Alexander Monakov

@ValdikSS It appears to be caused by uBlock's noscript shim when javascript is disabled for t.co: github.com/uBlockOrigin/u…

@ValdikSS Ah, sorry for a moment I forgot uBlock has a "disable on this page" button. Still reproducible with the disable enabled.

Sorcerer's Apprentice Syndrome is a network protocol flaw in TFTP. It was named after Goethe's 1797 poem "Der Zauberlehrling", because the details of its operation resemble the disaster that befalls the sorcerer's apprentice: an ever-growing replication of every packet in TX.

@ValdikSS Amazingly, clicking on this link in Tweetdeck opens Wikipedia page for 'Sorcerer' instead, presumably because somewhere in the convoluted json/js apparatus the URL is terminated on the apostrophe in "Sorcerer's".

@rrika9 Ah. Let me add that with blend+blackframe+metadata filter combo you can get ffmpeg to just print numbers of differing frames without transcoding: superuser.com/questions/1615…

@_monoid I was half thinking maybe I can diagnose this myself, but I guess there's too much about this only AMD knows. It's also not deterministic, so…

is there a way with vaapi to do hardware and software video decoding in parallel and compare the results? I keep getting artifacts on my *takes deep breath* Mesa Gallium driver 22.3.6 for AMD Radeon Graphics (rembrandt, LLVM 15.0.7, DRM 3.49, 6.2.1-arch1-1)

@rrika9 It's a nice food for thought :) Can you explain more why you want to do hw+sw decoding in parallel? I would use Ctrl-H in mpv to toggle hw decoding around the scene with artifacts.

@_monoid oh, thanks a lot! I didn't expect to get a reply honestly

@rrika9 Yeah. If this doesn't involve accelerated rendering, I would launch a nested server: Xephyr :1 -resizeable -no-host-grab and then 'set env DISPLAY :1' in gdb before 'run'.

this is apparently working as intended, you just need to make sure to never debug step between the display_grab and display_ungrab calls

this apparently freezes X11 until you kill gdb: $ gdb --args zenity --question (gdb) break gdk_x11_window_foreign_new_for_display (gdb) r

@Wunkolo @pshufb You probably do want a dependency-breaking instruction, but you cannot use the 512-bit vpcmp like that here: it produces the all-ones result in a k-reg (a vector mask register).

@pshufb Something tells me `pcmpeq{b,w,d} xmm0, xmm0` is a bit faster at making -1 since it's a special hardware idiom.

#AVX512 gem - how to get uint32/uint64=1? VPTERNLOGD $0xff, Z0, Z0, Z0 # set Z0 words to -1 VPABSD Z0, Z0 # this turns -1 into 1

@cr1901 @stephentyrone In the (non-default) rounding-to-zero mode, sure, but how do you know that 1.0 is never the nearest representable floating-point number to cos(x)?

@stephentyrone Sure, it makes sense that "you can't actually get sin/cos except cos(0) to ever equal 1 exactly using floating point". I just never took that train of thought to its logical conclusion :'D!

@David3141593 @ilyakurdyukov From the other subthread it seems the PC program is a filesystem and syslog server, and the libc on the phone is the (remote) filesystem and syslog client.

@ilyakurdyukov or is it more of a long-running daemon that's "driven" by the phone-end?

@ilyakurdyukov Hm, on second thought I might have mis-guessed and you implement software on both sides and so control the protocol yourself? Then you can go with 'ikdb'! (riffing off 'adb', Android Debug Bridge)

@ilyakurdyukov I guess the program is specific to a particular protocol or vendor, so I'd suggest to put one of those in the name first, like sprd-xfer for example.

@ilyakurdyukov I think the need to remove individual device match entries without blacklisting a module wholesale came up so rarely that it is not implemented. It would need to be an in-kernel facility to block binding of built-in modules, not in userspace tools like udev or modutils.

@ilyakurdyukov Yes, it's for unbinding a bound driver, not preventing binding in the first place, and it seems libusb can automate that for you — have you seen libusb_set_auto_detach_kernel_driver? github.com/libusb/libusb/…

@ilyakurdyukov You can unbind a specific device from a driver by writing the name of the device to the 'unbind' file of the driver in the sysfs (device names are visible as symlinks in the directory with the 'unbind' file).

@shafikyaghmour I also love they are bug-compatible with each other in allowing to skip initialization without resorting to extensions: godbolt.org/z/vcbc7K4n4

I really love that clang and gcc when using: -ftrivial-auto-var-init=zero Still issue a diagnostic when using an uninitialized variable with: -Wuninitialized godbolt: godbolt.org/z/bzT196h4M #cplusplus

@whisperity @JoshuaG19367959 @shafikyaghmour In case of GCC you'd lose that bet: the synthetic initialization is injected before the Wuninit pass runs, and yet it can be distinguished from an explicit initialization: godbolt.org/z/eKvPKGccE

@JoshuaG19367959 @shafikyaghmour I wouldn't be so sure that this was some thoughtfully engineered design. My bet is that due to the complexity of the compiler, simply the effects of the -f gets applied way after -Ws has been synthesised already.

@ioccc Append .rss to the account URL to get an RSS feed: @ioccc.rss" target="_blank" rel="nofollow noopener">fosstodon.org/@ioccc.rss

2/2 However we plan to post more detailed updates, especially when an contest is underway, directly to #mastodon: @ioccc@fosstodon.org Non-mastodon users can bookmark and follow: @ioccc" target="_blank" rel="nofollow noopener">fosstodon.org/@ioccc

@marc_b_reynolds @markpapadakis Maybe there should be a shorthand for '-fno-math-errno -fno-trapping-math -ffp-contract=off'. -ffine-math?

@markpapadakis --ffast-math: not even once. -fno-math-errno: every damn time. (unless you're using visual-c and then: may your gods have mercy on your eternal essence)

Wasted over an hour again thanks to -ffast-math. Don’t ever use -ffast-math (or -Ofast that implies —ffast-math, etc).