[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference ricks::dechips

Title:Hudson VLSI
Notice:For Digital Chip Data - CHIPBZ::PRODUCTION$:[DS_INFO...]
Moderator:RICKS::PHIPPS
Created:Wed Feb 12 1986
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:701
Total number of notes:4658

632.0. "EV6 exception handling" by STKAI1::FEHRM (Bjorn Fehrm @ugo) Sat Jan 11 1997 06:05

T.RTitleUserPersonal
Name
DateLines
632.1ROCK::BANNONSat Jan 11 1997 19:2734
632.2OK, so what does that bring usSTKAI1::FEHRMBjorn Fehrm @ugoMon Jan 13 1997 03:2327
632.3No change requiredKAMPUS::NEIDECKEREUROMEDIA: Distributed Multimedia ArchivesMon Jan 13 1997 07:0710
632.4ROCK::BANNONMon Jan 13 1997 21:3133
632.5EV6 page size and virtual address range?BRSVMS::bro-ras-mod2.bro.dec.com::CornelisRoger - OMS Brussels - 856-7612Thu May 15 1997 11:0815
Hi,

I'm in the process of "upgrading" a presentation about EV4 and EV5 to EV6, for 
our upcoming DECUS symposium.  Most of the information I need, I already found 
in the �P Forum presentation, and the �P Report.  But there are two numbers I am 
unable to find, although IMHO there's no reason to keep them secret.

Page size:		Is the page size still 8 kB?
Virtual address range:	43 bits?

Can someone confirm/deny/correct these numbers?

Thanks,

Roger
632.6ROCK::BANNONThu May 15 1997 11:563
    
    Page size:	8Kb
    Virtual address range: 48 bits
632.748b capable, not requiredAD::MCLELLANThu May 15 1997 14:545
    The 48b virtual address mode is enabled via a mode bit and 
    requires supporting page tables/tb fill flows.  Early users
    will likely stay with the 43b mode.
    
    Ed 
632.8What is a CAM?BRSVMS::bro-ras-mod4.bro.dec.com::CornelisRoger - OMS Brussels - 856-7612Thu May 15 1997 19:1520
.7>     The 48b virtual address mode is enabled via a mode bit and 

Nice solution.

I have another question:
I don't really understand the slide notes on page 13 of A264UP1.PPT.

"
key fcn ofmapper is 80cams ... every cycle 8 virtual i comes out and do 8x80 
cams
virtual comes out of physicall ...
doing 640 cams per cycle ...
" etc.

It would definitely help if I would know what CAM stands for - but I sure 
wouldn't mind some help with the rest of the text/calculations on that page;-}

Thanks,

Roger
632.9DECCXL::OUELLETTEmudseason into blackfly seasonThu May 15 1997 19:3012
A CAM is a Contents Addressed Memory.
For a TLB you have a CAM and a RAM.
You present the CAM with some bits of the address.
If the mapping is present in the CAM, a hit line
fires.  That causes the right line of the RAM
to send you its contents.

The quote you have above hardly parses using an English grammar.
640 cams per cycle... doesn't make much sense.  Maybe something
else is meant.

R.
632.10register mappingAD::MCLELLANFri May 16 1997 09:5619
    I don't have the slides, but the comment is attempting to describe the 
    register mapping hardware.  In order to support 4 instructions per
    cycle you have up to 8 source and 4 destination registers.  To avoid 
    unnecessary pipeline stalls (due to WAR and WAW register dependancies), 
    we provide more physical registers than the 32 architecturally defined, 
    virtual registers.  This virtual to physical mapping requires that each 
    cycle we need to compare all incoming source register specifiers to the 
    complete register map state to determine which physical register holds 
    the source register data.  Since there are 80 physical registers, we 
    need to compare 8 source registers against 80 physical register
    mappings to determine the physical location of each source - so 8 X 80 
    compares take place each cycle.  The CAM, as described earlier, is a 
    comparitor which can also store state.  Since the virt-phy mapping 
    changes as you write new destination registers, you need to constantly 
    update the values being compared.  Doing this every 2ns is a real trick. 
    And it simultaneously occurs for fp registers (4 X 72). 
    
    Ed 
    
632.11DECCXL::OUELLETTEmudseason into blackfly seasonFri May 16 1997 15:211
Now that's pretty neat...  Thanks for the explanation Ed.
632.12BRSAXP::brodhcp2-95.bro.dec.com::CornelisRoger - OMS Brussels - 856-7612Thu May 22 1997 07:334
Thanks!

Roger