Solution to Extended Precision Natural Log (ELN) Issue
Steven Zoppi
I figured out what is going on and am looking forward to getting the clean build sources from Brian when he has time for sim and the patches he knows are “good”.
In the meantime …
Root Cause
The issue is an apparent overflow problem in the assembler in the XFLC instruction which I will debug when I have a proper build environment.
The algorithm requires the largest EP constant CLOSE to 1 without actually BEING 1. This means that the assembly of the following should work just fine:
XFLC 0.9999999999
But it does not. The assembler generates: 0080 0000 0000
Modifying the operand to 0.9999999994 causes the assembler to generate the proper constant: 0080 1FFF FFFF
Solution
For anyone who needs to build DMS from “scratch” … the problem can be easily fixed by modifying ONE line in s2eln.asm
Line 162 (ID:S2E01620) in Source s2eln.asm:
FONE XFLC 0.9999999999 FLT PT ONE S2E01620
Needs to be changed to this:
FONE XFLC 0.9999999994 FLT PT ONE S2E01620
Full System Test/Verification
Run:
1. Mkdms
2. Ibm1130 loaddms
3. exit sim
4. Run ibm1130
5. Submit this job:
// JOB
// FOR
*ONE WORD INTEGERS
*LIST ALL
*EXTENDED PRECISION
*IOCS (TYPEWRITER)
C
C TEST OF FLOATING POINT PRECISION
C---------------------------------
C
X = 3.0
Y = ALOG(X)
WRITE(1,1000) X,Y
1000 FORMAT('1 X=',F16.9,' LN(X)=',F16.9)
CALL EXIT
END
// XEQ L
6. Enjoy the result:
1 X= 3.000000001 LN(X)= 1.098612289
Wait, IAR: 0000002A (4c80 BSC I ,0028 )
Thanks to Carl and Brian for some hints and helping me dust off some rusty brain cells.
Regards,
Steve
Eddy Quicksall
Hmm … in my run, which gave the correct answer, I used the EMU compiler with its library. I don’t see S2ELN in the map so maybe EMU made a correction there.
Here is the map for that:
The FORTRAN program
*ONE WORD INTEGERS
*EXTENDED PRECISION
*LIST ALL
*IOCS(TYPEWRITER)
C
C TEST OF FLOATING POINT PRECISION
C---------------------------------
C
PAUSE 1234
X = 3.0
Y = ALOG(X)
WRITE(1,1000) X,Y
1000 FORMAT('X=',F16.9,' LN(X)=',F16.9)
CALL EXIT
END
Here is the map:
R 41 6AE8 (HEX) WDS UNUSED BY CORE LOAD
TYPE NAME ENADR LOAD CIDN LNTH TVADR
ILS ILS02 00B3 00B3 0002
ILS ILS04 00C4 00C4 0002
LIBF SWRT 026F 0244 0002 7FD3
LIBF SIOF 0358 0244 0002 7FD0
LIBF SCOMP 038F 0244 0002 7FCD
LIBF SFIO 0425 0244 0002 06B8 7FE2
LIBF WRTYZ 08FC 08FC 0003 00EA 7FDF
LIBF ESTO 09EA 09E6 0002 7FD9
LIBF ELD 0A00 09E6 0002 0030 7FDC
CALL INIT$ 0A1D 0A16 0002 0122 7FFF
LIBF PAUSE 0B38 0B38 0003 0016 7FD6
CALL EALOG 0B4E 0B4E 0002 0094 7FFE
LIBF IFIX4 0BEC 0BE2 0002 002A 7FCA
LIBF IFIX 0C0C 0C0C 0002 002C 7FC7
LIBF FLOT4 0C38 0C38 0002 000C 7FC4
LIBF FLOAT 0C44 0C44 0002 000C 7FC1
CALL IOCS2 0C5D 0C50 0002 000E 7FFD
CALL CXCLR 0C5E 0C5E 0002 0008 7FFC
CALL NOIRQ 0C66 0C66 0003 0088 7FFB
CALL LKIRQ 0CC0 0C66 0003 7FF5
CALL MASK 0CEE 0CEE 0003 001A 7FFA
CALL UNMK 0CF3 0CEE 0003 7FF8
CALL JCLOK 0D08 0D08 0002 003A 7FF9
CALL ZCOR$ 0D42 0D42 0003 0034 7FF7
CALL LADSI 0D76 0D76 0002 001A 7FF6
CALL $TRJC 0D90 0D90 0002 0070 7FF4
CALL $TRND 0DEA 0D90 0002 7FF1
LIBF DIV4 0E1B 0E00 0002 008E 7FBE
CALL RSVDR 0E8E 0E8E 0002 00BA 7FF3
CALL RELDR 0ED8 0E8E 0002 7FF2
CALL $FILE 0F48 0F48 0002 0282 7FF0
LIBF EGETP 11CA 11CA 0002 0012 7FBB
LIBF EADD 11E7 11DC 0002 0062 7FB8
LIBF NORM 123E 123E 0003 002A 7FB5
LIBF EDIV 126C 1268 0002 005C 7FB2
LIBF XMD 12C4 12C4 0003 0042 7FAF
LIBF EMPY 130A 1306 0002 0082 7FAC
CALL IGAUX 1388 1388 0002 0056 7FEF
LIBF TRAP$ 13DE 13DE 0002 005E 7FA9
LIBF FARC 143C 143C 0002 0022 7FA6
LIBF XDD 145E 145E 0003 004A 7FA3
CALL CXCHN 14A8 14A8 0002 0012 7FEE
CORE LOAD NAME IS- T2
7F6E (HEX) IS $XR3X TR VCTR ADDR
0214 (HEX) IS THE EXECUTION ADDR
X= 3.000000001 LN(X)= 1.098612289
Eddy
Brian Knittel
--
You received this message because you are subscribed to the Google Groups "IBM1130" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ibm1130+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Steven Zoppi
Thank you Brian! Looking forward to the new distro.
-Steve
From: ibm...@googlegroups.com [mailto:ibm...@googlegroups.com] On Behalf Of Brian Knittel
Sent: Sunday, July 31, 2016 8:55 AM
To: ibm...@googlegroups.com
Subject: Re: [IBM1130] Solution to Extended Precision Natural Log (ELN) Issue
The issue would not shown up on any "real" disk images. The bug was in the cross assembler I write to bring up DMS from source code back around 2001. We found the bug last week. A full updated source code distro will go out this coming week.
Eddy Quicksall
Once you get the new distro, can someone please send me the DCI so I can try it on my emulation?
--
Eddy Quicksall
Actually, can you please send me the DCI for both?
Eddy Quicksall
Regarding this … I did not have to change any IBM code and someone already said my disassembly matched your code. I did have a bug in the early days when Intel came out with the 486 and that was because the 486 treated a shift left on the 8086 differently. And my, if I remember correctly, my bug showed up in one of the extended precision routines.
Eddy
From: ibm...@googlegroups.com [mailto:ibm...@googlegroups.com] On Behalf Of Steven Zoppi
Sent: Sunday, July 31, 2016 12:06 AM
To: ibm...@googlegroups.com
--
Steven Zoppi
The disassembly of the instruction sequence matched. The constant generated by the cross-assembler (at label FONE) did not … and it turns out, it was kind of an important one.
Brian Knittel
--