Skip to content

[mlir][linalg] Add folder for linalg.index #136640

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Apr 22, 2025
Merged

[mlir][linalg] Add folder for linalg.index #136640

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
492439a
[mlir][linalg] Add folder for `linalg.index`
kuhar Apr 22, 2025
b3e5afe
Update test
kuhar Apr 22, 2025
cbcdb33
Simplify
kuhar Apr 22, 2025
12ba5a3
Add assert message
kuhar Apr 22, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions mlir/include/mlir/Dialect/Linalg/IR/LinalgOps.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,6 +88,7 @@ def Linalg_IndexOp : Linalg_Op<"index", [Pure]>,

let assemblyFormat = [{ $dim attr-dict `:` type($result) }];
let hasVerifier = 1;
let hasFolder = 1;
}

def Linalg_SoftmaxOp : Linalg_Op<"softmax",
Expand Down
13 changes: 13 additions & 0 deletions mlir/lib/Dialect/Linalg/IR/LinalgOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2283,6 +2283,19 @@ LogicalResult IndexOp::verify() {
return success();
}

OpFoldResult IndexOp::fold(FoldAdaptor adaptor) {
auto linalgOp = cast<LinalgOp>((*this)->getParentOp());

// Index of unit dims is always 0.
SmallVector<int64_t, 4> loopBounds = linalgOp.getStaticLoopRanges();
uint64_t dim = getDim();
assert(dim < loopBounds.size());
if (loopBounds[dim] == 1)
return IntegerAttr::get(IndexType::get(getContext()), 0);

return OpFoldResult{};
}

/////// Operations corresponding to library calls defined with Tablegen ////////

#include "mlir/Dialect/Linalg/IR/LinalgNamedStructuredOps.yamlgen.cpp.inc"
Expand Down
80 changes: 80 additions & 0 deletions mlir/test/Dialect/Linalg/canonicalize.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -305,6 +305,86 @@ func.func @self_copy(%arg0 : memref<2x3x?x4xf32>) {
}

// -----

// CHECK: func @fold_linalg_index_tensor_static
func.func @fold_linalg_index_tensor_static(%0: tensor<4x16xi32>, %1: tensor<1x16xi32>,
%2: tensor<4x1xi32>) -> tensor<4x1xi32> {
// CHECK-NEXT: linalg.generic
// CHECK: %[[IDX_0:.+]] = linalg.index 0 : index
// CHECK-NOT: linalg.index 1
// CHECK: %[[IDX_2:.+]] = linalg.index 2 : index
// CHECK: %[[ADD:.+]] = arith.addi %[[IDX_0]], %[[IDX_2]]
// CHECK: %[[CAST:.+]] = arith.index_cast %[[ADD]]
// CHECK: linalg.yield %[[CAST]]
%res = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2) -> (d0, d2)>,
affine_map<(d0, d1, d2) -> (d1, d2)>,
affine_map<(d0, d1, d2) -> (d0, d1)>],
iterator_types = ["parallel", "parallel", "reduction"]}
ins(%0, %1 : tensor<4x16xi32>, tensor<1x16xi32>)
outs(%2 : tensor<4x1xi32>) {
^bb0(%lhs: i32, %rhs: i32, %out: i32):
%idx0 = linalg.index 0 : index
%idx1 = linalg.index 1 : index
%idx2 = linalg.index 2 : index
%add0 = arith.addi %idx0, %idx1 : index
%add1 = arith.addi %add0, %idx2 : index
%int = arith.index_cast %add1 : index to i32
linalg.yield %int : i32
} -> tensor<4x1xi32>
return %res : tensor<4x1xi32>
}

// -----

// CHECK: func @fold_linalg_index_tensor_dynamic
func.func @fold_linalg_index_tensor_dynamic(%0: tensor<?x1xi32>,
%1: tensor<?x1xi32>) -> tensor<?x1xi32> {
// CHECK-NEXT: linalg.generic
// CHECK: %[[IDX_0:.+]] = linalg.index 0 : index
// CHECK-NOT: linalg.index 1
// CHECK: %[[CAST:.+]] = arith.index_cast %[[IDX_0]]
// CHECK: linalg.yield %[[CAST]]
%res = linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
affine_map<(d0, d1) -> (d1, d1)>],
iterator_types = ["parallel", "parallel"]}
ins(%0 : tensor<?x1xi32>)
outs(%1 : tensor<?x1xi32>) {
^bb0(%lhs: i32, %out: i32):
%idx0 = linalg.index 0 : index
%idx1 = linalg.index 1 : index
%add = arith.addi %idx0, %idx1 : index
%int = arith.index_cast %add : index to i32
linalg.yield %int : i32
} -> tensor<?x1xi32>
return %res : tensor<?x1xi32>
}

// -----

// CHECK: func @fold_linalg_index_memref
func.func @fold_linalg_index_memref(%0: memref<1x?xi32>, %1: memref<1x?xi32>) {
// CHECK-NEXT: linalg.generic
// CHECK-NOT: linalg.index 0
// CHECK: %[[IDX_1:.+]] = linalg.index 1 : index
// CHECK: %[[CAST:.+]] = arith.index_cast %[[IDX_1]]
// CHECK: linalg.yield %[[CAST]]
linalg.generic {indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
affine_map<(d0, d1) -> (d1, d1)>],
iterator_types = ["parallel", "parallel"]}
ins(%0 : memref<1x?xi32>)
outs(%1 : memref<1x?xi32>) {
^bb0(%lhs: i32, %out: i32):
%idx0 = linalg.index 0 : index
%idx1 = linalg.index 1 : index
%add = arith.addi %idx0, %idx1 : index
%int = arith.index_cast %add : index to i32
linalg.yield %int : i32
}
return
}

// -----

// CHECK-LABEL: func @fold_fill_reshape()
func.func @fold_fill_reshape() -> tensor<6x4xf32> {
%zero = arith.constant 0.0 : f32
Expand Down
5 changes: 2 additions & 3 deletions mlir/test/Dialect/Linalg/vectorize-tensor-extract.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -278,12 +278,11 @@ func.func @vectorize_nd_tensor_extract_load_1d_column_vector_using_gather_load(%
// CHECK: %[[EMPTY:.*]] = tensor.empty() : tensor<8x1xf32>
// CHECK: %[[B1:.*]] = vector.broadcast %[[CST_3]] : vector<8xindex> to vector<1x8xindex>
// CHECK: %[[ADDI_ARG1:.*]] = arith.addi %[[ARG1]], %[[ARG1]] : index
// CHECK: %[[B2:.*]] = vector.broadcast %[[ADDI_ARG1]] : index to vector<1xindex>
// CHECK: %[[MULI_1:.*]] = arith.muli %[[B1]], %[[CST_0]] : vector<1x8xindex>
// CHECK: %[[MULI_2:.*]] = arith.muli %[[MULI_1]], %[[CST]] : vector<1x8xindex>
// CHECK: %[[T:.*]] = vector.transpose %[[MULI_2]], [1, 0] : vector<1x8xindex> to vector<8x1xindex>
// CHECK: %[[B3:.*]] = vector.broadcast %[[B2]] : vector<1xindex> to vector<8x1xindex>
// CHECK: %[[ADDI:.*]] = arith.addi %[[B3]], %[[T]] : vector<8x1xindex>
// CHECK: %[[B2:.*]] = vector.broadcast %[[ADDI_ARG1]] : index to vector<8x1xindex>
// CHECK: %[[ADDI:.*]] = arith.addi %[[B2]], %[[T]] : vector<8x1xindex>
// CHECK: %[[GATHER:.*]] = vector.gather %[[ARG0]][%[[C0]], %[[C0]], %[[C0]]] [%[[ADDI]]], %[[CST_2]], %[[PASSTHRU]] : tensor<8x128x768xf32>, vector<8x1xindex>, vector<8x1xi1>, vector<8x1xf32> into vector<8x1xf32>
// CHECK: vector.transfer_write %[[GATHER]], %[[EMPTY]][%[[C0]], %[[C0]]] {in_bounds = [true, true]} : vector<8x1xf32>, tensor<8x1xf32>

Expand Down
Loading