On 7/15/2011 8:14 PM, Edward Jaffe wrote:
On 7/14/2011 12:55 PM, Tom Marchant wrote:
On Thu, 14 Jul 2011 08:40:37 -0600, Steve Comstock wrote:
yes, there is some penalty
for AMODE switching.
What penalty is that, Steve? Do BASSM and BSM run significantly
slower than BASR/BALR and BR?
There is no way the processor can know in advance which bits will be on in a
branch target register, so its seems likely that the pipeline must be flushed
when 'surprise' AMODE switching occurs for pointer-defined linkage. However, if
the BASSM/BSM is executed frequently enough, it's also possible the branch
history/prediction logic in the processor can guess the right target AMODE a
significant percentage of the time to minimize such 'surprises'.
I would not expect this to be an issue at all with the SAMxx instructions.
The real issue is what LE uses to switch AMODE when the
setting is AMODE31(OFF); it may be more than just a BASSM
or SAMxx. I don't really know, but I somehow expect overhead
and complexity in that scenario (COBOL interacting with non-LE
Assembler, as specified in the OP)
--
Kind regards,
-Steve Comstock
The Trainer's Friend, Inc.
303-393-8716
http://www.trainersfriend.com
* Special promotion: 15% off on all DB2 training classes
scheduled by September 1, taught by year end 2011
* Check out our entire DB2 curriculum at:
http://www.trainersfriend.com/DB2_and_VSAM_courses/DB2curric.htm
----------------------------------------------------------------------
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
