[XLA] b/243182930 LiteralBase::Reshape

[bhack]

Click to expand!

Issue Type

Feature Request

Source

source

Tensorflow Version

master

Custom Code

No

OS Platform and Distribution

No response

Mobile device

No response

Python version

No response

Bazel version

No response

GCC/Compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current Behaviour?

Can you expand a little bit over b/243182930?
In OSS we don't see the details of this ticket:

tensorflow/tensorflow/compiler/xla/literal.cc

Lines 967 to 970 in 3d32b9f

	if (shape().is_dynamic()) {
	// TODO(b/243182930): We should consider supporting dynamic reshape.
	return Unimplemented("Dynamic reshape is not implemented.");
	}

What is required?

Standalone code to reproduce the issue

Do you have any expected failing test to add in https://github.com/tensorflow/tensorflow/blob/master/tensorflow/compiler/xla/literal_test.cc?

Relevant log output

No response

[bhack]

It was introduced with c958838

[sachinprasadhs]

@JoshVarty, Could you please comment on this issue, since it is related to your commit. Thanks!

[bhack]

@JoshVarty, Could you please comment on this issue, since it is related to your commit. Thanks!

Thanks, I would to mention the commit's author but it was masked by tensorflower-gardener in this case.

[bhack]

Just to add more context we are landing on this from keras-team/keras-cv#895 (comment)

[JoshVarty]

Sorry for the slow response, I missed this message until I cleaned out my inbox today.

I've been making a few changes to LiteralBase to better support dynamic literals. Prior to c958838 if you had some dynamic literal of shape f32[<=1024] and you called SetDynamicSize(0, {}, 1024) it would be silently converted into a static literal. However, it really should continue to be treated as a dynamic literal (just one that happens to be fully populated).

I added that comment just to indicate that dynamic reshape isn't currently supported, but that we could consider adding support for it. That said, I don't think such a change is trivial so I don't think there are any immediate plans to support it yet.

If you'd like I could probably add a test that covers this Unimplemented. Let me know what you think.

[bhack]

Any follow-up since 2020?

https://groups.google.com/a/tensorflow.org/g/mlir/c/_X48poNcbDI/m/jCC8BWIICQAJ

[JoshVarty]

There has been progress on support for dynamic shapes within XLA since 2020 but there has been no progress on Dynamic Reshape for XLA Literals as far as I'm aware.

[bhack]

here has been no progress on Dynamic Reshape for XLA Literals as far as I'm aware.

Is it something that could be contributed? Do we have any pointer to contribute this?

[tilakrayal]

Hi,

Thank you for opening this issue. Since this issue has been open for a long time, the code/debug information for this issue may not be relevant with the current state of the code base.

The Tensorflow team is constantly improving the framework by fixing bugs and adding new features. We suggest you try the latest TensorFlow version with the latest compatible hardware configuration which could potentially resolve the issue. If you are still facing the issue, please create a new GitHub issue with your latest findings, with all the debugging information which could help us investigate.

Please follow the release notes to stay up to date with the latest developments which are happening in the Tensorflow space.