On Mon, Apr 6, 2020 at 7:36 PM Artem Belevich wrote:
>
> Yup 32-bit vis 64-bit does dis trip linux tests, too some time back. We ended
> up passing explicit target triples to avoid host architecture leaking into
> the test.
>
> On a side note, shouldn't we pass -nocudalib via /clang: ?
Yes, tha
Yup 32-bit vis 64-bit does dis trip linux tests, too some time back. We
ended up passing explicit target triples to avoid host architecture leaking
into the test.
On a side note, shouldn't we pass -nocudalib via /clang: ?
On Mon, Apr 6, 2020 at 7:05 AM Hans Wennborg wrote:
> Oh no, the warning
Oh no, the warning is a red herring. The problem is I'm doing a 32-bit
build and the triple is nvptx-nvidia-cuda, not nvptx64-nvidia-cuda.
f8e1fc20cb3 should fix.
On Mon, Apr 6, 2020 at 3:54 PM Hans Wennborg wrote:
>
> I'm seeing this failure when trying to build the Windows package for
> http:/
I'm seeing this failure when trying to build the Windows package for
http://llvm.org/builds (yes, it's been a while). Not sure why it
hasn't been on the bots (maybe the Windows bots don't build the nvptx
target). Anyway, the error comes from "nocudalib" not being a valid
clang-cl option. Is it supp
