Mixing memory models CMM, LMM and possibly cog?

pmrobert · 2014-08-07 12:48

My application has one large supervisory loop running in the startup cog that is not as time sensitive as other cogs and runs properly and acceptably in CMM mode. I have other tasks in other cogs that require at least LMM speed. They also run properly in CMM mode but at an unacceptably degraded speed. Is it possible to mix these two models with the added caveat that it would be very nice if fcache would still be available for the LMM cogs? Is it possible to mix LMM, CMM and cog models? The time sensitive/intensive tasks run well when in fcache but are small enough to fit in cog mode. I suspect it's a matter of providing proper compilation and linking commands.

-Mike

jazzed · 2014-08-07 12:58

Hi,

If your functions are small enough, they can run entirely at fcache speed regardless of memory model.

For example the function below is used in the SimpleLibraries simpletext.h library. It mixes C and ASM, but your code does not need to do that.

The __attribute_((fcache)) is a GCC thing that tags the function to run in fcache if possible. CMM fcached code is limited to 64 longs.

__attribute__((fcache)) static void _outbyte(int bitcycles, int txmask, int value)
{
  int j = 10;
  int waitcycles;


  waitcycles = CNT + bitcycles;
  while(j-- > 0) {
    /* C code is too big and not fast enough for all memory models.
    // waitcycles = waitcnt2(waitcycles, bitcycles); */
    __asm__ volatile("waitcnt %[_waitcycles], %[_bitcycles]"
                     : [_waitcycles] "+r" (waitcycles)
                     : [_bitcycles] "r" (bitcycles));


    /* if (value & 1) OUTA |= txmask else OUTA &= ~txmask; value = value >> 1; */
    __asm__ volatile("shr %[_value],#1 wc \n\t"
                     "muxc outa, %[_mask]"
                     : [_value] "+r" (value)
                     : [_mask] "r" (txmask));
  }
}

pmrobert · 2014-08-07 13:01

Thanks, Steve. I have a function that is much too large to fit into fcache, works very nicely under LMM but is too slow under CMM. I should have been clearer on this - my apologies. Does anyone have any insight on this issue?

-Mike

David Betz · 2014-08-07 14:07

pmrobert wrote: »

My application has one large supervisory loop running in the startup cog that is not as time sensitive as other cogs and runs properly and acceptably in CMM mode. I have other tasks in other cogs that require at least LMM speed. They also run properly in CMM mode but at an unacceptably degraded speed. Is it possible to mix these two models with the added caveat that it would be very nice if fcache would still be available for the LMM cogs? Is it possible to mix LMM, CMM and cog models? The time sensitive/intensive tasks run well when in fcache but are small enough to fit in cog mode. I suspect it's a matter of providing proper compilation and linking commands.

-Mike

Unfortunately, it isn't possible to directly use multiple memory models in the same program. A program is linked with a particular version of the C runtime library that is compiled for that memory model. If you wanted to use more than one memory model in a single program you'd have to have copies of each function compiled with each memory model. You can, however, compile small programs using the COG memory model and link them with your LMM or CMM program to be launched in other COGs. You can find an example of this in the PropGCC repository under demos/toggle/cog_c_toggle.

pmrobert · 2014-08-07 14:58

David, thank you, it's appreciated. I was on the edge of using a dual Prop solution and now that's where I'm headed. One for high speed, high resolution functions and the other for CMM lower speed human interface stuff. It's all good!

-Mike

abecedarian · 2014-08-07 20:56

pmrobert wrote: »

David, thank you, it's appreciated. I was on the edge of using a dual Prop solution and now that's where I'm headed. One for high speed, high resolution functions and the other for CMM lower speed human interface stuff. It's all good!

-Mike

It all makes more sense now. But I'm still confused.

SRLM · 2014-08-08 06:13

abecedarian wrote: »

It all makes more sense now. But I'm still confused.

What you confused about?

abecedarian · 2014-08-11 13:51

SRLM wrote: »

What you confused about?

pmrobert floated an email to me about using two prop's on the EFI thing I've been planning with him. I wasn't sure of the intention of it at the time but now it makes a little more sense- but it's still a bit over my head.

pmrobert · 2014-08-11 16:13

Apologies! We are running out of cogs and memory if we wish to implement all the things we discussed. An additional Prop would add a total of 6 cogs, etc., to the available resources. One cog per chip would be needed to facilitate interchip comms. I'll shoot you some email after I get back home tomorrow. I'm currently in a 3rd floor condo across the street from NAS Key West (Boca Chica) watching F18s take off, land and maneuver about 300 feet overhead so a bit preoccupied right now.... :-)

abecedarian · 2014-08-11 23:34

I hope you have a plan, 'cause I know next to nothing, and even less about interfacing two Prop's together, much less synchronizing said Props, signals and such.
So, my already precarious position causing second-guessing and commitment issues is about to grow exponentially.

Mixing memory models CMM, LMM and possibly cog?

Comments