Skip to content
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

Fix permute op lowering to TOSA transpose #157

Merged
merged 1 commit into from
Apr 16, 2024
Merged

Fix permute op lowering to TOSA transpose #157

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
12 changes: 7 additions & 5 deletions lib/Conversion/TorchToTosa/TorchToTosa.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2766,21 +2766,23 @@ LogicalResult ConvertAtenOp<AtenPermuteOp>::matchAndRewrite(
op,
"Only ranked tensor types with static shapes are currently supported");

SmallVector<int64_t> dimListInt;
if (!matchPattern(adaptor.getDims(), m_TorchListOfConstantInts(dimListInt)))
SmallVector<int64_t> dimListInt64;
if (!matchPattern(adaptor.getDims(), m_TorchListOfConstantInts(dimListInt64)))
return rewriter.notifyMatchFailure(
op, "Only constant dimensions are currently supported");
SmallVector<int32_t> dimListInt32;
roberteg16 marked this conversation as resolved.
Show resolved Hide resolved
copy(dimListInt64, std::back_inserter(dimListInt32));

int64_t selfRank = selfType.getRank();
// TODO: If this is already verified on the op then we can drop checking here.
for (auto &d : dimListInt) {
for (auto &d : dimListInt32) {
d = toPositiveDim(d, selfRank);
if (!isValidDim(d, selfRank))
return rewriter.notifyMatchFailure(op, "Not all dims are valid");
}

auto transposeDimsConst = mlir::tosa::getConstTensor<int64_t>(
rewriter, op.getOperation(), dimListInt, {selfRank});
auto transposeDimsConst = mlir::tosa::getConstTensor<int32_t>(
rewriter, op.getOperation(), dimListInt32, {selfRank});

rewriter.replaceOpWithNewOp<tosa::TransposeOp>(
op, getTypeConverter()->convertType(op.getType()), adaptor.getSelf(),
Expand Down
4 changes: 2 additions & 2 deletions test/Conversion/TorchToTosa/basic.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -797,8 +797,8 @@ func.func @torch.aten.ne.Tensor$basic(%arg0: !torch.vtensor<[?,?],f32>, %arg1: !
// CHECK: %[[VAL_3:.*]] = torch.constant.int 2
// CHECK: %[[VAL_4:.*]] = torch.constant.int 0
// CHECK: %[[VAL_5:.*]] = torch.prim.ListConstruct %[[VAL_4]], %[[VAL_3]], %[[VAL_2]] : (!torch.int, !torch.int, !torch.int) -> !torch.list<int>
// CHECK: %[[VAL_6:.*]] = "tosa.const"() <{value = dense<[0, 2, 1]> : tensor<3xi64>}> : () -> tensor<3xi64>
// CHECK: %[[VAL_7:.*]] = tosa.transpose %[[VAL_1]], %[[VAL_6]] : (tensor<3x4x2xf32>, tensor<3xi64>) -> tensor<3x2x4xf32>
// CHECK: %[[VAL_6:.*]] = "tosa.const"() <{value = dense<[0, 2, 1]> : tensor<3xi32>}> : () -> tensor<3xi32>
// CHECK: %[[VAL_7:.*]] = tosa.transpose %[[VAL_1]], %[[VAL_6]] : (tensor<3x4x2xf32>, tensor<3xi32>) -> tensor<3x2x4xf32>
// CHECK: %[[VAL_8:.*]] = torch_c.from_builtin_tensor %[[VAL_7]] : tensor<3x2x4xf32> -> !torch.vtensor<[3,2,4],f32>
// CHECK: return %[[VAL_8]] : !torch.vtensor<[3,2,4],f32>
// CHECK: }
Expand Down
Loading