Merge pull request #374 from Xilinx/bump_to_37263b6c

[AutoBump] Merge with fixes of 37263b6 (Sep 04) (21)

Author: mgehre-amd

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

@@ -1579,7 +1579,7 @@ def Tosa_ConcatOp : Tosa_InferTensorTypeOp<"concat"> {
 //===----------------------------------------------------------------------===//
 // Operator: pad
 //===----------------------------------------------------------------------===//
-def Tosa_PadOp : Tosa_InferShapedTypeOp<"pad", [InferTensorType]> {
+def Tosa_PadOp : Tosa_InferShapedTypeOp<"pad"> {
   let summary = "Pads a tensor with value specified.";
 
   let description = [{
@@ -1619,13 +1619,6 @@ def Tosa_PadOp : Tosa_InferShapedTypeOp<"pad", [InferTensorType]> {
   let hasCanonicalizer = 1;
   let hasFolder = 1;
   let hasVerifier = 1;
-
-  let extraClassDeclaration = [{
-    /// Returns true when two result types are compatible for this op;
-    /// Method used by InferTypeOpInterface.
-    static bool isCompatibleReturnTypes(TypeRange l, TypeRange r);
-  }];
-
 }
 
 //===----------------------------------------------------------------------===//

mlir/lib/Dialect/Tosa/IR/TosaOps.cpp

@@ -800,8 +800,6 @@ LogicalResult tosa::PadOp::inferReturnTypeComponents(
     MLIRContext *context, ::std::optional<Location> location,
     PadOp::Adaptor adaptor,
     SmallVectorImpl<ShapedTypeComponents> &inferredReturnShapes) {
-
-  Type inputType = getElementTypeOrSelf(adaptor.getInput1());
   ShapeAdaptor inputShape(adaptor.getInput1().getType());
   ShapeAdaptor paddingShape(adaptor.getPadding().getType());
   SmallVector<int64_t> outputShape;
@@ -822,17 +820,15 @@ LogicalResult tosa::PadOp::inferReturnTypeComponents(
     }
 
     outputShape.resize(paddingShape.getDimSize(0), ShapedType::kDynamic);
-    inferredReturnShapes.push_back(
-        ShapedTypeComponents(outputShape, inputType));
+    inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
     return success();
   }
 
   DenseIntElementsAttr paddings;
   // If the paddings value is not a constant, all dimensions must be dynamic.
   if (!matchPattern(adaptor.getPadding(), m_Constant(&paddings))) {
     outputShape.resize(inputShape.getRank(), ShapedType::kDynamic);
-    inferredReturnShapes.push_back(
-        ShapedTypeComponents(outputShape, inputType));
+    inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
     return success();
   }
 
@@ -852,39 +848,21 @@ LogicalResult tosa::PadOp::inferReturnTypeComponents(
                           paddingValues[i * 2 + 1]);
   }
 
-  inferredReturnShapes.push_back(ShapedTypeComponents(outputShape, inputType));
+  inferredReturnShapes.push_back(ShapedTypeComponents(outputShape));
   return success();
 }
 
-LogicalResult PadOp::verify() {
-  ShapedType inputType = llvm::cast<ShapedType>(getInput1().getType());
-  if (inputType.hasRank() && inputType.getRank() == 0) {
-    return emitOpError() << "input tensor rank must not be 0";
-  }
+LogicalResult tosa::PadOp::verify() {
+  RankedTensorType inputType = getInput1().getType();
+  RankedTensorType outputType = getOutput().getType();
+  TensorType paddingType = getPadding().getType();
 
-  ShapedType paddingType = llvm::cast<ShapedType>(getPadding().getType());
-  if (paddingType.hasRank()) {
-    if (paddingType.getRank() != 2) {
-      return emitOpError() << "paddings must be a tensor of rank 2";
-    }
-    if (inputType.hasRank() && !paddingType.isDynamicDim(0) &&
-        inputType.getRank() != paddingType.getDimSize(0)) {
-      return emitOpError() << "paddings must be a tensor of shape ["
-                           << inputType.getRank() << ", 2]";
-    }
-    if (!paddingType.isDynamicDim(1) && paddingType.getDimSize(1) != 2) {
-      return emitOpError() << "paddings must be a tensor of shape ["
-                           << inputType.getRank() << ", 2]";
-    }
+  if (inputType.getRank() != outputType.getRank())
+    return emitOpError() << "expect same input and output tensor rank.";
+
+  if (paddingType.hasRank() && paddingType.getRank() != 2)
+    return emitOpError() << "expect 'padding' tensor rank equal to 2.";
 
-    DenseIntElementsAttr paddings;
-    if (matchPattern(getPadding(), m_Constant(&paddings))) {
-      if (llvm::any_of(paddings,
-                       [](auto val) { return val.getSExtValue() < 0; })) {
-        return emitOpError() << "number of pad elements must be positive";
-      }
-    }
-  }
   return success();
 }
 
@@ -1441,7 +1419,6 @@ REDUCE_SHAPE_INFER(tosa::ReduceProdOp)
 REDUCE_SHAPE_INFER(tosa::ReduceSumOp)
 #undef REDUCE_SHAPE_INFER
 COMPATIBLE_RETURN_TYPES(tosa::ConcatOp)
-COMPATIBLE_RETURN_TYPES(tosa::PadOp)
 #undef COMPATIBLE_RETURN_TYPES
 
 template <typename T>

mlir/test/Dialect/Tosa/invalid.mlir

@@ -105,83 +105,36 @@ func.func @test_pad_non_const(%arg0: tensor<13x21x3xf32>, %arg1: tensor<3x2xi32>
 
 // -----
 
-func.func @test_pad_non_const(%arg0: tensor<13x21x3xi8>, %arg1: tensor<i8>) -> tensor<?x?x?xi8> {
+func.func @test_pad_non_const(%arg0: tensor<13x21x3xi8>, %arg1: tensor<i8>) -> tensor<13x21x3xi8> {
   %0 = "tosa.const"() {value = dense<[[0, 0], [0, 1], [0, 1]]> : tensor<3x2xi32>} : () -> tensor<3x2xi32>
   // expected-error@+1 {{'tosa.pad' op pad_const of pad is not constant}}
-  %1 = tosa.pad %arg0, %0, %arg1 : (tensor<13x21x3xi8>, tensor<3x2xi32>, tensor<i8>) -> tensor<?x?x?xi8>
-  return %1 : tensor<?x?x?xi8>
+  %1 = tosa.pad %arg0, %0, %arg1 : (tensor<13x21x3xi8>, tensor<3x2xi32>, tensor<i8>) -> tensor<13x21x3xi8>
+  return %1 : tensor<13x21x3xi8>
 }
 
 // -----
 
-func.func @test_pad_output_shape_mismatch(%arg0: tensor<13x21x3xf32>) -> tensor<13x21x3xf32> {
-  %0 = "tosa.const"() {value = dense<[[1, 1], [1, 1], [1, 1]]> : tensor<3x2xi32>} : () -> tensor<3x2xi32>
-  // expected-error@+2 {{'tosa.pad' op failed to infer returned types}}
-  // expected-error@+1 {{'tosa.pad' op inferred type(s) 'tensor<15x23x5xf32>' are incompatible with return type(s) of operation 'tensor<13x21x3xf32>}}
-  %1 = tosa.pad %arg0, %0 : (tensor<13x21x3xf32>, tensor<3x2xi32>) -> tensor<13x21x3xf32>
-  return %1 : tensor<13x21x3xf32>
-}
-
-// -----
-
-func.func @test_pad_type_mismatch(%arg0: tensor<13x21x3xf32>) -> tensor<15x23x5xi32> {
-  %0 = "tosa.const"() {value = dense<[[1, 1], [1, 1], [1, 1]]> : tensor<3x2xi32>} : () -> tensor<3x2xi32>
-  // expected-error@+2 {{'tosa.pad' op failed to infer returned types}}
-  // expected-error@+1 {{'tosa.pad' op inferred type(s) 'tensor<15x23x5xf32>' are incompatible with return type(s) of operation 'tensor<15x23x5xi32>}}
-  %1 = tosa.pad %arg0, %0 : (tensor<13x21x3xf32>, tensor<3x2xi32>) -> tensor<15x23x5xi32>
-  return %1 : tensor<15x23x5xi32>
-}
-
-// -----
-
-func.func @test_pad_incorret_padding_rank(%arg0: tensor<13x21xf32>) -> tensor<13x21xf32> {
-  %0 = "tosa.const"() {value = dense<[0, 1]> : tensor<2xi32>} : () -> tensor<2xi32>
-  // expected-error@+1 {{'tosa.pad' op paddings must be a tensor of rank 2}}
-  %1 = tosa.pad %arg0, %0 : (tensor<13x21xf32>, tensor<2xi32>) -> tensor<13x21xf32>
-  return %1 : tensor<13x21xf32>
-}
-
-// -----
-
-func.func @test_pad_incorret_padding_shape(%arg0: tensor<13x21xf32>) -> tensor<13x21xf32> {
-  %0 = "tosa.const"() {value = dense<[[0, 0], [0, 1], [0, 1], [1, 1]]> : tensor<4x2xi32>} : () -> tensor<4x2xi32>
-  // expected-error@+1 {{'tosa.pad' op paddings must be a tensor of shape [2, 2]}}
-  %1 = tosa.pad %arg0, %0 : (tensor<13x21xf32>, tensor<4x2xi32>) -> tensor<13x21xf32>
-  return %1 : tensor<13x21xf32>
-}
-
-// -----
-
-func.func @test_pad_incorret_padding_shape(%arg0: tensor<13x21xf32>) -> tensor<13x21xf32> {
-  %0 = "tosa.const"() {value = dense<[[0, 0, 0, 1], [0, 1, 1, 1]]> : tensor<2x4xi32>} : () -> tensor<2x4xi32>
-  // expected-error@+1 {{'tosa.pad' op paddings must be a tensor of shape [2, 2]}}
-  %1 = tosa.pad %arg0, %0 : (tensor<13x21xf32>, tensor<2x4xi32>) -> tensor<13x21xf32>
-  return %1 : tensor<13x21xf32>
-}
-
-// -----
-
-func.func @test_pad_negative_padding(%arg0: tensor<13x21xf32>) -> tensor<?x?xf32> {
-  %0 = "tosa.const"() {value = dense<[[0, 0], [0, -1]]> : tensor<2x2xi32>} : () -> tensor<2x2xi32>
-  // expected-error@+1 {{'tosa.pad' op number of pad elements must be positive}}
-  %1 = tosa.pad %arg0, %0 : (tensor<13x21xf32>, tensor<2x2xi32>) -> tensor<?x?xf32>
-  return %1 : tensor<?x?xf32>
+func.func @test_pad_io_rank_mismatch(%arg0: tensor<13x21xf32>, %arg1: tensor<2x2xi32>) {
+  // expected-error@+1 {{'tosa.pad' op expect same input and output tensor rank.}}
+  %1 = tosa.pad %arg0, %arg1 : (tensor<13x21xf32>, tensor<2x2xi32>) -> tensor<13x21x3xf32>
+  return
 }
 
 // -----
 
-func.func @test_pad_incorrect_input(%arg0: f32, %arg1: i32) -> f32 {
-  // expected-error@+1 {{'tosa.pad' op operand #0 must be ranked tensor of number values, but got 'f32'}}
-  %1 = tosa.pad %arg0, %arg1 : (f32, i32) -> f32
-  return %1 : f32
+func.func @test_pad_invalid_padding_rank(%arg0: tensor<13x21xf32>, %arg1: tensor<2xi32>) {
+  // expected-error@+1 {{'tosa.pad' op expect 'padding' tensor rank equal to 2.}}
+  %1 = tosa.pad %arg0, %arg1 : (tensor<13x21xf32>, tensor<2xi32>) -> tensor<13x21xf32>
+  return
 }
 
 // -----
 
-func.func @test_pad_zero_rank_input(%arg0: tensor<f32>, %arg1: tensor<i32>) -> tensor<f32> {
-  // expected-error@+1 {{'tosa.pad' op input tensor rank must not be 0}}
-  %1 = tosa.pad %arg0, %arg1 : (tensor<f32>, tensor<i32>) -> tensor<f32>
-  return %1 : tensor<f32>
+func.func @test_pad_invalid_padConst_rank(%arg0: tensor<13x21xf32>, %arg1: tensor<2x2xi32>) {
+  %0 = "tosa.const"() {value = dense<3.14> : tensor<1xf32>} : () -> tensor<1xf32>
+  // expected-error@+1 {{'tosa.pad' op operand #2 must be 0D tensor of number values, but got 'tensor<1xf32>'}}
+  %1 = tosa.pad %arg0, %arg1, %0 : (tensor<13x21xf32>, tensor<2x2xi32>, tensor<1xf32>) -> tensor<13x21xf32>
+  return
 }
 
 // -----