torch / distro

Torch installation in a self-contained folder
BSD 3-Clause "New" or "Revised" License
556 stars 484 forks source link

build bot, with publicly available binaries? #139

Open hughperkins opened 8 years ago

hughperkins commented 8 years ago

Havent installed cutorch for ages, but having to now, to fix an issue. Takes .... a really long time :-P

Cant help pondering, doesnt the existing travis buidls provide the functionality to upload builds to an s3 bucket? Then isntallation could essentially be downloading the prebuilt binaries?

bo01ean commented 8 years ago

I think the compile for these tools is very machine specific, so pre-built binaries will not be performance oriented.

If you need something really quick, check out NVIDIA docker: https://github.com/NVIDIA/nvidia-docker

And a cutorch: https://github.com/Kaixhin/dockerfiles/tree/master/cuda-torch/cuda_v7.5

I am also hacking a script to install these Dockerfiles locally on the host machine with CUDA 8RC and torch ... I'll link around when it is done.

Kaixhin commented 8 years ago

Actually cutorch is taking so long it's made the builds on the Docker Hub time out for a while. If anyone has any ideas that would be appreciated! A CUDA 8.0 version is ready to go, I just need to sort this problem out first...

I did actually try Travis, but it runs out of resources, so unfortunately that's out.

hughperkins commented 8 years ago

If you reduce the number of sort expansions, you can drop the time a lot. Ditto for apply3 macro expansions, which take combinatorial time to compile, but are used rarely.

On 6 October 2016 23:24:24 BST, Kai Arulkumaran notifications@github.com wrote:

Actually cutorch is taking so long it's made the builds on the Docker Hub time out for a while. If anyone has any ideas that would be appreciated! A CUDA 8.0 version is ready to go, I just need to sort this problem out first...

I did actually try Travis, but it runs out of resources, so unfortunately that's out.

You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub: https://github.com/torch/distro/issues/139#issuecomment-252105726

Sent from my Android device with K-9 Mail. Please excuse my brevity.

hughperkins commented 8 years ago

(basically remove the case 3 eg https://github.com/hughperkins/cutorch/blob/master/lib/THC/THCApply.cuh#L517 , and you'll knock off a lot of time)