Skip to content

Remove num dim usages in xnnpack flatbuffer #19013

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
lucylq wants to merge 1 commit into
base: main
Choose a base branch
from
+12 −27
Open

Remove num dim usages in xnnpack flatbuffer #19013

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
39 changes: 12 additions & 27 deletions backends/xnnpack/runtime/XNNCompiler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -319,24 +319,9 @@ Error defineTensor(
ET_CHECK_OR_RETURN_ERROR(
tensor_value != nullptr, InvalidProgram, "Deserialized tensor is null");

ET_CHECK_OR_RETURN_ERROR(
tensor_value->num_dims() == 0 || tensor_value->dims() != nullptr,
InvalidProgram,
"Tensor dims is null but num_dims is %u",
tensor_value->num_dims());

if (tensor_value->dims() != nullptr) {
ET_CHECK_OR_RETURN_ERROR(
tensor_value->num_dims() == tensor_value->dims()->size(),
InvalidProgram,
"Tensor num_dims %u does not match dims array size %u",
tensor_value->num_dims(),
tensor_value->dims()->size());
}

Copy link
Copy Markdown
Contributor Author

@lucylq lucylq Apr 21, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ah this was recently added but if we use tensor_value->dims()->size() directly then we also don't need the check on tensor_value->num_dims()..

// Get tensor dims, here we need to use a vector in order
// to properly convert the uint32_t* to size_t*. For scalar tensors
// (num_dims == 0), dims() is permitted to be null per the check above.
// Get tensor dims, here we need to use a vector in order to properly
// convert the uint32_t* to size_t*. Scalar tensors (rank 0) are permitted
// to have a null dims vector; in that case dims_data is empty.
std::vector<size_t> dims_data;
if (tensor_value->dims() != nullptr) {
dims_data = flatbufferDimsToVector(tensor_value->dims());
Expand Down Expand Up @@ -386,7 +371,7 @@ Error defineTensor(
status = xnn_define_tensor_value(
/*subgraph=*/subgraph_ptr,
/*datatype=*/getDataType(tensor_value->datatype()),
/*num_dims=*/tensor_value->num_dims(),
/*num_dims=*/dims_data.size(),
/*dims=*/dims_data.data(),
/*data=*/buffer_ptr,
/*external_id=*/tensor_value->external_id(),
Expand Down Expand Up @@ -421,7 +406,7 @@ Error defineTensor(
status = xnn_define_dynamically_quantized_tensor_value(
/*subgraph=*/subgraph_ptr,
/*datatype=*/xnn_datatype_qdint8,
/*num_dims=*/tensor_value->num_dims(),
/*num_dims=*/dims_data.size(),
/*num_nonbatch_dims=*/1, // always do per token quantization
/*dims=*/dims_data.data(),
/*external_id=*/XNN_INVALID_VALUE_ID, // always internal value id
Expand All @@ -435,7 +420,7 @@ Error defineTensor(
status = xnn_define_tensor_value(
/*subgraph=*/subgraph_ptr,
/*datatype=*/fp_datatype,
/*num_dims=*/tensor_value->num_dims(),
/*num_dims=*/dims_data.size(),
/*dims=*/dims_data.data(),
/*data=*/buffer_ptr,
/*external_id=*/tensor_value->external_id(),
Expand Down Expand Up @@ -476,7 +461,7 @@ Error defineTensor(
/*datatype=*/getDataType(tensor_value->datatype()),
/*zero_point=*/qparams->zero_point(),
/*scale=*/qparams->scale(),
/*num_dims=*/tensor_value->num_dims(),
/*num_dims=*/dims_data.size(),
/*dims=*/dims_data.data(),
/*data=*/buffer_ptr,
/*external_id=*/tensor_value->external_id(),
Expand Down Expand Up @@ -521,7 +506,7 @@ Error defineTensor(
/*datatype=*/dtype,
/*zero_point=*/zero_point,
/*scale=*/scale,
/*num_dims=*/tensor_value->num_dims(),
/*num_dims=*/dims_data.size(),
/*channel_dim*/ qparams->channel_dim(),
/*dims=*/dims_data.data(),
/*data=*/buffer_ptr,
Expand Down Expand Up @@ -599,7 +584,7 @@ Error defineTensor(
/*datatype=*/datatype,
/*zero_point=*/zero_point,
/*scale=*/scale_data,
/*num_dims=*/tensor_value->num_dims(),
/*num_dims=*/dims_data.size(),
/*channel_dim=*/qparams->channel_dim(),
Comment on lines 584 to 588
Copy link

Copilot AI Apr 21, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PerChannelGroupQuant path still dereferences tensor_value->dims()->Get(0/1) earlier in this case block without any null / length validation. Since the earlier num_dims/dims consistency checks were removed, a malformed (or malicious) flatbuffer with dims == nullptr (or a shorter-than-expected dims vector) can now crash here instead of returning InvalidProgram. Please add an ET_CHECK_OR_RETURN_ERROR(dims_data.size() >= 2, InvalidProgram, ...) (and use dims_data[0]/dims_data[1]) before computing output_channels/input_channels so this remains robust against corrupted inputs.

Copilot uses AI. Check for mistakes.
/*block_size=*/qparams->group_size(),
/*dims=*/dims_data.data(),
Expand All @@ -613,8 +598,8 @@ Error defineTensor(
auto qparams = qtensor_value->quant_params_as_PerTokenDynamicQuant();
ET_LOG(
Debug,
"define quant tensor (dynamic): num_dims: %i, num_nonbatch_dims: %i\n",
tensor_value->num_dims(),
"define quant tensor (dynamic): num_dims: %zu, num_nonbatch_dims: %i\n",
dims_data.size(),
qparams->num_nonbatch_dims());
ET_CHECK_OR_RETURN_ERROR(
buffer_ptr == nullptr,
Expand All @@ -623,7 +608,7 @@ Error defineTensor(
status = xnn_define_dynamically_quantized_tensor_value(
/*subgraph=*/subgraph_ptr,
/*datatype=*/getDataType(tensor_value->datatype()),
/*num_dims=*/tensor_value->num_dims(),
/*num_dims=*/dims_data.size(),
/*num_nonbatch_dims*/ qparams->num_nonbatch_dims(),
/*dims=*/dims_data.data(),
/*external_id=*/tensor_value->external_id(),
Expand Down
Loading