SHA checksums

SHA256: 43d7abcd4aa8049d8fd75538344a374104765e81e17b4a6314cee2c0160e4412  mesa-25.2.2.tar.xz
SHA512: 69cd7ed33d0856e227a0219a9d09e5eb2fad28162276b5971bff4f7102cf58b0d914131ad6543562cba230ce6fb9b2076d3b156b1b8c375e74722a143aee5379  mesa-25.2.2.tar.xz

New features

• None

Bug fixes

• radv: watching vp9 encoded video with vulkan hwdecode result in artifacts
• ci: zink-venus-lavapipe errors
• [ANV][DG2][BMG] 3DMark Solar Bay Misrendering
• lavapipe defaults to memfd when multiple export types are requested
• anv: Simple vulkan compute shader causes Intel GPU hang due to excessive loop unrolling
• INTEL_DEBUG=spill_fs regression
• NIR validation failed after nir_lower_io in DOOM The Dark Ages
• ACO validation failed in DOOM: The Dark Ages - Undef operand to `p_parallelcopy`
• ACO validation failure in DOOM: The Dark Ages Shader
• Dragon Age Veilguard / Ability Wheel Targeting Visual Distortion
• [AMD] glTexturePageCommitmentEXT triggers an error if level is higher than 0
• missing sparse synchronization in zink
• System Crashes when starting VR on rx 9070 (xt)

Changes

Aleksi Sapon (1):

• draw: fix missing line viewport transformation

Alyssa Ross (1):

• gfxstream: guest: don't use transitional LFS64 API

Antonio Ospite (2):

• nak/nouveau: silence errors about never used methods
• compiler/rust: fix errors about hiding elided lifetime

Ashley Smith (1):

• mesa: Fix support for GL_EXT_shader_clock

Boris Brezillon (1):

• panvk: Fix disjoint image memory binding

Caio Oliveira (2):

• brw: Fix folding case for MAD instruction with all immediates
• brw: Fix checking sources of wrong instruction in opt_address_reg_load

Connor Abbott (2):

• vulkan/queue: Fix VkTimelineSemaphoreSubmitInfo sanitization
• tu: Fix CmdBindTransformFeedbackBuffersEXT size handling

Daniel Schürmann (1):

• aco/ra: don't clear lateKill operands in get_reg_create_vector()

David Rosca (7):

• radeonsi/vcn: Disable H264 encode 8x8 transform when CABAC is disabled
• radv/video: Disable H264 encode 8x8 transform when CABAC is disabled
• radeonsi/vcn: Disable H264/5 constrained intra pred with rate control
• radeonsi/vcn: Fix compatibility with old FW for encode
• radeonsi/vcn: Fix HEVC encode cu_qp_delta with old FW
• radeonsi/vcn: Fix HEVC encode transform_skip with old FW
• radv/video: Fix VP9 loop filter and segmentation params

Eric Engestrom (4):

• docs: add sha sum for 25.2.1
• .pick_status.json: Update to c0e51bcf24c7905a2fdc272194744a5fdfe8c345
• .pick_status.json: Mark 85310e912c5a2fc542618a11b9e38ea9c433ed18 as denominated
• .pick_status.json: Mark a34756bbedacdf2558c966466a71bfd3715fa295 as denominated

Eric R. Smith (2):

• mesa: fix off by one in MSRTT handling
• panvk: revised occlusion query pointer fix

Erik Faye-Lund (1):

• pan/lib: clamp format size to 4

Faith Ekstrand (10):

• subprojects: Stop calling add_languages() in paste-1-rs/meson.build
• nak: Add a nak_qmd_size_B() query
• nak/hw_runner: Allow for variable sized QMDs
• nvk: Allow for larger QMDs
• nak/qmd: QMD versions 4.0 and 5.0 are both 384B
• nak: NAK_MAX_QMD_SIZE_B should be 384
• compiler/rust: Fix the DFS loop detection algorithm
• lavapipe: Always use dma-buf for external memory when we can
• nir: Add an option to make lower_phis_to_regs_block() less clever
• nak,nir: Use a simpler version of phis_to_regs_block in lower_cf

Georg Lehmann (6):

• nir/uub: fix exclusive scans
• aco: fix ra validation for flat/global/scratch/ds load sbyte_d16
• aco/optimizer: don't apply packed clamp to v_fma_mix
• aco/optimizer: don't create undef copies from p_create_vector
• ac/nir: do not assume mesh cull flag is 1bit
• nir/lower_io: fix boolean output stores

Hans-Kristian Arntzen (2):

• nvk: Return 0 for opaque memory capture replay.
• nvk: Avoid passing garbage data in descriptor buffers for UBOs.

Iván Briano (1):

• anv: fix capture/replay of sparse images with descriptor buffer

Jarrett Johnson (1):

• nir: tag cls variable as maybe unused

Jesse Natalie (1):

• nir: Add missing #include for c99_alloca.h

Jianxun Zhang (2):

• anv: No compression on host memory allocation (xe2)
• anv: Fix PAT entry in importing (xe2)

Job Noorman (4):

• ir3: don't vectorize nir_op_sdot_4x8_iadd[_sat]
• ir3: emit descriptor prefetch in block dominated by its sources
• ir3/cf: don't swap signedness of (sat) instructions
• ir3: use nir_lower_bit_size for 8-bit bit_count

Josh Simmons (1):

• util: Fix `BITSET_EXTRACT` out-of-bounds read

Julia Zhang (1):

• pps: init driver in OnSetup

Karol Herbst (2):

• aux/trace: move fence_server calls outside the locked area
• rusticl/event: fix create_and_queue for deps in error states

Lionel Landwerlin (9):

• brw: fix analysis dirtying with pulled constants
• anv: fix uninitialized return value
• brw: remove uniform from opt_offsets
• brw: fix INTEL_DEBUG=spill_fs
• brw: fix broadcast opcode
• anv: temporary disable KHR_maintenance8
• Revert "anv: enable non uniform texture offset lowering"
• Revert "brw: move texture offset packing to NIR"
• anv: fix pipeline barriers with pre-rasterization stages

Mark Collins (1):

• freedreno/drm: Only initialize memory data source when Perfetto is active

Mary Guillemard (2):

• nouveau/headers: Fix nv_push rust push_inline_data implementation
• hk: Return 0 for opaque memory capture replay

Mike Blumenkrantz (14):

• zink: break out unflushed batch waiting into separate function/mechanism
• zink: pass ctx to sparse bind functions
• zink: when sparse unbinding, always wait on main timeline semaphore
• zink: trigger fb unbind barrier on resolve images too
• zink: fix sizing on resolve resource array
• zink: update resized swapchain depth buffer layout while blitting
• zink: unify/fix clear flushing
• zink: also set msrtss stencil
• zink: always flush clears when doing single-aspect blit to avoid data loss
• zink: fix some weird indentation in update_binds_for_samplerviews()
• zink: flag resources for layout eval in update_binds_for_samplerviews()
• zink: zero db offset on batch reset
• zink: don't increase db scale when resizing a db up to the current scale
• kopper: unwrap screen before checking cpu flag

Nataraj Deshpande (1):

• anv: add feature flags for linearly tiled ASTC images

Philipp Zabel (1):

• rusticl: Fix hidden lifetime warnings

Pierre-Eric Pelloux-Prayer (2):

• mesa/st: always use base_serialized_nir for draw
• nir/opt_varyings: fix build with PRINT_RELOCATE_SLOT

Pohsiang (John) Hsu (1):

• gallium/pipebuffer: fix multithread issue on pb_slab_manager_create_buffer

Qiang Yu (1):

• mesa: fix glTexPageCommitmentARB and glTexturePageCommitmentEXT level check

Rob Clark (1):

• drirc: Work around ANGLE brokeness

Robert Mader (2):

• gallium: Set and count all extra samplers
• nir: Fixup 10/12 bit SW decoder YCbCr formats

Sagar Ghuge (4):

• anv: Add missing ACCELERATION_STRUCTURE_READ in barrier handling
• anv: Enable CS stall for ACCELERATION_STRUCTURE_COPY stage
• anv: Add missing L3 flushes
• anv: Apply pipe flushes for outstanding PC bits

Samuel Pitoiset (4):

• radv: dirty some states from graphics pipeline earlier
• radv: add missing L2 invalidate cache flush for non-coherent images
• radv: fix hashing graphics pipeline when no stages are compiled
• radv/rt: fix a potential issue with RADV_PERFTEST=dmashaders

Tapani Pälli (1):

• anv: change some image qualifiers as coherent for Last Of Us

Trigger Huang (1):

• virtio/vdrm: add ENABLE_DRM_AMDGPU for c_args

Valentine Burley (1):

• ci/crosvm: Retry all curl errors when downloading kernel

Yiwei Zhang (4):

• panvk: ensure wsi memory is bound at offset 0
• vulkan/android: amend a missing case for IMPLEMENTATION_DEFINED AHB
• vulkan: handle wsi private data properly
• anv: fix broken utrace