Quite easy now that the yak is shaved.
For info on the feature: http://llvm.org/devmtg/2016-11/Slides/Nemet-Compiler-assistedPerformanceAnalysis.pdf
Makes much easier to visualize optimization decisions (loop vectorizations, inlining decisions etc..). I plan to use it for tuning profile-guided inlining.
This was only supported for Mach-O, but I happen to be interested in another object format.
Details
Details
Diff Detail
Diff Detail
- Repository
- rL LLVM
Event Timeline
Comment Actions
Generally looking file, but I'll let people who are familiar with this feature to sign off.
What is expected to happen if you pass only -opt-remarks-filename but no -opt-remarks-with-hotness?
Comment Actions
The idea being that without hotness the opt-diag emitter pass does not require BlockFrequencyInfo which may cost some comple-time.