<snip> I'm doing an analysis of one of our application programs. As part of this, I'm seeing the COBOL code generate a lot of PACK/UNPACK instruction pairs.
Does anyone have an idea on the performance impact of all these (potential millions of executions in a single run)? </snip> A long time ago, on a processor far far away, I ran into this exact issue. To be specific, on a S370/138 I was responsible for correcting (not my code) a program that absolutely maxed the 370/138. Upon examining the program I found that all input, linkage section and working storage fields were defined in display format. The processing logic was read, process, call subprogram, process, return to main, clean-up, write. In essence, each numeric field used was packed/unpacked a minimum of 8 times. After modification only the working storage and linkage section variables to an apporpriate internal format (comp or comp-3), this same program was not noticed when processed. As a bonus, it ran about 20% faster that before the mods. NOTE: I did not change one line of logic, only data descriptions. It turns out that there were also some logic changes that also sped up the program some more. ISTR that the total improvement was on the order of 40% elapsed time. The impact of mutiple pack/unpack inctructions is significant. In the case of fixed point decimal operations, there are also CVD/CVB instructions associated with the PACK and UNPACK operations. (2X CVD/CVB for each PACK and UNPACK. As you can see, the amount of "useless processing" grows geometrically. Before you change a line of code, change the data representations to the appropriate internal format and prepare to be astounded at the improvement. HTH, ---------------------------------------------------------------------- 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
