[Mlir-commits] [mlir] [mlir][vector] Refactor WarpOpScfForOp to support unused or swapped forOp results. (PR #147620)

[Igor Zamyatin]

================
@@ -1745,19 +1749,18 @@ struct WarpOpScfForOp : public WarpDistributionPattern {
       : WarpDistributionPattern(ctx, b), distributionMapFn(std::move(fn)) {}
   LogicalResult matchAndRewrite(WarpExecuteOnLane0Op warpOp,
                                 PatternRewriter &rewriter) const override {
-    auto yield = cast<gpu::YieldOp>(
+    auto newWarpOpYield = cast<gpu::YieldOp>(
         warpOp.getBodyRegion().getBlocks().begin()->getTerminator());
-    // Only pick up forOp if it is the last op in the region.
-    Operation *lastNode = yield->getPrevNode();
+    // Only pick up `ForOp` if it is the last op in the region.
+    Operation *lastNode = newWarpOpYield->getPrevNode();
     auto forOp = dyn_cast_or_null<scf::ForOp>(lastNode);
     if (!forOp)
       return failure();
-    // Collect Values that come from the warp op but are outside the forOp.
-    // Those Value needs to be returned by the original warpOp and passed to
-    // the new op.
+    // Collect Values that come from the `WarpOp` but are outside the `ForOp`.
+    // Those Values need to be returned by the new warp op.
     llvm::SmallSetVector<Value, 32> escapingValues;
-    SmallVector<Type> inputTypes;
-    SmallVector<Type> distTypes;
+    SmallVector<Type> escapingValueInputTypes;
+    SmallVector<Type> escapingValuedistTypes;
----------------
Garra1980 wrote:

dist->Dist

https://github.com/llvm/llvm-project/pull/147620