RE: Intel and Microsoft provide higher 32-bit applications performance on Itanium

From: Tom Linden (tom_at_kednos.com)
Date: 01/31/04


Date: Sat, 31 Jan 2004 05:25:35 -0800


  -----Original Message-----
  From: Bill Todd [mailto:billtodd@metrocast.net]
  Sent: Friday, January 30, 2004 10:11 PM
  To: Info-VAX@Mvb.Saic.Com
  Subject: Re: Intel and Microsoft provide higher 32-bit applications
  performance on Itanium

  "JF Mezei" <jfmezei.spamnot@istop.com> wrote in message
  news:401ACDA2.9BD6FEE@istop.com...
> > applications will be carried out in software. At run time, the IA-32
> > Execution Layer software translates the IA-32 application code into
> > native Itanium architecture code and allows it to run as
  native code.
>
> Considering that half the IA64 architecture resides in the EPIC
  compilers,
  is
> it fair to state that any code compiled on and for 8086 would
  not be able
  take
> advantage of any of the fancy performance stuff of IA64 since
  there would
  be
> no explicit parralelism built into the code ?

  Actually, a software mechanism (akin to FX!32) *can* probably mung what it
  interprets from the x86 binary code to take more advantage of
  EPIC features
  (such as they are) than the current hardware emulation can. That
  may be one
  of the reasons it's faster, for that matter.

  IIRC FX!32 had a profiling-like facility which could massage the
  munged code
  based on execution characteristics. Now, I think that I vaguely remember
  that the new Itanic software emulator doesn't produce a permanent native
  binary (to keep with the x86 binary), but I also seem to remember that it
  *may* still improve the native binary as it executes. Oh, hell - someone
  who actually knows something about all this would be a much
  better source of
  information than I am: anyone? Anyone?

Actually, FX!32 cosisted of two parts, the emulator and the optimizer. The
latter actually symthesized an acyclic graph of the program and subsequently
ran a flow analysis, which could be scheduled to run successively to produce
better native alpha code. Of course, you really can't buils a symbol tree
brom this to do, e.g., liftime analysis but it was still pretty impressive.

  - bill

  ---
  Incoming mail is certified Virus Free.
  Checked by AVG anti-virus system (http://www.grisoft.com).
  Version: 6.0.566 / Virus Database: 357 - Release Date: 1/22/2004

---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.566 / Virus Database: 357 - Release Date: 1/22/2004


Relevant Pages

  • RE: non-system but priviledged account OK for installing layered products?
    ... DECMessageQ was written in PL/I and BEA needed to have the product avaialble ... Incoming mail is certified Virus Free. ... Checked by AVG anti-virus system. ...
    (comp.os.vms)
  • ENC: SED question
    ... Incoming mail is certified Virus Free. ... Checked by AVG anti-virus system. ... Outgoing mail is certified Virus Free. ...
    (SunManagers)
  • RE: New OpenVMS-to-Itanium FAQ
    ... comparison to PL/I ... >Incoming mail is certified Virus Free. ... Outgoing mail is certified Virus Free. ... Checked by AVG anti-virus system. ...
    (comp.os.vms)
  • RE: Do we need a DCL debugger?
    ... >> containing the user written lexical functions, ... >Incoming mail is certified Virus Free. ... Outgoing mail is certified Virus Free. ... Checked by AVG anti-virus system. ...
    (comp.os.vms)
  • RE: A whopping 50 percent... ???
    ... < to make those fewer ticks much slower ticks. ... < Incoming mail is certified Virus Free. ... < Checked by AVG anti-virus system. ... Outgoing mail is certified Virus Free. ...
    (comp.os.vms)