| *** tpb has joined #symbiflow | 00:00 | |
| *** Degi_ has joined #symbiflow | 00:05 | |
| *** borisnotes has quit IRC | 00:09 | |
| *** Degi has quit IRC | 00:09 | |
| *** Degi_ is now known as Degi | 00:09 | |
| *** gsmecher has quit IRC | 00:19 | |
| *** futarisIRCcloud has quit IRC | 01:12 | |
| *** Degi has quit IRC | 01:15 | |
| *** Degi has joined #symbiflow | 01:33 | |
| *** az0re has quit IRC | 01:38 | |
| *** citypw has joined #symbiflow | 02:05 | |
| *** zkms has quit IRC | 02:43 | |
| *** zkms has joined #symbiflow | 02:44 | |
| *** Bertl is now known as Bertl_zZ | 03:14 | |
| *** FFY00 has quit IRC | 03:17 | |
| *** FFY00 has joined #symbiflow | 03:19 | |
| *** FFY00 has quit IRC | 03:27 | |
| *** FFY00 has joined #symbiflow | 03:28 | |
| *** tcal has quit IRC | 03:36 | |
| *** tcal has joined #symbiflow | 03:48 | |
| *** futarisIRCcloud has joined #symbiflow | 04:28 | |
| *** _whitelogger has quit IRC | 05:57 | |
| *** _whitelogger has joined #symbiflow | 05:59 | |
| *** az0re has joined #symbiflow | 07:08 | |
| *** PiaMuehlen has joined #symbiflow | 07:15 | |
| *** PiaMuehlen has quit IRC | 07:18 | |
| *** robert2 has joined #symbiflow | 07:33 | |
| *** robert2 has left #symbiflow | 07:33 | |
| *** rw1nkler has joined #symbiflow | 07:34 | |
| -_whitenotifier-9- [symbiflow-arch-defs] rw1nkler opened issue #1442: OpenTitan support - https://git.io/JfLtA | 07:44 | |
| *** wavedrom has quit IRC | 08:19 | |
| -_whitenotifier-9- [edalize] acomodi opened issue #37: Rebase symbiflow branch on current upstream master - https://git.io/JfLmK | 08:29 | |
| *** OmniMancer has joined #symbiflow | 08:55 | |
| *** borisnotes has joined #symbiflow | 09:12 | |
| *** epony has quit IRC | 10:10 | |
| *** epony has joined #symbiflow | 10:18 | |
| *** Bertl_zZ is now known as Bertl | 10:49 | |
| -_whitenotifier-9- [symbiflow-arch-defs] kkumar23 opened issue #1443: Branch : Quicklogic : Install at custom location - https://git.io/JfLCK | 12:14 | |
| mithro | acomodi: When you get a moment, I would like to discuss stats and FPGA Tool Perf | 14:57 |
|---|---|---|
| sf-slack | <acomodi> @mithro: Sure | 14:59 |
| sf-slack | <acomodi> mithro: now is good for me | 14:59 |
| ZirconiumX | And I'm around if it matters any, mithro | 15:05 |
| mithro | acomodi: See the last section I added to https://docs.google.com/document/d/16L50pyS3RjYStvRKRoWyVac7NoZKyu45fpPQXJqeKYo/edit# | 15:08 |
| tpb | Title: SymbiFlow FPGA Tool Performance (Xilinx Performance) - Google Docs (at docs.google.com) | 15:08 |
| mithro | ZirconiumX: You do seem better at stats than myself, so would appreciate feedback too | 15:09 |
| ZirconiumX | It's less about that, more about applying the little bit of existing knowledge I have on these things :P | 15:09 |
| sf-slack | <acomodi> mithro: as far as I understand, we would need to add tests in tool-perf to iteratively check an algorithm (in this case the a PnR tool) to verify the assumptions | 15:13 |
| sf-slack | <acomodi> for assumption 1: run a design X times and verify that the X results are the same (or at least are between an interval of acceptance) | 15:14 |
| sf-slack | <acomodi> for assumption 2: run a design X times and for each run change the seed and verify that the output changes as well | 15:15 |
| sf-slack | <acomodi> I assume that the output in this case would be CPD, correct? | 15:16 |
| ZirconiumX | mithro: it's "MARLANN" as in "multiply-accumulate and rectified-linear accelerator for neural network" | 15:17 |
| *** borisnotes has quit IRC | 15:25 | |
| *** gsmecher has joined #symbiflow | 15:26 | |
| mithro | acomodi: So for assumption 1 -- we need to run the tool X times with identical input and check we get identical output | 15:35 |
| mithro | acomodi: For assumption 2 -- I think we need to run a whole bunch of runs until we have a high level of confidence | 15:36 |
| mithro | Test 3 is the big one | 15:36 |
| ZirconiumX | mithro: Can I suggest an alternative alternative test 3? | 15:37 |
| mithro | ZirconiumX: Sure | 15:38 |
| mithro | ZirconiumX: the discussion section is trying to explain my high level thinking around why test 3 is a reasonable thing to think about | 15:39 |
| ZirconiumX | Assumption 3: Place-and-route results have approximately normally distributed maximum frequencies. | 15:39 |
| ZirconiumX | (this holds true in my experience) | 15:39 |
| ZirconiumX | Alt. Alt. Test 3: Determine the mean of the normal distribution of maximum frequencies. This will tell you how likely it is that a tool will produce a design that satisfies the timing constraints. | 15:40 |
| ZirconiumX | My assumption here is that the seed is effectively a random input, and related seeds do not produce related results. | 15:42 |
| ZirconiumX | Which seems a much safer assumption than the thinking you outline | 15:42 |
| ZirconiumX | mithro: ^ | 15:42 |
| *** rw1nkler has quit IRC | 15:51 | |
| *** wavedrom has joined #symbiflow | 16:27 | |
| *** borisnotes has joined #symbiflow | 16:31 | |
| mithro | ZirconiumX: I believe you that there is a distribution of maximum frequencies - I'm not sure that I believe it is normally distributed | 16:33 |
| mithro | I wouldn't be surprised to see binomial type distributions | 16:34 |
| ZirconiumX | I think Fmax should be treated as continuous | 16:34 |
| ZirconiumX | Because there's no inherent yes/no here | 16:35 |
| ZirconiumX | So binomial is definitely the wrong one here | 16:35 |
| *** citypw has quit IRC | 16:37 | |
| mithro | ZirconiumX: I've definitely seen designs where the optimizer gets stuck with fmax distributed around two seperate fmax values | 16:39 |
| ZirconiumX | That would be bimodal, not binomial :P | 16:41 |
| ZirconiumX | https://docs.google.com/spreadsheets/d/1ENDF78QmjoYobu_oep0v74wwrDMk1FxxXDlbJEW_Flg/edit?usp=sharing | 16:45 |
| tpb | Title: ABC Fmax distributions - Google Sheets (at docs.google.com) | 16:45 |
| ZirconiumX | Here's some data I have to hand | 16:45 |
| ZirconiumX | You can definitely see normals in there | 16:45 |
| ZirconiumX | Perhaps it's more like a mixture model | 16:47 |
| mithro | ZirconiumX: yes, I have definitely seen normal distributions around an fmax | 16:47 |
| litghost | None of those look like Gaussian distributions? I'd expect a normality test to return poor results on all of them? | 16:47 |
| *** tcal_ has joined #symbiflow | 16:50 | |
| ZirconiumX | I'm not entirely sure how to do a normality test here | 16:52 |
| ZirconiumX | And sadly I can't overlay plots | 16:52 |
| *** tcal_ has quit IRC | 16:53 | |
| *** tcal_ has joined #symbiflow | 16:53 | |
| *** burgerchamp has joined #symbiflow | 17:17 | |
| hackerfoo | ZirconiumX: Colab (Jupyter + Python) is much better at generating complex plots: https://colab.research.google.com/drive/1fZ2j40GhPJHU35KogyPTcomyff2Fxmnl | 17:26 |
| ZirconiumX | Ah, I see | 17:27 |
| hackerfoo | Unrelated: I'm seeing about 800GB memory usage for 80 runs of VPR for the Titan tests, so plan on 10G per run for tests, CI, etc. | 17:27 |
| hackerfoo | I had to switch to m1-megamem-96 to make effective use of the cores. | 17:29 |
| litghost | hackerfoo: Most of that memory is just the gaussian blur test | 17:30 |
| litghost | hackerfoo: Are you running with titan nightly or weekly? | 17:30 |
| hackerfoo | It's not e.g. LU230 is at 17GB right now. | 17:31 |
| hackerfoo | I'm running nightly, but with various flag settings. | 17:32 |
| hackerfoo | There isn't just one offender, there are many designs that are well over 10GB. | 17:33 |
| hackerfoo | The ones that use the most memory also tend to run longer. | 17:33 |
| *** tcal_ has quit IRC | 17:35 | |
| hackerfoo | I think using megamem (there's only one) is a good idea, because ultramem is too much, and with highmem you can only safely use about half the cores. | 17:35 |
| hackerfoo | I'm still only running 80 threads per machine, but I'm going to do 95 (leaving one for other tasks) next time. | 17:38 |
| hackerfoo | For 950 threads total. | 17:38 |
| *** syed has joined #symbiflow | 17:41 | |
| *** syed has quit IRC | 17:54 | |
| *** OmniMancer has quit IRC | 18:13 | |
| *** adjtm_ has joined #symbiflow | 19:23 | |
| *** borisnotes has quit IRC | 19:24 | |
| *** borisnotes has joined #symbiflow | 19:25 | |
| *** QDX45 has quit IRC | 19:25 | |
| *** QDX45 has joined #symbiflow | 19:25 | |
| *** adjtm has quit IRC | 19:26 | |
| *** LisaMarie has joined #symbiflow | 19:35 | |
| *** LisaMarie has quit IRC | 19:38 | |
| *** adjtm_ has quit IRC | 19:45 | |
| *** adjtm has joined #symbiflow | 19:45 | |
| *** tcal_ has joined #symbiflow | 20:58 | |
| *** tcal_ has quit IRC | 21:08 | |
| *** tcal_ has joined #symbiflow | 21:08 | |
| *** tcal_ has quit IRC | 21:12 | |
| *** tcal_ has joined #symbiflow | 21:19 | |
| *** tcal_ has quit IRC | 21:37 | |
| *** AkiraMay has joined #symbiflow | 21:44 | |
| *** AkiraMay has quit IRC | 21:47 | |
| *** tcal_ has joined #symbiflow | 21:51 | |
| * hackerfoo uploaded an image: Screen Shot 2020-04-24 at 2.56.03 PM.png (83KB) < http://sandbox.hackerfoo.com:8008/_matrix/media/v1/download/sandbox.hackerfoo.com/RAlketALySXsyDObkjGmFWFE > | 21:57 | |
| hackerfoo | Nice level load at ~80% (that's ten machines.) The dip is from switching machine types. | 21:58 |
| *** tcal_ has quit IRC | 22:03 | |
| *** boris_ has joined #symbiflow | 22:55 | |
| *** borisnotes has quit IRC | 22:58 | |
| *** boris_ has quit IRC | 23:35 | |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!