Failing uplifts due to test_maxpool2d

[kmitrovicTT]

Uplift PR failed due to

Investigating this locally I found that running pytest -svv tests/TTIR/test_maxpool2d.py::test_maxpool2d gives

tests/TTIR/test_maxpool2d.py::test_maxpool2d[act_shape0] ⠁ Initializing Chip                                                                                                                                                       
  Detecting chips (found 1)                                                                                                                                                                                                        
2024-11-13 08:45:34.241 | INFO     | SiliconDriver   - Detected 1 PCI device : [0]
2024-11-13 08:45:34.276 | INFO     | SiliconDriver   - Software version 6.0.0, Ethernet FW version 6.9.0 (Device 0)
                 Always | FATAL    | Buffer size and page size should be larger than 0 bytes!
FAILED

indicating

E       RuntimeError: TT_FATAL @ /proj_sw/user_dev/kmitrovic/work/tt-xla/third_party/tt-mlir/src/tt-mlir/third_party/tt-metal/src/tt-metal/tt_metal/impl/buffers/buffer.cpp:31: size != 0 and page_size != 0
E       info:
E       Buffer size and page size should be larger than 0 bytes!

[LPanosTT]

I have a fix for tests/TTIR/test_maxpool2d.py::test_maxpool2d in tenstorrent/tt-mlir#1249. As for tests/TTIR/test_maxpool2d.py::test_resnet_maxpool2d this PCC failure arises from the reshape inserted on the input and output of conv2d/maxpool2d ops. One of the reshapes results in garbage data. I will make a unit test for the TTNN team and file an issue for them.

[LPanosTT]

I've created the metal issue causing the resnet maxpool test to fail: tenstorrent/tt-metal#15048

[kmitrovicTT]

@LPanosTT can we close this issue? I referenced only test_maxpool2d fail in it, not the resnet one.

[LPanosTT]

@kmitrovicTT sure

[LPanosTT]

@kmitrovicTT The resnet maxpool test will pass if we do not include the convolution before, this is because it changes what reshapes are inserted into the ttnn graph down the line. I will edit this tt-xla test and perform an uplift.

[LPanosTT]

@uazizTT @kmitrovicTT uplift in here #69