[mlir][Interfaces] Add interface methods to allow reifying single result/single dim of result. (#162924)

Current implementation of `reifyResultShapes` forces all
implementations to return all dimensions of all results. This can be
wasteful when you only require dimensions of one result, or a single
dimension of a result. Further this also creates issues with using
patterns to resolve the `tensor.dim` and `memref.dim` operations since
the extra operations created result in the pattern rewriter entering
an infinite loop (eventually breaking out of the loop due to the
iteration limit on the pattern rewriter). This is demonstrated by some
of the test cases added here that hit this limit when using
`--resolve-shaped-type-result-dims` and
`--resolve-ranked-shaped-type-result-dims`. To resolve this issue the
interface should allow for creating just the operations needed. This
change is the first step in resolving this.

The original implementation was done with the restriction in mind that
it might not always be possible to compute dimension of a single
result or one dimension of a single result in all cases. To account
for such cases, two additional interface methods are added

- `reifyShapeOfResult` (which allows reifying dimensions of
  just one result), has a default implementation that calls
  `reifyResultShapes` and returns the dimensions of a single result.
- `reifyDimOfResult` (which allows reifying a single dimension of a
  single result) has a default implementation that calls
  `reifyDimOfResult` and returns the value for the dimension of the
  result (which in turn for the default case would call
  `reifyDimOfResult`).

While this change sets up the interface, ideally most operations will
implement the `refiyDimOfResult` when possible. For almost all
operations in tree this is true. Subsequent commits will change those
incrementally.

Some of the tests added here that check that the default
implementations for the above method work as expected, also end up
hitting the pattern rewriter limit when using
`--resolve-ranked-shaped-type-result-dims`/
`--resolve-ranked-shaped-type-result-dims`. For testing purposes, a
flag is added to these passes that ignore the error returned by the
pattern application (this flag is left on by default to maintain
current state).

Changes required downstream to integrate this change

1. In operation definitions in .td files, for those operations that
   implement the `ReifyRankedShapedTypeOpInterface`.

```
def <op-name> : Op<..., [...,
    DeclareOpInterfaceMethods[ReifyRankedShapedTypeOpInterface]]>
```

should be changed to

```
def <op-name> : Op<..., [...,
    DeclareOpInterfaceMethods[ReifyRankedShapedTypeOpInterface, [
        "reifyResultShapes"]]]>
```

---------

Signed-off-by: MaheshRavishankar <[email protected]>

Author: MaheshRavishankar

mlir/include/mlir/Dialect/Bufferization/IR/BufferizationOps.td

@@ -28,7 +28,8 @@ class Bufferization_Op<string mnemonic, list<Trait> traits = []>
 
 def Bufferization_AllocTensorOp : Bufferization_Op<"alloc_tensor",
     [AttrSizedOperandSegments, BufferizableOpInterface,
-     DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>]> {
+     DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+       "reifyResultShapes"]>]> {
   let summary = "allocate buffer for a tensor";
 
   let description = [{
@@ -219,7 +220,8 @@ def Bufferization_MaterializeInDestinationOp
     : Bufferization_Op<"materialize_in_destination",
         [AllElementTypesMatch<["source", "dest"]>,
          BufferizableOpInterface, DestinationStyleOpInterface,
-         DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+         DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+           "reifyResultShapes"]>,
          DeclareOpInterfaceMethods<SubsetOpInterface,
             ["operatesOnEquivalentSubset", "operatesOnDisjointSubset"]>,
          DeclareOpInterfaceMethods<SubsetInsertionOpInterface,

mlir/include/mlir/Dialect/Linalg/IR/LinalgOps.td

@@ -94,7 +94,8 @@ def Linalg_IndexOp : Linalg_Op<"index", [Pure]>,
 def Linalg_SoftmaxOp : Linalg_Op<"softmax",
     [DestinationStyleOpInterface,
      PredOpTrait<"input and output have same element type", TCopVTEtIsSameAs<0, 1>>,
-     DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+     DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface,
+      ["reifyResultShapes"]>,
      DeclareOpInterfaceMethods<AggregatedOpInterface, ["decomposeOperation"]>,
      DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
      DeclareOpInterfaceMethods<TilingInterface,

mlir/include/mlir/Dialect/Linalg/IR/LinalgRelayoutOps.td

@@ -35,7 +35,8 @@ class Linalg_RelayoutOp<string mnemonic, list<Trait> traits = []> :
         DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
         DestinationStyleOpInterface, LinalgRelayoutOpInterface,
         ConditionallySpeculatable, NoMemoryEffect,
-        DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+        DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+          "reifyResultShapes"]>,
         TypesMatchWith<"result type matches type of dest",
                    "dest", "result",
                    "$_self">])> {

mlir/include/mlir/Dialect/MemRef/IR/MemRefOps.td

@@ -1783,7 +1783,8 @@ class MemRef_ReassociativeReshapeOp<string mnemonic, list<Trait> traits = []> :
 def MemRef_ExpandShapeOp : MemRef_ReassociativeReshapeOp<"expand_shape", [
     DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
     DeclareOpInterfaceMethods<MemorySpaceCastConsumerOpInterface>,
-    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>]> {
+    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface,
+      ["reifyResultShapes"]>]> {
   let summary = "operation to produce a memref with a higher rank.";
   let description = [{
     The `memref.expand_shape` op produces a new view with a higher rank whose

mlir/include/mlir/Dialect/MemRef/Transforms/Passes.td

@@ -164,6 +164,11 @@ def ResolveRankedShapeTypeResultDimsPass
     implement the `ReifyRankedShapedTypeOpInterface` in terms of
     shapes of its operands.
   }];
+  let options = [
+    Option<"errorOnPatternIterationLimit", "error-on-pattern-iteration-limit", "bool",
+           /*default=*/"true",
+           "Throw an error when pattern rewriter hits iteration limit">,
+  ];
   let dependentDialects = [
     "memref::MemRefDialect", "tensor::TensorDialect"
   ];
@@ -177,6 +182,11 @@ def ResolveShapedTypeResultDimsPass : Pass<"resolve-shaped-type-result-dims"> {
     `ReifyRankedShapedTypeOpInterface` in terms of shapes of its
     operands.
   }];
+  let options = [
+    Option<"errorOnPatternIterationLimit", "error-on-pattern-iteration-limit", "bool",
+           /*default=*/"true",
+           "Throw an error when pattern rewriter hits iteration limit">,
+  ];
   let dependentDialects = [
     "affine::AffineDialect", "memref::MemRefDialect", "tensor::TensorDialect"
   ];

mlir/include/mlir/Dialect/Tensor/IR/TensorOps.td

@@ -131,7 +131,9 @@ def Tensor_CastOp : Tensor_Op<"cast", [
 def Tensor_ConcatOp : Tensor_Op<"concat",
     [Pure,
      DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
-     DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>]> {
+     DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+       "reifyResultShapes"]>,
+     ]> {
   let summary = "tensor concatenation operation";
   let description = [{
     The "concat" operation constructs a tensor out of a variadic list of input
@@ -261,7 +263,8 @@ def Tensor_DimOp : Tensor_Op<"dim", [
 
 def Tensor_EmptyOp : Tensor_Op<"empty",
     [Pure,
-     DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>]> {
+     DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+       "reifyResultShapes"]>]> {
   let summary = "empty tensor operation";
 
   let description = [{
@@ -358,7 +361,8 @@ def Tensor_ExtractOp : Tensor_Op<"extract", [
 
 def Tensor_ExtractSliceOp : Tensor_OpWithOffsetSizesAndStrides<"extract_slice", [
     DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
-    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+      "reifyResultShapes"]>,
     AttrSizedOperandSegments,
     Pure,
     OffsetSizeAndStrideOpInterface
@@ -740,7 +744,8 @@ def Tensor_GatherOp : Tensor_Op<"gather", [
 def Tensor_GenerateOp : Tensor_Op<"generate", [
     DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
     RecursiveMemoryEffects,
-    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+      "reifyResultShapes"]>,
     SingleBlockImplicitTerminator<"mlir::tensor::YieldOp">]> {
   let summary = "Creates a dynamically sized tensor from elements";
   let description = [{
@@ -835,7 +840,8 @@ def Tensor_InsertOp : Tensor_Op<"insert", [
 
 def Tensor_InsertSliceOp : Tensor_OpWithOffsetSizesAndStrides<"insert_slice", [
     DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
-    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+      "reifyResultShapes"]>,
     AttrSizedOperandSegments,
     DestinationStyleOpInterface,
     Pure,
@@ -1256,7 +1262,8 @@ def Tensor_CollapseShapeOp : Tensor_ReassociativeReshapeOp<"collapse_shape"> {
 
 def Tensor_PadOp : Tensor_Op<"pad", [
     DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
-    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, [
+      "reifyResultShapes"]>,
     AttrSizedOperandSegments,
     Pure,
     SingleBlockImplicitTerminator<"mlir::tensor::YieldOp">]> {
@@ -1764,7 +1771,8 @@ def Tensor_ScatterOp : Tensor_Op<"scatter", [
 
 def Tensor_SplatOp : Tensor_Op<"splat", [
     DeclareOpInterfaceMethods<OpAsmOpInterface, ["getAsmResultNames"]>,
-    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+    DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface, 
+      ["reifyResultShapes"]>,
     Pure,
     TypesMatchWith<"operand type matches element type of result",
                    "aggregate", "input",

mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td

@@ -2219,7 +2219,8 @@ def Tosa_TileOp : Tosa_InferShapedTypeOp<"tile"> {
 // Operator: transpose
 //===----------------------------------------------------------------------===//
 def Tosa_TransposeOp : Tosa_InferShapedTypeOp<"transpose",
-                [DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
+                [DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface ,
+                                           ["reifyResultShapes"]>,
                  AllElementTypesMatch<["input1", "output"]>]> {
   let summary = "Transpose operator.";
 

mlir/include/mlir/Interfaces/InferTypeOpInterface.h

@@ -33,6 +33,10 @@ using ReifiedRankedShapedTypeDims = SmallVector<SmallVector<OpFoldResult>>;
 LogicalResult
 reifyResultShapes(OpBuilder &b, Operation *op,
                   ReifiedRankedShapedTypeDims &reifiedReturnShapes);
+FailureOr<SmallVector<OpFoldResult>>
+reifyShapeOfResult(OpBuilder &b, Operation *op, int resultIndex);
+FailureOr<OpFoldResult> reifyDimOfResult(OpBuilder &b, Operation *op,
+                                         int resultIndex, int dim);
 
 /// Adaptor class to abstract the differences between whether value is from
 /// a ShapedType or ShapedTypeComponents or DenseIntElementsAttribute.

mlir/include/mlir/Interfaces/InferTypeOpInterface.td

@@ -361,20 +361,76 @@ def ReifyRankedShapedTypeOpInterface :
   let methods = [
     InterfaceMethod<
       /*desc=*/[{
-        Reify the shape of the result of an operation (typically in terms of the
-        shape of its operands).
+        Reify the shapes of all the result of an operation (typically in terms 
+        of the shape of its operands).
 
         `reifiedReturnShapes` is populated with one vector per op result. Each
         of those vectors contains an OpFoldResult for each dimension of the
         shaped type. The given builder may be used to insert ops that compute
         result shapes.
 
-        If the shape of a particular result cannot be computed it must be empty.
+        If the shape of a particular result cannot be computed it in terms of
+        its operands it must be left empty. If any dimension of the result cannot
+        be computed it must be set to OpFoldResult().
       }],
       /*retTy=*/"::llvm::LogicalResult",
       /*methodName=*/"reifyResultShapes",
       /*args=*/(ins "::mlir::OpBuilder &":$builder,
-        "::mlir::ReifiedRankedShapedTypeDims &":$reifiedReturnShapes)
+        "::mlir::ReifiedRankedShapedTypeDims &":$reifiedReturnShapes),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{ return ::mlir::failure(); }]
+    >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Reify the shape of a single result of an operation (typically in terms 
+        of the shape of its operands).
+
+        Returns the shape of a single result of the operation as a
+        `SmallVector<OpFoldResult>`, one per dimension of the shaped type. The
+        given builder may be used to insert ops that compute result shapes.
+
+        If any dimension of the result cannot be computed it must be set to
+        OpFoldResult().
+      }],
+      /*retTy=*/"::llvm::FailureOr<::llvm::SmallVector<::mlir::OpFoldResult>>",
+      /*methodName=*/"reifyShapeOfResult",
+      /*args=*/(ins "::mlir::OpBuilder &":$builder,
+        "int":$resultIndex),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        ReifiedRankedShapedTypeDims reifiedShapes;
+        if (failed(cast<ReifyRankedShapedTypeOpInterface>($_op.getOperation()).reifyResultShapes(builder, reifiedShapes)))
+          return failure();
+        if (resultIndex < 0 || resultIndex >= static_cast<int>(reifiedShapes.size()))
+          return $_op.emitOpError("invalid result index");
+        return reifiedShapes[resultIndex];
+      }]
+    >,
+    InterfaceMethod<
+      /*desc=*/[{
+        Reify the shape of a dimension of a given result of an operation
+        (typically in terms of the shape of its operands).
+
+        Returns the shape of a specific dimension of a result of the operation as
+        an OpFoldResult. The given builder may be used to insert ops that compute
+        the shapes.
+
+        If the dimension of the result cannot be computed the method must return
+        `failure()`.
+      }],
+      /*retTy=*/"::llvm::FailureOr<::mlir::OpFoldResult>",
+      /*methodName=*/"reifyDimOfResult",
+      /*args=*/(ins "::mlir::OpBuilder &":$builder,
+        "int":$resultIndex, "int":$dim),
+      /*methodBody=*/"",
+      /*defaultImplementation=*/[{
+        auto shapes = cast<ReifyRankedShapedTypeOpInterface>($_op.getOperation()).reifyShapeOfResult(builder, resultIndex);
+        if (failed(shapes))
+          return failure();
+        if (dim < 0 || dim >= static_cast<int>((*shapes).size()))
+          return $_op.emitOpError("invalid dimension");
+        return (*shapes)[dim];
+      }]
     >
   ];
 }

mlir/lib/Dialect/MemRef/Transforms/ResolveShapedTypeResultDims.cpp

@@ -90,17 +90,16 @@ struct DimOfReifyRankedShapedTypeOpInterface : public OpRewritePattern<OpTy> {
     if (!dimIndex)
       return failure();
 
-    ReifiedRankedShapedTypeDims reifiedResultShapes;
-    if (failed(reifyResultShapes(rewriter, dimValue.getOwner(),
-                                 reifiedResultShapes)))
+    FailureOr<OpFoldResult> replacement = reifyDimOfResult(
+        rewriter, dimValue.getOwner(), dimValue.getResultNumber(), *dimIndex);
+    if (failed(replacement))
       return failure();
-    unsigned resultNumber = dimValue.getResultNumber();
-    // Do not apply pattern if the IR is invalid (dim out of bounds).
-    if ((size_t)(*dimIndex) >= reifiedResultShapes[resultNumber].size())
-      return rewriter.notifyMatchFailure(dimOp, "dimension is out of bounds");
-    Value replacement = getValueOrCreateConstantIndexOp(
-        rewriter, dimOp.getLoc(), reifiedResultShapes[resultNumber][*dimIndex]);
-    rewriter.replaceOp(dimOp, replacement);
+    // Check if the OpFoldResult is empty (unreifiable dimension).
+    if (!replacement.value())
+      return failure();
+    Value replacementVal = getValueOrCreateConstantIndexOp(
+        rewriter, dimOp.getLoc(), replacement.value());
+    rewriter.replaceOp(dimOp, replacementVal);
     return success();
   }
 };
@@ -166,12 +165,14 @@ namespace {
 struct ResolveRankedShapeTypeResultDimsPass final
     : public memref::impl::ResolveRankedShapeTypeResultDimsPassBase<
           ResolveRankedShapeTypeResultDimsPass> {
+  using Base::Base;
   void runOnOperation() override;
 };
 
 struct ResolveShapedTypeResultDimsPass final
     : public memref::impl::ResolveShapedTypeResultDimsPassBase<
           ResolveShapedTypeResultDimsPass> {
+  using Base::Base;
   void runOnOperation() override;
 };
 
@@ -195,14 +196,22 @@ void memref::populateResolveShapedTypeResultDimsPatterns(
 void ResolveRankedShapeTypeResultDimsPass::runOnOperation() {
   RewritePatternSet patterns(&getContext());
   memref::populateResolveRankedShapedTypeResultDimsPatterns(patterns);
-  if (failed(applyPatternsGreedily(getOperation(), std::move(patterns))))
+  auto result = applyPatternsGreedily(getOperation(), std::move(patterns));
+  if (errorOnPatternIterationLimit && failed(result)) {
+    getOperation()->emitOpError(
+        "dim operation resolution hit pattern iteration limit");
     return signalPassFailure();
+  }
 }
 
 void ResolveShapedTypeResultDimsPass::runOnOperation() {
   RewritePatternSet patterns(&getContext());
   memref::populateResolveRankedShapedTypeResultDimsPatterns(patterns);
   memref::populateResolveShapedTypeResultDimsPatterns(patterns);
-  if (failed(applyPatternsGreedily(getOperation(), std::move(patterns))))
+  auto result = applyPatternsGreedily(getOperation(), std::move(patterns));
+  if (errorOnPatternIterationLimit && failed(result)) {
+    getOperation()->emitOpError(
+        "dim operation resolution hit pattern iteration limit");
     return signalPassFailure();
+  }
 }