jrmuizel has quit [Remote host closed the connection]
ninolein has quit [Ping timeout: 264 seconds]
ninolein_ has joined #lima
jrmuizel has joined #lima
jrmuizel has quit [Remote host closed the connection]
yuq825 has joined #lima
libv_ has joined #lima
libv has quit [Ping timeout: 245 seconds]
libv has joined #lima
libv_ has quit [Ping timeout: 248 seconds]
libv has quit [Ping timeout: 264 seconds]
libv has joined #lima
libv_ has joined #lima
libv has quit [Ping timeout: 264 seconds]
libv_ has quit [Ping timeout: 264 seconds]
libv has joined #lima
jrmuizel has joined #lima
yuq8251 has joined #lima
jrmuizel has quit [Remote host closed the connection]
yuq825 has quit [Ping timeout: 272 seconds]
libv_ has joined #lima
libv has quit [Ping timeout: 264 seconds]
_whitelogger has joined #lima
dddddd has quit [Remote host closed the connection]
libv has joined #lima
libv_ has quit [Ping timeout: 258 seconds]
griffinp has quit [Ping timeout: 248 seconds]
griffinp has joined #lima
griffinp has quit [Ping timeout: 245 seconds]
bhoj- has quit [Ping timeout: 272 seconds]
Net147 has quit [Read error: Connection reset by peer]
Net147 has joined #lima
Net147 has quit [Read error: Connection reset by peer]
enty has joined #lima
ente has quit [Read error: Connection reset by peer]
Net147 has joined #lima
mardination has joined #lima
<mardination>
and i also say, you gonna have to manage your code your own, and make the code your own, if i will do it the world is upside down and rebellion and terror attacks take off.
mardination has quit [Quit: Leaving]
guillaume_g has quit [Remote host closed the connection]
guillaume_g has joined #lima
nighty has quit [Quit: Disappeared in a puff of smoke]
nighty has joined #lima
enty has quit [Quit: Reconnecting]
ente has joined #lima
ente has joined #lima
ente has quit [Changing host]
ente has quit [Quit: Reconnecting]
ente has joined #lima
ente has quit [Changing host]
ente has joined #lima
ente has quit [Read error: error:1408F10B:SSL routines:ssl3_get_record:wrong version number]
ente has joined #lima
ente has quit [Changing host]
ente has joined #lima
ente has quit [Client Quit]
ente has joined #lima
ente has quit [Client Quit]
ente has joined #lima
griffinp has joined #lima
bhoj has joined #lima
bhoj has quit [Client Quit]
griffinp has quit [Client Quit]
yuq8251 has quit [Quit: Leaving.]
libv_ has joined #lima
griffinp has joined #lima
libv has quit [Ping timeout: 245 seconds]
bhoj has joined #lima
yuq825 has joined #lima
mardination has joined #lima
<mardination>
anyway mali-t604 for instance is similar more to intel chips, where things work on quads of 2x2 not in quarter waves
<mardination>
in other words, threads or work-item number is based of register usage there even though i have heard it's constant on intel
<mardination>
"per core hence 256/4 on register usage 0" quads are running
<mardination>
when register usage is 64 registers
<mardination>
then they use only 8threads i.e 2quads
<mardination>
when 128 registers than 4 threads i.e two quads
<mardination>
so the queues internally are 256 entries in total for 4core mali t-640
<mardination>
it is solid, weak but it does the job brilliantly
<mardination>
err t604
mardination was banned on #lima by rellla [mardination!*@*]
mardination was kicked from #lima by rellla [mardination]
<hellsenberg>
rellla: i am afraid you'll have to deal with manual bans for now. Or maybe ask freenode staff
<hellsenberg>
I wish I knew of a better approach :-/
afaerber has quit [Quit: Leaving]
jrmuizel has joined #lima
<libv>
no, bans are not needed
<libv>
why they did not work, i do not know
<libv>
but bans are not necessary
<libv>
just ignore mardinator, there's no other way
<libv>
his mind is all over the place, and that's why he does this, if you ignore him, he will just blurt stuff out and then go away again
jrmuizel has quit [Remote host closed the connection]
<libv>
trying to use power is not only not effective, as he just gets another address, it is not balanced compared to the overhead, and it is just overhead, he causes
<libv>
plus, i know how power gets abused against people who do not hold the party line, cfr, the handful of people who made fd.o what it is today
<libv>
and yours truly
<libv>
that's two reasons why mardinator should just be ignore
<libv>
just use /ignore yourself
dddddd has joined #lima
<libv>
i have approached mardinator myself, and he now spends some time ranting at me
<libv>
instead of ranting at the channel
maciejjo has quit [Remote host closed the connection]
afaerber has joined #lima
malaga123 has joined #lima
<malaga123>
yeah i am sorry for wrong info, 16x16 is the max tile or quad group
<malaga123>
but that did not diserve a ban, i had no code for mali gpu as a cycle accurate simulator
<malaga123>
so handling that gpu is clear to me, very easy infact for any of the model, later look into powervr and adreno if there is enough info for them too, those all probably are similar
malaga123 has quit [Quit: Leaving]
jrmuizel has joined #lima
xdarklight has joined #lima
maciejjo has joined #lima
jrmuizel has quit [Quit: Leaving...]
jrmuizel has joined #lima
jrmuizel has quit [Remote host closed the connection]
jrmuizel has joined #lima
libv has quit [Read error: Connection reset by peer]
libv has joined #lima
sibelius has joined #lima
<sibelius>
but what are you talking about, this all over the place mind is the only rational thinker btw. in this crew. http://fileadmin.cs.lth.se/cs/Education/EDAN35/guestLectures/ARM-Mali.pdf it is written there, so still internally 32x32 , cause it has a dual scheduling like VLIW has, 64/2 is in fact 32
<sibelius>
Tile size 16x16 (4x4-32x32 internally)
<sibelius>
says in this pdf.
<sibelius>
evergreen hence is one to one match for the scheduler
<sibelius>
i told you long time ago, to change the column to a new column from any of the position, inside the row of any, you need to issue two instructions in a row, which by definition is in-order issue
<sibelius>
and the column is changed to the second of arbitrated wfid of course of them
sibelius has quit [Remote host closed the connection]
sibelius has joined #lima
sibelius has quit [Read error: Connection reset by peer]
sibelius has joined #lima
<sibelius>
this *is* the only rule involved, once you issue two in sequence the column is changed, once you do not even in full pipeline mode, it switches to greedy-then-oldest
<sibelius>
which i call a line or row based wfid rendering of the matrix
<sibelius>
rendering/computing
<sibelius>
basically it means the queues per core on MALI is 32x32 matrix sort of, the queue controller where you are in the matrix, is determined by, whether you schedule in-order or out-of-order, in-order definition is anything arbitrated from and issued two in sequence
<sibelius>
out-of-order is anything but this
<sibelius>
hence to make matters easier, the last item on every line should be some special toggler, first item per every line also, this toggles the diagonal to block or succeed
<sibelius>
it is called some sort of difficult name of factorization of matrix, i have that bookmarked also, it is well known trick, they did slightly differently
<sibelius>
Cholesky factorization of tasks in the matrix
<sibelius>
put the sucker in your mouths nastu conspirants, and fucking shut up for once or get a brain
<sibelius>
i have a solution for you if you ain't gonna learn, then you go to a shop by some rope hang it to the roof somehow, and hang yourself idiots
<sibelius>
that way we do not have to come after you
sibelius has quit [Remote host closed the connection]
sibelius has joined #lima
sibelius has quit [Client Quit]
<hellsenberg>
libv: very true, will take into account. thanks!
JeansF_ has joined #lima
<JeansF_>
what you have caused is a fucking war idiots!!!!! And you can not read with your thick occulars information that i give you, where any of the last detail is put on golden plates, but joller has thick occulars on his nose and a conspiring attitude to hire estonian agents against real man
<JeansF_>
you have daily basis stalked me with using military forces and interpol infra. You clowns do not seem to understand anything what you do, you do not even see
<JeansF_>
one way or another you scammers get your payday, that happens ONE FUCKING WAY OR ANOTHER, just like in the movie independence day where laws were broken and military equipment used for wrong purpose.
JeansF_ has quit [Quit: Leaving]
jbrown has quit [Ping timeout: 252 seconds]
JeansF_ has joined #lima
<JeansF_>
another google idiot hellsenberg who is that other moran with mental disability joller jonkerj
<JeansF_>
incredible another dutch handycap
<JeansF_>
bunch of quasimodos who want to have a programming image but instead are cockblockers right?
<JeansF_>
you ain't gonna acheive anything in chess when you do not straighten up, all you acheive is another life as a worm behaving like this.
JeansF_ has quit [Client Quit]
jbrown has joined #lima
drod has joined #lima
Wizzup has quit [Ping timeout: 245 seconds]
Wizzup has joined #lima
drod1 has joined #lima
drod has quit [Ping timeout: 272 seconds]
guillaume_g has quit [Quit: Konversation terminated!]
jrmuizel has quit [Remote host closed the connection]
jrmuizel has joined #lima
arti has quit [Ping timeout: 252 seconds]
arti_ has joined #lima
JeansF_ has joined #lima
<JeansF_>
well i am not a mathematician but , i round about can read it fine, it is a modified cholesky factorization or decomposition, where it is modified to to work on blocks/tiles of the submatrices of the same size, it also called algorithm-by-blocks, the most important table is in 5.2.2 called tasks
<JeansF_>
which demonstrates exactly what i have been saying, cause in contrast to you , I can read code too.
<JeansF_>
allthough as you notice rows and columns are the other way around there, but this is based of how someone presents the algo.
<JeansF_>
what you want to express as the tile in that algorithm is arbitary, it can be entry of the issue queue, it can be a separate GPU, a functional unit, or separate core/CU wtf. needed
<JeansF_>
it is just fixed to be 3x3 tiled matrix, with some singletriangular
<JeansF_>
matrix or whatever in this
jrmuizel has quit [Remote host closed the connection]
<JeansF_>
such method itself is no work at all to bring in, as i have been talking, not for GPU nor CPU to the kernel scheduling, only an instruction stream needs to be manipulated minimally with
<JeansF_>
on a cpu the driver is kernel, where you need to append a header to the scheduled blob only correctly, it is fewer than 1week job to get right for both
<JeansF_>
are you telling me, that you do not know this monkeys?
<JeansF_>
was it the clockblocking.edu or teddynecrophilia.edu where you got your education from fucking perverse bastards? maybe robert tomasulo should had been taught otherwise
<JeansF_>
it is 50years ago taken place this theory roughly or so
<JeansF_>
türa idiootide raisk!
JeansF_ has quit [Quit: Leaving]
afaerber has quit [Quit: Leaving]
afaerber has joined #lima
drod1 has quit [Quit: Ухожу я от вас (xchat 2.4.5 или старше)]