[COFF, ARM64] Add __byteswap intrinsics

[efriedma]

lib/Headers/arm64intr.h

50 ↗

(On Diff #181657)

These should be available for all Microsoft targets, I think.

Please define these as inline functions or proper builtins, rather than macros; a macro is very likely to break something. (For example, if someone includes stdlib.h, which declares "unsigned long _byteswap_ulong(unsigned long);", replacing _byteswap_ulong with __builtin_bswap32 will cause a compile error.)

[efriedma]

LGTM

test/Headers/ms-arm64-intrin.cpp

14–29

While you're here, please fix all four tests to just check the IR.

[efriedma]

Before you commit, please verify this actually compiles if you include both this intrin.h and stdlib.h from the Microsoft SDK, in either order.

[mgrang]

Verified that this compiles when both intrin.h and stdlib.h from the Microsoft SDK are included, in either order.

lib/Headers/intrin.h

 /*----------------------------------------------------------------------------*\
 |* MS AArch64 specific
 \*----------------------------------------------------------------------------*/
 #if defined(__aarch64__)
 unsigned __int64 __getReg(int);
 long _InterlockedAdd(long volatile *Addend, long Value);
 int _ReadStatusReg(int);
 void _WriteStatusReg(int, int);
+
+static inline unsigned short _byteswap_ushort (unsigned short val) {
+  return __builtin_bswap16(val);
+}
+static inline unsigned long _byteswap_ulong (unsigned long val) {
+  return __builtin_bswap32(val);
+}
+static inline unsigned __int64 _byteswap_uint64 (unsigned __int64 val) {
+  return __builtin_bswap64(val);
+}
 #endif
 
 /*----------------------------------------------------------------------------*\
 |* Privileged intrinsics
 \*----------------------------------------------------------------------------*/
 #if defined(__i386__) || defined(__x86_64__)
 static __inline__ unsigned __int64 __DEFAULT_FN_ATTRS
 __readmsr(unsigned long __register) {

test/Headers/ms-arm64-intrin.cpp

-// RUN: %clang_cc1 -triple arm64-windows \
+// REQUIRES: aarch64-registered-target
+
+// RUN: %clang_cc1 -triple arm64-windows -O1 \
 // RUN: -fms-compatibility -fms-compatibility-version=17.00 \
 // RUN: -ffreestanding -fsyntax-only -Werror \
-// RUN: -isystem %S/Inputs/include %s -S -o - 2>&1 | FileCheck %s
-
-// REQUIRES: aarch64-registered-target
+// RUN: -isystem %S/Inputs/include %s -S -o - -emit-llvm 2>&1 \
+// RUN: | FileCheck %s
 
 #include <intrin.h>
 
-void f() {
-// CHECK: nop
+void check_nop() {
+// CHECK: "nop"
   __nop();
 }
+
+unsigned short check_byteswap_ushort(unsigned short val) {
+// CHECK: call i16 @llvm.bswap.i16(i16 %val)
+  return _byteswap_ushort(val);
+}
+
+unsigned long check_byteswap_ulong(unsigned long val) {
+// CHECK: call i32 @llvm.bswap.i32(i32 %val)
+  return _byteswap_ulong(val);
+}
+
+unsigned __int64 check_byteswap_uint64(unsigned __int64 val) {
+// CHECK: call i64 @llvm.bswap.i64(i64 %val)
+  return _byteswap_uint64(val);
+}