Turbine Camp

[pdhirajkumarprasad]

Contact:

• @vivekkhandelwal1 (Indian Time Zone)
• @AmosLewis (US Pacific Time Zone)
• @zjgarvey (US Central Time Zone)
• @marbre (Central EU Time Zone)

Onnx Ops( Known missing features)

Unassigned

• CastLike (saturate ) #840
• GridSample (mode - Cubic, paddding_mode only zero is supported)
• RoiAlign (roi max pooling without default, coordTfMode and sampling_ratio)
• NonMaxSuppression (center_point_box ) #844
• LpPool (auto_pad) #838
• DFT (onesided, axis input tensor) #841

In Progress

• AveragePool (auto_pad ) #835
• ConvInteger (auto_pad) #836
• Col2Im (5D input tensor)
• MaxUnpool (output_shape) #858
• Upsample (unsupported interpolation mode other than nearest, linear) #834
• SplitToSequence (keepdims) #837

Completed

• ConvTranspose (auto_pad) #839
• ScatterElements (reduction : max, min, add) onnx.ScatterElements #823
• TfIdfVectorizer (IDF and TFIDF mode) #833
• MaxPool (auto_pad, storage_order, indices output) #843
• Resize (antialias, axes, keep_aspect_ratio_policy , extrapolation_value, tf_crop_and_resize, coord tf mode, bicubic mode) #842

Onnx Ops( Known Quality Issues)

• AffineGrid
• Bernoulli
• Cast
• DequantizeLinear
• GRU Quality Issues #856
• MelWeightMatrix
• QuantizeLinear
• RandomUniformLike

Other tasks:

llvm/torch-mlir#3796

[knwng]

Seems LSTM also has quality issue. But it seems the testcases themselves are buggy(wrong # of inputs & outputs). Besides, RNN only has 1 testcase rn. I'm afraid it can't cover all the corner cases.

[vivekkhandelwal1]

Seems LSTM also has quality issue. But it seems the testcases themselves are buggy(wrong # of inputs & outputs). Besides, RNN only has 1 testcase rn. I'm afraid it can't cover all the corner cases.

Yeah, the issue for LSTM is already opened here #315.

[habcode]

Hello @zjgarvey @pdhirajkumarprasad , I would like to contribute to this repository and thinking of taking ownership of Cast or
DequantizeLinear OP. Would you please share some step by step guide on how to get started?

[zjgarvey]

Hello @zjgarvey @pdhirajkumarprasad , I would like to contribute to this repository and thinking of taking ownership of Cast or
DequantizeLinear OP. Would you please share some step by step guide on how to get started?

Hi!

1. clone a fork of llvm/torch-mlir and build it following the directions in docs/development.md in that repository.
2. Find which tests are failing for these ops (or which attribute combinations are not supported). It would be easiest to set up node tests for your torch-mlir build by cloning nod-ai/SHARK-TestSuite, navigating to alt_e2eshark and setting up with the instructions in the README.md. Then run python run.py -m cl-onnx-iree -v -t test_deq or test_cast to run the onnx node tests.
3. Find the torch-onnx-to-torch conversion logic in torch-mlir (under lib/Conversion/TorchOnnxToTorch/).
4. Figure out what is going wrong and try to fix it.