IBM System/370 Manual page 17

• Extended Precision Floating Point
Precision of up to 28 hexadecimal digits, approximately equal to
34 decimal digits, is provided by the extended precision data
format. Extended precision is achieved by using two doublewords
(16 bytes) to represent an extended precision floating-point number
instead of using one doubleword as is done in long form
representation. Fourteen hexadecimal digits, or up to 17 decimal
digits, of precision is provided by the long floating-point format.
Seven floating-point instructions are included in the extended
preciSion feature. They provide addition, subtraction, and
multiplication operations for extended precision data, using a
pair of floating-point registers, and the ability to round from
long to short form or from extended to long form.
• General Purpose Instructions
Six general purpose instructions, which will be of benefit to both
control and processing program performance, have been added to
the Model 165 standard instruction set.
SHIFT AND ROUND DECIMAL, using a single instruction, provides right
or left shifting of packed decimal data. This instruction can
save 6 to 18 bytes of instruction storage and instruction execution
time for each decimal shift and round operation performed in
commercial processing.
MOVE LONG provides for the movement of up to 16 million bytes from
one location in storage to another with a single instruction,
thereby removing the current limitation of 256 bytes per move.
This instruction can eliminate the necessity of multiple move
instructions or the inclusion of move subroutines. The format
and operation of MOVE LONG facilitates efficient record blocking
and deblocking, field padding, and storage clearing, operations
frequently performed in commercial processing.
COMPARE LOGICAL LONG can be used to compare logically two fields
of up to 16 million bytes in length, thus removing the current
256-byte limit on byte compares. In addition, when an unequal
compare occurs, the two characters that caused the inequality are
identified.
The MOVE LONG and COMPARE LOGICAL LONG instructions are
interruptable. Thus, when an I/O operation terminates during their
execution, the interrupt can be taken, and the channel is not held
up awaiting termination of what might be a lengthy move or compare.
COMPARE LOGICAL, STORE, and INSERT CHARACTERS UNDER MASK
instructions provide byte addressability within the general purpose
registers and permit nonword-size data that is not on a word
boundary to be compared to data in a register, loaded into a
register~ and stored from a register. These three instructions
can be of most benefit to control program programmers, to compiler
writers, and to others who must manipulate processor storage
addresses.
High-Speed Multiply Feature (Optional)
This feature substantially increases the internal performance of
fixed- and floating-point multiply operations. With use of this feature
a long precision floating-point multiply operation improves from 1.81
to .61 microseconds. A fixed-point mUltiply operation improves from
.78 to .42 microseconds.
10