On Thu, 2 Nov 2017 20:48:23 +0000, Feller, Paul wrote: >Our shop worked with IBM on a compiler performance issue which ended up >creating APAR PI84408. The official PTF UI51349 is also now out. This fixed >an issue with the compiler using lots more CPU then we thought is should when >using the NUMCHECK and SSRANGE options. For us the fix reduced the CPU time >when using the NUMCHECK and SSRANGE options from around 10 minutes to around 3 >minutes. Naturally this also reduced the overall run time. ... > It's astonishing/dismaying that range checking could cause a factor of 3 or 4 increase in *compile* time (less astonishing for execution time). Perhaps this isn't a case for which IBM optimizes the compiler.
>... What started this was if we ran a compile without the NUMCHECK and SSRANGE >options the compile took around 45 to 50 seconds of CPU time. Unfortunately >the APAR PI84408 is part of a larger group of fixes rolled into the PTF. > Here, from an ISV developer's PoV, I laud IBM's behavior. Testing all possible combinations of patches exponentially increases the resource requirement for testing. Or, if the testing resource is limited, the validation is proportionally less thorough, so patch interactions may go undiscovered. I know Ed G. has a contrary view. -- gil ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
