[MLIR][Shape] Allow `get_extent` to operate on extent tensors and indices

[frgossen]

Summary

Depends On D84433

mlir/include/mlir/Dialect/Shape/IR/ShapeOps.td

 
 def Shape_GetExtentOp : Shape_Op<"get_extent", [NoSideEffect]> {
   let summary = "Gets the specified extent from a shape or extent tensor";
   let description = [{
     Gets the extent indexed by `dim` from the `shape` operand. If the shape is
     an error then it returns an error size.
   }];
   let arguments = (ins Shape_ShapeOrExtentTensorType:$shape,
-                       Shape_SizeType:$dim);
-  let results = (outs Shape_SizeType:$extent);
-  let assemblyFormat = "$shape `,` $dim `:` type($shape)  attr-dict";
+                       Shape_SizeOrIndexType:$dim);
+  let results = (outs Shape_SizeOrIndexType:$extent);
+  let assemblyFormat = "$shape `,` $dim `:` type($shape) `,` type($dim) `->` "
+                       "type($extent) attr-dict";
 
   let builders = [
     // Builder that allows passing a constant dimension as a simple integer.
     OpBuilder<"OpBuilder &builder, OperationState &result, Value shape, "
               "int64_t dim">
   ];
 
   let extraClassDeclaration = [{
     /// Get the `dim` value as integer if it is constant.
     Optional<int64_t> getConstantDim();
   }];
 
   let hasFolder = 1;
+  let verifier = [{ return ::verify(*this); }];
 }
 
 def Shape_IndexToSizeOp : Shape_Op<"index_to_size", [NoSideEffect]> {
   let summary = "Converts a standard index to a shape size";
   let description = [{
     Converts a standard index to a `shape.size`. This operation and its
     inverse, `size_to_index`, facilitate index conversion between the standard
     and the shape dialect.

mlir/lib/Dialect/Shape/IR/Shape.cpp

 //===- Shape.cpp - MLIR Shape Operations ----------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
 #include "mlir/Dialect/Shape/IR/Shape.h"
 
 #include "mlir/Dialect/Traits.h"
 #include "mlir/IR/Builders.h"
 #include "mlir/IR/DialectImplementation.h"
 #include "mlir/IR/PatternMatch.h"
 #include "mlir/IR/StandardTypes.h"
 #include "mlir/include/mlir/Dialect/StandardOps/IR/Ops.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/Support/raw_ostream.h"
 
 using namespace mlir;
 using namespace mlir::shape;
 
 namespace {
 #include "ShapeCanonicalization.inc"
OpFoldResult FromExtentsOp::fold(ArrayRef<Attribute> operands) {
   Builder builder(getContext());
   return builder.getIndexTensorAttr(extents);
 }
 
 //===----------------------------------------------------------------------===//
 // GetExtentOp
 //===----------------------------------------------------------------------===//
 
+static LogicalResult verify(GetExtentOp op) {
+  Type shapeTy = op.shape().getType();
+  Type dimTy = op.dim().getType();
+  Type extentTy = op.extent().getType();
+  bool errorPropagationPossible =
+      shapeTy.isa<ShapeType>() || dimTy.isa<SizeType>();
+  if (errorPropagationPossible) {
+    if (!extentTy.isa<SizeType>())
+      op.emitError()
+          << "if at least one of the operands can hold error values then the "
+             "result must be of type `size` to propagate them";
+  } else {
+    if (extentTy.isa<SizeType>())
+      op.emitError() << "if none of the operands can hold error values then "
+                        "the result must be of type `index`";
+  }
+  return success();
+}
+
 Optional<int64_t> GetExtentOp::getConstantDim() {
-  if (auto constSizeOp = dim().getDefiningOp<ConstSizeOp>()) {
+  if (auto constSizeOp = dim().getDefiningOp<ConstSizeOp>())
     return constSizeOp.value().getLimitedValue();
-  }
+  if (auto constantOp = dim().getDefiningOp<ConstantOp>())
+    return constantOp.value().cast<IntegerAttr>().getInt();
   return llvm::None;
 }
 
 OpFoldResult GetExtentOp::fold(ArrayRef<Attribute> operands) {
   auto elements = operands[0].dyn_cast_or_null<DenseIntElementsAttr>();
   if (!elements)
     return nullptr;
   Optional<int64_t> dim = getConstantDim();
   if (!dim.hasValue())
     return nullptr;
   if (dim.getValue() >= elements.getNumElements())
     return nullptr;
   return elements.getValue({(uint64_t)dim.getValue()});
 }
 
 void GetExtentOp::build(OpBuilder &builder, OperationState &result, Value shape,
                         int64_t dim) {
   auto loc = result.location;
   auto dimAttr = builder.getIndexAttr(dim);
-  Value dimValue = builder.create<ConstSizeOp>(loc, dimAttr);
-  build(builder, result, shape, dimValue);
+  if (shape.getType().isa<ShapeType>()) {
+    Value dim = builder.create<ConstSizeOp>(loc, dimAttr);
+    build(builder, result, builder.getType<SizeType>(), shape, dim);
+  } else {
+    Value dim =
+        builder.create<ConstantOp>(loc, builder.getIndexType(), dimAttr);
+    build(builder, result, builder.getIndexType(), shape, dim);
+  }
 }
 
 //===----------------------------------------------------------------------===//
 // RankOp
 //===----------------------------------------------------------------------===//
 
 static LogicalResult verify(shape::RankOp op) {
   Type argTy = op.shape().getType();

mlir/test/Conversion/ShapeToStandard/shape-to-standard.mlir

 }
 
 // -----
 
 // Express `get_extent` as `std.dim` when it relies directly on the outcome of a
 // `shape_of` operation.
 // CHECK-LABEL: @get_extent_shape_of
 // CHECK-SAME:  (%[[ARG:.*]]: tensor<2x3xf32>, %[[IDX:.*]]: index) -> index
-func @get_extent_shape_of(%arg : tensor<2x3xf32>, %idx : !shape.size)
-    -> !shape.size {
+func @get_extent_shape_of(%arg : tensor<2x3xf32>, %idx : index) -> index {
   // CHECK: %[[RESULT:.*]] = dim %[[ARG]], %[[IDX]] : tensor<2x3xf32>
   // CHECK: return %[[RESULT]] : index
   %shape = shape.shape_of %arg : tensor<2x3xf32> -> tensor<?xindex>
-  %result = shape.get_extent %shape, %idx : tensor<?xindex>
-  return %result : !shape.size
+  %result = shape.get_extent %shape, %idx : tensor<?xindex>, index -> index
+  return %result : index
 }
 
 // -----
 
-// Express `get_extent` as `std.extract_element` when it relies directly on the
-// outcome of a `from_extent_tensor` operation.
+// Express `get_extent` as `std.extract_element`.
 // CHECK-LABEL: @get_extent_from_extent_tensor
 // CHECK-SAME: (%[[EXTENTS:.*]]: tensor<?xindex>, %[[IDX:.*]]: index) -> index
-func @get_extent_from_extent_tensor(%extents : tensor<?xindex>,
-                                    %idx : !shape.size) -> !shape.size {
+func @get_extent_from_extent_tensor(%extents : tensor<?xindex>, %idx : index)
+    -> index {
   // CHECK: %[[RESULT:.*]] = extract_element %[[EXTENTS]][%[[IDX]]] : tensor<?xindex>
   // CHECK: return %[[RESULT]] : index
-  %shape = shape.from_extent_tensor %extents : tensor<?xindex>
-  %result = shape.get_extent %shape, %idx : !shape.shape
-  return %result : !shape.size
+  %result = shape.get_extent %extents, %idx : tensor<?xindex>, index -> index
+  return %result : index
 }
 
 // -----
 
 // Lower `any` to its first operand.
 // CHECK-LABEL: @any_of_three
 // CHECK-SAME:  (%[[A:.*]]: tensor<?xindex>, %[[B:.*]]: tensor<?xindex>, %[[C:.*]]: tensor<?xindex>) -> tensor<?xindex>
 func @any_of_three(%a : tensor<?xindex>,

mlir/test/Dialect/Shape/canonicalize.mlir

func @nonfoldable_num_elements(%shape : !shape.shape) -> !shape.size {
   %num_elements = shape.num_elements %shape
   return %num_elements : !shape.size
 }
 
 // -----
 
 // Basic folding.
 // CHECK-LABEL: func @basic
+func @basic() -> index {
+  // CHECK: constant 2 : index
+  %0 = shape.const_shape [0, 1, 2] : tensor<?xindex>
+  %c2 = constant 2 : index
+  %1 = shape.get_extent %0, %c2 : tensor<?xindex>, index -> index
+  return %1 : index
+}
+
+// -----
+
+// Should not fold.
+// CHECK-LABEL: func @out_of_bounds
+func @out_of_bounds() -> index {
+  // CHECK: shape.const_shape
+  // CHECK: shape.get_extent
+  %0 = shape.const_shape [0, 1, 2] : tensor<?xindex>
+  %c3 = constant 3 : index
+  %1 = shape.get_extent %0, %c3 : tensor<?xindex>, index -> index
+  return %1 : index
+}
+
+// -----
+
+// Should not fold.
+// CHECK-LABEL: func @not_const
+func @not_const(%arg0: tensor<?xindex>) -> index {
+  // CHECK: shape.get_extent
+  %c3 = constant 3 : index
+  %0 = shape.get_extent %arg0, %c3 : tensor<?xindex>, index -> index
+  return %0 : index
+}
+
+// -----
+
+// Basic folding.
+// CHECK-LABEL: func @basic
 func @basic() -> !shape.size {
   // CHECK: shape.const_size 2
-  %0 = shape.const_shape [0, 1, 2] : tensor<?xindex>
+  %0 = shape.const_shape [0, 1, 2] : !shape.shape
   %c2 = shape.const_size 2
-  %1 = shape.get_extent %0, %c2 : tensor<?xindex>
+  %1 = shape.get_extent %0, %c2 : !shape.shape, !shape.size -> !shape.size
   return %1 : !shape.size
 }
 
 // -----
 
 // Should not fold.
 // CHECK-LABEL: func @out_of_bounds
 func @out_of_bounds() -> !shape.size {
   // CHECK: shape.const_shape
   // CHECK: shape.get_extent
-  %0 = shape.const_shape [0, 1, 2] : tensor<?xindex>
+  %0 = shape.const_shape [0, 1, 2] : !shape.shape
   %c3 = shape.const_size 3
-  %1 = shape.get_extent %0, %c3 : tensor<?xindex>
+  %1 = shape.get_extent %0, %c3 : !shape.shape, !shape.size -> !shape.size
   return %1 : !shape.size
 }
 
 // -----
 
 // Should not fold.
 // CHECK-LABEL: func @not_const
-func @not_const(%arg0: tensor<?xindex>) -> !shape.size {
+func @not_const(%arg0 : !shape.shape) -> !shape.size {
   // CHECK: shape.get_extent
   %c3 = shape.const_size 3
-  %0 = shape.get_extent %arg0, %c3 : tensor<?xindex>
+  %0 = shape.get_extent %arg0, %c3 : !shape.shape, !shape.size -> !shape.size
   return %0 : !shape.size
 }
 
-
 // -----
 // cstr_eq with non-constant but known equal shapes can be removed.
 // CHECK-LABEL: func @f
 func @f(%arg0 : !shape.shape) {
   // CHECK-NEXT: shape.const_witness true
   // CHECK-NEXT: consume.witness
   // CHECK-NEXT: return
   %0 = shape.cstr_eq %arg0, %arg0, %arg0

mlir/test/Dialect/Shape/invalid.mlir

 
 // -----
 
 func @rank(%arg : tensor<?xindex>) {
   // expected-error@+1 {{if operand is of type `tensor<?xindex>` then the result must be an index}}
   %1 = shape.rank %arg : tensor<?xindex> -> !shape.size
 }
 
+// -----
+
+func @get_extent_error_free(%arg : tensor<?xindex>) -> !shape.size {
+  %c0 = constant 0 : index
+  // expected-error@+1 {{if none of the operands can hold error values then the result must be of type `index`}}
+  %result = shape.get_extent %arg, %c0 : tensor<?xindex>, index -> !shape.size
+  return %result : !shape.size
+}
+
+// -----
+
+func @get_extent_error_possible(%arg : tensor<?xindex>) -> index {
+  %c0 = shape.const_size 0
+  // expected-error@+1 {{if at least one of the operands can hold error values then the result must be of type `size` to propagate them}}
+  %result = shape.get_extent %arg, %c0 : tensor<?xindex>, !shape.size -> index
+  return %result : index
+}
+

mlir/test/Dialect/Shape/ops.mlir

 
 func @shape_eq_on_mixed(%a : tensor<?xindex>, %b : !shape.shape) -> i1 {
   %result = shape.shape_eq %a, %b : tensor<?xindex>, !shape.shape
   return %result : i1
 }
 
 func @get_extent_on_shape(%arg : !shape.shape) -> !shape.size {
   %c0 = shape.const_size 0
-  %result = shape.get_extent %arg, %c0 : !shape.shape
+  %result = shape.get_extent %arg, %c0 :
+      !shape.shape, !shape.size -> !shape.size
   return %result : !shape.size
 }
 
-func @get_extent_on_extent_tensor(%arg : tensor<?xindex>) -> !shape.size {
+func @get_extent_on_extent_tensor(%arg : tensor<?xindex>) -> index {
+  %c0 = constant 0 : index
+  %result = shape.get_extent %arg, %c0 : tensor<?xindex>, index -> index
+  return %result : index
+}
+
+func @get_extent_on_mixed_operands(%arg : tensor<?xindex>) -> !shape.size {
   %c0 = shape.const_size 0
-  %result = shape.get_extent %arg, %c0 : tensor<?xindex>
+  %result = shape.get_extent %arg, %c0 : tensor<?xindex>, !shape.size -> !shape.size
   return %result : !shape.size
 }
 
 func @any() {
   %0 = shape.const_shape [1, 2, 3] : !shape.shape
   %1 = shape.const_shape [4, 5, 6] : !shape.shape
   %2 = shape.any %0, %1 : !shape.shape
   %3 = shape.const_shape [1, 2, 3] : tensor<?xindex>
   %4 = shape.const_shape [4, 5, 6] : tensor<?xindex>
   %5 = shape.any %3, %4 : tensor<?xindex>
   return
 }