[flang-commits] [flang] [flang][OpenMP] Fix LASTPRIVATE for iteration variables (PR #69773)

[Leandro Lupori via flang-commits]

================
@@ -349,34 +354,55 @@ void DataSharingProcessor::insertLastPrivateCompare(mlir::Operation *op) {
           }
         }
       } else if (mlir::isa<mlir::omp::WsLoopOp>(op)) {
-        mlir::Operation *lastOper = op->getRegion(0).back().getTerminator();
-        firOpBuilder.setInsertionPoint(lastOper);
-
         // Update the original variable just before exiting the worksharing
         // loop. Conversion as follows:
         //
         //                       omp.wsloop {
         // omp.wsloop {            ...
         //    ...                  store
-        //    store       ===>     %cmp = llvm.icmp "eq" %iv %ub
-        //    omp.yield            fir.if %cmp {
-        // }                         ^%lpv_update_blk:
+        //    store       ===>     %v = arith.addi %iv, %step
+        //    omp.yield            %cmp = %step < 0 ? %v < %ub : %v > %ub
+        // }                       fir.if %cmp {
+        //                           fir.store %v to %loopIV
----------------
luporl wrote:

The `fir.store` here is needed. Considering the example of the previous conversation, `loopIV` is updated in the beginning of the loop body only, to have the same value as `iv`, but we need it to hold `v` instead (`iv` + `step`). Also, `^%lpv_update_blk` copies `loopIV` to `ia` and not `iv` to `loopIV`. In the end this store gets optimized, because `^%lpv_update_blk` generates a load `loopIV` right after it.

https://github.com/llvm/llvm-project/pull/69773