search

flatironinstitute / cufinufft

Nonuniform fast Fourier transforms of types 1 and 2, in 1D, 2D, and 3D, on the GPU
Other
83 stars 18 forks source link

Compatibility CUDA 5.5 #109

Closed Anesh20 closed 3 years ago

Anesh20 commented 3 years ago

1) Will cufinufft compile under CUDA 5.5 and gcc 4.4? 2) Also do I need python in order to compile the package? If seems like this based on the readme file. Thanks!

MelodyShih commented 3 years ago

Hi Anesh20,

  1. Yes, it will compile under CUDA 5.5 + GCC 4.4.
  2. No. You don't need it if you are not using the python interface.

Are you facing any problem installing the library?

Anesh20 commented 3 years ago

Hi Melody, Yes, I tried to compile and got this error msg: "nvcc fatal : unknown option 'std' " Thanks Anesh

garrettwrong commented 3 years ago

5.5 is quite old (currently cuda is 11ish..). I don't recall building against anything before 8, and I think that required a minor tweak... Iirc major release 5 will be limited to Kepler and prior cards ( so six and prior years ago or so)...

I think it is safe to remove flags related to c and c++ standards from the Makefile in this case. Likely it is complaining about that. Unfortunately I don't have anything that old to test against. You may find the performance is not inspiring on older cards. Please let us know what you find. Thanks

On Mon, May 10, 2021, 5:15 PM Anesh20 @.***> wrote:

Hi Melody, Yes, I tried to compile and got this error msg: "nvcc fatal : unknown option 'std' " Thanks Anesh

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/flatironinstitute/cufinufft/issues/109#issuecomment-837354039, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALMMC5AMZYOLMUC2QICPDDLTNBEGLANCNFSM44M67GFA .

Anesh20 commented 3 years ago

I removed all std instances in the Makefile and had to removed other syntaxes not available under 5.5 e.g. -Wno-deprecated-gpu-targets, --default-stream per-thread and sm_50 and greater. Now I am stuck with lots of errors (Output.txt) and am not sure how to proceed. The Makefile I used is also attached: Makefile.txt Any advice will be appreciated. Thanks!

ahbarnett commented 3 years ago

Dear Anesh, I really suggest you follow Garrett's advice and upgrade to a more recent CUDA, which may make your life easier in other ways too if you're doing anything on GPUs. 5.5 is now 8 yrs old, and let's just say we don't support it... Best, Alex

On Mon, May 10, 2021 at 10:08 PM Anesh20 @.***> wrote:

I removed all std instances in the Makefile and had to removed other syntaxes not available under 5.5 e.g. -Wno-deprecated-gpu-targets, --default-stream per-thread and sm_50 and greater. Now I am stuck with lots of errors (Output.txt https://github.com/flatironinstitute/cufinufft/files/6456010/Output.txt) and am not sure how to proceed. The Makefile I used is also attached: Makefile.txt https://github.com/flatironinstitute/cufinufft/files/6456021/Makefile.txt Any advice will be appreciated. Thanks!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/flatironinstitute/cufinufft/issues/109#issuecomment-837644469, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACNZRSRS3N5YLFBPPIB7JYDTNCGTRANCNFSM44M67GFA .

-- *---------------------------------------------------------------------~^`^~._.~' |\ Alex H. Barnett Center for Computational Mathematics, Flatiron Institute | \ http://users.flatironinstitute.org/~ahb 646-876-5942

Anesh20 commented 3 years ago

Thanks Alex for the advice. Unfortunately the software I am currently using was build on cuda 5.5. That's why I cannot upgrade to the latest version. Anesh

janden commented 3 years ago

Looks like the errors mentioned in your error log are only defined for cuFFT>=8, so your two options are to (1) recompile the other software you are running for a more recent version of CUDA or (2) modify the cuFinufft source code to remove those lines in cufinufft_errors.h. I would recommend option (1) since there are likely to be more incompatibilities with such an old version of CUDA (they just won't show up until you fix this first problem).

At any rate, to answer your original question, CUDA 5.5 is not supported. Closing.