Skip to content

First set of files pulled from VM Workshop VMARC archive #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
james4765 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

First set of files pulled from VM Workshop VMARC archive #4

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions vmworkshop-vmarcs/1995/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,3 @@
## Source

Code from VMARCs distributed as part of the 1995 VM Workshop.
68 changes: 68 additions & 0 deletions vmworkshop-vmarcs/1995/cray2i95/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,68 @@
# ibm-cray(3,L) - IBM-Cray data conversion

"", 3 June 1991


<a name="purpose-"></a>

# Purpose

Converts arrays of floating point numbers between IBM and Cray data formats.

<a name="library"></a>

# Library

faux, Cornell's Fortran Auxiliary Subroutine Library (libfaux.a)

<a name="syntax"></a>

# Syntax

```
call cfdc(ibm, cray, n)

call cfcd(cray, ibm, n)

where:
ibm Array of IBM floating point numbers, REAL*8 values.
cray Array of CRAY floating point numbers.
n Number of elements in arrays to convert, integer.
```

<a name="description"></a>

# Description

cfdc converts floating point double precision numbers to Cray floating point numbers

cfcd converts floating point Cray numbers to floating point double precision IBM numbers

Routines are written in VS FORTRAN and thus are available for both CMS and AIX/370 systems. They could be compiled by either 'fortvs' command under VM/CMS or 'fvs' command under AIX/370.

Precision of the conversion is the best possible. When floating point formats differ in number of bits for mantissa, which affects data precision, the least mantissa bits are rounded. When formats differ in number of bits for exponent, which affects the range of data, the biggest possible value or zero value is assigned if the old number can't be represented.

Routines convert data in memory and must be invoked with three arguments: input array name, output array name, and number of elements in each of two arrays. When called from a C program, the last parameter must be passed by the pointer. The same variable name can be used for the first and the second arguments. This means that conversion can be done "in place" thus saving memory space, if necessary.

The table below shows a summary of conversion time per element in microseconds of virtual cpu time on a 3090 J processor. Each entry represents the average value of one thousand conversions of an array with a thousand elements.

```
----------------------------------
| Routine name | Time to convert |
----------------------------------
| cfdc | 0.83 |
| cfcd | 0.75 |
----------------------------------
```

<a name="notes"></a>

# Notes

1. Non-normalized numbers with zero exponents are kept intact.

2. In IBM to CRAY conversion, precision in the mantissa could be lost by rounding off the least significant bits. 0 &lt;= |error| &lt;= .18E-14 (From 5 to 8 least significant bits out of 56 mantissa bits could be rounded.)

3. CRAY to IBM conversion does not incur the lost of mantissa accuracy.

4. CRAY values that don't fit IBM standard are converted to either the biggest IBM values (positive or negative) or to zero.
119 changes: 119 additions & 0 deletions vmworkshop-vmarcs/1995/cray2i95/cfcd.assemble
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,119 @@
* ------------------------------------------------------------------- *
* *
* copyright - virtual machine / system product *
* contains restricted materials of cornell university *
* 5664-167 (c) copyright cornell university 1990 *
* licensed materials - property of cornell university *
* refer to copyright instructions *
* form number g120-2083 *
* ------------------------------------------------------------------- *
* processor - assebmler release 5.0 5/1/83 under aix/370 *
* *
* dependences - none *
* *
* attributes - reenterant *
* *
* entry point - _cfcd_ *
* *
* errors - not found *
* *
* 06/16/89 valery i. garger, technology integration *
* group, cnsf, cornell university *
* ------------------------------------------------------------------- *
* change log: some minor changes and bugs fixed 09/13/89 *
* converted to aix/370 05/01/90 *
* valery garger *
* ------------------------------------------------------------------- *
* convert floating point, cray 64-bit to ibm 64-bit (fortran real*8). *
* call this subroutine by: *
* *
* call cfcd(cray, ibm, n) *
* *
* input: cray array of 64-bit cray floating point numbers. *
* n number of elements in cray to convert, integer. *
* output:ibm array of ibm floating point numbers, real*8 values. *
* *
* format (bits, left to right): | exponent bias: *
* sign exponent mantissa | *
* ibm 1 7 56 | 64 hex *
* cray 1 15 48 | 16384 binary *
* | *
* latency: 1.064 microsecond per conversion of one element *
* on one of the ibm 3090-600e processors. *
* *
* usage notes: *
* 1. cray values that doesn't fit ibm standard are converted to *
* either the biggest ibm values (positive or negative) or to zero. *
* 2. non-normalized numbers with zero exponents are kept intact *
* (suspected to be integer values). *
* 3. conversion does not incur the lost of mantissa accuracy. *
* ------------------------------------------------------------------- *
_cfcd_ csect
entry _cfcd_
b 34(,15)
dc al1(6+22)
dc cl6'cfcd '
dc cl22'05-01-90/vesion#1.9'
stm 2,15,x'10'(13)
lr 12,13
la 11,x'68'
slr 13,11
st 12,4(13)
lr 12,15
using _cfcd_,12
* -----------------------------------------------
lr 4,2 pointer to n
lr 3,1 pointer to ibm
lr 2,0 pointer to cray
l 4,00(,4) g4: n
next ds 0h do j = 1,n
lm 8,9,00(2) gr8,gr9: for tiss, now intact
lr 6,8
n 6,=x'80000000' gr6: sign
lr 7,8
n 7,=x'7fff0000' gr7: exponent
lr 1,8
n 1,=x'00008000' gr1: nrmbit
or 1,7
bz done if so, keep intact
n 8,=x'0000ffff' tiss(1)
expon ds 0h
srl 7,16
s 7,=f'16128' exp = exp - 16384 + 256
lr 5,7
n 5,=f'3' .and. last two bits
bz *+8 avoid if mod(exp,4)=0
a 7,=f'4' +1 to next to two last bits
sra 7,2 see whole new exponent
ltr 7,7
bnm expos to exponet is positive
sr 8,8
sr 9,9
b strsgn store zero w/sign
expos ds 0h exp is pos
c 7,=f'127'
bnp exprng to exponent in a range
l 8,=x'7fffffff'
l 9,=x'ffffffff'
b strsgn
exprng ds 0h
sll 7,24 exponent on place
tissa ds 0h
sldl 8,5 at least on 5 bits left
ltr 5,5
bnz *+8
la 5,4
s 5,=f'1'
bz *+8
sldl 8,00(5) logically double
store ds 0h
or 8,7
strsgn or 8,6
done stm 8,9,00(3)
la 2,8(2)
la 3,8(3)
bct 4,next enddo
* --------- epiloge -----------------------------------------------
exit lm 2,14,x'78'(13)
br 14
end
Loading