Mesa 21.3.7 Release Notes / 2022-02-23

Mesa 21.3.7 is a bug fix release which fixes bugs found since the 21.3.6 release.

Mesa 21.3.7 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 21.3.7 implements the Vulkan 1.2 API, but the version reported by the apiVersion property of the VkPhysicalDeviceProperties struct depends on the particular driver being used.

SHA256 checksum

b4fa9db7aa61bf209ef0b40bef83080999d86ad98df8b8b4fada7c128a1efc3d  mesa-21.3.7.tar.xz

New features

  • None

Bug fixes

  • lavapipe: dEQP-VK.spirv_assembly.instruction.compute.float16.arithmetic_3.step fails

  • ANV: Bad output from TransformFeedback . Regression from Mesa 21. Something to do with VB+XFB -> VB+XFB dependency?


Alyssa Rosenzweig (4):

  • pan/bi: Avoid *FADD.v2f16 hazard in optimizer

  • pan/bi: Avoid *FADD.v2f16 hazard in scheduler

  • pan/bi: Lower swizzles on CSEL.i32/MUX.i32

  • panvk: Use more reliable assert for UBO pushing

Bas Nieuwenhuizen (1):

  • radv: Fix preamble argument order.

Connor Abbott (1):

  • ir3/spill: Fix simplify_phi_nodes with multiple loop nesting

Dave Airlie (3):

  • lavapipe: fix sampler + sampler view leaks.

  • lavapipe: reference gallium fences correctly.

  • crocus: fix leak on gen4/5 stencil fallback blit path.

Emma Anholt (1):

  • i915g: Initialize the rest of the “from_nir” temporary VS struct.

Eric Engestrom (2):

  • .pick_status.json: Update to dabba7d7263be6ffb6f3676465e92c65952fa824

  • .pick_status.json: Mark b07372312d7053f2ef5c858ceb1fbf9ade5e7c52 as denominated

Ian Romanick (9):

  • gallivm/nir: Call nir_lower_bool_to_int32 after nir_opt_algebraic_late

  • nir: All set-on-comparison opcodes can take all float types

  • intel/fs: Don’t optimize out 1.0*x and -1.0*x

  • spriv: Produce correct result for GLSLstd450Step with NaN

  • spirv: Produce correct result for GLSLstd450Modf with Inf

  • spirv: Produce correct result for GLSLstd450Tanh with NaN

  • nir: Properly handle various exceptional values in frexp

  • nir: Produce correct results for atan with NaN

  • nir: Add missing dependency on

Faith Ekstrand (1):

  • anv: Call vk_command_buffer_finish if create fails

Jonathan Gray (1):

  • dri: avoid NULL deref of DrawBuffer on flush

Lionel Landwerlin (2):

  • nir: fix lower_memcpy

  • anv/genxml/intel/fs: fix binding shader record entry

Marcin Ślusarz (1):

  • anv: don’t set color state when input state was requested

Marek Olšák (1):

  • ac/surface: add more elements to meta equations because HTILE can use them

Mike Blumenkrantz (4):

  • lavapipe: use util_pack_color_union() for generating clear colors

  • aux/draw: fix llvm tcs lane vec generation

  • zink: always set VkPipelineMultisampleStateCreateInfo::pSampleMask

  • zink: always invalidate streamout counter buffer if not resuming

Nanley Chery (1):

  • iris: Don’t fast clear with the view format

Pavel Ondračka (1):

  • r300: fix transformation of abs modifiers with negate

Qiang Yu (3):

  • radeonsi: workaround Specviewperf13 Catia hang on GFX9

  • radeonsi: fix depth stencil multi sample texture blit

  • glx: fix pbuffer refcount init

Samuel Pitoiset (1):

  • radv/winsys: fix initializing debug/perftest options if multiple instances

Tapani Pälli (5):

  • intel/genxml: add PIPE_CONTROL field for L3 read only cache invalidation

  • anv: invalidate L3 read only cache when VF cache is invalidated

  • iris: invalidate L3 read only cache when VF cache is invalidated

  • iris: fix a leak on surface states

  • mesa/st: always use DXT5 when transcoding ASTC format

Thierry Reding (2):

  • tegra: Use private reference count for sampler views

  • tegra: Use private reference count for resources

Timur Kristóf (1):

  • radv: Disable IB2 on compute queues.

Yiwei Zhang (1):

  • venus: properly destroy deferred ahb image before real image creation