Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]>, Lukas =?utf-8?q?Döllerer?= <[email protected]> Message-ID: In-Reply-To: <llvm.org/llvm/llvm-project/pull/[email protected]>
aengelke wrote: Not the author (collaborator on the project), but I think this should be fine. Some more context: as part of our research project on Wasm PGO, we found that single-bit hints are in general not too useful (e.g. for LLVM-based runtimes like WAMR). In particular, there's no universally good value for the thresholds when to emit the hints (thresholds good for one application ended up being bad for another) and bad hints actively harm performance -- the current defaults of 50%/50% are especially bad and almost always harmful. For our subsequent work, we therefore moved to 7-bit hints, which can be encoded as single-byte LEB128 integer; this yielded much better results. But this is not 100% compatible with the current spec (just backwards compatible); pushing spec changes was out of scope for us. https://github.com/llvm/llvm-project/pull/146230 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
