enable the fallback to cpu for adaptive_avg_pool2d and max_pool2d #84
Conversation
…pt function 2. add check on fused node in the graph
torch_ipex/csrc/cpu/CustomOPs.h
Outdated
| #if defined(_DEBUG) | ||
| TORCH_WARN(e.what()); | ||
| #endif | ||
| } |
torch_ipex/csrc/cpu/CustomOPs.h
Outdated
| TORCH_WARN(e.what()); | ||
| #endif | ||
| } | ||
| if (input.device().type() == c10::DeviceType::DPCPP){ |
There was a problem hiding this comment.
Seems like incorrect indent style
|
LGTM. Please refine the code style. |
|
@pinzhenx r u okay with this fix? |
tests/cpu/test_lazy_reorder.py
Outdated
| y_cpu, | ||
| y_dpcpp) | ||
|
|
||
| self.assertEqual("cpu", y_cpu.device.type) |
There was a problem hiding this comment.
this check could be removed
| self.assertEqual(device, y_dpcpp.device.type) | ||
|
|
||
| def test_adaptive_avg_pool2d_backward_not_divisible(self): | ||
| ipex.enable_auto_dnnl() |
There was a problem hiding this comment.
is it possible to merge the forward and backward ut into a single one?
There was a problem hiding this comment.
Other tests in test_lazy_reorder.py separate the backward from the forward. Should we align with other ut in this file?
There was a problem hiding this comment.
okay, that's just fine.
This code was copied from test_mkldnn.py where we had tons of copy-pasted code there.
tests/cpu/test_lazy_reorder.py
Outdated
| y_dpcpp.backward() | ||
| self.assertEqual(x_cpu.grad, x_dpcpp.grad) | ||
|
|
||
| self.assertEqual("cpu", x_cpu.grad.device.type) |
torch_ipex/csrc/cpu/CustomOPs.h
Outdated
| } | ||
| } catch(std::exception& e) { | ||
| #if defined(_DEBUG) | ||
| TORCH_WARN(e.what()); |
There was a problem hiding this comment.
nit: 4 more spaces needed here, indent according to the catch block
also, a space between catch catch (std::exception& e)
|
LGTM. Thanks! |
…tel#84) 1. move "bn folding" and "prepack conv weight" to the hooked jit script function 2. add check on fused node in the graph
fix #79