LLVM 4.0 Causes Slow Performance For RadeonSI?
Several times in the past few weeks I've heard Phoronix readers claim the LLVM 4.0 SVN code causes "slow performance" or has rendering issues. Yet it's gone on for weeks and I haven't seen such myself, so I decided to run some definitive tests at least for the OpenGL games most relevant to our benchmarking here.
Readers have claimed many times recently that LLVM 4.0's AMDGPU back-end is buggy or causing bad performance. Even yesterday it was brought up again and to well known AMD developer Marek, he commented, "That's interesting. Are you sure LLVM 4.0svn causes the low performance?"
I decided to run some Black Friday benchmarks comparing Mesa 13.1-dev Git when built against LLVM 3.9 and then against built against LLVM 4.0 SVN as of this week... Tests with an R9 Fury.
More results via this OpenBenchmarking.org result file, but for better or worse, I don't see any big performance letdowns from LLVM 4.0 SVN's AMDGPU back-end nor with any of those tests had I noticed any glaring rendering differences. Feel free to share your own experiences by commenting on this article in the forums.
Readers have claimed many times recently that LLVM 4.0's AMDGPU back-end is buggy or causing bad performance. Even yesterday it was brought up again and to well known AMD developer Marek, he commented, "That's interesting. Are you sure LLVM 4.0svn causes the low performance?"
I decided to run some Black Friday benchmarks comparing Mesa 13.1-dev Git when built against LLVM 3.9 and then against built against LLVM 4.0 SVN as of this week... Tests with an R9 Fury.
More results via this OpenBenchmarking.org result file, but for better or worse, I don't see any big performance letdowns from LLVM 4.0 SVN's AMDGPU back-end nor with any of those tests had I noticed any glaring rendering differences. Feel free to share your own experiences by commenting on this article in the forums.
27 Comments