( ESNUG 454 Item 8 ) -------------------------------------------- [04/28/06]
Subject: ( ESNUG 451 #9 ) Magma follow-up to Blast Plan Pro user critique
> So in general, I'd say we were pretty happy. It could have been better.
> But I felt it was a step forward for us. I think over time, we'll see
> things improve, as more of our engineers become familiar with the Magma
> tools and our internal code base improves. Unlike AVNT/SNPS, Magma does
> seem to have some people actively working on this tool, and we are seeing
> progress. Of course, we'll always say its too slow. :)
>
> - Jeff Echtenkamp
> Broadcom Corporation Irvine, CA
From: Gnana Kanagaratnam <gnana=user domain=magma-da spot gone>
Hi John,
I'm the Magma Blast Plan Pro product engineer who worked with Jeff on this
tapeout and would like to comment on his Blast Plan Pro post.
First off I want to thank Jeff and his team for all the help and advice
they gave us last year. Their feedback has been used to improve the latest
Blast Plan Pro 5.0 that shipped last year. I'm sure Jeff will be pleased
with our progress.
Before I get into specific issues I should note that Jeff and his team do
not build your average hierarchical designs. Their chip was a monster. It
had about 6 million instances which are divided into about 10-15 top level
blocks implemented in Blast Fusion. Chip assembly is by abutment, with
typically just a small amount of distance between top level blocks -- as a
result, the ports need to be perfectly aligned and placement needed to take
into account congestion through the hierarchy so that all the blocks are
routable.
> The biggest thing we saw, was to get a good pin assignment using a virtual
> flat placement, it means first off, your placement has to be right. Then
> your routing has to be right. Then your pin assignment has to be right.
> We found that roughly 80% of the work to get a good pin placement was to
> get a good virtual flat placement, where the cells connecting to your top
> level net were near where the pin would eventually be. If you didn't have
> this, you were screwed.
Jeff's design pushes the limits of IP re-use. His design has about 200 hard
macros plus the 6 million instances that were placed from the top level
context (what Jeff calls virtual flat placement). There are times when the
pin alignment will not be 100% orthogonal to the block edge. In nearly
every case we've examined, this has been due to excessive congestion caused
by either macros or block topology causing funneling of nets into strange
corners.
We've made a number of improvements in the latest version of Blast Plan Pro.
The new flow is pretty simple, all the user needs to do is define the
non-default spacing rules for the top level nets then run global placement
to do a quick flat placement of all the logic and macros. The global router
now respects the non-default rules that were set earlier. Since we need to
make sure that the pins are 100% horizontally or vertically aligned we flag
the 'channel' areas in such a way that the global router can't jog
left/right or up/down - thus forcing proper alignment.
Getting the pins in the right place and 100% aligned is like the chicken
and egg debate.
I think Jeff said it best in one of his previous posts:
"Floorplanning is highly subjective. Ask 10 engineers from 10 companies
what the most important thing in a floorplanner is, and you'll get 10
different answers."
Jeff's right, the chances that the out-of-the-box functionality will address
your specific needs are about zero, especially if you're building complex
chips with unique requirements.
> In general, their TCL interface allowed us to do a lot of things in an
> efficient and repeatable manner which we couldn't have done in Milkyway.
Using Magma Tcl the user can dive into any level of detail needed to get the
job done and done on time. No more Perl scripts to parse DSPF, DEF, LEF,
Verilog -- you just do the whole shooting match right in Blast Fusion - the
same Tcl is used to drive floorplanning, right down to block implementation.
We'll keep working on improving Blast Plan Pro, but chances are we'll never
be able to stay ahead of all our clever users. So I expect that Jeff and
his team will still find some reason to dig into Magma Tcl to finish off
their next super-complex creation.
- Gnana Kanagaratnam
Magma DA Santa Clara, CA
Index
Next->Item
|
|
