[Mlir-commits] [flang] [mlir] [mlir][OpenMP] rewrite conversion of privatisation for omp.parallel (PR #111844)

[Kiran Chandramohan]

================
@@ -0,0 +1,262 @@
+! RUN: %flang_fc1 -fopenmp -emit-llvm %s -o - | FileCheck %s
+
+! Compinational testing of control flow graph and builder insertion points
+! in mlir-to-llvm conversion:
+!   - mixing multiple delayed privatisations and multiple reductions
+!   - multiple blocks in the private alloc region
+!   - private alloc region has to read from the mold variable
+!   - firstprivate
+!   - multiple blocks in the private copy region
+!   - multiple blocks in the reduction init region
+!   - reduction init region has to read from the mold variable
+!   - re-used omp.private ops
+!   - re-used omp.reduction.declare ops
+!   - unstructured code inside of the parallel region
+!   - needs private dealloc region, and this has multiple blocks
+!   - needs reduction cleanup region, and this has multiple blocks
+
+! This maybe belongs in the mlir tests, but what we are doing here is complex
+! enough that I find the kind of minimised mlir code prefered by mlir reviwers
----------------
kiranchandramohan wrote:

```suggestion
! enough that I find the kind of minimised mlir code preferred by mlir reviewers
```

https://github.com/llvm/llvm-project/pull/111844