Kiran Bulusu’s Blog

OK as promised earlier, here are the part2 top level synthesis -cpu module I’m assuming that the block level synthesis using bottom-up flow is done using part-1 and all Glassboxes (Magma’s Terminology for abstracts which are BTW efficient that ILM’s etc and are very suitable for implementing multi-million instances designs) are generated.

This is continuation of Posts 1 and 2 which adress the block level synthesis and top level synthesis of the opensparct2 processor . This post concentrates on building the floorplan/SDC etc suitable for block level P&R implementation. Due to short of time and bandwidth, I couldnt get a chance to build the floorplan for full chip “cpu” module, but to keep the momentum going and for enthusiasts, I built the flow for the “rtx” module which is quite big ( around 7M+ instances ) below and can act as good testcase to build the reference flow. Anyone who has the full cpu fix-time-final volcano from your top level synthesis run, can load it in place rtx or alternatively import the full chip netlist and follow the rest of the flow.

Folks can re-use the P&R implementation script for block level  from my Nova Processor Core RTL-GDSII post .

We are using channel style and the script below basically does the following :

Who said Nokia N95 VOIP + Magic Jack doesnt work

I have to admit that it took me 2 days to figure this out, but honestly
Nokia SIP client works. I googled and most posts in many symbian/nokia forums
suggested that Magic Jack never works with Nokia . Some blamed that Nokia SIP stack
implementation is broken, some said MJ isnt friendly etc…

Its not true,Some VOIP (especially windows based softphones ) can figure out
the DNS-SRV records (yes you read it right. Its not DNS, but its DNS-SRV DNS+STUN server )
and “realm” strings for the client to work.

1. Avoid communication black outs. This is a must. Please always keep the customer in loop. Whether its pre-sales or post-sales. Your customer needs to be aware of the progress you are making and you have to get his acknowledgment. This is tricky at times when the product is not fully matured yet and we need to fine tune before it starts delivering results. Even then , you have to present this “initial” results and then keep showing progress. This is very important.

If the customer doesnt know what you are working on, how much efforts you have made , he might be compelled to look at another company. You have make that “human” connection and keep him in the loop.

2. Stick to the findings of discovery process: If you deviate from this, customer will not agree when you present the results. The intent of whole “discovery process” is defeated. If you have to change for some reason, you need to get clear acknowledgment from customer.

3. Pre-Sales is very different from Post-Sales . Pre-Sales is to prove that product works and solves customers issues identified in “discovery process”. Deployment is real stress test to the product and can open doors to other products or open doors to the same product with other groups or departments . So the same strategies which worked for pre-sales might not work in deployment.

For anyone interested in learning processor architectures and synthesizing processor cores, they can refer to www.opensparc.net . Sun Micro has donated its entire sparc processor core to Opensource.  They even provide scripts for Synthesis using Synopsys DC and for Verification as well.

Due to my previous back ground and interest in Processor Architectures, given my current CAD background, I was very curious to learn and synthesize these cores. Also, the synthesis scripts are for to use with TALUS platform from Magma. The scripts will do the bottom-up synthesis & top level synthesis, Scan Insertion, Timing Optimization and will write a sized netlist .

Please note that, these scripts are not meant to use for any commerical purposes and strictly for personal/learning purposes only and provided “as-is” . Neither the scripts I provided below nor the work is endorsed and has nothing to do with my employer . Additionally please make sure you read all the disclaimers/license agreements when you download the cores.

Due to short of bandwidth and resources in mind, I’m only publishing logic synthesis scripts. May be time permitting, I will publish the P&R scripts in Part 2 series.

OpenSparc-T2 has the following features and this runs at 1.4GHz (714ps clock period )
8 Sparc cores, 8 threads each
Shared 4MB L2, 8-banks, 16-way associative
Two 10/1 Gb Enet ports w/onboard packet classification and filtering
One PCI-E x8 1.0 port

OpenSparc T2 Layout . Courtesy opensparc.net

Main modules in the design are :
CCX – Crossbar
CCU – Clock control
DMU/PEU – PCI Express
EFU – Efuse (redundancy)
ESR – Ethernet SERDES
FSR – FBDIMM SERDES
L2B – L2 write-back buffers
L2D – L2 Data  L2D – L2 Data
L2T – L2 tags
MCU – Memory controller
MIO – Miscellaneous I/O
PSR – PCI-Express SERDES
RDP/TDS/RTX/MAC – Ethernet
SII/SIO – I/O datapath in/out to memory
SPC – Sparc core
TCU – Test control unit

PCI-E is not provided with the binary when you download and its a seperate IP.
cpu  is the top level module name .

The following source browser might help everyone when navigating through the hierarchy and for understanding RTL .