Roman Lebedev

[Roman Lebedev @LebedevRI]

The sky is always bluer on the other side https://t.co/LUsEmHCi85

[Roman Lebedev @LebedevRI]

RIP, Dad. I can only hope your were happy in your life. May your next reincarnation perhaps be a higher one, or at least with a better life, whenever/wherever/whatever it is. Perhaps we may talk again, in my lifetime, even if only in an 'Astral' realm.

[Roman Lebedev @LebedevRI]

#PipeWire 25 bits is enough for everyone. @PipewireP https://t.co/2FzbbDcr66

[Roman Lebedev @LebedevRI]

Better late than never? Coming to a #NinjaBuildSystem near you, at last! Kudos to @esgamesorg! https://t.co/CLW2dq5Oq1 @phoronix

[Roman Lebedev @LebedevRI]

It took longer than i thought to get rid of this eyesore, but here we are at last! Even though the code has already been fuzz-clean for "eternity", actually having proper bounds checks on everything is a major milestone at last. @pixlsus @darktable_org https://t.co/wvE7DJ4DA1

[Roman Lebedev @LebedevRI]

#EBU #R128 #EBUR128 Best audio player? https://t.co/s3ydlTRYdZ

[Roman Lebedev @LebedevRI]

@llvmorg @SylvestreLedru @alexreinking Fascinating times :) #LLVMWorldDomination

[Debian upload @DebianUpload]

New upload: halide 12.0.1-1 by Roman Lebedev into unstable - Closes: https://t.co/wP1HftBeXH... https://t.co/75xx23cBDN

[Debian New Packages @DebianNew]

New package: halide Version: 12.0.1-1 by Roman Lebedev ... https://t.co/ukPZxWcSxy

[Roman Lebedev @LebedevRI]

@llvmorg @AMD @AMDRyzen #Zen3 #LLVM #Clang Interesting results, showing just how much proper per-CPU tuning matters. Free compute performance boost! You're welcome?

[Phoronix @phoronix]

An Early Look At #LLVM #Clang 13 Performance On @AMD #Zen3

-- This initial round of Clang 13 compiler benchmarking on a @Tyan @AMDServer #EPYC

https://t.co/6qFSCqoOb3

[Roman Lebedev @LebedevRI]

@pervognsen @matt_dz It is 4/7cy: https://t.co/5wBXRsk2c9 +1cy is coming from https://t.co/zHXEW0D4kT I didn't measure that particular number, it could be wrong, esp. since mov are 0cy. I'm not sure how to confirm this though. https://t.co/pMBZpavH0o measures @ 6.5cy/load+store. What's store latency?

[Roman Lebedev @LebedevRI]

@llvmorg @llvmweekly @phoronix @AMD @AMDRyzen
6 months for `-march=znver3` is better than 6 years it took for `-march=bdver2` :) Though, it actually only took me ~42 days since i started needing said sched model.

[Phoronix @phoronix]

The @AMDServer @AMDRyzen @AMD #Zen3 Scheduler Model Finally Added To #LLVM / #Clang

https://t.co/FnfkwIUZDr

[Roman Lebedev @LebedevRI]

@johnregehr That transform was a fun one to come up with!
Gaining a way to actually prove it (and that was made possible about the same time) was perhaps even more awesome, thanks @vpshufb & others!

[Roman Lebedev @LebedevRI]

@llvmorg @llvmweekly @SonarSource @GAnnCampbell @gaudol It took 3 years, but #LLVM's #clangtidy finally has a function cognitive complexity metric checker! I'm hopeful that having it in a widely-used #clang-driven linter makes good #c/#cpp easier to write!
https://t.co/rTgphR7pb2

[Roman Lebedev @LebedevRI]

Achievement unlocked: branch predictors are fun https://t.co/PFUeMzPBHa

[Roman Lebedev @LebedevRI]

@pixlsus @darktable_org @llvmorg Now that https://t.co/6mtwTRABFb fixed #llvm to not prevent vectorization by hoisting common code into loop header, i've *finally* found some energy to rewrite #RawSpeed's Canon .CR2 decoding code, with good perf results https://t.co/tmJY3690Ki

[Roman Lebedev @LebedevRI]

@johnregehr Since there is already structure-aware LLVM IR fuzzing, the very obvious next step would be to integrate alive2 into instcombine fuzzer - having a way to tell if IR transformation is a miscompile or not is great, but it requires preexisting IR to function..

[Roman Lebedev @LebedevRI]

@johnregehr Circling back to this, increment/decrement support done too: https://t.co/al2vgn1atl
i think mostly only bit field handling is missing..
(did not check volatile/atomic/enum status)

[Roman Lebedev @LebedevRI]

More UB fun, with live miscompilations in-the-wild: applying non-zero (in C - any) offset to null pointer is undefined behavior in C++
https://t.co/gsL1nY1bN3 @johnregehr @kayseesee @Philip_Reames #llvm #clang #ubsan

[Roman Lebedev @LebedevRI]

@johnregehr "[LangRef] Document forward-progress requirement"
https://t.co/mP758O3BGl

[Roman Lebedev @LebedevRI]

Who doesn't love benchmarking? @runtime

[Google Open Source @GoogleOSS]

Congratulations to our Open Source Peer Bonus winners! https://t.co/LA1MsnsSQf