#5389: Move ttnn.repeat_interleave to c++ #8961
Conversation
ayerofieiev-tt
requested review from
eyonland,
arakhmati,
cfjchu,
xanderchin and
TT-BrianLiu
as code owners
| struct RepeatInterleave { | ||
| static inline const std::array<TensorSchema, 1> input_tensor_schemas() { | ||
| return {ttnn::TensorSchema{ | ||
| 2, // min rank |
There was a problem hiding this comment.
should be 4 according to the actual code
ayerofieiev-tt
force-pushed
the
ay/issue-5389-repeat-interleave
branch
from
c8d2e27
to
ac9da88
Compare
| ) | ||
| ); | ||
|
|
||
| // tests/ttnn/unit_tests/operations/test_repeat_interleave.py proves that it should work over dim 1 too |
There was a problem hiding this comment.
Who can help to understand why this fails?
I suspect that the way I compare with expected result is wrong due to tiling or interleaving or..
| namespace data_movement { | ||
| namespace test { | ||
|
|
||
| void run_repeat_interleave_test(tt::tt_metal::Device* device, const uint32_t repeats, const uint32_t dim) { |
There was a problem hiding this comment.
This is not the best way to write op test. @tt-asaigal directed to tests/tt_eager/ops/test_bmm_op.cpp, but I wanted to explore this path to better understand whats going on under the hood
There was a problem hiding this comment.
I think @eyonland had an template example of a ttnn unit test
| @@ -26,4 +29,20 @@ class TTNNFixture : public ::testing::Test { | |||
|
|
|||
| void TearDown() override { tt::Cluster::instance().set_internal_routing_info_for_ethernet_cores(false); } | |||
| }; | |||
|
|
|||
| class TTNNFixtureWithDevice : public TTNNFixture { | |||
There was a problem hiding this comment.
this helps to make sure that the device is properly closed even if the test throws an unhandled exception.
| module, | ||
| ttnn::repeat_interleave, | ||
| R"doc( | ||
| repeat_interleave(input_tensor: ttnn.Tensor, repeats : int, dim: int = 0) -> ttnn.Tensor |
There was a problem hiding this comment.
Important change to the inteface - repeats only accept int.
Before this change, the interface lies like it can support Tensor or Int, but in reality it only supported Int
There was a problem hiding this comment.
https://pytorch.org/docs/stable/generated/torch.repeat_interleave.html
We should probably flag this in a separate issue to note the difference relative to torch
ayerofieiev-tt
force-pushed
the
ay/issue-5389-repeat-interleave
branch
from
ac9da88
to
5b8752c
Compare
ayerofieiev-tt
force-pushed
the
ay/issue-5389-repeat-interleave
branch
from
5b8752c
to
3ea17c1
Compare
What happens
Moving ttnn.repeat_interleave to C++ as a part of #5389
Todo: