[COFF, ARM64] Add MS builtins __dmb, __dsb, __isb

[mgrang]

Related clang patch: https://reviews.llvm.org/D36111

[mstorsjo]

Out of curiousity (I'm not familiar enough with this aspect of LLVM internals) - what practical difference does this make - wouldn't the existing llvm level intrinsic be usable as such already? (I do see that the same is done for __builtin_arm_dmb and __dmb etc for ARM already, so I'm sure it's right - just curious where it makes a difference.)

[efriedma]

GCCBuiltin/MSBuiltin emit tables which get used by clang IR generation. See Intrinsic::getIntrinsicForGCCBuiltin/Intrinsic::getIntrinsicForMSBuiltin.

[mstorsjo]

In D36110#826676, @efriedma wrote:

GCCBuiltin/MSBuiltin emit tables which get used by clang IR generation. See Intrinsic::getIntrinsicForGCCBuiltin/Intrinsic::getIntrinsicForMSBuiltin.

Thanks for the explanation!

This change looks ok to me. I guess this is a change that isn't too easy to add tests for so I guess it's ok as is?

[compnerd]

Might be a good idea to add tests for the IR itself that we lower it properly.

[efriedma]

LGTM.

There aren't any useful tests to add here because the generated tables are only used by clang.

@compnerd We already have tests for llvm.aarch64.dmb etc.

llvm/trunk/include/llvm/IR/IntrinsicsAArch64.td

 //===----------------------------------------------------------------------===//
 // HINT
 
 def int_aarch64_hint : Intrinsic<[], [llvm_i32_ty]>;
 
 //===----------------------------------------------------------------------===//
 // Data Barrier Instructions
 
-def int_aarch64_dmb : GCCBuiltin<"__builtin_arm_dmb">, Intrinsic<[], [llvm_i32_ty]>;
-def int_aarch64_dsb : GCCBuiltin<"__builtin_arm_dsb">, Intrinsic<[], [llvm_i32_ty]>;
-def int_aarch64_isb : GCCBuiltin<"__builtin_arm_isb">, Intrinsic<[], [llvm_i32_ty]>;
+def int_aarch64_dmb : GCCBuiltin<"__builtin_arm_dmb">, MSBuiltin<"__dmb">, Intrinsic<[], [llvm_i32_ty]>;
+def int_aarch64_dsb : GCCBuiltin<"__builtin_arm_dsb">, MSBuiltin<"__dsb">, Intrinsic<[], [llvm_i32_ty]>;
+def int_aarch64_isb : GCCBuiltin<"__builtin_arm_isb">, MSBuiltin<"__isb">, Intrinsic<[], [llvm_i32_ty]>;
 
 }
 
 //===----------------------------------------------------------------------===//
 // Advanced SIMD (NEON)
 
 let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
   class AdvSIMD_2Scalar_Float_Intrinsic