<cr1901_modern>
ZipCPU: Path bug == "yosys-config returns the wrong directory on Windows." But I need to test something before actually submitting a report
seldridge has joined #yosys
<cr1901_modern>
it's possible I forgot to run "make PREFIX=/path/to/root" while compiling
<ZipCPU>
I just ask because Clifford and I were chasing some path bugs today.
<ZipCPU>
I wasn't using Windows though
* cr1901_modern
goes ahead and checks
<ZipCPU>
We were looking into include paths
<ZipCPU>
I was surprised to discover that yosys allows code in multiple directories, and include paths between them
<cr1901_modern>
Why wouldn't it?
<ZipCPU>
Exactly ... I just hadn't used that feature before.
<ZipCPU>
Well, that and .... there were some bugs when using the feature with Verific ... initially, as I recall, there was no way to create new paths. Then once we fixed that, everytime you specified an include path it would get reset after a single read command.
<ZipCPU>
But ... that was just my morning's work ;)
<cr1901_modern>
include paths are broken on ISE too if memory serves. I.e. you can only add one
<tpb>
Title: initial port of SoC to the Lattice HX8K breakout board · maikmerten/spu32@5e3401f · GitHub (at github.com)
<maikmerten>
turns out it helps to actually double-check pin assignments and not misreading the somewhat wacky silk screen on that board
<maikmerten>
in the icestorm toolchain, is there something more than toying around with the seeds to tweak timings? My design somewhat fluctuates between 45 MHz and 56 MHz depending on minor tweaks (pin assignments, minor logic tweaks far away from the critical path) so I guess "placement and routing luck" has a role here
jwhitmore has joined #yosys
<ZipCPU>
maikmerten: Not sure what to say. The arachne-pnr placer isn't really timing driven, so ... it gets different timing's depending upon the random variables used to generate the placement.
<maikmerten>
well, I guess that answers the question quite fine, actually ;-)
<daveshah>
You can pick a seed using -s, or use -r to use a random seed each time
<daveshah>
If you want to experiment without even modifying the design at all
<maikmerten>
I think -r may be very useful if arachne-pnr and icetime are wrapped into a nice little script with a loop to retry if there's no timing closure
<daveshah>
Because arachne-pnr is only single-threaded, it would make sense also to run 4-8 times in parallel too
<maikmerten>
indeed
<daveshah>
I've seen a Makefile which did this once
<daveshah>
Haha, you could add two of the most researched and difficult parts of PnR - parallelisation and timing driven - to arachne in 10 lines of bash...
<maikmerten>
of course, that would be parallelisation with no net speed gain ;-)
<awygle>
but improved QoR which is still a win
<maikmerten>
sure :-)
X-Scale has quit [Ping timeout: 244 seconds]
X-Scale has joined #yosys
maikmerten has quit [Quit: Ex-Chat]
seldridge has quit [Ping timeout: 240 seconds]
dxld has quit [Quit: Bye]
dxld has joined #yosys
seldridge has joined #yosys
digshadow has quit [Quit: Leaving.]
digshadow has joined #yosys
tinyfpga has quit [Ping timeout: 256 seconds]
tinyfpga has joined #yosys
digshadow has quit [Quit: Leaving.]
digshadow has joined #yosys
knielsen has quit [Read error: Connection reset by peer]