All

 

Congratulations to Steve Comstock with http://www.trainersfriend.com/  for posting first solution to problem #13 using 5 packed decimal instructions.  See the source and generated output here:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

Some follow-up questions and comments after reading all the posts today regarding the ZMFACC contest on the http://www.listserv.uga.edu/archives/asm370.html:

 

1.       In response to the question about the purpose of the contest and the trade-off between fewest instructions, fastest execution, least memory, and ease of maintenance, I have updated the web site to say the primary goals are to have some fun and learn more about mainframe assembler.  Hopefully in the process folks will also share what they think makes for maintainable assembler code.  Certainly I agree that when writing assembler code for a project where the code will live on indefinitely and probably be maintained by someone else, the first priority may well be making the code as understandable and portable as possible including such things as defining the derivation and limits of any constants, avoiding tricks which may be machine dependant and may make future upgrades and/or porting to other languages much more difficult.

2.      In response to the statement that floating point should not be used in the solution to this problem, I have a question.  What method of rounding is “correct” for this problem?  In Steve’s packed decimal solution he only used 1 extra decimal place to determine rounding.  Using the new packed decimal instructions, it is possible to use any number of “extra” digits to perform rounding up to 34 digits using the 128 bit extended format.   If additional digits are included and there is a carry into the 2 decimal digits required, the result is different.  Perhaps one of the most common methods for performing this function is using a COBOL statement like DIVIDE TOTAL-PRICE BY QUANTITY GIVING UNIT-PRICE ROUNDED.  What kind of code does this generate and does the COBOL standard say exactly how many extra digits or bits will be used in performing the rounding.  I can remember in my distant past, having issues with this sort of thing when simply upgrading from one COBOL compiler version to another on the same machine let along switching between platforms with different hardware and software.  If I am reading this document correctly it would appear that the COBOL “Standard” calls for using 128 bit decimal floating point to handle ROUNDED in a way that is portable across machines and compilers: http://www.cobolstandard.info/j4/files/05-0063.doc

 

Don Higgins

mailto:don@...

http://don.higgins.net