Notices by pinskia (pinskia@hachyderm.io), page 3

Today has been a wash. I only finally getting through all of the patches and bug reports. Still have not finished replying to the review of https://gcc.gnu.org/pipermail/gcc-patches/2024-November/669065.html . But at least it was approved with some changes; still need to go and do those changes.
On the glibc side, just waiting on getting approvals to be able to push things myself. But at least those patch was approved.

@thomasjwebb I still don't know which one I should use really. Since I consider myself non-gender. I present partly fem partly masc.
I feel uncomfortable using the men's bathroom some of the time. Using the women's bathroom is even worse.

@Marty @riotnrrd BY the way LLM and LLVM are different things too :). LLM is large language model that is the AI/ML model that will be used.

Ading a new internal checker in GCC. Oh you would think it would just work. WRONG, Things are not as well defined as you would think. Punt until GCC 16 because well stage3. At least I filed bug to track this progress.

@mutinyc Just thinking how fascists/racists went from "ragtime is just noise", "jazz is just noise" to "rock is just noise" to "rap is just noise" and even more recently
"electronic/house/dubstep is just noise".

@left_adjoint Depends on the level you want to do.
For compiler level; I found the ISA documents at https://developer.arm.com/documentation/ddi0602/2023-12/Base-Instructions?lang=en help a lot (and the software performance guides; I don't have a link right now).
For kernel level the arm ARM (for armv8.x-a) I found useful to understand how things work. But almost of this is very low level and almost never useful for a beginging to begin with.

From one bug report becomes 3. This is what I get from looking into fixing something.

@sidereal @hipsterelectron @ciggysmokebringer Since free source success, capitalists have turned towards the subscription model for software sales. Basically buying software is not useful alone, you need to buy the support instead. Now IBM was the fist to push this model in the 70s way before anyone else learned about it. Though they did with hw support. But hw support is no longer being handled that way because most hw doesn't need that kind of setup any more.
Cloud is a mainframe...

@sidereal @hipsterelectron @ciggysmokebringer " I appreciate linux but damn, it seems they have created a lot of value for the capitalists while not getting much leverage back."
So I am going to have to disagree. Only because I remember when every hw vendor had their own Unix system. They did that explictly for capitalism reasons. Some still have it (IBM and Oracle), though both also now support Linux ecosystem.

@FinalOverdrive And don't get me started on why autoconf is around in the first place. Most of the pro-cmake/ninja folks don't know of the time when you had to also deal with all of those OS's and executable formats.
a.out, a.outb, coff, xcoff, elf, etc.

@FinalOverdrive We don't do enough story telling. So many of youth don't realize the time when every major hardware vendor (and even some software vendor) had their own closed source Unix OS; AIX (IBM), HPUX (HP), SCO, IRIX (SGI), Solaris (Sun), Tru64 (DEC), etc. (not to mention embedded OS or phone OS)

@camelcdr Though looking at it again ARMv9-a's SVE should be able to optimize it. in a reasonible fashion I think. But neither GCC nor LLVM is able to handle it with SVE either.

@diedofheartbreak "the energy consumption is temporary" is same as oh that mine over there is just temporary and will not cause any damage. It is the exactly the same as capitalists talking about long term damage; even though the mine (polution) might be short term, it has unkown long term ecological effects. Plus renewables still has an ecological effect that most folks still don't take into account . E.g. mining rare earth, daming up water.

@camelcdr With a patch I am working on, aarch64 can vectorize this but only with -fno-vect-cost-model. The code generated is bad. Looks like GCC does not realize it could unroll the loop 4x to get a reasonible code generation (or with my patch just 2x).

@fay59 that is why underlying types were added ...
because this is an implementation defined behavior.

GCC documents it here:
https://gcc.gnu.org/onlinedocs/gcc-14.2.0/gcc/Structures-unions-enumerations-and-bit-fields-implementation.html

I can't find where clang documents their implementation defined behaviors. (I didn't look for MSVC documentation).

@lenary I am hoping next year I will be in San Diego to meet folks in person. Hopefully it will be after you have moved too :).

@velzie I looked into the git history and it was included in the initial checkin (into cvs which then was converted into git) back in 1993. So it was added a long time ago and the history on it looks to be lost. The 42 that is used by the xor is definitely a reference to `hitchhiker's guide to the galaxy`.

blah, I forgot some of GCC's IR is not valid for pointer types :) e.g. BIT_IOR and BIT_AND are not valid for pointer types.
https://gcc.gnu.org/PR117646 .
Lucky the fix is simple; just leaving it broken until tomorrow.

@paul @adam_chal well there are different implementations of tan too. Some have a slow path included and some don't. Even different versions of glibc are different. see https://sourceware.org/bugzilla/show_bug.cgi?id=15267 for talking about the slow path.
I am not 100% sure FMA is the only difference here though. But it does change the ULP so does the slow path in some cases.

@paul @adam_chal I suspect the difference is because on x86_64 the default libc does not use FMA (since it is not part of x86_64v1) while on aarch64 (since FMA is part of armv8-a) will use FMA.
Especially when it is 2up difference.
Fused Multiple Add can make a huge difference since the addition is done in infinite precision after the multiply and only rounded afterwards.