Closed Jipok closed 1 year ago
@Jipok Could you try pulling the latest commit and see if that fixes it? I'm not sure how I didn't run into this before.
Also, I really recommend setting XBPS_PKG_OPTIONS
in the etc/conf
file in the repo to reduce the time you spend building and the resources needed to build:
XBPS_PKG_OPTIONS=~gfx803,~gfx900,~gfx906,~gfx908,~gfx90a,~gfx1010,~gfx1011,~gfx1012,~gfx1030,~gfx1031,~gfx1100,~gfx1101,~gfx1102
Just remove the '~' (tilde) from the GPU architecture(s) you wish to build for, i.e., to build for a gfx900 device you would need:
XBPS_PKG_OPTIONS=~gfx803,gfx900,~gfx906,~gfx908,~gfx90a,~gfx1010,~gfx1011,~gfx1012,~gfx1030,~gfx1031,~gfx1100,~gfx1101,~gfx1102
More information on the etc/conf
file here: https://github.com/void-linux/void-packages#configuration
I'm happy to help with any other issues if you encounter any more.
I got:
But fixed by deleting line "librocblas.so.3 rocBLAS-5.5.0_1". M.b. something from previous build attempt.
Well, I skipped this package and installed hipBLAS
and MIOpenGEMM
for myself. That was enough for building llama.cpp.
Regarding your issue with building hipSPARSE
, I'm not entirely sure what the issue is. I can reproduce it on my system, however, running ./xbps-src pkg hipSPARSE
a few times in a row seems to eventually get it working. There was also an unrelated typo in the template for this package which I have now fixed.
however, running
./xbps-src pkg hipSPARSE
a few times in a row seems to eventually get it working
Any technology that is sufficiently complicated is indistinguishable from magic. Complexity does not denote advancement.
This seems like a simple mistake. But I don’t understand how the void build system works at all and I can’t understand what I need to fix.
Do you have any plans to publish a binary version of the package (that you compiled)? Building this garbage from AMD is not an easy task.