[Runtime][MatchShape] Type error: Cannot convert from type ' DLTensor* ' to ' ffi.Shape ' #18546

locnd182644 · 2025-12-05T07:58:06Z

Summary

This commit fixed error : "Type error: Cannot convert from type ' DLTensor* ' to ' ffi.Shape ' ".
Error occurred at runtime when the first args of the MatchShape function (in runtime/vm/builtin.cc) is DLTensor*. If (auto opt_nd = args[0].as()) is false, args[0] (DLTensor*) will try convert to Shape.

Reproduce

RPC

After over RPCWarppedFunc, Tensor will convert to DLTensor *

Cell In[3], line 189
    186 remote_input = tvm.runtime.tensor(input_data, dev)
    188 # Run inference on remote device
--> 189 output = vm["main"](remote_input)
    191 # Extract result (handle both tuple and single tensor outputs)
    192 if isinstance(output, tvm.ir.Array) and len(output) > 0:

File python/tvm_ffi/cython/function.pxi:904, in tvm_ffi.core.Function.__call__()

File ~/Programming/tvm/src/runtime/rpc/rpc_module.cc:141, in tvm::runtime::RPCWrappedFunc::operator()(tvm::ffi::PackedArgs, tvm::ffi::Any*) const()
    139   }
    140   auto set_return = [this, rv](ffi::PackedArgs args) { this->WrapRemoteReturnToValue(args, rv); };
--> 141   sess_->CallFunc(handle_, ffi::PackedArgs(packed_args.data(), packed_args.size()), set_return);
    142 }
    143 

File ~/Programming/tvm/src/runtime/rpc/rpc_endpoint.cc:1116, in tvm::runtime::RPCClientSession::CallFunc(void*, tvm::ffi::PackedArgs, std::function<void (tvm::ffi::PackedArgs)> const&)()
   1114 void CallFunc(PackedFuncHandle func, ffi::PackedArgs args,
   1115               const FEncodeReturn& fencode_return) final {
-> 1116   endpoint_->CallFunc(func, args, fencode_return);
   1117 }
   1118 

File ~/Programming/tvm/src/runtime/rpc/rpc_endpoint.cc:906, in tvm::runtime::RPCEndpoint::CallFunc(void*, tvm::ffi::PackedArgs, std::function<void (tvm::ffi::PackedArgs)>)()
    904   handler_->SendPackedSeq(args.data(), args.size(), true);
    905 
--> 906   code = HandleUntilReturnEvent(true, encode_return);
    907   ICHECK(code == RPCCode::kReturn) << "code=" << RPCCodeToString(code);
    908 }

File ~/Programming/tvm/src/runtime/rpc/rpc_endpoint.cc:746, in tvm::runtime::RPCEndpoint::HandleUntilReturnEvent(bool, std::function<void (tvm::ffi::PackedArgs)>)()
    744     }
    745   }
--> 746   code = handler_->HandleNextEvent(client_mode, false, setreturn);
    747 }
    748 return code;

File ~/Programming/tvm/src/runtime/rpc/rpc_endpoint.cc:134, in tvm::runtime::RPCEndpoint::EventHandler::HandleNextEvent(bool, bool, std::function<void (tvm::ffi::PackedArgs)>)()
    132 }
    133 case kProcessPacket: {
--> 134   this->HandleProcessPacket(setreturn);
    135   break;
    136 }

File ~/Programming/tvm/src/runtime/rpc/rpc_endpoint.cc:409, in tvm::runtime::RPCEndpoint::EventHandler::HandleProcessPacket(std::function<void (tvm::ffi::PackedArgs)>)()
    407 case RPCCode::kException:
    408 case RPCCode::kReturn: {
--> 409   this->HandleReturn(code, setreturn);
    410   break;
    411 }

File ~/Programming/tvm/src/runtime/rpc/rpc_endpoint.cc:473, in tvm::runtime::RPCEndpoint::EventHandler::HandleReturn(tvm::runtime::RPCCode, std::function<void (tvm::ffi::PackedArgs)>)()
    471     msg = "RPCError: Error caught from RPC call:\n" + msg;
    472   }
--> 473   LOG(FATAL) << msg;
    474 }
    475 

File ~/Programming/tvm/include/tvm/runtime/logging.h:321, in tvm::runtime::detail::LogFatal::~LogFatal()()
    319 #endif
    320   [[noreturn]] ~LogFatal() TVM_THROW_EXCEPTION {
--> 321     GetEntry().Finalize();
    322     throw;
    323   }

File ~/Programming/tvm/include/tvm/runtime/logging.h:337, in tvm::runtime::detail::LogFatal::Entry::Finalize()()
    335     }
    336     [[noreturn]] TVM_NO_INLINE dmlc::Error Finalize() TVM_THROW_EXCEPTION {
--> 337       InternalError error(file_, lineno_, stream_.str());
    338 #if DMLC_LOG_BEFORE_THROW
    339       std::cerr << error.what() << std::endl;

RPCError: Error caught from RPC call:
Cannot convert from type `DLTensor*` to `ffi.Shape`

C++

Input is DLTensor *

std::cout << "--- TVM C++ TorchMLP Example ---" << std::endl;
int64_t shapes[] = {1, 1, 28, 28};
...
// Load binary data
std::vector<float> input_img = load_bin(argv[1], shapes[0] * shapes[1] * shapes[2] * shapes[3]);

// Create DLTensor for input
DLTensor* dl_tensor = new DLTensor();
dl_tensor->data = malloc(sizeof(float) * input_img.size());
dl_tensor->device = device;
dl_tensor->ndim = sizeof(shapes) / sizeof(shapes[0]);
dl_tensor->dtype = DLDataType{kDLFloat, 32, 1};
dl_tensor->shape = shapes;
dl_tensor->strides = nullptr;
dl_tensor->byte_offset = 0;

memcpy(dl_tensor->data, input_img.data(), sizeof(float) * input_img.size());

// Run the main function
tvm::ffi::Array<tvm::ffi::Tensor> ouput_temp = (*main)(dl_tensor).cast<tvm::ffi::Array<tvm::ffi::Tensor>>();

Resolved

Instead of converting to Tensor, try_cast to DLTensor*, then create Shape

…* ' to ' ffi.Shape ' - Error occurred at runtime when the first args of the MatchShape function (in runtime/vm/builtin.cc) is DLTensor*. If (auto opt_nd = args[0].as<Tensor>()) is false, args[0] (DLTensor*) will try convert to Shape.

gemini-code-assist · 2025-12-05T07:58:19Z

Summary of Changes

Hello @locnd182644, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request addresses a critical runtime type conversion error within the MatchShape function, which previously led to failures when DLTensor* was passed as an argument. By enhancing the argument parsing to correctly interpret and process DLTensor* inputs, the change prevents crashes in RPC and C++ contexts. This improves the overall robustness and compatibility of the runtime system with various tensor representations, ensuring smoother execution of operations involving shape matching.

Highlights

Fix MatchShape Type Error: Resolved a runtime type error in the MatchShape function where it failed to correctly convert a DLTensor* argument to ffi::Shape. This issue specifically occurred in scenarios involving RPC calls and direct C++ usage where DLTensor* is passed as the first argument.
Improved Argument Handling: The MatchShape function's argument parsing logic has been updated to first attempt a try_cast to DLTensor*. If successful, it now constructs the ffi::Shape directly from the DLTensor*'s shape and ndim fields, ensuring proper handling of this tensor representation.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request addresses a runtime type error in MatchShape when it receives a DLTensor* argument. The fix correctly handles this case. However, my review identifies a critical regression where the support for tvm::runtime::Tensor arguments is unintentionally removed. I've provided a code suggestion to restore this functionality while keeping the fix, ensuring that MatchShape can correctly handle Tensor, DLTensor*, and ffi::Shape arguments as intended.

gemini-code-assist · 2025-12-05T07:59:17Z

src/runtime/vm/builtin.cc

+  if (auto opt_nd = args[0].try_cast<DLTensor*>()) {
+    DLTensor* ptr = opt_nd.value();
+    input_shape = ffi::Shape(ptr->shape, ptr->shape + ptr->ndim);
  } else {
    input_shape = args[0].cast<ffi::Shape>();
  }


The current change correctly adds support for DLTensor* arguments, which fixes the reported issue. However, it removes the existing support for tvm::runtime::Tensor arguments by replacing the check instead of adding to it. This will break existing code that passes a Tensor to MatchShape.

The logic should be updated to handle Tensor, DLTensor*, and ffi::Shape arguments, as indicated by the updated comment on line 125. You can chain the checks to support all three types.

if (auto opt_tensor = args[0].as<Tensor>()) { input_shape = opt_tensor.value().Shape(); } else if (auto opt_dltensor = args[0].try_cast<DLTensor*>()) { DLTensor* ptr = opt_dltensor.value(); input_shape = ffi::Shape(ptr->shape, ptr->shape + ptr->ndim); } else { input_shape = args[0].cast<ffi::Shape>(); }

I think this review makes sense. Could you update the code like this?

I replaced Tensor to DLTensor* because I think tensor can cast to DLTensor*. This review makes sense because "as" operator cost is less than that of "cast" operator or other reason, could you tell me know.

The main reason here is not related to the performance difference between as<Tensor>() and try_cast<DLTensor*>(). The issue is type coverage and backward compatibility. Tensor cannot be assumed to always cast safely to DLTensor*. Also, existing callers still pass tvm::runtime::Tensor as an argument. If we replace the Tensor branch with DLTensor*, the previous behavior may break.

I understand. I will update the code like this review

@tlopex Thank you for explaining. I updated the code.

cbalint13 · 2025-12-06T22:55:21Z

@tvm-bot rerun

tlopex · 2025-12-10T06:04:43Z

src/runtime/vm/builtin.cc

+  if (auto opt_nd = args[0].try_cast<DLTensor*>()) {
+    DLTensor* ptr = opt_nd.value();
+    input_shape = ffi::Shape(ptr->shape, ptr->shape + ptr->ndim);
  } else {
    input_shape = args[0].cast<ffi::Shape>();
  }


I think this review makes sense. Could you update the code like this?

…* ' to ' ffi.Shape ' - Edit code like review

tlopex · 2025-12-22T14:31:23Z

Hi @locnd182644 This pr is the same. Could you retrigger the CI? Thanks!

locnd182644 · 2025-12-22T16:08:22Z

@tlopex This PR has completed the check

gemini-code-assist bot reviewed Dec 5, 2025

View reviewed changes

locnd182644 mentioned this pull request Dec 5, 2025

[Bug] RPCError: Cannot convert from type DLTensor* to ffi.Shape`` when run How to: Cross Compilation and RPC #18526

Closed

tlopex requested changes Dec 10, 2025

View reviewed changes

[Runtime][MatchShape] Type error: Cannot convert from type ' DLTensor…

e214996

…* ' to ' ffi.Shape ' - Edit code like review

locnd182644 added 2 commits December 22, 2025 21:50

Rerun again

493441d

Rerun

2b1d72c

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Runtime][MatchShape] Type error: Cannot convert from type ' DLTensor* ' to ' ffi.Shape ' #18546

[Runtime][MatchShape] Type error: Cannot convert from type ' DLTensor* ' to ' ffi.Shape ' #18546

locnd182644 commented Dec 5, 2025

Uh oh!

gemini-code-assist bot commented Dec 5, 2025

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Dec 5, 2025

Uh oh!

tlopex Dec 10, 2025

Uh oh!

locnd182644 Dec 10, 2025

Uh oh!

tlopex Dec 11, 2025

Uh oh!

locnd182644 Dec 11, 2025

Uh oh!

locnd182644 Dec 11, 2025

Uh oh!

cbalint13 commented Dec 6, 2025

Uh oh!

tlopex Dec 10, 2025

Uh oh!

tlopex commented Dec 22, 2025

Uh oh!

locnd182644 commented Dec 22, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

[Runtime][MatchShape] Type error: Cannot convert from type ' DLTensor* ' to ' ffi.Shape ' #18546

Are you sure you want to change the base?

[Runtime][MatchShape] Type error: Cannot convert from type ' DLTensor* ' to ' ffi.Shape ' #18546

Conversation

locnd182644 commented Dec 5, 2025

Summary

Reproduce

RPC

C++

Resolved

Uh oh!

gemini-code-assist bot commented Dec 5, 2025

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Dec 5, 2025

Choose a reason for hiding this comment

Uh oh!

tlopex Dec 10, 2025

Choose a reason for hiding this comment

Uh oh!

locnd182644 Dec 10, 2025

Choose a reason for hiding this comment

Uh oh!

tlopex Dec 11, 2025

Choose a reason for hiding this comment

Uh oh!

locnd182644 Dec 11, 2025

Choose a reason for hiding this comment

Uh oh!

locnd182644 Dec 11, 2025

Choose a reason for hiding this comment

Uh oh!

cbalint13 commented Dec 6, 2025

Uh oh!

tlopex Dec 10, 2025

Choose a reason for hiding this comment

Uh oh!

tlopex commented Dec 22, 2025

Uh oh!

locnd182644 commented Dec 22, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants