[Mlir-commits] [mlir] [mlir][sparse_tensor] Migrate `SparseIterationToScf.cpp` to dialect conversion (PR #121054)

[Peiming Liu]

================
@@ -178,46 +190,47 @@ namespace {
 
 /// Sparse codegen rule for number of entries operator.
 class ExtractIterSpaceConverter
-    : public OneToNOpConversionPattern<ExtractIterSpaceOp> {
+    : public OpConversionPattern<ExtractIterSpaceOp> {
 public:
-  using OneToNOpConversionPattern::OneToNOpConversionPattern;
+  using OpConversionPattern::OpConversionPattern;
   LogicalResult
-  matchAndRewrite(ExtractIterSpaceOp op, OpAdaptor adaptor,
-                  OneToNPatternRewriter &rewriter) const override {
+  matchAndRewrite(ExtractIterSpaceOp op, OneToNOpAdaptor adaptor,
+                  ConversionPatternRewriter &rewriter) const override {
     Location loc = op.getLoc();
-    const OneToNTypeMapping &resultMapping = adaptor.getResultMapping();
 
     // Construct the iteration space.
-    SparseIterationSpace space(loc, rewriter, op.getTensor(), 0,
+    SparseIterationSpace space(loc, rewriter,
+                               getSingleValue(adaptor.getTensor()), 0,
                                op.getLvlRange(), adaptor.getParentIter());
 
     SmallVector<Value> result = space.toValues();
-    rewriter.replaceOp(op, result, resultMapping);
+    rewriter.replaceOpWithMultiple(op, {result});
----------------
PeimingLiu wrote:

Sry that I did not follow the MLIR 1toN changes, but it seems wired to use `{result}` when `result` is already a vector. What is the expected type here?

https://github.com/llvm/llvm-project/pull/121054