[mlir][tosa] Require operand/result tensors of at least rank 1 for some operations #131335
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…me operations This commit updates the following operations (operands/results) to be of at least rank 1 such that it aligns wih the expectations of the specification: - ARGMAX (input) - REDUCE_ALL (input/output) - REDUCE_ANY (input/output) - REDUCE_MAX (input/output) - REDUCE_MIN (input/output) - REDUCE_PRODUCT (input/output) - REDUCE_SUM (input/output) - CONCAT (each input in input1/output) - PAD (input1/output) - REVERSE (input1/output) - SLICE (input1/output) - TILE (input1/output) - TRANSPOSE (input1/output) In addition to this change, PAD has been updated to allow unranked tensors for input1/output, inline with other operations. Change-Id: I703c398e91c0a68366bab8e0778eb01a80588ebe
|
@llvm/pr-subscribers-mlir-tosa @llvm/pr-subscribers-mlir Author: Luke Hutton (lhutton1) ChangesThis commit updates the following operations (operands/results) to be of at least rank 1 such that it aligns wih the expectations of the specification:
In addition to this change, PAD has been updated to allow unranked tensors for input1/output, inline with other operations. Full diff: https://github.com/llvm/llvm-project/pull/131335.diff 5 Files Affected:
diff --git a/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td b/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td
index b79993f48b379..0c99dd6130c2a 100644
--- a/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td
+++ b/mlir/include/mlir/Dialect/Tosa/IR/TosaOps.td
@@ -41,7 +41,7 @@ def Tosa_ArgMaxOp : Tosa_InferShapedTypeOp<"argmax"> {
}];
let arguments = (ins
- Tosa_Tensor: $input,
+ Tosa_TensorAtLeast1D: $input,
I32Attr: $axis,
DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
@@ -1629,12 +1629,12 @@ def Tosa_ReduceAllOp : Tosa_InferTensorTypeOp<"reduce_all"> {
}];
let arguments = (ins
- Tosa_Tensor:$input,
+ Tosa_TensorAtLeast1D:$input,
I32Attr:$axis
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -1668,12 +1668,12 @@ def Tosa_ReduceAnyOp : Tosa_InferTensorTypeOp<"reduce_any"> {
}];
let arguments = (ins
- Tosa_Tensor:$input,
+ Tosa_TensorAtLeast1D:$input,
I32Attr:$axis
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -1707,13 +1707,13 @@ def Tosa_ReduceMaxOp : Tosa_InferTensorTypeOp<"reduce_max"> {
}];
let arguments = (ins
- Tosa_Tensor:$input,
+ Tosa_TensorAtLeast1D:$input,
I32Attr:$axis,
DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -1748,13 +1748,13 @@ def Tosa_ReduceMinOp : Tosa_InferTensorTypeOp<"reduce_min"> {
}];
let arguments = (ins
- Tosa_Tensor:$input,
+ Tosa_TensorAtLeast1D:$input,
I32Attr:$axis,
DefaultValuedAttr<Tosa_NanPropagationAttr, "\"PROPAGATE\"">:$nan_mode
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -1789,12 +1789,12 @@ def Tosa_ReduceProductOp : Tosa_InferTensorTypeOp<"reduce_product"> {
}];
let arguments = (ins
- Tosa_Tensor:$input,
+ Tosa_TensorAtLeast1D:$input,
I32Attr:$axis
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -1828,12 +1828,12 @@ def Tosa_ReduceSumOp : Tosa_InferTensorTypeOp<"reduce_sum"> {
}];
let arguments = (ins
- Tosa_Tensor:$input,
+ Tosa_TensorAtLeast1D:$input,
I32Attr:$axis
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -1872,12 +1872,12 @@ def Tosa_ConcatOp : Tosa_InferTensorTypeOp<"concat"> {
}];
let arguments = (ins
- Variadic<Tosa_Tensor>:$input1,
+ Variadic<Tosa_TensorAtLeast1D>:$input1,
I32Attr:$axis
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -1923,13 +1923,13 @@ def Tosa_PadOp : Tosa_InferShapedTypeOp<"pad"> {
}];
let arguments = (ins
- Tosa_RankedTensor:$input1,
+ Tosa_TensorAtLeast1D:$input1,
Tosa_Shape:$padding,
Tosa_ScalarTensor:$pad_const
);
let results = (outs
- Tosa_RankedTensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -1996,12 +1996,12 @@ def Tosa_ReverseOp: Tosa_Op<"reverse", [
}];
let arguments = (ins
- Tosa_Tensor:$input1,
+ Tosa_TensorAtLeast1D:$input1,
I32Attr:$axis
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -2028,13 +2028,13 @@ def Tosa_SliceOp : Tosa_InferShapedTypeOp<"slice"> {
}];
let arguments = (ins
- Tosa_Tensor:$input1,
+ Tosa_TensorAtLeast1D:$input1,
Tosa_Shape:$start,
Tosa_Shape:$size
);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -2058,11 +2058,11 @@ def Tosa_TileOp : Tosa_InferShapedTypeOp<"tile"> {
}];
let arguments = (ins
- Tosa_Tensor:$input1,
+ Tosa_TensorAtLeast1D:$input1,
Tosa_Shape:$multiples);
let results = (outs
- Tosa_Tensor:$output
+ Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
@@ -2093,12 +2093,12 @@ def Tosa_TransposeOp : Tosa_InferShapedTypeOp<"transpose",
}];
let arguments = (ins
- Tosa_Tensor:$input1,
+ Tosa_TensorAtLeast1D:$input1,
DenseI32ArrayAttr:$perms
);
let results = (
- outs Tosa_Tensor:$output
+ outs Tosa_TensorAtLeast1D:$output
);
list<Availability> availability = [
diff --git a/mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td b/mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td
index 0038d8c386ca7..67011f22fbe2a 100644
--- a/mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td
+++ b/mlir/include/mlir/Dialect/Tosa/IR/TosaTypesBase.td
@@ -101,6 +101,10 @@ def AllDimensionsAreSizeOne : And<[
IsRankedTensorTypePred,
CPred<"::llvm::all_of(::llvm::cast<::mlir::RankedTensorType>($_self).getShape(), [](auto v) { return v == 1; })">]>;
+def AtLeastRankOne : And<[
+ IsRankedTensorTypePred,
+ CPred<"::llvm::cast<::mlir::RankedTensorType>($_self).getRank() >= 1">]>;
+
class TosaTensorOf<
list<Type> allowedTypes, string summary = "tosa-conformant tensor">
: TensorOf<allowedTypes, [Or<[HasNo0Dimensions, IsUnrankedTensorTypePred]>], summary>;
@@ -183,6 +187,9 @@ def Tosa_TensorUpto4D : AnyTypeOf<[
def Tosa_Int32TensorUpto4D : AnyTypeOf<[
Tosa_UnrankedTensor, TosaTensorRankOf<[Tosa_Int32], [0,1,2,3,4]>]>;
+def Tosa_TensorAtLeast1D : AnyTypeOf<[
+ Tosa_UnrankedTensor, TosaRankedTensorOf<[Tosa_AnyNumber], [AtLeastRankOne]>], "tosa-conformant tensor of at least rank 1", "::mlir::TensorType">;
+
//===----------------------------------------------------------------------===//
// Generic scalar, vector, or tensor of a particular type.
//===----------------------------------------------------------------------===//
diff --git a/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp b/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
index 97a3009a20302..cdba332792eb0 100644
--- a/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
+++ b/mlir/lib/Dialect/Tosa/IR/TosaOps.cpp
@@ -1354,8 +1354,13 @@ LogicalResult tosa::PadOp::verify() {
}
}
- RankedTensorType inputType = getInput1().getType();
- RankedTensorType outputType = getOutput().getType();
+ RankedTensorType inputType =
+ llvm::dyn_cast<RankedTensorType>(getInput1().getType());
+ RankedTensorType outputType =
+ llvm::dyn_cast<RankedTensorType>(getOutput().getType());
+ if (!inputType || !outputType)
+ return success();
+
auto paddingRank = cast<tosa::shapeType>(getPadding().getType()).getRank();
if (inputType.getRank() != outputType.getRank())
diff --git a/mlir/test/Dialect/Tosa/canonicalize.mlir b/mlir/test/Dialect/Tosa/canonicalize.mlir
index 3bc438e465e1d..077a6cee0a1bb 100644
--- a/mlir/test/Dialect/Tosa/canonicalize.mlir
+++ b/mlir/test/Dialect/Tosa/canonicalize.mlir
@@ -915,29 +915,6 @@ func.func @fold_abs_abs(%arg0: tensor<?x1xf32>) -> tensor<?x1xf32> {
// -----
-// CHECK-LABEL: @fold_reduce_rank_zero
-func.func @fold_reduce_rank_zero() {
- // CHECK-NOT: tosa.reduce_min
- // CHECK-NOT: tosa.reverse
- %0 = tensor.empty() : tensor<i32>
- %1 = tosa.reduce_min %0 {axis = 0 : i32} : (tensor<i32>) -> tensor<i32>
- %2 = tosa.reverse %0 {axis = 0 : i32} : (tensor<i32>) -> tensor<i32>
- return
-}
-
-// -----
-
-// CHECK-LABEL: @fold_tile_rank_zero
-func.func nested @fold_tile_rank_zero() -> tensor<i32> {
- // CHECK-NOT: tosa.tile
- %0 = tensor.empty() : tensor<i32>
- %cst = tosa.const_shape { values = dense<> : tensor<0xindex> } : () -> !tosa.shape<0>
- %1 = tosa.tile %0, %cst : (tensor<i32>, !tosa.shape<0>) -> tensor<i32>
- return %1 : tensor<i32>
-}
-
-// -----
-
// CHECK-LABEL: @reshape_quant_nofold
// check that segfault is fixed
func.func @reshape_quant_nofold() -> tensor<1x1x1x1xi32> {
@@ -1015,12 +992,12 @@ func.func @cast_quant_nofold() -> tensor<!quant.uniform<i8:f32, 3.07574046018999
// -----
// CHECK-LABEL: @reverse_quant_fold
-func.func @reverse_quant_fold() -> tensor<!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>> {
- // CHECK: %[[CST:.*]] = "tosa.const"() <{values = dense<0> : tensor<i8>}> : () -> tensor<!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>
+func.func @reverse_quant_fold() -> tensor<1x!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>> {
+ // CHECK: %[[CST:.*]] = "tosa.const"() <{values = dense<0> : tensor<1xi8>}> : () -> tensor<1x!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>
// CHECK: return %[[CST]]
- %0 = "tosa.const"() {values = dense<0> : tensor<i8>} : () -> tensor<!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>
- %1 = "tosa.reverse"(%0) { axis = 0 : i32 } : (tensor<!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>) -> tensor<!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>
- return %1 : tensor<!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>
+ %0 = "tosa.const"() {values = dense<0> : tensor<1xi8>} : () -> tensor<1x!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>
+ %1 = "tosa.reverse"(%0) { axis = 0 : i32 } : (tensor<1x!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>) -> tensor<1x!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>
+ return %1 : tensor<1x!quant.uniform<i8:f32, 3.0757404601899907E-5:-128>>
}
// -----
diff --git a/mlir/test/Dialect/Tosa/invalid.mlir b/mlir/test/Dialect/Tosa/invalid.mlir
index a488c051dcd3b..2dc749422c12d 100644
--- a/mlir/test/Dialect/Tosa/invalid.mlir
+++ b/mlir/test/Dialect/Tosa/invalid.mlir
@@ -452,9 +452,9 @@ func.func @test_reduce_sum_invalid_axis(%arg0 : tensor<2x3x4xf32>) -> () {
// -----
-func.func @test_reduce_min_invalid_output_rank(%arg0 : tensor<i32>) -> () {
+func.func @test_reduce_min_invalid_output_rank(%arg0 : tensor<1xi32>) -> () {
// expected-error@+1 {{'tosa.reduce_min' op expect output tensor rank to be equal to input tensor rank}}
- %0 = tosa.reduce_min %arg0 {axis = 0 : i32} : (tensor<i32>) -> tensor<1x10xi32>
+ %0 = tosa.reduce_min %arg0 {axis = 0 : i32} : (tensor<1xi32>) -> tensor<1x10xi32>
return
}
@@ -1852,3 +1852,72 @@ func.func @test_maxpool2d_unexpected_output_width(%arg0: tensor<1x32x32x8xf32>)
(tensor<1x32x32x8xf32>) -> tensor<1x32x2x8xf32>
return %0 : tensor<1x32x2x8xf32>
}
+
+// -----
+
+func.func @test_scalar_argmax(%arg0: tensor<i32>) -> tensor<i32> {
+ // expected-error@+1 {{'tosa.argmax' op operand #0 must be tosa-conformant tensor of at least rank 1, but got 'tensor<i32>'}}
+ %0 = tosa.argmax %arg0 {axis = 0 : i32} : (tensor<i32>) -> tensor<i32>
+ return %0 : tensor<i32>
+}
+
+// -----
+
+func.func @test_scalar_reduce_all(%arg0: tensor<i1>) -> tensor<i1> {
+ // expected-error@+1 {{'tosa.reduce_all' op operand #0 must be tosa-conformant tensor of at least rank 1, but got 'tensor<i1>'}}
+ %0 = tosa.reduce_all %arg0 {axis = 0 : i32} : (tensor<i1>) -> tensor<i1>
+ return %0 : tensor<i1>
+}
+
+// -----
+
+func.func @test_scalar_inputs_concat(%arg0: tensor<f32>, %arg1: tensor<f32>) -> tensor<2xf32> {
+ // expected-error@+1 {{'tosa.concat' op operand #0 must be variadic of tosa-conformant tensor of at least rank 1, but got 'tensor<f32>'}}
+ %0 = tosa.concat %arg0, %arg1 {axis = 0 : i32} : (tensor<f32>, tensor<f32>) -> tensor<2xf32>
+ return %0 : tensor<2xf32>
+}
+
+// -----
+
+func.func @test_scalar_pad(%arg0: tensor<f32>) -> tensor<f32> {
+ %0 = "tosa.const"() {values = dense<3.14> : tensor<1xf32>} : () -> tensor<1xf32>
+ %padding = tosa.const_shape {values = dense<0> : tensor<6xindex>} : () -> !tosa.shape<6>
+ // expected-error@+1 {{'tosa.pad' op operand #0 must be tosa-conformant tensor of at least rank 1, but got 'tensor<f32>'}}
+ %1 = tosa.pad %arg0, %padding, %0 : (tensor<f32>, !tosa.shape<6>, tensor<1xf32>) -> tensor<f32>
+ return %1 : tensor<f32>
+}
+
+// -----
+
+func.func @test_scalar_reverse(%arg0: tensor<f32>) -> tensor<f32> {
+ // expected-error@+1 {{'tosa.reverse' op operand #0 must be tosa-conformant tensor of at least rank 1, but got 'tensor<f32>'}}
+ %0 = tosa.reverse %arg0 {axis = 0: i32} : (tensor<f32>) -> tensor<f32>
+ return %arg0 : tensor<f32>
+}
+
+// -----
+
+func.func @test_scalar_slice(%arg0: tensor<f32>) -> tensor<f32> {
+ %0 = tosa.const_shape {values = dense<[]> : tensor<0xindex>} : () -> !tosa.shape<0>
+ %1 = tosa.const_shape {values = dense<[]> : tensor<0xindex>} : () -> !tosa.shape<0>
+ // expected-error@+1 {{'tosa.slice' op operand #0 must be tosa-conformant tensor of at least rank 1, but got 'tensor<f32>'}}
+ %2 = tosa.slice %arg0, %0, %1 : (tensor<f32>, !tosa.shape<0>, !tosa.shape<0>) -> tensor<f32>
+ return %2 : tensor<f32>
+}
+
+// -----
+
+func.func @test_scalar_tile(%arg0: tensor<f32>) -> tensor<*xf32> {
+ %cst = tosa.const_shape { values = dense<[]> : tensor<0xindex> } : () -> !tosa.shape<0>
+ // expected-error@+1 {{'tosa.tile' op operand #0 must be tosa-conformant tensor of at least rank 1, but got 'tensor<f32>'}}
+ %0 = tosa.tile %arg0, %cst: (tensor<f32>, !tosa.shape<0>) -> tensor<*xf32>
+ return %0 : tensor<*xf32>
+}
+
+// -----
+
+func.func @test_scalar_output_transpose(%arg0: tensor<*xf32>) -> tensor<f32> {
+ // expected-error@+1 {{'tosa.transpose' op result #0 must be tosa-conformant tensor of at least rank 1, but got 'tensor<f32>'}}
+ %1 = tosa.transpose %arg0 {perms = array<i32: 2, 0, 1>} : (tensor<*xf32>) -> tensor<f32>
+ return %1 : tensor<f32>
+}
|
GeorgeARM
approved these changes
Mar 14, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This commit updates the following operations (operands/results) to be of at least rank 1 such that it aligns with the expectations of the specification:
In addition to this change, PAD has been updated to allow unranked tensors for input1/output, inline with other operations.