[mlir][LLVMIR][OpenMP] fix dominance for reduction init block

It was incorrect to set the insertion point to the init block after
inlining the initialization region because the code generated in the
init block depends upon the value yielded from the init region. When
there were multiple reduction initialization regions each with multiple
blocks, this could lead to the initilization region being inlined after
the init block which depends upon it.

Moving the insertion point to before inlining the initialization block
turned up further issues around the handling of the terminator for the
initialization block, which are also fixed here.

This fixes a bug in llvm#92430 (but the affected code couldn't compile
before llvm#92430 anyway).

Author: tblah

mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp

@@ -388,8 +388,18 @@ static LogicalResult inlineConvertOmpRegions(
     // be processed multiple times.
     moduleTranslation.forgetMapping(region);
 
-    if (potentialTerminator && potentialTerminator->isTerminator())
-      potentialTerminator->insertAfter(&builder.GetInsertBlock()->back());
+    if (potentialTerminator && potentialTerminator->isTerminator()) {
+      llvm::BasicBlock *block = builder.GetInsertBlock();
+      if (block->empty())
+        // this can happen for really simple reduction init regions e.g.
+        // %0 = llvm.mlir.constant(0 : i32) : i32
+        // omp.yield(%0 : i32)
+        // because the llvm.mlir.constant (MLIR op) isn't converted into any
+        // llvm op
+        potentialTerminator->insertInto(block, block->begin());
+      else
+        potentialTerminator->insertAfter(&block->back());
+    }
 
     return success();
   }
@@ -1171,6 +1181,8 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder,
       }
     }
 
+    builder.SetInsertPoint(initBlock->getFirstNonPHIOrDbgOrAlloca());
+
     for (unsigned i = 0; i < opInst.getNumReductionVars(); ++i) {
       SmallVector<llvm::Value *> phis;
 
@@ -1183,7 +1195,10 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder,
       assert(phis.size() == 1 &&
              "expected one value to be yielded from the "
              "reduction neutral element declaration region");
-      builder.SetInsertPoint(initBlock->getTerminator());
+
+      // mapInitializationArg finishes its block with a terminator. We need to
+      // insert before that terminator.
+      builder.SetInsertPoint(builder.GetInsertBlock()->getTerminator());
 
       if (isByRef[i]) {
         // Store the result of the inlined region to the allocated reduction var