From: Bob Supnik Newsgroups: alt.sys.pdp10 Subject: KS10 fp behavior with frac = -0 Date: Tue, 20 Mar 2001 10:38:30 -0500 Organization: occasional Message-ID: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> X-Newsreader: Forte Agent 1.8/32.548 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Complaints-To: newsabuse@supernews.com Lines: 21 Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news-ge.switch.ch!news.maxwell.syr.edu!newshub2.home.com!newshub2.rdc1.sfba.home.com!news.home.com!sn-xit-01!sn-post-01!supernews.com!corp.supernews.com!not-for-mail Xref: chonsp.franklin.ch alt.sys.pdp10:4209 Debugging diagnostic DSKCA, I ran into the following case: input = 577000,,000000 fsc by 1 expected output = 5754000,,000000 The input is a negative floating point number with sign = 1, fraction = 0, and exponent = 201. The processor reference manual, and the FSC instruction description, says that if the input fraction is 0, FSC clears the result. Yet the result is non 0. The microcode seems to treat sign = 1, frac = 0 as a valid but unnormalized negative number; that is, as though the fraction = 400000000 and the exponent is one higher. So the normalized equivalent of the input is 576400,,000000, and the output follows immediately. Question: is this the general treatment of sign = 1, fraction = 0, throughout the architecture, or unique to the KS10 or FSC? /Bob Supnik ###### From: Rich Alderson Newsgroups: alt.sys.pdp10 Subject: Re: KS10 fp behavior with frac = -0 Date: 20 Mar 2001 14:38:06 -0500 Organization: Systems Administration, XKL LLC, Redmond WA 98052 Lines: 28 Sender: alderson+news@panix6.panix.com Message-ID: References: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> NNTP-Posting-Host: panix6.panix.com X-Trace: news.panix.com 985117086 17653 166.84.0.231 (20 Mar 2001 19:38:06 GMT) X-Complaints-To: abuse@panix.com NNTP-Posting-Date: 20 Mar 2001 19:38:06 GMT X-Newsreader: Gnus v5.7/Emacs 20.7 Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news-ge.switch.ch!enews.sgi.com!newsfeed.mathworks.com!panix!news.panix.com!not-for-mail Xref: chonsp.franklin.ch alt.sys.pdp10:4161 Bob Supnik writes: > Debugging diagnostic DSKCA, I ran into the following case: > > input = 577000,,000000 > fsc by 1 > expected output = 5754000,,000000 [ snip ] > Question: is this the general treatment of sign = 1, fraction = 0, > throughout the architecture, or unique to the KS10 or FSC? On a Toad-1 (XKL-1 CPU), we see the expected result: DDT 1[ 0 577000,,000000 fsc 1,1$x <> 1[ 575400,,0 I can't check this on a KL. -- Rich Alderson alderson+news@panix.com "You get what anybody gets. You get a lifetime." --Death, of the Endless ###### Newsgroups: alt.sys.pdp10 From: htimsc@moc.rric.pcuu.orobenots (Chris Smith) Subject: Re: KS10 fp behavior with frac = -0 References: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> Organization: The act or process of organizing or being organized Date: Wed, 21 Mar 2001 03:18:39 GMT Message-ID: Lines: 37 Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!newsfeed-zh.ip-plus.net!news.ip-plus.net!news.tesion.net!news.belwue.de!news.uni-stuttgart.de!news.stealth.net!news.maxwell.syr.edu!feedeast.news.agis.net!us.telia.net!howland.erols.net!news-out.nntp.airnews.net.MISMATCH!cabal10.airnews.net!news.airnews.net!cabal11.airnews.net!news.airnews.net!egsner!stoneboro!htimsc Xref: chonsp.franklin.ch alt.sys.pdp10:4156 Bob Supnik writes: | Debugging diagnostic DSKCA, I ran into the following case: | | input = 577000,,000000 | fsc by 1 | expected output = 5754000,,000000 | Question: is this the general treatment of sign = 1, fraction = 0, | throughout the architecture, or unique to the KS10 or FSC? A microcode emulator might tell. I know that 000000,,000000 fsc'd by anything is 000000,,000000. That result makes some sense, as you say, but I find it really odd. There's something like it in a Knuth exercise -- his floating point is sign and magnitude, but he briefly discusses two's complement. With the (to me odd) interpretation of sign=1, fraction=0 allowed and meaning -1.0. The way I had thought PDP-10s worked is much cleaner. Normalized positive numbers have a 1 bit, so negating them doesn't carry into the exponent. So the exponent is always one's complement and the fraction is always two's complement. The above unnormalized zero has a two's complemented exponent. It's a special case never otherwise seen, and I can't imagine it making sense to spend hardware or microcode worrying about it. I can imagine that the post-operation normalization step of add-like instructions could turn 576000,,000000 into 575400,,000000. It is much more difficult to imagine that all instructions are prepared to handle this abberant -1.0 fraction. Some subset of PDP-10's -- KI? -- could not do integer 1B35 / 1B35 = 1. I'd lay decent odds that 577000000000 FDVR 577000000000 doesn't produce +1.0 on any PDP-10. ###### From: Rich Alderson Newsgroups: alt.sys.pdp10 Subject: Re: KS10 fp behavior with frac = -0 Date: 21 Mar 2001 15:51:34 -0500 Organization: Systems Administration, XKL LLC, Redmond WA 98052 Lines: 25 Sender: alderson+news@panix2.panix.com Message-ID: References: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> NNTP-Posting-Host: panix2.panix.com X-Trace: news.panix.com 985207894 13412 166.84.0.227 (21 Mar 2001 20:51:34 GMT) X-Complaints-To: abuse@panix.com NNTP-Posting-Date: 21 Mar 2001 20:51:34 GMT X-Newsreader: Gnus v5.7/Emacs 20.7 Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news-ge.switch.ch!newsfeeds.belnet.be!news.belnet.be!news.tele.dk!144.212.100.101!newsfeed.mathworks.com!panix!news.panix.com!not-for-mail Xref: chonsp.franklin.ch alt.sys.pdp10:4218 htimsc@moc.rric.pcuu.orobenots (Chris Smith) writes: > Some subset of PDP-10's -- KI? -- could not do integer 1B35 / 1B35 = 1. > I'd lay decent odds that 577000000000 FDVR 577000000000 doesn't produce > +1.0 on any PDP-10. XKL-1 CPU: !xddt DDT 1[ 0 577000,,0 2[ 0 577000,,0 fdvr 2,1$x <> 1[ 577000,,0 2[ 201400,,0 $f; 1.0000000E+00 ^Z ! -- Rich Alderson alderson+news@panix.com "You get what anybody gets. You get a lifetime." --Death, of the Endless ###### Newsgroups: alt.sys.pdp10 From: htimsc@moc.rric.pcuu.orobenots (Chris Smith) Subject: Re: KS10 fp behavior with frac = -0 References: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> Organization: The act or process of organizing or being organized Date: Wed, 21 Mar 2001 23:30:15 GMT Message-ID: Lines: 18 Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news-ge.switch.ch!feed2.news.luth.se!luth.se!logbridge.uoregon.edu!news.maxwell.syr.edu!howland.erols.net!news-out.nntp.airnews.net.MISMATCH!cabal10.airnews.net!news.airnews.net!cabal11.airnews.net!news.airnews.net!egsner!stoneboro!htimsc Xref: chonsp.franklin.ch alt.sys.pdp10:4240 Rich Alderson writes: | 2[ 201400,,0 $f; 1.0000000E+00 Man, I can't get anything right today. It's still odd. However, orange phone book, 3rd edition, SRM, p. 2-33: A number with a 1 in bit 0 and 0s in bits 9-35 is not simply an incorrect representation of zero, but rather an unnormalized "fraction" with value -1. This unnormalized number can produce an incorrect answer in any operation. This may explain the apparent inconsistency between KS10 FSC and the instruction description if you consider "if the fractional part of AC is zero" to be false for this -1 case. (Under this interpretation, 400000000000 is -2^127.) ###### From: Bob Supnik Newsgroups: alt.sys.pdp10 Subject: Re: KS10 fp behavior with frac = -0 Date: Thu, 22 Mar 2001 08:57:57 -0500 Organization: occasional Message-ID: References: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> X-Newsreader: Forte Agent 1.8/32.548 MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Complaints-To: newsabuse@supernews.com Lines: 30 Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news-ge.switch.ch!news.imp.ch!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!newshub2.home.com!newshub2.rdc1.sfba.home.com!news.home.com!sn-xit-01!sn-post-01!supernews.com!corp.supernews.com!not-for-mail Xref: chonsp.franklin.ch alt.sys.pdp10:4258 Chris, The 'orange book', SRM, is that a different manual from AD-H391A-T1 Processor Reference Manual? If so, has someone scanned it and placed it online anywhere? Thanks /Bob On Wed, 21 Mar 2001 23:30:15 GMT, htimsc@moc.rric.pcuu.orobenots (Chris Smith) wrote: >Rich Alderson writes: > >| 2[ 201400,,0 $f; 1.0000000E+00 > >Man, I can't get anything right today. > >It's still odd. However, orange phone book, 3rd edition, SRM, p. 2-33: > > A number with a 1 in bit 0 and 0s in bits 9-35 is not simply an > incorrect representation of zero, but rather an unnormalized > "fraction" with value -1. This unnormalized number can produce > an incorrect answer in any operation. > >This may explain the apparent inconsistency between KS10 FSC and >the instruction description if you consider "if the fractional part >of AC is zero" to be false for this -1 case. > >(Under this interpretation, 400000000000 is -2^127.) ###### Newsgroups: alt.sys.pdp10 From: htimsc@moc.rric.pcuu.orobenots (Chris Smith) Subject: Re: KS10 fp behavior with frac = -0 References: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> Organization: The act or process of organizing or being organized Date: Thu, 22 Mar 2001 15:14:43 GMT Message-ID: Lines: 30 Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news-ge.switch.ch!newsfeeds.belnet.be!news.belnet.be!newsfeed.icl.net!newsfeed.skycache.com!Cidera!portc03.blue.aol.com!howland.erols.net!news-out.nntp.airnews.net.MISMATCH!cabal10.airnews.net!news.airnews.net!cabal11.airnews.net!news.airnews.net!egsner!stoneboro!htimsc Xref: chonsp.franklin.ch alt.sys.pdp10:4244 Bob Supnik writes: | The 'orange book', SRM, is that a different manual from AD-H391A-T1 | Processor Reference Manual? If so, has someone scanned it and placed | it online anywhere? It's probably an older version of that manual. The orange book is the 'decsystem10 assembly language handbook,' comprising six things, the first of which is the DECsystem-10 System Reference Manual, order no. DEC-10-HGAD-D. First edition May 1968, second edition Decemeber 1971: This edition has been expanded to provide system reference information for a DECsystem-10 with KA10 or KI10 central processors. The KI10 material has been incorporated into the text throughout. I don't know of any online version. The whole phone book is 975 pages long and went for $5.00. Hm, there's also an older white phone book, the 'pdp10 reference handbook'. It has the first edition, KA-only, system reference manual. Which has the same text about negative floating point numbers with zero fractions. And both of them have Appendix D, Algorithms (retitled 'KA10 Algorithms'), spelling out in detail what the floating point instructions do. According to it, KA's did the same thing with that bizarre aberration of a number. Bit string. I'd scan it in if I had a scanner. Maybe somebody has both. ###### Path: chonsp.franklin.ch!not-for-mail From: Neil Franklin Newsgroups: alt.sys.pdp10 Subject: Re: KS10 fp behavior with frac = -0 Date: 22 Mar 2001 23:36:28 +0100 Organization: My own Private Self Lines: 32 Message-ID: <6u4rwlfn2r.fsf@chonsp.franklin.ch> References: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> NNTP-Posting-Host: chonsp.franklin.ch X-Trace: chonsp.franklin.ch 985300588 561 10.0.3.2 (22 Mar 2001 22:36:28 GMT) X-Complaints-To: news@chonsp.franklin.ch NNTP-Posting-Date: 22 Mar 2001 22:36:28 GMT X-Newsreader: Gnus v5.7/Emacs 20.4 Xref: chonsp.franklin.ch alt.sys.pdp10:4262 htimsc@moc.rric.pcuu.orobenots (Chris Smith) writes: > Bob Supnik writes: > > | The 'orange book', SRM, is that a different manual from AD-H391A-T1 > | Processor Reference Manual? If so, has someone scanned it and placed > | it online anywhere? > > It's probably an older version of that manual. The orange book is > the 'decsystem10 assembly language handbook,' comprising six things, > the first of which is the DECsystem-10 System Reference Manual, > order no. DEC-10-HGAD-D. First edition May 1968, second edition > Decemeber 1971: > > This edition has been expanded to provide system reference information > for a DECsystem-10 with KA10 or KI10 central processors. The KI10 > material has been incorporated into the text throughout. The KI-10 version exists? Great! Up to now I only have seen KL/KS and XKL versions. Sort of not exactly ehat is needed for KI cloning. > I'd scan it in if I had a scanner. Maybe somebody has both. Or arrange to have it go to someone with an scanner. -- Neil Franklin, neil@franklin.ch.remove http://neil.franklin.ch/ Hacker, Unix Guru, El Eng HTL/FH/BSc, Sysadmin, Roleplayer, LARPer ###### From: bsupnik@us.inter.net (Bob Supnik) Newsgroups: alt.sys.pdp10 Subject: Re: KS10 fp behavior with frac = -0 Date: 23 Mar 2001 00:42:05 GMT Organization: rarely Lines: 30 Message-ID: <99e64t$4a2@news-central.tiac.net> References: <5stebtkek798jr8gg240h6rtaar8qav8ef@4ax.com> NNTP-Posting-Host: ip147.bedford16.ma.pub-ip.psi.net Mime-Version: 1.0 Content-Type: Text/Plain; charset=US-ASCII X-Newsreader: WinVN 0.99.9 (Released Version) (x86 32bit) Path: chonsp.franklin.ch!pfaff.ethz.ch!news-zh.switch.ch!news-ge.switch.ch!news.maxwell.syr.edu!news-peer1.tiac.net!posterchild2.tiac.net!news@tiac.net Xref: chonsp.franklin.ch alt.sys.pdp10:4271 Thanks, Al Kossow has scanned in the 1970 edition, I'll check Appendix D. In article , htimsc@moc.rric.pcuu.orobenots says... > >It's probably an older version of that manual. The orange book is >the 'decsystem10 assembly language handbook,' comprising six things, >the first of which is the DECsystem-10 System Reference Manual, >order no. DEC-10-HGAD-D. First edition May 1968, second edition >Decemeber 1971: > > This edition has been expanded to provide system reference information > for a DECsystem-10 with KA10 or KI10 central processors. The KI10 > material has been incorporated into the text throughout. > >I don't know of any online version. The whole phone book is 975 pages >long and went for $5.00. > >Hm, there's also an older white phone book, the 'pdp10 reference >handbook'. It has the first edition, KA-only, system reference >manual. Which has the same text about negative floating point numbers >with zero fractions. > >And both of them have Appendix D, Algorithms (retitled 'KA10 Algorithms'), >spelling out in detail what the floating point instructions do. >According to it, KA's did the same thing with that bizarre aberration >of a number. Bit string.