search

intel / torch-xpu-ops

Apache License 2.0
17 stars 8 forks source link

[Evaluated] issues report for test_convolution_xpu.py #348

Open PenghuiCheng opened 1 month ago

PenghuiCheng commented 1 month ago

šŸš€ The feature, motivation and pitch

  1. torch.backends.mkldnn.flags() got an unexpected keyword argument 'deterministic'---fixed cases:

    test/xpu/nn/test_convolution_xpu.py::TestConvolutionNNDeviceTypeXPU::test_conv_double_backward_xpu_float64

  2. Tensor-likes are not close! -----fixed cases:

    
TestConvolutionNNDeviceTypeXPU.test_Conv2d_depthwise_naive_groups_xpu_float16
TestConvolutionNNDeviceTypeXPU.test_Conv3d_depthwise_naive_groups_xpu_float16

  3. check_random_bounds handles only integral, floating-point and boolean types---fixed cases:

    TestConvolutionNNDeviceTypeXPU.test_conv_cudnn_nhwc_xpu_complex64

  4. largeTensorTest didn't support XPU device ---fixed cases:

    TestConvolutionNNDeviceTypeXPU::test_conv_large_xpu
TestConvolutionNNDeviceTypeXPU::test_conv_large_nosplit_xpu
TestConvolutionNNDeviceTypeXPU::test_conv_large_batch_1_xpu
TestConvolutionNNDeviceTypeXPU::test_conv3d_64bit_indexing_xpu
TestConvolutionNNDeviceTypeXPU::test_conv3d_large_batch_1_xpu
TestConvolutionNNDeviceTypeXPU::test_conv_transposed_large_xpu

  5. NotImplementedError: The operators "convolution_add_relu, convolution_relu, ". cases: TestConvolutionNNDeviceTypeXPU::test_cudnn_convolution_add_relu_xpu_float16 TestConvolutionNNDeviceTypeXPU::test_cudnn_convolution_add_relu_xpu_float32 TestConvolutionNNDeviceTypeXPU::test_cudnn_convolution_relu_xpu_float16 TestConvolutionNNDeviceTypeXPU::test_cudnn_convolution_relu_xpu_float32

Alternatives

No response

Additional context

No response

PenghuiCheng commented 1 month ago

Log: convolution.log

PenghuiCheng commented 6 days ago

1~4, are fixed, and 5 is not in 2.5 plan.