I wanted to close the loop on this, and frankly, vent a little bit. I played around and played around with different C++ and LE options -- nothing seemed to make a big difference. So finally I wrote a trivial, 3-machine instruction, do-nothing ADATA exit in HLASM. It turns out the overhead is not in C++ and LE -- it's in the COBOL compiler calling exits! Here are the numbers -- CPU seconds, details below -- for various situations:
NOADATA: 1.91 CPU seconds ADATA to DUMMY: 1.935 CPU seconds ADATA to DASD: 2.04 CPU seconds ADATA to DUMMY, 3-instruction exit: 2.82 CPU seconds -- an overhead of over 45 percent! Details: REGION=0M, COBOL 6.2, up-to-date on PTFs, typical options, OPT(0), 52,063 lines, 34,012 statements, z14-M01, TCB+SRB time per IEF033I. All tests run on a reasonably quiet LPAR. Grrrrr! Charles On Sat, 30 Dec 2023 16:36:07 -0600, Charles Mills <charl...@mcn.org> wrote: >I have a working Enterprise COBOL compiler exit written in XLC C++. It >functions "correctly" but it is consuming more CPU time than expected, and I >have this suspicion that it is going through LE initialization on each entry. > >The COBOL doc says of exit modules "The Enterprise COBOL compiler >automatically manages a preinitialized Language Environment at compile time, >and calls compiler exits within this environment. Therefore, the following >rules apply: Compiler exits are run as subprograms instead of main programs. >Compiler exits must not include logic for explicitly initializing or >terminating Language Environment." > >I have not done anything real special in the C++ code. My entry point is *not* >a "main" function and is declared extern "OS" void. I am pre-binding and >linking pretty much as I have done for other C++ programs. I tell the binder >simply INCLUDE, ENTRY and NAME with no reference to CEESTART. > >Is there something "special" that I should be doing differently? > >Phrasing the question differently, if an LE program loads an LE program by >module name and then calls it, how should that called program be compiled and >bound so as to avoid repeated LE initialization? ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN
