Hi list,

Mesa 19.3.0-rc5 is now available as the latest release in the 19.3 series. This
is a pretty small release, likely due to tomorrow being a major US holiday. The
majority of the changes are for radv, but there's a few other bits and pieces
here too: v3d, r600, freedreno, and old intel, just to name a few.

Dylan

Shortlog
========


Alejandro Piñeiro (1):
      v3d: adds an extra MOV for any sig.ld*

Bas Nieuwenhuizen (2):
      radv: Do not change scratch settings while shaders are active.
      radv: Allocate cmdbuffer space for buffer marker write.

Dave Airlie (1):
      llvmpipe/ppc: fix if/ifdef confusion in backport.

Dylan Baker (1):
      VERSION: Bump version for -rc5

Eric Engestrom (1):
      vulkan: delete typo'd header

Gert Wollny (1):
      r600: Disable eight bit three channel formats

Hyunjun Ko (1):
      freedreno/ir3: fix printing output registers of FS.

Ian Romanick (1):
      intel/fs: Disable conditional discard optimization on Gen4 and Gen5

Jose Maria Casanova Crespo (1):
      v3d: Fix predication with atomic image operations

Timothy Arceri (3):
      radv: add some infrastructure for fresh forks for each secure compile
      radv: add a secure_compile_open_fifo_fds() helper
      radv: create a fresh fork for each pipeline compile

Yevhenii Kolesnikov (2):
      glsl: Enable textureSize for samplerExternalOES
      meson: Fix linkage of libgallium_nine with libgalliumvl

Zebediah Figura (1):
      Revert "draw: revert using correct order for prim decomposition."


git tag: mesa-19.3.0-rc5

Also: Mesa 19.3-RC5 Brings RADV Secure Compile Update, Other Fixes

As part of the exciting benchmark week and our ongoing tests of Intel Ice Lake on Linux, this next piece has been driven out of curiosity... While recently I posted new benchmark results of Intel Haswell to Ice Lake laptop performance, what about going further back like to the days of Nehalem? Here is that comparison of Core i7 Nehalem to Core i7 Ice Lake including power / performance-per-Watt data, thermal, and performance-per-MHz data too. Enjoy this fun comparison for how the Intel mobile performance on Ubuntu has evolved over the past decade.

The Nehalem part used is the ten-year-old Core i7 720QM "Clarksfield" processor. This CPU offers four cores / eight threads, 1.6GHz base frequency, 2.8GHz turbo frequency, a 6MB cache, and a 45 Watt TDP. Clarksfield is the mobile variants while Lynnfield made up the desktop side for the 45nm Nehalem microarchitecture.

• Radeon ROCm 2.10 Released With SUSE 15 SP1 Support, rocBLAS For Complex GEMM

  While AMD announced ROCm 3.0 earlier this month at SuperComputing 19 as the next major iteration to Radeon Open Compute, it looks like they aren't quite ready to ship it and instead released ROCm 2.10.

  ROCm 3.0 support was noted in their SC19 press release with having new HIP-clang improvements for better running CUDA codes on Radeon GPUs, expanded HPC programming model support with the likes of OpenMP / NAMD / LAMMPS, and new support for system/workload deployment tools like Kubernetes. ROCm 3.0 sounds like it should be a great release, thus surprising today they diverted to releasing ROCm 2.10.

• Wayland's Weston 8.0 Bringing Direct-Display Extension, HDCP On DRM Back-End + More

  Simon Ser has stepped up again to manage the upcoming release of the Weston 8.0 Wayland reference compositor. No Wayland update itself is planned with nothing real to release at this point, but Weston 8.0 should arrive before the end of January.

  For making a January release target for Weston 8.0, Simon has laid out plans for the initial alpha release around 6 December, a beta release on 20 December, and the first release candidate on 17 January. If all goes well with Weston 8.0 RC1, the actual release could be on or around 24 January.

• X.Org's Modesetting Driver Gets Smarter - Queries Mesa For Which GL Driver To Use

  There is a list of PCI ID driver mappings that has needed to be maintained for correlating graphics hardware to the respective DRI driver that should be loaded. That's been another maintenance burden and rather vintage, but now a modern solution has landed thanks to Intel's Ken Graunke. Since earlier this year has been the MESA_query_driver EGL extension for being able to query the name of the running 3D/OpenGL driver on the system. EGL_MESA_query_driver is now being used for dynamically querying the driver that was loaded by Mesa rather than having to duplicate the PCI ID driver mapping logic within X.Org Server.

  Besides the burden of having to maintain multiple graphics PCI ID mappings in 2019, the motivation for Graunke to get this change made is for Intel's transition to the "Iris" Gallium3D driver he has been leading. With Intel Gen12 / Tiger Lake graphics, only Iris is supported and starting with Mesa 20.0 they intend to default to this Gallium3D driver over the "i965" classic driver. Rather than having to work this logic into their mapping for the xorg-server, EGL_MESA_query_driver can simply leverage the logic from Mesa's driver loader.

Last week marked the release of Blender 2.81 with one of the shiny new features being the OptiX back-end for the Cycles engine to provide hardware-accelerated ray-tracing with NVIDIA RTX graphics processors. Long story short, OptiX is much faster for Blender than using NVIDIA's CUDA back-end -- which already was much faster than the OpenCL support within Blender. For your viewing pleasure today are benchmarks of 19 different graphics cards looking at the CUDA performance from Maxwell to Pascal to Turing and then for the RTX graphics cards also the OptiX performance.

OptiX is only supported with the NVIDIA RTX graphics cards but it offers a significant boost to the rendering performance. This NVIDIA-designed API for exploiting their RT cores introduced with Turing yields an impressive speed-up for Blender render times in common benchmark scenes. For more background information on OptiX with Blender 2.81 can be found via this Blender.org blog post from the summer.