New Mesa

[Grogan]

Here's a thread I haven't started yet

Mesa 24.1.0 is released now
https://docs.mesa3d.org/relnotes/24.1.0.html

Code: Select all

New features

    VK_EXT_map_memory_placed on RADV, ANV and NVK

    VK_KHR_shader_subgroup_rotate on RADV and ANV and NVK

    VK_KHR_load_store_op_none on RADV, ANV, NVK and Turnip

    VK_KHR_line_rasterization on RADV, ANV, NVK and Turnip

    VK_KHR_index_type_uint8 on RADV, ANV, NVK and Turnip

    VK_KHR_shader_expect_assume on all Vulkan drivers

    VK_KHR_shader_maximal_reconvergence on RADV, ANV and NVK

    VK_KHR_shader_quad_control on RADV

    OpenGL 4.6 on Asahi

    OpenGL ES 3.2 on Asahi

    Mali G610 and G310 on Panfrost

    Mali T600 on Panfrost

    VK_KHR_shader_subgroup_uniform_control_flow on NVK

    alphaToOne/extendedDynamicState3AlphaToOneEnable on RADV

    VK_EXT_device_address_binding_report on RADV

    VK_EXT_external_memory_dma_buf for lavapipe

    VK_EXT_queue_family_foreign for lavapipe

    VK_EXT_shader_object on RADV

    VK_EXT_nested_command_buffer on NVK and RADV

    VK_EXT_queue_family_foreign on NVK

    VK_EXT_image_drm_format_modifier on NVK

Bug fixes

    anv: unbounded shader cache

    radv: Crash due to nir validation fail in Enshrouded

    bisected: turnip: deqp regressions

    android: sRGB configs no longer exist after !27709

    [24.1-rc4] fatal error: intel/dev/intel_wa.h: No such file or directory

    vcn: rewinding attached video in Totem cause [mmhub] page fault

    When using amd gpu deinterlace, tv bt709 properties mapping to 2 chroma

    ci: switch from CI_JOB_JWT to id_tokens

    VCN decoding freezes the whole system

    [RDNA2 [AV1] [VAAPI] hw decoding glitches in Thorium 123.0.6312.133 after https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28960

    nvk: Tracker issue for gamescope support

    nvk: Implement VK_EXT_image_drm_format_modifier

    WSI: Support VK_IMAGE_ASPECT_MEMORY_PLANE_i_BIT_EXT for DRM Modifiers in Vulkan

    [Regression][Bisected] EGL/Wayland: QT applications terminated by SIGSEGV (Address boundary error) when using dGPU

    radv: Enshrouded GPU hang on RX 6800

    NVK Zink: Wrong color in Unigine Valley benchmark

    panforst: T604 issue with using u32 for flat varyings

    [anv] FINISHME: support YUV colorspace with DRM format modifiers

    gen9/11 test became flaky: piglit.spec.!opengl 1_4.blendminmax

    24.0.6: build fails

    mesa 24 intel A770 KOTOR black shadow smoke scenes

    [bisected][regression] kitty fails to start due to `glfwWindowHint(GLFW_SRGB_CAPABLE,true)`

    r600: bisected 5eb0136a3c561 breaks a number of piglits

    Graphical glitches in RPCS3 after updating Vulkan Intel drivers

    [R600] OpenGL and VDPAU regression in Mesa 23.3.0 - some bitmaps get distorted.

    VAAPI radeonsi: VBAQ broken with HEVC

    radv/video: 10-bit support

    radv: vkCmdWaitEvents2 is broken

    Zink: enabled extensions and features may not match

    glRasterPos: “Assertion `prog->base_serialized_nir’ failed.” if a shader is loaded from the shader cache

    radv: mesa-9999/src/amd/vulkan/radv_image_view.c:147: radv_set_mutable_tex_desc_fields: Assertion `(plane->surface.u.gfx9.surf_pitch * plane->surface.bpe) % 256 == 0’ failed.

    ACO doesn’t hide lds_param_load latencies

    ACO doesn’t form a VMEM clause for image stores in one case on GFX11

    r600: Valheim hangs CAYMAN gpu (regression/bisected)

    r600: Artifacts in Oxygen Not Included around air ducts and pipes (regression, bisected)

    radv: UMR wave parsing format is outdated

    radv: GetImageMemoryRequirements2 does not look at VkImagePlaneMemoryRequirementsInfo

    RADV, regression : Objects randomly appear/disappear on Unreal Engine 4 titles using D3D12 backend on Polaris

    mesa 23.1.0-rc3 flickering textures/lighting in Unreal 4 games Polaris10

    ACO tests SIGSEGV in debian-vulkan job with LTO enabled

    radv: Address binding report for images is incorrect.

    blorp: avoid dirtying push constants in 3D

    anv: flaky vkd3d-proton test_buffer_feedback_instructions_sm51

    FTBFS: commit aaccc25a4dd9ccfc134e51a7e81168334d63a909 broke mesa snapshot build

    d3d12_screen.cpp:60:10: fatal error: ShlObj.h: No such file or directory

    r300: crash when compiling some GSK shaders

    anv: vkd3d-proton test_stress_suballocation failure

    d3d12: Zwift renders with bad textures/lighting

    nir_opt_remove_phis breaks divergence analysis

    intel: Require 64KB alignment when using CCS and multiple engines

    NVK: Misrendering with Civilization 6

    radv: RDR2 might need zerovram

    intel-clc build failure, i think?

    Issues rendering gtk4 window decorations on v3d on Fedora-40/mesa-24.0

    clc: Failure when linking with llvm+clang 18.1 (-Dshared-llvm=disabled)

    LLVM-18 build issue

    vulkan/wsi/x11: VK_SUBOPTIMAL_KHR is never reported by the swapchain

    Broken vaapi encoding on Radeon RX 6900XT

    RUSTICL creating a shared reference to mutable static is discouraged and will become a hard error

    anv: GPU hang on Assassin’s Creed Valhalla while running benchmark

    nvk: dota 2 crashes after ~5 seconds in game

    dzn: conflicting defines with DirectX headers 1.613.0

    VAAPI: Incorrect HEVC block size reported with radeonsi

    radv: WWE 2K24 has very quirky DCC issues on RDNA2

    anv: Dirt 5 crashes at tryCreatingPipelineStateFromCache

    freedreno: remove headergen2

    freedreno: remove headergen2

    vulkan/wsi: crash in dEQP-VK.wsi.wayland.swapchain.simulate_oom.min_image_count

    Document that Zink on MoltenVK is not expected to work

    KiCAD 3D Viewer - rounded pads rendered incorrectly (texture mapping or stencil test error)

    OpenSCAD rendering incorrect and inconsistent on radeonsi

    intel/fs: regression on MTL with 64bit values in UBO

    ci: split debian-build-testing?

    [freedreno] Black background on SuperTux Kart with postmarketOS and Oneplus 6T

    [radv] Half-Life Alyx renders solid black for reflective surfaces

    iris: iris_resource_get_handle returns wrong modifier

    [RX 7900 XTX] Helldivers 2 cause GPU reset

    radeon: Crash in radeon_bo_can_reclaim_slab

    regression/bisected: commit 4e3f3c10e14d8778781c81f39ced659b5ec2f148 broke mesa snapshot build

    RV530 renders improperly at non 4:3 resolutions.

    anv: new cooperative matrix failures with CTS 1.3.8.0

    nvk: Missing implementation of VkImageSwapchainCreateInfoKHR and VkBindImageMemorySwapchainInfoKHR

    mesa > 23.1.9 [opencl,video_cards_nouveau] fails to build due to missing symbol vl_video_buffer_is_format_supported

    intel/meson: Make intel_stub_gpu work with `meson devenv`

    Follow-up from “iris: Fix plane indexing and handling on image import”

    nvk,nak: Implement shaderStorageImageMultisample

    nvk,nak: Implement VK_KHR_shader_subgroup_uniform_control_flow

    `[gfxhub0] no-retry page fault` triggered by `AMD_TEST=testdmaperf` on gfx90c APU

    nvk: glcts hangs

    v3d: Line rendering broken when smoothing is enabled

    PowerVR reports minMemoryMapAlignment of 64

    RADV: GPU crash when setting ‘RADV_DEBUG=allbos’

    [intel] mesa ftbfs with time_t64

    d3d12_resource.cpp:307:49: error: no matching function for call to ‘ID3D12Heap::GetDesc()’

    radv regression between a337a0c8072d0be487e43c2b7b132e003c6d5a5e and 83f741124b66818053b6b1b2f7e42f5217a27004

    [build failure] [armhf] - error: #error “_TIME_BITS=64 is allowed only with _FILE_OFFSET_BITS=64”

    R400 should have native support for sin/cos in VS

    [radv] Crash when VkGraphicsPipelineCreateInfo::flags = ~0u

    intel: all workarounds disabled with ATS skus

    vulkan: GPL now broken

    Gen4 assertion `force_writemask_all’ failed.

    src/gallium/auxiliary/rtasm/rtasm_x86sse.c:198:10: runtime error: store to misaligned address 0x7fabba0cd011 for type ‘int’, which requires 4 byte alignment

    [radv] Holographic projection texture glitch in Rage 2

    RustiCL: Callbacks are not called upon errors

    MTL: regressions in vulkancts due to BO CCS allocations

    zink: spec@ext_external_objects@vk-image-overwrite fail

    vaapi: radeonsi: surface_region.{x,y} is not honored in processing when source is RGB

    nvk: Implement VK_EXT_shader_object

    nvk: Implement VK_EXT_graphics_pipeline_library

    turnip: UBWC disabled for MSAA

    KHR-Single-GL46.arrays_of_arrays_gl.AtomicUsage fails on MTL

    GTF-GL46.gtf42.GL3Tests.texture_storage.texture_storage_texture_as_framebuffer_attachment fails on MTL

    nvk: Implement VK_KHR_maintenance5

    [intel][anv][build][regression] - genX_grl.h:27:10: fatal error: grl/grl_cl_kernel.h: No such file or directory

    RX 6600 VDPAU not recognizing HEVC_MAIN_10 correctly

    Running an app on another AMD GPU (offload, DRI_PRIME) produces corrupted frames on Wayland.

    regression in radeonsi since 9aa205668bcbf701f8f694551c284cd8e4cc17a3 (crashes in vbo_save_playback_vertex_list)

    clang/libclc related Mesa build failures

    Ninja Install Error

    anv: add a dri config to enable implicit fencing on external memory interop

    VDPAU declares a texture as “immutable” without also setting its ImmutableLevels attribute.

    Segfault in glsl_to_nir.cpp nir_visitor::visit when assigning interface block

    [rusticl]WARNING: Project targets ‘>= 1.1.0’ but uses feature deprecated since ‘1.0.0’: module rust has been stabilized. drop “unstable-” prefix from the module name

    RX6600 hardware HEVC video decode fails for VDPAU but works for VA-API. (Can lock up GPU!)

    Rusticl panics when getting program build logs using opencl.hpp

    ue5 game issues lighting Rog Ally 7080u (z1e)

    Missing textures in RoboCop: Rogue City with mesh shaders enabled

    Intel/anv: Allow pre-compiled shader caches to be reused across multiple devices

    radv: Multiview PSO forgets to export layer in some cases.

    -Dintel-rt=enabled fails to build on 32-bit

    MTL: regressions in vulkancts due to BO CCS allocations

    intel: build failures

    regression/bisected commit 4de62731f4db56360026cbb6a3b8566f86f22466 broke HW acceleration in the Google Chrome

    i386 intel build failure: meson.build:45:6: ERROR: Unknown variable “prog_intel_clc”.

    rusticl: clEnqueueFillBuffer (among others) fails on buffers created from GL object.

    MTL raytracing regression

    [ANV/DG2] Unexpectedly slow replay of RenderDoc frame capture of Resident Evil 4 Remake

    zink: flickering artifacts in Selaco

    [ADL] gpu hang on dEQP-VK.synchronization.internally_synchronized_objects.pipeline_cache_graphics

    Turnip spam on non-turnip devices

    Intermittent compiler failures when building valhall tests

    panfrost: graphical artifacts on T604 (T600)

    Dying Light native artifacts on Intel A770

    r300: Amnesia: The Dark Descent heavy corruption

    [ANV/DG2] Age of Empires IV fullscreen “banding” artefacts

    [mtl][anv] dEQP-VK.pipeline.monolithic.depth.format.d32_sfloat.compare_ops.* failures when run multithreaded

    [mtl][anv] flaky tests in pipeline.monolithic.extended_dynamic_state*stencil_state_face* series

    Broken colors/dual-source blending on PinePhone (Pro) since 23.1.0

    r600/sfn: “Indexed locks of kcache banks 14 and 15 are ignored” in the ALU clause documentation

    turnip: Logarithmic-time subgroup reductions using brcst.active and getlast.w8 instructions on a6xx gen4+

    GTF-GL46.gtf42.GL3Tests.texture_storage.texture_storage_compressed_texture_data regression

    microsoft/compiler: Missing globally-coherent logic

    Regression between 23.0.4 and 23.1.0: texture glitches in osgEarth

    [Broadcom] Warning when runnin every OpenGL game on Vulkan using ZINK

    radeonsi unsynchronized flips/tearing with KMS DRM rendering on 780M

    radeonsi has an unchecked hard dependency on libelf

    DR crashes with mesa 24 and rusticl (radeonsi)

    Piglit tests assert on gen9 with zink

    vlc crashes when playing 1920x1080 video with Radeon RX6600 hardware acceleration and deinterlacing enabled.

    [radeonsi] Regression: graphical artifacting on water texture in OpenGOAL

    Assertion when creating dmabuf-compatible VkImage on Tigerlake

    Palworld fails to launch on Intel Arc unless “force_vk_vendor” is set to “-1”.

    panfrost: implement line smoothing

    r300: backend DCE fails in piglit glsl-vs-copy-propagation-1.shader_test

    [AMDGPU RDNA3] Antialiasing is broken in Blender

    MTL: vulkan cooperative matrix tests gpu hang on MTL

    nvk: Implement VK_KHR_zero_initialize_workgroup_memory

    Assassin’s Creed Odyssey wrong colors on Arc A770

    VAAPI: EFC on VCN2 produces broken H264 video and crashes the HEVC encoder

    etnaviv, modesetting, and glxgears

    The Finals fails to launch with DX12 on Intel Arc unless “force_vk_vendor” is set to -1.

    nvk: `VK_KHR_zero_initialize_workgroup_memory` and `VK_KHR_shader_subgroup_extended_types` not marked as complete in features.txt

    nvk: Implement variableMultisampleRate

    VA-API CI tests freeze

    radv: games render with garbage output on RX5600M through PRIME with DCC

    Warning when use ALIGN over uint64_t and uintptr_t

A more changelog-like list of changes follows on the release notes page.

Too bad I'm on Master/Main (it's 24.2 branch) or I'd be excited

(It's better for me to stay on their main tree, as it's easier to track changes and if something is NFG, easier to track down the commit. They are pretty careful, it's rare that I would experience significant breakage.)

[Grogan]

Oh, and this change right here

microsoft/compiler: Missing globally-coherent logic

That sounds just like Microsoft

I could not resist.

[Grogan]

Mesa 24.1.1 is out now.
https://docs.mesa3d.org/relnotes/24.1.1.html

Bug fixes listed (these are more posted bugs resolved) but detailed changes below which have more value.

--------------------------------------------------------------------------------------------------------

I am on their main branch, and hadn't done a build in a while because I was troubleshooting that display problem and didn't want to screw around and have to log off and mode switch etc. I did one tonight and it broke (at least) my Starfield game. What looked like incorrectly rendered meshes (I didn't try to step off the world, I avoided those areas until I quit lol) and terrain that didn't render. I quit right out of the game and came back, and it was still fuckt like that.

Going back to my last Mesa package "24.2.0_devel.189807.7c07f1cdfb6" from May 27 fixed it.

That looks pretty serious (likely wouldn't just affect that game, but it was only the second I tried) and I don't think I need to be the one to bisect it. I'll try again in a few days and if it doesn't get fixed or I don't see a related bug report I'll try to figure it out.

[Grogan]

I did a new mesa checkout today, and it's still broken for me with respect to Starfield. I wanted to do a recompile against current system rather than just install my old mesa build, and didn't want to figure out what commit isn't broken for me (my head isn't into debugging this right now) so I'm going back to stable Mesa 24.1.1 for a while. I'll try again next week or so.

[Grogan]

... and here it is

Starfield invisible terrain on a 7900 XTX
https://gitlab.freedesktop.org/mesa/mesa/-/issues/11324

(bisected to commit be49b02f so I know where this is now too)

I'm glad somebody else did this (I should have but I was preoccupied)

There's also a potential fix posted by Timur Kristof (related issue) that I could make a patch from.

[Grogan]

Yep, this fixes it (tessellation related)

Code: Select all

diff --git a/src/amd/common/ac_nir_lower_tess_io_to_mem.c b/src/amd/common/ac_nir_lower_tess_io_to_mem.c
index 3a0acddc70b3873f6484699f1a3acf1dec3a7440..8d33503c9eef9ec20a7473ef86f3f6098abd6291 100644
--- a/src/amd/common/ac_nir_lower_tess_io_to_mem.c
+++ b/src/amd/common/ac_nir_lower_tess_io_to_mem.c
@@ -803,7 +803,13 @@ hs_store_tess_factors_for_tes(nir_builder *b, tess_levels tessfactors, lower_tes
    nir_def *offchip_offset = nir_load_ring_tess_offchip_offset_amd(b);
    nir_def *zero = nir_imm_int(b, 0);
 
-   if (st->tcs_tess_level_outer_mask) {
+   /* For linked shaders, we must only write the tess factors that the TES actually reads,
+    * otherwise we would write to a memory location reserved for another per-patch output.
+    */
+   const bool tes_reads_outer = st->tes_inputs_read & VARYING_BIT_TESS_LEVEL_OUTER;
+   const bool tes_reads_inner = st->tes_inputs_read & VARYING_BIT_TESS_LEVEL_INNER;
+
+   if (st->tcs_tess_level_outer_mask && tes_reads_outer) {
       const unsigned tf_outer_loc = hs_output_vram_map_io_location(b->shader, false, VARYING_SLOT_TESS_LEVEL_OUTER, st);
       nir_def *vmem_off_outer = hs_per_patch_output_vmem_offset(b, st, NULL, tf_outer_loc * 16);
 
@@ -813,7 +819,7 @@ hs_store_tess_factors_for_tes(nir_builder *b, tess_levels tessfactors, lower_tes
                            .access = ACCESS_COHERENT);
    }
 
-   if (tessfactors.inner && st->tcs_tess_level_inner_mask) {
+   if (tessfactors.inner && st->tcs_tess_level_inner_mask && tes_reads_inner) {
       const unsigned tf_inner_loc = hs_output_vram_map_io_location(b->shader, false, VARYING_SLOT_TESS_LEVEL_INNER, st);
       nir_def *vmem_off_inner = hs_per_patch_output_vmem_offset(b, st, NULL, tf_inner_loc * 16);

[Grogan]

I was expecting 24.2.0-rc5 today, but it's 24.2.0 instead

* [new tag] mesa-24.2.0 -> mesa-24.2.0

Release notes, even.
https://docs.mesa3d.org/relnotes/24.2.0.html

Distributors don't consider that "stable" anymore until .1 though.

[Grogan]

Going through that changelog, I just realized (This has been in my mesa builds for a while lol) that the new shader cache database method (it's actually Valve's "fossilize" cache, *.foz) is now the default. This caught my eye.

New Features:
...
...
* Defaulting to a new shader cache implementation, reducing filesystem overhead.

To confirm that meant what I thought it did, I found the actual "changes" in the log. Yep! I can get rid of MESA_DISK_CACHE_SINGLE_FILE=1 from my env.

Bas Nieuwenhuizen (5):

radv: Use zerovram for Enshrouded.

util/disk_cache: Fix cache marker refresh.

util/disk_cache: Delete the old multifile cache if using the default.

util/cache_test: Add tests for old cache deletion.

relnotes: Add an entry about the new cache default.

That's good news (and it's good that mesa will delete the old cache if present... unless overridden to use the old 256 directory, 40,000+ file garbage heap, presumably by an environment variable)

P.S. Actually no, it's something different again. Fuck this, it's still too silly.

Now I have:

~/.cache/mesa_shader_cache (created, but currently empty... we'll see later if it populates with anything.)
~/.cache/mesa_shader_cache_db

Which is 50 directories (part0 through part49), each containing a .db and an index (.idx). Multiply that for every Proton game that sets its own cache location variable. How is that better? After I figure out exactly what this does now, I'm putting my MESA_DISK_CACHE_SINGLE_FILE=1 variable back, which still uses a single database for each of opengl 64 bit, opengl 32 bit and RADV 64 bit and 32 bit.

(side note: I'm glad I used a wildcard, mesa_shader_cache* for my cleanup script )

P.S. mesa_shader_cache doesn't seem to get populated, it's probably just created as an artifact of how it deletes the old style cache. So far, only mesa_shader_cache_db/part0/mesa_cache.db has grown appreciably in size (vulkan shaders). So we'll let it handle them like this for now. I'm not sure why it creates all those databases, but I'll find out after I play a bunch of different types of games. (opengl, 32 bit, 64 bit, vulkan etc.)

[Grogan]

Looks like Mesa 24.2.1 is a release now (not considered stable until .1 now). A new tag in the last few hours, I had updated but didn't get to the build, so I updated the tree again and there it was. (Not what I'm building though, I want to go back to the main tree now). No release notes yet.

I'd expect rolling distros to have Mesa 24.2.1 soon, with the new features of 24.2.

[Grogan]

I accomplished what I set out to do, latest Mesa 24.3.0-devel from git, latest firmware... vulkan video enabled by default (for at least RDNA3 architecture... others may have to enable with environment variables) and I can use it in mpv as the "gpu-next" driver. It works well, and I can confirm it's using vulkan from the output of lsof.

Also:

Code: Select all

[grogan@nicetry ~]$ mpv --hwdec=help | grep vulkan
  vulkan (h264-vulkan)
  vulkan (hevc-vulkan)
  vulkan (av1-vulkan)
  vulkan-copy (h264-vulkan-copy)
  vulkan-copy (hevc-vulkan-copy)
  vulkan-copy (av1-vulkan-copy)

Too bad it doesn't work in Cosmic. It looks like this is SUPPOSED to be working with Wayland, but I get the old Bad Window (invalid window parameter) X11 error. We'll see about a git build of MPV soon.

mpv-gpu-next.jpg (89.45 KiB) Viewed 111586 times

P.S. Non RDNA3 or non-latest VCN firmware, or even non-latest Mesa can still test vulkan video with this variable. I'd probably guess that it works OK with RDNA2

Code: Select all

RADV_PERFTEST=video_decode

with luck

[Grogan]

So there have been some big code cleanups in Mesa it seems (current 24.3.0-devel git) Not only removing gallium-omx (OpenMax media support) as I disable that, but other things too. At first I thought something was wrong, as my Arch packages reported a 90 Mb saving during install. Then I noticed that my packages were about 5 megs smaller each (mesa and lib32-mesa). Nope, nothing was amiss, that was only a good thing.

Similar today, I built the same sources on Gentoo and my mesa build is a lot smaller.

They may have gotten rid of a lot of unnecessary headers and things... obviously it's something that compresses a lot for the on-disk size to be that much smaller than the package archives.

[Grogan]

Mesa 24.3.0 tagged today. I was expecting a -rc3 but they jumped it. The site hasn't been updated, but release notes are in the git tree in the 24.3.0 branch.

https://gitlab.freedesktop.org/mesa/mes ... _type=tags

[Grogan]

* [new tag] mesa-24.3.1 -> mesa-24.3.1

That means it's considered "stable" and distros should start picking it up.

[Grogan]

Well, I'm back on mesa main (25.0.0-devel) again now. I tried sticking with the stable 24.3, but stable moves too slowly.

New everything... vulkan loader (which was complaining that my vulkan implementation doesn't support its new 1.4 spec), libdrm, kernel driver, firmware etc. but not mesa. I wanted to recompile it anyway on the system, so I might as well get it up to current.

[Grogan]

I noticed the Mesa 24.3.2 release tagged today

* [new tag] mesa-24.3.2 -> mesa-24.3.2

Release notes:
https://docs.mesa3d.org/relnotes/24.3.2.html

New features

None

Bug fixes

crash on video playback

anv: bellwright needs force_vk_vendor=-1 %command% to launch

Up to 60% perf drop in SynMark DrvRes benchmark

Memory leak on closing and re-opening X11 windows

anv: `MESA: warning: INTEL_HWCONFIG_MIN_GS_URB_ENTRIES (2) != devinfo->urb.min_entries[MESA_SHADER_GEOMETRY] (0)`

d3d12: va-api: build failure regression since 24.3.0-rc1 with MinGW GCC and clang

anv: Marvel Rivals XeSS crash, game needs force_vk_vendor=-1 env variable

aco: two nir_shader_clock are miss optimized to one for GFX12

aco: nir_ddx/ddy v_interp optimization does not work on GFX12

anv: vkd3d-proton hangs on TGL with test_multisample_rendering

More details under "Changes"

I don't think I'm going to revert back to that, I'm having very good performance with the 25.0.0-devel git build I have. Also Vulkan 1.4 spec for when things start to make use of the extensions (e.g. development of vkd3d-proton and dxvk)

[Grogan]

Mesa 25.0.0 released today. I was expecting another release candidate, but it's a release after -rc3. When Mesa goes -rc, I usually get off the main branch when it bumps to the next (e.g. 26.0.0-devel), because that's when they may start breaking eggs.

Maybe I'll use the release for a few days before I go back to main/master (I still want to call it "git master" though it's not politically correct anymore)

No release notes posted yet, but I have them in my source tree:

New features
------------

- cl_khr_depth_images in rusticl
- Vulkan 1.4 on radv/gfx8+
- VK_KHR_dedicated_allocation on panvk
- VK_KHR_global_priority on panvk
- VK_KHR_index_type_uint8 on panvk
- VK_KHR_map_memory2 on panvk
- VK_KHR_multiview on panvk/v10+
- VK_KHR_shader_non_semantic_info on panvk
- VK_KHR_shader_relaxed_extended_instruction on panvk
- VK_KHR_vertex_attribute_divisor on panvk
- VK_KHR_zero_initialize_workgroup_memory on panvk
- VK_KHR_shader_draw_parameters on panvk
- VK_KHR_shader_float16_int8 on panvk
- VK_KHR_8bit_storage on panvk
- VK_EXT_4444_formats on panvk
- VK_EXT_global_priority on panvk
- VK_EXT_global_priority_query on panvk
- VK_EXT_host_query_reset on panvk
- VK_EXT_image_robustness on panvk
- VK_EXT_pipeline_robustness on panvk
- VK_EXT_provoking_vertex on panvk
- VK_EXT_queue_family_foreign on panvk
- VK_EXT_sampler_filter_minmax on panvk
- VK_EXT_scalar_block_layout on panvk
- VK_EXT_tooling_info on panvk
- depthClamp on panvk
- depthBiasClamp on panvk
- drawIndirectFirstInstance on panvk
- fragmentStoresAndAtomics on panvk/v10+
- sampleRateShading on panvk
- occlusionQueryPrecise on panvk
- shaderInt16 on panvk
- shaderInt64 on panvk
- imageCubeArray on panvk
- VK_KHR_depth_clamp_zero_one on RADV
- VK_KHR_maintenance8 on radv
- VK_KHR_shader_subgroup_rotate on panvk/v10+
- Vulkan 1.1 on panvk/v10+
- VK_EXT_subgroup_size_control on panvk/v10+
- initial GFX12 (RDNA4) support on RADV


Bug fixes
---------

- radeonsi: regression with running DaVinci Resolve under rusticl since 666a6eb871d5dec79362bdc5d16f15915eb52f96
- [ANV][LNL] - Black Myth: Wukong (2358720) - Corruption is visible near the edge of water.
- [ANV][LNL] - Hogwarts Legacy (990080) - Pixelated corruption is visible when looking out at the water.
- radv/video/h265: pps.flags.transform_skip_enabled_flag = 1 randomly hangs GPU
- [ANV][LNL] - Steel Rats (619700) - Game crashes after opening logos play before reaching main menu
- nvk: Implement host-only descriptors
- Gnome-shell Wayland fails to start with segfault at modifier-less driver
- [ANV][LNL] - DYNASTY WARRIORS: ORIGINS (2384580) - Dithered transparency has vertical bands.
- AMD Radeon R9 270 randomly causes video playback applications to crash with "amdgpu: The CS has been rejected"
- Rendering issues on GravityMark with RadeonSI ACO
- i915: multiple tests assert with tgsi_ureg.h:893: ureg_swizzle: Assertion \`reg.File != TGSI_FILE_NULL' failed.
- shaders/closed/steam/deus-ex-mankind-divided/260.shader_test fails NIR validation
- shaders/closed/steam/deus-ex-mankind-divided/260.shader_test fails NIR validation
- panvk : vk_pipeline_cache_object_deserialize: Assertion \`reader.current == reader.end && !reader.overrun' failed.
- 46a8d5e7ef61735416d0c54886a7a9930621ae2c causes a permission denied spam
- [BUILD] Build Failure: Implicit Function Declaration 'timespec_sub_saturate' (loader_wayland_helper.c)
- intel genX_acceleration_structure: missing dependency to bvh/header.spv.h
- KHR_subgroup glsl parsing broken
- intel: add config options to disable ELK compiler bits
- a618: godot-tps-gles3-high trace reproducible flakes
- radv: mesh shader depth-only rendering is broken
- anv: Enable VK_FORMAT_A4R4G4B4_UNORM_PACK16_EXT for Android 15
- Using a buffer allocated on a rx 6800XT for scanout on a Ryzen 7950X results in glitches
- Systemfreeze from mesa version 1:24.3.0-1-x86_64 and above with Chromium and derivatives [and more or less all other graphic related things]
- msm_kgsl.h:560:21: error: expected ‘:’, ‘,’, ‘;’, ‘}’ or ‘__attribute__’ before ‘*’ token
- [radeonsi] VC1 hardware decoding over vaapi outputs green screen
- consecutive glDrawPixels do not reflect a changed pixel mapping
- Crashing while Processing Shaders in Marvel Rivals on Mesa 24.3.2 & Mesa 24.3.3
- Assertion \`nir_cf_node_get_function(&block->cf_node)->structured' failed
- r300: Conditional jump or move depends on uninitialised value in Xnine.mova test
- anv: Mesh shaders with two OpSetMeshOutputsEXT instructions are not supported
- hasvk: apps crash since "intel/compiler: Remove usage of variable length arrays"
- nir_validate should check metadata
- anv: vkcube(pp) segfault in multi-GPU config, apparent vkCreateSwapchainKHR failure
- anv,regression: Black square artifacts in Fenyx Rising on BMG
- [anv] Cyberpunk visual corruption on BMG
- [ANV][LNL] - Cyberpunk 2077 (1091500) - Flickering mesh during benchmark.
- Intel Arc A770: Crosshair in THE FINALS renders too large
- 3d render issues in Chromium after 1:24.3.1-3 update over 1:24.2.7-1 of mesa package
- intel/compiler: Out of bounds read in brw_eu_compact.c
- intel/compiler: Out of bounds read in brw_eu_compact.c
- egl,dri2: Segfault when running wayland clients on non-default GPU
- anv,regression: Visual glitches in Ghost of Tsushima on BMG
- anv, regression: Resident Evil 2 d3d12 freezes in main menu on a Arc b580
- radeonsi: fails to build with libc++
- Random mesa crashes in kwin_wayland on a 6600XT
- enc->enc_pic.enc_pic_order_cnt_type always zero even if pic->pic_order_cnt_type non-zero that application set
- [anv] Visual corruption in Cyberpunk on LNL and BMG
- [anv] Borderlands 3 visual corruption on BMG
- [ANV] LNL triangle corruption on clothing in HogwartsLegacy-trace-dx12-1080p-ultra
- Intel: Dark graphical glitches on cars and characters on Disney Speedstorm
- Regression in VA-API decoding
- freedreno: fails to build with Android NDK 27c
- hk_cmd_draw.c32: error: expression in static assertion is not constant
- anv/gfx12: Enable non-zero fast clears for non-FCV CCS_E
- gen12: 5% regression in factorio
- 32-bit: error: format ‘%lx’ expects argument of type
- regression;bisected;FTBFS: commit b13e2a495e9e3da56add7d852ca01b2cd7eef52d breaks x86_32 mesa build
- glxext.c: error: 'struct glx_screen' has no member named 'frontend_screen'
- regression;bisected;FTBFS: commit ae76a6a04596bfdbd37bab165bc5f2a5ff60d389 breaks x86 mesa build
- Can't allocate dpb buffer on firefox
- Segmentation fault resetting a query pool used to get BLAS properties
- libvulkan_lvp link fails if glslangValidator is not installed
- lvp acceleration structure broken on \`main` but not on \`staging/24.x`
- radv: warning that "radv is not a conformant Vulkan implementation" on Navi 32
- [anv][UHD630] DXVK 2.5 - 2.5.2 with DXVK_HUD=compiler or DXVK_HUD=fps freezes the game or the entire system (Works without compiler/fps HUD, DXVK 2.4.1 works fine)
- Licenses seems incomplete/misleading
- anv: Symbol clash in intel_batch_decoder build when expat not available
- glcts failures on LNL/BMG
- Lavapipe vulkan 1.4 support?
- d3d12 vaapi: thread safety issues
- anv: Missing textures and glitches in It Takes Two (game)
- [anv][bisected] GravityMark segfault when enabling u-trace on RT workload
- features.txt does not have a Vulkan 1.4 section despite some drivers already supporting the new version
- Black screen bug that only affects AMD
- Failure to correctly decode H.264, possibly specific to use of array output view
- X1-85: Portal 2: Bottom of portal gun disappears
- X-Plane 12: Prop disc rendering regression
- Errors when enumerating devices create incorrect expecations
- Resident evil 3 remake hanging - f8b584d6 regression
- R6700XT: QP value doesn't affect output when using CQP rate control w/ H264/H265 VAAPI encoders
- Bug in Mesa headers: \`error: redefinition of typedef 'GLsync'`
- nak: Crash when starting The First Descendant
- [r300] Regression in f424ef18010 breaks wayland on RS480M
- anv: Missing text in Age of Mythology Retold on a Arc b580
- RustiCL: and Clover broken with 9b7ea720c93 (!32713 (merged))
- nvk: Artifact Classic crash at loading screen
- radeonsi VAAPI - vc-1 interlaced decoding garbled on Polaris
- VDPAU AV1 hardware decoding broken for Mesa 25.0.0-devel
- mesa: st_glsl_to_nir call to nir_opt_fragdepth might not be valid with MSAA
- rusticl: warning: pointers cannot be transmuted to integers during const eval
- rusticl: warning: pointers cannot be transmuted to integers during const eval
- X1-85: Half Life 2 water rendering artifacts
- crash on video playback
- anv: Allow buffer compression for vkd3d by default?
- anv: bellwright needs force_vk_vendor=-1 %command% to launch
- [anv] Possible regression from !31269
- Up to 60% perf drop in SynMark DrvRes benchmark
- Memory leak on closing and re-opening X11 windows
- SIVPE errors on GPU-based screen recording (Radeon 890M)
- d3d12: va-api: build failure regression since 24.3.0-rc1 with MinGW GCC and clang
- anv: Marvel Rivals XeSS crash, game needs force_vk_vendor=-1 env variable
- anv: \`MESA: warning: INTEL_HWCONFIG_MIN_GS_URB_ENTRIES (2) != devinfo->urb.min_entries[MESA_SHADER_GEOMETRY] (0)`
- aco: two nir_shader_clock are miss optimized to one for GFX12
- aco: opengl buffer blit test fail when using aco on GFX12
- aco: nir_ddx/ddy v_interp optimization does not work on GFX12
- VAAPI b_depth 2 causes "manage_dpb_before_encode UVD - Failed to find ref0" error
- regression;bisected;FTBFS: commits 37d47913437e2e9f72283ea8bffce00efc40fce2 and e67e44522f4f5de4fcde53ad0fb75e396ef31f52 breaks x86 mesa build
- anv: Enable storage image compression on TGL
- zink: zink_create_quads_emulation_gs doesn't write primitive ID
- DZN/DXIL doesn't validate GTK shaders
- black screen and "Failed to add framebuffer" error in wayland compositors when not filtering dmabuf formats with ccs modifiers on intel graphics when upgrading to mesa 24.3.0
- nir: nir_opt_if_merge_test fails validation with NIR_DEBUG=validate_ssa_dominance
- radv: Vulkan AV1 video decode glitches
- radv: support RGP captures for purely compute pipelines
- regression;bisected: c49a71c03c9166b0814db92420eadac74cbc4b11 leads to artifacts if on top of launched game (in full screen mode) show list running apps (Hold Alt + Tab)
- !32067 broke piglit "spec\@egl_khr_create_context\@no-error context gl"
- Intel: Re-enable bo cache in iris driver (Xe2)
- [amdgpu][regression] GPU Hang/Reset Triggered by Several Applications
- ANV: X4 Foundations crashes with vkAllocateDescriptorSets -12
- About twenty vulkan-samples cases will crash caused by the same error while running on PanVK
- Firestorm crashes on startup with Mesa 24.3
- anv: Use-after-free detected by AddressSanitizer while running dEQP-VK
- GPU process crash via WebGPU shader - UAF in mesa gcm_schedule_early_instr at src/compiler/nir/nir_opt_gcm.c:477
- radv: DCC causes glitches in Red Dead Redemption 2
- A5xx rendering issues with firefox
- [ANV][Regression] Broken rendering in Flycast + Per-Pixel Alpha Sorting
- [TGL][anv] Performance regression in Dota 2 replay
- vtn: OpTypeStruct in kernel parameters trigger assertion in glsl_types.h
- anv: Assertion failure in \`dEQP-VK.image.extended_usage_bit_compatibility.image_format_list.s8_uint_optimal_transfer_src_bit`
- radv: Resident Evil 6 Benchmark Tool has artifacts on 7900 XTX when DCC is enabled, game launched on 4K monitor without scaling and with FullHD settings
- [AMD RX 6700 XT] Artifacts while upscaling games in fullscreen mode
- Distorted pixelated graphics with Radeon RX 7900 XT with some games
- Total War Warhammer 2 Graphical Glitch
- Glitching artifacts in tile shaped patterns on 6700 XT, when using upscaled fullscreen game on labwc
- anv: Page fault when using MTL simulator in dEQP-VK.ray_tracing_pipeline.data_spill.report_intersection.float32
- mesa_cache_db.c:316:33: error: call to undeclared function 'mremap'
- [trunk] shaders fail hard in openmw after cbfc225e2bda2c8627a4580fa3a9b63bfb7133e0
- u_perfetto.h:33:9: error: unknown type name 'clockid_t'; did you mean 'clock_t'?
- brw_fs_opt_copy_propagation incorrectly handles size changes of uniforms
- RADV Command buffer reuse doesn't reinitialize is_secondary
- Virgl:Qcom sa8155 GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS/GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS is too small to run antutu benchmark apk
- nouveau paraview msaa corruption 23.1 bisected regression
- mesa fails to build due to missing SPV_ENV_UNIVERSAL_1_6 symbol

[Grogan]

* [new tag] mesa-25.0.1 -> mesa-25.0.1

Mesa 25.0.1 now, with releasenotes published, even (relnotes since 25.0.0)

https://docs.mesa3d.org/relnotes/25.0.1.html

Now distros will start picking this up.

I would imagine they want to get this out before they have to migrate to new git hosting. Gitlab is losing their main back end hosting and apparently is going to be down for a week. That's going to sting a bit, it'll be crippling for development of a lot of projects. Mesa, Xorg, Wayland and other freedesktop projects are hosted on that gitlab setup.

New features

None

Bug fixes

Zink: Kopper’s present thread causes Wayland protocol races

GLmatrix needs aligned malloc

Lavapipe crashes if no Position is output in mesh shader

[RADV/aco][regression][bisected] - Avowed (2457220) - GPU hangs near Watermill outside of Dawnshore

radv/sqtt: assertion “layout transition marker should be only emitted inside a barrier marker”

[radv] Glitchy ground geometry regression in Total War Warhammer III on RX 7600

More detailed list of changes in the release notes as always.

[Grogan]

* [new tag] mesa-25.0.2 -> mesa-25.0.2

Mesa 25.0.2 now. No published release notes, but they are in my source tree:

Mesa 25.0.2 is a bug fix release which fixes bugs found since the 25.0.1 release.

Mesa 25.0.2 implements the OpenGL 4.6 API, but the version reported by
glGetString(GL_VERSION) or glGetIntegerv(GL_MAJOR_VERSION) /
glGetIntegerv(GL_MINOR_VERSION) depends on the particular driver being used.
Some drivers don't support all the features required in OpenGL 4.6. OpenGL
4.6 is **only** available if requested at context creation.
Compatibility contexts may report a lower version depending on each driver.

Mesa 25.0.2 implements the Vulkan 1.4 API, but the version reported by
the apiVersion property of the VkPhysicalDeviceProperties struct
depends on the particular driver being used.

New features
------------

- None


Bug fixes
---------

- RADV: logic used to avoid running on CDNA is faulty
- [LNL/BMG] Assassin's Creed Valhalla trace replay hang
- X11 + Zink on NVK flickers older frames in Firefox based browsers
- Vulkan conformanceVersion is reported as 0.0.0.0 in Mesa 25.0.0
- VRAM Abnormal use on mesa 25.0
- [radv][regression] Multiple games detect the wrong amount of vram
- Resident Evil 2 Remake flickers
- OpConstantNull not supported for OpTypeCooperativeMatrixKHR
- v3dv: vkcube-wayland crashes on raspberry pi 5 kernel 6.12 and latest mesa
- GMSH Visualization Fails with radeonsi:can't compile a main shader part, Fedora 41 AMD 7900xt
- AMD VDPAU deinterlacing SIGSEGV
- radv: vkd3d-proton test_primitive_restart_list_topology_stream_output randomly fails on NAVI2X
- Mesa 24.1 introduced a Vulkan problem with DOOM 2016 on AMD 780M GPU
- nouveau & zink+nvk: Flashing in Firefox and Thunderbird on Hyprland


Changes
-------

Aaron Ruby (1):

- gfxstream: Downgrade log severity when enabling params in LinuxVirtGpu

Alyssa Rosenzweig (2):

- pan/mdg: call nir_lower_is_helper_invocation
- nir/lower_helper_writes: fix stores after discard

Ashley Smith (1):

- panfrost: Reset syncobj after use to avoid kernel warnings

Bas Nieuwenhuizen (1):

- radv: Move support check out of winsys.

Dave Airlie (1):

- radv/video: don't try and send events on UVD devices.

David Rosca (4):

- gallium/vl: Fix video buffer supported format check
- Revert "frontends/vdpau: Alloc interlaced surface for interlaced pics"
- frontends/vdpau: Fix creating deinterlace filter for interleaved buffers
- gallium/vl: Return YUV plane order for single plane formats

Eric Engestrom (6):

- docs: add sha sum for 25.0.1
- .pick_status.json: Mark 61b0955308d720a6fa065e7a414d16999f7ffd03 as denominated
- .pick_status.json: Mark 534436f8635e63a30e4d7af4837dad35cfa361ad as denominated
- .pick_status.json: Update to 61feea6954a7526836ccbd30c657e6afc11fb4f5
- .pick_status.json: Mark 551770ccf8bdb1e5fa45ddac854535edf2b31a22 as denominated
- meson: announce that clover is deprecated (slated for removal)

Erik Faye-Lund (2):

- docs/features: add missing panvk feature
- panvk: correct VkPhysicalDeviceProperties::deviceName

Faith Ekstrand (9):

- util/box: Add a intersect_2d helper
- zink: Use pipe_box helpers for damage calculations
- zink: Set needs_barrier after transitioning to QUEUE_FAMILY_FOREIGN
- zink: Check queue families when binding image resources
- nvk: Allow rendering to linear images with unaligned strides
- nil: Relax alignment requirements for linear images
- vtn: Support cooperative matrices in OpConstantNull
- egl/x11: Re-order an if statement
- egl/kopper: Update the EGLSurface size after kopperSwapBuffers()

Ganesh Belgur Ramachandra (1):

- amd: use 128B compression for scanout images when drm.minor <63

Georg Lehmann (3):

- radv: enable invariant geom for DOOM(2016)
- aco/gfx11.5: remove vinterp ddx/ddy path
- aco/ra: disallow vcc definitions for pseudo scalar trans instrs

Ivan A. Melnikov (1):

- gallium/radeon: Make sure radeonsi PCI IDs are also included

Job Noorman (2):

- ir3: fix false dependencies of rpt instructions
- ir3: keep inputs at start block when creating empty preamble

John Anthony (1):

- panvk: Avoid division by zero for vkCmdCopyQueryPoolResults

José Roberto de Souza (1):

- intel/common: Retry GEM_CONTEXT_CREATE when PXP have not finished initialization

Karol Herbst (6):

- rusticl/program: implement CL_INVALID_PROGRAM_EXECUTABLE check in clGetProgramInfo
- rusticl/program: pass options by reference
- rusticl/program: loop over all devices inside Program::build
- rusticl/program: rework build_nirs so it only touches devices we care about
- rusticl/program: fix building kernels
- nir/serialize: fix decoding of is_return and is_uniform

Lionel Landwerlin (3):

- anv: fix non page aligned descriptor bindings on <Gfx12.0
- brw: fix spilling for Xe2+
- brw: ensure VUE header writes in HS/DS/GS stages

Lucas Stach (2):

- etnaviv: rs: fix slow/fast clear transitions
- etnaviv: fix ETNA_MESA_DEBUG=no_early_z

Marek Olšák (1):

- Revert "ac/nir: clamp vertex color outputs in the right place"

Mary Guillemard (2):

- pan/bi: Fix out of range access in bi_instr_replicates
- pan/bi: Ensure we select b0 with halfswizzle in va_lower_constants

Matt Turner (1):

- glsl: Add missing break

Maíra Canal (1):

- v3dv: don't overwrite the primary fd if it's already set

Mel Henning (1):

- nvk: Don't zero imported memory

Mike Blumenkrantz (1):

- zink: fix refcounting of zink_surface objects

Natalie Vock (2):

- radv/rt: Guard leaf encoding by leaf node count
- radv/rt: Flush L2 after writing internal node offset on GFX12

Patrick Lerda (2):

- r600: fix cayman main non-deterministic behavior problem
- r600: update the software fp64 support

Pierre-Eric Pelloux-Prayer (1):

- st/mesa: fix nir_load_per_vertex_input parameter

Rebecca Mckeever (1):

- panvk: Add STORAGE_IMAGE_BIT feature for formats supporting sampled images

Rhys Perry (1):

- aco: insert dependency waits in certain situations

Rob Clark (2):

- tc: Add missing tc_set_driver_thread()
- freedreno: Wait for imported syncobj fences to be available

Samuel Pitoiset (6):

- ac,radv: add a workaround for a hw bug with primitive restart on GFX10-GFX10.3
- radv: fix a GPU hang with inherited rendering and HiZ/HiS on GFX1201
- radv/amdgpu: fix device deduplication
- radv: update conformance version
- aco: do not apply OMOD/CLAMP for pseudo scalar trans instrs
- radv: emit a dummy PS state for noop FS on GFX12

Seán de Búrca (1):

- rusticl/mem: don't create svm_pointers slice from null raw pointer

Sviatoslav Peleshko (2):

- anv: Add full subgroups workaround for the shaders that use shared memory
- drirc: Apply assume_full_subgroups_with_shared_memory to Resident Evil 2

Timothy Arceri (1):

- util/u_idalloc: fix util_idalloc_sparse_alloc_range()

Yiwei Zhang (4):

- venus: fix a memory corruption in query records recycle
- lavapipe: set availability bit for accel struct host queries
- lavapipe: fix accel struct device query copy
- venus: fix to ignore dstSet for push descriptor

[Grogan]

* [new tag] mesa-25.1.0-rc1 -> mesa-25.1.0-rc1

I tried master (25.1.0-devel) a few nights ago (Friday?), and found myself with a strange problem. A reproducible crash, every time I looked down a zip line in Sniper Elite Resistance. Not riding the zip line (I wouldn't get the chance) just looking down in its direction. Only the zip lines, not the actual scenery near them. Indoor or outdoor. A fairly serious crash, that would freeze the game and within several seconds, it would crash me out of X and down to console. "The CS was canceled because the context was lost." That's a fault in the kernel driver (amdgpu) but it was indeed Mesa causing it, going back to 25.0.3 fixed it.

I'd better see if this is fixed or report it. Thing is, I don't know where it happened during 25.1.0-devel because I was using the stable until then. It would be needle-in-haystack to bisect.

[Grogan]

Nope... same problem. Look down a zip line, and the driver crashes. "amdgpu: the CS has canceled because the context was lost. This context is innocent". It will probably be because looking down a zip line starts to pre-render the scene in anticipation of sliding down it. If it's happening there reproducibly (any zip line), there's bound to be problems in other games so I just went back to Mesa 25.0.3 for now.

That's one of my newest games that I'm actively playing every night and the next Mesa is headed towards breaking it.

I'm having second thoughts about getting involved this time. The last time I chased an issue, multiple occlusion queries ordering incorrectly, I bisected it thinking it should be easy to fix, but I chased it for close to a year reverting the code, trying again when something looked promising or when asked if the issue was still happening, reporting and then finally, "sorry about that folks" and the guy finally reverted that change for the GPUs that had a problem with it, long after I built a new rig and didn't care anymore (moreover it wasn't even the GPUs themselves, but on older chipsets... I temporarily had the RX 570 card in here and could no longer reproduce it). I'm not sure I want to sink that much time into trying to bisect this etc.

[Grogan]

Actually I missed this (I didn't go back far enough in "issues"), it was already reported two weeks ago

https://gitlab.freedesktop.org/mesa/mesa/-/issues/12901

Someone knowledgeable in the thread bisected it to a commit that was merged 3 weeks ago. I'm glad

I may have some workarounds. It looks like I can probably revert the offending changes, but he also said "ACO_DEBUG=nosched" avoided the pagefaults. That is "disable pre-RA, ILP and VOPD instruction scheduling". Yep, sure I know what all that stuff is! Like the back of my hand I tell you

[Grogan]

I just tested a patch ("pull request") that seems to have fixed that zip line crash in Sniper Elite Resistance. I just made the code changes manually to aco_insert_exec_mask.cpp in 25.1.0-rc1 instead of switching to main, took out RADV_DEBUG=nosched, and loaded some the maps with zip lines that were crashing me and looked down them and rode them.

[Grogan]

Mesa 25.1.0 released today, with release notes:
https://docs.mesa3d.org/relnotes/25.1.0.html

This won't be a big change for me because I've been using the release candidates.

New features

cl_khr_spirv_linkonce_odr in rusticl
storagePushConstant16 on panvk
storageInputOutput16 on panvk
VK_KHR_depth_stencil_resolve on panvk
VK_KHR_separate_depth_stencil_layouts on panvk
VK_EXT_separate_stencil_usage on panvk
VK_KHR_sampler_ycbcr_conversion on panvk/v10+
VK_EXT_ycbcr_2plane_444_formats on panvk/v10+
VK_EXT_ycbcr_image_arrays on panvk/v10+
VK_KHR_imageless_framebuffer on panvk
VK_KHR_uniform_buffer_standard_layout on panvk
VK_EXT_border_color_swizzle on panvk
VK_MESA_image_alignment_control on NVK
shaderFloat16 on panvk
VK_KHR_shader_subgroup_uniform_control_flow on panvk/v10+
VK_KHR_shader_maximal_reconvergence on panvk/v10+
VK_EXT_device_memory_report on RADV
VK_KHR_shader_subgroup_extended_types on panvk/v10+
shaderStorageImageExtendedFormats on panvk
VK_KHR_display on panvk
VK_EXT_display_control on panvk
EXT_shader_framebuffer_image_fetch on v3d
EXT_shader_framebuffer_image_fetch_coherent on v3d
KHR_blend_equation_advanced on v3d
KHR_blend_equation_advanced_coherent on v3d
KHR_partial_update on etnaviv
VK_KHR_line_rasterization on panvk
VK_EXT_line_rasterization on panvk
shaderImageGatherExtended on panvk
textureCompressionBC on panvk
VK_EXT_sample_locations on RADV for GFX10+
GL_ARB_shader_clock on panvk
VK_KHR_shader_float_controls on panvk
VK_KHR_shader_float_controls2 on panvk/v10+
storageBuffer8BitAccess on panvk
storagePushConstant8 on panvk
uniformAndStorageBuffer8BitAccess on panvk
MSAA with 8 and 16 sample counts on panvk
VK_KHR_spirv_1_4 on panvk/v10+
Mali G720 and G725 on Panfrost and panvk
dualSrcBlend on panvk
VK_KHR_dynamic_rendering_local_read on panvk
VK_EXT_subgroup_size_control on panvk/v10+
VK_KHR_format_feature_flags2 on panvk
shaderStorageImageReadWithoutFormat on panvk
shaderStorageImageWriteWithoutFormat on panvk
VK_EXT_direct_mode_display on panvk
Vulkan 1.2 on panvk/v10+

Bug fixes

RADV:RX 9070:Mesa-25.0.5 GTA 5 Enhanced GPU HANG
[ANV/DG2] nvpro-samples/vk_raytracing_tutorial_KHR/ray_tracing_reflections renders not all reflections
radv: Flickering in Kingdom Come: Deliverance II
RADV regression causes severe glitches in Hunt Showdown 1896 on Polaris
Z-Fighting in Tomb Raider IV - VI Remastered Linux
[anv] VK_ERROR_DEVICE_LOST on Linux 6.13.8 while playing Dota 2 on Intel Graphics
Variable Rate Shading (VRS) produces very aliased results on RADV with an AMD gpu
WWE 2k23 small “artifacts”
RadeonSI - ACO does not spill enough SGPRs despite low VGPR pressure in some scenarios
RADV: Performance Regression (~18%) on Vega 64 (GFX9) in Cyberpunk 2077 caused by commit 08918f08805f (MR 34361)
radeonsi: CL conformance test `vector_swizzle` fails since 177427877bb50ad7ba24abfa13e55a2684d804df
[macOS] glapi_gentable.c:46:10: fatal error: ‘glapitable.h’ file not found
commit 90faadae regression - Warhammer 40,000: Space Marine 2 crashing
RADV: GPU hangs always at certain places in the game “Sniper Elite: Resistance”
Random mesa crashes in kwin_wayland on a 6600XT
Patch to fix clinfo on rusticl
radv/aco: Ghost of Tsushima hangs and causes gpu resets on RDNA 3 GPU
mesa-vulkan-driver-git.x86_64 causes strange colored rectangle artifacts in Final Fantasy XIV
NVK - Out of video memory error when starting Marvel Rivals
NVK - Out of video memory error when starting Marvel Rivals
RADV: Performance regression in Elden Ring on GFX8/Polaris
RADV: Performance regression in Elden Ring on GFX8/Polaris
Vulkan issues after sleeping on 9070 XT
ring gfx_0.0.0 timeout after waking from sleep - RX 9070
Confidential issue #12324
Confidential issue #12946
svga: how do I test this driver?
brw: new Xe2 CTS failures
[NVK] NAK assert in The Last of Us Part 2 shader
anv: S.T.A.L.K.E.R. 2: Heart of Chornobyl crashes after starting a new game
[ANV][LNL] - Lost Records: Bloom & Rage (1902960) - Title hangs on launch and subsequently crashes to desktop.
[BMG] Intel b580 battlemage: Fort Solis (Unreal Engine game) boots to menu, hangs while loading after hitting continue from the main menu
[ANV][LNL] - NINJA GAIDEN 2 Black (3287520) - Environment assets are incorrectly rendered or missing.
[ANV][LNL] - The Headliners (3059070) - Title hangs a few minutes after launch.
anv, regression: Invisibly blinking cliffs & rocks in Satisfactory DX12 on BMG
[Feature request] Add HK support for VK_EXT_queue_family_foreign vulkan extension
vk/overlay: output_file option failing
The Last of Us Part I GPU hang on gfx1201
glsl: IR validation fails on some shaders after lower_precision (breaks GLSL-to-NIR)
[bisected, LNL] brw: 341e5117ecbc (“brw/nir: Treat load_const as convergent”) regresses arb_gpu_shader5-interpolateAtOffset on LNL
vulkan regression mesa 24.3.4 to 25.0.0.rc3 with broadcom
AMD: broken Minetest trace on Vega iGPU
All OpenGL applications segfault
radv: nir_opt_varyings.c deduplicate_outputs: Assertion `list_index == 0’ failed.
vulkan/wsi: memory leak from wsi_CreateSwapchainKHR
[RADV][RDNA3][Phoenix3][APU] NARAKA: BLADEPOINT (1203220) gpu hang reproducible (ice/water regression mesa 24.1 bisected SAMPLE_MASK_TRACKER_WATERMARK=15) random (maybe other apps/games)
GPU hangs running Octopath Traveler II with 780M
GPU crash on Radeon 780M with Tales of Arise
The Last of Us - shadows flickering on gfx1201 without nohiz flag
brw: Hit unreachable nir_op_fsign case that brw_nir_lower_fsign missed
Panvk:Add Support for BCn (BC1–BC7) Texture Compression
anv: Dark pattern overlayed on objects in Eve Online DX11 mode on BMG
radv/aco: Strobing artifacts in Pacific Drive
aco: Distorted light halos, Star Citizen
Mesa 25 removes VA-API encoding for R9 390
debian-android build is huge
isl: CPCBCompressionEnable is now LosslessCompressionEnable
Video stuttering / anv: extend implicit fencing support
anv, bmg: Visual issues in AC Origins, Odyssey and Fenyx Rising when dxvk doesn’t export PointSize
[ANV][LNL] - A Game About Digging A Hole (3244220) - Title throws an assertion failure on launch.
[ANV][BMG] - A Plague Tale: Requiem - Severe shadow flicker
[ANV][LNL] - Hogwarts Legacy (990080) - Flickering artifacts visible on magical barrier.
brw: regression on Gfx9 dEQP-VK
HTML Docs fail to build from source with Sphinx 8.2.1
libglapi.dll ain’t built anymore under Windows
anv/video: Timestamps are exposed in video encode queue, but it crashes
Getting a crash with manually built llvmpipe (OpenGL)
[RadeonSI] Blender assetshelf icons are borken in mesa >= 25.0.0
radeonsi regression after 24.3.4
misc OpenGL CTS failures
glBindVertexBuffer regression due to ID reuse
RADV: logic used to avoid running on CDNA is faulty
[ANV][LNL] - Company of Heroes 3 (1677280) - Circular banding is present on screen during gameplay.
video hardware acceleration issue with “VA”
[LNL/BMG] Assassin’s Creed Valhalla trace replay hang
Vulkan conformanceVersion is reported as 0.0.0.0 in Mesa 25.0.0
X11 + Zink on NVK flickers older frames in Firefox based browsers
VRAM Abnormal use on mesa 25.0
[radv][regression] Multiple games detect the wrong amount of vram
Resident Evil 2 Remake flickers
OpConstantNull not supported for OpTypeCooperativeMatrixKHR
v3dv: vkcube-wayland crashes on raspberry pi 5 kernel 6.12 and latest mesa
GMSH Visualization Fails with radeonsi:can’t compile a main shader part, Fedora 41 AMD 7900xt
AMD VDPAU deinterlacing SIGSEGV
radv: vkd3d-proton test_primitive_restart_list_topology_stream_output randomly fails on NAVI2X
Mesa 24.1 introduced a Vulkan problem with DOOM 2016 on AMD 780M GPU
radeonsi: Firefox fails assertion requiring binding vertex elements before vertex_buffers
nouveau & zink+nvk: Flashing in Firefox and Thunderbird on Hyprland
Zink: Kopper’s present thread causes Wayland protocol races
GLmatrix needs aligned malloc
Lavapipe crashes if no Position is output in mesh shader
radeonsi/video: Allocate video buffers with modifiers 7f7206f1a9d brake ‘mplayer -vo vdpau…’ on GFX8 (Polaris 20)
[RADV/aco][regression][bisected] - Avowed (2457220) - GPU hangs near Watermill outside of Dawnshore
radv/sqtt: assertion “layout transition marker should be only emitted inside a barrier marker”
mesa_shader_cache directory is created even if mesa_shader_cache_db is used.
nak: Fold i2b(b2i(x))
[radv] Glitchy ground geometry regression in Total War Warhammer III on RX 7600
NVK: Implement shaderSharedInt64Atomics
nvk: Implement VK_MESA_image_alignment_control
radeonsi: regression with running DaVinci Resolve under rusticl since 666a6eb871d5dec79362bdc5d16f15915eb52f96
[ANV][LNL] - Black Myth: Wukong (2358720) - Corruption is visible near the edge of water.
[ANV][LNL] - Hogwarts Legacy (990080) - Pixelated corruption is visible when looking out at the water.
radv/video/h265: pps.flags.transform_skip_enabled_flag = 1 randomly hangs GPU
turnip falls with `assertion “errno == ETIME” failed`
[ANV][LNL] - Steel Rats (619700) - Game crashes after opening logos play before reaching main menu
nvk: Implement host-only descriptors
Gnome-shell Wayland fails to start with segfault at modifier-less driver
[ANV][LNL] - DYNASTY WARRIORS: ORIGINS (2384580) - Dithered transparency has vertical bands.
AMD Radeon R9 270 randomly causes video playback applications to crash with “amdgpu: The CS has been rejected”
ci: a306 jobs use a307 files
Rendering issues on GravityMark with RadeonSI ACO
Expose low latency encoding for radv vulkan video encode
i915: multiple tests assert with tgsi_ureg.h:893: ureg_swizzle: Assertion `reg.File != TGSI_FILE_NULL’ failed.
ci: debian-build-testing fails in nightly
shaders/closed/steam/deus-ex-mankind-divided/260.shader_test fails NIR validation
shaders/closed/steam/deus-ex-mankind-divided/260.shader_test fails NIR validation
panvk : vk_pipeline_cache_object_deserialize: Assertion `reader.current == reader.end && !reader.overrun’ failed.
46a8d5e7ef61735416d0c54886a7a9930621ae2c causes a permission denied spam
[BUILD] Build Failure: Implicit Function Declaration ‘timespec_sub_saturate’ (loader_wayland_helper.c)
anv: `MESA: warning: INTEL_HWCONFIG_MIN_GS_URB_ENTRIES (2) != devinfo->urb.min_entries[MESA_SHADER_GEOMETRY] (0)`
intel genX_acceleration_structure: missing dependency to bvh/header.spv.h
ci: what’s going on with zink-venus-lavapipe ?

[Grogan]

Mesa 25.1.1 today, which means distros should start putting it out

https://docs.mesa3d.org/relnotes/25.1.1.html

Bug fixes

radv: VK_EXT_primitive_topology_list_restart bug
[RADV][GFX9] Recent Mesa-git broken on AMD Vega 64 with ring sdma0 timeouts when launching DXVK games
ACO: IR Validation error “SDWA operand selection size” triggered by compute shader on VEGA20
RADV: Gibberish output with llama.cpp (Vulkan compute) on Radeon VII (Vega20) with Mesa 25.1.0, works on 25.0.5
Blending broken in game SkinDeep
Radeon R5 (Mullins) H264 VA-API encoding acceleration doesn’t work
nvk: lib_rs_gen.py requires `rustfmt`
radv: vkd3d-proton test failure with predication + EXT_dgc
mesa-25.0.4 aborts Xserver due to ACO “Unsupported opcode” v_lshlrev_b16
In SkinDeep, GL_LINES causes GL_INVALID_OPERATION with radeonsi and llvmpipe
radv: UB and artifacts when copying a `COMBINED_IMAGE_SAMPLER` with an immutable sampler
RADV: Dynamic state multiple viewport corruption
[drm:amdgpu_uvd_cs_pass2 [amdgpu]] *ERROR* )Handle 0x48780001 already in use!
glGetInternalformativ returns incorrect information for GL_STENCIL_INDEX8
RadeonSI: Psychonauts rendering regression since !29895
[r600g] Rejected CS when using dolphin’s GPU texture decoder
radeonsi: Assertion `src_bit_size == bit_size’ failed. when running without MESA_GLSL_DISABLE_IO_OPT=1
radeonsi vdpau + Packed YUY2 = assert
Indiana Jones and The Great Circle, Graphical corruption on 9070 XT.
glPushAttrib/glPopAttrib broken with glColorMaterial and ligthing

More detailed changelog below, at release notes link.


Funny nowadays that so many things aren't "stable" until they reach a .1 release. Semantics, really. Other projects keep their versioning below 1.0, like foo 0.5.9 being the latest stable.