Skip to content

[TorchToTosa] Support flattened index_put updates in TOSA lowering #4561

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

Open
ahagro01 wants to merge 2 commits into
base: main
Choose a base branch
from
+26 −3
Open

[TorchToTosa] Support flattened index_put updates in TOSA lowering #4561

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
aaa448c
Support flattened index_put updates in TOSA lowering
ahagro01 May 4, 2026
11de2ef
remove passing from XFAIL
ahagro01 May 6, 2026
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
8 changes: 6 additions & 2 deletions lib/Conversion/TorchToTosa/TosaLegalizeCommon.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -603,9 +603,13 @@ std::optional<Value> convertScatterNdOp(PatternRewriter &rewriter,

// fillK: range of each index, total number of fillInput(could be scatter)
// after flattened k = 1*1*3 = 3
for (int i = 0; i < ND; i++) {
fillK *= fillValuesType.getShape()[i];
int64_t fillNumElements = 1;
for (int64_t dim : fillValuesType.getShape()) {
fillNumElements *= dim;
}
if (fillNumElements % C != 0)
return std::nullopt;
fillK = fillNumElements / C;
SmallVector<int64_t, 3> tosaFillValuesShape({N, fillK, C}); // {1,3,1}

// Reshape/Flatten fillValues to 3d tensor
Expand Down
1 change: 0 additions & 1 deletion projects/pt1/e2e_testing/xfail_sets.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3825,7 +3825,6 @@
"IndexPutImpl1DFloatAccumulateModule_basic",
"IndexPutImpl1DIntAccumulateModule_basic",
"IndexPutImpl2DFloatAccumulateModule_basic",
"IndexPutImpl2DImplicitModule_basic",
"IndexPutImpl2DIndexModule_basic",
"IndexPutImpl2DNoneIndexStaticModule_basic",
"IndexPutImpl3DFloatAccumulateModule_basic",
Expand Down
20 changes: 20 additions & 0 deletions test/Conversion/TorchToTosa/basic.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2746,6 +2746,26 @@ func.func @torch.aten.diag_embed$basic(%arg0: !torch.vtensor<[2,3,4],f32>) -> !t

// -----

// CHECK-LABEL: func.func @torch.aten.index_put_hacked_twin_flattened_updates(
// CHECK: %[[SCATTER:.*]] = tosa.scatter
// CHECK-SAME: (tensor<1x6x1xf32>, tensor<1x6xi32>, tensor<1x6x1xf32>) -> tensor<1x6x1xf32>
// CHECK: %[[RESHAPE:.*]] = tosa.reshape %[[SCATTER]]
// CHECK-SAME: (tensor<1x6x1xf32>, !tosa.shape<3>) -> tensor<1x2x3xf32>
// CHECK: torch_c.from_builtin_tensor %[[RESHAPE]] : tensor<1x2x3xf32> -> !torch.vtensor<[1,2,3],f32>
func.func @torch.aten.index_put_hacked_twin_flattened_updates(
%arg0: !torch.vtensor<[1,2,3],f32>,
%arg1: !torch.vtensor<[6],si64>,
%arg2: !torch.vtensor<[6],si64>,
%arg3: !torch.vtensor<[6],si64>,
%arg4: !torch.vtensor<[6],f32>) -> !torch.vtensor<[1,2,3],f32> {
%indices = torch.prim.ListConstruct %arg1, %arg2, %arg3 : (!torch.vtensor<[6],si64>, !torch.vtensor<[6],si64>, !torch.vtensor<[6],si64>) -> !torch.list<vtensor>
%false = torch.constant.bool false
%0 = torch.aten.index_put.hacked_twin %arg0, %indices, %arg4, %false : !torch.vtensor<[1,2,3],f32>, !torch.list<vtensor>, !torch.vtensor<[6],f32>, !torch.bool -> !torch.vtensor<[1,2,3],f32>
return %0 : !torch.vtensor<[1,2,3],f32>
}

// -----

// CHECK-LABEL: func.func @torch.aten.index.Tensor_hacked_twin(
// CHECK-SAME: %[[VAL_0:.*]]: !torch.vtensor<[2,4,2],si64>,
// CHECK-SAME: %[[VAL_1:.*]]: !torch.vtensor<[],si64>) -> !torch.vtensor<[4,2],si64> {
Expand Down
Loading