[llvm] [mlir] [mlir][OpenMP] Allow composite SIMD REDUCTION and IF (PR #147568)

[Tom Eccles via llvm-commits]

================
@@ -5370,58 +5370,90 @@ void OpenMPIRBuilder::unrollLoopHeuristic(DebugLoc, CanonicalLoopInfo *Loop) {
 
 void OpenMPIRBuilder::createIfVersion(CanonicalLoopInfo *CanonicalLoop,
                                       Value *IfCond, ValueToValueMapTy &VMap,
+                                      LoopAnalysis &LIA, LoopInfo &LI, Loop *L,
                                       const Twine &NamePrefix) {
   Function *F = CanonicalLoop->getFunction();
 
-  // Define where if branch should be inserted
-  Instruction *SplitBefore = CanonicalLoop->getPreheader()->getTerminator();
-
-  // TODO: We should not rely on pass manager. Currently we use pass manager
-  // only for getting llvm::Loop which corresponds to given CanonicalLoopInfo
-  // object. We should have a method  which returns all blocks between
-  // CanonicalLoopInfo::getHeader() and CanonicalLoopInfo::getAfter()
-  FunctionAnalysisManager FAM;
----------------
tblah wrote:

Just to clarify: these passes are removed from createIfVersion because there is already a LoopAnalysis run in the calling function.

Further I wanted to use the same `Loop` instance in both so that I could add the new block to it.

https://github.com/llvm/llvm-project/pull/147568