[PATCH] D136560: llvm-reduce: Try to turn calls into something else

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/tools/llvm-reduce/deltas/ReduceOpcodes.cpp:139
+  if (DataArg)
+    return B.CreateStore(DataArg, PtrArg, true);
+  return B.CreateLoad(CB->getType(), PtrArg, true);
----------------
aeubanks wrote:
> why volatile?
volatile acts more like a call with a pointer. It's treated as a pointer capture, has stricter scheduling. Plus we may be emitting a load/store from null in the default address space, so this makes that defined. If the volatile doesn't matter, the volatile reduction can remove it later


================
Comment at: llvm/tools/llvm-reduce/deltas/ReduceOpcodes.cpp:194
+    case 3:
+      return B.CreateIntrinsic(Intrinsic::fshl, {CB->getType()}, Arguments);
+    default:
----------------
aeubanks wrote:
> IMO a call to an intrinsic is less reduced than a call to some arbitrary external function.
> Perhaps it's finally time to create a flag for reducing IR optimizing for codegen?
An intrinsic has known behavior, an arbitrary external call doesn't. From a semantics perspective, that's more understandable 


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D136560/new/

https://reviews.llvm.org/D136560