Uplift third_party/tt-mlir to 2025-03-01

[vmilosevic]

This PR uplifts the third_party/tt-mlir to the

[github-actions]

	Tests	Passed ☑️	Skipped ⚠️	Failed ❌️
TT-Torch Tests	435 ran	426 passed	7 skipped	2 failed

Test	Result
TT-Torch Tests
pytest
test_conv2d.test_conv2d[dtype1-1-64-3-256-256-7-7-2-2-3]	❌ failure
test_conv2d.test_conv2d[dtype1-1-128-64-128-128-2-2-2-2-0]	❌ failure

[codecov-commenter]

❌ 2 Tests Failed:

Tests completed	Failed	Passed	Skipped
428	2	426	7

View the full list of 2 ❄️ flaky tests

tests.torch.test_conv2d::test_conv2d[dtype1-1-128-64-128-128-2-2-2-2-0]

Flake rate in main: 42.86% (Passed 4 times, Failed 3 times)

Stack Traces | 0.459s run time

batch_size = 1, output_channels = 128, input_channels = 64, input_height = 128
input_width = 128, filter_height = 2, filter_width = 2, stride_h = 2
stride_w = 2, padding = 0, dtype = torch.float32

    @pytest.mark.parametrize(
        "batch_size, output_channels, input_channels, input_height, input_width, filter_height, filter_width, stride_h, stride_w, padding",
        ((1, 64, 3, 256, 256, 7, 7, 2, 2, 3), (1, 128, 64, 128, 128, 2, 2, 2, 2, 0)),
    )
    @pytest.mark.parametrize("dtype", [torch.bfloat16, torch.float32])
    def test_conv2d(
        batch_size,
        output_channels,
        input_channels,
        input_height,
        input_width,
        filter_height,
        filter_width,
        stride_h,
        stride_w,
        padding,
        dtype,
    ):
        class Basic(nn.Module):
            def __init__(self):
                super().__init__()
                self.conv2d = nn.Conv2d(
                    input_channels,
                    output_channels,
                    kernel_size=(filter_height, filter_width),
                    stride=(stride_h, stride_w),
                    padding=padding,
                    bias=False,
                    dtype=dtype,
                )
    
            def forward(self, x):
                return self.conv2d(x)
    
>       verify_module(
            Basic(),
            input_shapes=[(batch_size, input_channels, input_height, input_width)],
            input_data_types=[dtype],
            required_atol=10,
        )

tests/torch/test_conv2d.py:47: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
tt_torch/tools/verify.py:252: in verify_module
    _verify_torch_module(
tt_torch/tools/verify.py:146: in _verify_torch_module
    ret = tt_mod(*inputs)
.../venv/lib/python3.11.../nn/modules/module.py:1736: in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
.../venv/lib/python3.11.../nn/modules/module.py:1747: in _call_impl
    return forward_call(*args, **kwargs)
.../venv/lib/python3.11.../torch/_dynamo/eval_frame.py:465: in _fn
    return fn(*args, **kwargs)
.../venv/lib/python3.11.../nn/modules/module.py:1736: in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
.../venv/lib/python3.11.../nn/modules/module.py:1747: in _call_impl
    return forward_call(*args, **kwargs)
tests/torch/test_conv2d.py:44: in forward
    def forward(self, x):
.../venv/lib/python3.11.../torch/_dynamo/eval_frame.py:632: in _fn
    return fn(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <tt_torch.dynamo.backend.Executor object at 0x7fb440b68610>
inputs = (Parameter containing:
tensor([[[[-0.0005,  0.0335],
          [-0.0514, -0.0460]],

         [[-0.0241,  0.0168],
   ...  0.1049,  ...,  0.4944,  0.0473, -0.0291],
          [-0.1079, -0.3563, -0.4745,  ..., -0.4814, -0.4681,  0.4106]]]]))
new_inputs = (Parameter containing:
tensor([[[[-0.0005,  0.0335],
          [-0.0514, -0.0460]],

         [[-0.0241,  0.0168],
   ...  0.1049,  ...,  0.4944,  0.0473, -0.0291],
          [-0.1079, -0.3563, -0.4745,  ..., -0.4814, -0.4681,  0.4106]]]]))
input = tensor([[[[-0.1674,  0.2599, -0.3615,  ...,  0.2775, -0.2959,  0.0303],
          [-0.2474,  0.1794,  0.4380,  ..., -0...,  0.1049,  ...,  0.4944,  0.0473, -0.0291],
          [-0.1079, -0.3563, -0.4745,  ..., -0.4814, -0.4681,  0.4106]]]])
input_type = torch.float32

    def __call__(self, *inputs):
        new_inputs = ()
        for input in inputs:
            # Handle scalar inputs.
            if not hasattr(input, "dtype"):
                assert (
                    type(input) is not bool
                ), "Conversion for scalar boolean is not supported."
                new_inputs = new_inputs + ((input),)
                continue
    
            # Apply type conversion if required.
            input_type = input.dtype
            if input_type in self.type_conversion.keys():
                new_inputs = new_inputs + (
                    (input.to(dtype=self.type_conversion[input_type])),
                )
                continue
    
            # No conversion required.
            new_inputs = new_inputs + ((input),)
    
        inputs = new_inputs
    
        if self.compiler_config.compile_depth == CompileDepth.EXECUTE:
            assert self.binary is not None, "Binary must be set for EXECUTE mode"
>           return tt_mlir.run(inputs + self.graph_constants, self.binary)
E           RuntimeError: std::get: wrong index for variant

tt_torch/dynamo/backend.py:558: RuntimeError

tests.torch.test_conv2d::test_conv2d[dtype1-1-64-3-256-256-7-7-2-2-3]

Flake rate in main: 42.86% (Passed 4 times, Failed 3 times)

Stack Traces | 1.01s run time

batch_size = 1, output_channels = 64, input_channels = 3, input_height = 256
input_width = 256, filter_height = 7, filter_width = 7, stride_h = 2
stride_w = 2, padding = 3, dtype = torch.float32

    @pytest.mark.parametrize(
        "batch_size, output_channels, input_channels, input_height, input_width, filter_height, filter_width, stride_h, stride_w, padding",
        ((1, 64, 3, 256, 256, 7, 7, 2, 2, 3), (1, 128, 64, 128, 128, 2, 2, 2, 2, 0)),
    )
    @pytest.mark.parametrize("dtype", [torch.bfloat16, torch.float32])
    def test_conv2d(
        batch_size,
        output_channels,
        input_channels,
        input_height,
        input_width,
        filter_height,
        filter_width,
        stride_h,
        stride_w,
        padding,
        dtype,
    ):
        class Basic(nn.Module):
            def __init__(self):
                super().__init__()
                self.conv2d = nn.Conv2d(
                    input_channels,
                    output_channels,
                    kernel_size=(filter_height, filter_width),
                    stride=(stride_h, stride_w),
                    padding=padding,
                    bias=False,
                    dtype=dtype,
                )
    
            def forward(self, x):
                return self.conv2d(x)
    
>       verify_module(
            Basic(),
            input_shapes=[(batch_size, input_channels, input_height, input_width)],
            input_data_types=[dtype],
            required_atol=10,
        )

tests/torch/test_conv2d.py:47: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
tt_torch/tools/verify.py:252: in verify_module
    _verify_torch_module(
tt_torch/tools/verify.py:146: in _verify_torch_module
    ret = tt_mod(*inputs)
.../venv/lib/python3.11.../nn/modules/module.py:1736: in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
.../venv/lib/python3.11.../nn/modules/module.py:1747: in _call_impl
    return forward_call(*args, **kwargs)
.../venv/lib/python3.11.../torch/_dynamo/eval_frame.py:465: in _fn
    return fn(*args, **kwargs)
.../venv/lib/python3.11.../nn/modules/module.py:1736: in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
.../venv/lib/python3.11.../nn/modules/module.py:1747: in _call_impl
    return forward_call(*args, **kwargs)
tests/torch/test_conv2d.py:44: in forward
    def forward(self, x):
.../venv/lib/python3.11.../torch/_dynamo/eval_frame.py:632: in _fn
    return fn(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <tt_torch.dynamo.backend.Executor object at 0x7fb443568250>
inputs = (Parameter containing:
tensor([[[[-0.0006,  0.0442, -0.0679,  ..., -0.0318,  0.0221, -0.0016],
          [ 0.0654, -0.... -0.4588,  ..., -0.3514, -0.4965,  0.3036],
          [ 0.3726, -0.2234,  0.2158,  ..., -0.2494, -0.0489, -0.4841]]]]))
new_inputs = (Parameter containing:
tensor([[[[-0.0006,  0.0442, -0.0679,  ..., -0.0318,  0.0221, -0.0016],
          [ 0.0654, -0.... -0.4588,  ..., -0.3514, -0.4965,  0.3036],
          [ 0.3726, -0.2234,  0.2158,  ..., -0.2494, -0.0489, -0.4841]]]]))
input = tensor([[[[-0.3725, -0.3886,  0.4439,  ..., -0.4821,  0.4407, -0.2102],
          [-0.0409, -0.3014, -0.1411,  ..., -0..., -0.4588,  ..., -0.3514, -0.4965,  0.3036],
          [ 0.3726, -0.2234,  0.2158,  ..., -0.2494, -0.0489, -0.4841]]]])
input_type = torch.float32

    def __call__(self, *inputs):
        new_inputs = ()
        for input in inputs:
            # Handle scalar inputs.
            if not hasattr(input, "dtype"):
                assert (
                    type(input) is not bool
                ), "Conversion for scalar boolean is not supported."
                new_inputs = new_inputs + ((input),)
                continue
    
            # Apply type conversion if required.
            input_type = input.dtype
            if input_type in self.type_conversion.keys():
                new_inputs = new_inputs + (
                    (input.to(dtype=self.type_conversion[input_type])),
                )
                continue
    
            # No conversion required.
            new_inputs = new_inputs + ((input),)
    
        inputs = new_inputs
    
        if self.compiler_config.compile_depth == CompileDepth.EXECUTE:
            assert self.binary is not None, "Binary must be set for EXECUTE mode"
>           return tt_mlir.run(inputs + self.graph_constants, self.binary)
E           RuntimeError: std::get: wrong index for variant

tt_torch/dynamo/backend.py:558: RuntimeError

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

[github-actions]

	Tests	Passed ☑️	Skipped ⚠️	Failed ❌️
TT-Torch Tests	435 ran	426 passed	7 skipped	2 failed

Test	Result
TT-Torch Tests
pytest
test_conv2d.test_conv2d[dtype1-1-64-3-256-256-7-7-2-2-3]	❌ failure
test_conv2d.test_conv2d[dtype1-1-128-64-128-128-2-2-2-2-0]	❌ failure

[github-actions]

	Tests	Passed ❌️	Skipped	Failed
TT-Torch Tests	0 ran	0 passed	0 skipped	0 failed

Test	Result
No test annotations available

[github-actions]

	Tests	Passed ❌️	Skipped	Failed
TT-Torch Tests	0 ran	0 passed	0 skipped	0 failed

Test	Result
No test annotations available