You can override the detected compute version with an argument to
cudacompile.
terralib.cudacompile(function_table,debug,computeversion)
So replace 61 with a number that it accepts.
-- Zach
On Tue, Dec 13, 2016 at 7:14 PM, Michael Mara <[email protected]> wrote:
> I have some terra code that compiles cuda kernels, and works quite well on
> my machine and others with GPUs over a year old. I tried to run the same
> code on a computer with a GeForce 1080 and got the following error:
>
>
> D:\Projects\DSL\terra\src\tcuda.cpp:143: nvvm error reported (7)
> libnvvm : error: -arch=compute_61 is an unsupported option
>
> stack traceback:
> [C]: in function 'toptximpl'
> D:\Projects\DSL\terra\src\cudalib.lua:77: in function 'toptx'
> [string "<string>"]:240: in function 'cudacompile'
>
>
> Has anyone else run into this? I'm assuming the version of nvvm is older
> than the "6.1" compute architecture, and the 1080 is a "6.1" architecture,
> and so it just kind of dies. Anyone know of any work arounds besides
> updating nvvm?
>
> -Michael
>
> _______________________________________________
> terralang mailing list
> [email protected]
> https://mailman.stanford.edu/mailman/listinfo/terralang
>
>
_______________________________________________
terralang mailing list
[email protected]
https://mailman.stanford.edu/mailman/listinfo/terralang
