| --- Log opened Sun Aug 04 00:00:56 2013 | ||
| stekern | I'm looking for a (linux) benchmark that would stress the mmu, does anyone know of such? | 07:19 |
|---|---|---|
| stekern | we have pretty decent coremark score under linux now btw, 1.66/MHz | 07:21 |
| stekern | we get the same with sw tlb reload though, since the test doesn't cause any tlb misses | 07:22 |
| juliusb | oh, that's a bit unexpected! I would have thought it would cause them | 09:45 |
| juliusb | I have a nice neat TCM interface module now | 10:02 |
| juliusb | http://pastie.org/8205042 | 10:03 |
| juliusb | so it goes cpu->tcm_if->bus_if | 10:04 |
| stekern | and if it's not used, it just acts as a passthrough I assume? | 13:55 |
| stekern | and coremark works so, that it preruns the tests a couple of rounds to determine how many rounds it should run the actual test | 13:56 |
| stekern | so caches and tlbs are nicely preloaded | 13:57 |
| stekern | so, actually, the idea of trying to cheat by running it twice to have things loaded in cache wouldn't make a difference | 13:59 |
| juliusb | ah of course! | 14:13 |
| juliusb | (tcm passthrough) yes, a generate loop in there makes sure no logic is ineferred | 14:14 |
| juliusb | having trouble with a combinatorial loop though, according to verilator | 14:15 |
| juliusb | unoptflat thing, at least | 14:15 |
| juliusb | xst doesn't think it's a combinatorial loop (does it report them, though - has been a while since I've used it. Synplify certainly does!) | 14:15 |
| stekern | I think it does, but verilator seems more sensitive | 18:19 |
| stekern | I got around a couple by using assign a = b ? c : d; instead of always @* muxes | 18:21 |
| stekern | but IMO gerilator was just confused, marking those as comb loops | 18:22 |
| stekern | verilator even | 18:22 |
| stekern | hmm, interesting, linux 3.11 doesn't boot on my "default" atlys orpsoc (the one I had at orconf 2012) | 20:31 |
| stekern | with or1200 | 20:32 |
| stekern | it boots with mor1kx | 20:32 |
| stekern | a lot of things that might be off though, I have a lot of hacks ontop of jonas tree | 20:33 |
| stekern | and I haven't tried with a or1200 build of the same orpsoc I'm running now | 20:33 |
| stekern | but at least the hw tlb reload stuff boots on the atlys, that was I wanted to try ;) | 20:34 |
| stekern | seems to be my stuff, jonas tree boots on or1200 | 21:03 |
| stekern | interesting enough, *that* crashes mor1kx | 21:03 |
| --- Log closed Mon Aug 05 00:00:57 2013 | ||
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!