[COFF, ARM64] Add __byteswap intrinsics

[efriedma]

lib/Headers/arm64intr.h

50

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

18–81

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/arm64intr.h

typedef enum
   _ARM64_BARRIER_NSH   = 0x7,
   _ARM64_BARRIER_NSHST = 0x6,
   _ARM64_BARRIER_NSHLD = 0x5,
   _ARM64_BARRIER_OSH   = 0x3,
   _ARM64_BARRIER_OSHST = 0x2,
   _ARM64_BARRIER_OSHLD = 0x1
 } _ARM64INTR_BARRIER_TYPE;
 
+#define _byteswap_ushort __builtin_bswap16
+#define _byteswap_ulong  __builtin_bswap32
+#define _byteswap_uint64 __builtin_bswap64
+
 #endif /* __ARM64INTR_H */
 #endif /* _MSC_VER */

test/Headers/ms-arm64-intrin.cpp

 // RUN: %clang_cc1 -triple arm64-windows \
 // 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
 
+// RUN: %clang_cc1 -triple arm64-windows \
+// RUN: -fms-compatibility -fms-compatibility-version=17.00 \
+// RUN: -ffreestanding -fsyntax-only -Werror \
+// RUN: -isystem %S/Inputs/include %s -S -o - -emit-llvm 2>&1 | FileCheck %s --check-prefix CHECK-IR
+
 // REQUIRES: aarch64-registered-target
 
 #include <intrin.h>
 
 void f() {
 // CHECK: nop
   __nop();
 }
+
+unsigned short check_byteswap_ushort_1() {
+// CHECK-IR: ret i16 256
+  return _byteswap_ushort(1);
+}
+unsigned short check_byteswap_ushort_2() {
+// CHECK-IR: ret i16 512
+  return _byteswap_ushort(2);
+}
+unsigned short check_byteswap_ushort_3() {
+// CHECK-IR: ret i16 768
+  return _byteswap_ushort(3);
+}
+unsigned short check_byteswap_ushort_4() {
+// CHECK-IR: ret i16 1024
+  return _byteswap_ushort(4);
+}
+unsigned short check_byteswap_ushort_5() {
+// CHECK-IR: ret i16 1280
+  return _byteswap_ushort(5);
+}
+
+unsigned long check_byteswap_ulong_1() {
+// CHECK-IR: ret i32 16777216
+  return _byteswap_ulong(1);
+}
+unsigned long check_byteswap_ulong_2() {
+// CHECK-IR: ret i32 33554432
+  return _byteswap_ulong(2);
+}
+unsigned long check_byteswap_ulong_3() {
+// CHECK-IR: ret i32 50331648
+  return _byteswap_ulong(3);
+}
+unsigned long check_byteswap_ulong_4() {
+// CHECK-IR: ret i32 67108864
+  return _byteswap_ulong(4);
+}
+unsigned long check_byteswap_ulong_5() {
+// CHECK-IR: ret i32 83886080
+  return _byteswap_ulong(5);
+}
+
+unsigned __int64 check_byteswap_uint64_1() {
+// CHECK-IR: ret i64 72057594037927936
+  return _byteswap_uint64(1);
+}
+unsigned __int64 check_byteswap_uint64_2() {
+// CHECK-IR: ret i64 144115188075855872
+  return _byteswap_uint64(2);
+}
+unsigned __int64 check_byteswap_uint64_3() {
+// CHECK-IR: ret i64 216172782113783808
+  return _byteswap_uint64(3);
+}
+unsigned __int64 check_byteswap_uint64_4() {
+// CHECK-IR: ret i64 288230376151711744
+  return _byteswap_uint64(4);
+}
+unsigned __int64 check_byteswap_uint64_5() {
+// CHECK-IR: ret i64 360287970189639680
+  return _byteswap_uint64(5);
+}