Current Server Time: 15:50 (Central Europe)

#apertus IRC Channel Logs

2019/08/08

Timezone: UTC


00:01
Fares
left the channel
00:13
Nira
changed nick to: Nira|away
01:50
Y_G
joined the channel
01:54
BAndiT1983|away
changed nick to: BAndiT1983
02:15
Fares
joined the channel
02:15
Fares
left the channel
02:31
BAndiT1983
changed nick to: BAndiT1983|away
03:42
Y_G
left the channel
03:48
Bertl
off to bed now ... have a good one everyone!
03:48
Bertl
changed nick to: Bertl_zZ
03:48
apurvanandan[m]
Good night :)
05:08
BAndiT1983|away
changed nick to: BAndiT1983
05:38
BAndiT1983
changed nick to: BAndiT1983|away
08:28
namibj
I'm sorry I didn't yet write the email as I was told.
12:29
danieel
left the channel
12:32
danieel
joined the channel
13:00
Bertl_zZ
changed nick to: Bertl
13:00
Bertl
morning folks!
13:06
se6ast1an
good day
13:07
se6ast1an
once you got your morning coffee/tea/mana I think apurvanandan[m] would appreciate some assistance with the code for the temperature stress test
13:07
se6ast1an
I have a thermal camera ready to take measurements
13:07
Bertl
excellent!
13:09
apurvanandan[m]
Good morning Bertl
14:00
Amman
joined the channel
14:04
Amman
Greetings
14:09
se6ast1an
hi there
14:10
Amman
Hey. How is it going?
14:22
Amman
left the channel
15:06
intrac
left the channel
15:07
intrac
joined the channel
15:09
Bertl
off for now .. bbs
15:09
Bertl
changed nick to: Bertl_oO
15:14
intrac
left the channel
15:15
intrac
joined the channel
15:20
BAndiT1983|away
changed nick to: BAndiT1983
15:59
Bertl_oO
changed nick to: Bertl
17:41
BAndiT1983
changed nick to: BAndiT1983|away
18:36
apurvanandan[m]
Bertl, Yes the issue was in FIFO logic only :) now the simpler ft601 controller is working, and I am myself amazed how much useless code has been removed. The svf size got halved too.
18:38
Bertl
great!
18:38
apurvanandan[m]
Mainly the buffering of data was being used due to wrong handling of clock edges
18:39
apurvanandan[m]
But unfortunately the previous issue is still there
18:39
Bertl
can you generate a simulation VCD of the signals invovled for current setup?
18:39
apurvanandan[m]
I lose values due to fifo getting full
18:40
apurvanandan[m]
Yes, I sending the VCD in a moment
18:42
Nira|away
changed nick to: Nira
18:46
BAndiT1983|away
changed nick to: BAndiT1983
19:04
apurvanandan[m]
https://drive.google.com/drive/folders/1Gw3Ei6eRlRSOe8W10eacfuRnK_bJQuQS?usp=sharing
19:04
apurvanandan[m]
The vcd file and wlf file
19:05
apurvanandan[m]
https://gist.github.com/apurvanandan1997/ed2d060b068373d4ffae777930ff69ce
19:05
apurvanandan[m]
The final working code
19:08
Bertl
why do we have the 60MHz clock now?
19:09
Bertl
and I do not see the fifo signals in the VCD
19:09
apurvanandan[m]
They are unnconnected, I didn't remove signals which are useful for future testing
19:09
apurvanandan[m]
From the testbench
19:10
apurvanandan[m]
FIFO out directly connected to FT601 data out using a wire
19:11
Bertl
how about fifo_rd_en and fifo_*emp?
19:12
Bertl
the FT_WR seems to happen two cycles after the FT_TXE, shouldn't that be one cycle?
19:13
apurvanandan[m]
That doesn't matter
19:13
apurvanandan[m]
https://ibb.co/r5DD2Zf
19:13
Bertl
yes, but you lose a cycle there
19:14
apurvanandan[m]
How, if we don't enable write txe will be forever 0
19:14
Bertl
now for the real world test setup, the FIFO is now 1024 words deep?
19:14
Bertl
and still you lose data?
19:14
apurvanandan[m]
Yes
19:15
Bertl
what size are the USB transfers?
19:15
apurvanandan[m]
I am at same state as two days earlier
19:16
Bertl
also, the data dump you showed me last time had errors on precisely the same position every time
19:16
apurvanandan[m]
When I stop data creation on fifo being full, I don't lose data but when I don't I lose 100 words every milion words
19:16
apurvanandan[m]
Yes that was a issue of clock edge and is solved
19:17
Bertl
how much did you lose before the fifo depth change?
19:17
apurvanandan[m]
Approximately samw
19:17
apurvanandan[m]
same*
19:19
apurvanandan[m]
Yes one thing I noticed now, when I stop data generation I don't lost any word, but when I stop on fifo full condition I lose approx thousand words every 200Gbits of dataa received
19:20
apurvanandan[m]
Let me rephrase
19:20
Bertl
so first, there is a big difference betwenn 100/1mio and 1000/200Gbit
19:21
apurvanandan[m]
In one case I stop data generation when FIFO gets almost full then no loss
19:21
Bertl
obvious
19:21
apurvanandan[m]
When I stop on FIFO full condition, I lose 100 words every 200Gbits
19:22
apurvanandan[m]
*1000
19:22
Bertl
this sounds like a bug somewhere
19:22
apurvanandan[m]
When I don't stop I lost 1000s words per million words
19:22
Bertl
i.e. fifo full should be sufficient to prevent any loss if you stop data generation
19:23
apurvanandan[m]
Regarding the two cycle delay after TXE
19:24
apurvanandan[m]
I have to say that fifo takes one cycle to give new data, but the FT601_wr needs to be high just along with new data
19:24
Bertl
so one cycle is enough, as I suspected
19:24
apurvanandan[m]
So a cycle where no change is visible in vcd
19:25
apurvanandan[m]
That is one second cycle you give data right?
19:25
Bertl
i.e. set fifo read in one cycle and the write in the next
19:25
Bertl
will come along with the data
19:25
Bertl
anyway, you have plenty of other problems, was just making a comment there
19:26
Bertl
now have you tested 100% throughput with the new setup?
19:26
apurvanandan[m]
https://ibb.co/r5DD2Zf
19:26
apurvanandan[m]
I meant the wr_n is perfectly aligned with new data
19:26
apurvanandan[m]
Yes tested ok
19:27
Bertl
so, once again, just to recap, 100% throughput, 3.2Gbit works without data loss
19:27
Bertl
no need to ever stop any data generation, etc
19:27
apurvanandan[m]
Let me check once more
19:27
Bertl
FIFO never gets full, etc
19:28
apurvanandan[m]
Just to be double sure
19:28
Bertl
yeah, make it triple :)
19:49
apurvanandan[m]
umm, there is a slight change in simulation waveform
19:51
apurvanandan[m]
FT601_wr_n is now one cycle before ie txe change on one clock, wr_n on next, data on after wr_n
19:53
apurvanandan[m]
This is due to the fact that once txe is low then the packet currently out of fifo is not transmitted
20:14
BAndiT1983
changed nick to: BAndiT1983|away
20:25
illwieckz
left the channel
20:38
illwieckz
joined the channel
21:05
illwieckz
left the channel
21:21
Bertl
apurvanandan[m]: any news on the double/triple check?
21:21
illwieckz
joined the channel
21:22
apurvanandan[m]
Not working :(
21:23
Bertl
which means?
21:23
apurvanandan[m]
FIFO full condition is always required
21:24
apurvanandan[m]
Otherwise lot of data gets miss
21:24
Bertl
so what was the 'no dataloss' case you reported?
21:24
apurvanandan[m]
I am checking for the bug
21:25
apurvanandan[m]
ie stop writing when FIFO gets full
21:26
apurvanandan[m]
Just give me an hour or so, to try out things
21:26
Bertl
okay
21:56
se6ast1an
off to bed
21:56
se6ast1an
nn
21:58
Bertl
sleep well
22:15
apurvanandan[m]
Bertl, Can I attach a counter that pause the transfer after every 256 words?
22:17
Bertl
I do not see why not?
22:17
apurvanandan[m]
Okay
22:20
Bertl
what's the idea behind this?
22:22
apurvanandan[m]
To eliminate all cases when fifo gets empty in between of transfer. So I start with around 300 words in fifo and send 256 words and then again wait for fifo to fill upto 300
22:23
Bertl
okay
22:23
apurvanandan[m]
Maybe I am wrong, but testing this won't hurt