Daniel Vetter has sent in his latest drm-intel-next pull request for DRM graphics driver code targeting the Linux 3.19 kernel.
There's just 2~3 weeks until DRM-Next is closing early again for Linux 3.19 so Vetter on Monday sent in another pull request. There's already been new code queued up that includes initial hardware enablement for Skylake while more code is ready now.
The latest batch of Intel DRM changes for the next kernel cycle include suspend/resume/freeze/thaw unification work, WA list improvements, more kernel documentation of the i915 driver code, 180 rotation support for cursors, and many fixes all over the place.
Vetter posted his atomic mode-setting patch series in latest form on Sunday. There's the helper libraries for migrating over to atomic mode-setting and the other core/driver interface changes for this work. The description on his latest patch series is quite lengthy so check it out if you're wanting to learn some more. These patches though don't offer the actual atomic mode-setting ioctl to expose to user-space.
The same Intel Core i7 4770K system used for yesterday's Windows vs. Linux graphics benchmarks were used when benchmarking the GeForce GTX 780 Ti, 970, and 980 graphics cards. Windows 8.1 Pro x64 had all available system updates at the time and was running the NVIDIA 344.48 WHQL binary driver that was their latest release at the time of testing. When running Ubuntu 14.10 x86_64 on the system with its Linux 3.16 kernel, the NVIDIA 343.22 driver was used. The 343.22 driver was the latest publicly available proprietary Linux driver at the time of testing and their first to support the GTX 970/980 under Linux. All of the same hardware was used under each operating system and each OS was with its software default settings as were the driver settings.
In an interesting change of events, Code Aurora on the behalf of the Qualcomm Innovation Center has added Adreno A4xx product support to the Freedreno-spawned DRM/KMS "MSM" driver.
Rob Clark started the Freedreno project over two years ago as a reverse-engineered project around Qualcomm's Adreno hardware. At the time Rob was working for Texas Instruments but now is employed by Red Hat. The Freedreno driver has largely been developed just by Rob with contributions by a few others, but without any official support from Qualcomm. Freedreno is to Adreno hardware as Nouveau is to NVIDIA hardware. Like Nouveau, Rob developed Freedreno code through clean-room reverse engineering.
For those curious how the latest open-source Intel Linux graphics driver is performing against Intel's newest closed-source Windows OpenGL driver, we've put Ubuntu 14.10 (including a second run with the latest Linux kernel / Mesa) against Microsoft Windows 8.1 with the newest Intel GPU driver released earlier this month.
As I've been mentioning on Twitter, fresh Windows 8.1 vs. Ubuntu 14.10 benchmarks are coming in the days ahead with Intel HD Graphics, AMD Radeon, and NVIDIA GeForce hardware. With the AMD/NVIDIA tests it includes the closed-source, binary drivers for Linux too. The Intel Haswell graphics results are up first since that's the most straight forward with Intel's Linux graphics support only coming through via their open-source driver.
These results are much more interesting than the earlier two-disk HDD benchmarks now using solid-state storage and having bought four Intel Series 530 120GB SSDs for making this an interesting RAID comparison. Four of the Intel SSDSC2BW120A4K5 solid-state drives were used in their 120GB capacity. Each of these solid-state drives retail for $75~80 USD and features sequential reads up to 540MB/s and sequential writes up to 480MB/s with its Serial ATA 3.0 interface. The 2.5-inch SSD 530 Series drive is rated by a five-year warranty and uses 20nm Intel NAND MLC memory.
Kristian's latest patches being made public are enabling support for vertex shaders to be generated using Intel's SIMD8 scalar back-end for Broadwell hardware and newer. "With Broadwell we have the option to run vertex shaders in scalar (SIMD8) mode which potentially gives us better throughput and more vertices per thread dispatch. This patch series implements this by repurposing our [fragment shader] backend to also work for vertex shaders."
This patch series for VS SIMD8 support deals with just under one thousand lines of code. This should hopefully lead to a performance win, but of course we're still waiting for Broadwell hardware to actually arrive. The Broadwell ultrabooks / convertible tablets should hopefully not be too many weeks out now in the US (so far it's mostly just the Lenovo Yoga 3 Pro available) and it will be interesting to see how they perform compared to Haswell but the Broadwell desktop CPUs won't see the light of day until 2015. The Broadwell Linux support should be fairly rock solid by now with the open-source Intel developers working on the hardware enablement for more than one year with most work now just being about more fine tuning and optimizations.
As a follow-up to last week's Ubuntu 14.04 vs. Ubuntu 14.10 AMD Performance Comparison and yesterday's Radeon R9 290: Gallium3D vs. Catalyst driver comparison, here's taking things further in looking at the performance of the open-source AMD Radeon Linux graphics driver in several configurations while compared against the closed-source AMD Catalyst graphics driver as found on Ubuntu 14.10.
For those stuck running on the R300g driver, which supports the ATI Radeon X1000 (R500) series and older GPUs, you really should consider upgrading your graphics card and likely your system. But if you're set on using the R300g driver going into the foreseeable future, you might as well upgrade Mesa.
While the R300 Gallium3D driver seldom sees new commit activity to mainline Mesa, as of today the VDPAU state tracker support was re-enabled for the R300g driver for its very basic video playback support.