Mesa CI Begins Making Use Of Mold Linker For "Substantial" Performance Improvement
For speeding up the actual Mesa continuous integration (CI) process itself with frequently building new revisions of Mesa3D, their CI infrastructure is beginning to make use of the Mold linker as a high performance alternative to the GNU Gold and LLVM LLD linkers. This is yielding a "substantial" performance improvement in tests for being able to turnaround CI jobs faster and in turn allowing Mesa developers to be more efficient.
A month ago was a issue raised over re-evaluating the linker options for Mesa's continuous integration. Mold was raised there was it would "bring another speedup, but of course it can get abandoned at any moment." The abandonment issue is over Mold currently being a standalone open-source project led by Rui Ueyama, the developer also originally involved with LLVM's LLD.
As I've been covering in many Phoronix articles over the past year of Mold coming together nicely, Mold's performance is looking great and far ahead of alternatives. Mesa developer Mike Blumenkrantz commented around the Mold prospects at the time:
Some bugs were uncovered with trialing Mold for linking Mesa and those issues since addressed in upstream Mold.
With this merge as of yesterday, Mesa's continuous integration is now using Mold for linking on x86_64 and AArch64. Its description of the change is simply:
This performance boost is just around the speed for linking when compiling Mesa but doesn't affect the run-time graphics driver performance. It also doesn't impact the default configuration for Mesa builds by users, but is just about their CI setup. In any event now using Mold for Mesa CI builds allow them to turn around jobs quicker and in turn allowing Mesa developers to be more productive and efficient. Plus their CI jobs being more efficient may also end up help them with their previously ballooning cloud CI costs.
Given the widespread performance advantages of Mold for linking, hopefully more open-source projects evaluate possibly using this GNU Gold and LLD alternative.
A month ago was a issue raised over re-evaluating the linker options for Mesa's continuous integration. Mold was raised there was it would "bring another speedup, but of course it can get abandoned at any moment." The abandonment issue is over Mold currently being a standalone open-source project led by Rui Ueyama, the developer also originally involved with LLVM's LLD.
As I've been covering in many Phoronix articles over the past year of Mold coming together nicely, Mold's performance is looking great and far ahead of alternatives. Mesa developer Mike Blumenkrantz commented around the Mold prospects at the time:
The performance improvement is substantial, and I think we would get a noticeable decrease in build times for many jobs. On my system here, linking gallium goes from ~20 seconds to ~1 second.
Some bugs were uncovered with trialing Mold for linking Mesa and those issues since addressed in upstream Mold.
The Mold linker's logo...
With this merge as of yesterday, Mesa's continuous integration is now using Mold for linking on x86_64 and AArch64. Its description of the change is simply:
mold is a fancy new linker that's really fast.
This performance boost is just around the speed for linking when compiling Mesa but doesn't affect the run-time graphics driver performance. It also doesn't impact the default configuration for Mesa builds by users, but is just about their CI setup. In any event now using Mold for Mesa CI builds allow them to turn around jobs quicker and in turn allowing Mesa developers to be more productive and efficient. Plus their CI jobs being more efficient may also end up help them with their previously ballooning cloud CI costs.
Given the widespread performance advantages of Mold for linking, hopefully more open-source projects evaluate possibly using this GNU Gold and LLD alternative.
10 Comments