Change branch to branch relative wherever possible. This is an easy change and is surprising how much it helps.
If only branching to a subroutine from one location, replace bas/jas with j. You can return to the label of the J+4. (I have some macros for calling subroutines that make it easy.) Aggressively remove all possible Store Into Instruction Stream (SIIS), including where you're saving registers and the boundary between end of code and first data. Moving to base-less code helps because then the assembler will flag some SIIS, since you have no base register for the code. Switching to rsect is harder unless you go all in on re-entrant code design. As noted by Abe, now there are more instructions for using immediate data, and some can hold halfwords. Immediate beats storage. The secret* MVCOS instruction is easier to use than MVCL if the from/to length is the same and no overlap, but it isn't faster than MVC. * I say secret because the documentation is hidden in the "control instructions" chapter with even though it works find for general use if you set r0 to zero first. -----Original Message----- From: IBM Mainframe Discussion List <[email protected]> On Behalf Of Abe Kornelis Sent: Friday, August 28, 2026 1:03 PM To: [email protected] Subject: Re: Efficiency of Instructions Steve, the quality of your algorithm beats instruction A versus B many times over. I have often found that how you organise your data, both on storage devices and in memory during processing, can make a decisive difference in performance. For optimising your product code, it might be more relevant to load the code into non-pageable storage. Pay attention to pipeline optimisation - at least avoid sequences that degrade pipeline and cache efficiency. Individual instructions - although tempting to discuss ;-) are the least concern. That matters only in tight loops executed billions of times a day. Maintaining a queue of reusable storage elements to avoid repeated STORAGE obtain/release sequences matters a lot more than what instructions you use to clear that storage. Major performance characteristics are determined at the earliest design stages. I have found that, as a programmer, one can only try to sub-optimise to minimise the mistakes of the architects :-( Therefore, to optimise your code, zoom out as far as your responsibility/pay grade allows. Start from there. Then work your way down into the details of your design and finally, as a last resort, optimise just your own code. I wish you success! Kind regards & happy programming, Abe Kornelis ========== On 28/08/2026 19:10, Steve Thompson wrote: > I am considering how to get software products to run using fewer MSUs.... > > Is there such a thing as a functional Characteristics manual these > days that describes the behaviors of an instruction on/in a particular > machine level. > > I started thinking like this after dealing with COBOL 6.x and arch > level and the like.... > > Example: At one time, I knew that using up to 6 MVC instructions was > actually faster than an MVCL. (Assuming I have remembered the number > of MVCs correctly). > > I knew that an MVCL is an interruptible instruction, and I think it > was the completion of a Unit of operation where it could be > interrupted. I didn't know what a unit of op was in that case. Was > it interruptible at 256 byte boundaries or 1K boundaries? > > Since IBM doesn't make PLAS (or PLX, or whatever it is called today) > available to us software developers I was just doing some blue sky > thinking and trying to decide if we have this kind of info, if the > juice would be worth the squeeze. > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
