Shmuel Metz (Seymour J.) wrote:
In <listserv%[email protected]>, on 08/12/2010
at 01:09 AM, Supra Uche <[email protected]> said:
I think that new instructions run applications more efficiently than
the previous ones.
Not quite. New code *may* run faster if it uses the new instructions;
old code will not be affected.
When there is a new hardware generation, IBM favors "new" workloads so
some older instructions will run much slower (relatively speaking).
For example, TR and TRT are "dogs" now because they're implemented in
millicode. They used to be "blazing" fast when they were implemented in
hardware prior to the 9672 G4s. Similarly, EX is a "dog" on the latest
machines. It used to be much, much faster. Specifically:
o On z9 LA executes in .6 cycles. EX/NOPR executes in 9 cycles -- 15
times slower than LA.
o On z10 LA executes in .7 cycles. EX/NOPR executes in 62 cycles (88
times slower than LA) except when executed in a tight loop, in which
case it is only 16 cycles or 23 times slower than LA. (Experiments seem
to suggest that EX on z10 is being aided by branch prediction circuitry
which won't help most of the time.)
You can find this information published in Session 2215: "z10 User
Experience" from the proceedings for SHARE in Denver, August 2009.
--
Edward E Jaffe
Phoenix Software International, Inc
831 Parkview Drive North
El Segundo, CA 90245
310-338-0400 x318
[email protected]
http://www.phoenixsoftware.com/
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
