XLA development

Contact owners and managers

1–30 of 246

This group is for discussions by developers working on the implementation of XLA, which is Tensorflow's accelerated linear algebra compiler framework.

This group is not intended for Tensorflow end-user support. To report bugs in Tensorflow or XLA, please use the Tensorflow GitHub issue tracker. To ask questions about Tensorflow or XLA, please ask on stackoverflow.com using the "tensorflow" tag.

0 selected

art....@gmail.com

11/24/22

Memory Safe Computations with XLA

Hello everyone, I have created a PR for TensorFlow/XLA. That PR is the code from this paper, and I

unread,

Memory Safe Computations with XLA

Hello everyone, I have created a PR for TensorFlow/XLA. That PR is the code from this paper, and I

11/24/22

Simon King, Adrian Kuegel2

9/20/22

Why disabling fusion_merger HLO pass speeds up my model?

Hi Simon, currently we have no good way of evaluating perfectly how profitable a certain fusion is.

unread,

Why disabling fusion_merger HLO pass speeds up my model?

Hi Simon, currently we have no good way of evaluating perfectly how profitable a certain fusion is.

9/20/22

Moshe Maor, ta...@google.com2

9/16/22

XLA GSPMD and Keras

Hi, Yes, it is. The sharding APIs described here lower to GSPMD. On Thursday, May 19, 2022 at 3:59:03

unread,

XLA GSPMD and Keras

Hi, Yes, it is. The sharding APIs described here lower to GSPMD. On Thursday, May 19, 2022 at 3:59:03

9/16/22

Yanming Wang, … Stephan Herhut6

9/5/22

Performance regression using multi-threaded compilation

Looks like it. Thanks Yanming for fixing the issue! Cheers Stephan On Wed, Aug 31, 2022 at 10:21 AM

unread,

Performance regression using multi-threaded compilation

Looks like it. Thanks Yanming for fixing the issue! Cheers Stephan On Wed, Aug 31, 2022 at 10:21 AM

9/5/22

Joel, … Leary, Chris7

6/30/22

Cost of xla::Reshape

Thanks for the thorough response. I'll need some time to digest it On Tuesday, June 28, 2022 at 2

unread,

Cost of xla::Reshape

Thanks for the thorough response. I'll need some time to digest it On Tuesday, June 28, 2022 at 2

6/30/22

Noor, Abdul Rafae, … Justin Lebar7

6/7/22

Maintaining Gelu operation from Keras in XLA

Thanks, that does help. I am a little confused how, in the system you describe, you're still

unread,

Maintaining Gelu operation from Keras in XLA

Thanks, that does help. I am a little confused how, in the system you describe, you're still

6/7/22

5/21/22

Are more specific functions faster?

Hi all, Are functions that have a subset of the functionality of other functions faster, or just

unread,

Are more specific functions faster?

Hi all, Are functions that have a subset of the functionality of other functions faster, or just

5/21/22

5/8/22

Generalising Map: am I able to contribute?

Hi all, I raised this ticket https://github.com/tensorflow/tensorflow/issues/55246 for generalising

unread,

Generalising Map: am I able to contribute?

Hi all, I raised this ticket https://github.com/tensorflow/tensorflow/issues/55246 for generalising

5/8/22

Joel, Peter Hawkins2

4/7/22

What's the reasoning behind implementation-defined function behaviour?

It's just saying that we don't promise anything about the values in the "other"

unread,

What's the reasoning behind implementation-defined function behaviour?

It's just saying that we don't promise anything about the values in the "other"

4/7/22

Garvit Kulshreshtha IN, t kevin3

4/4/22

Questions about XLA instruction metadata

Hi Kevin. Thanks for the quick response (Sorry I missed your reply) I am using TF 2.6. I checked the

unread,

Questions about XLA instruction metadata

Hi Kevin. Thanks for the quick response (Sorry I missed your reply) I am using TF 2.6. I checked the

4/4/22

Joel, Peter Hawkins3

4/1/22

How best to execute computations?

Thank you Peter for the quick response. > I would not use the non-local client APIs any more Are

unread,

How best to execute computations?

Thank you Peter for the quick response. > I would not use the non-local client APIs any more Are

4/1/22

3/23/22

CompileRequest too large

Hello XLA Devs, I have an issue where the CompileRequest failed because of the xla message > 2GB.

unread,

CompileRequest too large

Hello XLA Devs, I have an issue where the CompileRequest failed because of the xla message > 2GB.

3/23/22

Abdul Rafae Noor, … Mehdi AMINI7

3/22/22

Building Tensorflow from source with Local LLVM Repo

Thank you all for your suggestions. I was able to get my llvm repository compiled with Tensorflow/XLA

unread,

Building Tensorflow from source with Local LLVM Repo

Thank you all for your suggestions. I was able to get my llvm repository compiled with Tensorflow/XLA

3/22/22

Joel, George Karpenkov2

3/22/22

Should questions still be directed to stack overflow?

Hi Joel, Despite these instructions, I don't think anyone is answering XLA questions on SO, so

unread,

Should questions still be directed to stack overflow?

Hi Joel, Despite these instructions, I don't think anyone is answering XLA questions on SO, so

3/22/22

Daniel Morandini

3/16/22

Questions on Dynamic Reshape Operation

Greetings XLA developers, We're trying to implement DynamicReshape operation on the Nx frontend.

unread,

Questions on Dynamic Reshape Operation

Greetings XLA developers, We're trying to implement DynamicReshape operation on the Nx frontend.

3/16/22

3/1/22

Questions about accuracy verification of XLA compiled subgaph

Hi XLA devs, I successfully replay an clustered subgraph from ResNet50 by using XLA on both a

unread,

Questions about accuracy verification of XLA compiled subgaph

Hi XLA devs, I successfully replay an clustered subgraph from ResNet50 by using XLA on both a

3/1/22

art....@gmail.com, Peter Hawkins4

2/9/22

Compiling Tensorflow/JAX with custom XLA optimisation passes on TPU

[Wrong references in previous email; the order is wrong] Interesting. The introduced optimisations

unread,

Compiling Tensorflow/JAX with custom XLA optimisation passes on TPU

[Wrong references in previous email; the order is wrong] Interesting. The introduced optimisations

2/9/22

Moshe Maor, George Karpenkov8

2/3/22

"multi-device function on TPUs"

XLA_CPU is just an experiment to demonstrate the issue. I bumped into this behavior when I was trying

unread,

"multi-device function on TPUs"

XLA_CPU is just an experiment to demonstrate the issue. I bumped into this behavior when I was trying

2/3/22

t kevin, … dim...@google.com4

1/7/22

Question regards to DFSMemoryScheduler

hi Dimitris ,Sanjoy, Thanks for the reply. Indeed this "mysterious" implementation works

unread,

Question regards to DFSMemoryScheduler

hi Dimitris ,Sanjoy, Thanks for the reply. Indeed this "mysterious" implementation works

1/7/22

12/27/21

Idris backed by XLA

Hi XLA devs, Just wanted to mention that I've successfully run XLA from the Idris programming

unread,

Idris backed by XLA

Hi XLA devs, Just wanted to mention that I've successfully run XLA from the Idris programming

12/27/21

Marcello Maggioni, George Karpenkov3

12/2/21

Adding rounding_mode support to convert() HLO

Thanks George, I'll send the CL then. Talking with David we might want to hold on the "_FAST

unread,

Adding rounding_mode support to convert() HLO

Thanks George, I'll send the CL then. Talking with David we might want to hold on the "_FAST

12/2/21

t kevin, George Karpenkov2

11/30/21

difference between TF_XLA_FLAGS=--tf_xla_auto_jit=2 and tf.config.optimizer.set_jit(True)

Hi Kevin, They are identical and they both enable autoclustering. TF_XLA_FLAGS=--tf_xla_auto_jit=2 is

unread,

difference between TF_XLA_FLAGS=--tf_xla_auto_jit=2 and tf.config.optimizer.set_jit(True)

Hi Kevin, They are identical and they both enable autoclustering. TF_XLA_FLAGS=--tf_xla_auto_jit=2 is

11/30/21

Moshe Maor, George Karpenkov13

11/29/21

StatefulPartitionedCall on an XLA device

> Yes, does it make sense to define an XLA device for your backend, and then force compilation of

unread,

StatefulPartitionedCall on an XLA device

> Yes, does it make sense to define an XLA device for your backend, and then force compilation of

11/29/21

t kevin, … Sanjoy Das4

11/21/21

Question regards to the design of GpuInstructionFusion pass

Hi George, Sanjoy Thanks for the clarification. Much appreciated. Have a nice day. Kevin Sanjoy Das

unread,

Question regards to the design of GpuInstructionFusion pass

Hi George, Sanjoy Thanks for the clarification. Much appreciated. Have a nice day. Kevin Sanjoy Das

11/21/21

Moshe Maor, George Karpenkov2

11/11/21

XlaLaunch and xla.compile()

Hi Moshe, xla.compile is already marked as deprecated, jit_compile=True is the preferred alternative.

unread,

XlaLaunch and xla.compile()

Hi Moshe, xla.compile is already marked as deprecated, jit_compile=True is the preferred alternative.

11/11/21

Simon King, George Karpenkov2

11/11/21

Would input batch size affect auto clustering and op fusion?

hi Simon, it seems that it's too late to look at HLO computations, since there's difference

unread,

Would input batch size affect auto clustering and op fusion?

hi Simon, it seems that it's too late to look at HLO computations, since there's difference

11/11/21

Simon King, Sean Moriarity3

10/29/21

Understanding performance difference of two calculation methods on CPU

Hi Sean, Thanks for your insights of model performance on GPU. I am very interested in the

unread,

Understanding performance difference of two calculation methods on CPU

Hi Sean, Thanks for your insights of model performance on GPU. I am very interested in the

10/29/21

Joel Berkeley, … Adrian Kuegel5

8/20/21

XLA for C/Idris

and I guess all the dependencies for "client" are covered in the tensorflow/compiler/xla/

unread,

XLA for C/Idris

and I guess all the dependencies for "client" are covered in the tensorflow/compiler/xla/

8/20/21

Naveen Swamy, George Karpenkov6

8/6/21

Breaking of XLA Computation

Could you provide a code snippet of how do you use it? On Fri, Aug 6, 2021 at 2:22 PM Naveen Swamy

unread,

Breaking of XLA Computation

Could you provide a code snippet of how do you use it? On Fri, Aug 6, 2021 at 2:22 PM Naveen Swamy

8/6/21

Moshe Maor, George Karpenkov8

8/3/21

XLA device and autoclustering

Hi Moshe, > I do see explicit registration for compile/run (that are required for the 'main

unread,

XLA device and autoclustering

Hi Moshe, > I do see explicit registration for compile/run (that are required for the 'main

8/3/21

Search

Clear search

Close search

Google apps

Main menu