This currently is.

Possibly an MF=L in the static area.

MF=L in the DSECT

Copy static to DSECT and use MF=E

Or am I missing something obvious?


On Tue, 27 Aug 2024 17:37:58 +0200 Bernd Oppolzer <[email protected]>
wrote:

:>Some topic drift:
:>
:>for one of my customers, I had to change some very old ASSEMBLER 
:>programs (1980 era) because of
:>serious SIIS issues ... programs modifying static areas near the 
:>instruction stream, thereby invalidating
:>cache lines and leading to performance issues.
:>
:>One of the problem areas were IBM macros such as TIME DEC (IIRC), which 
:>modify their inline parameter list.
:>The examples can be found on the web, I think. Other areas are inhouse 
:>macros, which had the same or
:>similar issues.
:>
:>There is a IMO clever idea at this customer's site:
:>
:>if a global macro symbol &GBREN is set, all macros put their temporary 
:>storage in a certain DSECT
:>which is then allocated/getmained at program startup (together with the 
:>save area) and freed at the end.
:>The base register is 13 ... which is needed anyway to address the save 
:>area.
:>
:>This way, the SIIS problems disappear at the same time, because the 
:>target addresses of the store
:>and move instructions of the macros now are far away from the 
:>instruction stream.
:>
:>I fixed most of the SIIS problems of my customer by "cheating" and 
:>telling the programs that they were
:>RENT (by setting &GBREN), when in fact they were not. And then I 
:>supplied a static area instead of the
:>dynamic area (which is used with true RENT) and addressed this area 
:>using "dependent USING".
:>
:>My question now:
:>
:>what do you think of the idea asking IBM to provide a similar technique 
:>(an optional parameter,
:>like MF=L and MF=E, which is present on almost all macros) on their 
:>macros, so that the
:>definition of the work areas can be put in another (separate) DSECT ? 
:>This would IMO make
:>the creation of RENT programs which include IBM macros so much easier 
:>... and help us
:>solve some SIIS problems.
:>
:>Kind regards
:>
:>Bernd
:>
:>
:>
:>Am 27.08.2024 um 15:10 schrieb Peter Relson:
:>> ...
:>>
:>> Will the macros ever be "changed" from using "LA" or "LAE" to some 
long-displacement form unconditionally? Surely "no". That's the same 6-byte vs 
4-byte consideration.
:>> Conditionally? Maybe. But only if a formal request is submitted. And it's 
more likely to happen for those macros that are tool-generated (as many of the 
macros created since the mid 80's are).
:>>
:>> ...
:>
:>----------------------------------------------------------------------
:>For IBM-MAIN subscribe / signoff / archive access instructions,
:>send email to [email protected] with the message: INFO IBM-MAIN

--
Binyamin Dissen <[email protected]>
http://www.dissensoftware.com

Director, Dissen Software, Bar & Grill - Israel

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to