| --- Log opened Wed May 06 00:00:37 2015 | ||
| olofk | stekern: Nvidia? That sounds really cool | 05:02 |
|---|---|---|
| bandvig | olofk: What about Nvidia? (it looks like not all night post were logged) | 08:40 |
| olofk | bandvig: It turns out that all Nvidia GPUs are just OpenRISC CPUs | 08:41 |
| bandvig | olofk: :) Where did you read that? | 08:42 |
| olofk | When I grow up I'm going to stop lying :) | 08:44 |
| -!- Netsplit *.net <-> *.split quits: arokux | 08:58 | |
| -!- Netsplit over, joins: arokux | 09:00 | |
| GeneralStupid | okay, 39% of my FPGA is used by my core:) | 09:33 |
| GeneralStupid | Is there anything i can do to reduce the used area? For example, did he already uses the FPGAs built in DSP's? | 09:34 |
| olofk | GeneralStupid: The .map.rpt file (and maybe others) should give some clues about what resources have been used | 09:55 |
| olofk | You can also fire up the quartus gui and take a look at the netlist | 09:55 |
| GeneralStupid | but at the moment i have nothing, just 4 GPIO units and the "normal" mor1k | 09:56 |
| GeneralStupid | olofk: i would like to tune fusesoc a bit. I think it will be good of you can add "module" (and connect them to wb) Configurations in already working projects and tell fusesoc to regenerate the wishbone related files... | 10:05 |
| olofk | GeneralStupid: Can you elaborate a bit? | 10:05 |
| GeneralStupid | olofk: i think of this workflow: checkout with fusesoc, add a user defined core with extensions to the wb_intercon.conf, call fusesoc to create that wb files again. | 10:07 |
| olofk | GeneralStupid: The part of regenerating the wb_intercon files automatically is on my todo list. I have written up a proposal for a generic way of letting cores specify that they have files that needs to be regenerated | 10:11 |
| GeneralStupid | ahh okay sounds nice :) | 10:12 |
| olofk | Do I understand you correctly that you also want to have the additions to wb_intercon.conf inside the slave cores themselves (such as in gpio, or uart)? | 10:12 |
| olofk | Alright! Looks like my u-boot image loader works now | 10:12 |
| olofk | Bye bye bin2binsizeword! | 10:12 |
| GeneralStupid | olofk: i dont know how exactly i would concatenate it exactly but thats what i wanted to do | 10:13 |
| GeneralStupid | i will thx | 10:13 |
| olofk | I don't think you should put that information inside the slave cores. The top-level core should be the only one to decide on the register map | 10:14 |
| GeneralStupid | is there an easy way to switch between the mor1k implementations? | 10:44 |
| GeneralStupid | (coffeeflavours:))? | 10:45 |
| olofk | I think you just need to set the parameter OPTION_CPU0 to something else. Not sure how well maintained the other pipelines are though | 10:45 |
| olofk | If you are looking to save resources I suggest you decrease the cache size and turn off MMUs if you don't plan to run Linux | 10:46 |
| GeneralStupid | but i have to change the verlilog code for that? | 10:48 |
| GeneralStupid | ok ok i see it looks well documentated into the sourcecode | 10:48 |
| olofk | Just your top-level file | 10:48 |
| GeneralStupid | IMMU? | 10:50 |
| GeneralStupid | frome enabled to None | 10:50 |
| olofk | Yep. That will bring down the resources a bit. You can disable DMMU too | 10:51 |
| GeneralStupid | ICACHE - Instruction cache? | 10:51 |
| GeneralStupid | DCACHE data cache? | 10:51 |
| GeneralStupid | ok now iam down from 35% (just 2 GPIO modules) to 31% | 10:55 |
| olofk | The cache sizes are controlled by *_BLOCK_WIDTH, SET_WIDTH and WAYS | 10:57 |
| GeneralStupid | what is a good size? | 10:59 |
| olofk | Not sure | 11:00 |
| GeneralStupid | (i dont understand the WAYS) | 11:00 |
| olofk | http://en.wikipedia.org/wiki/CPU_cache#Associativity | 11:01 |
| GeneralStupid | okay | 11:01 |
| GeneralStupid | thx, i think i try it this way first :) iam on 30% | 11:03 |
| olofk | Not sure why you need to optimize it further. You still got 70% to play with :) | 11:06 |
| GeneralStupid | iam also not 100% sure what i want to do with it right now. I had an idea but it looks like its too much work to realize it. | 11:08 |
| olofk | Oh god, how I hate branch delay slots | 11:10 |
| olofk | I wonder if anyone has analyzed the cycles saved on branch delay slots vs the extra debugging time caused by them | 11:11 |
| GeneralStupid | :-D | 11:12 |
| GeneralStupid | i already changed my ram to 8mb in my template, but i cant fetch it because of or1200 svn and the jtag bug i had in de1 too | 14:40 |
| stekern | olofk: not as cool as openrisc ;) | 14:51 |
| stekern | but yeah, it's pretty exciting | 14:52 |
| GeneralStupid | stekern: where can i change the RAM amount in orf1k ? in verilog :) | 15:25 |
| stekern | GeneralStupid: what RAM? | 15:43 |
| GeneralStupid | the sdram | 15:50 |
| GeneralStupid | it was 32mb and i changed it but i did not change it into my build | 15:50 |
| stekern | iirc, it doesn't matter as long as you don't access it | 15:57 |
| GeneralStupid | can i control this if i use malloc? | 15:57 |
| stekern | most likely | 16:03 |
| olofk | GeneralStupid: What jtag bug? | 18:20 |
| olofk | and or1200 shouldn't be fetched from svn anymore | 18:21 |
| olofk | Have you update orpsoc-cores? | 18:21 |
| olofk | 'fusesoc update' should take care of that for you | 18:21 |
| GeneralStupid | fusesoc update does nothing | 18:22 |
| olofk | GeneralStupid: Do you run it from your workspace directory? | 18:36 |
| -!- Netsplit *.net <-> *.split quits: ed-jones, tfh, arokux, poke53281 | 18:48 | |
| -!- Netsplit over, joins: ed-jones, poke53281, tfh | 18:49 | |
| -!- Netsplit over, joins: arokux | 18:49 | |
| -!- Netsplit *.net <-> *.split quits: ed-jones, tfh, poke53281 | 19:18 | |
| -!- Netsplit over, joins: ed-jones, poke53281, tfh | 19:19 | |
| GeneralStupid | olofk: yes | 19:55 |
| olofk | Anyone good with C preprocessing macros? | 20:44 |
| GeneralStupid | i could give it a try | 20:48 |
| olofk | I just want to byte-reverse a number | 20:49 |
| olofk | It's something weird going on here | 20:51 |
| olofk | I think I found a way around it now | 20:52 |
| olofk | ahhh | 20:53 |
| olofk | I probably needed some more parenthesis | 20:53 |
| olofk | yep. That was it | 21:00 |
| olofk | Alright. The uImage loader is pushed to the de0_nano system in orpsoc-cores | 21:13 |
| olofk | wb_ram still doesn't work with quartus though. Need to push the fix for that as well | 21:17 |
| olofk | some other day | 21:17 |
| GeneralStupid | its really hard to get this running and generic | 21:56 |
| --- Log closed Thu May 07 00:00:39 2015 | ||
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!