[Mlir-commits] [mlir] [mlir][OpenMP] inscan reduction modifier and scan op mlir support (PR #114737)

[Sergio Afonso]

================
@@ -4591,6 +4591,10 @@ convertHostOrTargetOperation(Operation *op, llvm::IRBuilderBase &builder,
       .Case([&](omp::AtomicCaptureOp op) {
         return convertOmpAtomicCapture(op, builder, moduleTranslation);
       })
+      .Case([&](omp::ScanOp) {
+        return op->emitError()
+               << "not yet implemented: " << op->getName() << " operation";
+      })
----------------
skatrak wrote:

I don't think this is necessary, since the default case already does this. With my previous comment I was referring to adding a case to `checkReduction` in `checkImplementationStatus` triggering a not-yet-implemented error if the `reduction_mod` is set to anything but "default". Something like this:
```c++
auto checkReduction = [&todo](auto op, LogicalResult &result) {
  if (isa<omp::TeamsOp, omp::SimdOp>(op)) {
    if (!op.getReductionVars().empty() || op.getReductionByref() ||
        op.getReductionSyms())
      result = todo("reduction");
  }
  if (op.getReductionMod() && op.getReductionMod().getValue() != omp::ReductionModifier::defaultmod)
    result = todo("reduction with modifier");
};
```
You'd just need to also make sure that the updated `checkReduction` is called for all operations that can take `reduction` clauses: parallel, teams, sections, loop, wsloop, simd and taskloop.

https://github.com/llvm/llvm-project/pull/114737