mesa-22.1.0 (wait for i965 and nouveau availability)

[Bruce Dubbs]

New major version.

[pierre]

From Mesa 22.0.0 Release Notes

Mesa 22.0.0 is a new development release. People who are concerned with stability and reliability should stick with a previous release or wait for Mesa 22.0.1.

[Douglas R. Reno]

This package has been split into two, with "legacy" drivers (some of which are on our list) have been moved to a separate package.

[Douglas R. Reno]

Hello everyone,

The eighth and final bugfix release, 21.3.8, is now available.

Please upgrade to the 22.0 series if your hardware supports it, or wait
for the announcement for the upcoming Amber branch for legacy hardware.

Cheers,
  Eric

We probably need to wait for the "Amber" branch.

[Bruce Dubbs]

Now version 22.0.1.

[Bruce Dubbs]

Mesa 22.0.1 Release Notes / 2022-03-29

Bug fixes

• freedreno: crash in PUBG
• [regression][bisected] MSVC: Build failure in libmesa_util when targeting x86 32-bit
• A crash in radeonsi driver
• freedreno: deqp cts fails

Changes

• panfrost: Emulate GL_CLAMP on Bifrost
• pan/bi: Handle non-2D arrays
• panfrost: Fix primitive restart with 32-bit indices
• pan/bi: Don't analyze helper reqs in !frag shaders
• anv, iris: Add Wa_16011411144 for DG2
• ir3/ra: Sanitize parallel copy flags better
• util/bitset: Fix off-by-one in bitset_set_range
• ir3/lower_spill: Fix corner case with oob offsets
• zink: workaround depth texture mode alpha.
• lavapipe: remove broken workaround for zink depth texturing.
• crocus: don't map scanout buffers as write-back
• docs: add release notes for 22.0.0
• ci: Fixup some lvp expected passes
• pick/ui: Fix mypy issue
• pick/core: drop the main_sha field
• pick/core: Add proper type annotation for update_commits callback
• pick/core: Add a method for updating commits
• pick/core: Add a MANUAL_RESOLUTION resolution
• pick/ui: correctly handle MANUAL_RESOLUTION
• core: don't pass the ui back into commit.apply
• docs: fixup zink gl 4.3 requirements
• virgl: Fix texture transfers by using a staging resource
• panfrost: Don't initialise the trampolines array
• lavapipe: Reset the free_cmd_buffers list in TrimCommandPool
• anv: Allow MSAA resolve with different numbers of planes
• spirv: Properly mangle generic pointers
• util: fix msvc build
• iris: Fix MOCS for copy regions
• anv: emit timestamp & availability using the same part of CS
• anv: flush tile cache with query copy command
• amd: add support for gfx1036 and gfx1037 chips
• llvmpipe: fix debug print iterating in set_framebuffer_state
• lavapipe: clamp clear attachments rects
• llvmpipe: clamp surface clear geometry
• zink: map R8G8B8X8_SRGB -> R8G8B8A8_SRGB
• lavapipe: update multisample state after blend state
• lavapipe: fix pipeline creation for blend and zs states
• aux/trace: rzalloc the context struct
• zink: lower dmod on AMD hardware
• zink: use 64bit mask for xfb analysis
• zink: store the correct number of components for xfb packing outputs
• zink: correctly set xfb packed output offsets
• lavapipe: use the correct value for dynamic render resolve attachment indexing
• lavapipe: skip format checks for EXTENDED_USAGE
• lavapipe: run nir_opt_copy_prop_vars during optimization loop
• anv: fix xfb usage with rasterizer discard
• anv: fix CmdSetColorWriteEnableEXT for maximum rts
• anv: fix some dynamic rasterization discard cases in pipeline construction
• lavapipe: always clone shader nir for shader states
• gallivm: fix oob image detection for cube/1dArray/2dArray/cubeArray
• llvmpipe: fix occlusion queries with early depth test
• zink: force-add usage when adding last-ref tracking
• zink: only update usage on buffer rebind if rebinds occurred
• nir/lower_is_helper_invocation: create load_helper_invocation instr with bitsize=1
• util/slab: add slab_zalloc
• radeonsi: don't clear framebuffer.state before dcc decomp
• mesa: Fix discard_framebuffer for fbo vs winsys
• freedreno/ir3: Don't try re-swapping cat3 srcs
• intel/fs: Add Wa_22013689345
• intel/fs: Add Wa_14014435656
• radv: add reference counting for descriptor set layouts
• meson: restore private requires to libdrm in dri.pc file

[Bruce Dubbs]

We need to get a complete version of mesa or another package to supplement this one.

[Bruce Dubbs]

Now version 22.0.3.

[Bruce Dubbs]

Now version 22.1.0.

[Xi Ruoyao]

Hopefully I can try to diagnose the i965 issue next week.

[Bruce Dubbs]

To summarize the issue, my current thought is that i965 is not really needed for Intel Haswell and later because the iris driver is used. The issue comes up because xorg-server, not mesa, finds the Intel chipset and looks for the i965 file but cannot find it and complains in the Xorg.0.log.

If the above is correct, there are three options. We can ignore the warning, document it, or disable it. We can also wait for the next version of xorg-server to see if upstream addresses the issue.

[Bruce Dubbs]

The actual error messages are:

(EE) AIGLX error: dlopen of /opt/xorg/lib/dri/i965_dri.so failed (/opt/xorg/lib/dri/i965_dri.so: cannot open shared object file: No such file or directory)
(EE) AIGLX error: unable to load driver i965

The message is generated in the xorg-server file glx/glxdricommon.c, function glxProbeDriver, lines 326 to 338.

I'll note that if I copy the i965_dri.so file from the previous version of mesa into /opt/xorg/lib/dri/, then the error message goes away.

[Douglas R. Reno]

The Nouveau patch was not needed for me when testing gnome-maps again (this was the application that caused me to have to add it). We should be good to drop it

xdriinfo output:

screen 0: nouveau

[Bruce Dubbs]

Fixed at commit bee072a039, but needs review.

[pierre]

Using this ticket for review. I'd like to make to points:

• I think virgl may become good to use with virtual machines at some point (or maybe already, I have not tested the last versions of qemu). I am not sure it would be picked up automatically, so maybe keep the instructions for it, either in a note or in a paragraph.
• swrast can be an emergency solution when the driver for the machine hardware is somehow broken. But I guess (hope) it is always built. If not, I'd really recommend to build it.

[Xi Ruoyao]

Replying to pierre:

Using this ticket for review. I'd like to make to points:

• I think virgl may become good to use with virtual machines at some point (or maybe already, I have not tested the last versions of qemu). I am not sure it would be picked up automatically, so maybe keep the instructions for it, either in a note or in a paragraph.
• swrast can be an emergency solution when the driver for the machine hardware is somehow broken. But I guess (hope) it is always built. If not, I'd really recommend to build it.

Both swrast and virgl are enabled by default on x86 (32-bit or 64-bit).

I'm still using -Dgallium-drivers=iris,swrast (or amdgpu,swrast) for my systems to avoid building unnecessary things.

[pierre]

Replying to Xi Ruoyao:

Replying to pierre:

Using this ticket for review. I'd like to make to points:

• I think virgl may become good to use with virtual machines at some point (or maybe already, I have not tested the last versions of qemu). I am not sure it would be picked up automatically, so maybe keep the instructions for it, either in a note or in a paragraph.
• swrast can be an emergency solution when the driver for the machine hardware is somehow broken. But I guess (hope) it is always built. If not, I'd really recommend to build it.

Both swrast and virgl are enabled by default on x86 (32-bit or 64-bit).

I'm still using -Dgallium-drivers=iris,swrast (or amdgpu,swrast) for my systems to avoid building unnecessary things.

Thanks for the info. I am not sure about the viglrenderer instructions. Maybe move them to qemu...

[pierre]

A third point: why removing (optional) for /usr/share/doc/mesa-&mesa-version;? We say:

If desired, install the optional documentation by running
the following commands as the root user:

So for me, this directory is only optionally created.

[pierre]

A small update: our mesa-demos url still references the ftp server. The mesa demos can be found at ​https://archive.mesa3d.org/demos/ now. (I have made this change)

[Bruce Dubbs]

As far as swrast and virgl go, swrast_dri.so, kms_swrast_dri.so, and virtio_gpu_dri.so are built.

I forgot to remove d3dadapter9.so. (fixed)

We do not need to add 'optional' to contents if the instructions say something is optional.