================
@@ -2142,85 +2178,58 @@ void NeonEmitter::genOverloadTypeCheckCode(raw_ostream
&OS,
OS << "#endif\n\n";
}
-void NeonEmitter::genIntrinsicRangeCheckCode(raw_ostream &OS,
- SmallVectorImpl<Intrinsic *> &Defs) {
- OS << "#ifdef GET_NEON_IMMEDIATE_CHECK\n";
+inline bool
+NeonEmitter::areRangeChecksCompatable(const ArrayRef<ImmCheck> ChecksA,
+ const ArrayRef<ImmCheck> ChecksB) {
+ // If multiple intrinsics map to the same builtin, we must ensure that the
+ // intended range checks performed in SemaArm.cpp do not contradict each
+ // other, as these are emitted once per-buitlin.
+ //
+ // The arguments to be checked and type of each check to be performed must be
+ // the same. The element types may differ as they will be resolved
+ // per-intrinsic as overloaded types by SemaArm.cpp, though the vector sizes
+ // are not and so must be the same.
+ bool compat =
----------------
SpencerAbson wrote:
We actually had this approach initially, the problem with it is that this issue
only prevalent for NEON intrinsics yet the class is also used for SVE/SME.
Defining **equality** for SVE/SME and NEON based on a definition of
**compatibility** that only applies to NEON intrinsics didn't seem right at the
time.
https://github.com/llvm/llvm-project/pull/100278
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
