Patch as posted by xenu applied.
Thanks for creating it.
Git Clone URL: | https://aur.archlinux.org/mesa-minimal-git.git (read-only, click to copy) |
---|---|
Package Base: | mesa-minimal-git |
Description: | OpenCL support in rust for mesa drivers (git version) |
Upstream URL: | https://www.mesa3d.org |
Licenses: | MIT AND BSD-3-Clause AND SGI-B-2.0 |
Conflicts: | opencl-clover-mesa, opencl-rusticl-mesa |
Provides: | opencl-driver, opencl-rusticl-mesa |
Submitter: | shoober420 |
Maintainer: | Lone_Wolf |
Last Packager: | Lone_Wolf |
Votes: | 11 |
Popularity: | 0.003826 |
First Submitted: | 2020-12-10 00:38 (UTC) |
Last Updated: | 2024-09-11 18:17 (UTC) |
Patch as posted by xenu applied.
Thanks for creating it.
Created a short patch for current build issue with llvm14 (based on https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12715)
--- a/src/amd/llvm/ac_llvm_helper.cpp
+++ b/src/amd/llvm/ac_llvm_helper.cpp
@@ -60,7 +60,7 @@
llvm::Argument *A = llvm::unwrap<llvm::Argument>(arg);
llvm::AttributeList AS = A->getParent()->getAttributes();
unsigned ArgNo = A->getArgNo();
- return AS.hasAttribute(ArgNo + 1, llvm::Attribute::InReg);
+ return AS.hasParamAttr(ArgNo, llvm::Attribute::InReg);
}
--- a/src/gallium/frontends/clover/llvm/codegen/common.cpp
+++ b/src/gallium/frontends/clover/llvm/codegen/common.cpp
@@ -233,8 +233,8 @@ namespace {
}
} else {
- const bool needs_sign_ext = f.getAttributes().hasAttribute(
- arg.getArgNo() + 1, ::llvm::Attribute::SExt);
+ const bool needs_sign_ext = f.getAttributes().hasParamAttr(
+ arg.getArgNo(), ::llvm::Attribute::SExt);
args.emplace_back(module::argument::scalar, arg_api_size,
target_size, target_align,
until that MR gets added.
I had an issue with opengl games freezing after a short time and causing a GPU reset (Radeon 6900XT). Same game run with zink worked fine. Downgrading llvm/clang to version 12 and rebuilding this package resolved the issue. Anyone else run into this?
Depends on which functionality you need.
libunwind is for debugging
lmsensors adds graphics related sensors to lmsensors database
gallium-opencl enables opencl through libclc, but mainly works on polaris (like RX 580) and earlier cards.
If you feel you don't need / want that functionality, you can disable them and simplify dependencies a bit .
Oops! you're right. I cloned that one instead!
Thanks for maintaining this really! Now I want to go more minimal with only AMD stuff, should I only edit the vulkan-drivers
and gallium-drivers
or are there extra stuff to modify in PKGBUILD as well?
This package has a makedepend on llvm-minimal-git , which conflicts & provides clang, compiler-rt and llvm .
It should refuse to build if any other llvm/clang version is present .
Are you use you were using this package and not https://aur.archlinux.org/packages/mesa-minimal+-git ?
(there's a + in its pkgname)
I think clang
should be added to makedepends. I had compilation error without it.
The recently added patch '0002-fix-ac_build_atomic_rmw-with-LLVM-13.patch' can be removed again as it was merged upstream (and results in an error during prepare).
Thank you for instructions, I have added same patch found in llvm bug tracker (proposed by you), to allow me to build.
I've added a patch that does allow building opencl again.
Incase you still want to disable opencl , replace icd
with disabled
in the line -D gallium-opencl=icd \
Pinned Comments
Lone_Wolf commented on 2023-05-22 12:07 (UTC) (edited on 2024-09-06 21:16 (UTC) by Lone_Wolf)
Build order
N.B. these packages are closely tied together, make sure you build all of them in a short period of time (worst case on my system is around 3 hours) .
Build frequency
I aim to build everything (including the lib32 part) atleast once a week.
How often you build this is a personal choice, but once a month is in my opinion the absolute minimum .
In that timeframe mesa will have seen almost 1k commits, llvm/clang gets more.
Lone_Wolf commented on 2021-01-22 18:36 (UTC) (edited on 2021-01-22 18:39 (UTC) by Lone_Wolf)
Why does this package hard depend on llvm-minimal-git ?
performance
archlinux repo packages are build with
-march=x86-64 -mtune=generic
which works on lots of machines but makes limited use of modern processor capabilities. For many packages this has little impact, but with llvm my experience is different.My local builds for llvm / mesa are done with
-march=native
and this has a noticeable effect on their performance.How big the benefit of this is depends heavily on the exact hardware you use. Worse, the software setup also impacts this. The only way to find out if it benefits your system/software setup is to try it out yourself.
easier maintenance and troubleshooting
Since i started my first mesa trunk package late in 2010 I have maintained versions without any llvm, one llvm implementation, split versions, singular versions, versions supporting multiple llvm implementations , switch from libgl hacks libglvnd to allow mesa & nvidia to cooperate etc.
Depending on one llvm variant in a non-splitted singular version results in a simple PKGBUILD that is easy to maintain.
Troubleshooting is also much easier if maintainer uses the same llvm variant as users.
If people feel those reasons are not good enough to hard depend on llvm-minimal-git , maybe I should transfer ownership .
Lone_Wolf commented on 2021-01-09 15:02 (UTC)
Why does this exist ?
Basically mesa/mesa-git build almost everything they can build.
This package tries to build just enough so everyone can use it, but disables older and/or unused components.
Check https://bbs.archlinux.org/viewtopic.php?id=261629 for a discussion about this package.