Shop OBEX P1 Docs P2 Docs Learn Events
BeMicroCV... ? What? — Parallax Forums

BeMicroCV... ? What?

rjo__rjo__ Posts: 2,114
edited 2014-09-23 23:09 in Propeller 1
First of all, I like the BeMicroCV(I got the i7 version... industrial C7) and I would buy it again... for the simple reason that I don't have to carry an extra power supply around with me.
It wouldn't be at the top of the list, but it would be on the list.

There are some issues... if this were a Parallax product, I would know for sure that I was doing something wrong. But since it isn't, I'm not so sure.

First of all, the compile time didn't drop when comparing Parallax's official sources with my 4CogP1v. On both the DE2-115 and Nano, the compile time dropped by roughly half.

So... for example... 4 minutes on the Nano becomes 20 minutes+ on the BeMicro.

Second... the upper end on the clock stays the same.

I compared all of the specs and it sure seems like I should be able to squeeze a few more clocks out of the BeMIcroCV than I can on the Nano,
but the BeMicroCV fails above 140MHz... just like the Nano.

I clicked every option that looked like it should help... the compile time went through the roof but the clocking didn't improve.

Is it possible that Altera's Cyclone V compiles worse and performs exactly the same as the Cyclone IV? It has to be me. I have to be missing something... right?

Rich

Comments

  • SeairthSeairth Posts: 2,474
    edited 2014-09-23 16:09
    You're not imagining things. Compiling for a Cyclone V takes longer than for a Cyclone IV. The big time difference is in the time the Fitter takes. I'm assuming this is because the Cyclone IV has relatively simple and fine-grained LEs, while the Cyclone V is composed of ALMs (which are comparatively more complex, more course-grained. and less abundant). In order to make efficient use of the ALMs, the fitter must figure out where it can pack multiple pieces of logic into partially-used ALMs. My guess is that a Cyclone V with significantly more ALMs could allow for a much faster fitting (at the cost of less efficient resource utilization).

    As for the clocks, that's still a mystery to me. I've been tweaking SDC files for the last day and I still don't understand much of it.
  • rjo__rjo__ Posts: 2,114
    edited 2014-09-23 16:11
    Thanks you very much. I get this very cold feeling inside when I think I am missing something... and I was feeling really cold about this:)
  • rjo__rjo__ Posts: 2,114
    edited 2014-09-23 16:13
    on the bright side... it does have about a million bits left over when I compile the 4Cog version...
  • jmgjmg Posts: 15,173
    edited 2014-09-23 16:18
    rjo__ wrote: »
    First of all, the compile time didn't drop when comparing Parallax's official sources with my 4CogP1v. On both the DE2-115 and Nano, the compile time dropped by roughly half.

    So... for example... 4 minutes on the Nano becomes 20 minutes+ on the BeMicro.

    Are you saying the 4Cog and 8Cog designs both take the same time to compile on BeMicro ?

    rjo__ wrote: »
    Second... the upper end on the clock stays the same.

    I compared all of the specs and it sure seems like I should be able to squeeze a few more clocks out of the BeMIcroCV than I can on the Nano,
    but the BeMicroCV fails above 140MHz... just like the Nano.

    As #2 says, the change in structure of the CV will affect the fitter times, but one would expect a 4 Cog design to have spare room and less logic, so the fitter has less work to do.
    You could try lowering the target fit speeds, as most fitters will try harder closer to the max.
    Chip mentioned something about multi-cycle paths not being fully supported yet IIRC ?
  • rjo__rjo__ Posts: 2,114
    edited 2014-09-23 16:51
    Are you saying the 4Cog and 8Cog designs both take the same time to compile on BeMicro ?

    Same ball park.

    What you and Seairth say about the fitter is true... but I also noticed longer times in every phase, the timequest module was particularly noticeable.

    To tell you the truth, I am not bothered by the compile times. If that is what they are... that is what they are. It is the fact that the clocking didn't change (when the specs nearly all seem better for the BeMicroCV) that has me perplexed.

    My understanding from here on the forum, was that we are limited on the cyclone IV chips by the PLL maximums... they can't do 300MHz. the cyclone v's should be able to do a 300 MHz PLL... but that is just my understanding.
    So if the PLL was the issue... it should be solved by the cyclone V.

    We are using the default 50MHz clock... there is a 2x clock(100MHz)... and I wonder if we have to use that to get anywhere with this issue.
  • rjo__rjo__ Posts: 2,114
    edited 2014-09-23 16:53
    I have had the day to play... a rare occasion. But family matters call. I'll try to check in later.

    Thanks

    Rich
  • SeairthSeairth Posts: 2,474
    edited 2014-09-23 17:08
    Does anyone have a properly configured SDC file for the 8-cog build? I want to compare it to the one I put together.
  • AleAle Posts: 2,363
    edited 2014-09-23 23:09
    The CV compiler is slow :(, compared to the the lattice compiler where the code is in the same like 10 times slower... :(. Well, call it progress or something related :). Activating multicore compiling, with 2 cpus brings like 15 % time reduction, with 4 cpus like 40 % reduction, ymmv.
Sign In or Register to comment.