| *** tpb has joined #symbiflow | 00:00 | |
| *** gsmecher has quit IRC | 00:21 | |
| *** gromero has quit IRC | 00:28 | |
| *** Degi_ has joined #symbiflow | 00:45 | |
| *** Degi has quit IRC | 00:49 | |
| *** Degi_ is now known as Degi | 00:49 | |
| *** FFY00_ has quit IRC | 01:01 | |
| *** cjearls has joined #symbiflow | 01:20 | |
| sf-slack | <cjearls> Is there a place to find yosys optimization strategies? I have a design that is taking up 160% of the logic slices in my orangecrab, and I was hoping to find some yosys settings to reduce the utilization enough that it'll fit | 01:24 |
|---|---|---|
| *** umarcor has joined #symbiflow | 01:37 | |
| *** curtosis[away] has joined #symbiflow | 02:22 | |
| *** curtosis[away] has quit IRC | 02:44 | |
| *** ASHR has joined #symbiflow | 03:32 | |
| *** ASHR has quit IRC | 04:31 | |
| *** ASHR has joined #symbiflow | 06:59 | |
| *** kgugala_ has joined #symbiflow | 08:35 | |
| *** kgugala_ has quit IRC | 08:35 | |
| *** kgugala_ has joined #symbiflow | 08:35 | |
| *** kgugala has quit IRC | 08:36 | |
| tnt | cjearls: That's not a optimization strategy you need, it's a miracle. | 08:45 |
| Lofty | cjearls: you're running with -abc9 right? | 09:37 |
| Lofty | Also consider -abc9 -dff | 09:37 |
| tnt | I don't have that much faith in abc9 to go from 160% LC usage to something placable/routable. | 09:57 |
| tnt | At that point you need a whole new architecture for whatever you're doing. | 09:57 |
| umarcor | agree. pnr needs to be done. even 80-90% may be difficult to fit. | 10:10 |
| Lofty | tnt: ABC9 and a more accurate delay target can do a *lot* to reduce area | 10:51 |
| *** maartenBE has quit IRC | 10:52 | |
| *** maartenBE has joined #symbiflow | 10:57 | |
| tnt | Huh, I have never seen any such drastic area reduction on any real world design by using abc9. At most I think I've seen 10% LUT reduction which doesn't even translate to 10% LC reduction since a bunch would have been packed with an adjacent FF. | 11:21 |
| *** bjorkintosh has quit IRC | 11:42 | |
| sf-slack | <vastarparaakshar> Kintex-7 support is on the planning? | 12:28 |
| Lofty | tnt: are you setting ABC9 delay targets or just passing "-abc9" and assuming the best? | 12:42 |
| tnt | Lofty: not sure what you mean ... I pretty much always run my designs however fast they'll go. | 12:54 |
| *** FFY00_ has joined #symbiflow | 13:01 | |
| Lofty | That might be why you're experiencing not much in the way of area reduction | 13:11 |
| Lofty | To people who have other constraints it can help a lot | 13:12 |
| tnt | How do you specify the target frequency ? synth_ecp5 doesn't have anything in help | 13:16 |
| Lofty | It's not a target frequency but target delay in picoseconds, which you pass to -D | 13:37 |
| Lofty | (synth_ecp5 -D) | 13:38 |
| Lofty | I normally turn target frequency to picoseconds then halve it to account for interconnect delay | 13:38 |
| *** infinite_recursi has joined #symbiflow | 14:49 | |
| *** infinite_recursi has quit IRC | 14:50 | |
| *** gromero has joined #symbiflow | 15:55 | |
| cjearls | Lofty: I'm getting an unknown option error when running "synth_ecp5 -nowidelut -abc9 -dff -D 80000 -json gsd_orangecrab.json -top gsd_orangecrab" for the -D option, and I can't find it in the synth_ecp5 documentation | 17:10 |
| Lofty | Ah, the joys of Yosys having different options for different flows | 17:11 |
| Lofty | *sigh* | 17:11 |
| Lofty | "scratchpad -set abc9.D 80000; synth_ecp5 -nowidelut -abc9 -dff -json gsd_orangecrab.json -top gsd_orangecrab" | 17:11 |
| cjearls | Alright, it's synthesizing. Fingers crossed | 17:14 |
| cjearls | Lofty: Thanks for your help, I appreciate it | 17:15 |
| Lofty | I don't know if it can give you a design that *fits*, but perhaps it'll save you a bit of work | 17:22 |
| cjearls | Lofty: I'm getting an error "ERROR: Assert `cell->parameters.empty()' failed in passes/techmap/abc9_ops.cc:460.", is that a bug? | 17:24 |
| Lofty | Aaah, this one again | 17:24 |
| Lofty | So it would seem, yes | 17:24 |
| cjearls | So that would be submitted as an issue to yosys? | 17:25 |
| Lofty | Somebody else was meant to have submitted it | 17:25 |
| cjearls | I don't see any issues related to the assert on that line or in that file on Github | 17:28 |
| Lofty | I know | 17:30 |
| Lofty | I'll go file it now | 17:31 |
| cjearls | Oh, ok. Do you need any information from me to file it? Sounds like it's a known bug | 17:31 |
| Lofty | cjearls: It got reported to me personally by another person, so I reduced a testcase from that | 17:32 |
| Lofty | Temporary workaround: remove `-dff` from `synth_ecp5` | 17:32 |
| cjearls | Alright, I'm rerunning | 17:34 |
| cjearls | No miracles today, we're down to 167% utilization | 17:45 |
| *** futarisIRCcloud has joined #symbiflow | 18:02 | |
| mithro | cjearls: It is probably good to check that things like DSPs / blockrams / lutrams are being inferred correctly | 18:55 |
| *** ASHR has quit IRC | 19:11 | |
| *** FFY00_ has quit IRC | 19:37 | |
| *** FFY00_ has joined #symbiflow | 19:38 | |
| *** rvalles has quit IRC | 19:51 | |
| *** rvalles has joined #symbiflow | 19:51 | |
| *** futarisIRCcloud has quit IRC | 20:11 | |
| cjearls | mithro: Ok, how would I go about doing that? | 20:40 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!