[Mlir-commits] [mlir] Refactor LoopFuseSiblingOp and support parallel fusion (PR #94391)

[llvmlistbot at llvm.org]

================
@@ -113,3 +115,56 @@ LogicalResult detail::verifyLoopLikeOpInterface(Operation *op) {
 
   return success();
 }
+
+LoopLikeOpInterface mlir::createFused(LoopLikeOpInterface target,
+                                      LoopLikeOpInterface source,
+                                      RewriterBase &rewriter,
+                                      NewYieldValuesFn newYieldValuesFn,
+                                      FuseTerminatorFn fuseTerminatorFn) {
+  auto targetIterArgs = target.getRegionIterArgs();
+  std::optional<SmallVector<Value>> targetInductionVar =
+      target.getLoopInductionVars();
+  SmallVector<Value> targetYieldOperands(target.getYieldedValues());
+  auto sourceIterArgs = source.getRegionIterArgs();
+  std::optional<SmallVector<Value>> sourceInductionVar =
+      *source.getLoopInductionVars();
+  SmallVector<Value> sourceYieldOperands(source.getYieldedValues());
+  auto sourceRegion = source.getLoopRegions().front();
+
+  FailureOr<LoopLikeOpInterface> maybeFusedLoop =
+      target.replaceWithAdditionalYields(rewriter, source.getInits(),
+                                         /*replaceInitOperandUsesInLoop=*/false,
+                                         newYieldValuesFn);
----------------
srcarroll wrote:

@ftynse  i can't find your original comment, i guess because i moved the code. but it was about expanding auto and checking for failures. i have done so here and below with `fusedLoop` and the optional induction vars

https://github.com/llvm/llvm-project/pull/94391