New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Build Tensorflow version that detects CPU instruction set at runtime and lights-up/down #25590
Comments
Maybe using MKL can help with this. |
@ericstj, |
Is this flavor of TensorFlow produced and published as part of regular releases? If you have a link to the binaries we could take a look. That would enable projects like https://github.com/SciSharp/TensorFlow.NET (used by https://github.com/dotnet/machinelearning) to redistribute these. cc @Oceania2018 @michaelgsharp |
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you. |
I responded, and asked a question. I don't believe this should be stalled. |
Hi @ericstj , I am going through the backlogs and observed this issue can be clarified to you or may be you might have already known this. Right now |
I understand it's not built by default. This issue was a request that it be added to the set of official binaries so that users might benefit from it without having to build it themselves. |
System information
Describe the feature and the current behavior/state.
Currently Tensorflow cross-compiles for different instruction sets and will warn if the CPU supports instructions that the TF build does not use, and fail to load if the CPU does not support an instruction set that the TF build uses. This makes it impossible to build an application that runs on a variety of hardware and ensure that it achieves optimal results for that hardware (or even runs at all).
I understand that TF supports cross-compilation and developers can build their own library that works best for their hardware, but this doesn't solve the case where an application developer wants to ship an application that uses TF and runs on a variety of hardware.
I understand that having multiple codepaths with runtime light-up could increase the size of TF, that could be dealt with by making this a separate flavor/configuration of TF, eg: "portable" build, and that could be published as a binary zip/tarball along side the current builds.
Will this change the current api? How?
No
Who will benefit with this feature?
Applications and libraries redistributing tensor flow binaries to run on a variety of hardware.
The text was updated successfully, but these errors were encountered: