log☇︎
25 entries in 0.702s
asciilifeform: i really gotta genesis my rdtsc asmism thingie for gnat
asciilifeform: mircea_popescu: not that i see it as particularly likely, but afaik cannot be guaranteed not to, at least not w/out using inline asm, e.g. RDTSC on x86
asciilifeform: zx2c4: to observe it, you will have to hand-emplace rdtsc around it , and run on properly doctored inputs
asciilifeform: but yes this calls for a proper test, with rdtsc, then buncha MUL, then rdtsc again
asciilifeform: this prolly calls for a proper test with rdtsc. but i'm saving this for 'dessert'
asciilifeform: just rdtsc and add the result to [somewhere] every other instruction.
a111: Logged on 2017-03-18 11:03 Framedragger: http://btcbase.org/log/2017-03-10#1624858 << re rdtsc etc, i'm curious, does `grep -E '^model name|^cpu MHz' /proc/cpuinfo` consistently report to you mhz which are (+/-1 a mhz or so) equal to model frequency? (i know you can turn off linux powersaving etc features)
Framedragger: (also due to modern out-of-order execution in cpu pipelines, may need to force all commands before the rdtsc call to be 'flushed', i.e. to have completed before the counter instruction is called; just noting for self)
Framedragger: http://btcbase.org/log/2017-03-10#1624858 << re rdtsc etc, i'm curious, does `grep -E '^model name|^cpu MHz' /proc/cpuinfo` consistently report to you mhz which are (+/-1 a mhz or so) equal to model frequency? (i know you can turn off linux powersaving etc features) ☝︎☟︎
Framedragger: http://stackoverflow.com/questions/19941588/wrong-clock-cycle-measurements-with-rdtsc don't kill me plz
Framedragger: asciilifeform: something about needing to discard 'invalid values' with rdtsc (even when set up correctly - apparently best to 'pin' particular cpu, etc etc..) oh god.
asciilifeform: (cpu will stop to do ??????, maybe pack key logger buffer for later monthly upload to obummer, and undetectably, but rdtsc counter still ticks, it is lowtech)
mircea_popescu: yeah. ok, ok. rdtsc.
mircea_popescu: still, on contemplation, hard to argue against rdtsc
asciilifeform: rdtsc is the ticket, it is how profilers time, and the only timer that is really worth twopence on x86.
Framedragger: i actually did check what was available before deciding on clock measurement yknow. but yah, need to try rdtsc
asciilifeform: if your cpu dun have rdtsc, throw it out.
a111: Logged on 2016-09-14 12:01 Framedragger: "rdtsc is not guaranteed to be available on every CPU, or to run at a constant rate, *or be consistent between different cores.*" (emphasis mine). `get_cycles()` is recommended, but from cursory look it seems that on some architectures it uses rdtsc internally? madness.
a111: 3 results for "rdtsc", http://btcbase.org/log-search?q=rdtsc
Framedragger: !#s rdtsc
Framedragger: !# rdtsc
asciilifeform: Framedragger: try rdtsc
a111: Logged on 2016-09-14 12:01 Framedragger: "rdtsc is not guaranteed to be available on every CPU, or to run at a constant rate, *or be consistent between different cores.*" (emphasis mine). `get_cycles()` is recommended, but from cursory look it seems that on some architectures it uses rdtsc internally? madness.
Framedragger: "rdtsc is not guaranteed to be available on every CPU, or to run at a constant rate, *or be consistent between different cores.*" (emphasis mine). `get_cycles()` is recommended, but from cursory look it seems that on some architectures it uses rdtsc internally? madness. ☟︎☟︎
Framedragger: (linux appears to require `rdtsc` to be able to return something akin to tick count. asciilifeform maybe knows if this is a doomed affair.)