Mesa 26.3.0 Release Notes / 2026-11-09

Mesa 26.3.0 is a new development release. People who are concerned with stability and reliability should stick with a previous release or wait for Mesa 26.3.1.

Mesa 26.3.0 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 26.3.0 implements the OpenCL 3.1 API, but the version reported by the CL_DEVICE_VERSION, CL_DEVICE_NUMERIC_VERSION and CL_DEVICE_OPENCL_C_ALL_VERSIONS clGetDeviceInfo queries depends on the particular driver being used.

Mesa 26.3.0 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.

SHA checksums

SHA256: 91d729aa68bb89e4ec857841300bbe922bcb3e52b1f97566ca83930d3b28343a  mesa-26.3.0.tar.xz
SHA512: b4abd266ecaf813ef078a0461bcdf7c03d2ecddeb5b6e1c51054fb0afe0d830c0f27fc78f0effe68df5ea7b48f76f4e6ca36ed9afe98513f0e4bf8e0bbbe96dd  mesa-26.3.0.tar.xz

New features

  • VK_AMD_anti_lag on RADV

  • VK_IMG_filter_linear_2d on pvr

  • VK_KHR_maintenance7 on pvr

  • Vulkan 1.3 on panvk/v6-7

  • VK_KHR_maintenance4 on panvk/v6-7

  • VK_KHR_maintenance5 on panvk/v6-7

  • VK_KHR_maintenance7 on panvk/v6-7

  • VK_KHR_maintenance8 on panvk/v6-7

  • VK_KHR_shader_float_controls2 on panvk/v6-7

  • VK_KHR_shader_subgroup_extended_types on panvk/v6-7

  • VK_EXT_subgroup_size_control on panvk/v6-7

  • VK_ARM_shader_core_builtins on panvk/v6-7

  • VK_ARM_shader_core_properties on panvk/v6-7

  • VK_ARM_scheduling_controls on panvk/v6-7

  • shaderCoreBuiltins on panvk/v6-7

  • schedulingControls on panvk/v6-7

  • VK_KHR_synchronization2 on pvr

  • VK_KHR_copy_memory_indirect on panvk

  • VK_KHR_maintenance6 on panvk/v6-7

  • VK_KHR_spirv_1_4 on panvk/v6-7

  • VK_EXT_device_memory_report on v3dv

  • shaderUniformBufferArrayDynamicIndexing on pvr

  • shaderStorageBufferArrayDynamicIndexing on pvr

  • shaderSampledImageArrayDynamicIndexing on pvr

  • shaderStorageImageArrayDynamicIndexing on pvr

  • shaderInputAttachmentArrayDynamicIndexing on pvr

  • shaderUniformTexelBufferArrayDynamicIndexing on pvr

  • shaderStorageTexelBufferArrayDynamicIndexing on pvr

  • VK_KHR_video_encode_feedback2 on RADV

  • VK_EXT_global_priority on pvr

  • VK_EXT_global_priority_query on pvr

  • VK_KHR_global_priority on pvr

  • VK_KHR_cooperative_matrix on panvk

  • Support for Mali G68 GPU on Panfrost

  • cl_ext_float_atomics on iris, llvmpipe and zink

  • cl_khr_semaphore on iris

  • cl_khr_external_semaphore on iris

  • cl_khr_external_semaphore_sync_fd on iris

  • VK_EXT_inline_uniform_block on pvr

  • VK_KHR_shader_fma on pvr

  • fragmentStoresAndAtomics on pvr

  • VK_EXT_cooperative_matrix_maintenance1 on radv, lavapipe

  • VK_KHR_device_address_commands on nvk

  • Support for G715 and G615 GPUs on Panfrost and PanVK

  • GL_ARB_texture_multisample on etnaviv/HALTI5

  • VK_KHR_shader_constant_data on panvk

  • Support for BXM-4-64 (36.52.104.182) on pvr

  • VK_KHR_shader_float_controls2 on pvr

  • VK_KHR_performance_query on GFX12

  • GLSL 130 on etnaviv/HALTI2

  • GL_EXT_texture_integer on etnaviv/HALTI2

  • GL_ARB_shader_bit_encoding on etnaviv/HALTI2

  • GL_ARB_texture_rgb10_a2ui on etnaviv/HALTI2

  • VK_EXT_primitive_topology_list_restart on pvr

  • VK_KHR_incremental_present on panvk

  • VK_EXT_swapchain_colorspace on panvk

  • VK_KHR_push_descriptor on pvr

  • VK_EXT_image_sliced_view_of_3d on pvr

  • VK_EXT_image_compression_control on panvk

  • VK_KHR_shader_untyped_pointers on HK

  • VK_KHR_maintenance10 on HK

  • VK_KHR_shader_fma on HK

  • VK_EXT_primitive_restart_index on HK

  • GL_EXT_color_buffer_float on etnaviv/HALTI5

  • GL_EXT_conservative_depth on etnaviv/HALTI5

  • GL_EXT_draw_buffers_indexed on etnaviv/HALTI5

  • GL_EXT_EGL_image_storage on etnaviv/HALTI5

  • GL_EXT_EGL_image_storage_compression on etnaviv/HALTI5

  • GL_EXT_shader_integer_mix on etnaviv/HALTI5

  • GL_EXT_sRGB_write_control on etnaviv/HALTI5

  • GL_EXT_texture_compression_rgtc on etnaviv/HALTI5

  • GL_EXT_texture_sRGB_decode on etnaviv/HALTI5

  • GL_EXT_texture_sRGB_R8 on etnaviv/HALTI5

  • GL_EXT_texture_sRGB_RG8 on etnaviv/HALTI5

  • GL_EXT_texture_storage_compression on etnaviv/HALTI5

  • GL_MESA_framebuffer_flip_y on etnaviv/HALTI5

  • GL_MESA_shader_integer_functions on etnaviv/HALTI5

  • GL_OES_draw_buffers_indexed on etnaviv/HALTI5

  • GL_OES_EGL_image_external_essl3 on etnaviv/HALTI5

  • GLES 3.0 on etnaviv/HALTI5

  • VK_QCOM_rotated_copy_commands on Turnip

  • VK_KHR_maintenance10 on panvk

  • VK_KHR_map_memory2 on v3dv

  • VK_KHR_unified_image_layouts on v3dv

  • VK_INTEL_shader_integer_functions2 on panvk

  • VK_KHR_internally_synchronized_queues on NVK

  • VK_KHR_maintenance11 on NVK

  • VK_EXT_primitive_restart_index on NVK

  • occlusionQueryPrecise on pvr

  • EGL_NV_context_priority_realtime on radeonsi

  • VK_KHR_maintenance11 on panvk

  • VK_EXT_transform_feedback on pvr

Bug fixes

  • “TY the Tasmanian Tiger 3” always crashes in the same cutscene on amd gpus

  • # va: H.264 encode spins forever in vaRenderPicture on some packed SPS headers (vl_rbsp misreads emulation prevention bytes)

  • 26.2.x build fails with armhf

  • ACO: s_bitset0 retargeted to exec drops the VCC condition (26.2 regression from 984aa6e0855, RADV, RDNA2 to RDNA4, KytyPS5)

  • ANV: Assert on vkGetPhysicalDeviceImageFormatProperties2

  • Ambient occlusion is broken in The Chronicles of Riddick - Assault on Dark Athena

  • Asahi OpenCL compiler packing assertion failed

  • Blockland crashes when shader quality turned up

  • Confidential issue #15052

  • Confidential issue #15670

  • Confidential issue #16496

  • Confidential issue #16630

  • D3D12: last vertex’s non-zero-offset attribute fetched as 0 with a tightly-packed interleaved vertex buffer (AMD D3D12 only, regression since 24.1)

  • Fog in OpenGL 1.x behaves oddly in Mesa

  • Follow-up from “isl: disable MCS compression on R9G9B9E5”

  • GL stencil reference query does not update immediately

  • GPULostDevice on 6800 XT (mesa-26.0.X)

  • Intel Meteor Lake: Unigine Heaven and Valley trigger GPU hangs with Mesa 26.2.0–26.2.4; work with 26.1.8

  • LLVM 23, riscv64, compile error gallium/auxiliary/gallivm/lp_bld_init_orc.cpp (API change?)

  • Last Of Us Part 2 Remastered has glitches on hair strands - since Mesa 26.0.1

  • Memory bloat via glTexSubImage2D

  • Mesa 26.2.1 Ground Truth Ambient Occlusion Broken on No Man’s Sky

  • Mesa 26.2.1: radeonsi no longer loads for GCN1 / Pitcairn when using radeon driver.

  • Mesh shader uniform/constant rendering issue

  • NIR goto_if pass generates broken code in some cases

  • NIR: loop unrolling should not create phis for constants defined in the loop

  • NVK memory leak in CmdCopyBufferToImage in 26.2.0, 26.2.1 and 26.2.2.

  • RADV NIR optimization bug

  • RADV regression in Mesa 26.2.x – No Man’s Sky has broken/dark sunlight on RX 7900 XT

  • RADV/RX 9070 XT: WUCHANG: Fallen Feathers gpu ring reset

  • RADV: GPU page fault (null pointer) with RT emulation on Polaris (RX 580 2048SP). Issue between 26.1.6 and 26.2.1

  • RADV: Missing optimizations for typical uaddCarry patterns

  • RadeonSI VA-API: vertical color discontinuity / split on YouTube (AV1/VP9) with RX 7900 XTX

  • Radv/RadonSI: si_set_rasterized_prim config error

  • Ray tracing tests started failing after following commit

  • Recording on AMD GPU using gpu-screen-recorder with VAAPI H.264 backend produces corrupted output since Mesa 26.0.0

  • Regression. VTK polydata. gl_PrimitiveID requires explicit geometry shader with versions 25.3.6 and later.

  • Regression: D3D12 driver and DRI broken under WSL in 25.0.4

  • Regression: HEVC hardware decoding (VA-API) fails with “invalid UVD command 5!” on Stoney Ridge (amdgpu) in Mesa 26+

  • SIGILL with llvmpipe on ARMv9(?) without SVE

  • Spiderman remasterd crashes/fails with ray tracing

  • Spike in mmap count for vk_cmd_queue regressed dEQP-VK.api.object_management.max_concurrent#command_buffer_*

  • Textures got swapped/glitched on Wuthering Waves game.

  • Unable to compile from master since update of rust-bindgen 0.73 (Arch Linux)

  • V3D driver (Mesa) does not expose GL_EXT_shader_texture_lod, breaking raspberry Pi5 WebGL in Firefox (e.g. Google Maps with Street View)

  • VK_VALVE_video_encode_rgb_conversion always outputs limited color range

  • Witcher 3 Remastered - Graphical/visual bug that act like an overlay

  • Witchfire - black artifacts

  • X-Plane 12 (2014780) Flickering artifacts on textures with Intel Arc B580

  • Zink-on-radv: Regression in the unvanquished trace on Vangogh

  • [ANV] Flickering black blocks in Project Wingman

  • [ANV][ARC] Space engineers 2 Artifacts

  • [ANV][ARL][bisected] Visual graphics glitches in Counter-Strike 2 starting in Mesa 26.2.x

  • [ANV][BMG] llama.cpp Flaky unit test results with latest Mesa on Arc B70 (Regression)

  • [ANV][BMG]: A few dEQP-VK.texture.mipmap.cubemap.image_view_min_lod.* fails

  • [ANV][PTL] llama.cpp Vulkan token generation slower on Arc B390 (Q3_K/Q6_K/Q8_0) (Regression)

  • [ANV][PTL][jay] Counter-Strike 2 triggers an assertion error when launched in a debug driver.

  • [DG2] ANV: Enshrouded (1203620) - character invisible, transparent billboards rendered as opaque spheres on discrete Arc GPU

  • [Forts RTS] Corrupt rendering of compressed textures

  • [GR Breakpoint] Inverted frustum culling for grass meshes on Linux

  • [NVK/Kepler] NAK compiler panic in assign_regs.rs: “Failed to find free register” when launching wgpu application (Zed)

  • [Navi10][RadeonSI/RADV] NGG culling causes nondeterministic GPU hangs; nonggc fixes both OpenGL and Vulkan

  • [RADV] GFX8: Black screen in UE5 DX12 (Half Sword) after commit 15990c78b8e (“allow DCC for concurrent images without DCC stores”)

  • [RADV] Insurgency Sandstorm - black Z-fighting geometry

  • [RADV][regression] PiP scope in the game OPERATOR causes foliage to use PiP as texture

  • [Regression] [RADV] Control Ultimate Edition – RDNA3

  • [Turnip/A650] Hangs in several several games due to some missing shader(?) workaround

  • [VAOn12] Thread safety issues

  • [amdgpu] Little Inferno rendering issues starting with Mesa 21.1

  • [anv][jay][bmg] Monster Hunter World (582010): Texture flickering with INTEL_JAY=all

  • [anv][ptl] GPU hang / xe engine reset in Warframe (Mesa 26.3.0-devel)

  • [anv][regression] Genshin Impact broken after !43158 with descriptor heap path

  • [freedreno] GTK 4 ngl renderer hangs on FD306A

  • [meson] rust crate downloading for subbprojects blocked

  • [r300/r600][EGL]: The r300/r600 driver is failing to load the gbm platform properly on certain older iMac systems

  • [radeonsi] Minecraft Voxy mod: Distant water rendering issue on Vega iGPU and RDNA2

  • [radeonsi] eglinfo ‘libEGL warning: failed to get driver name for fd -1’

  • [radeonsi][RDNA4] Blade strangers shows a black screen

  • [radv] Anisotropic filtering glitch in Wolfenstein II: The New Colossus

  • [radv] Regression causes GPU timeouts in Final Fantasy XVI on RDNA1

  • [radv][bisected][regression] Broken rendering in Persona 3 Reload menu on RDNA2 in gamescope

  • `hevc_vaapi` on AMD (radeonsi/gfx1150) stalls to ~1 fps when input frames carry A/53 (CEA-608/708) closed-caption side data

  • a530: broken arb_uniform_buffer_object struct-array-array-array-struct and array-struct-array-struct

  • ac/radv/radeonsi: PS restricted to 2 of 6 CUs on Rembrandt 660M with fully harvested SA0 (~50% graphics perf loss)

  • aco: illegal VOP3 v_cndmask_b32 with two SGPR operands on GFX6-9 in emit_scaled_op (fp32 denorm preserve) → wrong results

  • anv/jay: loop-invariant and loop-carried 16-bit values copied out and back every loop iteration

  • anv: 860d2493214 (“intel/genxml: more sharing between host/device gen code”) broke dgc tests

  • anv: DG2 functional failures with CCS_E and aliasing

  • anv: Test fails in simulation due to coarse pixel shading state

  • anv: Tune buffer compression enabling for DXVK

  • anv: VK_EXT_calibrated_timestamps returns a GPU timestamp ahead of CPU time on Arc B390 (Panther Lake), crashing Unreal Engine 5.8

  • anv: big lock contention for shader allocation

  • anv: bottom-of-pipe timestamp latched before vkCmdDispatch completes (Arc B390 / Panther Lake, Mesa 26.1.4), breaking wgpu compute-pass timings

  • anv: cooperative matrix loads from shared memory return the wrong tile on Arc B390 (Xe3/PTL), matmul results are wrong

  • anv: dEQP-VK.info.device_mandatory_features fails due to partial exposure of VK_NV_cooperative_matrix2

  • anv: revisit uncached setting for indirect data on ARL/Xe2+

  • asahi: 16-bit blending into RGBA32F produces zero due to 32-bit tilebuffer load

  • asahi: 64-bit multiply of a sign-extended sub-32-bit value produces wrong result

  • asahi: Mario Kart WiiCompiled graphical glitches

  • asahi: Some textures do not render correctly in Portals 2

  • assert in radeonsi doing glReadPixels with invalid draw framebuffer

  • brw accumulator crashes Android RenderEngine

  • brw/decoder: fragment shader decoding issue

  • brw: segfault in brw_opt_register_coalesce

  • ci/cl-cts: Basic: work_item_functions_out_of_range failure

  • ci_run_n_monitor: add a way to select a stress duration time

  • d3d12 (WSL2, NVIDIA RTX 5090): self-deadlock in d3d12_bo_new(), reclaim under pb_slab manager mutex hangs RPCS3 on compressed texture upload

  • dEQP-VK.image.swapchain_mutable.wayland.2d.r8g8b8a8_srgb_r8g8b8a8_unorm_draw_texture_format_list crash

  • freedreno/a8xx: KHR-{GL46,GLES32}.core.tessellation_shader.tessellation_shader_tc_barriers.barrier_guarded_read_calls

  • freedreno: garbled text rendered by pdf.js on snapdragon x1 elite

  • gbm: support YUV and non-renderable allocations

  • i915 GPU hang on Alder Lake-HX GT1 (8086:4688) with Mesa 26.2.0 OpenGL — works on 26.1.6

  • intel/compiler: regression from fc8b1928cbc21099d1428299bf7d0c86da9f7b65

  • iris: Blender wireframe rendering broken

  • kopper: non-dmabuf sw texture update is broken

  • kraid/v14+: Support .zext in disassembler

  • kraid: Fragment Shaders

  • kraid: Fragment var/tex fusion

  • kraid: Hook up OpICmpMulti and implement 64-bit comparisons

  • kraid: Implement branch threading to get rid of empty blocks

  • kraid: Optimize texture projectors on v11+

  • kraid: Optimize to TEX_DUAL

  • kraid: Plumb the Kraid disassembler through to the C tools

  • kraid: Tag last uses on sources

  • kraid: Use IADD for swaps instead of XOR

  • kraid: Wire up texture .skip modifier

  • lavapipe: GALLIUM_TRACE broken

  • lavapipe: maintenance8 depth/stencil <-> color vkCmdCopyImage copies only sample 0 of multisampled D32_SFLOAT_S8_UINT and D24_UNORM_S8_UINT images

  • lavapipe: new RT fails

  • libvulkan_intel.so crash with X-Plane 12.4.3 on Intel Arc B580

  • llvmpipe/lavapipe: freed device memory is never released from the memfd

  • nir_opt_copy_prop_vars validation failure with Rusticl OpenCL kernel shaders

  • nircache causes glxgears/eglgears to segfault under Zink

  • nvk: new mesa is causing GDM to crash

  • panfrost: Optimize CubeMaps on v11+

  • panvk broke unrelated NixOS desktop

  • panvk: Need to mention PVR_I_WANT_A_BROKEN_VULKAN_DRIVER=1 to the environment to enable PanVK for GPUs that aren’t conformant

  • program@execute@builtin@builtin-*-mad_sat-1.0.generated fail with rusticl on radeonsi

  • r300: constant inlining might be broken on big endian

  • r300: rc_pair_regalloc optimization is causing fail of do_while_dynamic_iterations.mixed_break_continue_fragment

  • r300: use lower_to_scalar (and add fix for regressions) or create optimization that can replicate improvemnent

  • r600 , bisected , sfn : Assertion failed on Minecraft with R700

  • radeonsi/VCN: AV1/HEVC encode reports a coded size larger than the coded buffer, client segfaults reading the mapped buffer

  • radeonsi/VCN: HEVC VAAPI encode corrupts P-frames on Van Gogh (gfx1033) since Mesa 26.0 — I-frames and H.264 unaffected

  • radeonsi/va: vaGetImage to I420/YUYV fails with UNSUPPORTED_RT_FORMAT since Mesa 26.2 on GPUs with VPE (Radeon 890M)

  • radeonsi: GPU page fault (SQC) / context loss when using Iris shaderpacks on RX 9070 XT (gfx1201)

  • radeonsi: black screen in picom after ae36b9e3459766a953fe6318545463f9a773d119

  • radeonsi: incorrect rendering in Doom 3: BFG

  • radeonsi: null-deref in validate_io() during glUseProgramStages on separable, SPIR-V-ingested program pipeline

  • radv (GFX12): WSI swapchain memory for explicit-LINEAR-modifier images is allocated with AMDGPU_GEM_CREATE_GFX12_DCC, so the exported dmabuf’s contents are not what the modifier declares

  • radv ignores count of 0 in vkCmdDrawIndirect

  • radv/amdgpu: padding VM page (RADEON_FLAG_VM_PAD_1PAGE) is never unmapped on BO destroy; VA reuse then fails with “va … conflict” on GFX9

  • radv/ci: RAPHAEL is using very old MEC firmwares

  • radv: NULL deref creating a VK_QUERY_TYPE_RESULT_STATUS_ONLY_KHR query pool with no video profile

  • radv: NULL deref in radv_get_vgt_shader_key after binding a compute shader object

  • radv: PCIe 4.0 x4 fails the “PCIe 3.0 x8” bandwidth check for VRAM command buffers by 4 MB/s (rounded per-lane constants); up to 13% decode loss in llama.cpp

  • radv: PixelInterlockOrderedEXT critical sections of consecutive draws in one subpass are not ordered (Radeon 780M, GFX11)

  • radv: UNREACHABLE clearing DEPTH|STENCIL when the dynamic render pass has no stencil attachment

  • radv: VkDescriptorUpdateTemplate takes a pointer to the set_layout memory

  • radv: acceleration structure update (refit) of AABB geometry loses intersection candidates (NAVI32, Mesa 26.1.4)

  • radv: angle/angle_trace_tests/pubg_mobile_battle_royale rendering regression

  • radv: descriptor heaps do not support non-uniform indexing on buffer pointers

  • radv: failed vdrm_device_connect on a virtio-gpu node aborts all device enumeration, hiding a working amdgpu GPU

  • radv: sdma/transfer queue copies to DCC-compressed image sometimes corrupt the image

  • radv: vkCopyMemoryToImage segmentation fault with block-compressed formats when memoryRowLength is non-zero

  • radv: vk_device_finish assert when two queue create infos share a family index but differ in flags

  • rusticl + v3d assertion failure

  • rusticl: could not initialize SVM support with radeonsi,zink

  • rusticl: freedreno: Darktable very slow processing (due to stalls?)

  • segfault v3d raspberry Pi5 h265 hw decoding / regression mesa 26.1.3/26.1.4

  • tu/anv: pDependencyInfos dereferenced as a single pointer instead of an array in CmdWaitEvents2

  • tu: assertions in tu_formats.cc: Assertion `fmt.fmt != FMT6_NONE’ failed

  • turnip: dEQP-VK.sparse_resources.* crash with Venus

  • v3dv: gpu hang when an inline uniform block comes before a sampled image

  • vulkan,radv: when using descriptor heaps, input attachments in SPIR-V are incorrectly detected as read only instead of read write

  • vulkan/runtime: GetPipelineBinaryDataKHR incorrectly assumes `pPipelineBinaryDataSize` must be 0 initialized

  • zink on radv: spec@arb_timer_query@timestamp-get regression after switching to gfx-ci/linux

  • zink/adl: flakiness on dEQP-GLES3.functional.occlusion_query.scissor_depth_write_stencil_write_stencil_clear

  • zink/ci: switch all traces jobs to surfaceless+gbm

  • zink: draws old vertex data after buffer update with threading

  • zink: emulated multisampled render to texture breaks the next draw

  • zink: wrong pipeline after a vertex-layout-only change without VK_EXT_vertex_input_dynamic_state

Changes

3mro777o (1):

  • nir/opt_varyings: use nir_src_rewrite when shrinking varying arrays

Abir Deol (1):

  • gallivm: Don’t assume invocation 0 is active in callees

Adam Stylinski (1):

  • nv30: fix an issue when this push buffer is NULL

AeroModes (1):

  • meson: fix GNU qsort_r detection for GCC

Ahmed Hesham (2):

  • pan/rusticl: Update the last OpenCL conformance version passed value

  • pan/rusticl: Add CL latest conformance version to features.txt

Aitor Camacho (29):

  • kk: workaround MSL compiler issue with volatile coherent qualifiers

  • kk: remove deprecated storeActionOptions in macOS27

  • kk: call opt_licm to lighten loop executions

  • wsi/metal: Make CAMetalLayer’s residencySet resident for presenting queue

  • wsi/metal: always force 3 drawables for smoother presentation

  • kk: Wait on drawable availability before rendering to it

  • kk: Enable triangle merge for multisample rendering to prevent CTS failures

  • kk: release drawable after presentation signal

  • kk: Fix icd json api version

  • wsi/metal,kk: Work around Metal4 presentation crash

  • kk: Improve workaround 14

  • kk: Fix simultaneous use command buffers

  • wsi/metal: Correctly report minImageCount instead of forcing 3 drawables always

  • kk: Add Foundation framework to bridge

  • kk: Fixes workaround 14 in macOS26 with a new one workaround 17

  • kk: Fix host imported buffer offsets

  • kk: Move kk_buffer Metal members into its own struct

  • kk: Don’t split compute encoders when possible to add a marker

  • kk: rework how we disable triangle merge when needed

  • kk: narrow forced depth write application

  • kk: use fb fetch and avoid breaking render passes in barriers when possible

  • kk: Workaround for intermittent memory_model failures in macOS 26 with M3+

  • kk: add base to shared data access

  • kk: move constant data out of scratch

  • kk: One MTL4CommandBuffer per VkCommandBuffer

  • wsi/metal: set extended range for Display P3 with float formats

  • kk: Update conformance version to 1.4.6.2

  • kk: Adopt new input attachment intrinsics

  • kk: Fix miptail computation due to behaviour difference in OS versions

Aksel Hjerpbakk (2):

  • panvk: encode cs NOP payload and type

  • panvk/decode: implement nop strings

Alejandro Piñeiro (7):

  • docs/features: v3d and panfrost supports GL_EXT_color_buffer_half_float and GL_EXT_texture_norm16

  • vtn_bindgen2: avoid unused result warning on fread call

  • mesa_clc: avoid unused result warning on fread call

  • drm-shim: avoid unused result warning on write call

  • llvmpipe: avoid unused result warning on asprintf call

  • pan/kmod: avoid unused result warning on read call

  • panfrost: avoid unused result warning on asprintf call

Aleksi Sapon (1):

  • wsi/metal: use correct macOS 26 version macro

Alexander Kanavin (1):

  • src/util/u_math.c: do not use arm fpu instructions if fpu is not available

Allen Ballway (3):

  • intel/gen: Use global namespace size_t

  • anv: Fix ASTC emulation sync in CopyMemoryToImage

  • intel/compiler: fix bounds check in Wa_18035690555

Alyssa Rosenzweig (273):

  • jay: simplify if/else emit logic

  • jay: add missing physical CFG edge

  • jay: fix unreachable block pass

  • jay: uniformize bti

  • nir: add i/umad_32x16_intel opcodes

  • jay: fuse integer MAD

  • jay/lower_pre_ra: improve MAD immediate handling

  • jay: optimize multiplication with 32-bit constant

  • jay: fuse mad with 32-bit constant integer mul

  • jay: fuse more integer MAD

  • jay/lower_pre_ra: allow 2 immediates

  • jay/lower_scoreboard: fix flag scoreboarding

  • jay/simd_width: handle 64-bit broadcast_flag

  • jay: don’t region DPAS

  • jay: fix UACCUM use for swaps

  • jay: defeature UACCUM

  • jay: fix alignments for ugpr vectors

  • jay: fix simd splits of vectorized uniform code

  • jay: don’t touch float controls after branch

  • jay: fix derivatives in non-uniform control flow

  • jay/to_binary: fix small ugpr vectors

  • jay/opt_propagate: fix cmod prop with uniform vecs

  • jay/opt_propagate: fix cmp+ballot+32-bit and

  • jay/opt_propagate: fix copyprop of UGPR vectors

  • jay/lower_scoreboard: refactor regdist keys

  • jay: regather liveness in RA

  • jay/dag: add transpose routine

  • jay/dag: add jay_dag_foreach_edge iterator

  • jay/schedule: generalize choose_inst

  • jay/schedule: add phi src special handling

  • jay/schedule: consider multiple scheduler phases

  • jay/schedule: add a non-pressure mode

  • jay/schedule: refactors for forward scheduling

  • jay/schedule: don’t reorder timestamps

  • jay/schedule: add latency scheduler

  • jay/schedule: rename schedule_pressure -> schedule

  • jay/schedule: honour nosched within jay_schedule

  • jay/schedule: schedule for latency

  • jay/schedule: add post-RA scheduling

  • nir: allow 64-bit scalar indices

  • jay: fix MAD restrictions

  • jay: add and use jay_ugpr_per_gpr

  • anv/nir_lower_descriptor_heap: zero sampler_index

  • gen/validate: check for MAD restriction

  • jay: skip nir_opt_16bit_tex_image

  • jay: restrict shuffle strides more

  • jay/partition: drop anti-fragmentation heuristic

  • jay: adjust lane ID calculation

  • jay/schedule: bias towards accumulator use

  • jay: remove braces

  • jay/register_allocate: fix uaccum handling

  • jay: fix uniform accumulator indexing

  • jay: avoid invalid broadcasts

  • jay: lower vote_any after opt_uniform_subgroups

  • intel/gen: print sr0 subregisters

  • brw,elk: remove comment so old it makes no sense

  • brw: expose brw_nir_lower_int64

  • nir: add intrinsics for jay scan/reduce lowering

  • nir: add sample_positions_intel sysval

  • jay/validate: relax broadcast_imm validation

  • jay/assign_accumulators: use MAC more

  • jay: vectorize ugpr copies

  • jay: ingest vectorized ALU

  • jay: handle 64-bit ugpr vectors

  • jay/validate: handle 64-bit gpr-as-ugpr-vector

  • jay/to_binary: simplify region handling expression

  • jay: implement intrinsics for scan/reduce

  • jay: lower scan/reduce in NIR

  • jay: implement nir_intrinsic_last_invocation

  • jay: enable nir_opt_uniform_atomics

  • jay: implement nir_intrinsic_load_sample_positions_intel

  • jay: lower nir_intrinsic_load_sample_pos_from_id

  • jay: remove bogus restriction

  • jay: implement robust_ubo_ranges

  • jay: lower load_max_polygon_intel trivially

  • jay: fix null RT setting

  • jay: deal with unreachable loop back edges

  • jay: handle VS urb input handle

  • jay/schedule: try fewer schedules

  • jay/performance: fix accumulator latency reporting

  • jay: allow sr0.2 in the IR

  • jay: align demote handling with IGC

  • jay/schedule: ignore liveness post-RA

  • jay: rework last-use fields

  • jay: pep8

  • jay: bump max_def_length

  • jay: fix ingesting vectorized nir_op_mov

  • jay: fix d64 handling

  • jay: do not use transpose with byte aligned

  • jay: clang-format

  • jay: limit simd width with 16k of scratch

  • jay/lower_scoreboard: treat annotations as in-order

  • jay: generate bfi from shift+and+add

  • jay: optimize barriers a bit

  • jay: cache signal barrier payload

  • jay/lower_flags: try to reuse ARF moves

  • jay/arl: handle SIMD8 RT writes

  • jay/arl: handle SWSB for MATH

  • jay/arl: handle fs payload

  • jay: factor out jay_flag_type

  • jay: use flag_type for packing flags

  • jay/validate: use flag_type

  • jay: flip operands on and_sN_s32

  • jay/arl: use 16-bit for simd8 flags

  • jay/arl: # of flags

  • jay/arl: lower 32-bit multiply

  • jay/arl: set AccWrEn as necessary

  • jay/arl: port more RT write code from brw

  • jay/arl: handle load_layer_id

  • jay/arl: fork off enough HDC code for imageStore

  • jay/arl: impl front face

  • jay/arl: fix payload granularity

  • jay/arl: handle urb

  • jay/arl: handle urb eot

  • jay/arl: promote simd8 ballots to 16-bit

  • jay/arl: indexed urb

  • jay/arl: geometry shaders with vec4 urb

  • jay/arl: fb read

  • jay/arl: flag ratio

  • jay/arl: accumulator assignment in simd16

  • jay/register_allocate: deal with predicated EOT

  • jay: optimize phis with duplicated sources

  • jay: optimize out convergent 1-src booleans

  • jay: fix lcssa for loop-invariant divergent bools

  • jay: cache masked sampler state pointer

  • jay: use MAD for indirect samplers

  • jay: fix extra adds with indirect samplers

  • nir: add load_sample_pos_intel intrinsic

  • jay: lower nir_intrinsic_load_sample_pos in NIR

  • drm-shim: add all intel targets

  • jay: drop #include

  • jay/print: fix print with no block

  • jay/opt_propagate: be more graceful about immediates

  • jay/opt_propagate: generalize and/or opts to cmod

  • jay: improve inot optimizations

  • jay: bump JAY_PARTITION_BLOCKS

  • jay: drop LOOP_ONCE macro

  • jay: refactor opcode NoMask forcing

  • jay: eliminate a is_no_mask use

  • jay/opt_predicate: use op.no_mask

  • jay/to_binary: inline no_mask expression

  • jay: use I->dep.sbid consistently

  • jay: generalize predication

  • jay: combine predication impls

  • jay/print: plumb func into print

  • jay/print: clean up non-lu printing

  • jay/print: print r0 post-ra

  • jay/print: don’t indent blocks

  • jay/performance: fix .src latency calc

  • jay/performance: eliminate needless temp

  • jay/performance: use foreach helper

  • jay/performance: rename var for clarity

  • jay/performance: fix sbid tracking

  • jay/performance: add debug print

  • jay/performance: fix double counting of sbid latencies

  • jay/schedule: make latency weights less nonsense

  • jay: run vectorizer earlier

  • jay: remove pointless prog_data zeroing

  • jay/performance: A@1 does not stall all SBIDs

  • jay: remove an outdated assert

  • jay/schedule: distinguish edge types while scheduling

  • jay/schedule: weaken address register deps

  • jay/schedule: add post-acc scheduling phase

  • iris: lower gl_FragColor away

  • mesa/st: fix incorrect VS input bases

  • jay/print: skip .lu post-RA

  • jay: add jay_operand_type helper

  • jay: fix CVT using integer accumulators

  • jay: move num accum define

  • jay: eliminate jay_num_regs()

  • jay: do not use MAC for simd split instructions

  • jay/opt_propagate: fix incorrect cmod prop with abs

  • jay/to_binary: use more <1;1,0> regions

  • AGENTS.md: add initial file

  • jay: force helpers to write scratch

  • jay/partition: add some asserts

  • jay/partition: handle zero demand case

  • jay: fix alpha-to-coverage setting

  • jay/ra: fix vector_extract alignment

  • jay/assign_accumulators: deal with global accs

  • jay: rewrite the README

  • jay: split out jay_grf_per_gpr_at_width

  • jay: add simd splits to jay_inst

  • jay: simd split in a dedicated pass

  • jay: expand mul macro before jay_to_binary

  • jay: merge shuffle & vector_extract

  • jay: lower macros earlier

  • jay: split SIMD before SWSB

  • jay/ra: fix vectors falling off the reg file

  • jay/opt_postra_vectorize: vectorize at end of block

  • jay/opt_propagate: restrict send ugpr->gpr prop

  • jay/partition: ensure uniform SENDs fit

  • anv: align gl_NumWorkgroups pointer

  • intel: remove intel_use_jay_any_stage

  • intel: enable Jay by default on Xe2 and Xe3

  • intel: hide Jay unsupported platforms better

  • jay: update README

  • jay/arl: SIMD split texture instructions

  • jay/arl: 64-bit imm

  • jay/arl: generalize zip instruction

  • jay/arl: handle indirect images

  • jay/arl: bindless tgm

  • jay/arl: assign sbids to f64

  • jay/arl: packed offsets

  • jay/arl: uniform urb reads

  • jay/arl: tcs barrier

  • jay/arl: offset packed pixel coords outside simd16

  • jay/lower_helpers: don’t move half of a SEND

  • jay/lower_scoreboard: don’t add regdist to f64

  • jay/to_binary: relax quad swizzle

  • jay/to_binary: fix simd split with lane ID

  • people: sync Alyssa with .mailmap

  • bin: drop my failing unit test

  • nir/print: quiet nir prints for RT pipeline stages

  • jay/nir: fuse (x * 4) - y patterns

  • jay/ra: expand nr

  • jay/ra: rotate roundrobin less

  • jay/ra: do not roundrobin MEM

  • jay/opt_dead_code: fix null f64 destination issue

  • jay/lower_helpers: do not EOT from 16-bit

  • gallium: drop caps->endianness = LITTLE sets

  • jay: zero a0 earlier

  • jay: hoist SIMD split above post-sched passes

  • jay: add Wa_18035690555 workaround

  • jay: fix tracking of integer accumulators

  • jay: schedule after SIMD splitting

  • jay: rename jay_range to jay_footprint

  • jay: extract jay_footprint_{set,test} helpers

  • jay: extract jay_foreach_in_footprint

  • jay: make footprint helpers dumber (for now)

  • jay: track GPR footprint per GRF

  • jay: wire up generate_local_id

  • AGENTS: update for recent policy changes

  • jay/lower_spill: fix compressed instruction hazard

  • jay/ra: handle compressed instr restriction

  • jay/builder: add simd splitting knob to sends

  • jay: add SIMD splits to the partition

  • jay/register_allocate: account for SIMD splits

  • jay/simd_width: account for block SIMD width

  • jay/to_binary: account for partition simd splits

  • jay: pass sampler simd widths

  • jay/simd_width: handle splitting SENDs

  • jay/partition: fix corner case underalloc’ing stride 4

  • jay/partition: allocate blocks for split SENDs

  • jay: allow simd32 dual src on xe2+

  • jay: allow simd32 ray query

  • jay/arl: split TGM to simd8

  • jay/arl: split txd

  • jay/arl: bail on simd32 + demote

  • jay/arl: clamp send simd width

  • jay/arl: handle simd32 fbread

  • jay/arl: simd16 multisample sends

  • jay/lower_helpers: inherit simd width of the block

  • jay/lower_helpers: prefer earlier partitions

  • jay: track block weights

  • jay/performance: scale blocks by weights

  • jay: introduce 16-bit phis

  • jay: use more 16-bit moves

  • jay: promote 8-bit ineg

  • jay/stride: restrict CVT limitation

  • jay/stride: restrict OFFSET_PACKED_PIXEL_COORDS dst

  • jay/ra: add stride affinities

  • jay/ra: generalize try_find_free_reg

  • jay/ra: tweak global reg assignment

  • jay/ra: try to assign global with the right stride

  • jay/ra: reassign globally to avoid shuffles in loops

  • jay/ra: drop old stride heuristic

  • jay: handle convergent graphics sysvals

  • jay: handle shuffle of UGPR

  • jay: set nir_divergence_single_prim_per_subgroup

  • jay: assert we do not mov_indirect(ugpr)

  • jay/lower_scoreboard: fix excess SYNCs

  • jay: fix uniform stores in divergent frag shaders

Andrzej Datczuk (5):

  • ac/sqtt: increase bind_point bitfield to 2 bits in rgp_sqtt_marker_pipeline_bind

  • ac/rdp: expose elf_gfxip_level enum and conversion helper in the header

  • ac: add native RDF writer

  • radv/rra: switch to shared ac_rdf writer

  • ac/sqtt, radv, radeonsi: Fix SQTT buffer size alignment issues

Angell Fear (1):

  • panfrost: wait for in-flight jobs on JM context teardown without a kernel context

Anthony Shoumikhin (1):

  • torx: prefer ExecuTorch’s optimized CPU kernels, use the wheel in the docs

Arisa Snowbell (1):

  • radv: apply radv_gfx12_hiz_wa partial for Hunt: Showdown 1896

Arjob Mukherjee (1):

  • pvr: Implementation of VK_KHR_maintenance7

Ashish Chauhan (17):

  • pvr: add transform feedback hardware definitions

  • pvr: derive transform feedback state from compiler output

  • pvr: program DOUTI count from the fragment shader inputs

  • pdsc: add ST serialization support

  • pvr, pdsc: add xfb_main program interface

  • pvr, pdsc: add xfb_init program interface and wiring

  • pvr, pdsc: add xfb_end program interface and wiring

  • pvr, pdsc: add xfb_vertex_count program interface and wiring

  • pvr, pdsc: add xfb_query_begin program interface and wiring

  • pvr, pdsc: add xfb_query_end program interface and wiring

  • pvr: add transform feedback entry points support

  • pvr: make PPP state creation size based

  • pvr: add support for geometry only jobs

  • pvr: make query pool handling query type aware

  • pvr, pco: add support for different query result strides and report counts

  • pvr: add transform feedback stream query support

  • pvr: advertise VK_EXT_transform_feedback

Ashley Smith (4):

  • pan/bifrost: Fix shader clock selection

  • pan/genxml: Add YUV Writeback Format enum for YUV render targets

  • panfrost: no-op texture framebuffer barriers

  • mesa: only set _NEW_BUFFERS from _mesa_resize_framebuffer when size changed

Ashwin Bhat (2):

  • lavapipe: fix MSVC Release unused locals used only in assert

  • spirv: fix MSVC Release unused locals used only in assert

Axel Gneiting (1):

  • nak: Fix missing scoreboard for uniform Ldcg on SM120

Benjamin Cheng (27):

  • ac/vcn_enc: Disable var slice for preencode with VCN4

  • vl/proc: Guard compositor creation with HAVE_GFX_COMPUTE

  • radv: implement queryResultStatusSupport for enc queues

  • vk/query_pool: save encode_feedback2 flags

  • vk/query_pool: save video profile info

  • radv: Refactor video encode queries

  • ac/video: Add feedback2 quirk caps

  • radv: Support video_encode_feedback2 (frame) stats

  • radv/video: Fix interaction with KHR_extended_flags

  • ac/video: Don’t take codec level from kernel

  • radv/video: Use vk_video_get_h264_level

  • ac: Fix guardband calculation for large viewports

  • ac/uvd_dec: Only send SESSION_CONTEXT_BUFFER if needed

  • vtn: Mark Location decorated variables with explicit_location

  • glsl: Set location to -1 for SPIR-V built-ins

  • mesa/shader_query: Handle SPIR-V built-ins in validate_io

  • radeonsi: try to clarify RECTLIST/RECT_2D differences

  • mesa/main: Add extra_new_buffers for STENCIL_REF

  • util/video: Account for MaxFS in h264 level derivation

  • radeonsi: set stencil clear to 0 when enabling TC compat

  • ac/video: Adjust HEVC min_qp for VCN1/VCN2

  • radv/query: Allow creating “noop” pools

  • frontends/va: Remove ChromaToPipe

  • frontends/va: Support NV16 and P210

  • frontends/va: Add HEVC RExt support

  • radv/video_enc: fix qp map va calculation

  • radv/video_enc: fix VCN5 qp map buffer size

Birk Skyum (1):

  • freedreno: Don’t double-close imported syncobj fd

Bo Hu (2):

  • vulkan-snapshot: Delay host deletion of several Vulkan handle types

  • gfxstream/codegen: Remove obsolete frame_buffer.h from decoder includes

Boris Brezillon (12):

  • panfrost: Get rid of unused PAN_BO_ACCESS_xx flags

  • panfrost: Drop the unused dump_headerblock() helper

  • panfrost: Record access to plane1+ in case of multi-planar resources

  • panfrost: Add a panfrost_resource_wait() helper

  • panfrost: Move BO access tracking to the context layer

  • panfrost: Make BO label/flags update threadsafe

  • pan/kmod: Make sure all executables live in the same 4G VA range

  • panvk: Revisit the heap selection logic

  • panvk: Don’t downcast CS func address to a 32-bit VA

  • panvk: Don’t force non-executable buffers to live in the first 4G

  • pan/ci: Use a new kernel to fix the G925 flakes

  • panvk: Consider non memory allocation failures as driver incompatibility

Brandon Jones (1):

  • brw: use ACCESS_NON_TEMPORAL to Expose L1S_L3UC Functionality

Caio Oliveira (95):

  • intel/gen: Pass LSC op to ex-desc helpers

  • intel/gen: Store parsed LSC descriptor directly

  • intel/gen: Omit null dst in translated LSC stores

  • intel/gen: Add LSC UGM 2D block support

  • intel/executor: Add block2d example for Xe2

  • brw: Add register coalesce overlap regression test

  • brw: Make sure all coalesced source slots are filled

  • nir: Account for cmat memory accesses in copy_prop_vars

  • util: Remove string shader statistic support

  • anv: Include build and device identity in shader binary UUID

  • brw: Match fill/spill optimization scratch accesses by logical offset

  • intel/gen: Allow empty instruction options at the end

  • intel/executor: Simplify ts/gateway EOT emission

  • intel/mda: Add slice_cut_any

  • intel/executor: Use slice helpers for macro parsing

  • intel/executor: Reserve payload registers in docs and examples

  • intel/executor: Strip macro prefixes in dispatcher

  • intel/executor: Move manual to a source file

  • intel/executor: Allow multiple hardware threads

  • intel/executor: Rename executor_params to executor_run and use it more

  • intel/executor: Add per-run temporary context

  • intel/executor: Make memory handling explicit

  • intel/executor: Add explicit SIMD selection

  • intel/executor: Add SLM support

  • intel/gen: Make gen_inst_is_send public

  • intel/executor: Add automatic SWSB annotations

  • intel/gen: Add sampler message string parser

  • intel/executor: Add sampler surface support

  • intel/executor: Add support for thread groups

  • intel/gen: Fix Gfx11 3-src accumulator file encoding

  • intel/decoder: Handle unavailable binding table surfaces

  • nir, spirv: Match debug printf argument alignment with u_printf

  • nir: Align debug printf arguments in nir_printf_fmt

  • nir, spirv: Pad 3-component debug printf arguments to 4 components

  • brw: Print immediates with an explicit :TYPE suffix

  • brw: Print the is_scalar qualifier on registers

  • brw: Add a flags argument to brw_print_instructions

  • brw: Do not print a (null) dst in for instructions without operands

  • anv: Pass the correct RT shader key pointers

  • jay: Avoid use-after-free in jay_compile

  • anv: Copy accumulated resource requirements from jay RT shaders into the final prog_data

  • brw: Consider address register SWSB dependencies in lower_scoreboard

  • brw: Respect execution mask when resetting GS control data

  • intel: Preserve 64-bit hashes in shader dump filter

  • intel/jay: Respect INTEL_SHADER_DUMP_FILTER

  • iris: Initialize source hash in program data

  • anv: Close shader dump archive on device destruction

  • glsl: Store the type name as a pointer

  • util/u_cpu_detect: Emit the xgetbv mnemonic instead of its encoding

  • brw: Fix scalar image coordinate payloads

  • spirv: Discard excess image coordinate components

  • brw: Skip unspilling the destination when nothing is defined

  • spirv: Fix spec constant OpSelect for composite types

  • brw: Use reaching definitions when skipping destination fills

  • spirv: Narrow the VectorShuffle undef marker to the component bit size

  • spirv: Fix bounds check in OpSpecConstantOp CompositeExtract/Insert

  • nir, spirv: Widen boolean DebugPrintf arguments to 32-bit

  • nir: Add access flags to shared and global atomics

  • intel/compiler: Add atomic branch optimizations

  • spirv: Don’t apply mediump_16bit to operations with struct result

  • brw: Drop conservative_raster from prog_data

  • jay: Free skipped SIMD shader variants

  • jay: Track uses_fence in compute shaders

  • iris: Remove unused fs_config_param

  • intel: Remove unused per-primitive remap param

  • blorp, iris: Remove redundant driver-side VUE map setup

  • brw: Stop trying wider SIMD widths after a compile failure

  • spirv: Preserve all fp_math_ctrl bits in vtn_handle_alu default case

  • brw: Re-lower immediate offsets after mesh nir_opt_offsets

  • jay: Pick a single variant for compute shaders in Xe3+

  • intel: Don’t restrict grf_used to uint8_t in RT helpers

  • intel/compiler: Remove uses_is_indexed_draw field from prog_data

  • anv: Remove unused mem_ctx in internal shader deserialization

  • iris: Free compute compilation context on failure

  • iris: Split brw/elk paths in disk cache serialization

  • intel/compiler: Rename intel_shader_enums.h to intel_shader.h

  • intel: Add jay_compile_params

  • intel/compiler: Flatten brw_compile_params

  • intel/gen: Remove stale compact_all TODO

  • intel/gen: Allow parsing the {Compacted} annotations

  • intel/gen: Add compact support for roundtrip checker

  • jay/ra: Don’t penalize affinities that are at the tail of the block

  • intel/compiler: Fix cooperative matrix conversion signedness

  • brw/acc: Do not reassociate source modifiers across saturation

  • brw/acc: Don’t use accumulator for non-contiguous regions

  • jay: Update PHI uniform field when spilling its destination

  • jay: Be more precise when using SHUFFLE for GPR->UGPR fill

  • gen: Fix typo in validation for special restrictions

  • nir: Fix split cooperative matrix get-coordinate row offset

  • jay: Fix emission of dummy mov

  • jay: Don’t lower U32 movs to F32 when cmod is being used

  • jay: Handle uniform derivatives without varying temporaries

  • brw: Preserve the VGRF result of an accumulator substitution path

  • intel/mda: Preserve binary data in printraw

  • intel/compiler: Archive shader binaries as separate MDA objects

Calder Young (35):

  • nir/builder: Add nir_halt_if helper

  • nir: Make it easier to define private driver values in SLM

  • brw: Add pass to emulate Xe2+ active thread barriers on Gfx125

  • anv: Add drirc to emulate Xe2+ active thread barriers on Gfx125

  • brw: Add workaround pass for broken RE engine shader

  • anv: Add drirc option to workaround broken RE engine shader

  • anv: Enable the temporary RE engine workaround for several titles

  • jay: Avoid corrupting prog_data for TCS when barriers are used

  • jay: Clean up instruction selection logic for barriers

  • jay: Handle more combinations of barrier memory semantics

  • jay: Implement workaround for HSD 14020414266 and 22020984324

  • jay: Add a CHECK_TDR opcode to help lower interlock

  • jay: Add support for fragment shader interlock

  • jay: Unpack 16-bit vectors in SEND return payloads

  • jay: Add support for 16-bit sampler return payloads

  • jay: Fix broken FP16->INT conversion dst restriction

  • jay: Unset zero_inactive when cond_flag is optimized away

  • jay: Handle irregular vector sizes for LSC block loads

  • jay: Handle reconvergence on the physical CFG correctly

  • jay: Make sure fall-through edges are always the first successor

  • jay: Don’t clear regdists when the logical CFG doesn’t fall through

  • jay: Fix an additional reconvergence issue with the physical CFG

  • anv: Add debug options for skipping CHS and AHS during RT traversal

  • drm-uapi: Sync xe_drm.h

  • intel: Document the page fault reporting enums more

  • intel: Add page fault reporting enums for source id

  • intel: Decode page fault source ids on the Xe KMD

  • anv: Include page fault sources in device fault reports

  • jay/to_binary: Fix uniform accumulators in SIMD32

  • jay/scoreboard: Fix an issue with multiple nested loops

  • util: Add more ways to delete from dynarray

  • gen: Add helper for encoding bindless ex_desc immediates

  • jay: move lowering call out of nested if statement

  • jay: Add a variant of jay_shrink_sources for ISA sources only

  • jay: Encode the mlen and ex_mlen portions of descriptors post-RA

Caleb Callaway (1):

  • anv: hide Intel vendor ID for Cyberpunk

Callum Fare (1):

  • ir3: Fix potential out-of-range load_const_ir3 base offset

Carlos Santa (1):

  • intel/tools: Enable surface and sampler decoding in aubinator_error_decode

Casey Bowman (7):

  • pps: Default to the driver minimum sampling period

  • anv/pps: Update the minimum sampling period formula

  • anv: Disable MSAA depth write-through

  • intel/dev: add option for INTEL_THREADS_PER_EU_SET

  • jay: enable INTEL_THREADS_PER_EU_SET

  • brw: enable INTEL_THREADS_PER_EU_SET

  • docs: add documentation for INTEL_THREADS_PER_EU_* options

Caterina Shablia (1):

  • panfrost: remove redundant util_dynarray_init

Celeste “Coelacanthus” Liu (1):

  • nvk/meson: add build root dir to nvk_include_directories

Chris Healy (4):

  • panfrost: enable ASTC decode mode on Bifrost (v7)

  • panvk: enable VK_EXT_astc_decode_mode on Bifrost (v7)

  • panfrost/model: Add Bifrost pixel/texel/FMA rates

  • pan/bi: Make bi_gather_stats arith and texel rates model-aware

Christian Gmeiner (238):

  • panvk: Compile the bind queue code only once

  • panvk: Advertise sparse binding on the main queue family

  • mesa/st: Invalidate FS sampler views after PBO texture transfers

  • nir: Unvendor raw_vertex_id/raw_vertex_offset from panfrost

  • nir: Unvendor the panfrost XFB lowering pass

  • nir: Add options to nir_lower_xfb_to_stores

  • nir/opt_vectorize: Remove the phis that were combined

  • freedreno: Build blit shaders with nir_builder

  • etnaviv: Set the ASTC sRGB bit in the texture descriptor

  • etnaviv/drm-shim: Report zero performance monitor domains

  • etnaviv/drm-shim: Take features and limits from the hardware database

  • drm-shim: Add etnaviv

  • panvk: Fix num_workgroups patching for dynamic precomp dispatches

  • libpan: Rename copy.cl to libpan_copy.cl

  • panvk: Add indirect memory copy kernel to libpan

  • panvk: Implement CmdCopyMemoryIndirectKHR

  • panvk: Use exact dispatch sizing for indirect memory copies on v13+

  • panvk: Advertise VK_KHR_copy_memory_indirect

  • etnaviv: texture-desc: Set texture border color for CLAMP_TO_BORDER

  • etnaviv: Drop stale stream output relocs in etna_update_hwxfb(..)

  • etnaviv: Coalesce TFB buffer state emission

  • etnaviv: Set up TFB buffer address and size at bind time

  • lavapipe: Embed bindless sampler views by value

  • nir: Add cmat_muladd_pan intrinsic

  • pan/bi: Add MMUL cooperative matrix instructions

  • panvk: Lower cooperative matrix operations

  • panvk: Move panvk_nir_lower_tile_image(..) declaration to panvk_nir.h

  • panvk: Advertise VK_KHR_cooperative_matrix

  • etnaviv: Update headers from rnndb

  • etnaviv: Turn the sampler parameter loads into constants

  • etnaviv: Handle sRGB formats via util_format_linear(..)

  • etnaviv: Derive the shadow compare shader key from current sampler state

  • etnaviv: Keep the SE scissor rectangle within the hardware limit

  • etnaviv: Use the 12 bit vertex stride field of the NFE

  • ci: Run shader-db for etnaviv

  • ci: Run the shader-db job on etnaviv changes

  • st/mesa: Only drop ARB_depth_buffer_float on desktop GL

  • etnaviv: Update headers from rnndb

  • etnaviv: rs: Add the rgb10_a2 RS format

  • etnaviv: Cache the shader key and variant lookup

  • etnaviv: drm: Do not return a stale fence fd

  • etnaviv: Program the i2i destination type in src1

  • etnaviv: Emit TFB_COMMAND after the TFB buffer state

  • etnaviv: drm: Cache the submit bo table index

  • etnaviv: Update headers from rnndb

  • etnaviv: Stop checking the sample count in can_create_resource(..)

  • etnaviv/isa: Add txf_ms instruction

  • etnaviv: texture-desc: Support multisample textures

  • etnaviv: Implement texelFetch on multisample textures

  • etnaviv: nir: Set maximum sample count

  • etnaviv: blt: Allow multisample rendering to integer formats up to 32 bit

  • etnaviv: rs: Allow multisample rendering to integer formats up to 32 bit

  • etnaviv: Resolve stencil8 like the integer color formats

  • etnaviv: Report the hardware sample positions

  • etnaviv: Skip the dummy vertex element when the id pseudo-input is used

  • gallium: Let a driver take the sample coverage value directly

  • etnaviv: Program sample coverage and mask through PS.MSAA_CONFIG

  • etnaviv: Extract gpu_supports_texture_desc(..)

  • etnaviv: Accept multisample render formats without an RS entry

  • etnaviv: Expose multisample textures on halti5

  • etnaviv: nir: Call nir_opt_combine_stores(..)

  • panvk: Implement VK_KHR_shader_constant_data

  • etnaviv: hwdb: Add PE_A8B8G8R8 feature

  • etnaviv: Rename RB_SWAP PE format defines

  • etnaviv: Update headers from rnndb

  • etnaviv: Handle native RGBA PE format

  • etnaviv: Compute resource render compatibility once

  • etnaviv: hwdb: Add PE_32BPC_COLORMASK_FIX feature

  • etnaviv: Widen 32bpc RT colormasks without PE_32BPC_COLORMASK_FIX

  • etnaviv: Do not abort on unsupported stencil-only blits

  • etnaviv: Return success for etna_blit_stencil_fallback(..)

  • etnaviv: Use RS depth formats only on halti5+

  • etnaviv/ci: Update CI expectations

  • etnaviv/ci: Update CI expectations

  • etnaviv: Drop unreachable compute-only path in etna_reset_gpu_state(..)

  • etnaviv: Track dirty sampler states per slot

  • etnaviv: Only refresh dirty texture descriptor slots

  • bifrost: Support an inline constant pool on Valhall

  • kraid: Encode the pinned FAU page for instructions that require it

  • kraid: Support the inline constant pool

  • panvk: Wire up nir_opt_large_constants on Valhall

  • panvk: Support large constants on Bifrost

  • bifrost,kraid: Report the constant pool size in shader stats

  • util: Add u_hexdump_words(..)

  • bifrost,panvk: Dump the inline constant pool with the shader

  • etnaviv: blt: Clear recycled tile status buffers on the GPU

  • etnaviv/ci: Update flakes for gc7000-r6214

  • etnaviv: Report GLSL 1.30 on halti2 and later

  • etnaviv/ci: Update list of gles2 extensions

  • etnaviv: Render shared R/B swapped targets through a shadow

  • etnaviv: Drop the fragment shader R/B swap

  • etnaviv: Update headers from rnndb

  • etnaviv: texture-desc: Set the border color of the 128-bit companion

  • etnaviv: Apply the texture swizzle of 128-bit views in the shader

  • etnaviv: Clamp the texture offset LOD to the view levels

  • etnaviv: Emit the per-RT blend state on framebuffer changes

  • etnaviv: blt: Use the internal format for the masked clear bits

  • etnaviv: blt: Clear each 128-bit plane with its own clear bits

  • llvmpipe: Pass the sampler matrix to its helpers explicitly

  • llvmpipe: Make sampler matrix registration safe for concurrent use

  • llvmpipe: Load sampler matrix function pointers atomically

  • llvmpipe: Keep replaced sample_functions arrays alive

  • llvmpipe: Make sample function promotion safe for concurrent readers

  • llvmpipe: Move the sampler matrix to the screen

  • llvmpipe: Scope the texture/image handle functions to the screen

  • llvmpipe: Retire replaced memory through one trash list

  • llvmpipe: Reclaim retired sampler matrix memory

  • gallium: Add PIPE_CONTEXT_PARAM_MAG_SWITCHOVER_HALF

  • nir: Add lower_txl_mag_switchover to nir_lower_tex

  • etnaviv: Move magnification switch-over for shader LODs

  • etnaviv: Honor hw tileability when picking a super-tiled modifier

  • etnaviv: Do not compose a swizzle onto an inline immediate

  • etnaviv: rs: Take the sample scale into account for blit offsets

  • etnaviv: blt: Skip tile status for clears when it is not valid

  • etnaviv: Split the inline immediate encoding out of const_src(..)

  • etnaviv: Split BO upload out of etna_icache_upload_shader(..)

  • etnaviv: Split the uniform limit out of etna_compile_check_limits(..)

  • etnaviv: Carry shader constant data in a buffer read with LOAD

  • etnaviv: Spill literals to memory when the constant pool overflows

  • gallium/hud: Build shaders with NIR instead of TGSI

  • gallium/hud: Drop vs_color’s dead texcoord varying

  • etnaviv: Factor out level_size(..)

  • etnaviv: hwdb: Add TX_INTEGER_COORDINATE_V2 feature

  • etnaviv: Lower texelFetch without integer texture coordinates

  • etnaviv: blt: Handle partial clears of depth32f formats

  • etnaviv: rs: Use the internal format for clears

  • etnaviv: rs: Handle partial clears of depth32f formats

  • etnaviv: Resolve multisample float surfaces with u_blitter

  • etnaviv: Add uniforms for XFB emulation

  • etnaviv: Handle XFB intrinsics in the NIR backend

  • etnaviv: Emulate transform feedback with shader stores

  • etnaviv: Activate XFB emulation in the draw path

  • etnaviv: Emulate the primitives emitted query without HWTFB

  • etnaviv: Support gl_VertexID with 16 vertex attributes

  • etnaviv: Expose stream output on HALTI2+ without HWTFB

  • etnaviv: Reserve the vertex position output slot when it is unwritten

  • etnaviv: Clamp out-of-bounds texelFetch coordinates

  • panvk: Zero images built for device-level queries

  • panvk: Report image compression properties

  • panvk: Honor VkImageCompressionControlEXT

  • panvk: Advertise VK_EXT_image_compression_control

  • etnaviv: Update headers from rnndb

  • etnaviv/hwdb: Expose shader cache limits and tessellation feature

  • etnaviv: Derive the shader cache split from the core info

  • etnaviv: Derive the HALTI5 vertex shader cache state from core info

  • lavapipe: Use goto cleanup in lvp_CreateDevice()

  • lavapipe: Guard BDA handle arrays with bda_lock

  • lavapipe: Create bindless handles without the queue lock

  • llvmpipe: Make shader and variant counters atomic

  • lavapipe: Create shader CSOs without the queue lock

  • lavapipe: Delete shader CSOs on the queue thread

  • lavapipe: Remove the queue lock

  • lavapipe: Make the device queue an array

  • lavapipe: Make lazy pipeline CSO creation thread-safe

  • lavapipe: Expose two queues per family

  • etnaviv: Update headers from rnndb

  • etnaviv: Capture varyings not read by the fragment shader

  • etnaviv: Keep transform feedback active across flushes

  • etnaviv: Update headers from rnndb

  • etnaviv: blt: Use the TILED value for 4x4 tiled images

  • etnaviv: blt: Set UNK22 only on copies

  • etnaviv: hwdb: Expose BLT_8bpp_256TILE_FC_FIX

  • etnaviv: Skip tile status for 8 bpp MSAA without the BLT fix

  • etnaviv: Use 256B tile status for compressed MSAA on SMALL_MSAA cores

  • etnaviv: Keep both halves of the depth stencil fast clear value

  • etnaviv: Use single buffer mode 3 with stencil-only surfaces

  • etnaviv: Disable hardware ASTC

  • etnaviv: hwdb: Expose PE_RGBA16I_FIX

  • etnaviv: Factor out the render target output store helpers

  • etnaviv: Pack RGBA16 integer render targets into G32R32F

  • etnaviv: Set the overwrite bit only if every target is overwritten

  • docs/new_features: Document support for GLES3 on etnaviv/HALTI5

  • etnaviv/ci: Update expected GLES version

  • etnaviv/ci: Update list of extensions

  • docs: Add GLES3 line

  • etnaviv: hwdb: Add TX_BORDER_CLAMP_FIX feature

  • etnaviv: Write texture descriptor LOD addresses at compose time

  • etnaviv: Copy compressed resources on the CPU

  • etnaviv: Add a padded border shadow resource

  • etnaviv: Add etna_sampler_uses_border(..)

  • etnaviv: Add etna_sampler_view_resource(..)

  • etnaviv: Use a border shadow for CLAMP_TO_BORDER

  • util: Fix signed shift overflow

  • ac: Fix signed shift overflow

  • radeonsi: Fix signed shift overflow

  • etnaviv: Fix signed shift overflow

  • nvk: Fix signed shift overflow

  • r300: Fix signed shift overflow

  • radv: Fix signed shift overflow

  • util/blend: Add PIPE_ADVANCED_BLEND_KHR_MODES_MASK

  • gallium: Convert blend_equation_advanced cap to a mode bitmask

  • glsl: Let the caller pick the advanced blend modes to lower

  • mesa/st: Support partially native advanced blend modes

  • etnaviv: hwdb: Add PE_ADVANCE_BLEND_PART0 feature

  • etnaviv: Program native advanced blend modes

  • panvk: Move to CmdEndRendering2KHR(..)

  • panvk: Honor resolve transfer function flags

  • panvk: Advertise VK_KHR_maintenance10

  • etnaviv: Update valid_buffer_range for stream output buffers

  • etnaviv/ml: Do not discard input tensor buffers

  • etnaviv: Replace BOs instead of stalling on busy buffers

  • etnaviv: Implement invalidate_resource(..)

  • etnaviv/drm: Add some CPU tracepoints

  • etnaviv: Add some CPU tracepoints

  • etnaviv: Move disk_cache into etna_screen

  • lavapipe: Unwrap trace screen and context before llvmpipe calls

  • pan/bi: Lower imul_32x16 and umul_32x16 in NIR

  • kraid/hw_tests: Test OpIMul with a widened src1

  • kraid: Lower imul_32x16 and umul_32x16 in NIR

  • panvk: Advertise VK_INTEL_shader_integer_functions2

  • etnaviv: Skip render compatibility for external TS plane resources

  • etnaviv: Take a destination origin in etna_copy_resource_box(..)

  • etnaviv: Size transfer staging resources to the mapped box

  • etnaviv: Allocate cube map transfer staging as a 2D array

  • etnaviv: Use cached memory for read transfers on arm64

  • mesa: Drop const from the DrawGallium draw info

  • mesa: Fix draw preparation ordering

  • mesa: Do not update draw functions outside GL_RENDER

  • etnaviv/ci: Rework gles2+3 testing

  • etnaviv/ci: Update number of imx8mq-nitrogen boards in my farm

  • panvk: Report optimal image transfer granularity

  • panvk: Advertise VK_KHR_maintenance11

  • etnaviv: Track constant buffer changes per stage

  • etnaviv: Upload user constant buffers only when a shader reads them

  • etnaviv: blt: Program the source TS for mipmap generation

  • etnaviv: Update sampler sources before deriving state

  • etnaviv: Only update sampler sources the bound shaders use

  • etnaviv: blt: Re-validate tile status on full color clears

  • etnaviv: Allow unaligned vertex input on HALTI0 and newer

  • etnaviv/isa: Add denorm bit to ALU instructions

  • etnaviv/isa: Decode store destination register

  • etnaviv/isa: Add img_load and img_store tests

  • etnaviv/isa: Add img_store_3d instruction

  • etnaviv/isa: Add img_load_3d instruction

  • etnaviv/isa: Add img_addr instruction

  • etnaviv/isa: Add img_addr_3d instruction

  • etnaviv: Expose only the linear modifier for NV12

  • etnaviv: Do not set the destination for stores

Christoph Pillmayer (45):

  • pan: Fix BIFROST_MESA_DUMP_DIR

  • pan/kraid: Fix 32bit hw_runner builds

  • pan/perf: Fix 32bit build of panquick

  • kraid: Add opt_promote_consts

  • kraid/isa: Update isa for v9 to v15

  • nir: Add nir_move_options::move_all

  • pan: Use nir_move_options::move_all

  • pan: Inline specialized IDVS variants for v12+

  • pan/bi: Fix vector phis from spilling

  • kraid/encode_v9: Allow uint for SrcMod::BNot

  • kraid/ir: Make SrcMod EnumAsU8

  • kraid: Check mods and more swizzles for small constants

  • pan/bi: Fix binary shader dumps

  • kraid: Update isa files

  • kraid/proc: Allow querying fields used by expressions

  • kraid: Implement Display for EncodeError

  • kraid/proc: Store exact value in PhysicalField

  • kraid: Fix enum value default arch

  • kraid/proc: Add name and ident getters for InstrField

  • kraid/proc: Parse instruction syntax

  • kraid/proc: Store bitsize in FieldType::PcRelOffset[Un]signed

  • kraid/proc: Staging reg and name getters for Instr

  • kraid: Add SourceEncoding[64] implementations

  • kraid/proc: Add decoder generator

  • kraid: Add a disassembler

  • kraid: Add disassembler tests

  • kraid: Make decoder::try_decode return a Result

  • kraid: Separate field loading from InstrPrint

  • kraid: Try to decode instruction fields for try_decode

  • kraid: Check restricted fields during decoding

  • kraid: Support querying supported swizzles per source

  • kraid: Add ir::Src::resolve_imm(DataType)

  • kraid/small_constants: Use op_src_supported_swizzles

  • kraid/small_constants: Use Src::resolve_imm

  • kraid/ir: Make Src[Ref] comparable in tests

  • kraid: Add opt_normalize_constants

  • pan: Use kraid disassembler if available

  • kraid: Print disassembly if KRAID_DEBUG=print

  • kraid/decoder: Delete try_dec_* functions

  • kraid: Print correct stats for varying shader

  • pan/perf: Fix swapped block counts

  • pan/ds: Add missing percent unit

  • panvk/utrace: Fix payload assert

  • pan/ds: Report correct gpu_clock_id

  • kraid/ra: Fix round robin search start going OOB

Collabora’s Gfx CI Team (22):

  • Uprev VVL to 6727ae4734bf9e1e6f6954254bbc382687a3866d

  • Uprev Piglit to fa1953546a74ad0f91f0b13f7ca28a25706842fc

  • Uprev VVL to 8dc2d227ef6ca2fa7c4cca5711b0856b8dc18dff

  • Uprev ANGLE to c09abf1d3a9f3ff231a46fad806aa35ad2b3b3ad

  • Uprev VVL to 3f81f25aa04e97d94cdeb49442c9c8d79f926d95

  • Uprev Piglit to 65f44948b98d21f82ac119f8b3b5cfd866d2698a

  • Uprev VVL to b835cdcb74297baa81f049770962884f4d08d7be

  • Uprev ANGLE to be80ce591a481c12d60c50d6040d40c035b40a2b

  • Uprev Piglit to a705d674d11624a9ad9f670185b2dda9b1ee196f

  • Uprev ANGLE to 5ff3c3731a0558e73c6bf5ef739b585931e8c5e4

  • Uprev VVL to 84449ddbb2d0769b07483f1cf09fe30e64040e34

  • Uprev Piglit to 372590d39085e8640742341b5ec5a8a72ad1c56f

  • Uprev Piglit to 136db767d5eb88eb439574a25ba429926170da4b

  • Uprev ANGLE to 2c68772a0eb74c771e2b0b6c490869c0e98fc412

  • Uprev VVL to 1ff661d59fa63d580a03c5392dab9548f143b019

  • Uprev VVL to ad4ed518c3c9783b9c9ff912c205c987b17d7bf4

  • Uprev ANGLE to bda1785b8b16ef93269df45955a367d0d60aca45

  • Uprev Piglit to 81f9eac8cc0b8b12ace3629a24e934aac8f2c37c

  • Uprev VVL to 1d99d26ea11e1714c5bc729a6029a684703b5000

  • Uprev VVL to e48304c3d33abf4affd0cbd8944c8fb3e316a753

  • Uprev Piglit to c25575c9bfc82d48d6895af98eea5d9ed01f23f8

  • Uprev VVL to c93a48d963e7f311d8ca0511b5f41dc3abe63bff

Connor Abbott (74):

  • tu: Fix resetting command streams with writeable BOs

  • tu: Reshuffle shader hash

  • tu: Move all physical_device compiler options to a dedicated struct

  • tu: Fix compiler options padding size

  • tu: Fix condition for skipping emitting aprons

  • ir3/disasm: Check for invalid alias register

  • freedreno/qrisc: Fix indirect jump decoding

  • ir3: Don’t serialize fake a0.x writers in postsched

  • ir3: Don’t assume a delay for fake a0.x producers

  • ir3: Don’t schedule bar.g in a mova/(ul) sequence

  • ir3: Don’t schedule aliasing texture instructions between mova/(ul) on a7xx

  • ir3: Don’t use alias.tex inside predt/predf/prede sequence

  • tu: Use correct pointer for vis stream patchpoint cs fence

  • tu: Handle renderpasses with only some images subsampled

  • tu: Handle resolves in CmdBeginCustomResolveEXT

  • tu: Don’t write LRZ direction when not writing LRZ

  • freedreno/lrz: Move down fdl6_lrz_get_max_fdm_extra_size

  • freedreno/fdl: Rewrite fdl6_lrz_get_max_fdm_extra_size

  • freedreno/lrz: Fix fast-clear size computation

  • tu: Add missing WFI to tu6_clear_lrz()

  • tu: Extract out LRZ clear helper

  • tu: Set MSM_SUBMIT_SYNCOBJ_* when there are only internal syncobjs

  • ir3: Handle sparse loads in ir3_nir_lower_64b_image

  • nir: Don’t allow peephole_select on read_invocation

  • tu: Workaround 32B rollover bug in GPU firmware

  • nir: Add an access index to bindless_resource_ir3

  • freedreno: Mark SSBO/UBO loads and descriptors as speculatable

  • tu: Keep track of PARTIALLY_BOUND_BIT in descriptor sets

  • tu: Correctly set speculatability

  • ir3: Mark driver UBO loads as speculatable

  • ir3/opt_preamble: Consider speculatability correctly

  • ir3: Consider speculability when lowering UBOs to consts

  • ir3: Don’t use early preamble when the preamble is not speculatable

  • vulkan: Update headers/XML to 1.4.362

  • nir: Add texture_2_non_uniform/sampler_2_non_uniform

  • spirv: Mark non-uniformity of texture_2/sampler_2

  • ir3: Handle non-uniformity of texture_2/sampler_2

  • tu/knl: Plumb through iova alignment parameter

  • tu: Support VALVE_buffer_device_address_allocation_alignment

  • freedreno: Add subpass_fence_cleans_resolve property

  • tu: Don’t hardcode feedback_invalidate flushes

  • tu: Move up subpass dependencies with EXTERNAL source

  • tu: Implement SUBPASS_SLICE_FENCE flush

  • tu: Use SUBPASS(_SLICE)_FENCE for by-region dependencies

  • tu: Use SUBPASS(_SLICE)_FENCE for GMEM invalidation

  • nir: Improve load_input_attachment_coord

  • tu: Switch to using input attachment coord intrinsics

  • nir: Remove ir3-specific input attachment options

  • nir/lower_input_attachments: Drop sparse load handling

  • nir: Add input attachment intrinsics

  • vulkan/lower_descriptor_heaps: Handle input attachment load intrinsics

  • nir: Add input attachment flags to nir_tex_instr

  • vk/nir_convert_ycbcr: Move down assert

  • tu: Remove texture_index assert

  • anv/nir: Remove input attachment index from tex instrs

  • panvk/nir: Remove input attachment index from tex instrs

  • microsoft/compiler: Ignore texture_offset src for input attachments

  • nir/lower_tex: Copy input attachment information for fmask fetch

  • nir: Gather fb_fetch_output from new intrinsics

  • nir, vulkan: Use input attachment intrinsics

  • nir/lower_input_attachments: Pass through input attachment information

  • nir/lower_input_attachments: Stop getting index from the deref

  • tu: Lower input attachments using information from texture instruction

  • nir: Add access to rq_initialize intrinsic

  • vtn: Mark non-uniformity on ray query initialize

  • tu: Handle ray queries to non-uniform ASs

  • nir: Add VK_QCOM_image_processing+descriptor heap texture source types

  • vk/nir_lower_descriptor_heaps: Remove useless sampler check

  • vk/nir_lower_descriptor_heaps: Refactor texture handling

  • vk/nir_lower_descriptor_heaps: Handle VK_QCOM_image_processing

  • vulkan/graphics_state: Don’t get IAL state when not needed

  • tu: Don’t bail after the first instr when gathering push const range

  • tu: Fix bindless descriptor invalidate sequence

  • ir3: Only call nir_opt_shrink_vectors once

Cooper Morgan (3):

  • vulkan/wsi: keep present timing records that do not fit the output array

  • vulkan/wsi: keep GOOGLE display timing records that do not fit

  • anti-lag: convert GPU timestamps to ns in double precision

Corentin Noël (1):

  • mesa: fix variable shadowing

Cristina Murillo (1):

  • anv/video: set the AV1 encode bit depth if color config is not provided

Daivik Bhatia (3):

  • pan/perf: Don’t count the blend prologue

  • pan/va: Hoist independent instructions above WAITs

  • pan/va: Add tests for WAIT instruction hoisting

Daniel Almeida (12):

  • nvk/image: do not ask nil for a multi-planar format when picking a modifier

  • nvk/image: return the right plane for each memory plane aspect

  • nvk/video: add H.264 decode support in Rust

  • nvk/video: support layered DPB for H.264

  • nvk/video: set a non-zero decode watchdog timeout

  • nvk: implement events on the video queue

  • nvk: advertise VIDEO_DECODE_*_BIT_KHR for NV12

  • nvk/video: advertise VK_KHR_video_maintenance1

  • nvk/video: support VK_KHR_video_maintenance2 (inline session parameters)

  • nvk/video: report accurate video format properties

  • nvk: include the video decode layouts in the host image copy layout lists

  • nvk: add a nvk_video_enabled() helper

Daniel Lang (10):

  • etnaviv: add register addr comments to state emmissions

  • etnaviv: fix PE_RT_ADDR_{4,8}_PIPE addr comment

  • etnaviv: separate for loops in state emission

  • etnaviv: reorder state emission to optimize loads

  • etnaviv/layout: New library

  • etnaviv/layout: Add format table

  • etnaviv/layout: Add eil_format_is_emulated_z32f(..)

  • etnaviv/layout: Add eil_format_is_emulated_128bit(..)

  • etnaviv/layout: Add eil_format_emulated(..)

  • etnaviv: Drop translate_vertex_format_type(..)

Daniel Monteiro (5):

  • docs/panfrost: document PAN_I_WANT_A_BROKEN_VULKAN_DRIVER

  • ac: use os_localtime()

  • vulkan/rmv: use os_localtime()

  • radv: use os_localtime()

  • intel/mda: use os_localtime()

Daniel Schürmann (16):

  • nir: always return unsigned u2uN for integer downcasts in nir_type_conversion_op()

  • nir/opt_algebraic: remove some nir_i2iN pattern

  • nir/lower_int64: remove handling of i2iN downcasts

  • nir: special-case nir_i2i*() builder functions to emit u2u for downcasts

  • v3d/algebraic: use u2uN instead of i2iN for downcasts

  • panfrost/algebraic: use u2uN instead of i2iN for downcasts

  • panfrost: use nir_i2iN() in lower_bool_to_bitsize_instr()

  • nir/validate: Validate that i2iN opcodes are only used for sign-extension

  • util: move ringbuffer.h from anti-lag to util

  • util/ringbuffer: add ringbuffer_free_n() macro

  • vulkan/runtime: add vk_frame_pacer utility framework

  • radv: implement VK_AMD_anti_lag

  • radv: advertise VK_AMD_anti_lag

  • vulkan: delete anti-lag layer

  • nir/lower_continue_constructs: preserve phi sources when simplifying loop jumps

  • nir/lower_continue_constructs: preserve SSA dominance across nested exits

Daniel Stone (16):

  • egl/dri2: Call swap_buffers_with_damage without damage

  • egl/android: Implement swap_buffers_with_damage

  • egl/gbm: Implement swap_buffers_with_damage

  • egl/x11: Remove swrast non-damage swap_buffers

  • egl/kopper: Remove swap_buffers implementation

  • egl/x11/dri3: Remove swap_buffers hook

  • egl/wayland: Remove swap_buffers hook

  • egl/wayland/kopper: Remove swap_buffers hook

  • egl: Replace swap_buffers with swap_buffers_with_damage

  • dri: Push swap_buffers_with_damage fallback up

  • egl/wayland: Remove driSwapBuffers use

  • egl/x11: Remove driSwapBuffers use

  • glx/dri: Remove driSwapBuffers use

  • dri: Remove suffix from driSwapBuffersWithDamage

  • kopper: Remove non-damage kopperSwapBuffers

  • ci/panfrost: Add G610 and G925 traces

Daniel van Vugt (1):

  • gallium: Set a default max_texture_upload_memory_budget

Danylo Piliaiev (62):

  • tu: Custom resolve should always use AVOID_CCU layout

  • tu: AVOID_CCU should avoid depth CCU if custom resolve writes D/S

  • tu: Add msrtss to render_pass tracepoint

  • tu: Fix subsampled metadata and blit emission for separate stencil

  • tu: Fix gfx_write_access checking for TRANSFORM_FEEDBACK_COUNTER_READ_BIT

  • tu: Fix blit_cache_cleaned never being set to true

  • tu: Fix LRZ handling for VK_EXT_custom_resolve

  • tu: Dirty LRZ after changing attachment locations disable LRZ writes

  • nir: Include scalarized component offsets in UBO ranges

  • tu: Fix tu_event not being reset on creation

  • tu: Merge disable_write_for_rp from secondary to primary

  • tu: Fix memory leak of FDM patch-points

  • tu: Preserve LRZ for S/R and secondary write-disable

  • tu: Avoid retroactive LRZ disable for S/R and secondaries

  • freedreno/perfcntrs: Allow reserving unassociated counters

  • tu: Fix multi-pass raw perf queries (TU_DEBUG=perfcraw)

  • tu: Use PERFORMANCE_COUNTER_SCOPE_COMMAND_BUFFER for perfcraw

  • tu: Don’t expose BV_ counters in perfcraw queries

  • tu: Fix vkResetQueryPool for pipeline statistics query

  • tu: Simplify query pool resets

  • tu: Fix condition for implicit dep of separate DS resolve

  • tu: Fix CmdSetRenderingInputAttachmentIndices with custom resolve

  • tu: Resuming dynamic rendering should re-emit attachments state

  • tu: CmdSetRenderingInputAttachmentIndices should dirty LRZ

  • tu: Fix gmem not being disabled for some non-fb-space stages with sync2

  • tu: Fixup LRZ disabled_at_draw info

  • tu: Don’t store attachments before dynamic custom resolve subpass

  • tu: Fix renderpass tracepoints being lost with s/r chains

  • tu: Disable LRZ for depth resolve targets

  • tu: Don’t reuse LRZ when it is cleared with partial render area

  • freedreno,tu: Extract and make more accurate common fd_ticks_to_ns helper

  • freedreno/pps: Don’t give perfetto unconverted GPU timestamps

  • freedreno/computerator: Fix perfcounter calculations

  • tu: Wait for XFB query writes to land before making it available

  • tu: Synchronizing with CmdCopyQueryPoolResult should wait for mem writes

  • tu: Use p_atomic_read to read from query’s slot->available

  • tu: Fix PACKED + alignas(8) application to the same struct on older GCC

  • freedreno/pps: Queue perf counter stream samples

  • tu: Handle lazy memory in GetDeviceMemoryOpaqueCaptureAddress

  • tu: Fix missing primitives in FDM due to sample locations mismatch

  • tu: Move tu_render_pass_state_merge before draw cs merging

  • tu: Track sticky lrz_write_disabled in render pass state

  • tu: Allow LRZ writes in narrow “stencil tagging” case

  • tu: Make CmdClearAttachments record TU_ACCESS_BLIT_WRITE_GMEM access

  • util/xmlconfig: Add logNondefaultOptions to print non-default options

  • tu: Log drirc options on `TU_DEBUG=startup`

  • tu/kgsl: Prevent overflow in kgsl_queue_wait_fence

  • tu/kgsl: Distinguish timestamp and sync wait errors

  • tu/kgsl: Prevent indefinite wait times for 0 timeout waits

  • tu/kgsl: Always round up the wait timeout

  • tu/kgsl: Prevent OOB array access in kgsl_syncobj_foreach_state

  • tu/kgsl: Select the earliest timestamp for wait-any

  • tu/kgsl: Correctly interpret `poll` result in wait_any

  • tu/kgsl: Propagate syncobj export failures

  • tu/kgsl: Handle syncobj errors in queue_submit

  • tu/kgsl: Correctly merge timestamps and sync files

  • tu/kgsl: Handle FD-only waits in wait_any

  • tu/kgsl: Fix profile obj leak and check suballoc result

  • tu/kgsl: Handle errors around profiling buffer/ioctls

  • tu/a6xx: Set correct value for GRAS_MODE_CNTL

  • tu/a6xx: Invalidate shared consts before disabling them

  • freedreno: Fixup incorrectly copied A6XX_TPL1_DBG_ECO_CNTL for A650

Dave Airlie (42):

  • nir/coopmat: use args for flexible dimension lowering info.

  • vtn: add support for getting constant bool value.

  • vtn: make composite manipulation functions non-static

  • nir: add tensor layout/view enums.

  • nir: add cmat call for tensor load/store

  • vtn: add support for tensor layout/view and load/store operations

  • nir: add support for calculation tensor layout/view indexes.

  • nir: add flexdim support for tensor load/store

  • radv: add support for coopmat2 tensorAddressing

  • nir: add tensor load decode function pointer.

  • vtn: add decode option for block loads

  • nir: add block load decode support

  • radv: add support for coopmat2 cooperativeMatrixBlockLoads

  • nir: move load_ssbo_address to taking variable size src0

  • gallivm: implement load_ssbo_address intrinsic

  • lvp: add support for load_ssbo_address to vulkan lowering.

  • lavapipe: add tensorlayout and blockloads NV_cooperative_matrix2 support

  • vulkan: update spec to 1.4.359

  • radv/nir/cmat: align convert and transpose intrinsics.

  • nir: add nir_intrinsic_cmat_get_coordinate.

  • vtn: add support for cooperative matrix get coordinate

  • vtn: add cmm1 conversion support

  • vtn: rename aliased caps to EXT from NV

  • radv: add support VK_EXT_cooperative_matrix_maintenance1

  • lavapipe: add support for VK_EXT_cooperative_matrix_maintenance1

  • mesa/get: fix luid size and driver callback.

  • radv/ci: add khronos link to the bfloat16 dequant failures.

  • nir: add a central load struct field helper.

  • nir/coopmat: mark split_vars as hidden declaration

  • nir/coopmat: add a helper for cmat call var and use helpers in cmat code.

  • nir/coopmat: refactor reduce split code in advance of wg support

  • nvk: Add an nvk_video_session object

  • nvk: Add physical device queries for video

  • nvk: expose an H.264 video decode queue

  • anv: fix regression on coopmat per element operations feature.

  • nvk/nak/cmat: fix inconsistent enum name.

  • llvmpipe: handle indirect geometry shader inputs

  • llvmpipe: add vector width to the disk cache.

  • lavapipe: fix multisample depth<->color copies.

  • egl/surfaceless: don’t probe device name if fd is invalid.

  • llvmpipe: advance 1D framebuffer fetches at four lanes

  • lavapipe: fix pipeline cache size return and checks

David Rosca (76):

  • vulkan/video: Fix coding AV1 decoder/encoder_buffer_delay

  • vulkan/video: Don’t code AV1 decoder model info when not present

  • vulkan/video: Fix coding AV1 operating points

  • vulkan/video: Remove always true checks in encode_av1_seq_hdr

  • d3d12/video: Don’t reset batches in fence_wait

  • vulkan/video: Fix coding H265 ref pic list modification lists

  • vulkan/video: Fix coding H265 SPS pcm block sizes, inter ref pic set and lt refs

  • mr-label-maker: Label src/vulkan/runtime/vk_video as video

  • va: Remove unused upload_sampler function

  • va: Fix leak when vlVaUploadImage fails

  • va: Ensure templat is valid for temporary surfaces

  • va: Remove templat argument from vlVaHandleSurfaceAllocate

  • radeonsi: Stop forcing GTT with no gfx/compute

  • ac/video: Fix number of AV1 single refs

  • radeonsi/mm: Allow two refs without compound for AV1 encode

  • radeonsi/mm: Use encode picture alignment from video caps

  • ac/video: Update VCN 5.3 encode caps

  • ac/uvd: Use correct max_dw for destroy cmd

  • vulkan: Allow drivers to force internally synchronized queue

  • radv/video: Add ASSERTED to build_*_cmd return variables

  • radv/video: Implement destroy session for UVD < 6.3

  • radeonsi/mm: Don’t use HEVC per slice loop filter across slices on old FW

  • ac/uvd: Make create cmd work with duplicate handle on UVD < 6.3

  • drm-uapi: Update amdgpu_drm.h for userq changes

  • ac/gpu_info: Require amdgpu 3.65 to enable userq

  • winsys/amdgpu: Add userq support for timeline syncobjs

  • radeonsi: Enable NV_timeline_semaphore with userq

  • radv/video: Fix out of bounds reads of AV1 tile params

  • radeonsi/mm: Use correct feedback size for VCE

  • radv/video: Fix AV1 decode qmatrix params

  • radeonsi/mm: Fix AV1 decode qmatrix params

  • va: Don’t allow changing number of elements of coded buffer

  • va: Set a reasonable minimum size for coded buffer

  • va: Compare file descriptions to deduplicate objects when exporting surface

  • va: Parse all HEVC slice NAL unit types

  • va: Parse HEVC encode slice_temporal_mvp_enabled_flag

  • va: Fix parsing HEVC decode slice_temporal_mvp_enabled_flag

  • va: Fix parsing HEVC VPS timing info

  • radeonsi/mm: Return error when encode feedback size is zero

  • radeonsi/mm: Fix VCN encode allowed_max_bitstream_size

  • radeonsi/mm: Set PIPE_BIND_SHARED for video buffers

  • radeonsi: Don’t copy next pointer from template when creating texture

  • radeonsi: Support reallocate inplace for multi plane textures

  • radeonsi: Don’t force PIPE_BIND_SHARED for multi plane textures

  • pipe: Add reference to pipe_video_codec

  • va: Use refcounting for pipe_video_codec

  • va: Keep codec ref in surface/buffer for fence wait/destroy

  • va: Don’t allow NULL codec in vaEndPicture

  • ac/uvd: Don’t force h264 max dpb size on new kernel

  • va: Drop HEVC SEI parsing

  • va: Drop AV1 metadata OBU parsing

  • pipe/video: Remove HDR metadata

  • radv/video: Implement cu_qp_delta workaround for old FW

  • radv/video: Override sample_adaptive_offset_enabled_flag when not supported

  • radv/video: Fix setting session parameters override info

  • vulkan/video: Don’t update OBU size with NULL data_ptr

  • vl/bitstream: Fix coding ue() for large numbers

  • vl/bitstream: Mask value with bit count in vl_bitstream_put_bits

  • vl/bitstream: Don’t use internal buffer for size only use

  • vl/bitstream: Move overflow check to vl_bitstream_write_byte_start_code

  • radv/video: Fix handling size limit in GetEncodedVideoSessionParameters

  • va: Drop format support checks in vlVaPostProc

  • va: Don’t check format support in EndPicture for processing

  • radv/video: Only report BT709/Limited support for RGB conversion on VCN5

  • radeonsi/mm: Enable 2pass search center map on VCN 5.3

  • radeonsi/mm: Set error when failing to map bitstream buffer

  • radeonsi/mm: Don’t return feedback when allocating buffer fails

  • radeonsi/mm: Fix pipe_context leak when cs_create fails

  • radeonsi/mm: Check embedded buffer map fails

  • radeonsi/mm: Increase obu_size length to 2 bytes in AV1 sequence header

  • ac/uvd: Set uv pitch from surface instead of half y pitch

  • radv/video: Allow DRM format modifiers for video images on GFX6-8

  • ac/vcn_dec: Fix division by zero in AV1 film grain params

  • ac/vcn_dec: Clamp num_y/cb/cr_points and ar_coeff_lag in AV1 film grain params

  • radeonsi/mm: Set PIPE_BIND_SHARED also for video buffers with modifiers

  • va: Unmap coded buffer on encode failure

David Sanders (1):

  • radv: reset compute pipeline after IES in vkCmdExecuteGeneratedCommandsEXT

Denys Fedoryshchenko (1):

  • freedreno: Add missing <type_traits> include

Derek Lesho (13):

  • zink: Fix zink_bo_unmap synchronization for client pointer support.

  • zink: Catch all prior updates to source valid_buffer_range.

  • Update OpenGL headers for GL_MESA_map_buffer_client_pointer.

  • mesa,mapi: add MESA_map_buffer_client_pointer extension

  • zink: zink_buffer_map: factor out unneeded map_offset into trans->offset.

  • zink/ci: Skip KHR-GL46.direct_state_access.buffers_errors.

  • mesa: Implement MESA_map_buffer_client_pointer.

  • threaded_context: Implement MESA_map_buffer_client_pointer.

  • zink: Factor out thread-relevant context determination for uploader.

  • zink: Implement MESA_map_buffer_client_pointer.

  • threaded_context, zink: Introduce PIPE_MAP_SHORT_LIVED.

  • zink: zink_buffer_subdata: Set PIPE_MAP_SHORT_LIVED.

  • threaded_context: tc_buffer_subdata: Set PIPE_MAP_SHORT_LIVED.

Dhruv Mark Collins (14):

  • tu/autotune: Stabilize RP hash for replays

  • tu/autotune: Track binning pass for RP duration

  • tu/lrz: Fast-clear the part of LRZ that the flag RAM covers

  • tu/tile_config: fix extent compatibility check when merging FDM bins

  • tu: Fix inverted variable-descriptor check in can_speculate_descriptor_load

  • util/u_trace: Let tests reset the process-wide state

  • util/u_trace: Only close an open frame on context teardown

  • util/u_trace: Let the printers end tracepoint lines

  • util/u_trace: Assign an ID to each trace context and print it

  • util/u_trace: Support per-context trace files

  • util/u_trace: Serialize trace file writes

  • util/u_trace: Support gzip-compressed trace files

  • util/u_trace: Support per-process trace files

  • tu: Set SP_CHICKEN_BITS_2 bit 30 on A750 for HL: Alyx

Dmitry Baryshkov (98):

  • tu: limit KHR_internally_synchronized_queues to Vulkan 1.1+

  • tu: limit VALVE_fragment_density_map_layered to Vulkan 1.1 devices

  • ci: add Qualcomm LAVA lab

  • ci/lava: support deployment to flasher

  • ci/freedreno: add rules to schedule jobs to the Qualcomm lab

  • ci/lava: don’t override /lib symlink with LAVA_FIRMWARE

  • ci/lava: bump lava-job-submitter tag

  • ci/freedreno: add tests for the A702 Adreno GPU

  • freedreno/a702: drop passing test

  • freedreno/a2xx: index the fragment linkage by driver location

  • drm-shim: forward the glibc time64 entry points on 32-bit

  • panfrost/drm-shim: go through uintptr_t when casting a query pointer

  • nouveau/drm-shim: use a 64 bit constant for the VRAM BAR size

  • ci: build drm-shim in the arm test builds

  • freedreno/drm-shim: add a225

  • freedreno/a2xx: consume lowered IO directly

  • freedreno/a2xx: drop indirect IO deref lowering

  • freedreno/a2xx: remove the IO variables of lowered TGSI shaders

  • freedreno/a2xx: stop advertising GL_EXT_frag_depth

  • freedreno/a2xx: fix the window scissor losing a row and a column

  • freedreno/a2xx: don’t abort on a shader that does nothing

  • freedreno/registers: correct a2xx.xml against the vendor register file

  • freedreno/a2xx: read gl_FragCoord from the component asked for

  • freedreno/a2xx: let nir lower fsign

  • freedreno/registers: document how CP_DRAW_AUTO derives the vertex count

  • turnip: scale the transform feedback counter offset on a6xx

  • freedreno/a6xx: subtract the transform feedback buffer offset in CP_DRAW_AUTO

  • freedreno/a6xx: seed the streamout counter in the units the hardware uses

  • freedreno/a5xx: seed the streamout counter in the units the hardware uses

  • freedreno/drm-shim: add a650

  • drm-shim: add freedreno/a225

  • nir/lower_int_to_float: correct idiv from the remainder

  • freedreno/ci: drop a306 expectations shadowed by a skip

  • freedreno/ci: drop a306 skips that match no test

  • freedreno/ci: drop the inert test phase timeout on a306-piglit

  • freedreno: derive the workgroup dimensions from the device

  • tu: derive the workgroup dimensions from the device

  • nv30: fix abs source modifier on fragment program sources 1 and 2

  • freedreno/a5xx: implement depth clamp

  • freedreno/a5xx: implement dual-source blending

  • freedreno/ci: drop obsolete a530 expectations, flakes and skips

  • ir3: keep byte SSBO offsets before a6xx

  • freedreno/a5xx: only 2D-blit between compatible formats

  • freedreno/a5xx: expose ASTC HDR

  • freedreno/a5xx: program NUMNONPOSVAR with position’s VPC location

  • freedreno: don’t dereference a vertex buffer that isn’t there

  • ir3: accept the edge flag output instead of failing to compile

  • freedreno/a4xx+a5xx: clamp the texture buffer size

  • freedreno: keep the separate stencil out of the shadow blit

  • freedreno/a5xx: blit the separate stencil plane too

  • ir3: take the register file size from the device table

  • freedreno/a5xx: the register file holds 64 vec4

  • ir3: skip the phis of non-GPR registers in the minimum pressure bound

  • ir3: report an unreachable register limit rather than assert

  • ir3: keep the register banks inside the file without merged registers

  • ir3: print the register limits next to where they are used

  • ir3: fit every wave of a barrier workgroup in the register file

  • ci/freedreno: run the a530 pixel storage mode tests again

  • ci/freedreno: run the a530 tests skipped for being slow or hanging

  • freedreno: don’t spin forever when an RD dump cannot be opened

  • freedreno/a5xx: bound blits by the layer count, not the depth

  • freedreno/a5xx: advertise ARB/OES_copy_image

  • ci/freedreno: run half of a530 piglit per merge request

  • ci/freedreno: stop running the shader tests twice on a530

  • ci/freedreno: add a nightly unfractioned a530 GL job

  • ci/freedreno: update a530 expectations for the unfractioned dEQP run

  • freedreno/a2xx: keep the swizzle when a source modifier folds

  • freedreno/a2xx: honor the component a packed varying starts at

  • freedreno/a2xx: don’t type-pun the immediate values through a cast

  • freedreno/a2xx: mark texture fetch constants as textures

  • freedreno/a2xx: set the depth on a cube map fetch constant

  • freedreno/ci: a200: drop the cube mipmap flake

  • freedreno/a2xx: move the magnification switch-over for shader lods

  • tu: resolve wide formats with a shader on A702

  • freedreno/qrisc: fix hang on a650 / a660 firmware

  • freedreno/qrisc: implement A702 support

  • freedreno/qrisc: make BV and LPAC optional for a7xx+

  • freedreno/qrisc: handle Adreno X1-45 and 722

  • freedreno/common: flag a workaround for CP_DRAW_AUTO issue in a630_sqe

  • freedreno/registers: define format of the PC_AUTO_VERTEX_STRIDE register

  • freedreno/a6xx: workaround an issue in a630_sqe.fw

  • tu: workaround an issue in a630_sqe.fw

  • ci: bump FIRMARE_TAG, pulling in updated a630 firmware

  • freedreno/a5xx: set the sample count before the GMEM restore blits

  • freedreno/a5xx: write the linear centroid interpolator regid

  • ir3: keep the msaa key for shaders that interpolate at gl_SampleID

  • freedreno/a5xx: set the msaa shader key

  • ir3: read the a5xx sample position as f16

  • ir3: force per-sample execution on a5xx with a SampleID input

  • freedreno/a5xx: program the per-sample shading mode

  • freedreno/a5xx: emit the fragment sample-mask output

  • freedreno/a5xx: enable the centre 1/w interpolator

  • freedreno/a5xx: advertise the sample-shading extensions

  • ci: disable Qualcomm farm due to setup changes

  • ir3: do not let full-register writes reset half-register delays before a6xx

  • ir3: reserve image dimensions for read-write images on a5xx

  • freedreno/a4xx+a5xx: clear PIPE_BIND_OPENCL before checking format support

  • freedreno/ir3: wait for the indirect grid copy before loading it

Dmitry Osipenko (2):

  • virgl: Support tiled gbm modifiers

  • intel/virtio: Fix multi-queue sync

Dom Cobley (1):

  • v3d: don’t drop the last row in the sand30 blit

Dominique Belhachemi (5):

  • gallivm: orcjit: fix use-after-free in remove_jd() during exit

  • pco: fix compile failure for arrayLength() on non-power-of-two strides

  • pvr: raise maxImageDimension3D to 2048

  • pco: fix workgroup barrier race when reused across loop iterations

  • pvr: amend shared memory coeff reg count to use DWORD granularity

Dorinda Bassey (1):

  • magma-gpu-rs: change Event methods to &self

Duncan Brawley (1):

  • pco: Conditionally insert DITR/DITRP pre-fence only when necessary

Dylan Baker (39):

  • intel/compiler: Fix copy and paste bug

  • intel/compiler: Add assertion to help coverity out

  • intel/compiler: Avoid rechecking first character of string_view

  • egl: rewrite dispatchstubs generator to be more efficient

  • intel/compiler: Delete useless statement

  • intel/compiler: Fix typo in comment (dst -> src)

  • intel/anv: fix potential overflow before widen

  • intel/compiler/brw: Use moves instead of copies in scoreboarding pass

  • intel/gen: decode type of src1 in basic 2 source after setting IMM

  • intel/gen: Only set address_mode, negate, and abs for non-IMM cases

  • nir: Add an unreachable default to generated switch statement

  • docs: Add calendar entries for 26.3 release candidates.

  • gen_calendar_entries: fix datetime for newer Python

  • meson: Drop version check for GCC < 4.4.6

  • mr-label-maker: add support for intel/jay

  • intel/jay: Don’t attempt to do work when there is none

  • intel/dev: Add assert to ensure stage is not MESA_SHADER_NONE in intel_use_jay_for_stage

  • nir: Silence coverity warning about integer overflow

  • intel/jay: free count array after transposing DAG

  • VERSION: bump to -rc1

  • .pick_status.json: Update to 44fd0fde8187b5bb1e919625f1ba24cb36c18cbc

  • .pick_status.json: Update to f66af5ff99307e628b78de9250c8e948f25389ab

  • VERSION: bump for 26.3.0-rc2

  • .pick_status.json: Update to 86bc45c5f3fe6dee69c143ccd1567de6d98daaf0

  • meson: imgui-sdl3-vulkan needs the vulkan loader headers

  • .pick_status.json: Update to 04d7f733b89ae9cf1cc3544b9fc7d07d1b43878b

  • .pick_status.json: Update to f7dd61f78f83c9ffb3dc9c980cd9b303a5ce58b1

  • .pick_status.json: Update to e9177010db177e63966b6d1373efc2a6c582a428

  • .pick_status.json: Mark 42f3d153ce9fa0f656deec73faad51751e7ac117 as backported

  • .pick_status.json: Mark c06c5f9252a59c6d278c40d0f52a7b2c2b44e603 as backported

  • .pick_status.json: Mark e4e82b62140ab9c81a57543f48f8b8a2786f33a6 as denominated

  • .pick_status.json: Mark f3f63efe9ca77e197210d9ece06f8e5cedc0f853 as denominated

  • VERSION: bump to 26.3.0-rc3

  • .pick_status.json: Update to 0af57c3a99171880b177592d916b24e24aa0c53d

  • .pick_status.json: Update to 0618bfe0ffb0e92e55053b381fb1cbc844db9aa2

  • .pick_status.json: Update to f4871b8173d78afcc998c22f8b709a3b639b925f

  • .pick_status.json: De-nominate a large number of zink patches

  • .pick_status.json: Update to 7d52dbb1dd9f26a9d6dbf35569efd6bf274f4e97

  • .pick_status.json: Update to 5b2db4b8e3e3b8732f6bb076fd8da4764ac92b0b

Ella Stanforth (5):

  • pvr/ci: update ycbcr related axe flakes

  • pvr: enable VK_KHR_synchronization2

  • pvr: conditionally dump kernel words 9 through 11

  • pvr: remove unused pvr_pds.h includes

  • pvr: zalloc device

Emma Anholt (78):

  • freedreno: Don’t force image component A=1 substitution on R/RG textures.

  • mesa: Add a cap for drivers to set what counts for MAX_*_OUTPUT_COMPONENTS.

  • radv/ci: Update the traces checksum for radv-raven-traces-restricted

  • freedreno/ir3: Consistently use uint32_t for instruction counts.

  • freedreno/ir3: Actually pass cs_lock_unlock_quirk into the compiler.

  • freedreno/ir3: Size inputs[] for the actual sysvals in use currently.

  • tu: Fix VK_PIPELINE_CREATION_FEEDBACK_APPLICATION_PIPELINE_CACHE_HIT_BIT.

  • tu: Fix stitching together set layouts from graphics pipeline libraries.

  • tu: Add missing reservation of the extra safe-const VPC state for GSes.

  • tu: Fix the block texel view check for force_disable_linear_fallback.

  • tu: Publish proper aspect masks for D32S8 VkSparseImageMemoryRequirements2.

  • tu: Publish proper VkSparseImageFormatProperties2 for D32S8.

  • tu: Log autotune’s overrides as reasons sysmem/gmem was forced.

  • ir3: Use nir_opt_non_uniform.

  • tu: Log in TU_DEBUG=perf when we overflow VSC allocations.

  • tu: Add TU_DEBUG=gmem_warmup to preallocate a large VSC for GMEM rendering.

  • tu: Stop writing two copies of the FDM viewport/scissor state.

  • tu: Fix handling of pImmutableSamplers in tu_GetDescriptorSetLayoutSupport().

  • tu: Free through the right allocator on vkCreateDescriptorSetLayout fail.

  • tu: Free the new set on descriptor set allocation fragmentation failure.

  • tu: Properly distinguish pool fragmentation from pool exhaustion.

  • tu: Use memcpy instead of assuming opaque capture data pointers are aligned.

  • tu: Use STACK_ARRAY() for unbounded attachment_count-based allocations.

  • tu: Don’t advertise sample count > 1 for FDM/FSR usage.

  • tu: Don’t advertise buffer support for compressed formats.

  • tu: Don’t advertise sparse residency when the pdev can’t do it.

  • tu: Drop TRANSFER_SRC/DST from bufferFeatures.

  • tu: Fix 3D subresourceLayout.size and memcpy size.

  • tu: Fix the sType of the returned VkFragmentShadingRateKHRs.

  • tu: Drop !filter_width assertion for 8xx.

  • tu: Fix leaks of a few device mutexes.

  • tu: Fix destroy order of autotune vs perfcounters in failure path.

  • tu: Fix failure unwind ordering error for bin preamble.

  • tu/kgsl: Fix double-close on load failure.

  • tu/kgsl: Don’t leak the device/dma_fd on UBWC init failure.

  • tu/msm: Fix leak of fd on tu_try_enable_vm_bind() failure.

  • tu: Fix data race on memory_budget_props->heapUsage.

  • tu: Don’t try to use lazy allocation for MSRTSS if the device can’t.

  • tu: Return 0 for non-sparse image/buffer opaque capture data.

  • radv: Return 0 for non-sparse image/buffer opaque capture data.

  • anv: Return 0 for non-sparse buffer opaque capture data.

  • anv: Fix descriptor_heap sparse image capture/replay handling.

  • anv: Fix descriptor_heap non-sparse image capture/replay.

  • ir3: Start using nir_opt_fp_math_ctrl.

  • ir3: Start using nir_opt_phi_to_bool

  • ir3: Start using nir_opt_algebraic_distribute_src_mods.

  • ir3: Expose support for [ui]find_msb_rev (CLZ.B, CLZ.s).

  • ir3: Drop our custom lowering for find_msb.

  • ir3: Add support for nir_op_bitnz/bitz (ir3’s GETBIT).

  • ir3: Add support for nir_op_bfm (ir3’s MGEN).

  • ir3: Drop dead code for nir_op_[iu]hadd.

  • freedreno/ci: Update trace checksums for all of these ir3 changes.

  • ir3: Create IR3_REG_SRC_MODS to replace repeated lists of src mods.

  • ir3: Refactor out the “apply source mods to the immediate value” code.

  • ir3: Make source mod eval strictly follow hardware behavior for denorms.

  • ir3: Correct SABS evalution in the 16-bit case.

  • ir3/cp: Search existing consts to reuse them with different src mods.

  • nir/draw_helpers: Add support for polygon stippling in samplers_as_deref mode.

  • nir/draw_helpers: Add support for polygon stippling with io-lowered position.

  • softpipe: Advertise multi_draw_indirect.

  • mesa: Sink ST_PIPELINE_RENDER_STATE_MASK/st_prepare_draw() into DrawGallium*.

  • mesa: Lower polygon stipple for drivers that don’t support it.

  • draw: Remove the prior polygon stipple emulation.

  • svga: Retire its polygon stipple emulation in favor of mesa/st’s.

  • d3d12: Retire its polygon stipple emulation in favor of mesa/st’s.

  • i915: Actually set the stipple that the frontend gives us.

  • freedreno/tools: Add a5xx support to computerator.

  • intel: Don’t include intel_perf lib sources into intel_perf lib consumers.

  • nir/opt_varyings_tests: Move dedup test code body into the class.

  • nir/opt_varyings_tests: Move bicm_binary_alu test code body into the class.

  • nir/opt_varyings_tests: Move prop_const test code body into the class.

  • nir/opt_varyings_tests: Move prop_uniform code body into the class.

  • nir/opt_varyings_tests: Move prop_ubo code body into the class.

  • nir/opt_varyings_tests: Move prop_uniform_expr code body into the class.

  • nir/opt_varyings_tests: Move dead_input/output code body into the class.

  • freedreno: Ensure that the generated pack headers are stable between builds.

  • isaspec: Fix encoder output stability.

  • nir/opt_algebraic_test: Fix unit test codegen output stability.

Emre Cecanpunar (2):

  • kosmickrisp: don’t cache a NULL Metal compiler

  • jay: report scratch memory size in shader stats

Eric Engestrom (84):

  • VERSION: bump to 26.3

  • docs: reset new_features.txt

  • people: update my email address

  • docs: update calendar for 26.1.5

  • docs: add release notes for 26.1.5

  • docs: add sha sum for 26.1.5

  • docs: update calendar for 26.2.0-rc1

  • docs: update calendar for 26.2.0-rc2

  • pick-ui: require reverts to be explicitly nominated for backport

  • pick-ui: turn commit.date into a (cached) property

  • pick-ui: show MR number for additional context

  • meson: give names to libdrm drivers attributes checks

  • gallivm: drop #undef of never-defined symbols

  • mesa: inline bug report url

  • turnip/ci: drop invalid FREEDRENO_HANGCHECK_MS in a618 and a750 toml suites

  • mr-label-maker: deduplicate adding the ~CI label to any CI farm disable/restore

  • mr-label-maker: document affected drivers in case of ci-tron farms outage

  • ci: add missing .ci-farms/* for new farms

  • docs: update calendar for 26.2.0-rc3

  • docs: update calendar for 26.1.6

  • docs: add release notes for 26.1.6

  • docs: add sha sum for 26.1.6

  • broadcom/ci: document a couple more vc4 flakes

  • etnaviv/ci: document a few more flakes

  • freedreno/ci: document regressed tests, as well as some flakes

  • nvk/ci: document a vkd3d regression

  • nvk/ci: skip dEQP-VK.shader_object.binary.device_features.* as it now times out

  • llvmpipe/ci: document fixed test

  • panfrost/ci: document fixed tests

  • dozen/ci: document flakes

  • zink+radv/ci: sort navi10 flakes

  • zink+radv/ci: document more navi10 flakes

  • docs: update calendar for 26.2.0

  • docs: add release notes for 26.2.0

  • docs: add sha sum for 26.2.0

  • docs: add release dates for the 26.2.x cycle

  • ci: move virgl jobs from merge pipelines to pre-merge/nightly only

  • docs: update calendar for 26.1.7

  • docs: add release notes for 26.1.7

  • docs: add sha sum for 26.1.7

  • docs: update calendar for 26.1.8

  • docs: add release notes for 26.1.8

  • docs: add sha sum for 26.1.8

  • docs: update calendar for 26.2.1

  • docs: add release notes for 26.2.1

  • docs: add sha sum for 26.2.1

  • gfxstream/meson: drop unused include_directories()

  • ci-tron: bypass caching proxy for docker images

  • ci: update contact information for the imagination farm

  • ci: re-enable valve farms

  • docs: update calendar for 26.2.2

  • docs: add release notes for 26.2.2

  • docs: add sha sum for 26.2.2

  • llvmpipe/ci: update expectations for nightly jobs

  • zink+nvk/ci: update expectations for nightly jobs

  • lavapipe/ci: update expectations for nightly jobs

  • virgl/ci: update expectations for nightly jobs

  • meson: document what breaking bug bindgen 0.72.0 has

  • meson: document bindgen 0.73.0 & 0.73.1 as bad_versions

  • docs: update calendar for 26.2.3

  • docs: add release notes for 26.2.3

  • docs: add sha sum for 26.2.3

  • bin/ci_run_n_monitor: trivial whitespace fix

  • bin/ci_run_n_monitor: add `–no-new-job-after 12h`

  • meson: drop leftover option `d3d-drivers-path`

  • meson: drop leftover var dep_wl_protocols

  • meson/zink: drop unused var inc_ntv

  • radv/ci: disable gfx1201 & navi21 merge jobs

  • Apply 1 suggestion(s) to 1 file(s)

  • meson/mediafoundation: drop leftover var rc_compiler

  • meson/mediafoundation: drop unused mediafoundation.def

  • meson/gfxstream: drop leftover vars

  • docs: update calendar for 26.2.4

  • docs: add release notes for 26.2.4

  • docs: add sha sum for 26.2.4

  • ci: take kws farm offline for maintenance

  • docs/submittingpatches: drop duplicate spaces in lists

  • docs/submittingpatches: remove unenforced rule about number of lines in backported patches

  • docs/submittingpatches: clarify that only performance *regression* fixes can be backported

  • docs/submittingpatches: document that exceptions to the backport rules are possible and how

  • docs: update calendar for 26.2.5

  • docs: add release notes for 26.2.5

  • docs: add sha sum for 26.2.5

  • panfrost/ci: fix hw_jobs filter

Eric Guo (4):

  • pan/bi: relax load/store alignment constraints on Valhall (v9+)

  • pan/bi: enable late_lower_int64 on Valhall (v9+)

  • pan/bi: run nir_opt_gcm to hoist loop-invariant code

  • panfrost/perf: fix external bus width

Eric R. Smith (8):

  • pan: in register allocation, consider spilling the node being allocated

  • panfrost: restore max retry count for bifrost LCRA spills

  • pvr: move assert for buffer usage

  • pvr: implement push descriptors for powervr

  • pvr: add push descriptor templates

  • pvr: add VK_KHR_push_descriptor to the docs

  • pvr: update axe-1 CI results

  • panfrost: fix UB in pan_csf pointer based type-punning

Erik Faye-Lund (106):

  • mesa/main: use marshal_count if present

  • pan/ci: mark a few expected flakes

  • glapi: correct value for ALL_ATTRIB_BITS

  • glapi: store group for enums and params

  • mesa/glapi: trace bitfields symbolically

  • panvk: expose VK_KHR_spirv_1_4 on bifrost

  • panvk: expose vulkan 1.3 on bifrost

  • CODEOWNERS: update panfrost maintainers

  • mailmap: update my email

  • tgsi: remove LIT opcode

  • tgsi: remove DST opcode

  • tgsi: remove LOG/EXP opcodes

  • tgsi: remove more unused opcodes

  • main,glsl: add initial GL_EXT_YUV_target bits

  • glsl: add samplerExternal2DY2YEXT

  • glsl: add yuvCscStandardEXT type

  • glsl: add color-space keywords

  • glsl: add layout for EXT_YUV_target

  • glsl: emit error when combining layouts

  • glsl: loosen EXT_YUV_target restriction

  • glsl: emit yuv-layout related errors

  • glsl: add rgb_2_yuv and yuv_2_rgb

  • pan/genxml: add v11 XML

  • panfrost: build v11 support

  • panfrost: add v11 support

  • panvk: add v11 support

  • panfrost: enable v11 gpu models

  • docs: fix javascript-warning on search-page

  • pan/ci: document recent regression in xfails

  • pan/ci: update expected results

  • mesa/st: add missing license statement

  • update my email address etc

  • mr-label-maker: remove entries for non-existent things

  • mesa/glapi: report NULL-pointers consistently

  • tgsi: remove more unused opcodes

  • tgsi: remove ballot/vote opcodes

  • tgsi: remove bindless opcodes

  • tgsi: remove more unused opcodes

  • tgsi: remove LOD opcode

  • tgsi: remove even more opcodes

  • tgsi: remove dead opcode

  • etnaviv: remove tgsi leftover includes

  • lavapipe: remove needless tgsi includes

  • mesa/st: remove pointless tgsi includes

  • gallium/aux: remove pointless tgsi include

  • gallium/aux: drop needless ureg header

  • mesa/st: use nir_trim_vector

  • remove d3d10umd frontend

  • softpipe: remove defunct null-gs-tokens support

  • llvmpipe: remove defunct null-gs-tokens support

  • docs/panfrost: list G71 as supported

  • docs/panfrost: add missing trailing commas

  • docs/panfrost: do not limit to Bifrost

  • docs/panfrost: rephrase lima reference

  • docs/panfrost: switch to list-table syntax

  • docs/panfrost: list opencl as supported

  • docs: make sure a footnote-list has margin below

  • docs/panfrost: use footnotes for conformance status

  • docs/panfrost: link to v10 OpenCL submission

  • docs/panfrost: use note for experimental details

  • pan/compiler: do not misreport v4 as utgard

  • pan/compiler: drop incorrect UNUSED

  • pan/compiler: drop needless cruft

  • pan/compiler: drop unused gpu_id

  • pan/compiler: pass arch around when sufficient

  • pan/compiler: base texture lowering on arch

  • pan/compiler: base image lowering on arch

  • pan/compiler: base varying layout collection on arch

  • pan/compiler: base output lowering on arch

  • pan/compiler: base preprocessing on arch

  • pan/compiler: base metadata updating on arch

  • pan/compiler: pass arch to mem_vectorize_cb

  • pan/compiler: drop needless temporary

  • panfrost: pass arch around instead of gpu_id

  • pan/compiler: do not re-compute arch

  • pan/compiler: do not re-compute arch

  • pan/compiler: use unnormalized stats when unknown

  • panfrost: do not build or use libpan on midgard

  • pan/compiler: rework bifrost quirks

  • pan/compiler: old-blend is a v4 thing

  • pan/compiler: broken-lod is a v4 thing

  • pan/compiler: no-ooo is a v4 thing

  • panfrost: do not open-code MIDGARD_BROKEN_BLEND_LOADS detection

  • pan/compiler: remove unused midgard-quirks

  • pan/compiler: pass gpu_prod_id to midgard_get_quirks

  • pan/compiler: do not re-query midgard quirks

  • pan/compiler: pass arch and model to compiler

  • panfrost: make blend-cso code take pan_model

  • panfrost: make preload-cache take pan_model

  • panfrost: add pan_model_device_name helper

  • panfrost: correctly report immortalis

  • panfrost: plug precomp-shader leak

  • panvk: fix allow_merging_workgroups logic

  • correct link in copilot-instructions.md

  • panfrost: add ai-policy

  • panvk: drop redundant descriptor-copy

  • pan/lib: fix repeated width_px check

  • pan/lib: fix copy-pasta bug

  • pan/lib: do not fold into the same op twice

  • panfrost: free correct handle

  • pan/bi: handle tls-stores in helper analysis

  • vulkan/meta: correct image-resolve for 3D image

  • panvk: properly ignore depth-stencil aspectMask

  • vulkan/meta: fixup typo

  • panvk: track heap budget across instances

  • panvk: write_image_view_desc deals with images, not samplers

Fabio Piparo (1):

  • vc4: don’t reuse flags from before a thread switch

Fabricio Akio (1):

  • freedreno/a5xx: Reload textures, SSBOs and images after the program

Faith Ekstrand (289):

  • pan/nir: Use a single ALU op switch in lower_bool_to_bitsize

  • pan/nir: Cast 64-bit comparison results to 32 bits

  • kraid/nir: Implement 64-bit integer down-conversions

  • kraid: Rework comparison folding

  • kraid: Add OpICmpMulti and plumb it through

  • kraid/nir: Implement 64-bit integer comparisons

  • kraid/legalize: Be more strict about duplicate SSA values

  • kraid/legalize: Restructure legalize_vec_srcs()

  • kraid/ra: Rework register alignment constraints

  • kraid/ra: Rework the RA flow

  • kraid/swizzle: Implement Debug for Swizzle

  • kraid/swizzle: Add a SwizzleWord::modify() helper

  • kraid/swizzle: Add Swizzle::ZERO and an is_zero() helper

  • kraid: Handle Swizzle::ZERO in Src::word()

  • kraid/copy-prop: Handle zero swizzles

  • kraid/swizzle: Allow composing word with byte swizzles

  • kraid: Handle single-word swizzles in Src::swizzle()

  • kraid/hw_tests: Use Src::swizzle() for filling out sources

  • kraid: Validate swizzle invariants

  • nvk/queue: Clean up some error handling

  • nvk: Expose UBO/SSBO descriptor construction helpers

  • nvk: Store the actual image descriptor in image views

  • nvk: Refactor nvk_image_view_init()

  • kraid/ra: Rework eviction

  • kraid/ra: Kill sources and destinations up-front

  • kraid: Move the guts of ALU swizzling into a Builder::mkvec_vN helper

  • kraid/nir: Use the right number of components in tex dst_types

  • kraid/v9: Fix op_dst_supported_lanes() for vecsize dsts

  • kraid: Hide validation in release mode behind KRAID_DEBUG=validate

  • kraid/v9: Assert TEX write masks match destination types

  • kraid: Add dst_type and write_mask back to OpTexGradient

  • kraid/nir: Lower [iu]mul_high for 8 and 16-bit

  • kraid/nir: Implement fquantize2f16

  • pan/bi: Vectorize load/store before lower_mem_access_bit_sizes()

  • kraid: Move phis into their own file/module

  • kraid: Move PhiMap into phi

  • kraid: Add a concept of block prelude/postlude and some helpers

  • kraid: Don’t re-materialize constants interleaved with phis

  • kraid: Allow immediates on OpPhiSrc

  • kraid/validate: Validate block sections

  • kraid/validate: Print undefined SSA values

  • kraid: Don’t allow OpCopy with a non-trivial swizzle

  • kraid: Fix up the pressure model

  • kraid: Add a bitset to LiveSet

  • kraid: Panic if we don’t have enough registers

  • kraid/ssa: Add a concept of memory SSA values

  • kraid/ir: Add a MemRef type

  • kraid: Use the builder in lower_copy()

  • kraid: Handle MemRef in lower_copy()

  • kraid/liveness: Track reg and memory occupancy separately

  • kraid/ir: Improve the SSA use/def iterators

  • nak: Move UnionFind to the shared compiler crate

  • kraid: Copy the ssa repair pass from NAK

  • kraid: Add SSAValueAllocator::lookup_by_idx()

  • kraid/parallel_copy: Fix 16-bit swaps

  • kraid/parallel_copy: Support memory copies

  • kraid/ra: Plumb through a new Arena struct

  • kraid/ra: Fix v9-14 32 register mode

  • kraid/ra: Use 32-register mode on v9-v14

  • kraid/ra: Check with the Arena before allocating an SSA value

  • kraid/ra: Rework parallel copies

  • kraid/ra: Add a memory arena

  • kraid/ra: Consume the liveness

  • kraid/phi: Add a PhiWordSet struct

  • kraid: Add a spiller

  • kraid: Implement shader_clock

  • kraid/nir: Respect rounding modes on [ui]2f32

  • kraid: Add OpFExp16 and OpFLog16 and plumb them through

  • bifrost/nir: Lower fpow16 to fexp2+flog2

  • bifrost/nir: Don’t lower 16-bit flog2/fexp2/fpow for Kraid

  • kraid: Add OpCube* and plumb them through

  • compiler/rust/float16: Implement Display and Debug

  • compiler/rust/float16: Implement Sub

  • kraid/hw_tests: Move the float compare code and add f16 support

  • kraid/hw_tests: Add support for float ops in foldable_op()

  • kraid: Implement Foldable and add tests for a couple cube ops

  • bifrost/nir: Lower fexp2/flog2/fpow to scalar

  • kraid: Pass scale to FEXP.f32 instead of arg

  • kraid: Add a Src::zero() helper

  • kraid: Use Src:fneg_zero() in swizzle lowering

  • kraid: Explicitly specify size on immediates converted to Src

  • kraid: Implement From<u8> and From<u16> for Src

  • kraid: Use Src::from() instead of Src::imm_u8/16()

  • kraid: Implement derivatives

  • rusticl: Use unsafe(no_mangle)

  • rusticl: Stop taking references to temporaries

  • compiler/rust/cfg: Remap node edges in remove_unreachable()

  • pan: Rework PAN_USE_KRAID

  • pan: Add a pan_get_compiler_flags() helper

  • panvk: Pass pan_get_compiler_flags() to the disk cache

  • panfrost: Use pan_get_compiler_flags()

  • nir: Add load_shared_base_ptr to divergence analysis

  • compiler/rust/nir: Implement Send+Sync for nir_shader_compiler_options

  • kraid: Use nir_shader_compiler_options directly

  • kraid/legalize: Add a LegalizeFAU::new() helper

  • kraid/legalize: Legalize 64-bit loads of high-word FAUs

  • kraid/nir: Handle 8 and 16-bit load_push_constant

  • kraid/v9: FADD_IMM doesn’t support source modifiers

  • kraid/v9: Apply i_as_u() to OpIMul::dst_type

  • kraid: Add ICMP.v4[su]8

  • kraid: Allow MESA_SHADER_KERNEL for barriers

  • pan/bi: Promote UBO loads to FAU in NIR for Kraid

  • pan/bi/nir: Scalarize the rest of nir_op_b(any|all)_*

  • pan/bi/nir: FLOG/FEXP.f16 don’t exist prior to v12

  • pan/bi: Use kraid for OpenCL kernels when PAN_USE_KRAID=cs

  • kraid/isa: Add execution units to InstructionInfo

  • kraid/model: Add an exececution unit query

  • kraid/model: Add a way to get at the pan_model

  • kraid/ir: Add is_tls flags to OpLoad and OpStore to track scratch

  • pan/bi: Expose va_compute_alu_bound()

  • compiler/rust/bitset: Add a BitSet::len() method

  • kraid/v9: Fix up MOV.v2i16

  • kraid: Report shader statistics

  • nir,pan/bi: Add a new load_attr_pan intrinsic

  • pan/bi: Lower VS inputs in NIR

  • pan/bi: Drop the old load_input support

  • panvk: Drop lower_load_vs_input

  • nir,pan/bi: Dead code nir_intrinsic_load_attribute_pan

  • kraid/stats: Don’t count system FAU in uniforms

  • pan/nir_opt_push_ubo: Add an add_ubo_push() helper

  • pan/nir_opt_push_ubo: Record FAU indices as we pick push words

  • pan/nir_opt_push_ubo: Re-order pushed UBO entries

  • kraid/nir: Refactor get_alu_src()

  • kraid/nir: Extract immediate offsets when we can

  • kraid/ops: Fix printing of CSEL

  • kraid/nir: Emit OpIMul signless

  • kraid: Lower mediump I/O

  • kraid: Mark reconvergence

  • kraid: Add jump threading and fall-through optimization passes

  • kraid: Implement Debug for LiveSet

  • kraid: Improve spill assertions

  • kraid: Use a more precise register pressure calculation for OpCopy

  • kraid: Run DCE after SSA repair

  • kraid: Add OpMMul* and plumb them through

  • kraid/isa: Add full rename capability to meta enums

  • kraid/isa: Separate out narrowing DstLanes

  • kraid/validate: Add SSA values to ssa_vals earlier

  • kraid/ra: Add a fold_lanes() helper

  • kraid: Add f16 narrowing DstLanes modifiers

  • kraid: Add some FClamp helpers

  • kraid: Make pass!() pass on the return value

  • kraid: Add a destination modifier propagation pass

  • kraid: Stop double-storing immediates in opt_promote_constants()

  • kraid: Use early continues in opt_promote_constants()

  • kraid: Don’t promote supported immediates to FAU

  • meson,ci: Bump the minimum cbindgen version to 0.28.0

  • nak: Use unsafe(no_mangle)

  • nil: Use unsafe(no_mangle)

  • pan/bi: Calculate spills/fills in compute_spill_cost()

  • kraid/stats: Fix spill cost

  • nir,pan: Add a frame_arg_pan system value

  • panvk: Lower view_index and layer_id to frame_arg_pan

  • pan/bi: Drop support for load_layer_id

  • kraid: Drop support for load_layer_id

  • drm-uapi: Add NOUVEAU_FIFO_ENGINE_NVDEC

  • nouveau/winsys: Add video context support

  • nvk/nvkmd: Plumb through video decode contexts

  • nvk/video: Compute H.264 session layouts

  • nvk: Stop using enum nil_sample_layout

  • nvk: Add initial rust support

  • nouveau/push: Expose push_mthd_bits()

  • nvk/rust: Add an nvk_cmd_push() helper

  • nouveau/parser: Add a method_array_size() helper

  • nouveau/parser: Generate array methods for luma/chroma offsets

  • nil: Use GOB_2 tiling for video images

  • kraid/v9: Disallow LD_PKA.b1/b3

  • kraid/hw_tests: Add tests for byte LOAD

  • kraid/ra: Don’t unnecessarily over-write source byes

  • kraid/ra: Add a helper for getting a source alignment

  • kraid: Add a SSAValue indexed vector struct

  • kraid/ra: Use max_live.reg as the arena limit

  • kraid/ra: Add an alloc_regs helper function

  • compiler/rust/enum_as_u8: Implement some ops for U8EnumSet

  • kraid/ra: Add an is_aligned_pinned_range() helper

  • kraid/ra: Rework widening and over-allocation

  • kraid/ra: Rework choose_*_bytes()

  • kraid/ra: Use round robin allocation for registers

  • compiler/rust: Break the guts of UnionFind into a helper struct

  • compiler/rust: Implement IntoIterator for UnionFind

  • kraid/ra: Try to place SSAValues in phi webs in the same register

  • kraid/ra: Try to pre-collect vectors

  • kraid/ra: Use a loop continue for special RA cases

  • kraid/ra: Try to coalesce MKVEC when we can

  • kraid: Lower repeated phi sources before we RA

  • kraid/ra: Add an affinity map builder

  • kraid/ra: Don’t add live-out phi sources to the web

  • kraid: Include NOP.end in the postlude

  • kraid/spill: Use SSAValueIndexedVec for SSADistMap

  • kraid/liveness: Stop filtering out memory in calc_max_live_bytes()

  • kraid/liveness: Calculate max live bytes up-front

  • kraid: OpLea* have no effect on memory

  • kraid: Add a MemAccess::Const enum

  • kraid: Add a MemoryEffect::ConstRead enum

  • kraid/liveness: Add a LiveSet::insert_instr_bottom_up() helper

  • kraid: Add a pre-RA pressure scheduler

  • kraid/ra: Upgrade a comment to a doc comment

  • kraid/ra: Relax an assert in iter_ssa_bytes()

  • kraid/ra: Move iter_ssa_bytes() to SSARef::iter_zip_bytes()

  • kraid/ra: Use iter_zip_bytes() for OpRegOut affinities

  • kraid: Add an SSARef::iter_mut_zip_bytes() helper

  • kraid/ra: Handle mem arenas when building reg affinities

  • kraid/ra: Implement register affinities

  • kraid/ra: Use bytes_are_unpinned() instead of !bytes_are_pinned()

  • kraid/ra: Remove a couple unused pinning helpers

  • kraid/ra: Add a PinnedByteSet struct

  • kraid/ra: Pass the PinnedByteSet into choose_*_bytes() helpers

  • kraid/ra: Give GlobalRegAlloc its own pinned set

  • kraid/ra: Assign registers in 2 passes

  • kraid/ra: Allow sources and destinations to alias

  • kraid/schedule: Handle LD_VAR hidden register dependencies

  • kraid/schedule: Handle discard dependancies

  • compiler/rust/bitset: Add ConstBitSet::remove_range()

  • compiler/rust/bitset: Add ConstBitSet::contains_any_in_range()

  • kraid/ra: Implement clobber registers for OpBlendCall

  • kraid/ra: Propagate OpRegIn for OOB preload registers

  • kraid/copy_prop: Ensure the types match for source modifiers

  • kraid: Better document SSARef and SSAValue

  • kraid: Better document liveness analysis

  • kraid/liveness: Get rid of the traits

  • kraid/liveness: Drop the def_block_ip

  • kraid/bitview: Allow dead code

  • kraid: Delete some dead code

  • kraid: Set allow(dead_code) on a bunch of enum types

  • kraid: Allow a few dead methods

  • kraid/isa: Allow certain dead fields

  • kraid: Warn on dead code

  • kraid: Make the pan_model optional

  • kraid: Add OpV2F32ToV2F16 and plumb it through

  • kraid: Don’t assume 64-bit size for fwrite()

  • kraid: Don’t build HW tests on Android

  • ci: Use rust_ld in cross files instead of setting -C linker= ourselves

  • ci: Set bindgen_clang_arguments in the cross file

  • ci: Add cross-build Rust targets

  • ci: Enable Rust in the debian-x86_32 container

  • ci: Enable Kraid in cross builds

  • meson: Add a with_any_rust helper var

  • util,nouveau: Move bitview.rs to src/util/rust

  • kraid: Use the new mesa_util crate

  • kraid/proc/decoder: Fix up some indentation

  • kraid/disasm: Add a typedef for SourceEncoding[64]

  • kraid: Lower OpCopy to IADD whenever possible

  • kraid/ra: Get rid of the trivial RA

  • kraid/ra: Record whether or not a source is killed up-front

  • kraid: Set last_use on killed sources

  • kraid: Stop forcing dst.lanes in HW tests

  • pan/nir: Stop lowering inot for Kraid

  • kraid: Stop claiming indirect input support

  • kraid: Properly handle OpCopy of 64-bit FAU pre-v12

  • kraid: Set lower_hadd64

  • kraid/legalize: Small constants are never .w1

  • kraid/legalize: 64-bit sources always use 64-bit FAUs

  • pan/nir: Rework b2i/b2f lowering

  • kraid: Use OpIAdd::srcs[1] for swizzle lowering

  • kraid/ra: Prefer not moving source registers for vectors

  • meson: Add a with_any_panfrost helper var

  • meson,panfrost: Allow kraid for tools-only builds

  • kraid/isa: Move SourceEncoding into isa::v9::decode

  • meson,panfrost: Make building disassemblers optional

  • kraid: Fix sub-reg temporaries in legalize_src_swizzles()

  • nvk/queue: Split nvk_queue_state into SLM state and heap state

  • nvk/queue: Split heap binding into sampler and image helpers

  • nvk/queue: Add an nvk_queue_exec_cmd_buffer() helper

  • nvk: Drop the NOP from BeginCommandBuffer()

  • panfrost: Use full texture handles instead of offsets on v9+

  • panvk: Use full texture handles instead of offsets on v9+

  • pan/nir: Don’t use narrow_array_index with wide_indices

  • pan/nir: Only support bindless handles on v9+

  • nir/serialize: Allow more texops

  • nir: Add texop_dual_pan

  • kraid: Print the “.” before TEX write masks

  • kraid: Allow DataType::SR for destinations

  • kraid: Wire up TEX_DUAL

  • pan/nir: Try to lower to TEX_DUAL when using Kraid

  • nir: Use a switch for nir_tex_instr_src_size()

  • nir: Add nir_tex_src_var_(handle|buf_offset)_pan for panfrost

  • kraid: Don’t FAU-legalize OpBlendCall::secondary_color

  • kraid: Fix FAU legalization for ATEST and BLEND

  • kraid: Hook up VAR_TEX_*

  • pan/nir: Try to optimize to VAR_TEX_* for Kraid

  • panvk: Lower mediump on shader/function_temp immediately with Kraid

  • kopper: Don’t call swrast_update_tex_buffer() if we don’t have swrast

  • nir/lower_goto_ifs: Sort sets in debug prints

  • nir/lower_goto_ifs: Add some additional validation

  • util/set: Make some parameters const

  • util/set: Add a couple set union helpers

  • nir/lower_goto_ifs: Use the new set helpers

  • nir/lower_goto_ifs: Expand/reformat some of the debug prints

  • nir/lower_goto_ifs: Track outside blocks separately from break

  • nir/lower_multiview: Check variable modes before deleting

Felix DeGrood (13):

  • util/u_debug: add unsigned_option

  • brw: framework for selecting custom max register count

  • brw: add debug INTEL_THREADS_PER_EU_*

  • anv/rt: separate copies from bvh update

  • anv/rt: remove hashing of global_id

  • anv/rt: infrastructure for batching bvh encodes

  • anv/rt: add start_node_offset to encode.comp

  • anv/rt: enable multi-BVH encode

  • anv/rt: order bvh execution by node count, longest to shortest

  • anv/rt: split encode into multiple batched, breadth-first dispatches

  • jay/debug: allow fixing minimum vrt level

  • brw: simplify vrt fixing code

  • intel/measure: early exit if nothing to write

Filip Gawin (42):

  • nv30: fix 1 << 31 issues

  • nv30: fix another left shift cannot be represented in type ‘int’

  • r300: fixes for UB around left shifts

  • nv30: give more control to draw module for drawing vertex shaders

  • nv30: reorder binding steps to match draw module intended behavior

  • nv30: remove old results of my no longer working card

  • nv30: add new results for nv40 gpu

  • nv30: add CI results for NV34

  • nv30: add hacky, but working support for shader-db

  • nv30: create separable compiler configs for nv30/nv40 fs/vs

  • nv30: fix some code formatting in nvfx_fragprog.c

  • nv30: use lower_fceil

  • nv30: don’t use lower_flrp32 on nv30 fs

  • nv30: don’t use lower_fpow on nv30 fs

  • nv30: use lower_fsign on fs

  • nv30: don’t call lower_sat on nv40 vs

  • nv30: add todo comments

  • r300: VP convert error log for Saturate on r300/r400 into assert

  • r300: improve error printing in vertprog

  • r300: improve error printing in r500_fragprog_emit.c

  • amd/common: BITFIELD_MASK -> nir_component_mask

  • r300: BITFIELD_MASK -> nir_component_mask

  • nv30/ci: add notes about failing tests

  • nv30/ci: add notes about piglit tests

  • glapi: fix call to function compar_nr through pointer to incorrect function type

  • nv30: add piglit fails for nv40 (doing copying only on cpu)

  • nouveau/drm: fix UB applying non-zero offset 40 to null pointer

  • nv30/ci: update fails/flakes

  • nv50: fixes for left shifts (picked with ubsan)

  • nv30: add noticed flake

  • nv30/ci: add nv40 KHR results

  • nv30/ci: remove tests fixed by 4d02806659f20c17c1f96e17547fa789a2afed3c

  • nv30/ci: add recent regression

  • nv30/ci: update nv34 results

  • nv30/ci: add missing “,Fail” to nv43’s fails file

  • nv30: add missed fdot_replicates flag

  • nv30/ci: add flakes to fails file

  • nv30: port fix from nv30_validate_fb to nv40 blit copy

  • nv30: advertise blend_equation_separate and npot_textures on nv30 to get gles 2.0

  • nv30/ci: update nv43 piglit results

  • nv30/ci: remove duplicated line in nv43-fails

  • nv30: enforce that nv40’s blit is operation into vram domain

Francisco Jerez (6):

  • intel/brw/xe2+: Adjust render target cost factor in perf model to match observed behavior.

  • util/ra: Inform register selection callback whether the node is being colored optimistically.

  • brw/ra/xe3+: Implement custom RA select heuristic to make post-RA scheduling more effective without hurting thread parallelism.

  • intel/brw/xe3+: Consider trivial scheduling heuristic in pre-RA scheduling pass.

  • intel/compiler/xe2-3: Lower misaligned LSC access to avoid slow hardware path.

  • intel/compiler: Enforce restriction that destination hstride must not be zero for null destination.

Frank Binns (32):

  • pvr: rearrange some functions in pvr_arch_border.c

  • pvr: setup all format fields for custom border color entries

  • zink: gate some EXT_descriptor_indexing related code

  • pvr: re-enable VK_KHR_format_feature_flags2

  • pvr: re-enable VK_EXT_texel_buffer_alignment

  • pvr: re-enable VK_KHR_shader_expect_assume

  • pvr: increase maxPerStageDescriptorUniformBuffers value

  • pvr: increase maxImageArrayLayers value

  • pvr: check command buffer is in recording state in vkCmd{Begin,End}Rendering

  • pvr: support VK_FORMAT_FORMAT_A2R10G10B10_{S,U}NORM_PACK32 for vertex attribs

  • zink: enable renderpass tracking for PVR

  • pvr: add missing GPU architecture to device info for some BVNCs

  • pvr: fixup ordering of (has_)usc_itr_parallel_instances in dev info

  • pvr: add BXM-4-64 (36.52.104.182) to list of conformant GPUs

  • pvr/csbgen: remove unused control register fields

  • pvr/csbgen: add enum for CDM work item mode

  • pvr/csbgen: fix definition of PDS_CTRL

  • pvr/csbgen: add some missing feature checks to CR definitions

  • pvr/csbgen: add some missing feature checks to pbestate definitions

  • pvr/csbgen: reorder CDM_CONTEXT_RESUME_BUFFER defines for consistency

  • pvr/csbgen: add Compute Data Master control stream documentation

  • pvr/csbgen: add Control Register documentation

  • pvr/csbgen: add Internal Parameter Format documentation

  • pvr/csbgen: add Low-Latency Scheduling buffer documentation

  • pvr/csbgen: add Pixel Backend state format documentation

  • pvr/csbgen: add Programmable Data Sequencer documentation

  • pvr/csbgen: add Primitive Processing Pipeline state format documentation

  • pvr/csbgen: add Texture state format

  • pvr/csbgen: add Vertex Data Master control stream documentation

  • zink: replace blend and vertex elements state hashes with IDs

  • pvr: fix for job samples being initialised to 0

  • meson: bump required meson version to 1.5

Frank Bouwer (3):

  • pvr: Fix for depth stencil 2d array writes.

  • pvr: fix for multisampling with no attachments

  • pvr: enable feature precise occlusion query

Fyodor Kyslov (3):

  • gl: Add P210 support in virgl

  • ci: Uprev virglrenderer

  • vulkan: Add P210 (2-plane 422 10-bit) format support

Ganesh Belgur Ramachandra (3):

  • amd: add gfx1171

  • amd: import addrlib v10.9

  • amd/ci: update navi10 test expectations

Georg Lehmann (118):

  • nir/opt_dead_write_vars: handle atomics as reads

  • nir: return intrinsic in nir_store_reg

  • nir: fix divergence for deref_cast

  • nir/divergence: consider nir_var_mem_constant uniform

  • nir/opt_dead_write_vars: remove some shared writes after barriers

  • util/bitset: add EXTRACT64 helper to extract a uint64_t

  • radv/nir/opt_tid_function: split the pass in two steps

  • radv/nir/opt_tid_function: special case broadcasts

  • radv/nir/opt_tid_function: use max_subgroup_size

  • radv/nir/opt_tid_function: add helper to create ballot_mask

  • radv/nir/opt_tid_function: use ballot helper for bcsel

  • radv/nir/opt_tid_function: make reads_zero a bitset

  • radv/nir/opt_tid_function: rename fotid_context to shuffle_info

  • radv/nir/opt_tid_function: globally track which invocations of ssa defs are needed

  • radv/nir/opt_tid_function: clean up dpp16_row_shift opt

  • radv/nir/opt_tid_function: const fold partially used inverse_ballot

  • nir/opt_uub: optimize iand(iadd(a, b), #c) when num_lsb_zero(a) >= num_lsb(c)

  • nir/opt_uub: optimize shr(iadd(src0, src1), #c) when src1 is fully shifted away

  • radv/nir/opt_tid_function: fix pass description

  • radv/nir/opt_tid_function: allow all power of two x/y wg dims to be mapped to tid

  • radv/nir/opt_tid_function: optimize (local_index & #c) even with large workgroups

  • aco/live_var_analysis: make sure shared vgprs are within the encodable vgprs

  • aco/lower_to_hw: assert that we have space for shared vgprs

  • nir/unsigned_upper_bound: fix float to int conversions

  • nir: mark some AMD specific shuffles as subgroup ops

  • nir: add amd dpp8 and permlanex16 intrinsics

  • aco: implement permute_x16 and dpp8_swizzle

  • radv/nir/opt_tid_function: add option to create standard quad ops

  • radv/nir/opt_tid_function: try to use amd permute intrinsics

  • aco: do not require wqm for mbcnt

  • aco/optimizer: optimize smem after copy propagation

  • aco/optimizer: fix skip_smem_offset_align

  • aco/tests: test skip_smem_offset_align

  • nir: support phi sources in nir_rematerialize_deref_in_use_blocks

  • nir/to_lcssa: fix progress for derefs

  • nir/to_lcssa: move constants before the loop instead of creating a phi

  • nir/print: skip some more default cases of intrinsic indices to reduce visual noise

  • nir/opt_algebraic: optimize open coded ixor

  • nir/algebraic: don’t leak pattern conditions across passes

  • nir/instr_set: add callback to allow additional intrinsics

  • nir/instr_set: only allow demote/terminate for nir_opt_cse

  • nir/opt_cse: CSE subgroup ops per block

  • nir/opt_large_constants: support stores with partial write masks better

  • aco/optimizer: try to invert cndmask condition for more DPP, VOPD and smaller code size

  • aco/optimizer: support inverting more boolean ops

  • aco/tests: test cndmask condition invert

  • radv: disable nir_opt_reassociate_for_fma for VS if it’s not the last pre FS shader

  • aco/sched_vopd: create v_dual_and_b32 from v_bfe_u32

  • aco/sched_vopd: create v_dual_and_b32 from v_cvt_u32_u16

  • spirv: use EXT suffix for promoted NV_cmat2 opcodes

  • docs: add missing VK_EXT_cooperative_matrix_maintenance1 for radv, lvp

  • vc4: move clipping fail to flake

  • vulkan/util: fix strict aliasing violation in pNext iterator

  • vulkan/util: pass sType directly to vk_structure_type_size

  • vulkan: fix strict aliasing violations with vk_foreach_struct

  • nir/opt_shared_vars_to_subgroups: prepare for multiple subpasses

  • nir/opt_shared_vars_to_subgroup: add option to optimize shared access to shuffles

  • radv: optimize shared vars to shuffle

  • nir/opt_shared_vars_to_subgroup: detect loads that stay within subgroups

  • radv: fix lower_array_layer_round_even condition

  • amd/common: update TRUNC_COORD documentation

  • nir: fix strict aliasing violation in nir_def_instr_nonconst

  • nir/opt_algebraic: remove some redundant iands after explicit io lowering

  • nir: add a pass to optimize scalar arrays to vec arrays

  • radv: use nir_opt_scalar_array_vars_to_vec for shared

  • radv: use nir_opt_scalar_array_vars_to_vec for temps

  • nir/opt_uub: fix imax optimization

  • nir/opt_vectorize: use nir_scalar_resolved

  • nir/opt_barycentric: use nir_scalar_resolved

  • pan/nir/lower_tex: use nir_scalar_resolved

  • ir3: use nir_scalar_resolved

  • jay: use nir_scalar_resolved

  • ac/nir: use nir_scalar_resolved

  • nir: always resolve movs in nir_scalar_chase_alu_src

  • nir/gather_tcs_info: stop manually chasing scalars

  • nir/opt_offset: stop manually chasing scalars

  • ac/nir/fixup_smem_loads_null_prt: stop manually chasing scalars

  • radv: keep dcc compressed in GENERAL layout on GFX8-9

  • radv: remove TODOs for storage images with DCC

  • nir: optimize large constants in all functions

  • aco: avoid mubuf bounds checking for global

  • radv: enable emulated RT and fp16 by default for DOOM: The Dark Ages

  • radv/nir/lower_descriptors: fix disabling border colors for gather

  • radv/gfx10: work around broken descriptor type check

  • radv/ci: don’t skip test_undefined_descriptor_heap_mismatch_types on navi10

  • nir: reduce regs inserted in nir_lower_ssa_defs_to_regs_block

  • nir: add nir_fmin_imm and nir_fmax_imm helpers

  • nir: rework nir_tanh_emulated for RTZ correctness, and performance

  • radv: fix DB_SHADER_CONTROL tracking with mixed epilog/no-epilog draws

  • aco/sched_vopd: work around GFX11 VOPD interlock hazard

  • radv/ci: update restricted checksum for DB_SHADER_CONTROL fix

  • radv: use correct enum name for sampler MIP_FILTER

  • radeonsi: use correct enum name for sampler MIP_FILTER

  • aco/optimizer: do not truncate constants before resolving extracts

  • aco/optimizer: create more 16bit integer mad from adds

  • aco/optimizer: create 16bit integer mad from subs

  • aco/optimizer: create alignbyte from (a >> (b << 3))

  • aco/optimizer: apply neg/abs to trunc/rndne/floor/ceil/sin

  • aco/opcodes: add dest/src types for VOPD opcodes

  • aco/insert_fp_mode: handle both vopd opcodes

  • jay/nir_lower_fsign: use u2u instead of i2i for downcasts

  • aco/isel: emit split vector for multi component phis

  • vtn: add an option to force exact GLSLstd450Fma

  • radv: add an option to force exact GLSLstd450Fma

  • radv: force exact GLSLstd450Fma for No Man Sky

  • jay: optimize unsigned extract followed by i2f32

  • aco: fix constant bus violation for scaled transcendental op

  • nir: add uadd/usub opcodes that provide both the result and the carry bit

  • nir: add nir_opt_uadd_carry

  • aco/optimizer: copy propagate ine(b2i(a))

  • amd: switch to nir_opt_uadd_carry

  • pco: switch to combined add+carry opcodes

  • vc4: switch to combined add+carry opcodes

  • broadcom/compiler: switch to combined add+carry opcodes

  • vc4: set lower_int64_options to not support any 64bit alu

  • v3dv: set lower_int64_options to not support any 64bit alu

  • radv: expose fp16 on gfx8 for vkd3d-proton

  • aco/isel: fix sgpr sub dword nir_op_vec with undefs

Gert Wollny (11):

  • r600/sfn: lower BUF txf in nir when on PRE-EG

  • r600/sfn: replace fadd_rtne with fadd

  • r600: Convert TGSI selectors to NIR earlier

  • r600: Cache NIR selector info

  • r600: Cache fragment color output count

  • r600: Cache viewport state on shader selectors

  • r600/sfn: Stop updating TGSI shader info

  • r600: Cache LDS output masks from NIR

  • r600: Drop NIR to TGSI shader info scan

  • r600: Switch interpolation to glsl_interp_mode

  • r600/sfn: get rid of useless indirect_file tracking

Glenn Kennard (1):

  • nv30: Fix for loops in fragment shaders on nv40

Gurchetan Singh (25):

  • gfxstream: fix Fuchsia build

  • gfxstream: add build system TOML files

  • virtio: magma-gpu-rs: add android to supported systems

  • virtio/magma-gpu: continue with the renaming

  • magma-gpu-rs: bump buffer size

  • virtgpu_kumquat: define BLOB args

  • magma-gpu-rs: sync the tube implementation

  • gfxstream: hermetic: disable mallinfo2 for non-Android builds

  • gfxstream: add VK_EXT_HEADLESS_SURFACE_EXTENSION_NAME

  • intel: vulkan: remove unused <xf86drm.h>

  • intel: use align_calloc and align_free

  • anv: use u_memstream helper

  • anv: add dep_regex to anv_deps

  • anv: add missing regerror stub

  • anv: use os_mkdir and os_time_sleep in anv_util.c

  • intel/mda: use os_mkdir

  • intel/common: use simple_mtx_t in intel_aux_map

  • anv: use simple_mtx_t instead of pthread_mutex_t

  • anv/tests: use simple_mtx_t for device mutex

  • anv: use disk_cache_get_function_identifier

  • anv: remove unused anv_device_wait and anv_gem_wait

  • intel/dev: use strdup instead of strndup in scan_for_force_probe

  • gfxstream: delete duplicate functions

  • intel/dev: define INTEL_DEVICE_MAX_EU_MASKS and INTEL_DEVICE_MAX_SUBSLICE_MASKS

  • intel/dev: refactor update topology logic

Gustavo Noronha Silva (1):

  • venus: don’t run a thread’s TLS teardown from an unloaded driver

Gyeyoung Baek (10):

  • nir/coopmat: use nir_shader_get_entrypoint()

  • drm-shim: remove the BO’s `offset_map` entry on free

  • pvr/drm-shim: fix BO refcount leak

  • panfrost/drm-shim: fix BO refcount leak

  • lima/drm-shim: fix BO refcount leak

  • radeon/drm-shim: fix BO refcount leak

  • amdgpu/drm-shim: fix BO refcount leak

  • intel/drm-shim: fix BO refcount leak

  • drm-shim: fix inverted BO refcounting

  • pvr: fix sampler handle dereference with immutable samplers

HTRMC (1):

  • v3dv: keep descriptors aligned after inline uniform blocks

Hans-Kristian Arntzen (2):

  • radv: Prefer STRUCTURED OOB over STRUCTURED_OFFSET for texel buffers.

  • tu: Fix single texel alignment for non-POT texel buffers

Harri Nieminen (2):

  • broadcom/ci: remove nonexistent tests

  • r300/ci: remove nonexistent tests

Helen Koike (1):

  • vulkan/overlay: Draw on the present queue when possible

Herman Semenoff (6):

  • draw: avoid dirtying cache lines during vertex id reset

  • llvmpipe: optimize tile bin iteration using atomic fetch-and-add

  • radeonsi: eliminate redundant SWITCH_ON_EOI and wave splitting for single-SE GFX6

  • amd/common: use 256 threads per TCS workgroup on all GFX generations

  • winsys/amdgpu: cap bo_cache size and reduce size tolerance on dedicated VRAM GPUs

  • intel/i915: fix I915_PARAM_HAS_CONTEXT_FREQ_HINT check and low latency param value

Hugo223 (3):

  • turnip: add drirc option for DEVICE_LOCAL-only memory type

  • turnip: enable DEVICE_LOCAL-only memory type for RDR2

  • freedreno: Add support for Adreno 710 and 720

Hyunjun Ko (70):

  • intel/genxml: Update AVP commands for gen125

  • vulkan/video: Add a AV1 frame header packer

  • anv/video: Add av1 vdenc tables

  • anv/video: Fix to avoid null-pointer reference.

  • anv/video: Implement AV1 encoding on Gen125

  • anv/video: Enable KHR_video_encode_av1 on Gen125

  • anv/video: Add missing scratch buffers for H.265 encoding

  • anv/video: Handle low delay B frames for H.265 encoding

  • anv/video: Fix H.265 encode reference picture lists

  • anv/video: Fix to use proper QP for H265 encoding

  • anv/video: Fix to set proper values to missing attributes for H265 encoding

  • intel/common: Add to check whether HuC firmware is loaded.

  • intel/genxml: Set default length for HuC cmds

  • intel/genxml: Add MI_CONDITIONAL_BATCH_BUFFER_END to gen120

  • anv/video: Add HuC S2L helpers for H.265 decoding

  • anv/video: Implement H265 decoding with using HuC

  • anv: Enable H.265 decoding only when HuC firmware is loaded

  • intel/ci: Update relevant tests.

  • anv: Fix to fill the whole AVC FQM matrices

  • vk_video: Fix H.264 slice header bitstream packing

  • anv: Clamp the first H.264 slice first_mb_in_slice to 0

  • anv: Build MFX_AVC_REF_IDX_STATE from the slice reference lists

  • anv: Fill the inter mode costs for H.264 VDENC

  • anv: Fix H.264 B frame PAK programming

  • anv: Advertise maxL1ReferenceCount for H.264 encode

  • anv/video: Fix to update probabilities for VP9 decoding

  • anv/video: Make VP9 key/intra frame cmdbufs self-contained for prob reset

  • anv/video: Use fixed-role VP9 MV temporal buffers

  • anv/video: Derive VP9 UsePrevinFindMVReferences from the app flag

  • anv/video: Resolve VP9 LastFrameType at execution time

  • intel/ci: Update improvements.

  • intel/genxml: Rename HCP/HuC address fields to the general naming convention

  • anv: Macro out the vid_mem address/attributes boilerplate

  • anv/video: Share AV1 rowstore on-chip cache offsets between decoding and encoding

  • anv/video: Share the AV1 relative_dist helper between decoding and encoding

  • anv/video: Deduplicate the AV1 encode superblock and tile geometry

  • anv/video: Fix AV1 decode tile start array overrun

  • anv/video: Use the AV1 superblock size for the encode tile slice state

  • anv/video: Set DisableCDFUpdateFlag for AV1 encoding

  • anv/video: Fix HCP reference mapping for streams with more than 8 DPB refs

  • anv/video: Fix the MFX_AVC_REF_IDX_STATE entry encoding for H.264 encode

  • anv/video: Use the H.264 colocated MV buffers for VDENC B frames

  • anv/video: Fix to set OBU extension header info for AV1 encoding

  • anv/video: Fix size_limit calculation in GetEncodedVideoSessionParameters

  • anv/video: Increase the size of frame header

  • anv/video: Move all h264/5 tables to header files.

  • anv/video: Macro out the vdenc PictureFields boilerplate

  • anv/video: Split H.264 encode state emission into helpers

  • anv/video: Split HEVC encode state emission into helpers

  • intel/genxml: Fix the VDENC_PIPE_BUF_ADDR_STATE tail dwords on gen120

  • anv/video: Implement VkUpdateVideoSessionParametersKHR

  • anv/image: Allocate downscaled reference surfaces for video encode

  • anv/video: Use VDENC downscaled references for video encode

  • anv/video: Add more video session buffers for H265 encoding

  • anv/video: Override HEVC SPS max_transform_hierarchy_depth to 2

  • anv/video: Change uint8_t to int32_t for POC in HEVC encoding

  • anv/video: update VDENC cmd1/cmd2 for HEVC encoding

  • anv/video: Make rate control mode selection explicit in encode paths

  • anv/video: Store encode rate control state

  • anv/video: Add shared BRC session buffers for video encode

  • anv/video: Add HuC BRC DMEM and constant data definitions for H264 encode

  • anv/video: Implement the HuC BRC building blocks for H264 encode

  • anv/video: Enable the HuC BRC path for H.264 encode

  • anv/video: Add HuC BRC DMEM and constant data definitions for HEVC encode

  • anv/video: Implement the HuC BRC building blocks for HEVC

  • anv/video: Enable the HuC BRC multipass path for HEVC encode

  • anv/video: Expose CBR and VBR rate control for HEVC encode

  • anv/video: Move common BRC tables to new C file.

  • anv/video: Implement a helper to get the base address with specific alignment.

  • anv/video: Use helpers to set bitstream buffer base address/offsets

Iago Toral Quiroga (2):

  • panfrost: move a couple of passes from variant to shader creation time

  • v3dv: Fix early-ez on v71

Ian Forbes (1):

  • svga: Implement GL_ARB_shader_texture_image_samples

Ian Romanick (25):

  • brw: Ignore source modifiers in needs_dst_copy

  • brw: Handle empty top block in brw_nir_move_interpolation_to_top

  • brw/xe2: Use accumulator for SIMD16 USUB_SAT / ISUB_SAT lowering

  • brw: MAC has the same commutivity rules as MUL

  • brw/validate: Gfx11 can’t have accumulator src0 in 3-src instructions

  • brw: Enforce undocumented Gfx11 and Gfx12 restriction on src mods w/ accumulators

  • brw: Enforce Gfx9 restriction 3-source destination must be GRF

  • brw: Vectorize URB for task and mesh shaders after SIMD is set

  • brw: Refactor a big chunk of brw_opt_fill_and_spill to its own function

  • brw: Try harder at fill-to-fill propagation

  • nir/divergence: Handle SYSTEM_VALUE_DEVICE_INDEX

  • nir/divergence: Handle SYSTEM_VALUE_BARYCENTRIC_*

  • brw: Only HALT for terminate_if of non-helper pixels

  • brw: Use simpler instruction to clear flags for unconditional terminate/demote

  • brw: Start using the accumulator

  • nir/algebraic: Don’t generate 64-bit iadd3 when not supported

  • brw/acc: Fix up source modifiers on CSEL for Gfx11 and Gfx12

  • brw: Fix regions_overlap for accumulators

  • brw/mac: Don’t try to read an accumulator that cannot be implicitly written

  • brw: Handle multiple accumulators in scheduling

  • brw/acc: Use all available accumulators

  • brw/cmod: Teach cmod propagation about accumulators

  • brw: CSEL cannot mix type sizes even for immediates

  • brw: CSEL cannot have immediate src1

  • brw: Fix 16-bit type handling in constant combining

Icenowy Zheng (13):

  • docs/features: track VK_IMG_filter_linear_2d

  • pvr: add support for VK_IMG_filter_linear_2d

  • pco: fix ra emit_copies sidecases

  • pvr: set ycbcr sampler state words when any sampler word is present

  • pvr/ci: remove 3 plane YCbCr fails/flakes

  • gallivm: fix enablement of RISC-V Zbs extension

  • gallivm: generate Linux perf profiler information for JITLink

  • vulkan/wsi/headless: add support for KHR_present_{id2,wait2}

  • vulkan/wsi: support allocating more memory for buffers

  • meson: list MIPS(64) in llvm_has_mcjit list

  • llvmpipe: do not call LLVMLinkInMCJIT() when using OrcJIT

  • meson: link in orcjit instead of mcjit when using OrcJIT

  • gallivm: adapt for JITLink in LLVM 23

Igor Torrente (13):

  • panfrost/kmod: Adds `get_param` panfrost specialized function

  • panfrost/panthor: Add `bo_map` operation to kmod.ops

  • panfrost/panthor: Add `bo_export` operation to kmod.ops

  • panfrost/kmod: Add `submit_job` as a panfrost kmod operation

  • pan/kmod: Delegate more of the buffer import logic to backends

  • panfrost: Add prime_fd_to_handle to panfrost kmod ops

  • pan/kmod: Abstract sync operations

  • panfrost: replace the raw `drmSyncobj` calls with u_sync_provider.h

  • panvk/{jm,csf}: replace drmSyncobj operation with u_sync_provider.h

  • panfrost/kmod: Add backend selection

  • panvk: Move device compatibility checks to kmod

  • panfrost: Add VDRM backends

  • gallium: panfrost: Add vdrm driver descriptor

Ivaylo Spasov (4):

  • radv: close imported fd only after memory allocation succeeds

  • radv/android: fix fd leak on gralloc memory import failure

  • radv/android: fail image creation when the gralloc stride override fails

  • radv/android: support deferred swapchain image binding

Iván Briano (15):

  • anv: fix 2d-array to 3d blits

  • nir: add fadd_rtne opcode

  • elk: handle nir_op_fadd_rtne

  • brw: handle nir_op_fadd_rtne

  • jay: handle nir_op_fadd_rtne

  • nak: handle nir_op_fadd_rtne

  • microsoft/compiler: handle nir_op_fadd_rtne

  • nir_to_tgsi: handle nir_op_fadd_rtne

  • gallivm: handle nir_op_fadd_rtne

  • r600/sfn: handle nir_op_fadd_rtne

  • zink: don’t use nir_lower_dround_even

  • nir: use fadd_rtne in lower_round_even

  • anv: let’s not crash everywhere

  • anv: use companion cmdbuffer for MCS enabled blits

  • anv: fix push constant range gathering for dynamic offsets

Jaakko Jokinen (1):

  • nir: Add cases to nir_get_io_offset_src_number()

Jack Boykin (1):

  • radv/video: remove decode tier3

Jade Macho (1):

  • gallium/sw: Fix llvmpipe being chosen when not wanted

JaeHoon Lee (36):

  • v3dv: make room in the descriptor map for the no-sampler entries

  • v3dv: use the binning VS variant for the binning VPM config

  • v3dv: record the multiview geometry shader with its Vulkan stage bit

  • v3dv: record the no-op fragment shader with its Vulkan stage bit

  • nvk: free copy_memory_indirect_temps on command buffer destroy

  • v3dv: avoid restoring stale descriptor state after a meta op

  • v3d: don’t reallocate the occlusion query BO on every begin

  • v3dv: return a VkResult from pipeline_add_multiview_gs()

  • v3dv: propagate failure when getting a stage’s NIR fails

  • nvk: report fills from memory correctly

  • panfrost: Fix copyright header typos

  • v3dv: fix typo in comment (descripot -> descriptor)

  • v3dv: check for a failed default attribute values allocation

  • v3dv: drop the per-version wrapper for the default attribute values check

  • v3d: use the common default attribute values predicate

  • v3dv: release everything CreateDevice allocated when it fails

  • v3dv: free the default attribute values BO when its map fails

  • v3d: initialize the BO cache mutex

  • v3dv: initialize the BO cache mutex

  • v3dv: check for a failed device address context allocation

  • v3dv: propagate a failed executable data context allocation

  • v3dv: drop the unused u_memory.h include from the BO code

  • v3dv: report correct TMU spill and fill counts

  • broadcom/compiler: skip spill class updates without accumulators

  • broadcom/compiler: rebuild liveness once per spill batch

  • broadcom/compiler: drop obsolete V3D 3.x texture uniforms

  • broadcom/qpu: drop obsolete pre-4.0 UNIFA check

  • broadcom/compiler: update liveness incrementally

  • broadcom/compiler: skip duplicate liveness computation

  • v3dv: remove render pass fields from pipeline

  • v3dv: lower color input attachments to load_tile_image

  • v3dv: track color input self-dependencies

  • v3dv: keep CL jobs for color input self-dependencies

  • broadcom/compiler: fix ushr delay in NIR Scheduler

  • v3dv: handle memory barriers in self-dependencies.

  • v3dv: guard TLB input mappings for memory barriers

Jakob Sinclair (31):

  • kraid: Do not assert while encoding flow when slot is set

  • kraid: Optimize end of shaders

  • kraid: Do not encode empty shaders

  • pan/layout: Reserve GPU-visible CRC state

  • panvk: Add image CRC header plumbing

  • pan: Share CRC image-view eligibility

  • pan/fb: Emit CRC framebuffer descriptors

  • panvk: Propagate CRC header to framebuffer stores

  • panvk/csf: Patch crc_init at execution time for v11+

  • panvk/csf: Handle CRC during incremental rendering

  • panvk/csf: Invalidate CRC after non-rendering image changes

  • panvk: Enable CRC on v11-v13

  • panvk/csf: Enable ETE

  • panvk: Enable CRC/ETE on v14

  • panvk/csf: Add v10 CRC validity tracking

  • panvk/csf: Handle v10 CRC invalidation and incremental rendering

  • panvk: Enable CRC/ETE on v10

  • kraid/ir: Fix indentation for phis

  • kraid: Fix misc printing issues

  • kraid: Add max_reg_count function to model

  • kraid: Add MemoryEffect property to ops

  • kraid: Helper for register range instead of bytes

  • kraid: Add scheduling barriers

  • kraid: Implement improved message slot assignment

  • kraid: Add serial debug flag

  • docs/features.txt: fix PanVK arch qualifiers

  • kraid/ir: Pretty print small constants

  • kraid/ir: Pretty print small constants type

  • kraid/ir: Only print sources used for ShiftLop ops

  • kraid/ir: Pretty print all constants

  • kraid/ir: Add debug flag for showing raw constants

Jan Meisel (11):

  • nir/range_analysis: handle msad_4x8 in unsigned upper bound

  • radeonsi/vcn: fail feedback for truncated encodes

  • radv: fix RADV_PERFTEST=nircache enablement

  • radv: drop inlined functions before cleanup passes

  • vulkan/runtime: avoid rehashing shader stage hashes

  • nir/range_analysis: skip caching unsupported UUB ALU ops

  • util/dynarray: outline the growth slow path

  • util/hash_table: add clone_into helper

  • nir: reuse copy propagation hash table storage

  • nir: reuse copy propagation dynarray storage

  • nir/search: reset FP state for commutative match attempts

Janne Grunau (6):

  • asahi: nir: lower flrp64

  • asahi: ci: Drop no longer failing VK.wsi.xcb.present_timing test

  • panfrost: ci: Drop no longer failing VK.wsi.xcb.present_timing test

  • asahi: ci: Add failing b10g11r11 and e5b9g9r9 copy tests

  • hk: xfb: Avoid assertions in nir_slot_num_components

  • poly: Fix comment after moving passthrough_gs

Jason Le (1):

  • gfxstream: decode vkGetImageSubresourceLayout in the global state

Jason Macnak (66):

  • gfxstream: Fixes for extensions with type erased objects

  • Add codegen to generate host constants for supported exts

  • gfxstream: use gfxstream_vk_* for VkBuffer-s

  • gfxstream: use gfxstream_vk_* for VkFence-s

  • gfxstream: use gfxstream_vk_* for VkSemaphore-s

  • gfxstream: use gfxstream_vk_* for VkCommandPool-s

  • gfxstream: use gfxstream_vk_* for VkCommandBuffer-s

  • gfxstream: use gfxstream_vk_* for VkQueue-s

  • gfxstream: use gfxstream_vk_* for VkPhysicalDevice-s

  • gfxstream: use gfxstream_vk_* for VkDevice-s

  • gfxstream: use gfxstream_vk_* for VkInstance-s

  • gfxstream: remove now unused decleration

  • gfxstream: use gfxstream_vk_* for remaining trivial types

  • gfxstream: migrate macros to gfxstream_vk_private

  • gfxstream: migrate host handle accessing to gfxstream_vk_private

  • gfxstream: consolidate to a single object creation func

  • gfxstream: rename struct create and destroy funcs

  • gfxstream: use gfxstream_vk_*_from_handle funcs

  • gfxstream: replace handle mapping with explicit calls

  • gfxstream: remove unused guest marshal_*

  • gfxstream: consolidate declare macro

  • gallivm: truncate before zext

  • gfxstream: exclude dispatch only exts from support list

  • gfxstream: handle vkCmdBindDescriptorSets2 batching

  • gfxstream: populate VkBindMemoryStatus

  • gfxstream: use wire size for pNext_size when marshaling

  • gfxstream: pass through VkBufferUsageFlags2CreateInfo

  • gfxstream: sanitize VkWriteDescriptorSet for push descriptors

  • gfxstream: expand push descriptor template updates

  • gfxstream: support IUB in batched descriptor updates

  • gfxstream: handle null in batched descriptor handling

  • gfxstream: use heap index in device memory report

  • gfxstream: Enable VK_KHR_16bit_storage

  • gfxstream: Enable VK_KHR_relaxed_block_layout

  • gfxstream: Enable VK_KHR_shader_draw_parameters

  • gfxstream: Enable VK_KHR_variable_pointers

  • gfxstream: Enable VK_KHR_8bit_storage

  • gfxstream: Enable VK_KHR_draw_indirect_count

  • gfxstream: Enable VK_KHR_driver_properties

  • gfxstream: Enable VK_KHR_sampler_mirror_clamp_to_edge

  • gfxstream: Enable VK_KHR_separate_depth_stencil_layouts

  • gfxstream: Enable VK_KHR_shader_atomic_int64

  • gfxstream: Enable VK_KHR_shader_float_controls

  • gfxstream: Enable VK_KHR_spirv_1_4

  • gfxstream: Enable VK_KHR_uniform_buffer_standard_layout

  • gfxstream: Enable VK_EXT_descriptor_indexing

  • gfxstream: Enable VK_EXT_host_query_reset

  • gfxstream: Enable VK_EXT_sampler_filter_minmax

  • gfxstream: Enable VK_EXT_separate_stencil_usage

  • gfxstream: Enable VK_EXT_shader_viewport_index_layer

  • gfxstream: Enable VK_KHR_index_type_uint8

  • gfxstream: Enable VK_KHR_load_store_op_none

  • gfxstream: Enable VK_KHR_map_memory2

  • gfxstream: Enable VK_KHR_shader_expect_assume

  • gfxstream: Enable VK_KHR_shader_float_controls2

  • gfxstream: Enable VK_KHR_shader_subgroup_rotate

  • gfxstream: Enable VK_KHR_vertex_attribute_divisor

  • gfxstream: Enable VK_EXT_pipeline_robustness

  • gfxstream: Enable VK_EXT_robustness2

  • gfxstream: Enable VK_KHR_global_priority

  • gfxstream: Enable VK_KHR_maintenance5

  • gfxstream: populate VkAndroidHardwareBufferFormatProperties2ANDROID

  • gfxstream: migrate `ptr` to `gfxstream_vk_device_memory`

  • gfxstream: migrate `allocationSize` to `gfxstream_vk_device_memory`

  • gfxstream: flush before resource destruction

  • gfxstream: fix comment

Jeongik Cha (7):

  • gfxstream: advertise VK_KHR_xlib_surface

  • gfxstream: allow VK_EXT_vertex_attribute_divisor again

  • gfxstream: only query the image row pitch when it is used

  • gfxstream: expose VK_EXT_robustness2 on Linux guests

  • gfxstream: map more formats in getVirglFormat()

  • gfxstream: support non-dedicated dmabuf buffer exports on Linux guests

  • gfxstream: fix DRM format modifier emulation on Linux guests

Jeremy Chou (3):

  • r300: fix -Wint-in-bool-context warning in r300_fs

  • nir: Fix compiler warning

  • brw: Fix incompatible return type in systolic vectorize

Jeremy Gebben (12):

  • kk: Fix Vulkan to Metal stage translation for timestamps

  • kk: Fix up M5 robustness2 behavior when LOD is out of bounds

  • kk: Move settings from kk_device to kk_physical_device

  • kk: Disable allowGPUOptimizedContents on M5 for e5b9g9r9

  • kk: Re-enable workaround 2 for M5 in macOS 27

  • kk: Connect VK_EXT_debug_utils labels to Metal

  • kk: Wrap meta operations in debug groups

  • kk: Handle NULL object names in vkSetDebugUtilsObjectNameEXT()

  • kk: Add missing Vulkan stages to vkCmdWriteTimestamp() handling

  • kk: Fix debug labels on the post-gfx compute encoder

  • kk: Simplify timestamp query state

  • kk: Move the timestamp query stage map

Jeremy Huddleston (6):

  • glx/apple: only skip AppleGL election on macOS 26.0 through 26.5

  • llvmpipe: don’t create a screen when the process is not allowed to JIT

  • glx: add __glXForceEnableExtension()

  • glx/apple: implement swap control on top of kCGLCPSwapInterval

  • CODEOWNERS: Update CODEOWNERS to add myself for src/glx/apple

  • glx/apple: silence OpenGL deprecation warnings in libglx

Jesse Natalie (16):

  • d3d12: Disable vao fast path for AMD

  • mesa: Fix shared state bookkeeping for dynamic share list changes (wglShareLists)

  • wgl: Add driconf for XELive to force GDI compat

  • glapi: Add missing ARB_robustness aliases from GL4.5 promotion

  • d3d12: Stop using D3D12_IGNORE_SDK_LAYERS

  • dzn: Rearrange includes and stop using D3D12_IGNORE_SDK_LAYERS

  • vk/wsi/win32: Stop using D3D12_IGNORE_SDK_LAYERS

  • glsl: Enable default GLSL version overrides via driconf

  • driconf: Add entry for AutoDesk ReCap

  • d3d12: Reference the predication buffer from the batch

  • d3d12: Reference the destination of get_query_result_resource

  • d3d12: Remove promoted bos from the residency LRU

  • d3d12: Don’t re-enqueue a bo when the resident batch fills up

  • d3d12: Never claim residency we did not get

  • d3d12: Stop reclaim retries when the fence wait fails

  • d3d12: Reclaim buffer memory outside manager locks

Jessie Hao (1):

  • vulkan/android: add R16/R16G16 to AHB format equivalence table

Jhanani Thiagarajan (4):

  • driconf: Add device_version_regexp attribute for flexible device matching

  • anv: enable drirc device_regexp

  • anv/intel: plumb atomic-branch control through drirc prog key

  • anv: enable atomic-branch optimization for Mount And Blade

Jianxun Zhang (11):

  • isl: Update formats supporting LSC route

  • isl: Refactor comment of a function

  • iris: Fix an assertion when importing dmabuf

  • anv: Refactor check on CCS modifiers

  • anv: Set dedicated_only feature bit for CCS modifiers

  • intel/gen: Fix invalid data size in Load 2D Block tests

  • intel/gen: Add a restriction on 2D block load instruction

  • blorp: Remove hardcoded ‘gfx8’ from the hiz function name

  • brw: Add a checker on cmask of RT write messages (Xe3p+)

  • brw: Refactor function of emitting fs intrinsic

  • brw: Still setup src0 for alpha to coverage (Xe3p+)

Jiming Liu (1):

  • ac/virtio: Fix VIRTIO_SHARED flag masking

Jiyu Yang (4):

  • st: invalidate shader constants on clip plane state changes

  • clc: fix fd leak in open_clc_data() on map failure

  • pan: fix missing access tracking in batch read/write helpers

  • glsl: only expose tanh since GLSL1.30

Job Noorman (20):

  • ir3: move some helpers up in ir3.h

  • ir3: check all dsts in writes_addr[01]

  • ir3/sched: Handle fake writes to a0.x/a1.x

  • freedreno/computerator: attach perfcntr BO to CS

  • ir3: remove unnecessary ir3_nir_move_varying_inputs

  • ir3: add ir3_get_waves_per_wg helper

  • ir3: add quirk for prefetch sam.s2en needing helpers

  • ir3: interpret const @load/store_global_offset offset as unsigned

  • nir/lower_tex: copy texture/sampler_non_uniform flags

  • ir3: fix warning in release builds

  • ir3/lower_copies: prevent splitting when handling cycles

  • ir3/cp: allow cp’ing absneg into meta instructions

  • ir3/sched: add TODO about scheduling kill before bary.f

  • ir3/sched: don’t try to schedule kill after bary.f in different block

  • ir3: add foreach_main_block helper

  • ir3: fix QCTDD13523866 workaround

  • ir3: remove ir3_delayslots_with_repeat helper

  • ir3: add ir3_instr_prev/next helpers

  • ir3/lower_copies: add validation

  • tu: fix texel buffer emulation bounds check

JoeWang (5):

  • radv: use a compute copy for large GTT/host copies on dGPUs

  • radv: remove the VBO descriptor prefetch

  • radv: allow enabling the SDMA transfer queue per-application

  • amd: fix GFX12 SPM GCEA_SE counter aliasing with GCEA_CPWD

  • radv: derive the PWS ACQUIRE point from the barrier destination stage

Jordan Justen (33):

  • intel/executor: Add hw_regs shader parameter

  • intel/executor: Use hw_regs param to adjust internal/rsvd reg num

  • intel: Move brw_register_blocks to common intel_register_blocks

  • intel/common: Rework intel_register_blocks()

  • intel/executor: Set state for hw_regs param

  • intel/common: Add intel_register_blocks_supported()

  • intel/executor: Allow hw_regs based on intel_register_blocks_supported()

  • intel/executor: Add 256_hw_regs.lua example

  • intel/gen/validate: Allow DG2+ to access large grf registers

  • intel/executor: Support Large GRF mode on DG2 through Xe2

  • intel/dev: Drop 0xd744 PCI ID

  • intel/dev: Enable NVL-S/U PCI IDs (without FORCE_PROBE)

  • brw: Add Xe3P (GFX_VERx10 == 350)

  • anv: Support 64-bit addressing in shader heap

  • brw: Drop debug fprintf accidentally left in dfceddb18554

  • intel/gen: Use null operand params to calculate byte stride

  • brw/copy-prop: Handle Xe3P src1 in copy propagation optimization

  • brw: Update regioning lowering for Xe3P

  • brw: Allow Xe3P in brw_stage_has_packed_dispatch()

  • brw: Add `mov` & `srnd` swsb pipe exception for Xe3P

  • brw: Match dst/src1 type on Xe3P for assign_curb_setup robustness CMP

  • brw/builder: Allow accumulator as src to MAD for Xe3P

  • brw: MAC => MAD for Xe3P in emit_pixel_interpolater_alu_at_offset()

  • brw: Drop mac/mach/macl for Xe3P

  • brw: Add mullh instruction for Xe3P

  • brw: Lower mulh with Xe3P mul instruction

  • brw: Update spill for Xe3P

  • brw: Rely only on !has_integer_dword_mul for brw_lower_mul_dword_inst()

  • intel: Set has_integer_dword_mul in devinfo for Xe3P

  • intel/brw: Skip brw_opt_send_to_send_gather pass for efficient 64-bit mode

  • intel/gen/compact: Fix typo in Xe3P compact tables

  • intel/gen/compact: Fix UNCOMP_3SRC_SOURCE for Xe3P

  • intel: Add intel_force_probe driconf string option

Jose Maria Casanova Crespo (21):

  • broadcom/compiler: compute peak register pressure in O(n)

  • broadcom/compiler: spill register pressure in adaptive batches

  • vc4: save the fragment constant buffer around the YUV blit

  • vc4: unbind the textures around the blitter clears

  • vc4: don’t save the fragment constant buffer for the ZS surface clear

  • v3d: cap CL BO growth when doubling

  • v3dv: cap CL BO growth when doubling

  • broadcom/compiler: reuse spill costs within a spill batch

  • ci/broadcom: Update ci expectations after piglit uprev

  • broadcom/compiler: restrict fsat_signed/fclamp_pos to 32-bit

  • broadcom/compiler: handle 16-bit fsat

  • broadcom: pass simulator arg to build libbroadcom_v3d

  • v3d: flush jobs that queue too many draw calls

  • broadcom/qpu: decode fully the input unpack from mul FMOV

  • broadcom/qpu: decode the fround-group input unpack from raddr_b on V3D 7.1

  • broadcom/compiler: fill the old value when spilling a partial write

  • vc4: helper vc4_get_clip_window() for the clip window computation

  • vc4: skip draws whose clip window is empty

  • broadcom/compiler: keep robust SSBO stores scalar for bounds checking

  • v3dv: pad the shader assembly BO for the QPU instruction prefetch

  • v3d: pad shader uploads for the QPU instruction prefetch

Joseph Kogut (1):

  • radv: query heap usage in one request

José Roberto de Souza (89):

  • anv: Copy anv_reloc_list::uses_relocs when clonning reloc list

  • anv: Fix segfault when binding shaders to anv_cmd_buffer before emit any instruction

  • anv: Return early in anv_reloc_list_append() when relocs are not in use

  • anv: Program of async thread limits when VRT is disabled in runtime

  • iris: Program of async thread limits when VRT is disabled in runtime

  • blorp: Program of async thread limits when VRT is disabled in runtime

  • iris: Add function to compute and emit engine async threads limits

  • intel: Update URB deref_block_size value for GFX 20+

  • anv: Add missing ANV_BO_ALLOC_INTERNAL flags

  • anv: Add some comments around memory types

  • anv: Fix alloc flags for ANV_BO_SLAB_HEAP_LMEM_SMEM

  • anv: Allow bos with ANV_BO_ALLOC_LOCAL_MEM_CPU_VISIBLE to be allocated in slab

  • nir,spirv: Move vtn tanh emulation into nir_tanh_emulated()

  • glsl: Use nir_tanh_emulated()

  • nir: Add native ftanh opcode

  • compiler: Add support to TANH instruction

  • intel/gen: Add GEN_MATH_TANH

  • intel/brw: Add support to TANH instruction in Gfx 35

  • intel/brw: Replace tab by spaces to fix identation on brw_shader::assign_curb_setup()

  • intel/dev: Add intel_device_info_has_64bit_addressing()

  • intel/genxml: Add efficient 64bit instructions

  • intel/gen: Implement efficient64bit support

  • intel/brw: Implement SENDG instruction + LSC store

  • intel/brw/64bits: Add support for writes to BRW_SFID_URB

  • intel/brw/64bits: Add support for reads to BRW_SFID_URB

  • intel/brw/64bits: Implement fence message

  • intel/brw/64bits: Implement sendg barrier

  • intel/brw/64bits: Implement render target write message

  • intel/brw/64bits: Implement sampler message

  • intel/brw/64bits: Add support for bindless dispatch and retire

  • intel/brw/64bits: Add register spill and fill support

  • intel: Add new ‘Thread Group Dispatch Size’ encoding values for Compute Walker 2

  • genxml: prepare for 64bit state emission

  • intel: Switch blorp shader address from uint32 to uint64

  • intel: Add decoder support for efficient 64bit instructions

  • blorp: enable efficient 64bit support

  • anv: shader pool size bump for 4GiB pools

  • anv: Add support for efficient 64bit state emission

  • intel/brw: Fix lsc messages with negative global_offset in efficient 64bit addresing mode

  • iris: Use coherent framebuffer fetch helper in missed places

  • anv: Replace anv_shader_internal_get_pointer() by a more generic anv_shader_get_pointer()

  • intel/blorp: Replace hard-coded value in blorp_get_efficient_64bit_io_size()

  • intel/blorp: Add comment and new variables to better document efficient 64bit io

  • intel/brw: Set the scratch surface binding for efficient 64-bit accesses for nir_intrinsic_store/load_scratch_intel

  • iris: Add enable_efficient_64bit driconf

  • iris: Populate binding tables before emit push constants

  • iris: Add function to pin binder bo

  • iris: Return earlier in binder functions when 64bit addressing mode is enabled

  • iris: Set STATE_BASE_ADDRESS::BaseAddressdisable when 64bit addressing mode is enabled

  • iris: Prepare iris_record_state_size() to handle u64 and u32

  • iris: Don’t set heap base values in batch decoder when efficient 64bit is enabled

  • iris: Add warpers around u_upload_data_ref()/alloc_ref() that takes iris_state_ref as parameters

  • iris: Change stream_state(out_offset) and iris_state_ref::offset to u64

  • iris: Emit new 3D state instructions with 64-bit addresses

  • iris: Set blorp_config::use_efficient_64bit when running in this mode

  • iris: Replace iris_binding_table::size_bytes by total_surf_count

  • iris: Alloc, populate and emit render target binding table in 64bit addressing mode

  • iris: Add ubo_ranges lenght and used push regs to iris_nir_analyze_ubo_ranges()

  • iris: Add use_efficient_64bit to iris_base_prog_key and adjust paddings

  • iris: Set use_efficient_64bit in iris_xs_prog_key

  • iris/64bits: Add support to read data from uniform buffers

  • iris/64bits: Add rt_write_cb to calculate render target surface state address

  • iris: Replace iris_setup_binding_table() intel_device_info parameter by iris_screen

  • iris: Rewrite ubo and ssbo access to u64 surface address in 64bit addressing mode

  • intel/compiler: Skip intel_nir_lower_vec2_surface_sampler when the handle is already a single 64-bit

  • iris/64bits: Add support to populate IRIS_SURFACE_GROUP_IMAGE surfaces to binding table

  • iris/64bits: Add support to populate IRIS_SURFACE_GROUP_TEXTURE_* surfaces to binding table

  • intel/common: Add intel_sampler_state_size()

  • iris/64bits: Add sampler binding tables address to shader thread payload

  • iris: Rewrite image access to u64 surface address in 64bit addressing mode

  • iris: Rewrite texture access to u64 surface/sampler address in 64bit addressing mode

  • intel/compiler: Skip intel_nir_lower_vec2_surface_sampler when the texture handle is already a single 64-bit

  • iris: Split iris_compute_emit_engine_async_threads_limits() into 2 functions

  • iris: Split iris_upload_compute_walker() into 2 functions

  • iris/64bits: Add compute support to iris_load_eff_64bit_surfaces_base_address/sampler_base_address

  • iris/64bits: Add support for compute shaders

  • iris/64bits: Add support to nir_intrinsic_load_num_workgroups

  • iris: Refactor pin_scratch_space()

  • iris: Add scratch buffer handling for efficient 64bit addressing mode

  • iris: Add support for IRIS_SURFACE_GROUP_RENDER_TARGET_READ in iris_populate_64bit_binding_table()

  • intel/brw: Add efficient 64bit addresing mode support to lower_fs_load_output

  • iris: Rewrite intrinsic_load_output to u64 surface/sampler address in 64bit addressing mode

  • iris: Upload 3D sysvals before populating binding tables

  • iris: Split alloc_bo_from_cache() for each loop to another function

  • iris: Reduce madvise calls by adding a second cache state

  • intel/compiler: Add sendgc

  • intel/gen: Add sendg EOT handling to gen_find_shader_size_xe()

  • iris/64bits: Update push constants when binding or sampler is dirty

  • iris/64bits: Fix iris_populate_64bit_binding_table() handling when pin_only is set

Juan A. Suarez Romero (40):

  • v3dv: add TFU readead for all images and buffers

  • v3d: add TFU readahead for buffers

  • broadcom/ci: updated expected results

  • v3d: save fragment constants on sand8/sand30 blit

  • broadcom/ci: update expected results

  • broadcom/ci: update fractions for non-full jobs

  • broadcom/ci: update expected results

  • broadcom/ci: upgrade kernel in DuTs

  • broadcom/ci: clean CI unused jobs

  • broadcom/ci: update expected results

  • v3d: fix data in get UBO size

  • ci: run shader-db for supported v3d versions

  • broadcom/ci: reduce number of jobs for rpi4

  • broadcom/ci: update expected results

  • broadcom/compiler: flip varyings replaced through COORD_REPLACE

  • broadcom/compiler: set full texture coordinates for sprite points

  • broadcom/ci: restore kernel for vc4 pre-merge jobs

  • vc4: warn about unsupported polygon mode

  • v3d: use proper enum for primitive mode

  • vc4: use proper enum for primitive mode

  • v3d: fix key with non-fill polygon mode

  • broadcom/ci: enable OpenCL CTS tests

  • ci: igalia farm maintenance

  • Revert “ci: igalia farm maintenance”

  • v3dv: keep CPU mapping alive for the cached BOs

  • broadcom/ci: update expected results

  • v3dv: run nir_opt_access

  • broadcom/ci: update expected results

  • broadcom/ci: update kernel for CI

  • broadcom/ci: update expected test results

  • v3d: include filtering support per format

  • v3d: explicitly call util_gen_mipmap() on failure

  • broadcom/ci: update expected test results

  • broadcom/ci: update expected list

  • ci: disable Igalia’s farm due to maintenance

  • Revert “ci: disable Igalia’s farm due to maintenance”

  • v3d: make alpha-integer-only formats not renderable

  • vc4: check if MIN_LOD or MAX_LOD is used

  • vc4/ci: document some failures

  • egl/gbm: Ignore current from buffer in has_free_buffers

Julia Zhang (3):

  • radeonsi: avoid mapping timestamp BO for duplicate tracepoints

  • radeonsi,radv: avoid timestamp conversion overflow

  • amd/virtio: request deferred mapping for blob BOs

Julien Schueller (3):

  • st/xlib: Fix thread leak when using offscreen rendering via llvmpipe

  • glx: Fix leak of current context in glXDestroyContext

  • mesa/eval: Fix wrong auto-normal direction with reversed parameter ranges

Juston Li (2):

  • anv/android: Don’t disable AUX_BIT importing Xe2+ CCS mod images

  • util/android: Only read from debug/vendor prefixes on Android T+

Jyotiranjan Bhuyan (1):

  • radeonsi: invalidate I$/SMEM$/VMEM$ at IB start on gfx12

Kajal Kajal (3):

  • freedreno/a6xx: upload a dummy sampler when textures but no samplers bound

  • rusticl/kernel: emit image/texture barriers after a kernel launch

  • rusticl: fix wrong destination offset in Image::copy_to_buffer

Karmjit Mahil (28):

  • tu: Propagate vp state for binInfo

  • tu: Fix BO names which the kernel might reject

  • tu: Avoid EINVAL in tu_bo_set_kernel_name()

  • freedreno/msm: Fix bo name length truncation

  • freedreno,tu: Share more of the msm code

  • freedreno/drm,tu: Move some tu functions into msm_common.h

  • tu: Remove some simple helper/wrappers

  • tu: Fix GMEM_BASE GET_PARAM failure error message

  • freedreno,tu: Add FD_MSM_GEM_NAME_LENGTH

  • tu: Fix inline uniform update offset

  • tu: Fix sparse bind column merging

  • tu: Remove tu_sampler reference in VkDescriptorUpdateTemplate

  • tu: Bump editorconfig line length

  • util/blob: Add cpp wrapper classes

  • tu: Unify shader cache serialization/deserialization

  • tu: Force AMD vendor_id for MGS4

  • tu: Set `suballoc->bo` to `NULL` after finish in error path

  • tu: Don’t save errno before vm_bind_fence_lock unlock

  • tu: Add some error handling for failed `tu_bo_map()`

  • tu: Add missing `munmap()` in `kgsl_bo_user_map()`

  • tu: Add some error handling in `bo_init_new_ion_legacy()`

  • tu: Add some error handling in `kgsl_bo_init_dmabuf()`

  • tu: Add some error handling for perfcntr allocations

  • ir3: Add casts before moving ir3_disk_cache to cpp

  • ir3: Move ir3_disk_cache.c to cpp

  • ir3: Store mergedregs as a u8 bool

  • ir3: Unify disk shader cache serialization/deserialization

  • nir/opt_algebraic: Remove some remaining i2i casts

Karol Herbst (77):

  • mesa_clc: support 32 bit targets

  • meson/rusticl: fix typo in depfile for builtin shaders

  • rusticl/meta: mark builtin kernels SPIR-V as 64 bit

  • rusticl/mesa: compile 32 bit version

  • rusticl/program: add Program::from_spirv_with_devs

  • rusticl/meta: support 32 bit devices

  • rusticl/meta: split out ulong kernels

  • rusticl/memory: return 0 for CL_IMAGE_SLICE_PITCH also for 2d images

  • rusticl/kernel: add libclc source hash to kernel shader keys

  • vtn/opencl: fix libclc needing fp16 lowering to fp32

  • hwtest: fix tests as F2FP and IMNMX do not exist on Volta

  • nouveau: Fix return of dangling pointer in nouveau_fence_new

  • nak: add support for nanosleep

  • nak: add Ldcg

  • nak/qmd: add nak_set_invalidate_cache

  • nak: add support for QMD 1.7 on KEPLER_COMPUTE_B and newer

  • nak/hw_runner: move Runner::run_raw parameters into a struct

  • nak/hw_runner: allow multiple compute dispatches in one run

  • nak/hw_tests: add tests to verify QMDs can run in parallel or out of order

  • nak/hw_runner: refactor to allow for more manual testing

  • nak/hw_test: test QMD deps in more depth

  • nak: print shared memory usage in stats

  • nouveau/winsys: add nv_device_info::max_blocks_per_mp

  • clc: make libclc optional for configs not needing it

  • clc: use our downstream fork of libclc

  • rusticl: warn if we load not our own libclc fork

  • ci: download libclc artifacts from mesa-libclc

  • clc: add flags for fp32 atomics

  • rusticl: add basic support for cl_ext_float_atomics

  • iris: implement pipe_screen::semaphore_create

  • clc: move open_clc_data below map_clc_data and close_clc_data

  • clc: use the libclc file size from stat instead of seeking

  • clc: map the libclc early

  • clc: stop using the mtime as it is not reliable

  • intel/ci: skip a few long running CL test on ADL due to timeouts

  • nak: improve implementation of global_bounded

  • nak: call nir_opt_move_discards_to_top

  • nak: improve lea patterns

  • nak: swap iadd3 sources with src2 if src0 and src1 are both modified

  • nak: relax iadd3 patterns with is_only_used_by_iadd

  • nak: add iadd3 ineg with constant rules

  • nouveau/mme/tu104: add test and simulator code for DMA_READ and DMA_SYNC

  • nir/serialize: set info.spec to NULL

  • nvk: enable stencil compression

  • rusticl/event: properly implement CL_PROFILING_INFO_NOT_AVAILABLE error condition

  • rusticl/event: stop sharing a struct for event types

  • rusticl/event: move cmd_type into GPUEvent

  • rusticl/event: move queue into GPUEvent

  • rusticl/event: use atomics for profiling

  • rusticl/event: move deps into GPUEvent

  • rusticl/event: move profiling into GPUEvent

  • rusticl/event: extracts parts of call into call_inner

  • rusticl/event: change result type of call_inner to CLResult

  • rusticl/event: rework the public set_time to a mark_queued

  • rusticl/event: start abstracting the profiling implementation

  • rusticl/event: split up the Profiler implementation

  • rusticl/mesa: add needed interfaces for query buffer objects

  • rusticl/mesa: add needed interfaces for raw timestamp queries

  • rusticl/event: add GPU profiling support

  • meson: disable all warnings when compiling bindgen generated sources

  • rusticl/context: skip non SVM devices in clSVMAlloc

  • rusticl/platform: skip non SVM devices when creating the VM heap

  • nir/load_libclc: do not compare sys_path against NULL

  • nir/load_libclc: turn sys_path into an array

  • nir/load_libclc: add llvm-23 libclc paths

  • rusticl/program: skip pointless nir compilations

  • rusticl/event: add support for cleanup callbacks

  • rusticl/mem: implement clEnqueueSVMFree with a cleanup function

  • rusticl/event: support profiling of events without any work

  • rusticl/event: allow to create events without any work

  • util/queue: remove locked argument from util_queue_kill_threads

  • util/queue: remove locked argument from util_queue_adjust_num_threads

  • util/queue: Fix dead-lock between queue_finish and queue_kill_threads

  • rusticl/program: implement CL_INVALID_SPEC_ID error for clSetProgramSpecializationConstant

  • rusticl/spirv: add SPIRVBin::specialize

  • rusticl/program: fix spec constants

  • zink: cast last operands of SpvOpGroupNonUniformShuffle* ops to unsigned

Kelsi (1):

  • v3dv: fix typos in comments

Kenneth Graunke (53):

  • brw: Rework URB lowering max_push_bytes to be an URB read length

  • jay: Move VS URB setup back to jay_nir.c

  • intel: Make limit_trig_input_range a compiler-wide setting

  • iris: Drop limit_trig_input_range key field, repack keys

  • brw: Drop brw_nir_optimize call from brw_nir_apply_key

  • jay: Don’t call brw_nir_apply_key twice for geometry shaders

  • jay: Make simd_width not a pointer in jay_nir_lower_simd

  • jay: Use a switch in jay_nir_lower_simd

  • jay: Optimize load_subgroup_id to 0 for single thread workgroups

  • brw: Skip load_subgroup_size for known sizes when lowering ray queries

  • brw: Make brw_geometry_stage_dispatch_width available to C code

  • brw: lower vars to ssa after systolic vectorize pass

  • brw: Set nir->info.{min,max}_subgroup_size for geometry and RT stages

  • brw: Lower subgroup masks in brw_preprocess_nir

  • jay: Call brw_nir_apply_key later

  • jay: Split jay_process_nir into 3 pieces

  • jay: Stop using brw_nir_apply_key

  • brw: Refactor INTEL_SIMD_DEBUG helpers that take a dynamic SIMD index

  • jay: Rewrite SIMD selection

  • jay: Compile multiple SIMD variants

  • jay: Return multiple shader stats

  • jay: Delete load_subgroup_size translation

  • jay: Run nir_opt_dead_cf after lowering SIMD width and subgroup sizes

  • jay: Avoid propagating some immediates into uniform flags

  • jay: Implement FB write payloads with multiple file swaps

  • intel: Disable 16x MSAA support

  • intel/nir: Avoid vec8 loads/stores even on Xe2+

  • intel/nir: Allow V8 load/stores when known to be SIMD16 or smaller

  • intel/compiler: Don’t vectorize URB access to vec8 in SIMD32

  • nir: Set info->fs.color_is_dual_source for DUAL_SRC_BLEND var locations

  • intel: Drop non-SSA regs after NIR shader statistic

  • intel: Add VRT size and thread statistics

  • intel/compiler: Allow reordering deferred URB writes if no VS pulls

  • jay: Shrink geometry shader payload registers with invocations > 1

  • jay: Stop using JAY_NUM_PHYS_GRFS in a bunch of places

  • jay: Rename JAY_NUM_PHYS_GRF to JAY_MAX_PHYS_GRF

  • jay: Increase JAY_MAX_PHYS_GRF to 256

  • jay: Re-express UGPR spilling limit

  • jay: Move min_ugprs calculation out of the “spilling?” loop

  • jay: Move estimate_nonunif_grf calculation out of the “spilling?” loop

  • jay: Calculate the minimum number of GRFs to reserve for GPRs

  • jay: Select a VRT size

  • jay: Drop SIMD variants which decrease VRT occupancy

  • jay: Support Xe3+

  • jay: Rewrite algorithm for divying up GRFs between strides

  • jay: Be less conservative in our estimate of nonuniform GRF needs

  • jay: More aggressively schedule for pressure on Xe3

  • jay: Don’t require VECTOR_EXTRACT to align to the src vector size

  • jay: Don’t map null destinations to accums

  • jay: Fix SIMD selection when a workgroup fits within a size

  • Revert “jay: remove pointless prog_data zeroing”

  • jay: Avoid out-of-bounds bitset read for pinned registers

  • jay: Mask shuffle lane index to avoid out-of-bounds register file access

Konstantin Seurer (18):

  • vulkan: Remove some unnecessary guards

  • vulkan: Add a helper for querying the IR leaf node size

  • vulkan: Add quad node

  • vulkan: Make prefix_scan a shared helper

  • vulkan: Implement a late triangles -> quads pass

  • radv/bvh: Fix updating acceleration structures containing AABBs

  • radv,vulkan: Propagate the cull mask up the BVH

  • radv: Remove incorrect comment

  • radv/bvh: Use the correct bounds for instance nodes on GFX12

  • radv/bvh: Add box node encode helpers for GFX12

  • radv/bvh: Use filter instance nodes on GFX12

  • radv: Align bvh to 128 bytes on GFX12

  • radv: Avoid some VALU during ray traversal

  • radv/rt: Restore hit attribs after rejecting AABBs

  • lavapipe/rt: Restore hit attribs after rejecting AABBs

  • radv: Use pair compression for updatable AS on gfx12

  • radv: Fix build flags for updates

  • lavapipe: Fix DGC + conditional rendering

Kostadin Shishmanov (1):

  • aco: add missing <algorithm> include

Krzysztof Sokołowski (1):

  • lima: close the screen fd on destroy and on failed creation

Lakshman Chandu Kondreddy (5):

  • zink: fall back to sole pdev when DRM matching is unavailable

  • freedreno: Add chip-id support for A830v1

  • mesa/blit: don’t inherit color-blit swizzle in depth/stencil blit

  • freedreno: Add chip-id support for Adreno X2-85

  • ir3,freedreno: set api_wavesize=IR3_DOUBLE_ONLY for gen6+ compute shaders

Lakshmi Kumar (4):

  • gallium/va: fix build-order edge for libva win32 headers on Windows

  • d3d12: route AV1 film grain output to the film_grain_target surface

  • d3d12: scope AV1 film grain output redirect to reference-only configs

  • d3d12/video: reuse decoder heap on mid-stream resolution decrease

Lars-Ivar Hesselberg Simonsen (9):

  • nir/divergence: Add missing poly/pan intrinsics

  • pan/nir: Add divergent scratch lowering pass

  • pan/genxml/v9: Clean up format list

  • pan/genxml/v10: Clean up format list

  • pan/genxml/v12: Clean up format list

  • pan/genxml/v13: Clean up format list

  • pan/genxml/v14: Clean up format list

  • panvk/v10+: Fix unbound LD_ATTR.auto32

  • panvk: Fix incremental rendering for v14+

Leon Perianu (3):

  • pvr: enable global priority related extensions

  • pvr: Fix primitive restart for non-indexed draws

  • pvr: enable VK_EXT_primitive_topology_list_restart

Leon Seemann (2):

  • zink: update the pipeline when only the vertex input state changes

  • vbo: Fix extra vertex drawn on line loop end after buffer wrap

Linermao (1):

  • virgl: Close winsys fd after destroying screen

Lionel Landwerlin (212):

  • anv: flush accumulated barriers for top of TOP_OF_PIPE

  • iris: set 3DSTATE_SF::LastPixelEnable with smooth lines

  • anv: fix Wa_18040903259

  • anv: remove redundant entrypoint

  • anv: move post dispatch out of Gfx12.5 code

  • anv: track the last type of emitted commands

  • anv: workaround for missing barriers between compute shaders

  • anv: move RT post_dispatch_wa helper after the pre_dispatch_wa

  • anv: factor out compute breakpoint

  • brw/jay: move vs component packing to common code

  • brw: move variable off the stack

  • brw/nir: avoid generating invalid load_input

  • brw/nir: Don’t set ACCESS_CAN_REORDER/NON_WRITEABLE for VS inputs

  • brw/nir: Prepare URB lowering code for VERT_ATTRIB_*

  • brw/nir: augment load_attribute_payload_intel to express vector layout

  • intel/compiler: lower load_input in vertex shaders to load_urb if too large

  • anv/brw: get rid of push constant promotion limitation

  • anv/brw: add a compiler parameter to limit VF payload delivery

  • vulkan/runtime: fixup vk_shader leak on RT group recompile

  • anv: fix push buffer descriptor address relocation

  • brw: add missing INTEL_FS_CONFIG_PER_PRIMITIVE_REMAPPING handling

  • nir/builder: add deref/var atomic helpers

  • brw: fix wa_18019110168 lowering

  • anv: don’t release first allocated batch bo in command buffers

  • anv: fix leak in RT binding point

  • anv: move trace_rays_active to rt state

  • anv: shrink anv_gfx_dynamic_state

  • anv: move xfb tracking to graphics state

  • anv: shrink binding table hash tracking

  • anv: pack some bits in anv_attachment

  • anv: shuffle some bits in anv_cmd_graphics_state

  • anv: move some attachment clearing values off the graphics state

  • anv: rename anv_cmd_pipeline_state into anv_bind_point_state

  • anv: get rid of push constant data dirty tracking

  • anv: delay binding point state allocations

  • anv: fix barrier for Wa_1508744258 / Wa_14024015672

  • iris: fix barrier for Wa_1508744258 / Wa_14024015672

  • jay: copy resource_intel surface handle value

  • anv: fixup the logic dealing with STATE_BYTE_STRIDE

  • anv/brw/jay/iris: factor out source_hash

  • anv: only consider active view-capable queues for image views

  • zink/ci: skip occlusion queries on ADL/TGL

  • vulkan/runtime: remove assert on shader deserialization

  • vulkan/runtime: discard disk cache items when deserialization failed

  • anv/brw: fixup per shader compile options

  • anv: fix descriptor heap with non buffer aligned addresses pre Gfx12.5

  • vulkan/runtime: reindex internal bits

  • vulkan/runtime: add tracking of used binding model

  • vulkan/runtime: track pipeline layout for RT pipelines & binding

  • brw: fix non register aligned MOV_INDIRECT source

  • anv: unify descriptor set layout type and binding mode

  • anv: remove non-relevant PRM quote

  • anv: move driver push data out in front binding model data

  • anv: shrink push constant data for EXT_descriptor_buffer/EXT_descriptor_heap

  • anv: always ensure push constant is allocated for compute

  • anv: fix CmdWaitEvents2 dependencyFlags usage

  • anv: remove all code related to INTEL_performance_query not using metrics_library

  • anv: fix binding mode flush

  • anv: flush L1/L2 caches on end of command buffer

  • anv: remove unnecessary ssbo size intrinsic lowering

  • anv: rename a few helper functions to indicate when to use them

  • anv: stop allocating compute push constants in the general state pool

  • blorp: move enable_tbimr to blorp_config

  • blorp: stop using the general state heap

  • anv: stop using general state heap for simple_shader

  • anv: get rid of the remaining general state stream

  • anv/drirc: move indirect descriptor option to driver section

  • brw: handle load_ubo through memory_access on lsc platforms

  • anv: don’t forget to hash render pass color attachments

  • anv: move drirc instantiation to physical device

  • driconf: rename device attribute into device_name

  • brw: remove RT synchronous source

  • brw/jay: move stack_id handling for RT message payload to NIR

  • intel: fixup mcs sampler handle lowering

  • brw: remove unused helper

  • brw: pass shader key to pass tracker in compile_cs

  • blorp: stop using FRAG_RESULT_COLOR

  • intel: rename a NIR pass that is not tied to BRW

  • intel: extract lower fragment outputs out of jay

  • intel/nir: fixup RT write logic with missing alpha

  • intel/nir: disable sample mask write if we’re skipping it

  • intel/nir: replicate alpha with alpha_to_coverage & no sample mask

  • intel/nir: deal with lower_io with dual source

  • brw: help copy propagation for src0 alpha

  • brw: reuse jay’s lowering fs output pass

  • anv: move non push constant related value out of the push constant section

  • anv: someone doesn’t know its powers of 2

  • brw: use constant offset in extended descriptor for spill/fill

  • anv: fix 64bit values promoted to inline register

  • anv: fix push descriptor flushing

  • brw: pass shader keys around for RT lowering

  • brw: use sampler from incoming instructions for combiner

  • brw: use helper for base_offset lowering

  • brw: make txf combiner more robust to new instruction fields

  • brw/nir: prepare for render target writes in efficient 64bit mode

  • intel/gen: fixup restriction for scalar register MOVs

  • intel/brw/64bits: Add support for RT tracing

  • intel/brw/64bits: add missing scalar register dependency of send messages

  • nir: handle non 32bit resource_intel

  • brw: lower vec2 surface/sampler handles to 64bit values

  • brw: support surface index in efficient 64bit mode

  • blorp: reduce helper function arguments

  • blorp: use tex sources for sampler/texture

  • blorp: pass context for param initialization

  • anv: add a physical device property for efficient 64bit support

  • anv: split VA setting for indirect/non-indirect descriptors

  • anv: setup VA for efficient 64bit

  • anv: require bindless in efficient 64bit mode

  • anv: compile shader with efficient 64bit if enabled

  • anv: add a direct address format

  • anv: use vec2_index_32bit_offset address format for efficient 64bit

  • anv: efficient 64bit support for descriptor heap

  • anv: handle efficient 64bit resource_intel

  • anv: efficient 64bit render target support

  • anv: cleanup wa_18019110168 helper

  • anv: add support for relocated scratch surfaces in efficient 64bit mode

  • intel: remove old efficient 64bit defines

  • blorp: handle prebake surface states in efficient 64bit

  • blorp: compile VS with efficient 64bit if needed

  • anv: deal with attachment clearing in efficient 64bit

  • anv: fixup dynamic buffers with ANV_DEBUG=bindless

  • anv: try to avoid segfault on relocation append

  • anv: emit CPB setup at device initialization

  • anv: stop using compute blorp shaders on Xe2+

  • anv: only printout HuC warning when video is used

  • anv: fixup companion command buffer in efficient 64bit mode

  • brw: always let the driver handle -1 render targets

  • blorp: fixup indirect copies for efficient 64bit mode

  • anv: update the max uniform buffer range

  • isl: limit max programming buffer size on Xe2+

  • anv: fix synchronization of acceleration structure query writes

  • anv: fix another AS build barrier

  • vulkan/runtime: allow the driver to specific debug output

  • anv/bvh: make debug printf available to all shaders

  • anv: always print debug output to stderr

  • blorp: disable replicated sends in efficient64bit mode

  • brw: ensure repclear shader is not requested in efficient 64bit

  • blorp: enable SIMD32 clear shaders

  • intel/ds: trace num_layers in blorp operations

  • anv: add missing handling for PIPELINE_STAGE_2_COPY_INDIRECT

  • anv: remove transfer from render section in barriers

  • anv: respect bit_size/num_component from the IR with acceleration structures

  • anv/blorp: disable compute stats on internal work

  • anv: fix compute shader ThreadGroupDispatchSize value

  • blorp: fix shader kernel pointer with multi size kernels

  • gen/brw: differentiate register and memory data type sizes

  • brw: fix alignment of base offsets on Xe3P+

  • anv: add missing constant cache invalidation in DGC path

  • vulkan: bump headers to 1.4.363

  • nir/lower_input_attachments: add support for heap subpass image load

  • anv: fixup query clear/copy flushing on Xe2+

  • anv: limit flushing for transfer writes on RCS on Xe2+

  • anv: remove duplicate calls to flush generated draws

  • anv: simplify buffer_cp

  • anv: stop using ANV_PIPE_RT_BTI_CHANGE with RCC BTI+BTP hashing

  • anv: use helper for CS coherency check

  • anv: remove L3 flushing with indirect on Xe2+ again

  • anv: avoid CS stalls for indirect on Xe2+

  • anv: implement VK_INTEL_device_info

  • anv: disable DRM_XE_DEVICE_QUERY_ENGINE_CYCLES with UnrealEngine 5.8

  • vulkan/wsi: add missing usage2 flags

  • intel: wire more information for max immediate offset decision

  • anv: change detection of input attachments

  • intel: remove unused debug option

  • jay: respect INTEL_DEBUG=nocompact

  • brw/jay: add binary output to INTEL_DEBUG=mda

  • intel/compiler: add INTEL_DEBUG=no-jay as a compiler key

  • intel/compiler: add source hash at the end of the shaders

  • anv: keep OAG reports in host memory

  • anv: bound encode bvh global_id to avoid OOB accesses

  • anv: enable TRTT TLB invalidation on other engines than RCS/CCS

  • anv: report support for sparse binding on transfer queue

  • anv: enable image atomics with sparse

  • anv: enable dynamic indexing of input attachments

  • anv/iris: add support for realtime priority

  • anv: only upper bound encode bvh global_id

  • anv: align shader heap size allocation to alignment

  • anv/jay: fix debug archiver with resume shaders

  • anv: fix dynamic buffer base offset upload on <= Gfx12.0

  • anv: store availabilty before writing partial results

  • anv: fence MI availabilty writes

  • anv: track the query pool copies for resets

  • nir: fixup validation for nir_texop_resinfo_intel

  • nir/gather_info: set info.uses_resource_info_query for resinfo_intel

  • anv/pipeline_layout: use a different variable for set bti offset vs memory offset

  • anv: limit amount set variable writes

  • brw: move back early texture lowering to preprocess

  • anv: use a common indirect image descriptor structure

  • genxml: don’t generate global keyword in host repacking code

  • genxml: remove generated attributes

  • intel/genxml: more sharing between host/device gen code

  • intel/genxml: add a limited variant of unpack

  • anv: move internal kernel NIR calls out of genX

  • nir/lower_tex: enable special case lowering for txs usage

  • nir/intel/compiler: add an option to emit a special texop for lowered tex operations

  • anv/intel: add a software RESINFO implementation

  • vulkan/image: store maximumRequestedAlignment

  • anv: use the runtime’s alignment value

  • anv: add handling of VkFormatProperties4KHR

  • anv: switch to _2_ image create/usage flags

  • anv: handle VkImageCreateFlags2CreateInfoKHR/VkImageUsageFlags2CreateInfoKHR

  • anv: expose VK_KHR_extended_flags

  • genxml: fix repacking emission

  • genxml: remove unused code

  • genxml: remove non existing field type check

  • anv: reset texture_index

  • brw: fix oob source access

  • vulkan/wsi: fix a bunch of incorrect error code checks

  • anv: avoid leak on internal kernels

  • anv: fix potential set_layout null pointer access

  • intel/compiler: always run query optimization passes on supporting HW

  • anv: bail early on unsupported image format

Liviu Prodea (1):

  • clc: Fix static link with llvm and clang 23

Lokbondo Kung (2):

  • llvmpipe: fix ubsan shift exponent too large error

  • util: Fix misaligned address for ‘gc_slab’

Lorenzo Rossi (181):

  • kraid/ra: Fix re-swizzling an already swizzled 32-to-64 Src

  • kraid/swizzle: Fix Swizzle ToString crashing

  • kraid/swizzle: Allow SwizzleByte/Word::Zero.modify

  • kraid: Align hw_test compilation

  • panfrost/midgard: Fix fau max not initialized

  • kraid: Add From<f32> for SrcRef

  • kraid/builder: Add fadd and fma helpers

  • kraid: Add FExp.f32 and wire it up

  • kraid: Add OpFAddLScale with tests

  • kraid: Add OpFLogD and wire it up with tests

  • kraid: Add fsin/fcos and wire them up with tests

  • kraid: Split staging-register encodings into swizzle/lanes vs data_type forms

  • kraid/opt_copy_prop: Copy-prop IAdd

  • panfrost: Fix wrong layout reuse in user clip planes

  • kraid/encode: Fix BranchCombineOp

  • compiler/rust/smallvec: Implement retain()

  • kraid/legalize: Fix FAU User legalization for v13

  • kraid: Implement DisplayOp to split instruction printing

  • kraid: Print flow info after instruction name

  • kraid/ops: Use dot in bool_as_mod_str macro

  • glsl/gl_nir_lower_packed_varyings: propagate mediump inside structs

  • glsl/tests: Add mediump struct-field precision lowering comparison tests

  • compiler/rust/float16: Add a F16::classify() method

  • kraid/ops: Print the correct source in OpCubeFaceIdx

  • compiler/rust/float16: Add utility methods

  • kraid/foldable: Add float get/set utils

  • kraid: Add compiler workaround for f32 min/max

  • kraid/hw_test: Fix cmp_eq_f tests

  • kraid/ops: Various foldable fixes

  • kraid/hw_tests: Implement proper f32/f16 random sampling

  • kraid/hw_test: Print raw float value on error

  • kraid/ops: Fold and test float Ops

  • pan/bifrost: Add arch to tests

  • pan/compiler: Implement f2f16 fusion on valhall

  • panfrost/drm-shim: Fix g72 parameter error

  • pan/nir: Fix header static inline function

  • pan/compiler/stats: Fix ALU not being used in instruction bounds

  • nir/opt_algebraic: Fix fmul fadd compilation fixpoint

  • panvk: Lower VS load_view_index to zero when view_mask is zero

  • kraid/nir: Add preloaded regs for VS

  • kraid: Pretty print resource handles

  • kraid: Add OpLdAttr

  • kraid: Enable secondary shader invocation

  • kraid/encode_v9: Fix has_vecsize DstLanesSet

  • kraid: Pad variants to 128 bytes

  • kraid: Print store/load is_tls

  • kraid: Build variant without psiz

  • pan/bi: Propagate swizzle in bi_optimizer_result_type

  • pan/midgard: Fix 16-bit varying lowering for midgard

  • pan/nir_fuse_io: Add support for LD_ATTR

  • pan/compiler/lower_fs_outputs: Support gl_FragColor

  • pan/compiler: Stop lying about indirect input support

  • nir/lower_xfb_to_stores: Support non-zero store_output component

  • pan/compiler: Add pass that removes XFB after lowering

  • pan/compiler: Never trust highp flat types

  • panfrost,panvk: Remove trust_highp_types

  • pan/compiler/collect_varyings: Handle disagreement on alu_type

  • panfrost/shader: Only lower IO when needed

  • pan/compiler: Split pan_nir_resize_varying_io inputs in format and layout tables

  • pan/bifrost: Re-vectorize loads in the compiler

  • pan/compiler: Lower mediump varying floats to 16-bits early

  • panfrost: Lower gl_FragColor in the frontend only on Midgard

  • pan/compiler: Lower nopersp FS inputs in lower_fs_inputs

  • panfrost/compiler: Move PAN_AS_U32 in common code

  • panfrost: Add intrinsic for raster coverage, sample and centroid id

  • pan/compiler: Move FS input lowering in NIR

  • kraid/dst_mod_prop: Fix clamp remove check

  • kraid/encode_v9: FAddImm cannot narrow floats

  • panvk/ci: Route VS shaders through Kraid in Kraid job

  • panfrost: Always emit vertex attribute descriptors for arch < v9

  • panvk/jm: Always emit vertex attribute descriptors

  • panfrost/mid: Disable mediump lowering entirely

  • pan/compiler: Remove hand-rolled descriptor type smashing

  • pan/compiler: Optimize flat 16-bit load on linked shaders

  • pan/perf: Adjust varying rate calculations

  • pan/perf: Don’t count LD_VAR_SPECIAL.frag_z

  • panvk/jm: Fix unbound attribute reads

  • compiler/rust: Add U8EnumSet::remove()

  • kraid/ops: Fix Display for OpF32ToI32

  • kraid/dst_mod_prop: Fix wrong F32ToF16 dst data-type

  • kraid/nir: Fix miscompile for shift_and(x, shift, 0)

  • kraid/opt_copy_prop: Fix hardcoded 4 bytes in swizzle read source

  • kraid/opt_promote_consts: Respect inputs.promote_immediates

  • kraid/isa: Fix is_message not set for hardcoded messages

  • nir/intrinsics: Rename preloaded register for v15 compatibility

  • kraid: Clean up preloads

  • kraid/ra: Clean up clippy complaints

  • kraid/hw_tests: Apply Clippy suggestions

  • kraid/hw_tests: Switch to builder pattern for args

  • kraid/hw_tests: Move read-write data buffer layout to Rust

  • kraid/hw_tests: Implement simple buffer descriptors

  • kraid/hw_tests: Fix 64-bit test data address support

  • kraid/ir: Add RegRef::new utility

  • kraid/hw_tests: Add infrastructure for low-level hardware tests

  • kraid/hw_tests: Add LD_PKA hardware bug test

  • pan/nir_opt_push_ubo: Don’t reorder blend constants

  • pan/nir_opt_push_ubo: Fix crash for out-of-bounds UBO promotion

  • kraid/builder: Fix 16-bit derivatives

  • kraid/dst_mod_prop: Fix illegal FADD.f16 produced

  • kraid/stats: Use the actual encoded code_size instead of re-deriving it

  • kraid/isa: Wire up exec_time

  • kraid/stats: Wire up exec_time and MMUL-specific cycles

  • pan/bi: Add arch name for v13 and v14

  • pan/va: Update ISA.xml exec units

  • pan/va_perf: Adjust unit based on the GPU architecture

  • pan/va_perf: Account for instruction exec_time

  • kraid: Fix OpAdr definition order

  • kraid/ir: Add From<ALUType> for NumericType

  • kraid/data_type: Add missing untyped ints for LdVarFlat

  • kraid: Add OpLdVar family

  • kraid/message_slots: Handle LD_VAR hidden register dependancise

  • kraid/nir: Add get_f32_src helper to fix swizzle in 16-bit floats

  • kraid: Add OpATest

  • rust/nir: Add io_semantics index read

  • compiler/rust: Add an AsArrayLen trait

  • kraid: Refactor MAX_SRC_COUNT

  • kraid: Add OpBlend

  • kraid: Add OpDiscard

  • kraid: Add OpJump

  • kraid: Add OpZSEmit

  • kraid: Add LdTile and StTile

  • kraid/nir: Add cumulative coverage preload reg

  • kraid/nir: Add Fragment Shader preloads and FAUs

  • kraid: Add Blend-related preloads

  • kraid/ra: Implement fixed registers

  • kraid: Add pass that lowers blend calls

  • kraid: Add preceding wait message slot support

  • kraid: Add Blend shader support

  • panvk/ci: Route FS shaders through Kraid in Kraid job

  • kraid: Add varying hidden-register optimization algorithm

  • kraid: Ignore blend prologue in stats

  • kraid: Add opt_flow to DCE and merge flow

  • kraid/compile: Use valhall isa stats for empty shaders

  • kraid/compile: Remove lower_insert_byte

  • kraid: Shrink special FAURefs after load

  • kraid/lower_copy: Lower 64-bit copy with IADD instead of NO_SHIFT

  • kraid/spill: Fix spill heuristics

  • kraid/spill: Fix inverted spill heuristic next-use order

  • kraid/spill: Cleanup dead SSAs from predecessor’s liveset

  • panfrost: Organize reporting code

  • kraid: Wire up stats printing

  • kraid/stats: Refactor stats_count

  • kraid/ir: Refactor Shader printing into separate utility

  • kraid/stats: Print per-instr-stats if PRINT debug flag is on

  • kraid/v9: Add 8 and 16 bit variants to IToF32

  • kraid/v9: Add F16ToI32

  • kraid/v9: Add IToF16

  • kraid/v9: Add HAdd

  • kraid/v9: Add 8-bit variants to add/sub/abs

  • kraid/v9: Add 16-bit float ops

  • kraid/nir: Fix fquantize2f16

  • kraid/v9: Fix FRcp/FRsq .h1

  • kraid/v9: Wire up pack_half_2x16_split

  • kraid/v9: Implement IDVS varyings

  • kraid/hw_runner: Fix v9 compute_ep_limit

  • kraid: Remove OpFMul

  • kraid/legalize: Fix k0 legalization for v9-v11

  • kraid/hw_tests: Add lower_copy test

  • kraid/model: Add SmallConst helpers to search for swizzled immediates

  • kraid/lower_copy: Add fallback for v9-v10

  • pan/bi: Fix spill cost overestimation

  • kraid: Add pass to translate MUX into CSEL

  • kraid/spill: Fix overflow in liveness analysis

  • kraid/spill: Fix spill heuristics for loop live-through SSAs

  • kraid/opt_copy_prop: Fix zero-extension on 64-bit swizzles

  • kraid: Add 64-bit immediates

  • kraid/lower_mkvec_swz: Add 64-bit swizzles

  • kraid/hw_tests: Test 64-bit swizzles (and OpSwz)

  • kraid/opt_copy_prop: Add 64-bit immediates

  • kraid/opt_promote_consts: Promote 64-bit immediates

  • kraid/nir: Create Imm64 sources from 64-bit NIR constants

  • kraid/foldable: Use Src::resolve_imm instead of duplicating

  • mr-label-maker: Label src/panfrost/compiler/kraid

  • kraid/legalize: Treat fixed-dst regs as clobbers

  • kraid: Wire up tex skip modifier

  • kraid/hw_tests: Fix and assert on OOB test data access

  • kraid/compile: Set correct threads for empty shaders

  • kraid/copy_prop: Fix narrowing FAdd being ignored

  • kraid/dst_mod_prop: Fix clamp narrowed dst_type

  • kraid: Remove F32_TO_F16

  • kraid/nir: Use FAdd for v2f16 on v9-v10

Loïc Molinari (34):

  • panfrost/ci: Mark “spec@!opengl 1.4@copy-pixels” as flake

  • u_blitter: Fix out-dated draw_rectangle handler doc

  • panfrost: Pass primitive mode to funcs instead of full draw info

  • panfrost: Add infrastructure for fullscreen draw calls

  • panfrost: Emit draw call flags with dedicated func

  • panfrost: Blit with the FullScreenJob descriptor (v9)

  • panfrost: Blit with the RUN_FULLSCREEN instruction (v10)

  • panfrost: Blit with the RUN_FULLSCREEN instruction (v12)

  • panfrost: Blit with the RUN_FULLSCREEN instruction (v13)

  • panfrost: Blit with the RUN_FULLSCREEN instruction (v14+)

  • panfrost: Fix MTK tiled resources mapped as shader images

  • panfrost: Move blit funcs to pan_blitter

  • panfrost: Move clear funcs to pan_blitter

  • panfrost: Make legalization less verbose

  • panfrost: Shorten legalized blit func name

  • panfrost: Test render cond in legalized blit func

  • panfrost: Refactor blitter states saving

  • panfrost: Save render cond state in panfrost_blitter_clear()

  • panfrost: Setup HW tile buffer clear in dedicated func

  • panfrost: Legalize blit feedback loops

  • gallium: Add downscaling support to util_blitter_generate_mipmap()

  • panfrost: Generate mipmaps with util_blitter_generate_mipmap()

  • pan/lib: Add supports_msaa_average() to pan_mod_handler

  • pan/genxml: Add Downscale Mode field to non-YUV render targets (v10+)

  • panfrost: Add support for mipmap downscaling (v10+)

  • panfrost: Move resource_copy_region() implementation to pan_blitter

  • panfrost: Copy resource regions with util_blitter_copy_texture()

  • panfrost: Rename pan_legalize_format()

  • panfrost: Don’t convert format to linear before legalization

  • panfrost: Remove duplicated branches

  • panfrost: Add CPU trace to panfrost_blitter_draw_rectangle()

  • panfrost: Set active batch bit in panfrost_batch_init()

  • panfrost: Reserve last batch slot for compute

  • panfrost: Flush AFBC-P compute jobs

Lucas Stach (18):

  • etnaviv: clean up index buffer handling code a bit

  • etnaviv: move index buffer handling in draw_vbo after derived state handling

  • etnaviv: reserve state emission space early in draw_vbo

  • etnaviv: move sampler source update before draw space reservation

  • etnaviv/ci: update expectations after cmdstream reservation fixes

  • etnaviv: record implicitly flushed framebuffers in draw_vbo

  • etnaviv: remove etna_state_update call from forced flush

  • etnaviv: validate that there are no forced flushes happen during draw

  • etnaviv: rs: validate that there are no forced flushes during emission

  • etnaviv: blt: validate that there are no forced flushes during emission

  • etnaviv: don’t reserve space for shaders fetched from memory

  • etnaviv: drm: don’t skip submit if there is a in_fence

  • etnaviv: discard in_fence fd after flush

  • etnaviv: resume active queries after pending resources has been reset

  • etnaviv: drm: make fence fd checks more robust

  • etnaviv: make fence fd checks more robust

  • etnaviv: use correct define when emitting NTE state

  • etnaviv: remove dead ASTC handling in TE texture state emission

Lucie Hartmann (1):

  • gallivm: Query host CPU features on aarch64 too

Luigi Santivetti (2):

  • pvr: speculative fix for out-of-bound access of winsys_heaps

  • pvr: set the UM stride when querying heap static areas

Lynne (1):

  • anv/video: force VP9 probability context 0 for key, intra-only and ER frames

Maaz Mombasawala (4):

  • svga: Pass buffer handle for fd handles for resource_from_handle()

  • svga: Remove erroneous else-if in surface_get_handle()

  • svga: Remove dma_mm buffer manager

  • svga: Fix some vmw_printf() debugging messages.

Marc Alcala Prieto (2):

  • panfrost/model: Fix v11 model rates

  • panfrost/model: Fix G720 and G725 sfu rates

Marek Olšák (195):

  • radv: don’t expose memory types from AMD_device_coherent_memory without the ext

  • ac/llvm: switch to amdgcn.ballot intrinsic

  • ac/llvm: remove the optimization barrier from ac_build_ballot

  • ac/nir/lower_ps_early: make mrtN.alpha dead with alpha_to_one

  • ac/nir/lower_ps_early: port radv_nir_trim_fs_color_exports in here

  • nir/lower_io_indirect_loads: implement lowering indirect vertex index

  • radv: fix determining the raster prim for guardband

  • radv: fix determining the raster prim for line mode

  • radv: fix determining the dynamic raster prim for FS barycentrics

  • radv: fix determining the static raster prim for FS barycentrics and front_face

  • radv: get FRONT_FACE from a user SGPR if it’s dynamic but constant

  • radv: reduce PS_STATE_PS_ITER_MASK to 8 bits

  • radeonsi: add handling of PERSP_PULL_MODEL

  • radv: enable indirect GS inputs, don’t move all GS input loads to the beginning

  • radv: stop using nir_lower_io_vars_to_temporaries to move and lower FS inputs

  • mesa: initialize texture tiling to GL_OPTIMAL_TILING_EXT

  • mesa: add a debug option that allows allocating GL textures with linear tiling

  • radv: don’t enable PERSP VGPRs for POS_W_FLOAT_ENA needlessly

  • radv: replace SPI_PS_INPUT_ENA magic number with proper definitions

  • radeonsi: replace SPI_PS_INPUT_ENA magic numbers with proper definitions

  • nir/remove_outputs: don’t remove LAYER as varying unconditionally

  • nir/lower_sysvals_to_varyings: add lowering to intrinsics instead of variables

  • radv: always remove PSIZ and LAYER as varyings

  • radv: optimize layer_id with nir_opt_varyings

  • radv: fix incorrect and suboptimal point size output removal

  • radv: don’t lower for smooth lines if the raster prim is never lines

  • radv: simplify code determining whether smooth lines may be enabled

  • radv: replace PS_STATE_LINE_RAST_MODE with SMOOTH_LINES, make it accurate

  • nir/opt_varyings: fix incorrect counting of emit_vertex within a block

  • ac/gpu_info: fully explain the conformant_trunc_coord saga

  • aco: check spi_shader_z_format consistency between ACO and RADV

  • aco: implement aco_ps_epilog_info::lower_1bit_sample_mask_to_discard

  • radv: reduce radv_ps_epilog_key from 36 bytes to 28 bytes

  • radv: rename export_{depth/stencil/sample_mask} to has_*_output

  • radv: let the PS epilog determine DB_SHADER_CONTROL for mrtz, not the shader

  • radv: always export mrtz in the PS epilog if it’s present

  • radv: rename radv_graphics_state_key…has_epilog -> color_outputs_need_epilog

  • radv: rename radv_nir_trim_fs_color_exports -> radv_nir_trim_fs_exports

  • radv: don’t export depth from the PS epilog if it has no effect

  • radv: don’t export stencil from the PS epilog if it has no effect

  • radv: enable lowering sample mask with samples == 1 to discard in the PS epilog

  • radv: remove the non-epilog FS depth output if it has no effect

  • radv: remove the non-epilog FS stencil output if it has no effect

  • radv: lower non-epilog sample mask output to discard for samples == 1

  • radv: use the PS epilog for the FS depth output if dynamic state can remove it

  • radv: use the PS epilog for the FS stencil output if dynamic state can remove it

  • radv: use the PS epilog for the samplemask output if dynamic state can remove it

  • radv: fix alpha-to-coverage perf for GFX11+ when “A2C enable” is static state

  • radv: enable depth/stencil/samplemask mrtz optimizations for shader objects too

  • radv: fix barriers with NGG without exports on GFX10-11

  • radv: use VS_PARTIAL_FLUSH with PS_PARTIAL_FLUSH when waiting for waves

  • radv: only use VS_PARTIAL_FLUSH to wait for GDS

  • radv: handle missing src stage flags for pipeline barriers

  • radv: fix and improve vkCmdSetEvent2 / vkCmdResetEvent2

  • radv: allow NGG passthrough with per-vertex primitive ID export

  • radv: remove all pre-rast shader outputs when rasterizer discard is enabled

  • aco: allow NGG shaders without exports

  • ac/nir/lower_ngg: add an option to emit 0 primitives for rasterizer discard

  • radv: don’t gather any output masks with rasterizer discard

  • radv: remove all exports from NGG VS/TES/GS with static rasterizer discard

  • ac/nir/lower_ngg_mesh: emit 0 primitives if rasterizer discard is set

  • mesa: clamp level count to the allocated level count in mipmap generation

  • nir/lower_io_indirect_loads: skip lowering if no indirect input loads exist

  • nir/lower_io_indirect_loads: add an option to skip lowering convergent loads

  • radv: skip indirect draws with (max)drawCount == 0

  • docs: add a link to gpu-ratemeter

  • radv: fix RADV_FORCE_VRS (Persona 3 Reload)

  • docs: discourage AI-generated code comments, commit messages, gitlab comments

  • ac/nir: add ac_nir_lower_fs_input_loads for indirect FS input indexing

  • radv: scalarize IO even if optimizations are disabled

  • aco,radv,radeonsi: use ac_nir_lower_fs_input_prim_mask

  • radv: enable subgroup-convergent indirect FS input indexing

  • radv: always allow RADV_FORCE_VRS on gfx11+

  • radv: fix sRGB color clears via compute

  • radv: add a drirc option to select the image clear/copy/blit/resolve path

  • radv: allow RADV_PERFTEST=nggc to affect Raphael, mention it explicitly

  • radv: precompute HW representation of custom sample locations

  • radv: precompute XMAX_RIGHT_EXCLUSION/YMAX_BOTTOM_EXCLUSION early

  • radv: allow small primitive NGG culling for user sample locations when possible

  • radv: implement radv_nggc_face_is_ccw by swapping cull face bits

  • ac/nir: remove load_cull_ccw_amd intrinsic

  • radv: don’t generate face culling code if face culling is static & disabled

  • radv: disable shader face culling when Y inversion differs between viewports

  • radv: don’t use VARYING_BIT_VIEWPORT_MASK

  • radv: explain NGG culling + conservative raster, cull anyway with raster-discard

  • radv: don’t generate culling code if overestimated raster is static & enabled

  • radv: allow NGG small prim culling with multiple viewports if they’re compatible

  • ac,radv,radeonsi: rename cull front/face names to negative/positive determinant

  • amd/drm-shim: default shims to renoir

  • ac: sort devices in amdgpu_devices[] by gfx version and radeon_family

  • ac/tools: use | as delimiter in the tiling layout table

  • ac/tools: replace _ in device names with space in the tiling layout table

  • ac/tools: indicate ARRAY_MODE as “am” instead of “sw”

  • docs/amd/hw: document all image 2D tiling layouts in terms of X/Y bit swizzle

  • ac: explain FMASK

  • ac/surface: fix incorrect ac_surface_get_plane_stride for bpe=12

  • compiler: add UINT16 sampler and image types

  • radv: fix checking whether a FS is present

  • radeonsi: dump cs_blit shader keys when dumping shader keys

  • ac/nir/meta: use typedef for structs & unions

  • ac/nir/meta_cs_blit: explain it

  • ac/nir/meta_cs_blit: use nir_var_image for images, not nir_var_uniform

  • ac/nir/meta_cs_blit: expose the number of used user data terms

  • ac/nir/meta_cs_blit: expose dispatched num_invocations[] and clean up

  • ac/nir/meta_cs_blit: pack start_x/y/z more tightly

  • ac/nir/meta_cs_blit: pass the variable workgroup size via packed user data

  • ac/nir/meta_cs_blit: add an option to clear only sample 0 of MSAA images

  • ac/nir/meta_cs_blit: skip FMASK loads if FMASK is disabled or identity

  • ac/nir/meta_cs_blit: move the DCC image store support condition to radeonsi

  • ac/nir/meta_cs_blit: get image derefs only once

  • ac/nir/meta_cs_blit: add an option to get the src image from a sampler binding

  • ac/nir/meta_cs_blit: fix possible uint32_t wraparounds and undefined behavior

  • ac/nir/meta_cs_blit: pack user data SGPRs better to reduce their number

  • ac/nir/meta_cs_blit: add local variable dst_bpe

  • ac/nir/meta_cs_blit: indent code to prepare for 96-bit formats

  • ac/nir/meta_cs_blit: add 96-bit (RGB32) image clear/copy support

  • ac/nir: generate better code for ac_get_global_ids

  • ac,radv: don’t ignore START/STOP_PIPELINE_STATS on compute queues

  • ac: remove duplicated AC_BARRIER_ALL_COMPUTE flags

  • radv: rename gfx9_fence_bo -> eop_fence_bo, etc.

  • ac: rename cb_db_* variables to eop_* variables

  • ac: set AC_RGP_FLUSH_WAIT_ON_EOP_TS and add assertions

  • ac: don’t use GLM and GL1 bits for EOP events on gfx12

  • ac: add L2 flushing into the EOP event for GFX8

  • radv: allocate the EOP fence buffer on GFX6-10.3 and compute queues

  • ac,radv: add a new barrier flag AC_BARRIER_SYNC_BOTTOM_OF_PIPE

  • radv: use optimal flush flags to wait for query writes

  • radv: don’t invalidate the VMEM cache after queries and timestamps

  • radv: don’t invalidate L2 after timestamps

  • radv: don’t invalidate L2 in end_query

  • radv: reduce L2 flushing before COPY_DATA for queries

  • radv: reduce cache flushing in radv_query_shader

  • radv: sync BVH build with PFP_SYNC_ME to fix GPU hangs

  • ac/nir/clear_copy_buffer: add optimal (and optional) logic for compute queues

  • ac/nir/clear_copy_buffer: add ac_cs_clear_copy_buffer_dispatch::cpdma_supported

  • ac/nir/clear_copy_buffer: split a dst_is_sparse block into 2

  • ac/nir/clear_copy_buffer: execute assertions for overridden dword_per_thread

  • ac/nir/clear_copy_buffer: remove unused prefer_cp_dma_threshold

  • ac/nir/clear_copy_buffer: make validation failures fatal

  • radv: add drirc option radv_buffer_meta_path

  • radv: don’t use CP DMA on compute queues

  • ac/gpu_info: add a flag indicating the CP DMA unaligned copy perf issue

  • radv: enable syncing CP DMA on gfx6

  • radv: don’t wait for previous CP DMA operations before CP DMA

  • radv: set dma_is_busy correctly

  • radv: add radv_cmd_buffer::is_mec

  • radv: remove dead predication from CP DMA

  • radv: inline radv_cs_emit_cp_dma

  • radv: remove no-op code setting CP_DMA_SYNC for copies

  • radv: inline radv_cp_dma_prepare

  • radv: remove unused radv_cp_dma_prefetch

  • radv: don’t prefetch shaders via CP DMA on MEC

  • radv: don’t sync CP DMA on MEC

  • radv: remove the CP DMA sync after every buffer fill

  • radv: use AC_BARRIER_SYNC_CP_DMA in 2 commands

  • radv: use AC_BARRIER_SYNC_CP_DMA for barriers

  • radv: add radv_gfx11_full_size_cp_dma drirc option

  • radv: improve the comment before CP_DMA_SYNC

  • radv: don’t invalidate L2 before CMASK/HTILE/DCC CP DMA clears on GFX7-8

  • radv: enable L2 for CP DMA copies on GFX7-8 to skip L2 invalidation

  • radv: enable L2 for CP DMA clears on GFX7-8 to skip L2 invalidation

  • radv: don’t sync PFP->ME in radv_emit_cp_dma

  • radv: sync CP DMA by setting the SYNC bit in the last packet if it’s close

  • radv: always use compute for buffer meta ops if CP DMA can’t use L2

  • docs: s/patch/commit/ in most places

  • radv: adjust comments in radv_layout_fmask_compression

  • radv: fix compute MSAA copies with FMASK and radv_image_meta_path=2

  • radv: clean up radv_is_clear_rect_full

  • radv: move support_comp_to_single checking out of radv_is_clear_rect_full

  • radv: treat image_deref_descriptor_amd as descriptor use

  • radv: fix the fast color clear format

  • radv: simplify initial code in radv_cmd_clear_image

  • radv: use better formats for memcpy-like image copies

  • radv: always try to perform fast clear in radv_cmd_clear_image

  • radv: sink/deduplicate code into radv_cmd_clear_image

  • radv: determine the layer range sooner in gfx_or_compute_copy_image

  • radv: determine the formats sooner in gfx_or_compute_copy_image

  • radv: do DCC decompression sooner in gfx_or_compute_copy_image

  • ac: explain HTILE a bit

  • radv: fix to use HTILE fills instead of RMW shaders to restore previous behavior

  • radv: invalidate the VMEM cache before HTILE read-modify-write dispatches

  • radv: don’t ignore the layer range when clearing HTILE levels individually

  • radv: rename radv_expand_depth_stencil -> radv_decompress_depth_stencil

  • radv: add explicit options to preserve HTILE fields with radv_clear_htile

  • radv: rename radv_clear_htile -> radv_clear_htile_rmw

  • radv: decompress depth+stencil aspects together for Z/S copies on the gfx queue

  • radv: always clear HTILE after compute-based Z decompression

  • radv: move clearing the HiZ range to the end of radv_decompress_depth_stencil

  • radv: use RADV_HTILE_PRESERVE_* flags in radv_decompress_depth_stencil

  • radv: fix Z/S HTILE metadata fixups for Z copies

  • radv: make radv_fixup_copy_dst_metadata non-static

  • radv: use radv_meta_fixup_dst_metadata in radv_fixup_resolve_dst_metadata

  • amd/ci: update results for HTILE fixes

  • aco: fix packing negated 16-bit constants in neg_mul_to_i16_cb

  • ac/gpu_info: calculate min/max_good_cu_per_sa correctly

Markuss Broks (1):

  • panfrost: Advertise Mali-G68 support

Martin Abente Lahaye (1):

  • kopper: Add getCapability to Kopper loader for RGBA ordering

Martin Roukala (né Peres) (16):

  • zink/ci: renable the unvanquished traces on radv

  • ci: allow downstream pipelines to run only the wanted hardware jobs

  • docs/ci/kernel: update the kernel uprev documentation

  • freedreno/ci: correctly tag the qualcomm jobs

  • radv/ci: document more flakes

  • radv/ci: raise back the lockup timeout to the previous level

  • zink/ci: update the NVK expectations

  • nvk/ci: update expectations

  • radeonsi/ci: update expectations

  • ci: disable the valve farms

  • radv/ci: document more flakes in stoney-angle

  • radeonsi/ci: update the vangogh expectations

  • radv/ci: switch to the gfx-ci/linux kernels

  • Revert “radv/ci: disable gfx1201 & navi21 merge jobs”

  • Revert “ci: take kws farm offline for maintenance”

  • Revert “ci-tron: bypass caching proxy for docker images”

Marvin Friedrich (1):

  • mesa/st: don’t pass stale VBO offsets as user pointers in core profiles

Mary Guillemard (55):

  • nvk: Do not take base into account when lowering emulated attributes

  • nvk: Do CS invocations calculation in the macro

  • nvk: Implement CmdDispatchIndirect via indirect cmbufs

  • nvk: Reenable compression support on Turing with nouveau 1.4.3

  • nak: Add helpers for dependent QMD

  • nak: Add QMD dependence counter

  • nak: Move shader I/O lowering out of nak_postprocess_nir

  • nvk: Optimize varyings

  • nvk/ci: Update expectations

  • nak: Fix nvdisasm tests on SM70

  • nak: Disallow MUFU.TAN on SM70

  • nak: Skip scheduling infos on disassemble_instrs

  • nak: Add support in nvdisasm_tests for all gens supported by nvdisasm 12

  • nvk: Do not open primary fd when KHR_display is not enabled

  • nak: Add F2FP uniform variant

  • nak: Add a test for F2FP in nvdisasm_tests

  • nak: Do not encode a RZ for src2 on FMNMX

  • nak: Implement new uniform variants for SM120+ instructions

  • nak: Add test_float_ops in nvdisasm_tests

  • nak: Add a test for IABS in nvdisasm_tests

  • nak: Add a test for IMNMX in nvdisasm_tests

  • nvk: Expose BAR as host cached

  • Revert “nvk: Expose BAR as host cached”

  • asahi: Add the same workaround as Firefox for Zen browser

  • asahi/ci: Update failures after VKCTS 1.4.6.1 uprev

  • hk: Advertise VK_KHR_shader_untyped_pointers

  • hk: Move to CmdEndRendering2KHR

  • hk: Advertise maintenance10 format features

  • hk: Advertise VK_KHR_maintenance10

  • hk: Advertise VK_KHR_shader_fma

  • hk: Advertise VK_EXT_primitive_restart_index

  • docs/new_features: Add new extensions supported by HK

  • agx: Fix f2i8 downcast

  • nouveau/winsys: Stop using atomic_uint_fast32_t

  • util: Add p_atomic_dec_not_one helper

  • nvk: Advertise VK_KHR_internally_synchronized_queues

  • nvk: Implement VK_KHR_maintenance11

  • nvk: Implement VK_EXT_primitive_restart_index

  • nvk: Add system values for interlock

  • nak: Use official names for inner coverage and interlock sph bits

  • nak: Track interlock ticket reads via load_sysval_nv

  • nak/hw_tests: Adjust test_render_enable for pre-Turing

  • nvk/nvkmd: handle exec_count == 0 in nvkmd_nouveau_exec_ctx_exec

  • nouveau/mme: Add dependency on mthd for DREAD/DWRITE

  • nouveau/mme: Mark DREAD/DWRITE as having implicit imm value on Blackwell+

  • nouveau/mme: Adapt disassembler for DREAD/DWRITE changes on Blackwell+

  • nouveau/mme: Adapt simulator for DREAD/DWRITE changes on Blackwell+

  • nouveau/mme: Fix scratch_share_3d_to_compute test

  • nvk: Do not overwrite prev_subc if no method were emitted

  • vk/meta: Handle color to depth equivalents in convert_texel

  • nvk: Do not directly use depth formats for meta copies

  • agx: Skip signext optimisation in case of absolute

  • hk: Properly handle pCounterBuffers being NUL

  • hk: Stop adjusting the index buffer range

  • libagx: Fix libagx_memcpy_to_aligned handling of unaligned start

Matt Arsenault (1):

  • ac/llvm: Use amdgpu subarch triple for LLVM >= 24

Matt Turner (31):

  • mesa/st: use concrete formats for lowered YUV plane views

  • dri: use concrete formats for the GR88, 8888 and sRGB DRM formats

  • dri: allocate MSAA color buffers for drisw image loader buffers

  • llvmpipe: don’t skip draw’s viewport clipping without the linear rasterizer

  • mesa: use a byte-array format for GL_RGBA8 texture buffers

  • mesa: store boolean custom glGet values in value_bool

  • mesa: pack GL_MAX_COMBINED_DIMENSIONS without depending on byte order

  • dri: describe GR1616 as R16G16_UNORM

  • util: fix the Z32_FLOAT_S8X24_UINT clear value on big-endian

  • util/format: store R11G11B10_FLOAT and R9G9B9E5_FLOAT in host byte order

  • util/astc: read the 128-bit block as little-endian words

  • util/format: make the FXT1 codec independent of host byte order

  • gallivm: reinterpret image texels using the shader’s format qualifier

  • gallivm: don’t use the s3tc and rgtc JIT decoders on big-endian

  • gallivm: pick the right lane for image stores wider than 32 bits on big-endian

  • gallivm: fix first_active_invocation() on big-endian

  • gallivm: fix unpack_64_2x32_split for scalars on big-endian

  • gallivm: only use vmaxfp for 32-bit floats

  • gallivm: don’t use vminfp and vmaxfp when NaN operands must be ignored

  • gallivm: use VSX for rounding when available

  • gallivm: don’t use vminfp and vmaxfp when VSX is available

  • llvmpipe: test min and max with denormals in lp_test_arit

  • nir, glsl, gallium: keep the halves of lowered 64-bit IO in memory order

  • mesa/st: take lowered 16-bit YUV view formats from the plane resources

  • gallivm: don’t emit a scalar llvm.nearbyint on PowerPC

  • gallivm: truncate integer constants to the type width

  • mesa: zero pipe_memory_info before querying it

  • mesa: don’t read past the name when matching a resource without its [0]

  • nir: don’t hardcode dedup_entry size in static_assert

  • nir: allow nir_cmat_tensor_load to be smaller than its const_index slots

  • llvmpipe: hash cpu caps up to num_L3_caches, not 5 dwords

Matthieu Moatti (1):

  • pan/bi: Restore VA_MEMORY_ACCESS_FORCE for spills and refills

Mauro Rossi (3):

  • lavapipe: Fix gnu-empty-initializer error in 0409984f

  • jay: fix clang building error with braces

  • radv: Fix gnu-empty-initializer error in 6cad4ee1

Maíra Canal (2):

  • broadcom/ci: update RPi 3’s skip list

  • v3dv: drop the “accum_fd” assertion in merge_syncobjs()

Mel Henning (45):

  • nak: Serialize carry access in instr_sched_prepass

  • subprojects: Use CDN URLs for crates.io

  • vk: Round for WHOLE_SIZE in vk_common_CmdFillBuffer

  • vk: Split out helper for copy_buffer_to_image2

  • vk: Split out helper for copy_image_to_buffer2

  • vk: Split out helper for copy_buffer2

  • vk/meta: Support memory_to_image / image_to_memory

  • vk/meta: Support copy_memory

  • vk/meta: Support fill_memory

  • nvk: Implement vkCmdCopyMemoryToImage/ImageToMemory

  • nvk: Implement CmdCopyMemoryKHR

  • nvk: Implement CmdFillMemoryKHR

  • panvk: Implement vkCmdCopyMemoryToImage/ImageToMemory

  • panvk: Implement CmdCopyMemoryKHR

  • kk: Implement CmdFillMemoryKHR

  • vk/meta: Remove vk_meta_fill_buffer

  • ci: Add .gitattributes to .build-rules

  • docs: Use merge=union for new_features.txt

  • nvk: Implement CmdCopyQueryPoolResultsToMemoryKHR

  • nvk: Handle VkMemoryRangeBarriersInfoKHR

  • nvk: Implement CmdUpdateMemoryKHR

  • nvk: Implement CmdWriteMarkerToMemoryAMD

  • nvk: Advertise VK_KHR_device_address_commands

  • nvk: Implement WFI for the video queue

  • vk/meta: Skip helper invocations for buffer_to_image

  • vk/meta: Fix z offset on gfx buffer_to_image

  • nvk/meta: Allow gfx for CmdCopyMemoryToImage

  • vk/meta: Match b2i and i2i copy offset calculation

  • vk/meta: Expand comment in copy_image_prepare_gfx_push_const

  • nvk: Fix memory leak in compute save/restore

  • nvk: Fix leak of dev->heap and dev->replay_heap

  • nvk: Increase pushbuf limit in CmdBeginRendering

  • nir: Handle unaligned base in lower_mem_access_bit_sizes

  • nak: Only use copy for cbuf loads of >= 32 bits

  • nir: Add nir_format_unorm24_to_float

  • nvk: Assert there’s no zcull plane for CE copies

  • nvk: Don’t zero base_array_layer for 3D depth

  • nvk: Remove base_array_layer out param

  • vk/meta: Support copying into depth attachments

  • nvk: Use meta gfx for depth copies

  • nak: Ignore texture_offset for input attachments

  • nvk: Advertise FILTER_MINMAX for integer formats

  • nvk: Advertise blitting into depth/stencil

  • nil: Mark more formats as buffer formats

  • nil: Add a few missing formats

Mengyang Liu (2):

  • radv: remove redundant draw packets for multiview meta clear

  • aco: demote exec if condition is all-ones

Michael Cheng (18):

  • intel: Fix DEBUG_FS_SIMD mask

  • intel: Fix operator precedence in intel_simd_overridden

  • brw: add brw_fs_prog_key::prefer_simd32 to keep SIMD32 through the throughput tie

  • anv: add per-shader brw_prefer_simd32_fs driconf option

  • anv: add infrastructure to force tagged compute shaders to SIMD32

  • util/drirc: change min_value to max_value typo

  • anv: key the pipeline and disk caches on shader_binary_uuid

  • anv: add the compiler configuration to the cache key

  • intel,anv: add anv_disable_jay driconf option

  • anv: use brw for 3DMark Wild Life Extreme

  • intel/compiler: use LICM instead of GCM for jay

  • intel/compiler: skip code motion when linking shaders

  • anv: add an intel_code_motion driconf option

  • anv: use GCM instead of LICM for Palworld

  • anv: use GCM instead of LICM for Total War: Pharaoh on Xe3

  • intel/measure: don’t skip deferred file open in cpu mode

  • intel/compiler: add filter for speculative hoisting

  • anv: Revert titles using GCM via driconfs

Mike Blumenkrantz (188):

  • zink/ci: switch zink/anv jobs to surfaceless+i915

  • lavapipe: reset compute shader info in emit_compute_state()

  • lavapipe: zero compute state for every mesh draw

  • llvmpipe: force alignment for Windows texture allocations

  • ntv: improve deref type matching when converting vri derefs

  • vulkan/wsi: handle other flags in VkSwapchainFlagsSurfaceCapabilitiesEXT

  • lavapipe: implement independentResolve/independentResolveNone

  • lavapipe: independently calculate src/dst blit depth

  • lavapipe: add more layouts to lvp_host_copy_image_layouts

  • lavapipe: handle planar views as rendering attachments

  • llvmpipe: use os_free_aligned for non-vma case

  • lavapipe: update returned size in GetPipelineCacheData

  • draw: plumb through vk tess mode semantics

  • draw/llvmpipe: plumb through a bool to flip tess ccw

  • lavapipe: use mallinfo2() where available for better memory budget

  • lavapipe: account for memfd size in memory budget

  • zink: don’t claim VRAM on cpu implementations

  • zink: check swapchain timline when pruning

  • zink: stop leaking swapchains

  • zink: clamp present regions to 0x0 when out of bounds

  • zink: ci updates

  • zink/kopper: fix check for async presents

  • zink/kopper: wait on async presents when updating swapchain

  • zink/kopper: stop leaking swapchains harder

  • zink/kopper: do a full glthread/tc sync when setting swap interval

  • zink/kopper: use the actual current image extent when setting swap interval

  • zink: return on flush_frontbuffer if swapchain is kill

  • zink/kopper: fix return value in acquire

  • zink: remove extra fence wait in flush_frontbuffer

  • kopper: move event selection for present into kopperGetSyncValues

  • zink/kopper: add locking around ANI and QP

  • zink/kopper: fix invalid read during swapchain teardown

  • kopper: move kopper_interface.h into gallium

  • dri: drop dead extension tables, types, and fields

  • dri: dispatch dri_drawable backend calls on screen type

  • dri: fix kopper_init_screen stub signature

  • dri: pass loader interfaces as a struct, not an extension list

  • dri: drop __DRIextension base from loader-only interfaces

  • mesa: call st reset status and program binary code directly

  • dri/sw: drop trivial put_image wrappers

  • egl/dri2: drop get_dri_drawable from the platform vtable

  • dri: parse context attribs directly into st_context_attribs

  • dri: fold driCreateNewContextForAPI into driCreateNewContext

  • dri: merge dri_create_image_with_modifiers into dri_create_image

  • mesa: call renderbuffer storage allocation directly

  • gallium: replace pipe_frontend_screen::get_param with a flag

  • dri/sw: use dri_image_fence_sync instead of a local copy

  • dri: share image list binding between dri2, drisw, and kopper

  • kopper: stop including mesa_interface.h from kopper_interface.h

  • tc: add batch param to method calls

  • tc: align driver-thread renderpass_info onto new cacheline

  • tc: implement new pipe_context::draw_vbo_buffers

  • zink: implement pipe_context::draw_vbo_buffers

  • util/texcompress_rgtc: zero-initialize encode variables

  • ci: stop building with glx=xlib

  • vl: delete the xlib swrast winsys

  • gallium: delete the xlib GLX frontend and the glx=xlib option

  • gallium: delete the xlib sw winsys

  • mesa: delete _mesa_copy_context and _mesa_copy_texture_state

  • docs: remove the xlib driver page and references to glx=xlib

  • zink: make zink_bind_vertex_addresses() fill only the bindings in use

  • zink: bind_index_buffer helper using CmdBindIndexBuffer3KHR

  • zink: group zink_resource_object hot fields in the first cache line

  • zink: group hot zink_resource per-bind fields into one cacheline

  • vk/android: replace %#b with %#x

  • util/tc: add PIPE_FLUSH_INTERMEDIATE for internal flushes

  • zink: do not flush clears during pipe_context flush for INTERMEDIATE flushes

  • zink/sync: plug an extremely tricky hole in reordering logic

  • st/mesa,nir: optimize GL_CLAMP lowering

  • zink: disable sampler2DViewOf3D

  • zink: only add VK_IMAGE_CREATE_2D_VIEW_COMPATIBLE_BIT_EXT sometimes

  • zink: add ci flakes

  • util/disk_cache: only subtract an evicted item if unlink succeeded

  • util/disk_cache: put guard pages around the index mapping

  • util/disk_cache: return whether evict_lru_item removed a file

  • util/disk_cache: recompute an impossible size counter on open

  • util/disk_cache: recompute the size when eviction cannot help

  • util/disk_cache: remove stale .tmp files when recomputing the size

  • util/tests: add tests for multi-file disk cache size counter recovery

  • st/pbo: skip delete_compute_state for shaders never created

  • zink: clear the generated tcs nir pointer after it is freed

  • zink: check unflushed under the usage mutex in the unflushed wait

  • zink: take ralloc_lock around screen-parented rallocs

  • zink: stop parenting the screen’s semaphore arrays to the screen

  • zink: make fence state changes atomic for batch submit and reset

  • zink: block in fence finish until the batch has been submitted

  • zink: serialize unthreaded batch submission per screen

  • zink: check the submit count when matching a deferred fence

  • zink: lock access to the batch state’s tc fence list

  • zink: read the fence’s submit count before flushing the batch

  • zink: fix batch usage checks racing with the submit thread

  • zink: loop on the condition in the unflushed usage wait

  • zink: use an absolute deadline for the unflushed usage trywait

  • zink: use a monotonic condition variable for batch usage waits

  • zink: force sync after unbind, even with GENERAL layouts

  • zink: ensure descriptor buffers stay mappable without BAR alloc

  • zink: avoid constantly failing BAR allocations on systems without ReBAR

  • anv/video: list modifier tiling in decode output video format properties

  • glx: keep GLX windows and pbuffers client-side on NVIDIA X servers

  • glx: match server MSAA fbconfigs that differ only in tfp, accum, or sRGB

  • vbuf: rename fallback_always to needs_fallback

  • cso,u_vbuf: check the primitive mode per draw in cso_draw_vbo

  • mesa: take the direct tc draw path only for supported modes

  • cso,st/mesa: send vertex formats that need translation to u_vbuf

  • cso,u_vbuf: threaded drivers skip u_vbuf when unneeded

  • zink: update the pipeline when the draw mode needs a different one

  • zink: update the last vertex stage variant when its key changes

  • lavapipe: add ci skip/flake

  • tc: only scan in-flight buffer lists in tc_is_buffer_on_busy_list

  • tc: scan constant buffer bindings only up to the highest bound slot

  • tc: stop asking the driver about busy buffers after 8 busy replies

  • zink: limit how many batches a context queues for the flush thread

  • tc,zink: force driver flushes by batches since the last driver flush

  • util/u_queue: add a flag to wake blocked producers at half-empty

  • zink: get the screen from the resource in transfer_unmap

  • glx: don’t modify the server fbconfig while matching driver configs

  • util/sparse_array: inline the fast path of util_sparse_array_get

  • util: clear the lowest set bit in u_bit_scan without a shift

  • mesa: inline the buffer lookup in bind_buffer_object

  • zink: emit global binding barriers before setting batch usage

  • zink: count every held swapchain image against the acquire limit

  • zink: stop retrying swapchain acquires that cannot succeed

  • zink: stop spinning on swapchain readbacks that can’t acquire an image

  • zink: don’t acquire a new swapchain image to read presented content

  • zink: mark swapchain readback copies stale on GENERAL layout barriers

  • zink: create swapchain readback copies as stale

  • zink: use readback copies when a presented image cannot be reacquired

  • zink: wait for the acquire before presenting an image for readback

  • zink: recycle present semaphores when their image is acquired again

  • zink: resolve swapchain blit sources before any blit path reads them

  • zink: only try DISJOINT for multi-planar formats in find_good_mod

  • wgl: keep the GALLIUM_DRIVER name valid for driconf

  • wgl: don’t wait for the gpu on zink swaps

  • wgl: don’t choose pbuffer-only pixel formats for windows

  • anv+hasvk: keep shaderInt8 with no_16bit

  • lavapipe: only advertise VK_EXT_headless_surface with WSI

  • zink: only enable VK_EXT_headless_surface with VK_KHR_surface

  • mesa: raise INVALID_VALUE for bad msrtt texture sample counts

  • mesa: raise INVALID_ENUM for msrtt textargets that are not 2d or a face

  • mesa: raise the OVR_multiview draw errors for xfb and extra stages

  • mesa: allow multiview attachments of any texture level

  • mesa: give RenderbufferStorageMultisampleEXT the INVALID_VALUE sample limit

  • mesa: allow reading from EXT_multisampled_render_to_texture renderbuffers

  • mesa: don’t mix implicit and explicit multisample attachments

  • zink: create msrtss transients with only the attached level

  • zink: reload msrtss transients after transfer writes to the image

  • radeonsi: capture whole frames again with frame and file triggers

  • radeonsi: don’t start a capture while reading one back

  • radeonsi: keep the trace buffer when it can’t grow

  • radeonsi: start frame and file captures where a frame begins

  • radeonsi: re-emit the sqtt pipeline when shader states are emitted

  • mesa/st: pass PIPE_IMAGE_ACCESS bits to make_image_handle_resident

  • zink: share bindless handles between contexts

  • zink: track bindless image access per context

  • zink: drop bindless residency on context destroy

  • zink: fix bindless image buffer descriptors

  • zink: index bindless buffer arrays without the handle offset

  • zink: emit the dynamic line rasterization mode when it changes

  • zink: clear the last ds3 rasterizer and blend bits after emitting them

  • zink: consistently use atomics for ctx->robust_ctx_count access

  • zink: unlock the copy context when timestamp query creation fails

  • zink: don’t update unordered state in get_cmdbuf() in the presence of current usage

  • zink: create mesh full programs when replacing mesh separable programs

  • zink: fix the color write mask of blend states

  • zink: don’t read a missing coord src when lowering bindless textures

  • zink: use one bindless sampler variable per sampler type

  • st/mesa: don’t release the shared pixel map view after glDrawPixels

  • st/mesa: add deref sources to the second pixel map lookup

  • st/mesa: read the b and a pixel maps from the right channels

  • st/mesa: use a new pixel map texture for each update

  • zink: copy the spirv when compiling injected tcs

  • zink: add helpers for locking bo memory

  • zink: lock bo memory around memory binds

  • zink/kopper: fix stale swapchain indices shared between resources

  • st/mesa: let the color buffer define the framebuffer size

  • zink/kopper: use the current swapchain size when fixing up the framebuffer

  • zink: hold a reference on the context swapchain resource

  • mesa: free the trace dispatch table with the other tables

  • util/tc: don’t full_copy an rp info that has ended

  • util/tc: only set end_call for a blit that ends the info

  • util/tc: advance the driver rp info before the first draw or clear

  • util/tc: advance the driver rp info for indirect and vstate draws

  • util/tc: move the driver rp info advance call list into a helper

  • util/tc: write the resolve geometry before enqueueing the resolve

  • util/tc: don’t merge a resolve into a different target

  • util/tc: parse the rp info for mesh draws

  • util/tc: don’t end the rp info without an end_call

  • util/tc: reset the new rp info fence before linking it

Mikhail Mlhcnn (5):

  • panvk/csf: flush the FAU stores before indirect draws read them

  • panfrost: add the missing v11 cases to the arch switches

  • panvk/csf: give compute jobs the TLS size of the shared allocation

  • util/tc: reset end_call when rp info is reused after sync

  • util/tc: fix buffer over-read and thread race in staging texture_subdata

Mohammadzafar Ziya (2):

  • pvr: implement u_trace support

  • pvr: Instrument draw and transfer u_trace markers

Nanley Chery (32):

  • anv: Allow 2MB page oversubscription for 1MB+ BOs

  • anv: Add a non-NULL anv_instance to some anv tests

  • anv: Convert oversubscription ANV_DEBUG to drirc

  • anv: Rework the aux-disabling of disjoint images

  • anv: Share private BOs of overlapping images

  • anv: Allow CCS_E on more aliased images

  • anv: Allow aliasing of images with tiled modifiers

  • anv: Enable buffer compression on DXVK

  • anv: Drop the anv_enable_buffer_comp drirc option

  • anv: Allow CCS_E without a format list more often

  • isl: Use ternary in HALIGN selection on Xe2+

  • isl: Drop the HALIGN_16 restriction on Xe2+

  • isl: Drop HALIGN_128 restriction for Xe2+ CCS

  • iris: Prevent VALIGN_4 fast-clears for workaround

  • anv: Prevent VALIGN_4 fast-clears for workaround

  • isl: Avoid VALIGN_8 WA for compressed surfaces Xe2+

  • intel/isl: Support MSAA in subimage range helper

  • intel/isl: Split surface size calculations in two

  • intel/isl: Omit unused space when sizing mipmaps

  • intel/ci: Update expectation for RPL

  • intel/isl: Bump HALIGN from 16B to 32B on Xe2+

  • blorp: Convert fast clear color on all platforms

  • isl: Add entries to gfx12.5 compression format helper

  • isl: Add CCS_E support for three formats on gfx12.5

  • anv: Allow CCS_E without a format list on gfx12.5

  • intel/isl: Print more data with INTEL_DEBUG=isl

  • intel: Rework the standard-tiling helpers

  • anv,iris: Prefer 4K tiles for memory objects

  • intel/isl: Prefer larger tilings when sizes match

  • intel/isl: Allow Tile64 with 10% wasted padding

  • intel/ci: Update expectation for RPL

  • isl: Drop a standard tiling assertion

Natalie Vock (16):

  • radv/nir: Clean up descriptor index lowering

  • radv: Expose mutable acceleration structure descriptors

  • radv/gamma: Fix geometry counts

  • tool/gamma: Add virtual destructor for gamma_file_view

  • aco/validate: Allow p_startpgm in resume blocks as well

  • radv/nir: Loosen tail call parameter requirements for callers that don’t return

  • radv/rt: Apply call lowering passes to all stages

  • radv/rt: Inline any-hit shaders for CPS

  • radv/rt: Implement missing intrinsics for CPS

  • radv/rt: Make CPS signature independent from has_position_fetch

  • radv/rt: Make radv_nir_init_cps_function public

  • radv/rt: Use CPS function signature for raygen if CPS is used

  • vulkan/bvh: Use custom merge-sort for morton code sorting

  • vulkan/bvh: Add vk_bvh barrier helpers

  • vulkan,anv,tu,lvp,radv: Delete radix_sort_vk

  • radv: Document ACE DGC L2 invalidation on GFX8 better

Neha Bhende (5):

  • svga: fix shared memory index for svga driver

  • svga: fix shader leakage for emulated code path

  • svga: fix vertex buffer stride (#42)

  • svga: flip y for emulated code

  • svga: do y flip back only for non negative svga->curr.viewport[0].scale[1]

Nick Hamilton (6):

  • glthread: fix check for unroll draws using user VBOs when the ctx supports GLES

  • pvr, ci: Update axe failure list

  • zink: Handle max_varying where maxVertexOutputComponents value is too low.

  • pvr/ci: Update AXE failing list

  • pvr, ci: Update axe failing list

  • cso: Add quirk that stops adding samplers with border color to the cache

Nikita Konton (1):

  • radeonsi: invalidate I$/SMEM$/VMEM$ at IB start on all gfx levels

Oliver Clarke (3):

  • asahi/agx: Add shift mask for bitfield_extract and fused shift ops

  • asahi: Dump shaders for KHR_pipeline_executable_properties

  • radv: disable 8x MSAA for VRS in device properties

Olivia Lee (12):

  • panvk/csf: fix layered FB barrier in secondary cmdbufs

  • panvk/csf: support layered indirect draw in primary cmdbufs

  • panvk/csf: minor register assignment refactor in launch_draw

  • panvk/csf: support layered draws in inherited secondary cmdbufs

  • panvk/csf: advertise shaderOutputLayer

  • poly: add DrawIndirectByteCount support for GS indirect draws

  • panvk: fix null pointer dereference in GetInstanceProcAddr

  • panvk: fix invalid memcpy in nir_lower_descriptors

  • poly: add alignment parameter for heap allocation

  • panvk: hook up poly_heap

  • panvk/jm: switch indirect varying bufs to poly_heap

  • panvk: fix mark_all_access_non_uniform metadata preservation

Omar Rashwan (11):

  • anv: return VK_TIMEOUT instead of asserting on a held profiling lock

  • anv: don’t warn about a wrong-queue performance query with no stream open

  • intel/perf/xe: allow opening the global OA stream

  • intel/perf: add intel_perf_stream_map_oa_buffer()

  • anv: use the global OAG unit for KHR performance queries on Xe2+

  • anv: resolve OAG boundary reports before dropping the OA buffer mapping

  • anv: make KHR performance queries mutually exclusive with the metrics library

  • anv: avoid wrapping the OA ring over unresolved OAG boundaries

  • intel/perf/xe: cap the global stream ring at 64 MiB before Xe2

  • nir/instr_set: compare tex def when CSEing texture instructions

  • nir/instr_set: compare tex dest_type instead of bit size when CSEing texture instructions

Patrick Gaskin (1):

  • egl: don’t clobber an initialized display’s device in eglGetPlatformDisplay

Patrick Lerda (15):

  • r600: handle ssbo access coherent

  • mesa: implement max_combined_image_uniforms

  • r600: increase the number of shader buffers from 8 to 11

  • r600: update vs_as_ls sampler switch

  • r600: update split_double_load_ssbo

  • r600: disable hardware_gl_select

  • r600: implement TexInstr alt_const

  • r600: implement FetchInstr alt_const

  • r600: enable vertex image and ssbo as a first step

  • r600: refactor with struct dynamic_offset

  • r600: refactor and implement vertex atoms

  • r600: enable ssbos on the remaining shader stages

  • r600: add evergreen_emit_image_state nr_cbufs_clamp

  • r600: update the inputs to prevent an overlap

  • r600: fix barycenter coordinates overlap

Paul Grandperrin (2):

  • wsi/wayland: move wsi_wl_surface_get_present_modes() higher in file

  • wsi/wayland: send tearing hints when present mode changes to IMMEDIATE

Paulo Zanoni (11):

  • intel/isl: optimize find_next_divisor()

  • intel/isl: help gcc unify “num / i” and “num % i” into a single idiv

  • anv: bring back hdc/dataport flushes for transfer writes on Xe2

  • intel/jay: implement non-temporal stores

  • intel/isl: remove completely redundant call to isl_tiling_get_info()

  • intel/isl: don’t change tile_info in isl_tiling_get_intratile_offset_el

  • intel/isl: move some functions up in isl.c

  • intel/isl: extract isl_tile_info_get_intratile_offset_el()

  • intel/isl: extract isl_tile_info_get_intratile_range_el()

  • intel/isl: isl_surf_get_image_range_B_tile() now requires tile_info

  • intel/isl: shrink struct isl_tile_info by 8 bytes

Pavel Ondračka (49):

  • nv50_ir_ra: align B96 spill slots to vec4

  • nir/opt_vectorize: handle commutative ALU sources

  • r300: fix DXT1 RGB border color alpha

  • r300: fix varying limits on R300/R400

  • r300: limit in-flight texture upload memory

  • r300: preserve masked writes on microtiled color buffers

  • r300: replace private memory pool with linear allocator

  • r300: get rid of custom radeon_list implementation

  • r300: scalarize single-component bcsel updates

  • winsys/radeon: extract BO import ref helper

  • winsys/radeon: prefer a render node FD for internal work

  • winsys/radeon: reuse GEM_OPEN handles for display

  • r300: fall back for oversized blit rectangles

  • ci/deqp: backport GLES2 CTS fixes

  • r300: don’t propagate negation through absolute values

  • r300: emulate base-level NPOT cube sampling

  • r300: split tiled NPOT 3D uploads by layer

  • r300: lower NPOT wrap modes per coordinate

  • r300: clamp NPOT 3D coordinates to the logical edge

  • r300: clamp NPOT cube sampling to the logical face edge

  • r300: correct NPOT cube filter selection at the logical edge

  • r300: respect unnormalized sampler coordinates

  • docs/gallium: clarify npot_textures semantics

  • mesa: pass the context to _mesa_is_texture_complete

  • mesa: relax GLES2 NPOT texturing requirements

  • mesa: relax OpenGL 2.0 NPOT texturing requirements

  • st/mesa: relax TexSubImage and ReadPixels fast path NPOT requirements

  • dri: choose screen target based on supported API versions

  • ci: include drirc files in test artifacts

  • r300/ci: retain full NPOT Piglit coverage

  • r300: add +GL_OES_texture_npot override for GNOME Shell

  • r300: add +GL_OES_texture_npot override for X/glamor

  • r300: add +GL_OES_texture_npot override for SuperTuxKart

  • r300: stop advertising full NPOT texture support

  • r300: fine-tune LOD bias for nearest mip filtering

  • r300/ci: remove stale polygon offset fails

  • r300: preserve writers across loop exits

  • r300: fix packed sampler formats on big endian

  • r300: fix RGB565 CBZB clear values on big endian

  • r300: fix BE constant blend colors on R500

  • r300: fix BE constant blend colors for RGBA4

  • r300: fix BE texture border color packing

  • r300: remove stale TODO

  • r300: add npot override for kwin

  • r300: handle underflow and overflow when packing FP24 constants

  • r300: use 32-bit immediate indices when applying an index bias

  • r300: skip constant attributes when splitting a negative index bias

  • r300: clamp sampler LOD bounds before converting to integers

  • glsl: propagate invariance after lowering

Peng_Lx (1):

  • turnip/kgsl: close the dma-buf fd of our own allocations

Peter Quayle (2):

  • pvr: fix stream-out PDS data size units in CSB dump

  • pvr: increase the CSB dump nesting limit

Peyton Lee (2):

  • radeonsi/vpe: fix YUV422 output width issue

  • radeonsi/vpe: fix 3D LUT channel order and PQ tone mapping

Pierre-Eric Pelloux-Prayer (48):

  • radv: fix chained IB dump

  • radeonsi: fix typo in barrier flag check

  • radeonsi/sqtt: fix barrier flags for <= gfx8

  • radeonsi: fix typo in descriptors bitmask

  • radeonsi: add NULL-checks to si_create_vertex_state

  • radeonsi: add NULL check to si_semaphore_create

  • radeonsi: fix off-by-one cs shader indexing

  • radeonsi: give unique name to internal NIR shaders

  • radeonsi: keep the sqtt buffer mapped until ac_dump_rgp_capture is done

  • radeonsi/sqtt: NULL check cs before destroying them

  • radeonsi/sqtt: fix markers being truncated by 1

  • radeonsi: NULL check attribute_pos_prim_ring during init

  • radeonsi: add NULL check to si_create_compute_state_for_nir

  • radeonsi: avoid replacing global_buffers with a NULL ptr

  • radeonsi/test: update navi10 failures

  • winsys/radeon: replace enum radeon_map_flags by a plain define

  • radeon/winsys: pass debug flags to the screen_create callback

  • winsys/amdgpu, radeonsi: move debug option handing to radeonsi_screen_create

  • radeonsi: add meta debug options

  • radeonsi: add new debug option to force cache flushes

  • radeonsi: disable tc when AMD_DEBUG=safest is used

  • radeonsi: fix AMD_DEBUG=usellvm usage

  • loader: consider radeon in amd_predicate

  • radeonsi: remove reachable assert

  • radeonsi: pass si_shader to si_get_ir_cache_key

  • ac/info: disable userq support for SDMA

  • radv/amdgpu: return VK_ERROR_INCOMPATIBLE_DRIVER on device failure

  • radeonsi: remove invalid assert

  • radeonsi/gfx12: dont set BO metadata on import

  • winsys/amdgpu: fix import/destroy race again

  • vbo: remove unneeded preprocess tokens

  • vbo: move vbo_save_destroy call after display list deletion

  • vbo: extract two methods from compile_vertex_list

  • vbo: hoist _mesa_index_buffer

  • vbo: move in progress BO to a separate list

  • vbo: store existing VAO alongside the BO

  • vbo: track holes in existing BOs

  • vbo: fix warning and add error handling

  • vbo: rename no_current_update into restore_attrib_after_draw

  • vbo: use suballocation only for smaller buffer

  • vbo: remove dead code

  • vbo: deal with pipe_vertex_state immutability

  • vbo: make sure the BO is large enough

  • radeonsi: pass spm struct to spm functions

  • radeonsi: move sqtt data in a sub-struct

  • ac/sqtt: move radeonsi-only fields out of shared code

  • ac/sqtt: move trigger parameters to radeonsi

  • radeonsi/sqtt: allow submit-based capture

Pietro Negri (10):

  • pan: Use correct enum value for RT downscale mode

  • pan/nir: Add projection_enable flag to pan_va_tex_flags

  • kraid: Wire up projection_enable flag

  • pan/nir: Optimize Cubemaps on v11+

  • kraid/v14+: Support .zext in disassembler

  • kraid: Macro disasm_case rework

  • kraid/v14+: Add v14 zext test case

  • kraid: Optimize texture projectors on v11+

  • kraid: Use IADD for register swaps

  • kraid: Adds simple hw_test checking parallel copies

Pohsiang (John) Hsu (17):

  • mediafoundation: fix drain/flush when running ProcessDX12Context outside of m_lock, and allocator/buffer pool release.

  • mediafoundation: fix shadow variable in stats_buffer_manager.cpp, also remove using namespace std to avoid collision

  • mediafoundation: change default value for HEVC_LOG2_MAX_PIC_ORDER_CNT_LSB_MINUS4 to 10

  • d3d12: return a bigger encoding output buffer size for dx12 encoder.

  • mediafoundation: return a bigger output buffer size for hmft.

  • mediafoundation: removing dependency of pipe_screen from CD3D12BitstreamMFBuffer

  • d3d12: fix msvc build warning C4819

  • d3d12: return av1 sequence header for encoder

  • mediafoundation: shut down the event queue.

  • mediafoundation: fix av1 encode QP

  • d3d12: support AV1 420 10 bit encode

  • mediafoundation: support AV1 420 10 bit encode

  • d3d12: Allow importing resources as PIPE_USAGE_STAGING

  • mediafoundation: update version to 1.10

  • mediafoundation: fix missing pipe profile to spec profile conversion

  • mediafoundation: refactor checkmediatype

  • d3d12: add logic to log2_sao_offset_scale diagnostic checks when driver return 0 for allowed_log2_sao_offset_scale_luma_value.

Qiang Yu (1):

  • nir: model ftanh in opt_varyings and range analysis

Radu Costas (5):

  • pvr,ci: Amend mesa-ci fails lists

  • nir: Skip fminmax preserve_sz lowering if non-zero consts are used

  • pco: Use negative 0 in fceil and ffloor implementations

  • pco: Specific lowering for shader_float_controls2

  • pvr: Enable shader_float_controls2 extension

Rahul Mahantappa Bhadrashette (1):

  • radeonsi: emit MESHLET registers for mesh shaders in gfx10_emit_shader_ngg

Randy Palamar (2):

  • aco/insert_fp_mode: fix rounding mode for fma_mixhi_f16

  • docs: clarify that MESA_VK_TRACE_PER_SUBMIT is not compute only

Raviraj Uppal (1):

  • radv: reset the HiZ range in HTILE to [0, 1] after Z decompression

Rhys Perry (83):

  • people: update Natalie’s email

  • radeonsi: run nir_copy_prop after nir_lower_load_const_to_scalar

  • ac/nir/ngg: don’t shrink device-scope memory barriers

  • ac/nir/ngg: track uniformity with multiple set_vertex_and_primitive_count

  • nir/load_store_vectorize: rework barriers to use entries

  • nir/load_store_vectorize: recreate entry key when adding from predecessor

  • nir/tests: add more load/store vectorizer tests

  • vtn: don’t fail at uniformity decorations on variables and OpBufferPointer

  • radv: drop support for cooperativeMatrixRobustBufferAccess

  • radv: remove code for tracking cooperativeMatrixRobustBufferAccess

  • radv: add comment for Fossilize considerations for radv_compiler_info

  • radv: enable more features with shaderObject

  • ac/nir/lower_tex_coords: fix moving vector FS inputs loads

  • radv/nir: remove unsafe assertion

  • radv: run nir_opt_dead_cf before nir_vk_lower_ycbcr_tex

  • aco/ra: adjust adjust_max_used_regs for subdword regclasses

  • aco: don’t take do_pack_2x16 path with unaligned operands

  • aco: allow unaligned subdword copy definitions and operands

  • nir/cf: add nir_cf_list_detach_ssa

  • nir/simplify_loops: lower ssa to registers if it might be necessary

  • nir/simplify_loops: fix predication of an empty cf list

  • nir/tests: add some nir_simplify_loop tests

  • nir: update some comments

  • nir: add deref_buffer_address

  • ac/nir: implement load_ssbo_address

  • radv: use load_deref_buffer_address

  • radeonsi: remove load_ssbo_address lowering

  • ac/rgp: stop using strtok()

  • vulkan/rmv: stop using strtok()

  • mesa: stop using strtok()

  • vulkan/screenshot-layer: stop using strtok()

  • intel/dev: stop using strtok()

  • glx: stop using strtok()

  • broadcom: stop using strtok()

  • gallivm: stop using strtok()

  • glsl: stop using strtok()

  • radv: don’t set muladd_support_prefers_split when ffma is unsupported

  • radv: fix ds_att_aspects with depth-only or stencil-only renderpasses

  • radv: rework depth clamping

  • radv: disable depth clipping for slow depth clears

  • ac/nir_lower_global_access: stack allocate hash tables

  • nir/load_store_vectorize: stack allocate numlsb_ht

  • nir: add nir_is_op_nuw and nir_is_scalar_nuw

  • ac/nir: use nir_is_op_nuw/nir_is_scalar_nuw

  • nir/load_store_vectorize: recognize u2u64(iadd(…, …))

  • nir/load_store_vectorize: use nir_is_scalar_nuw

  • nir/uub: replace i2i64 with u2u64

  • nir/uub: reduce bit size of op(u2u64, u2u64)

  • nir/uub: more aggressively reduce bit size of op(u2u64, u2u64)

  • ir3: stop optimizing nir_tex_instr::offset_non_uniform

  • nir/non_uniform_access: stop using offset_non_uniform

  • nir: remove nir_tex_instr::offset_non_uniform

  • anv: remove now-unnecessary divergence analysis

  • radv: initialize shader_blake3 for shader objects

  • radv: hash descriptor set layouts in radv_hash_graphics_spirv_to_nir

  • aco,ac: add ac_shader_config::mem_ordered

  • radv: combine MEM_ORDERED

  • radv: use ac_shader_config::mem_ordered

  • aco: track what vmem types a barrier info contains

  • aco: only set MEM_ORDERED if it might be needed

  • radeonsi: use ac_shader_config::lds_size from ACO

  • radeonsi: only fill aco_shader_info::lds_size for compute stages

  • radeonsi: use ac_shader_config::mem_ordered

  • radeonsi: remove si_shader_mem_ordered

  • aco/isel: fix sub-dword emit_load() with VGPR dest and SMEM load

  • aco/isel: improve load_constant register type

  • ac/llvm: implement last_invocation

  • aco: move callback parameters into a struct

  • aco: add finish_program() helper

  • aco: simplify emit_program parameters

  • aco: have emit_program() write into aco_callback_params

  • aco: add more fields to aco_callback_params

  • aco: add function for creating an ELF

  • amd/common: parse lds_size from llvm binary config

  • radv: support ELF for ACO shader binaries

  • radv,ac: preserve float_mode for ACO ELfs

  • nir: don’t reindex SSA defs for NIR_DEBUG=print

  • nir: skip mov/vec in LSB/UUB analysis

  • aco/ra: check next register against bounds in subdword get_reg_simple path

  • radv: stop checking dirty bit in CmdExecuteCommands HiZ workaround

  • radv: flatten ifs in CmdExecuteCommands

  • aco: add a couple aco_alu_opcode_info methods

  • aco/opt_postRA: check for tied defs in try_optimize_branching_sequence

Rob Clark (95):

  • freedreno: Remove VS driver param patching

  • freedreno/a6xx: Split driver UBO upload and emit

  • freedreno/a6xx: Split VS driver params

  • freedreno/a6xx: Fix query barriers for gen8+

  • freedreno/a6xx: Early return if no flushes to add

  • freedreno/a6xx: Indirect buffer needs WFM on a8xx+

  • freedreno: Insert implicit BARRIER_INDIRECT_BUFFER

  • freedreno/a6xx: Don’t trucate subdraw size

  • tu: Don’t truncate subdraw size

  • freedreno: Fix separate-stencil tracking

  • freedreno/a6xx: Fix ms resolve blits

  • freedreno/a6xx: Fix a2d_blt_cntl setup

  • freedreno/a6xx: Rework emit_blit_setup() params

  • freedreno/a6xx: Handle blit mask

  • freedreno/a6xx: Use a2d for z24s8 blits

  • tu+freedreno: Add asserts for unsupported events

  • freedreno: Split gen7 and gen8 event table

  • freedreno: Add SUBPASS_FENCE, SLICE_SUBPASS_FENCE

  • freedreno/a6xx: Add SUBPASS_FENCE

  • ci: Move OpenCL-CTS SPIR to global skips list

  • ci: Skip CL_DEVICE_TYPE_CPU tests

  • ci/deqp-runner: Bump to 0.23.3 for CL CTS fix

  • nir/convert_address_format: Split convert_def into two passes

  • nir/convert_address_format: Handle non-deref sources

  • nir/validate: Make errors red

  • nir: Add missing lower_io case

  • Revert “freedreno: Don’t re-bind global buffers”

  • ci: Ensure debugfs is mounted

  • ci/cl-cts: Enable test_gl build

  • ci: Import OpenCL-CTS opencl_conformance_tests_quick.csv

  • ci/cl-cts: Disable test_gl until deqp-runner is fixed

  • ci/cl-cts: Wimpy Select

  • ci/cl-cts: Drop allocations tests

  • ci/cl-cts: Drop max_images and use_pitches

  • ci/cl-cts: Split int/uint/float tests

  • ci/cl-cts: Further split 1D/2D/3D

  • panfrost/ci: Switch to opencl_conformance_tests_mesa.csv

  • freedreno/ci: Enable CL CTS

  • ci: Uprev OpenCL-CTS

  • freedreno/ci: Document cl-cts test bugs

  • freedreno/fdl: Fix gen8 multi-plane descriptor base

  • freedreno/decode: Add ray_query test

  • freedreno/decode: Fix CP_REG_RMW handler name

  • freedreno/registers: Rename “64B”

  • freedreno/decode: Remove error message

  • freedreno/decode: __index should be on metatable

  • freedreno/decode: Handle scratch <-> reg

  • freedreno/decode: Handle CP_REG_RMW.DST_SCRATCH

  • freedreno/decode: Handle CP_REG_TO_MEM and CP_MEM_TO_REG

  • freedreno/decode: Handle CP_SCRATCH_WRITE

  • freedreno/decode: Add lua exception handling

  • freedreno/decode: Add missing CP_RUN_OPENCL case

  • freedreno/registers: Rename CP_MEM_TO_OC_MEM

  • tu: Fix gen8 TEX_LINE_OFFSET

  • freedreno/registers: Name WAIT_CACHE_FLUSH

  • freedreno/registers: Name ONE_REG_WR

  • freedreno/registers: Drop bogus variants attr

  • tu: Add tu_pkt7 builder

  • tu: Move reg_pair definition

  • tu: Add scratch-ram helper

  • tu: Use scratch-ram helper for tu_dispatch()

  • tu: Add CP_REG_RMW helper

  • tu: Add CP_REG_TO_MEM / CP_MEM_TO_REG helpers

  • tu: Add CP_COND_WRITE helper

  • freedreno/ir3: Lower scratch to scalar

  • ir3: Also vectorize scratch

  • freedreno/a6xx: Add perf warn if not pvtmem_per_wave

  • tu: Add perf warn if not pvtmem_per_wave

  • iris: Implement PIPE_QUERY_TIMESTAMP_RAW

  • freedreno: Add device quirks section

  • freedreno/ir3: Convert cs_lock_unlock_quirk to quirk

  • freedreno: Convert r8g8_faulty_fast_clear_quirk to quirk

  • freedreno/ir3: Implement QCTDD13523866 quirk

  • freedreno/ir3: Implement QCTDD10789828 quirk

  • freedreno/ir3: Implement QCTDD10204462 quirk

  • freedreno/ir3: Convert alias_mova_quirk to quirk

  • freedreno/ir3: Convert alias_predication_quirk to quirk

  • freedreno: Convert cmdbuf_start_a725_quirk to quirk

  • freedreno/ir3: Convert stsc_duplication_quirk to quirk

  • freedreno/ir3: Convert mov_half_shared_quirk to quirk

  • freedreno/ir3: Convert fs_must_have_non_zero_constlen_quirk to quirk

  • freedreno/common: Check for dup magic regs

  • freedreno/common: Add support to dump quirk summary

  • freedreno/common: Fix a702 chip-id

  • freedreno/decode: A bit of lastvals cleanup

  • freedreno/decode: Speed up partial decode

  • freedreno/registers: Sync xml changes back from kernel

  • freedreno: Move RB_RBP_CNTL to raw_magic_regs

  • freedreno: Implement QCTDD04536579

  • freedreno: Move RB_DBG_ECO_CNTL to raw_magic_regs

  • freedreno: Drop magic_regs

  • freedreno: Slightly better with_crb()

  • tu: Slightly better with_crb()

  • gbm: Allow GBM allocation of YUV surfaces

  • freedreno/a6xx: Fix fbfetch with msrtt

Rob Herring (Arm) (107):

  • gallium: Add explicit ML convolution activation range

  • etnaviv/ml: Reject clipped convolution activations

  • teflon: Compute explicit convolution activation ranges

  • ethosu: Emit explicit convolution activation ranges

  • teflon: Add ssd_mobilenet_v3 model

  • teflon: Add DS-CNN keyword spotting model

  • teflon: Add VWW4 model

  • teflon: Add wav2letter model

  • teflon: Normalize TFLite concatenation axes

  • teflon: Preserve fully connected ReLU activation

  • ethosu: Clamp fully connected ReLU outputs

  • ethosu: Fix U85 activation precision encoding

  • ethosu: Handle submission allocation failures

  • ethosu: Fix U85 scalar register opcode

  • ethosu: Clear zero points for average pooling

  • ethosu: Prefer wide pointwise blocks for NHWC tail outputs

  • ethosu: Trim padding folded into convolutions

  • ethosu: Preserve AvgPool state for U85 SUM pooling

  • ethosu: Match pooling scale quantization to Regor

  • ethosu: Track scalar feature map presence

  • ethosu: Read elementwise scalars by type

  • ethosu: Initialize feature maps in the IO region

  • ethosu: Emit U85 scalar feature map state

  • ethosu: Skip IFM2 state for unary elementwise ops

  • ethosu: Exclude scalar operands from memory accesses

  • ethosu: Account for IFM bit depth on U85

  • ethosu: Match FC scale precision with Regor

  • ethosu: Match elementwise multiply scale precision

  • ethosu: Align coefficient buffer allocations

  • ethosu: Load constant quantize inputs

  • ethosu: Fix concatenation output offsets

  • ethosu: Account for dilated kernel footprints

  • ethosu: Emit the scale for minimum pooling

  • ethosu: Use natural rounding for copy operations

  • ethosu: Fix rank-aligned Strided Slice offsets

  • ethosu: Match U85 aspect-ratio area fitting

  • ethosu: Reject operations without block configurations

  • ethosu: Fix U85 IFM allocation units

  • gallium: Mark exported ML tensors

  • teflon: Mark exported ML tensors

  • ethosu: Keep exported tensors in NHWC

  • ethosu: Enable U85 sparse convolution weights

  • ethosu: Improve U65 Conv1D block selection

  • ethosu: Fix flattened fully connected inputs

  • ethosu: Cache encoded weights within a subgraph

  • ethosu: Silence expected U85 area-fit failures

  • ethosu: Match fully connected batch layout

  • ethosu: Calculate block dependencies for all operations

  • ethosu: Support unaligned NHCWB16 feature maps

  • ethosu: Align U65 block search with Regor

  • ethosu: Route scalar constants through the coefficient allocator

  • ethosu: Prepare feature map allocation helpers

  • ethosu: Reset operation defaults for each concatenation input

  • ethosu: Move activation and LUT state to the common operation struct

  • ethosu: Support U85 activations and multiple LUT slots

  • ethosu: Support standalone pad operations

  • teflon: Support more TFLite PAD tensor encodings

  • ethosu: Reject unsupported large convolution strides

  • ethosu: Support internal reduce-sum pooling

  • ethosu: Support subtract operations

  • gallium: Add pipe ML softmax operation

  • teflon: Add softmax operation support

  • ethosu: Chain U85 activation feature maps

  • ethosu: Detect NPU wait dependencies by memory conflict

  • ethosu: Rework elementwise scale emission

  • ethosu: Emit LUT-backed and CLZ elementwise operations

  • ethosu: Add constant coefficient feature maps

  • ethosu: Add support for 8-bit softmax

  • gallium: Add pipe ML mean operation

  • teflon: Add TFLite mean operation support

  • ethosu: Lower spatial and channel-axis mean operations

  • ethosu: Fuse LUT activations into convolutions

  • ethosu: Unroll convolutions with large strides

  • ethosu: Fuse LUT activations through reshape

  • ethosu: Fuse LUT activations into elementwise operations

  • ethosu: Fuse LUT activations into fully connected operations

  • ethosu: Execute batched feature maps

  • teflon: Rebuild partitions for dynamic shapes

  • ethosu: Lower equal-sized SPLIT operations

  • teflon: Parse SPLIT operations

  • gallium: Add batch matrix multiplication operation state

  • teflon: Parse batch matrix multiplication operations

  • ethosu: Support BatchMatMul operations

  • ethosu: Support transpose operations

  • teflon: Parse Transpose permutations

  • gallium: Add reciprocal square root operation state

  • teflon: Parse reciprocal square root operations

  • ethosu: Support reciprocal square root

  • gallium: Add pack and unpack operation states

  • teflon: Parse pack and unpack operations

  • ethosu: Support pack and unpack operations

  • gallium: Add ResizeBilinear operation state

  • teflon: Parse ResizeBilinear operations

  • ethosu: Support ResizeBilinear

  • gallium: Add ArgMax operation state

  • teflon: Parse ArgMax operations

  • ethosu: Support ArgMax

  • gallium: Add space-batch operation states

  • teflon: Parse space-batch operations

  • ethosu: Fuse space-batch convolution patterns

  • pipe: Add raw-zero padding state

  • ethosu: Support raw-zero padding

  • pipe: Add ScatterND operation state

  • ethosu: Lower rectangular ScatterND operations

  • teflon: Expose ScatterND operations

  • ethosu: Generalize SPLIT output validation

  • teflon: Parse SplitV operations

Robert Krzemień (14):

  • anv: metrics library initialization

  • anv: metrics library deinitialization

  • anv: metrics library oa configuration

  • anv: metrics library query create & destroy

  • anv: metrics library query get data

  • anv: metrics library oa configuration activation

  • anv: metrics library stream marker

  • anv: metrics library query begin & end

  • anv: fixed query slot addr passed to metrics library

  • anv: fixed query slot addr passed to metrics library

  • anv: fix intel perf query cmds size

  • anv: improve intel perf query reset

  • anv: fix leak on error path in intel perf query creation

  • anv: return init failed if no metrics library

Robert Mader (3):

  • llvmpipe: Pass size to displaytarget_create_mapped()

  • kms-dri-sw: Use size and offset from whandle in create_mapped()

  • dri-sw: Use size and offset from whandle in create_mapped()

Robert Mazur (2):

  • pvr/csbgen: Add documentation support for generated headers

  • docs: Add CSBGEN API documentation for PowerVR

Robin Kertels (8):

  • kk: add function to dynamically set store actions to mtl bridge

  • kk: Set store action dynamically

  • kk: Remove redundant render pass descriptor load op change

  • kk: Add Metal img read usage for Vulkan storage usage

  • kk: Implement buffer marker AMD

  • kk: Fix double init in the Metal bridge

  • kk: Add bridge functions for resolve texture

  • kk: Do resolves using Metal render passes when possible

Robotwizard (1):

  • zink: switch EXT/KHR functions to the promoted/non-suffixed name

Rohan Garg (4):

  • pvr: use a load op for the resolve stencil load op

  • nir: derive color_is_dual_source from declared vars

  • llvmpipe: fix snorm blend when only alpha factor is inverse

  • lavapipe: drop LVP_SNORM_BLEND workaround

Romaric Jodin (2):

  • gallivm: bypass vector reduction intrinsic for f16 fmin/fmax on x86

  • gallivm: fix clamping of 32-bit unsigned pure integer image writes

Ruben Solvang (1):

  • llvmpipe: release freed memory from the allocation memfd

Ruitang Wang (1):

  • amd/common/virtio: fix import/free race that detaches a live res_id

Ryan Mckeever (8):

  • docs: advertise VK_KHR_multiview support for Bifrost

  • pan/bi: set branch reconvergence at loop exit points

  • panvk: enable Vulkan 1.1 and subgroups on Bifrost

  • panvk/jm: allow binding VK_NULL_HANDLE index buffers

  • panvk: advertise VK_KHR_maintenance6 on v6-7

  • pan/kmod: Don’t initialize pan_kmod_deferred_bo_sync with size=0

  • panvk/v6-7: Increase maxBoundDescriptorSets to 7

  • panvk/v6-7: include required layouts in host image copy layouts

Ryan Zhang (12):

  • u_gralloc: query alloc_size and layer_count from gralloc

  • vulkan/android: compute arrayPitch for multi-layer AHB imports

  • panfrost: allow explicit layout for array images

  • panvk/utrace: add dynamic alloc for utrace clone buffer

  • panvk/utrace: skip clone copy when timestamp buffer allocation fails

  • panvk/utrace: guard delete_buffer against NULL on clone alloc failure

  • panvk/utrace: handle clone CS heap exhaustion gracefully

  • wsi/wayland: skip feedback dispatch when nothing is outstanding

  • panvk: initialize CPU tracing categories

  • panvk/utrace: replace bogus assert with static_assert in perfetto begin event callback

  • panvk: validate descriptor counts in shader deserialization

  • panvk/csf: check malloc return values in queue submit storage init

Rémi Bernon (1):

  • mesa: Initialize the viewport even with no surfaces.

Sagar Ghuge (37):

  • anv: Compile main and resume shaders for Jay

  • intel/debug: Extend Jay debug options for RT stages

  • vulkan/bvh: Move pair triangles into header

  • vulkan/bvh: Add option to override pair triangles spirv

  • vulkan/bvh: Specialize pair triangle shader

  • anv: Override pair triangles shader

  • anv/rt: Add parent child count map instead of using pad field

  • anv/rt: Update the parent child count map for valid nodes

  • brw: Restrict the vector size based on exec size on Gfx125

  • brw: Respect LSC data sizes and address alignment

  • jay: Drop assertion which doesn’t hold true anymore

  • anv/rt: Update quads during update BVH pass

  • bvh/meson: Add -g option to glslang for debug information

  • anv/rt: Fix wrong copy size for out of place update

  • anv/rt: Use 64bit keys for sorting

  • anv: Bump up the subgroup size to 32

  • brw/rt: Avoid duplicating the build_accept_ray

  • brw/rt: Check if hit attribtute are really getting used

  • brw/rt: Accept hit directly based on condition

  • anv/rt: Drop unnecessary barrier in update shader

  • anv/rt: Enable propagate cull flags

  • anv: Add drirc option to disable dynamicStackIDControl

  • anv: Call nir_remove_dead_variables after lowering vars_to_ssa

  • vulkan: Move barriers before encode pass at driver level

  • anv: Fix WA_18040903259

  • anv/rt: Skip scratch AABB init

  • anv: Fix barrier after out of place copy AS

  • anv: Fix barrier after anv_init_update_scratch

  • anv: Change trampoline dispatch layout

  • anv/rt: Cache geometry transform in SLM

  • anv/rt: Write root bounding box directly to header in update shader

  • anv/rt: Increase update shader workgroup size to 256

  • anv: avoid redundant vertex loads in quad leaf update

  • anv: Use transfer_to_compute barrier during bvh clear

  • anv: Use anv_cmd_buffer_alloc_space for encode batch

  • intel: Handle MESA_SHADER_KERNEL stage in helpers

  • anv: Fix parent child count map size

Sahitya Kandru (1):

  • freedreno: Add chip support for a722

Sam Day (6):

  • ir3: don’t use MGEN.B for nir_op_bfm on a3xx

  • freedreno/ci: a306: mark flaky test

  • freedreno/drm: limit BO dumping on failed submits

  • freedreno/a3xx: flush HLSQ before indirect const loads

  • freedreno/ci: a306: reenable fixed tests

  • freedreno/a3xx: replace HLSQ_FLUSH guard with WFI

Samuel Pitoiset (304):

  • spirv: Update the JSON and headers

  • vulkan: update spec to 1.4.358

  • radv/ci: bump the timeout for POLARIS10/RENOIR VKCTS jobs slightly

  • radv/ci: skip WSI tests on VANGOGH

  • radv/ci: cleanup CI lists for WSI tests

  • spirv: fix handling OpAbortKHR

  • ac/gpu_info: fix detecting AMD BC250 as RDNA1

  • ac/gpu_info: only fix incorrect IP versions for the GFX queue

  • radv: fix invalid assertions in DGC when queues aren’t enabled

  • ci: uprev vkd3d

  • radv/ci: make more flakes on VANGOGH

  • radv: remove deprecated RADV_PERFTEST options

  • radv: remove deprecated RADV_DEBUG=nocompute

  • docs: fix reporting VK_KHR_index_type_uint8 on RADV

  • radv: fix enabling mesh shader queries when shaderObject is enabled

  • radv/ci: skip dEQP-VK.ray_tracing_pipeline.misc.maxrtinvocations*

  • wsi/display: fix a memleak when swapchain creation failed

  • wsi/display: free the previous display name when EDID is parsed

  • radv: add radv_physical_device_destroy()

  • radv: move drirc options to the physical device

  • radv: cleanup using DCC drirc options on GFX12

  • radv: fix computing BC formats in texels for HIC

  • radv/meta: fix resolving to 3D color images

  • radv: add a workaround for broken multi viewport with UE5

  • ci: uprev Fossilize

  • ci: add radv-validate-cache to verify precompilation doesn’t break

  • radv: fix redundant WSI destroy

  • radv/ci: add radv-validate-cache.sh to gitlab-ci-inc

  • amd/drm-shm: add a very simple tool that lists known devices

  • radv/ci: use all drm-shim devices in radv-validate-cache

  • radv: fix blitting for 2D array to 3D with a Z flip

  • radv/amdgpu: print error when setting pstate fails

  • radv/amdgpu: prevent a rare issue when setting pstate

  • radv: fix cmdbuf alignment with DGC when queues aren’t enabled

  • radv/ci: improve the comment about the custom border color on compute queue

  • radv/ci: skip more tests that use custom border color on compute queue

  • radv: reset color remapping states when CmdBeginRendering() is called

  • radv: add radv_cmd_buffer_end_rendering()

  • radv/ci: skip two more tests that use custom border color on compute

  • radv: try to clarify RECTLIST/RECT_2D differences

  • radv: set radv_disable_aniso_single_level=true for Wolfenstein II: The New Colossus

  • radv/rt: do not use the dynamic offset user SGPR with descriptor heap

  • radv/rt: slightly simplify declaring shader args for RT prolog

  • radv: tidy up declaring CS_TASK_DRAW_ID for task shaders

  • radv/ci: skip WSI tests on RENOIR too

  • radv/meta: fix using aspects for blits

  • nir: consider load_push_constant as potentially divergent

  • nir: add an ACCESS flag to load_push_constant

  • ac/nir: prevent push constants to be lowered to subdwords

  • ac/nir: set ACCESS_SMEM_AMD for load_push_constant when possible

  • ac/nir: consider that load_push_constant might have ACCESS_SMEM_AMD

  • radv/nir: do not always lower push constants to SMEM

  • radv: zero-allocate all Vulkan objects by default

  • ci: uprev vkd3d

  • radv: merge {gfx6,gfx10}_make_texture_descriptor() together

  • radv: remove dead code about UNDEFINED formats for Android

  • radv: only split the right aspect for depth/stencil internal views

  • radv: cleanup computing the guardband rasterization primitive

  • radv: remove unnecessary radv_write_data()

  • radv/nir: disable custom border color for task shaders

  • radv: disable custom border color for compute shaders on compute queue

  • radv: disable custom border color for RT shaders on compute queue

  • radv: add a drirc to force-enable custom border color on compute queue

  • ci: mark a bunch of tests as expected failures for RADV

  • radv: add radv_force_exclusive_image drirc

  • radv: set radv_force_exclusive_image for few native Vulkan games

  • zink/ci: fix wrongly identified failures on POLARIS10/NAVI10

  • radv/meta: use 2D_ARRAY view types for 3D images that are rendered

  • radv: fix copying VRS rates with a non-zero base layer correctly

  • radv: enable HTILE for mips+layers on GFX10+

  • util,intel: move parse_debug_bitset() to common

  • radv: rework RADV_DEBUG options using a bitset

  • radv: add radv_image_has_display_dcc()

  • radv: rework initializing display DCC

  • radv: improve a comment about DCC retiling with read-only images

  • radv: ignore external/foreign queues for concurrent images

  • radv: use RADV_QUEUE_FOREIGN during image layout transitions

  • radv: rework getting the queue family mask

  • vulkan: fix generating vk_expand_pipeline_stage_flags2()

  • radv: fix waiting for CP DMA for accel struct build

  • radv: make sure transitions related operations only run on GENERAL/COMPUTE

  • radv: move cache flushes around for some transitions related operations

  • radv: rework initializing depth metadata

  • radv: skip some depth metadata init for zero-initialized images

  • radv/meta: add layers support to the FS resolve path

  • radv/meta: use the FS resolve path for layers too

  • radv/ci: update list of expected failures for TAHITI/HAWAII

  • radv: ignore VGT_STREAMOUT_SYNC on compute queue

  • radv: ignore non compute cache flushes also for gang CS

  • radv: cleanup determining when DCC retile is needed

  • radv: slightly refactor the DCC retile path

  • radv: fallback to compute for DCC retiles on transfer queue

  • radv: stop excluding all QFOTs on the transfer queue

  • radv: fix sync events for STAGE_2_COPY_INDIRECT_BIT_KHR

  • radv: only consider PS_PARTIAL_FLUSH for STAGE_2_BLIT_BIT

  • radv: use vk_expand_src_stage_flags2() in radv_stage_flush()

  • radv: remove redundant STAGE_2_VERTEX_INPUT in radv_stage_flush()

  • radv: remove redundant STAGE_2_BOTTOM_OF_PIPE in radv_stage_flush()

  • radv: remove redundant STAGE_2_PRE_RASTERIZATION_SHADERS in radv_stage_flush()

  • radv: cleanup STAGE_2_ALL_TRANSFER stages in radv_stage_flush()

  • radv: remove redundant STAGE_2_ALL_{GRAPHICS,COMMANDS} for VS sync in radv_stage_flush()

  • radv: remove redundant STAGE_2_ALL_COMMANDS for CS sync in radv_stage_flush()

  • radv: remove redundant STAGE_2_ALL_COMMANDS for PS sync in radv_stage_flush()

  • radv: re-organize radv_stage_flush()

  • radv: ignore stage flags that are expanded for radv_stage_flush()

  • radv: validate pipeline stage flags in radv_stage_flush()

  • amd: only enable the addrlib SIMD path when supported

  • ci: store the output of vulkaninfo for all stages that use Vulkan

  • radv: refactor common pipeline stage masks

  • radv: remove unneccesary handling of transfer stages in write_event()

  • radv: use radv_get_src_stage_flags2() in write_event()

  • radv: use radv_get_src_stage_flags2() for radv_cp_dma_wait_for_stages()

  • radv: remove redundant STAGE_2_xx bits when waiting for CP DMA

  • radv: move CP DMA wait stages close to the other pipeline stage masks

  • zink/ci: mark spec@arb_timer_query@timestamp-get as flakes for RADV

  • vulkan: force exclusive images when queueFamilyIndexCount is 1

  • radv: use vk_expand_src_stage_flags2() in radv_gang_barrier()

  • radv: move pipeline stage related masks around

  • radv: remove redundant STAGE_2_BOTTOM_OF_PIPE in radv_gang_barrier()

  • radv: simplify the ACE->SDMA sync in radv_gang_barrier()

  • radv: simplify the GFX->ACE sync for CP DMA waits

  • radv: simplify waiting for CS workload during gang barriers

  • radv: use vk_expand_dst_stage_flags2() in radv_gang_barrier()

  • radv: remove redundant STAGE_2_TOP_OF_PIPE in radv_gang_barrier()

  • radv: simplify the GFX->ACE when GFX needs to block ACE

  • radv: simplify the SDMA->ACE when SDMA needs to block ACE

  • radv: use radv_get_src_stage_flags2() in radv_gang_barrier()

  • radv: add radv_get_dst_stage_flags2() and use it in radv_gang_barrier()

  • radv: expand HiZ after transfer operations on compute queue on GFX12

  • radv: rename the clear HiZ path

  • radv: refactor clearing HiZ with compute/SDMA

  • radv: expand HiZ after transfer operations on SDMA on GFX12

  • ac/surface: store more information about HiZ surfaces on GFX12

  • radv: stop clearing HiZ for the whole image with SDMA on GFX12

  • radv: add RADV_CMD_FLAG_PFP_SYNC_ME

  • radeonsi: remove useless checks in gfx10_emit_barrier()

  • radv: rework determining CB/DB cache events for GFX10+

  • radv: simplify emitting release/acquire PWS events on GFX11+

  • radv: directly use ac_emit_cp_wait_mem in radv_cs_emit_cache_flush()

  • radv: directly use ac_emit_cp_release_mem() in radv_cs_emit_cache_flush()

  • radv: reorganize gfx10_cs_emit_cache_flush() slightly

  • radv: ignore CS_PARTIAL_FLUSH with CB/DB TS events on GFX10+

  • radv: reorganize radv_cs_emit_cache_flush() slightly

  • radv: ignore {VS,PS}_PARTIAL_FLUSH with CB/DB TS events on GFX9

  • radeonsi: remove some redundant emit helpers

  • radeonsi: slightly rework PWS events in gfx10_emit_barrier()

  • radeonsi: add and use si_get_eop_bug_va()

  • radv: use radv_cs_emit_cache_flush() for the postamble CS on GFX6-7

  • radv: rework the ACQUIRE_MEM/PFP_SYNC_ME sequence for GFX6-9

  • radv: do not always emit INV_TS for CB+DB flushes on GFX9

  • radv: only flush L2 metadata when it’s dirty

  • radv/ci: update list of expected failures for HAWAII/POLARIS10

  • radv: remove RADV_DEBUG=forcecompress

  • radv: simplify a check in radv_use_dcc_for_image_late()

  • radv: simplify a check in radv_image_alloc_single_sample_cmask()

  • radv: allow DCC for concurrent images without DCC stores

  • ac/gpu_info: refine the has_taskmesh_indirect0_bug workaround

  • radv: remove the workaround for has_taskmesh_indirect0_bug

  • radv: fix L0_CACHE_HIT_RATIO perf counter on GFX11

  • radv: emit a full barrier when waiting for perf counters

  • radv: implement few perf counters on GFX12

  • radv: implement a workaround for perf counters on GFX12

  • radv: enable VK_KHR_performance_query on GFX12

  • radv/rt: adjust the barrier before radv_encode_triangles_gfx12()

  • radv: dirty PFP_SYNC_ME for internal operations when needed

  • radv: dirty PFP_SYNC_ME for pipeline barriers only when needed

  • radv: stop waiting for mesh shader when the apps wait for task shaders

  • radv: emit ACQUIRE_MEM in ME when possible

  • zink/ci: update list of expected failures for RADV/POLARIS10

  • radv: return VK_ERROR_DEVICE_LOST if any shader executed OpAbortKHR

  • radv: consider shader abort faults with VK_KHR_device_fault

  • radv: use vk_queue_set_lost() when submits hung

  • radv: rework setting queue as lost

  • radeonsi: tidy up {gfx6,gfx10}_emit_barrier()

  • radeonsi: rework si_cp_acquire_mem() slightly

  • radeonsi: rework handling common barrier events slightly

  • ac,radv: move and rename rgp_flush_bits to common code

  • ac,radv: add ac_rgp_flush_bits_to_barrier_marker()

  • radeonsi: rework sqtt barriers

  • radeonsi: wait for graphics shaders differently on GFX10-11.7

  • radv: program PA_SC_HISZ_CONTROL.ROUND also for noop FS

  • radv: fix late-z when MRTZ isn’t exported for fbfetch

  • radv: add radv_ps_writes_mrtz()

  • util/drirc: change deviceVersion to a 32-bit uint

  • radv: adjust descriptions of radv_disable_dcc_{mips,stores} driconf options

  • radv: configure deviceVersion for driconf

  • drirc_gen: add a way to ignore options based on device version min/max

  • radv: ignore radv_disable_dcc* driconf options when not applicable

  • radv: fix initializing the DCC predication value when stores aren’t supported

  • radv: fix initializing DCC with mipmaps on GFX8

  • radv: stop using radv_init_dcc() when not strictly necessary

  • radv: only consider a cleared DCC image as compressed if the value isn’t 0xff

  • radv: slightly rework initializing the DCC predicate

  • radv: disable binning correctly on GFX9-11

  • radv: fix a typo when computing the binning size on GFX10-11.7

  • radv: add radv_disable_binning_gfx11 and enable for Sunset Overdrive

  • nir/lower_abort: fix payload alignment to 8-bytes

  • radv: do not abort when the app specifies an invalid modifier

  • radv: remove a very old way to import RadeonSI images

  • radeonsi: replace si_context::context_roll by a boolean

  • ac,radv,radeonsi: adjust ac_emit_cp_acquire_mem() for GFX6-9

  • radv: remove RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER

  • radv: stop dirtying RADV_CMD_FLAG_INV_L2_METADATA on GFX12+

  • radv: stop dirtying RADV_CMD_FLAG_FLUSH_AND_INV_DB_META on GFX10+

  • radv: allow DCC decompress with compute on gang CS

  • radv: fallback to compute for DCC/HTILE compressed images on SDMA 5 and older

  • radv: simplify a check for compressed images with SDMA

  • radv: don’t DCC compress concurrent images with transfer without DCC stores

  • radv: stop disabling DCC/HTILE for CONCURRENT images on SDMA 5 and older

  • radv: do not abort when a GPU hang report is created with deviceFaultVendorBinary

  • radv: ignore dynamic color states for depth-only rendering

  • radv: fix depth/stencil aspects for resolves

  • radv: rework KHR_device_fault to report faults only once

  • radv: fix immutable samplers lifetime with descriptor update templates

  • amd: add ac_barrier()

  • radv: use the common ac_barrier() helper

  • radeonsi: use the common ac_barrier() helper

  • radeonsi: merge {gfx6,gfx10}_emit_barrier() together

  • radv: replace RADV_CMD_FLAG_xxx by AC_BARRIER_xxx

  • radv: replace RADV_PWS_xxx by AC_PWS_xxx

  • radeonsi: replace SI_BARRIER_xxx by AC_BARRIER_xxx

  • radv: fix dirtying graphics shaders when shader objects are bound

  • radv: always disturb the related pipeline bind point with radv_CmdBindShadersEXT()

  • radv: simplify radv_CmdBindShadersEXT() slightly

  • amd/drm-shim: bump DRM version to 3.59.

  • amd: requires DRM 3.59.0 (Linux kernel 6.11.2+)

  • amd: add AC_AMDGPU_DRM_{MAJOR,MINOR}

  • radv/amdgpu: remove a dead check for GFX12 DCC

  • ac,radv: remove has_gpuvm_fault_query

  • ac,radv,radeonsi: remove gfx12_supports_display_dcc

  • ac,radeonsi: remove has_default_zerovram_support

  • radv: remove zerovram support completely

  • radv: stop setting vk_zero_vram=true for a bunch of applications

  • radv: fix allocating queues with different flags

  • ac,radv: add new macros to set optimized registers

  • radv: track more context registers

  • radv: ignore aspects that have no effects during clears

  • ac/sdma: use the same cache policy as the kernel for some packets on 5.2

  • ac/sdma: workaround a cache coherency issue with SDMA 5.2 and DCC

  • radv: move game performance tuning to a separate drirc file

  • radv/rt: fix checking errors for imported binaries

  • radv: be slightly more robust against invalid/corrupted blob reads

  • ac,radv,radeonsi: add has_streamout_vgt_hang_bug

  • radv: slightly optimize the streamout VGT hang checks

  • radv: make more draw functions always inline

  • radv: slightly rework emitting graphics states

  • radv: hoist a NGG check to radv_emit_draw_registers()

  • radv: only iterate on bound shaders when emitting descriptors

  • radv: optimize debugging after draws

  • radv: pre-compute index size at bind time

  • radv: pass radv_draw_info as value to some functions

  • radv: use radv_cmd_set_primitive_restart_index() more

  • radv: slightly rework prefetching gfx shaders in the draw path

  • radv: remove RADV_PREFETCH_GRAPHICS

  • radv: stop guarding radv_gfx12_emit_buffered_regs() with GFX12 checks

  • radv/meta: make stencil reference a static state for color clears

  • radv/meta: stop setting pDepthStencilState as noop

  • radv/meta: stop restoring stencil reference for ds clears

  • amd: rename some SDMA defines for consistency

  • amd: declare more SDMA bits

  • ac/sdma: emit the default cache policy on SDMA 5.2-6.1

  • ac/sdma: emit the mall policy on SDMA 7.0

  • radv: bind a default depth/stencil state when it’s ignored

  • radv: clear unnecessary depth/stencil dynamic states correctly

  • radv: add a function to pre-compute dispatch parameters

  • radv: emit compute userdata separately from dispatch packets

  • radv: add a function that emits userdata for RT

  • radv: emit compute userdata before barriers

  • radv: use more buffer SH_REG for compute on GFX12

  • radv/rt: fix importing pipeline binaries

  • radv: remove redundant parameters in radv_cs_emit_compute_predication()

  • radv: rework conditional rendering emulation on ACE

  • ac/nir/meta: fix determining if sparse loads must be scalarized

  • radv/nir: fix determining if a VS input location is high 64-bit

  • zink/ci: mark one test as expected failure on RADV/NAVI31/ASAN

  • radv/ci: mark two tests as expected failures with vkd3d/tahiti

  • radv: fix tracking GE_CNTL for NGG <-> legacy roundtrip

  • radv: do not pass radv_draw_info as value

  • radv: check if the stride is set when binding VBOs

  • radv: remove a leftover TODO about barrier flags

  • radv: remove an incorrect comment when resetting a query pool

  • radv: fix partial results for some query types

  • radv: use AC_BARRIER_SYNC_BOTTOM_OF_PIPE for all queries that need sync

  • radv: stop waiting for individual queries with vkCmdCopyQueryPoolResults()

  • radv: do not set RADEON_SURF_DISABLE_DCC on GFX12

  • radv: do not allow DCC for dedicated allocations that don’t support it on GFX12

  • radv/amdgpu: unmap the first page used to implement VM padding

  • radv: fix the VMA heap size for the descriptor sets allocator

  • radv: fix a memory leak when allocating a pool entry failed

  • radv: fix the error code when allocating a pool entry failed

  • radv: stop over reserving the CS for indirect taskmesh draws

  • radv: emit userdata for task/mesh direct draws before barriers

  • radv: slightly refactor normal draws and multi-draws

  • radv: emit userdata for vertex direct draws before barriers

  • ci: uprev vkd3d

  • aco: re-run clang-format

  • radv: re-run clang-format

  • radv,aco: add re-format commit to .git-blame-ignore-revs

  • radv: allow multisampling with 2D CUBE_COMPATIBLE images with MSRTSS

  • radv: reset all tracked registers after executing a secondary

  • radv: fix emitting RSRC2 for vertex prologs

  • radv/rt: fix overwriting stack size on partial cache hits

  • nir: fix RANGE handling for load_ubo in lower_mem_access_bit_sizes

  • nir/opt_access: do not infer ACCESS_CAN_REORDER with shader interlock

Sangeetha Pokala (1):

  • freedreno: Enable A753 and A735

Sebastian Krzyszkowiak (1):

  • etnaviv: Always read clear color from SW meta when TS is exported

Sebastian Wick (4):

  • kms-dri-sw: Remove always-true whandle check and simplify control flow

  • kms-dri-sw: Handle drmPrimeFDToHandle errors in create_mapped()

  • kms-dri-sw: Move find_and_ref() to make it available to create_mapped()

  • kms-dri-sw: Deduplicate displaytargets by handle in create_mapped()

Serdar Kocdemir (7):

  • gfxstream: fix semaphore pruning with empty timeline semaphore lists

  • gfxstream: Fixes for physical device property overwrites

  • gfxstream: add some error logging

  • gfxstream: fix gfxstream_device leak in vkCreateDevice error paths

  • gfxstream: make generate-gfxstream-vulkan.sh standalone-safe

  • gfxstream: format with clang-format and enable enforcement

  • gfxstream: Deduplicate goldfish_vk_extension_struct_size_with_stream_features

Sergi Blanch Torne (16):

  • xfiles: update before uprev Piglit

  • xfiles: update expectations based on 2026-7-21 nightly

  • Uprev ljs to 393c4ac1c7456ee9903e92a71dd5ed97da1eff10

  • ci,crnm: profiles with stages

  • ci,crnm: profiles for Piglit uprev

  • ci,crnm: profiles for uprev ANGLE

  • ci,crnm: profiles for uprev VVL

  • ci: disable Collabora’s farm due to maintenance

  • Revert “ci: disable Collabora’s farm due to maintenance”

  • xfiles: update expectations based on 2026-9-28 nightly

  • ci: disable Collabora’s farm due to maintenance

  • Revert “ci: disable Collabora’s farm due to maintenance”

  • ci: temporarily disable dragonboard-410c

  • Revert “ci: temporarily disable dragonboard-410c”

  • crnm: organize the arguments

  • crnm: group argument definitions

Sergii Ushakov (1):

  • lavapipe: guard AHB/ANB imports with VK_USE_PLATFORM_ANDROID_KHR

Sid Pranjale (25):

  • v3d,v3dv: commonize timestamp query

  • v3dv: store shared data owner type and handle

  • v3dv: introduce a unique per-device bo identifier

  • v3dv: track self imported BOs

  • v3dv: store obj_type and obj_handle in v3dv_bo

  • v3dv: emit device memory reports in v3dv_bo_alloc/free

  • v3dv: advertise VK_EXT_device_memory_report

  • broadcom/cle: fix Cpp-incompatible anonymous struct pattern

  • v3dv: rename v3dv_bo::private to is_private

  • v3dv: track a job’s parent queue

  • v3dv: add identifiers to jobs

  • v3dv: track queue usage for cmd buffer perfetto trace markers

  • v3dv: add u_trace timestamp support

  • v3dv: add perfetto GPU tracepoints

  • v3dv: emit tracepoints during queue submission

  • tool/pps: register gpu.renderstages.broadcom data source

  • docs/perfetto: v3dv now supports render stages

  • radv: switch radv_device_memory to use vk_device_memory

  • broadcom/common: add common code for perfetto GPU tracepoints

  • v3dv: switch to common perfetto GPU tracepoint code

  • vulkan/debug_utils: add a common vk_device_memory_report_emit()

  • anv, panvk, pvr, radv, tu, v3dv, vn: use new device memory report function

  • vulkan/runtime: drop vk_emit_device_memory_report()

  • v3dv: Advertise VK_KHR_map_memory2

  • v3dv: advertise VK_KHR_unified_image_layouts

Sidong Yang (1):

  • rocket: support average pooling via depthwise convolution

Silvio Vilerino (8):

  • d3d12: Video Encode slice fallback to support AUTO if FULL_FRAME unavailable

  • mediafoundation: Add ETW for fence_wait_multiple call

  • d3d12: Do not use D3D12_VIDEO_ENCODER_PICTURE_CONTROL_CODEC_DATA1 as in/out param arg

  • d3d12: Fix d3d12 video encoder batch leaks

  • d3d12: Fix video encoder destruction path NULL pointer accesses

  • d3d12: Fix video decoder sync_completion releasing the wrong pool entry fence

  • d3d12: Fix video decoder destruction path NULL pointer accesses

  • d3d12: Fix d3d12 video decoder batch leaks

Simon Perretta (38):

  • pco: keep scalar writes for vs outputs

  • pco: temporarily disable vtxin allocation due to vec handling bug

  • pco, pvr: add support for ubo and ssbo dynamic indexing

  • pco, pvr: add support for image, ia, texel buffer dynamic indexing

  • pco: handle large hwreg offsets for mov_offset, op_atomic_offset

  • nir: generalize the src type of interleave to tuint

  • pco: drop i2i16 conversion for nir_interleave srcs

  • pco: pass shader data to preprocessing function

  • pco, pvr: add support for VK_EXT_inline_uniform_block

  • pvr: advertise KHR_shader_fma

  • nir/lower_helper_writes: add image deref support

  • pco: flush the cache for images accessed via atomics and non-atomics

  • pco: add debug option for disabling the dma cache

  • pco: skip propagating access for derefs without vars

  • pvr: pass enabled device features to the shader earlier

  • pco: consider robustness in load/store vectorization

  • pco: split out sample mask out pass

  • pco: use sample check for demote_samples

  • pco: fold sample mask in code into fs intrinsics

  • pco: use is_helper_invocation for discard/demote handling

  • pco: amend early_fragment_tests implementation

  • pco: make sure final depth is clamped and used for z-replicate

  • pco: amend atomic address calculation for 2d images

  • pco: ensure nir_global_atomic_pco ops are tracked

  • pco: use the source from src instead of io when building mappings

  • pco: defer lower_helper_writes until buffer writes/atomics are lowered

  • pvr, pco: consider sampleShadingEnable for fs coords/sample pos

  • pvr: advertise fragmentStoresAndAtomics

  • pco: special-case tanh inf handling on cores with rtz rounding

  • pvr: drop legacy compiler foreach macros

  • pvr: add zero and null pages

  • util/bitpack_helpers: add some unpack helpers

  • pvr: add new PDS compiler and pre-compilation utility

  • Revert “pco: special-case tanh inf handling on cores with rtz rounding”

  • pvr, pco: add support for VK_EXT_image_sliced_view_of_3d

  • pco: gather transform feedback info for vertex shaders

  • pco: point size handling for xfb

  • pvr/ci: update fails/skips related to xfb

Simon Ser (19):

  • EGL: sync headers with Khronos

  • egl/device: add support for EGL_EXT_device_type for software device

  • egl, gallium: implement EGL_EXT_device_type for DRM devices

  • asahi: advertise device type

  • crocus: advertise device type

  • etnaviv: advertise device type

  • freedreno: advertise device type

  • i915: advertise device type

  • iris: advertise device type

  • lima: advertise device type

  • nouveau: advertise device type

  • panfrost: advertise device type

  • r300: advertise device type

  • r600: advertise device type

  • radeonsi: advertise device type

  • v3d: advertise device type

  • vc4: advertise device type

  • zink: advertise device type

  • dri: bail out when missing swrast loader in drisw_init_screen()

Skipin Andrey (3):

  • dri3: ignore IdleNotify for a reused pixmap XID

  • wsi/x11: ignore IdleNotify for a reused pixmap XID

  • vl/dri3: ignore IdleNotify for a reused pixmap XID

Sonny Jiang (3):

  • radeonsi/vcn: Add vcn_5_0_2 support

  • radeonsi: always set is_format_supported in screen create

  • ac/gpu_info, radeonsi, radv: allow initialization without GFX or compute queues

Stefano Marchesi (1):

  • v3dv: make pipeline cache entry cap opt-in

Sun Haiyong (1):

  • llvmpipe: Raise llvmpipe max bits to 256 on loongarch.

Sushma Venkatesh Reddy (3):

  • intel/dev: Clamp PTL+ CS workgroup threads to 32

  • brw: Skip DPAS in has_invalid_dst_region

  • jay: Legalize flag sources of float and 64-bit instructions

Tacodiva (1):

  • vulkan/runtime: Fix bad assumption in GetPipelineBinaryDataKHR

Takuro Ashie (2):

  • panfrost: Reallocate BO when linear layout grows

  • panfrost: Preserve BO flags during linear conversion

Tanner Van De Walle (1):

  • nir/nir_builder: inline dst_bit_size calculation in assert

Tapani Pälli (23):

  • anv: remove usage of fcv bit and related

  • blorp: remove gfx125 related FCV_CCS_E cases

  • isl: fix FCV_CCS_E and ZCS usage asserts for gfx12, gfx125

  • intel/dev: update mesa_defs.json from workaround database

  • iris: implement Wa_16030144090 for iris driver

  • anv: implement Wa_16030144090 for anv driver

  • blorp: implement Wa_16030144090

  • iris: set VFG distribution mode according to Wa_16029281427

  • anv: set VFG distribution mode according to Wa_16029281427

  • anv: make it possible to toggle 3DSTATE_RASTER::ForceMultisampling

  • anv: change bindless debug env variable to be drirc option

  • anv: toggle anv_always_bindless for XPlane-12

  • brw: Expand RT thread-id assert version checks for Xe3P

  • anv: fix cmd_buffer_set_indirect_stride on ARL

  • anv: add additional heuristic to simd32 assumption pass

  • anv: choose correct pipeline mode for Xe2 query result copy

  • anv: enable brw_disable_subgroup_size_control only for brw

  • iris: update nir information within iris_force_dual_color_blend

  • compiler/nir: add missing intrinsic to avoid hitting assert

  • anv: change assume_full_subgroups drirc detection

  • anv/dric: set anv_assume_full_subgroups for Last Of Us 2

  • anv: add anv_force_exact_glsl_fma drirc option

  • anv/drirc: use anv_force_exact_glsl_fma for Project Wingman

Thomas H.P. Andersen (11):

  • nvk: wire the video engine into the queue engine and subchannel masks

  • nvk: add an NVK_EXPERIMENTAL=video flag

  • meson: readd ignore of unsafe_op_in_unsafe_fn warnings

  • nvk: fix error message typo

  • nvk: call vk_error on compatible struct

  • nvk: fix typo in error message

  • nvk: avoid leak in error path

  • nvk: drop duplicate include

  • nvk: drop duplicate asssert

  • nvk: use correct enum in code comment

  • nvk: fix ref count in error path

Thong Thai (8):

  • nir/meson: Add idep_nir_headers as a _libnir dependency

  • meson: Move dep_libdrm to be before dep_va

  • ci: add VIDEO_CODECS variable

  • ci: add check for VAAPI tag to set libva path

  • ci: decode-only/no-gfx-compute build testing

  • frontends/va: report processing entrypoint only if VPP is supported

  • meson: update libva wrap to 2.24.1

  • driconf: Add ignore_map_invalidate_buffer option

Timothy Arceri (35):

  • zink: fix swap interval changes being dropped

  • nir/opt_dead_write_vars: handle memcpy_deref as reads

  • nir/nir_opt_dead_write_vars: test correct behaviour with atomics/memcpy

  • util: add Blockland workaround for crash

  • util/mesa: add workaround to zero invalidated buffers

  • util: add workaround for Riddick using round() in glsl 1.20

  • llvmpipe: emit FS input vertex attributes in driver location order

  • llvmpipe: don’t rely on variable list order for per primitive outputs

  • zink: add is_unassigned_slot() helper

  • zink: assign the IO slot map in location order

  • nir/unlower_io_to_vars: stop sorting variables by location

  • mesa/st: approximate fog distance in the frag shader

  • glsl: optimize shared memory as derefs before lowering it

  • nir/gcm: move set_block_for_loop_instr() earlier

  • nir/gcm: dont touch loop instructions in a block that ends with a break

  • nir/gcm: store loop count for blocks inside if-statments

  • nir/gcm: add more heuristic for instructions in loops

  • Revert “intel/brw: keep texture ops in large loops”

  • Revert “nir/opt_gcm: add option to keep texture ops in large loops”

  • nir/gcm: add defensive break

  • nir/gcm: don’t hand nested loops the small loop free pass

  • nir/gcm: weigh loop hoists against what the loop actually holds

  • nir/gcm: let a driver say what a divergent value costs

  • nir/gcm: leave loops that barely run alone

  • intel/brw: measure loop pressure in GCM rather than guessing it

  • nir: temporarily disable floating point exception flags

  • util/xmlconfig: match blake3 against the PE executable under Wine

  • st/mesa: add a limit_max_texture_size driconf option

  • driconf: limit the max texture size for Blade Strangers

  • radeonsi: correctly detect and set mesh shaders as dirty

  • radeonsi: widen bind_history bitfield

  • gallium/tc/zink: widen rebind_mask to support use by mesh shaders

  • zink: fix rebinding support for mesh shaders

  • util/format: fix typo in the BPTC compressor

  • st/mesa: pass the edge flag vertex element to the draw module

Timur Kristóf (100):

  • ac/surface: Don’t set DCC numSamples for depth/stencil on GFX8

  • ac/gpu_info: Print modifiers for 16bpp, 32bpp, 64bpp

  • ac/surface: Check tiling version for DCC_RETILE flag

  • ac/surface: Split GFX9-12 format modifier code paths

  • ac/surface: Extract gfx6_select_3d_tile_idx() function

  • ac/surface: Add helpers to get tile and macrotile index on GFX6-8

  • ac/surface: Extract gfx6_fill_addr_info_from_surf() function

  • ac/surface: Extract is_displayable helper function

  • ac/nir/clear_copy_buffer: Fix last thread byte count

  • ac/nir/clear_copy_buffer: Change arguments to const

  • ac/nir/clear_copy_buffer: Add details to shader name

  • ac/nir/clear_copy_buffer: Use 64-bit offsets

  • ac/nir/clear_copy_buffer: Fix indentation to use 3 spaces style

  • ac/nir/clear_copy_buffer: Refactor to allow using different intrinsics

  • ac/nir/clear_copy_buffer: Only emit load_user_data_amd once

  • ac/nir/clear_copy_buffer: Allow placing user data before clear value

  • ac/nir/clear_copy_buffer: Mark the branch as always taken

  • radv/query: Don’t check buffer size before cache flush

  • radv/nir: Lower load_user_data_amd intrinsic to use SGPRs

  • radv/meta: Slightly reorganize meta fill and copy

  • aco/isel: Only allow first load to write the dst

  • radv: Use GL2_BYPASS flag for GFX preamble

  • radv: Use preemptable wait in gang preamble

  • radv: Use PFP for gang postamble wait and write

  • radv: Always bypass L2 cache for gang wait semaphore

  • radv: Emit task rings after cache flush and wait

  • radv: Wait for L2 cache writeback on ACE for SDMA+ACE gangs

  • radv: Invalidate L2 cache on ACE after waiting for SDMA

  • radv: Create separate BO for gang semaphore to avoid coherency and perf issues

  • aco/optimizer: Keep MUBUF SOFFSET on GFX6-7

  • aco/isel: Use SOFFSET when selecting buffer loads with index VGPR

  • radv/nir: Don’t fold constant offset into index VGPR for VS inputs anymore

  • radv/ci: Remove bounds checking failures from expected fails on Hawaii and Tahiti

  • ac/nir/ngg: Use ds_or_b32 for storing culling flags on Navi 10

  • ac/uvd_dec: Add and configure macro tiling related fields on GFX6-8

  • radeonsi/vce: Add tiling configuration fields

  • drm-uapi: Add AMD modifiers for GFX6-8

  • ac/surface: Skip surface modifier tests on GFX6-8

  • ac/surface: Add GFX6-8 specific modifier fields to ac_drm_fourcc.h

  • ac/surface: Support DRM format modifiers on GFX6-8

  • radv: Support DRM format modifiers on GFX6-8

  • radeonsi: Support DRM format modifiers on GFX6-8

  • amd/common: Only apply GFX6 TCS workaround when LBPW is enabled

  • ac/nir/clear_copy_buffer: Move CP DMA considerations into separate functions

  • ac/nir/clear_copy_buffer: Move dwords per thread calculation into separate function

  • ac/nir/clear_copy_buffer: Move preparing the clear user data into a separate function

  • ac/nir/clear_copy_buffer: Add option to prefer CP DMA below a threshold

  • ac/nir/clear_copy_buffer: Adjust dwords per thread on GFX11.5

  • ac/nir/clear_copy_buffer: Add option to put addresses in user data

  • ac/nir/clear_copy_buffer: Use only 1 user SGPR when clear size is 4 bytes

  • radv/meta: Create common fill/copy meta shader pipeline

  • radv/meta: Use common meta shader for buffer copy

  • radv/meta: Use common meta shader for buffer fill

  • radv/meta: Remove old fill and copy meta shaders and implementation

  • radv: Document default kernel driver for GFX6-7

  • ac/surface: Move getting legacy tiling index to ac_surface.h

  • ac/sdma: Move GFX9+ specific tile info to SDMA4+ specific code

  • ac/sdma: Fix tile index on SDMA2-3

  • ac/sdma: Clear macro tiling fields when not using macro tiling on GFX7-8

  • ac/sdma: Fix stencil tile split bytes for SDMA2-3

  • ac/sdma: Fixup linear sub window packet for SDMA 2-3

  • ac/sdma: Fixup tiled subwindow copy on SDMA2-3

  • ac/sdma: Fix tiled to tiled copy on SDMA2-3

  • ac/sdma: Fix SDMA WRITE packet for SDMA2-3

  • ac/sdma: Fix SDMA CONSTANT_FILL packet for SDMA2-3

  • ac/ib_parser: Add SDMA version field to IB parser

  • ac/ib_parser: Parse more SDMA packets and consider more versions

  • radv: Don’t use radv_cp_wait_mem() for gang semaphores in the command buffer

  • radv: Don’t use radv_cp_wait_mem() for events

  • radv: Reorganize gang preamble code, only use preemptable wait on GFX9+

  • radv: Reorganize gang postamble code, move wait flags to variable

  • radv: Delete radv_cp_wait_mem(), use ac_emit_cp_wait_mem() instead

  • ac/sdma: Work around Polaris hang on SDMA POLL_REGMEM packets

  • ac/sdma: Work around Polaris hang on SDMA FENCE packets

  • ac: Add WAIT_REG_MEM_UNCACHED_VI_MEC flag

  • radv: Use MTYPE=UC for gang semaphores on GFX8 compute queues

  • radv: Use MTYPE=UC in gang wait preamble on GFX8 compute queues

  • radv: Use MTYPE=UC in gang wait postamble on GFX8 compute queues

  • ac/sdma: Move T2T scanline copy determination to AC

  • ac/sdma: Move unaligned buffer copy determination to AC

  • ac/sdma: Consider T2T alignment requirements on SDMA2-3

  • ac/sdma: Consider buffer/image copy alignment requirements on SDMA2-3

  • ac/sdma: Use unaligned tiled buffer/image copy workaround to prevent page faults

  • radv/sdma: Fill ac_sdma_surf information on GFX6-8 too

  • radv/sdma: Take temp_bo_va as argument instead of a pointer to the BO

  • radv/sdma: Allocate more space for copy temp BO on SDMA2-3

  • radv/sdma: Split buffer/image and image/image copies on SDMA 2

  • radv/sdma: Implement dword-alignment for unaligned buffer image copies on SDMA2-3

  • radv: Optimize unaligned buffer/image copy on SDMA2-3

  • radv/sdma: Call buffer/image copy for T2T scanline copy

  • radv/sdma: Call buffer/image copy for unaligned copies

  • radv: Allow transfer queue on GFX7+ with experimental feature flag

  • radv/ci: Increase SDMA job timeout to 10 sec

  • radv/ci: Enable transfer queue on the Hawaii CI job

  • radv: Use compute copy/clear for sparse images on Fiji

  • ac/gpu_info: Enable sparse by default on GFX7+ dedicated GPUs

  • ac/nir/clear_copy_buffer: Add load_buffer_sparse_split() helper

  • ac/nir/clear_copy_buffer: Emit separate load for last thread, if necessary

  • ac/nir/clear_copy_buffer: Emit separate load when the first thread is the last, if necessary

  • ac/nir/clear_copy_buffer: Don’t round up the number of components for the first thread

Tomeu Vizoso (45):

  • rocket/tests: Add failures after increasing coverage

  • ethosu/tests: Add failures after increasing coverage

  • etnaviv/tests: Add failures after increasing coverage

  • ethosu: Match Regor’s block dependency calculation

  • ethosu: Keep fully connected feature maps in brick format

  • ethosu: Encode pooling OFM scale at the device register width

  • ethosu: Walk the strides of a brick-format feature map span

  • ethosu: Reject pooling with strides above three

  • ethosu: Keep concat and split tensors linear at unaligned offsets

  • ethosu: Decide split and unpack layout by channel offset alone

  • ethosu: Follow reshapes when choosing brick format

  • ethosu: Dump the command stream when it is compiled

  • torx: add AOT device discovery (pipe_loader_ml_probe)

  • torx: add ctypes bindings for Gallium ML structs (gallium.py)

  • torx: add graph partitioner

  • torx: add quantization support

  • torx: add ExIR-to-Gallium compiler and backend

  • torx: add ExecuTorch runtime backend

  • torx: add GTest-based test suite

  • torx: add documentation

  • torx: Update the Python Gallium ML bindings

  • torx: Don’t test the gallium ctypes bindings on every import

  • torx/tests: Compare the NPU output to the int8 outputs on the CPU

  • torx/tests: Check the int8 output on the CPU against the float

  • torx: Quantise convolution weights per channel

  • torx: Pad every tensor below rank 4 trailing-aligned

  • gallium: Document the tensor layout the ML API assumes

  • torx: Pass only int8 operations to drivers

  • torx: Check for Python dependencies at configure time

  • torx/tests: Bring the yolox-nano model from TI’s zoo

  • torx: Add helpers for operations with shared quantization

  • torx: Only fuse activations on convolutions and tensor additions

  • torx: Lower concatenation operations

  • torx: Lower nearest-neighbour resize operations

  • torx: Lower sigmoid operations

  • torx: Lower max pooling operations

  • torx/tests: Split Conv+Sigmoid in individual tests

  • torx: Decompose the boundary quantize with scalar operands

  • torx: Refactor detection of (de)quantize nodes

  • torx: Assert on operations we cannot lower

  • torx: Add a pass for annotating tensors with their memory layouts

  • torx: Use the dim_order annotation when translating tensors

  • torx: Use shared quantization on rank-3 layout operations

  • torx: Don’t transpose i/o tensors if not needed

  • torx: Lower layout operations with a rank of 3

Toshinari Morikawa (1):

  • egl: avoid calling loader_get_driver_for_fd with fd = -1

Tupili Krishna Gowtham Reddy (50):

  • freedreno/a6xx: enable f16 infinities for freedreno driver

  • tu: Don’t force R32_UINT when clearing MSAA E5B9G9R9 images

  • gallium/u_resource: add missing extern “C” guard

  • glsl: fix GLSL_SAMPLER_DIM_EXTERNAL_2D_Y2Y plumbing gaps

  • glsl,nir: introduce GLSL_TYPE_YUV_CSC_STANDARD_EXT

  • freedreno/ir3: handle GLSL_TYPE_YUV_CSC_STANDARD_EXT in RA size/align

  • intel/elk: handle GLSL_TYPE_YUV_CSC_STANDARD_EXT in two switches

  • mesa/st: enable GL_EXT_YUV_target when the driver supports YUV rendering

  • mesa/main: Add core support to attach YUV textures as TEXTURE_EXTERNAL_OES

  • mesa/main: Restrict operations on YUV and TEXTURE_EXTERNAL_OES textures

  • mesa/main: reject YUV color buffers in BlitFramebuffer

  • freedreno/a6xx: Advertise PIPE_BIND_LINEAR support in is_format_supported

  • nir, freedreno/ir3: Plumb YUV color output flag to ir3_shader_variant

  • freedreno/a6xx: add YUV clear support via RB_RESOLVE_CLEAR

  • gallium: add pipe_blend_state::is_yuv for YUV render targets

  • freedreno/a6xx: handle is_yuv blend state for GL_EXT_YUV_target

  • freedreno/a6xx: add YUV multiplanar support to GMEM MRT/restore/resolve

  • freedreno/a6xx: Route YUV color output to multiple MRTs

  • freedreno/a6xx: update draw state emission for YUV rendering

  • freedreno/a6xx: add YUV/NV12 support to the A2D 2D blitter

  • mesa/st: cache the plane-view resource-format check in gl_texture_object

  • mesa/st: don’t skip extra sampler view setup for natively-stored YUV textures

  • mesa/st: compute needs_yuv_plane_views for surface-based textures too

  • glsl, mesa/main: track Y2Y samplers in shader info and uniform storage

  • mesa/st, nir: plumb Y2Y sampler mask into NIR lowering options

  • nir/lower_tex: bypass CSC for Y2Y external samplers

  • freedreno/a6xx: enable UBWC for NV12 render targets

  • freedreno/a6xx: handle 422 YUV formats in the 2D blitter ifmt mapping

  • mesa/st: alias pt when no next-plane exists for natively-stored packed YUV

  • freedreno: fix GMEM allocation for packed YUV formats

  • freedreno/a6xx: fix GMEM resolve stride for packed YUV formats

  • freedreno/a6xx: enable TPL1_A2D_WINDOW_OFFSET on A7XX

  • freedreno: add bin-width alignment tracking for YUV

  • freedreno: reserve GMEM space for multiplanar YUV planes

  • freedreno/a6xx: add UV plane resolve for multiplanar YUV

  • freedreno/a6xx: enable UBWC NV12 GMEM rendering

  • freedreno: enforce per-plane resource invariant for multi-planar YUV

  • freedreno/fdl: map PIPE_FORMAT_NV12 to FMT6_NV12_4R

  • freedreno/fdl: add YUYV texture and render target format entry

  • docs: Add GL_EXT_YUV_target to features.txt

  • ci/freedreno: force gmem/sysmem piglit runs for ext_yuv_target tests

  • fdl: generalize UBWC block dimensions for all 2-plane YUV formats

  • freedreno/a6xx: allow YUV plane-format aliases in fd6_check_valid_format

  • fdl/fd6_view: fix NV12 texture descriptor format and chroma offset for UBWC

  • fdl: add fdl6_layout_multiplanar_image() for NV12-family plane stitching

  • freedreno/a6xx: add layout_multiplanar_resource screen hook

  • freedreno: allocate NV12 Y and UV planes into a single GEM BO

  • tu: Replace C-style casts with explicit C++ casts

  • gbm/dri/mesa: prepare native NV12 format support in GBM and DRI tables

  • gbm/dri: enable native YUYV format support in GBM and DRI format tables

Tyler Baker (1):

  • freedreno: Fix the A722 CCU count

Utkarsh Dalal (1):

  • freedreno: add 2 slice Adreno 840 chip id

Utku Iseri (6):

  • pan/genxml: fix SRs for RUN_FULLSCREEN on v13

  • pan/genxml: fix SRs for RUN_FULLSCREEN on v14

  • panfrost: v12+ fixes for run_fullscreen blits

  • pan: check full fb coverage on has_partial_tiles

  • pan: disallow CRC on AFBC on v14+

  • pan: allow CRC with AFBC

Valentine Burley (83):

  • freedreno/ci: Limit edgeflag and stipple skips to Piglit tests

  • ir3: Fix ballot_components for subgroups smaller than 32

  • ir3: Derive max_variable_workgroup_size from device limits

  • ir3: Compute subgroup_size from threadsize_base

  • tu: Use computed subgroup size

  • panvk/ci: Update expectations for g610-vk-asan following VK CTS uprev

  • panvk: Disable VK_KHR_internally_synchronized_queues on Vulkan 1.0

  • panvk/ci: Document another WSI flake

  • tu: Report correct maxFragmentInputComponents limits

  • zink: Use ShaderLayer capability for gl_Layer when available

  • zink/ci: Move glx@glx-make-current to flakes

  • ci/build: Increase timeout for debian-build-x86_64

  • ci/piglit: Switch all trace jobs to surfaceless+gbm

  • venus/ci: Drop no longer needed skips for Turnip

  • freedreno: Increase reg_size_vec4 for A702

  • freedreno: Fix VPC_RAST_STREAM_CNTL register layouts

  • venus: Avoid arithmetic on void pointers

  • subprojects: Update venus-protocol

  • venus: Enable -Werror=gnu-pointer-arith

  • docs/features: Mark VK_KHR_maintenance9 as done on Turnip

  • panvk/ci: Reduce panfrost-g925-vk:arm64 parallelism

  • panvk/ci: Add a Kraid job on G925

  • ci/debian: Install spirv-llvm-translator instead of building it

  • panvk/ci: Serialize nightly panfrost-g610-vk-asan:arm64 job

  • ci/rusticl: Update nightly expectations after spirv-llvm-translator bump

  • panfrost/ci: Remove retired mt8192-asurada-spherion-r0 device

  • ci: Update S3 kernel paths

  • freedreno/ci: Switch a200 to gfx-ci/linux kernel

  • tu: Fix pDependencyInfos indexing and barrier emission in CmdWaitEvents2

  • u_gralloc: Support depth/stencil HAL formats for AHB import

  • vulkan/android: Deep copy VkImageCompressionControlEXT for deferred AHB/ANB images

  • vulkan/android: Force linear for VK_IMAGE_COMPRESSION_DISABLED_EXT

  • tu: Honor VK_IMAGE_COMPRESSION_DISABLED_EXT in image creation

  • tu: Report compression status in vkGetImageSubresourceLayout2

  • tu: Report compression in vkGetPhysicalDeviceImageFormatProperties2

  • tu: Reject VK_FORMAT_S8_UINT AHB import

  • tu: Skip DRM modifier layout fixups for non-modifier images

  • tu: Advertise VK_EXT_image_compression_control(_swapchain)

  • tu/ci: Add Android coverage for VK_EXT_image_compression_control

  • ci: Update GL & GLES CTS

  • tu/ci: Skip more broken direct_drm tests on a702

  • tu: Allow multi-layer images with explicit DRM format modifiers

  • ci/deqp: Update VK CTS to 1.4.6.2

  • tu: Fix requiredSubgroupSizeStages enablement

  • tu: Use expose_double_threadsize for ray query subgroup size

  • tu: Enable subgroupQuadOperationsInAllStages where supported

  • tu: Advertise VK_EXT_shader_subgroup_vote

  • tu: Advertise VK_EXT_shader_subgroup_ballot

  • freedreno, tu: Limit XFB streams on GPUs lacking multi-stream query support

  • tu: Create patchpoints_ctx before emitting subpass begin in renderpass

  • tu: Ignore FDM offsets for subsampled images when shared viewport is active

  • tu: Do not bake static viewport/scissor when FDM is enabled

  • tu: Fix GMEM clears with an FDM offset on the blit-event clear path

  • tu/ci: Update a702 expectations after VK CTS uprev

  • lvp/ci: Update expectations after gitlab-runner update

  • spirv: Check all valid capabilities for Layer and ViewportIndex builtins

  • amd/ci: Use gfx-ci/firmware repo for CI-tron jobs

  • vulkan/runtime: Export runtime dependencies in interface dependencies

  • u_gralloc: Mark internal imapper helpers static

  • u_gralloc: Add and propagate dep_android_ui dependency

  • hermetic/android: Map ui dependency for u_gralloc

  • tu: Bump conformanceVersion

  • nir/lower_subgroups: Fix vote lowering with vector ballots

  • ir3: Lower quad votes

  • tu: Advertise VK_KHR_shader_quad_control

  • llvmpipe/ci: Add a surfaceless_egl piglit job

  • pan/afbc: Support X-component format variants in AFBC and blend tables

  • pan/mod: Introduce modifier layout capabilities query

  • panvk: Query modifier capabilities for format property advertisement

  • panvk: Enable AFBC for WSI images by default

  • panvk: Advertise VK_KHR_incremental_present

  • panvk: Advertise VK_EXT_swapchain_colorspace

  • ir3: Support load_view_index as a VS driver param

  • tu: Move down tu6_emit_empty_vs_params

  • tu: Implement software multiview via CPU draw duplication

  • tu: Enable VK_KHR_multiview and Vulkan 1.3 on all a6xx devices

  • bin/ci: Constrain anyio version to avoid conflict

  • freedreno/ci: Document recent a306 flakes

  • virtio/vdrm: Avoid arithmetic on void pointers

  • virtio/vdrm: Enable -Werror=gnu-pointer-arith

  • ci/deqp: Backport fix for MESA_map_buffer_client_pointer

  • lvp/ci: Skip timing out test

  • lavapipe: Guard VK_EXT_multisampled_render_to_swapchain with LVP_USE_WSI_PLATFORM

Vandomas (5):

  • wsi/metal: fix memoryTypeIndex in image memory allocation

  • zink: enable renderpass tracking for KosmicKrisp

  • zink: do not use implicit sync on kosmickrisp

  • zink: don’t draw or dispatch with a null pipeline

  • zink: put the buffer view cache key first

Victoria Brekenfeld (8):

  • ci: bump wayland-protocols to 1.49

  • meson: require wayland-protocols 1.49

  • wayland: use new stable name for dmabuf extension

  • loader: Split `loader_get_user_preferred_{device,fd}`

  • egl/wsi/wayland: keep the full `dmabuf_feedback` around

  • egl/wsi/wayland: Support wl-dmabuf v6

  • vulkan/wsi/wayland: Keep the whole `dmabuf_feedback` around

  • vulkan/wsi/wayland: Support wl-dmabuf v6

Vinson Lee (6):

  • aco/sched_vopd: initialize num_operands in VOPDInfo constructor

  • glsl: initialize pixel_local_storage and yuv in ir_variable constructor

  • glsl: initialize is_global in ir_variable_refcount_entry constructor

  • pps: Initialize Counter member variable units

  • egl: Fix int-to-void-pointer-cast warning on macOS

  • util: Fix NSEC_PER_SEC macro-redefined warning on macOS

Vinícius (1):

  • ac/gpu_info: make PCIe 4.0 and 5.0 lane bandwidth exact multiples of 3.0

Vlad Zahorodnii (1):

  • wsi/wayland: Add support for wl_fixes.ack_global_remove

Víctor Manuel Jáquez Leal (3):

  • anv/video: don’t report VK_IMAGE_CREATE_VIDEO_PROFILE_INDEPENDENT_BIT_KHR

  • radv/video: report hasOverrides on AV1 encoding

  • anv/video: report back codec overrides

Wei Zhao (1):

  • wsi/display: allocate the EDID display name only once

Xaver Hugl (1):

  • egl/wsi/wayland: ignore default_dmabuf_feedback_done

Xianzhong Li (2):

  • panfrost/csf: flush GPU caches after XFB compute in csf_launch_xfb

  • panfrost/csf: initialize flush_id to zero in the tiler OOM handler

Xueliang Zhang (1):

  • radeonsi: dirty DB_Z_INFO when the sample count changes without Z/S

Yassin Rezai (1):

  • freedreno: Match FD735 regardless of the speed bin fuse

Yi Ding (1):

  • panfrost: clear shader image mask on trailing unbinds

Yiwei Zhang (139):

  • venus: recycle idle sfb cmds only after async wait

  • panvk: amend missing vulkan_icd_link_depends

  • pvr: amend missing vulkan_icd_link_depends

  • lvp: add linker script to fix android symbols

  • venus: clean up sync2 treatment leftovers

  • venus: properly check sync2 enablement

  • venus: host image copy to scrub present_src layout if needed

  • venus: simplify sync feedback tracking

  • venus: update pnext fix tracking

  • venus: explicitly track if need to fix batch

  • venus: flatten sync feedback cmd counting

  • venus: deprecate fence feedback

  • venus: split ring submission to vn_queue_submission_do_submit

  • venus: skip empty batch submission

  • venus: drop vn_sync_payload_external from submission tracking

  • venus: extract vn_timeout_to_poll_timeout

  • venus/virtgpu: only signal non-zero initial value

  • venus/virtgpu/vtest: drop initial value from sync reset

  • venus: prepare for VN_SYNC_TYPE_SYNC

  • venus: migrate fence over to VN_SYNC_TYPE_SYNC

  • venus: relax SYNC_FD fence export requirement

  • venus: deprecate queue idle wait workaround

  • venus: rename to be explicit about sync fd semaphore

  • venus: vn_semaphore_(is|wait)_sync_fd to support VN_SYNC_TYPE_SYNC

  • venus: rename existing queue submission wait semaphore tracking

  • venus: count and prepare storage to scrub SYNC_FD signal semaphore

  • venus: extract syncs and scrub SYNC_FD signal semaphores

  • venus: ensure renderer sync fence is submitted between queue batches

  • venus: migrate SYNC_FD semaphore over to VN_SYNC_TYPE_SYNC

  • venus: relax SYNC_FD semaphore export requirement

  • ci/venus: update expectations

  • venus/virtgpu: hide has_timeline_sync behind a new perf option

  • venus/vtest: advertise timeline syncobj support

  • venus: drop vn_renderer_sync_flags

  • venus: add VN_SYNC_TYPE_TIMELINE_SYNC

  • venus: track queue internal array index within vn_device::queues

  • venus: count and init syncs from timeline semaphores

  • venus: implement host signal for TIMELINE_SYNC

  • venus: implement counter query for TIMELINE_SYNC

  • venus: add vn_wait_semaphores_legacy for legacy wait

  • venus: implement semaphore wait for TIMELINE_SYNC

  • venus: migrate timeline semaphore to VN_SYNC_TYPE_TIMELINE_SYNC

  • venus: document timeline semaphore implementation

  • venus: ensure cached vn_ring_submit batches are bounded

  • subprojects: pull in venus-protocol

  • venus: migrate to build with venus-protocol subproject

  • venus: remove in-tree venus-protocol

  • venus: support VK_KHR_maintenance8

  • venus: refactor to add vn_fill_queue_family_properties

  • venus: update queue family props related namings

  • venus: support VK_KHR_maintenance9

  • venus: support VK_KHR_maintenance10

  • venus: support VK_KHR_maintenance11

  • venus: amend VK_EXT_ycbcr_image_array advertisement

  • subprojects: uprev venus-protocol

  • venus: support VK_KHR_shader_constant_data

  • venus: support VK_EXT_shader_split_barrier

  • venus: support VK_EXT_shader_tile_image

  • venus: support VK_EXT_vertex_attribute_robustness

  • venus: support VK_EXT_zero_initialize_device_memory

  • venus: support VK_KHR_unified_image_layouts

  • venus: support VK_EXT_primitive_restart_index

  • venus: support VK_EXT_rgba10x6_formats

  • subprojects: amend missing venus-protocol version bump

  • vulkan/android: undo forcing linear for mutable format

  • vulkan/android: align vk_ahb_probe_format with returned AHB usage

  • vulkan/android: unify AHB usage bits calculation

  • vulkan/android: allow optimal tiling for unorm and srgb mutation

  • vulkan/util: improve __vk_append_struct

  • vulkan/util: improve vk_physical_device_check_device_features

  • vulkan/util/overlay: improve overlay layer

  • panvk/android: avoid pNext injection of caller ANB info

  • venus: ensure queue submission pNext chain ends properly

  • subprojects: uprev venus-protocol for strict aliasing updates

  • venus: fix remaining strict aliasing violations

  • venus/subprojects: uprev protocol to avoid header install

  • subprojects: uprev venus-protocol to v1.1.3

  • panvk: place coherent memory type before cached type

  • vulkan/android: error log upon AHB alloc or import failure

  • venus: log more upon failed gralloc interactions

  • pan/kmod: dummy BO is per VM for emulated sparse

  • vulkan/android: add a generic helper to retrieve AHB external format

  • vulkan/android: make efr resolve format mapping explicit

  • vulkan/android: support null color efr with legacy render pass

  • vulkan/android: support null color efr dynamic rendering - Part I

  • vulkan/android: support null color efr dynamic rendering - Part II

  • vulkan/android: support null color efr dynamic rendering - Part III

  • pan/format: prepare for YUV render target support

  • pan/desc: support YUV color attachment emit on v10+

  • pan/fb: drop single plane assert from pan_image_view_get_color_plane

  • pan/fb: support YUV render target emit on v10+

  • panvk: patch null color attachment and format for efr

  • panvk: advertise VK_ANDROID_external_format_resolve on v10+

  • venus: ignore imported SYNC_FD payload for signal semaphore inspection

  • docs/lvp: amend to advertise implemented extensions

  • venus: fix DMR on import failure

  • venus: amend TLS ring error check in pipeline cache retrieval

  • venus: fix a race condition for TLS ring creations

  • venus: return TIMEOUT instead of NOT_READY on fence wait timeout

  • venus: fix legacy semaphore wait values to align with semaphores

  • venus: early return for timeline counter zero wait under WAIT_ANY

  • venus: fix lockless query of descriptor size cache

  • venus: fix vk_command_pool adoption

  • venus: recycle sfb cmds when wait is done elsewhere

  • venus: avoid new sfb cmds being prematurely recycled

  • venus: avoid leaking the cpu timeline syncobj

  • venus/sim_syncobj: amend to set errno on wait timeout

  • venus/virtgpu: emulate fence query via wait without timeline sync

  • venus/virtgpu: avoid potential double BO destroy

  • venus: fix maint7 zero layeredApiCount with non-NULL pLayeredApis

  • venus: avoid UB in vn_get_ahb_image_props

  • venus: avoid null ptr dereference for zero external image handleType

  • venus: properly scrub VkGraphicsPipelineCreateInfo::renderPass

  • vulkan/runtime/meson: avoid duplicate header gen deps list

  • venus: fix pWaitDstStageMask when dropping imported semaphores

  • venus/sim_syncobj: convert absolute timeout back to relative

  • venus: avoid deadlock in cs upload for guest vram

  • venus: ignore VkImportMemoryFdInfoKHR if handleType is 0

  • venus: fix planar format host copy size calc

  • venus: consider buffer usage 2 when fixing external buffer pNext

  • venus: only call vn_wsi_sync_wait with valid sync fd

  • venus: push set template shall grab a ref on the desc set

  • venus: refactor out a vn_descriptor_set_layout_fix_create_info helper

  • venus: sanitize GetDescriptorSetLayoutSupport ignored fields

  • venus: amend a missing state in async set alloc

  • venus: properly fail mutable descriptor alloc

  • venus: respect descriptor template entry stride for accel struct

  • venus: fix missed cases in vn_descriptor_heap_get_descriptor_count

  • venus: descriptor size cache to only query supported types

  • venus: always chain dedicated reqs for buffer cache init

  • venus: image format and reqs cache to hash sType of pNext chain

  • venus: fill image mem reqs with cache locked

  • venus: terminate pNext properly in vn_fix_command_buffer_begin_info

  • vulkan/wsi/x11: use resolved present mode to calculate image count

  • kraid/ra: Fix shift overflow in SSAVecRepr::set_affinity

  • kraid/encode_v9: amend missing accum in OpICmpMulti::get_info

  • kraid/encode_v9: validate swizzle before src_info check

  • kraid/hw_tests: fix multiply overflow errors

  • kraid/hw_tests: use wrapping_sub instead of plain -1

Yves Auxier (1):

  • radeonsi: Implement EGL_NV_context_priority_realtime

Zan Dobersek (1):

  • tu: fix R64 image clears via non-generic clear code path

Zhang, Jianxun (7):

  • brw: New interface to get number of grf register blocks

  • iris: Use new interface to get number of register blocks

  • blorp: Use new interface to get number of register blocks

  • brw: Use new interface to get number of register blocks

  • anv: Use new interface to get number of register blocks

  • brw: Remove obsolete interface to get number of used registers

  • anv: Require dedicated allocation query on scanout images from WSI

adrian baker (26):

  • jay: Use new lane_id fn for load_subgroup_invocation.

  • jay: Various geometry shader fixes.

  • anv: Use quad primitives for BVH encoding

  • brw: expose mesh/task compile helpers to jay

  • brw: lower pack_64_2x32 for jay

  • brw: prohibit vec8 SENDs in SIMD32 shaders

  • brw: use jay-specific simd selection for all compute-like stages

  • nir: mark mesh_shader urb input handles as convergent

  • brw: disable coarse pixel shading lowering for Jay mesh shaders

  • jay: refactor emit_lsc_fence to not be strongly coupled to nir instr

  • jay: implement mesh/task shaders

  • nir: mark task and mesh shader urb outputs as convergent

  • jay: implement fs interop for mesh/task shaders

  • jay: lower interpolateAtOffset in NIR with correct perspective

  • jay: defer moving SEND ex_desc to address register

  • jay: port over brw_opt_txf_combiner

  • anv/rt: use 64-bit sizes for ray query shadow stacks

  • jay: fix convergent sampler SENDs w/ implicit LOD

  • jay: use addr reg for SEND ex_desc if provided

  • anv/rt: optimize bvh update quantized bounds calculation

  • jay: restrict offset_packed_pixel_coords to STRIDE_4

  • jay: add option for omitting mlen from SEND ex_desc

  • brw: disable ‘avoid ternary with two constants’

  • brw: use find_msb_rev instead of find_msb

  • anv,jay,brw: make intel_use_jay togglable per-shader

  • brw,jay: get rid of isub in brw_nir_opt_divergent_atomics

airidosas252 (1):

  • freedreno/fdl: Use NUM_ELEMENTS for A8XX buffer descriptors

benrichard-amd (1):

  • gallium/ddebug: exclude ddebug from non-gfx builds

bslette (1):

  • nak: never encode IDP.4A with an immediate source

compilersutra (1):

  • freedreno/drm-shim: add Adreno 735 (FD735) for shader-db

faetalize (2):

  • util/vl_vlc: allow removebits at the top of the bit buffer

  • util/rbsp: check each byte once for emulation prevention

genie1997 (1):

  • tu/kgsl: Force 8-channel macrotile mode on A635

hwandy (1):

  • anv: defer batch BO allocation for command buffers until recording

jiajia Qian (3):

  • pan/libpan: add a direct buffer->buffer copy kernel

  • panfrost: accelerate buffer->buffer copies with a compute shader

  • spirv,panfrost: emit precise OpenCL log/log2/exp2 as hardware transcendentals

joe.wang (5):

  • radv: enable 2D swizzle for 3D storage images on Ghost Recon Breakpoint

  • radv: also apply the DiRT 4 HiZ workaround profile to the native port

  • radv/gfx12: add full_rez HiZ workaround variant

  • radv: use the full_rez HiZ workaround for Sniper Elite 5

  • radv: enable 2D swizzle for 3D storage images on Total War: ROME

kajal (1):

  • rusticl/kernel: consider current dimension in prescale logic

kas5852 (1):

  • anv: use brw for RE Engine titles with emulated divergent barriers

kiritakekumi (1):

  • pco: Enable IMAD32 fusion in NIR

llyyr (1):

  • radv/amdgpu: fix BO list leak on command submit failure

matthewperiut (2):

  • venus: honor the virtio-gpu blob alignment

  • venus: skip virtio-gpu devices without 3D

mitul72 (1):

  • radv: Clear stale vertex user SGPR state when binding mesh shaders

mmes (2):

  • anv: fix upload of shader kernels spanning more than one heap BO

  • jay/lower_scoreboard: clear $.src after $.dst wait

mythilam (1):

  • radv/sqtt: fix RGP capture cancel when the SQTT buffer hits its limit

osman.yagci (3):

  • util: probe for _SC_PHYS_PAGES, not just for sysconf()

  • util/u_queue: fail util_queue_init when mtx_init or cnd_init fails

  • egl/dri2: destroy the display lock when the display is destroyed

rajaryan1492007 (1):

  • spirv: Don’t implicitly set ACCESS_NON_WRITEABLE for input attachments

rowanG077 (1):

  • agx: fix phi register shuffling order

sergiuferentz (3):

  • gfxstream: default unknown AHB dataspace to BT.709 (sRGB)

  • gfxstream: map DRM_FORMAT_ARGB8888/XRGB8888 to VK_FORMAT_B8G8R8A8_UNORM

  • gfxstream: Use GFXSTREAM_ERROR and GFXSTREAM_FATAL in cereal codegen

soka (6):

  • zink: check the unowned buffers before returning early

  • zink: check the primitive type after zink_batch_rp()

  • zink: bind the vertex buffers after zink_batch_rp()

  • zink: wait for dma-buf fences when the batch ends

  • zink: track batch dmabuf exports in a hash_table instead of a set

  • zink: wait for dma-buf readers only when writing

squidbus (35):

  • kk: Move nir_lower_io_vars_to_temporaries to early msl_preprocess_nir

  • kk: Advertise proper rendering granularity

  • wsi/metal: Implement present ID and present wait

  • kk: Advertise VK_KHR_present_(id/wait)(2)

  • kk: Remove unused mtl_drawable_present

  • kk: Determine supported samplers by Metal limit

  • kk: Support VK_EXT_custom_border_color and VK_EXT_border_color_swizzle

  • kk: Support VK_EXT_map_memory_placed

  • poly: Support emulating last provoking vertex when unrolling

  • kk: Support VK_EXT_provoking_vertex

  • kk: Support VK_EXT_image_view_min_lod

  • kk: Re-format code

  • kk: Do not pass layer count to Metal for attachment-less rendering

  • kk: Release drawable on command buffer free

  • kk: Simplify memory budget calculations

  • kk: Support VK_EXT_depth_clip_enable

  • kk: Disable workaround 8 and 11 on macOS 27

  • kk: Fix disable of workaround 8

  • kk: Pad image store value to 4 components

  • nir/opt_varyings: Change non-interpolated I/O to uint type.

  • kk: Enable nir_io_non_interpolated_as_uint

  • kk: Support depthBounds on M5+

  • kk: Support VK_EXT_sampler_filter_minmax on M5+

  • kk: Set MSL compile version based on OS

  • kk: Expose shaderSharedFloat32AtomicAdd for MSL >= 4.1

  • kk: Add feature check when setting depth bounds test

  • vk/meta: Add vk_meta_update_memory

  • kk: Enable VK_KHR_deferred_host_operations

  • kk: Implement vkCmdUpdateMemoryKHR and vkCmdCopyQueryPoolResultsToMemoryKHR

  • kk: Lower alpha-to-coverage and alpha-to-one in shader

  • kk: Raise max samplers to table size for M3+

  • kk: Fix ViewIndex when multiview is off

  • kk: Map triangle fan topology class

  • kk: Account for mip tail in level size corner padding

  • kk: Fix storageInputOutput16

sunshineinabox (1):

  • tu: implement VK_QCOM_rotated_copy_commands

utzcoz (18):

  • aco: use s_pack_hh with shifted constant when s_pack_hl is unavailable

  • aco: remove TODO in try_optimize_branching_sequence

  • virtio: magma-gpu-rs: accept a null device in virtgpu_kumquat_finish

  • gfxstream/guest: import host memory through an fd on Apple too

  • gfxstream/guest: do not define LINUX_GUEST_BUILD on non-Linux hosts

  • gfxstream/guest: let the Vulkan driver link on Mach-O

  • gfxstream/guest: do not define VK_ANDROID_native_buffer on Apple

  • util: make the event signaller and waiter explicit

  • util: hand the descriptors to Tube::send

  • magma-gpu-rs: sort the create_event_pair import

  • util: hand the descriptor to WritePipe::new

  • subprojects/rustix: enable the features an Apple backend needs

  • subprojects: add a wrap for mach2

  • util: implement the sys layer for Darwin

  • gfxstream/guest: stub the virtio-gpu backend on Apple

  • gfxstream/guest: present through the Metal WSI on Apple

  • gfxstream/guest: route vkMapMemory2 through the resource tracker

  • gfxstream: route the device buffer memory requirements query through global state

zhaqian (1):

  • radv/android: suggest narrow YCbCr range for AHB YCbCr formats

zsoli (2):

  • nir: add sample_mask_out_declared to shader info

  • radv: disable VRS when output sample mask is used