Thomas Fransham

[Thomas Fransham @fsfodx]

@lemire @geofflangdale Except they skipped kmask overloads in there API and punted it to the JIT to try and magic them up out of code pattens.

[Thomas Fransham @fsfodx]

@KooKiz WerFault may be using windows version of fork via the use of the PssCaptureSnapshot API

[Thomas Fransham @fsfodx]

@lemire Windows 11 has effectively made the same requirements recently as well https://t.co/vTbJOoW22l

[Bob Pony @TheBobPony]

So HUGE discovery found in Windows 11 Version 24H2, since build 25905.
A CPU with the instruction "POPCNT" IS NOW REQUIRED!
There are various system files requiring the POPCNT CPU instruction, from the Windows 11 kernel to the USB XHCI drivers.

Without POPCNT, it doesn't boot!

[Thomas Fransham @fsfodx]

@JokerEph @hansw2000 It might be "A Wishlist for Faster LLVM Back-ends" by Alexis Engelke. There was earlier talk at CGO "Building a Fast Back-End for LLVM" https://t.co/QzE73MHwUC

[Thomas Fransham @fsfodx]

@b0rk I think anything staged will still be kept around in the loose object store until something triggers a git gc.

[Thomas Fransham @fsfodx]

@SheriefFYI @UnadjustedGame Here's a link to that section https://t.co/z9dlGcVyuS. They've not split out most talks to separate videos yet so its in a huge livestream video.

[Thomas Fransham @fsfodx]

@rflaherty71 @BanchouBoo Doesn't it also skipped sending IO Read requests through the filesystem Minifilter stack on Windows 11 as well?

[Thomas Fransham @fsfodx]

@corsix @FelixCLC_ @ChrisGr93091552 I thought a lot of those issues went away with Sandybridge, except for 3 cycle decoder delay for ops with 16 bit immediate values.

[Thomas Fransham @fsfodx]

@mikulasflorek Are you not able to use the FFI system to call back in to the engine?

[Thomas Fransham @fsfodx]

@cmuratori @NoHatCoder For AMD you stuck having to calculate the frequency from the P0 PStateDef MSR i think

[Thomas Fransham @fsfodx]

@konradkokosa Have you added anything about the frozen object heap?

[Thomas Fransham @fsfodx]

@estama2 @pcwalton A single incremental GC step can sometimes take surprising amount of time if the GC is in its atomic phase and you have lots of userdata objects or weak tables. Lua 5.2 did at least fix the having to scan all userdata objects for finalizers.

[Thomas Fransham @fsfodx]

@eevee I don't think any code will be JIT'ed while a Lua debug hook active that debuggers normally set

[Thomas Fransham @fsfodx]

@molecularmusing @pervognsen @terrym4h might know, he mentioned here https://t.co/9nUTggOUwq about adding some similar inlining attributes 2 years ago

[Thomas Fransham @fsfodx]

@dotstdy @damageboy It looks like someone did do the work to make if(__builtin_unpredictable(a < b)) turn in to a cmov https://t.co/bWSXLaosdX but it never got merged.

[Thomas Fransham @fsfodx]

@ssylvan @VS_Debugger There is already an extension to plot values from an array in the debugger at least https://t.co/ALntw8UO6u

[Thomas Fransham @fsfodx]

@molecularmusing I have no idea VS 2022 always starts it for me it even with a solution that is just c++ projects.
There is handy window in Process Monitor(ctrl-t) that show the process tree of both live and exited process that could show the parent process of the web views if this happens again

[Thomas Fransham @fsfodx]

@dotstdy I find explorer sometimes just ends up spending a huge amount of time at the end of deleting a large amount of files just updating the CustomDestinations\recent file lists

[Thomas Fransham @fsfodx]

@PierreTerdiman @pervognsen There is API to query if it will use the slow kernel call path https://t.co/1gPvXsFHUo, QPC is also modified by the Hyper V TSC page https://t.co/H9EQwnmfJW https://t.co/oiYvRE76gc