Skip to content

Vulkan model with (adaptive) avgpool1d (or maxpool1d) fail to load #12224

Open
@GregoryComer

Description

@GregoryComer

🐛 Describe the bug

Some models containing avgpool1d ops delegated to Vulkan fail to load with following error: IndexError: vector::_M_range_check: __n (which is 18446744073709551614) >= this->size() (which is 4). Maxpool1d fails in the same way, as do the adaptive versions of both avgpool and maxpool.

import torch
from executorch.backends.vulkan.partitioner.vulkan_partitioner import VulkanPartitioner
from executorch.exir import to_edge_transform_and_lower, EdgeCompileConfig, to_edge
from executorch.extension.pybindings.portable_lib import _load_for_executorch_from_buffer

class Model(torch.nn.Module):
    def __init__(
        self,
        kernel_size=3,
        stride=None,
        padding=0,
        ceil_mode=False,
        count_include_pad=True,
    ):
        super().__init__()
        self.avgpool = torch.nn.AvgPool1d(
            kernel_size=kernel_size,
            stride=stride,
            padding=padding,
            ceil_mode=ceil_mode,
            count_include_pad=count_include_pad,
        )
        
    def forward(self, x):
        return self.avgpool(x)
        
model = Model()
inputs = (
    torch.randn(1, 3, 10),
)
eager_outputs = model(*inputs)

ep = torch.export.export(model.eval(), inputs)
print(ep)
lowered = to_edge_transform_and_lower(
    ep,
    partitioner=[VulkanPartitioner()],
    compile_config=EdgeCompileConfig(_check_ir_validity=False)
).to_executorch()
print(lowered.exported_program())

et_model = _load_for_executorch_from_buffer(lowered.buffer)
et_outputs = et_model([*inputs])[0]

print(f"Inputs: {inputs}")
print(f"Eager: {eager_outputs}")
print(f"ET:    {et_outputs}")

Outputs:

IndexError: vector::_M_range_check: __n (which is 18446744073709551614) >= this->size() (which is 4)
Traceback
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
Cell In[13], line 42
     35 lowered = to_edge_transform_and_lower(
     36     ep,
     37     partitioner=[VulkanPartitioner()],
     38     compile_config=EdgeCompileConfig(_check_ir_validity=False)
     39 ).to_executorch()
     40 print(lowered.exported_program())
---> 42 et_model = _load_for_executorch_from_buffer(lowered.buffer)
     43 et_outputs = et_model([*inputs])[0]
     45 print(f"Inputs: {inputs}")
IndexError: vector::_M_range_check: __n (which is 18446744073709551614) >= this->size() (which is 4)

Versions

Run on Meta internal master, Jul 3, fbcode/SwiftShader

cc @SS-JIA @manuelcandales @cbilgin

Metadata

Metadata

Assignees

No one assigned

    Labels

    backend testerThis bug was found by the backend test suite.module: vulkanIssues related to the Vulkan delegate and code under backends/vulkan/

    Type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions