Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!rutgers!cs.utexas.edu!tut.cis.ohio-state.edu!ucbvax!hplabs!hpfcso!hpfcda!jvm From: jvm@hpfcda.HP.COM (Jack McClurg) Newsgroups: comp.arch Subject: Re: Fast conversions, another urban myth? Message-ID: <5210001@hpfcda.HP.COM> Date: 25 Sep 89 22:21:38 GMT References: <832@dms.UUCP> Organization: Hewlett-Packard, Fort Collins, CO Lines: 26 From Mash's reference earlier I looked up a reference in the Hewlett-Packard Journal. On an IBM370 running COBOL programs, 1.6% of the instructions exucuted are decimal math and 5.9% of the time spent. This is from page 31 of the August 1986 Issue. From the January 1986 issue, here is the instruction sequence to perform a packed add: r1 and r2 contain packed decimal operands r3 contains 0x99999999 UADDCM 1,3,31 ; prebias operand into r31 ADD 2,31,31 ; perform binary add DCOR 31,31 ; correct result an unpacked add: r1 and r2 contain unpacked decimal operands r3 contains 0x96969696 r4 contains 0x0f0f0f0f r5 contains 0x30303030 ADD 3,1,31 ; prebias operand into r31 ADD 31,2,31 ; binary add into r31 DCOR 31,31 ; correct result AND 4,31,31 ; mask result OR 5,31,31 ; restore sum to unpacked decimal Jack McClurg jvm%hpfcda@hplabs.hp.com