
-    .Cases("armv7em", "armv7e-m", "cortex-m4")
-    .Cases("armebv7em", "armebv7e-m", "cortex-m4")
-    .Cases("thumbv7em", "thumbv7e-m", "cortex-m4")
-    .Cases("thumbebv7em", "thumbebv7e-m", "cortex-m4")
-    .Cases("armv8", "armv8a", "armv8-a", "cortex-a53")
-    .Cases("armebv8", "armebv8a", "armebv8-a", "cortex-a53")
-    .Cases("thumbv8", "thumbv8a", "cortex-a53")
-    .Cases("thumbebv8", "thumbebv8a", "cortex-a53")
-    .Case("ep9312", "ep9312")
-    .Case("iwmmxt", "iwmmxt")
-    .Case("xscale", "xscale")
-    // If all else failed, return the most base CPU with thumb interworking
-    // supported by LLVM.
-    .Default(0);
+  const char *result = nullptr;
+  if (MArch.startswith("arm") || MArch.startswith("thumb"))
+    result = llvm::StringSwitch<const char *>(MArch)
+      .EndsWith("v2", "arm2")
+      .EndsWith("v2a", "arm2")
+      .EndsWith("v3", "arm6")
+      .EndsWith("v3m", "arm7m")
+      .EndsWith("v4", "strongarm")
+      .EndsWith("v4t", "arm7tdmi")
+      .EndsWith("v5", "arm10tdmi")
+      .EndsWith("v5t", "arm10tdmi")
+      .EndsWith("v5e", "arm1022e")
+      .EndsWith("v5te", "arm1022e")
+      .EndsWith("v5tej", "arm926ej-s")
+      .EndsWith("v6", "arm1136jf-s")
+      .EndsWith("v6k", "arm1136jf-s")
+      .EndsWith("v6j", "arm1136j-s")
+      .EndsWith("v6z", "arm1176jzf-s")
+      .EndsWith("v6zk", "arm1176jzf-s")
+      .EndsWith("v6t2", "arm1156t2-s")
+      .EndsWith("v6m", "cortex-m0")
+      .EndsWith("v6-m", "cortex-m0")
+      .EndsWith("v7", "cortex-a8")
+      .EndsWith("v7a", "cortex-a8")
+      .EndsWith("v7-a", "cortex-a8")
+      .EndsWith("v7l", "cortex-a8")
+      .EndsWith("v7-l", "cortex-a8")
+      .EndsWith("v7s", "swift")
+      .EndsWith("v7-s", "swift")
+      .EndsWith("v7r", "cortex-r4")
+      .EndsWith("v7-r", "cortex-r4")
+      .EndsWith("v7m", "cortex-m3")
+      .EndsWith("v7-m", "cortex-m3")
+      .EndsWith("v7em", "cortex-m4")
+      .EndsWith("v7e-m", "cortex-m4")
+      .EndsWith("v8", "cortex-a53")
+      .EndsWith("v8a", "cortex-a53")
+      .EndsWith("v8-a", "cortex-a53")
+      .Default(nullptr);
+  else
+    result = llvm::StringSwitch<const char *>(MArch)
+      .Case("ep9312", "ep9312")
+      .Case("iwmmxt", "iwmmxt")
+      .Case("xscale", "xscale")
+      .Default(nullptr);
 
   if (result)
     return result;
 
+  // If all else failed, return the most base CPU with thumb interworking
+  // supported by LLVM.
   switch (Triple.getOS()) {
   case llvm::Triple::NetBSD:
     switch (Triple.getEnvironment()) {
Index: test/Driver/arm-cortex-cpus.c
===================================================================
--- test/Driver/arm-cortex-cpus.c
+++ test/Driver/arm-cortex-cpus.c
@@ -71,6 +71,10 @@
 // RUN: %clang -target arm -march=armv6m -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V6M %s
 // CHECK-V6M: "-cc1"{{.*}} "-triple" "thumbv6m-{{.*}} "-target-cpu" "cortex-m0"
 
+// RUN: %clang -target armv6m -mbig-endian -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V6M-BIG %s
+// RUN: %clang -target arm -march=armv6m -mbig-endian -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V6M-BIG %s
+// CHECK-V6M-BIG: "-cc1"{{.*}} "-triple" "thumbebv6m-{{.*}} "-target-cpu" "cortex-m0"
+
 // RUN: %clang -target armv7m -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V7M %s
 // RUN: %clang -target arm -march=armv7-m -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V7M %s
 // CHECK-V7M: "-cc1"{{.*}} "-triple" "thumbv7m-{{.*}} "-target-cpu" "cortex-m3"
@@ -79,6 +83,10 @@
 // RUN: %clang -target arm -march=armv7e-m -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V7EM %s
 // CHECK-V7EM: "-cc1"{{.*}} "-triple" "thumbv7em-{{.*}} "-target-cpu" "cortex-m4"
 
+// RUN: %clang -target armv7em -mbig-endian -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V7EM-BIG %s
+// RUN: %clang -target arm -march=armv7e-m -mbig-endian -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V7EM-BIG %s
+// CHECK-V7EM-BIG: "-cc1"{{.*}} "-triple" "thumbebv7em-{{.*}} "-target-cpu" "cortex-m4"
+
 // RUN: %clang -target armv6m-apple-darwin -arch armv6m -### -c %s 2>&1 | FileCheck -check-prefix=CHECK-V6M-DARWIN %s
 // CHECK-V6M-DARWIN: "-cc1"{{.*}} "-triple" "thumbv6m-{{.*}} "-target-cpu" "cortex-m0"