Package Details: mesa-git 23.1.0_devel.165442.9db7c1a509f.932463d268438ce945b21718552d92ab-1

Git Clone URL: https://aur.archlinux.org/mesa-git.git (read-only, click to copy)
Package Base: mesa-git
Description: an open-source implementation of the OpenGL specification, git version
Upstream URL: https://www.mesa3d.org
Keywords: mesa X11
Licenses: custom
Conflicts: libva-mesa-driver, mesa, mesa-libgl, mesa-vdpau, opencl-mesa, vulkan-intel, vulkan-mesa-layer, vulkan-radeon, vulkan-swrast
Provides: libva-mesa-driver, mesa, mesa-libgl, mesa-vdpau, opencl-driver, opencl-mesa, opengl-driver, vulkan-driver, vulkan-intel, vulkan-mesa-layer, vulkan-radeon, vulkan-swrast
Submitter: Krejzi
Maintainer: rjahanbakhshi
Last Packager: rjahanbakhshi
Votes: 174
Popularity: 2.76
First Submitted: 2014-06-19 21:33 (UTC)
Last Updated: 2023-01-18 21:09 (UTC)

Required by (1429)

Sources (3)

Pinned Comments

Lone_Wolf commented on 2020-04-23 12:26 (UTC) (edited on 2020-11-21 22:36 (UTC) by Lone_Wolf)

Mesa and llvm are closely tied together. Everytime the llvm mesa is built against changes/updates , mesa needs to be rebuilt.

I expect anyone building mesa-git against one of the llvm trunk variants to be able to do that themselves, but atleast some of the people that built against repo llvm don't understand how to deal with such a rebuild.

The rest of this post is meant for those people.

  • Verify if latest PKGBUILD requirements match llvm repo versions.
  • If they don't , post to alert me of this.
  • DELAY the update of llvm/clang suite and their lib32- counterparts
  • Once the PKGBUILD does match, download it.

Option A - best one

  • run pacman -Syu to ensure your system is in sync with your mirrror server
  • build mesa-git in a clean chroot

see https://wiki.archlinux.org/index.php/DeveloperWiki:Building_in_a_clean_chroot for the official way or https://aur.archlinux.org/packages/clean-chroot-manager for an alternative method.

  • install the newly built mesa-git
  • run pacman -Syu
  • ready

Option B - 2nd best

  • revert to repo mesa
  • pacman -Syu
  • build mesa-git
  • switch from mesa to mesa-git
  • ready

Lone_Wolf commented on 2019-07-09 13:43 (UTC) (edited on 2021-04-15 11:48 (UTC) by Lone_Wolf)

  • choosing which llvm variant to build against

WARNING : aur helpers don't support this method at all. They check .SRCINFO and that is set to build against repo llvm when it is uploaded.

Why would I want to use llvm development versions ?

For full functionality and latest features mesa trunk master needs to be build against llvm trunk master.
If you build against stable llvm things MAY work, but you're likely to lack some features and face stability issues.

Mesa-git uses a custom environment variable MESA_WHICH_LLVM for flexibility.
It has 4 values.
1 : aur llvm-minimal-git
2 : aur llvm-git
3: llvm-git from LordHeavy' unofficial repo
4: llvm from extra repo

I use value 1 95% of the time and set this in ~/.bash_rc .
If MESA_WHICH_LLVM is unset or empty, the default value of 4 (extra llvm) is used.

Lone_Wolf commented on 2018-01-30 14:06 (UTC) (edited on 2019-07-09 13:29 (UTC) by Lone_Wolf)

  • issues
Simple issues can be dealth with through aur comments, but sometimes discussion is needed.
Use https://bbs.archlinux.org/viewtopic.php?id=212819 for that.
  • main difference with stable repo package
[extra]mesa is a split package, but aur mesa-git isn't.
Basically with aur mesa-git you get everything in one package, while [extra]/mesa allows you to leave out some parts if you don't want/need them.
Having a single package reduces maintenance and makes switching from stable to mesa-git rather easy, though reverting back to stable can be tricky.
  • meson settings
Those who compare PKGBUILDs will notice I don't use arch-meson, but meson setup.
I disagree with some of the settings made by arch-meson (especially the buildtype and enabling LTO by default) and feel using meson setup is cleaner.
  • how to update

run pacman Syu update non-repo packages for llvm if you use them

build mesa-git, log out , restart X .

run glxinfo to verify basic OpenGL functionality

build lib32-mesa-git

run glxinfo32 to verify basic OpenGL functionality for multilib programs

Latest Comments

1 2 3 4 5 6 .. 105 Next › Last »

Lone_Wolf commented on 2023-02-05 13:10 (UTC) (edited on 2023-02-05 13:11 (UTC) by Lone_Wolf)

When llvm deprecated non-opaque pointers in the llvm 15 branch, it was found to affect atleast radeonsi and llvmpipe .

Also llvmpipe provides generic functionality that is used by other drivers than just swrast. MAYBE iris/crocus/anv are not affected.

I intend to test the workaround at https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21124 .

For those that can't wait : use the most recent build of llvm-minimnal-git/llvm-git with version number 16 you still have.

eclairevoyant commented on 2023-02-05 02:05 (UTC) (edited on 2023-02-05 02:18 (UTC) by eclairevoyant)

Since LLVM 17 forces opaque pointers, there's no point in setting the opaque pointer option to false, since that's not even possible. Hence I just deleted the offending lines by adding the following to the bottom of prepare():

cd mesa/src/gallium
sed -i '/LLVMContextSetOpaquePointers/d' drivers/llvmpipe/*.c auxiliary/draw/draw_llvm.c

This may actually break the driver, idk, but I don't use llvmpipe so I don't really have a chance to test it nor does it really matter for me. For those of us who use other drivers within mesa, this may be an acceptable workaround until it's fixed upstream.

Overall (as someone who knows next to nothing about Mesa or LLVM) I felt comfortable patching it this way since the context being operated on is an LLVMOpaqueContext, and actually fixing the code would take a lot more effort upstream that I am not willing to do for a driver that I don't use - maybe not many others use the driver either and that's why no one caught it?

Edit: There is a proper upstream fix (WIP) here, for anyone curious - https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18817. But I wouldn't expect that MR to work yet.

Jonezky commented on 2023-02-05 01:46 (UTC)

Tried to build the package using llvm-minimal-git but the build fails because of the llvm 17 forcing opaqure pointers https://gitlab.freedesktop.org/mesa/mesa/-/issues/8183 Any recomended workarounds until the fix is merged?

eclairevoyant commented on 2023-01-29 23:45 (UTC)

@ahmedmoselhi either edit the PKGBUILD directly to set the default (line 53) or don't use an AUR helper since Arch doesn't support them anyway

yurikoles commented on 2023-01-27 19:07 (UTC) (edited on 2023-01-27 19:09 (UTC) by yurikoles)

AsSalamu aleykum Ahmed,

If you read the pinned message about MESA_WHICH_LLVM carefully, you may see a notice at its very beginning:

WARNING : aur helpers don't support this method at all. They check .SRCINFO and that is set to build against repo llvm when it is uploaded.

ahmedmoselhi commented on 2023-01-27 17:20 (UTC)

I set export MESA_WHICH_LLVM=3 in ~/.bashrc but each time i build in pamac it cannot detect this value and gives me error about dependencies how can i set environment var to be used with pamac?

yurikoles commented on 2023-01-26 12:42 (UTC)

@szszoke,

In case of Radeon or software rendering on CPU, yes, since most development of them in mesa is always happening in parallel with LLVM.

szszoke commented on 2023-01-26 11:56 (UTC)

Now that llvm-15 is in extra is it still strongly recommended to build mesa-git against llvm trunk master?

rjahanbakhshi commented on 2023-01-26 11:15 (UTC)

@sclu1034,

0002-intel-fs-always-mask-the-bottom-bits-of-the-sampler-.patch has nothing to do with the commit you mentioned. I believe you're talking about 0001-anv-force-MEDIA_INTERFACE_DESCRIPTOR_LOAD-reemit-aft.patch that has been removed weeks ago.

sclu1034 commented on 2023-01-24 20:52 (UTC)

0002-intel-fs-always-mask-the-bottom-bits-of-the-sampler-.patch does not apply anymore, failing the build. Building commit 81b5f6a3fd570a3a0c4fab2b8a83eb41928dbc98.

The reject is:

--- src/intel/vulkan/genX_cmd_buffer.c
+++ src/intel/vulkan/genX_cmd_buffer.c
@@ -5955,6 +5955,11 @@ genX(flush_pipeline_select)(struct anv_cmd_buffer *cmd_buffer,
    }
 #endif

+#if GFX_VERx10 == 120
+   if (pipeline == _3D)
+      cmd_buffer->state.compute.pipeline_dirty = true;
+#endif
+
    /* From "BXML » GT » MI » vol1a GPU Overview » [Instruction]
     * PIPELINE_SELECT [DevBWR+]":
     *