To: J3                                                     J3/18-205r1
From: Steve Lionel & Dan Nagle & John Reid
Subject: Response to DIS comments GB043, 044, 045 (RANDOM_INIT)
Date: 2018-June-13

Reference: 18-007,14-184r4

Discussion
----------

DIS comment GB043 says:

"Subclause 16.7, para 4 says "It is processor
dependent whether each image uses a separate
random number generator, or if some or all images
use common random number generators." We
believe that the reason for this wording is to allow
implementations to use a separate seed on each
image, a single seed on each set of images
(perhaps those on single nodes) or a single seed
on all images (perhaps on a shared-memory
machine with a modest number of images).

The present wording for IMAGE_DISTINCT makes
sense only for the case of a separate seed on
each image - which image of a set sharing a seed
calls RANDOM_INIT has no bearing. Does a single
call of RANDOM_INIT initialize all the seeds? The a
current wording suggests that only one seed is
initialized. Subclause 16.7, para 4 also says "If
RANDOM_INIT or RANDOM_SEED is called in a
segment A, and RANDOM_INIT,
RANDOM_SEED, or RANDOM_NUMBER is called
in segment B, then segments A and B shall be
ordered." If only one seed is initialized, the calls
must be in a CRITICAL construct or its equivalent,
which would be very inefficient. It would be much
better to say that a single call initializes all the
seeds and that it does not matter which image
makes the call."

After extensive discussion, JOR recognizes that the current definition
of RANDOM_INIT is unworkable. Edits are supplied that provide the
following:

- Each image in a team that wishes to "participate" in a shared behavior
of RANDOM_NUMBER must call RANDOM_INIT with the same values for its
arguments. In this sense it is similar to a collective subroutine.
- All words that relate to implementation details ("seed" or "common
generator" are removed. Instead the text describes the sequences
obtained by calls to RANDOM_NUMBER after RANDOM_INIT is called. A
processor may choose to use a common generator in the cases where
REPEATABLE is false, but a program has no way of detecting this.
- The text in 16.7 about image segment ordering is removed, as it is
no longer relevant.

Comment GB044 says:

"In the definition of argument REPEATABLE, the last sentence says "If
it has the value false, the seed is set to a processor-dependent,
unpredictably different value on each call." The inclusion of the word
"unpredictably" has the effect of forbidding a vendor from documenting
how their pseudorandom number generator works. We believe that this
was not intended. "

JOR does not agree with this interpretation, and believes no such
restriction is implied.

Comment GB045 says:

"In the current definition [of the argument REPEATABLE], it is unclear
whether the value of the  seed is the same only for the current
execution or is repeatable over different runs on the same processor
with the same number of images. We think the latter was intended,
which is very useful when testing or verifying code."

We agree that this was the intention, but there remains the possibility
that changes in the execution environment may affect the sequence. An edit
and a note are added to address this.

Edits
-----

16.7p4, page 348 Standard Generic Intrinsic procedures

Replace the paragraph "If RANDOM_INIT ... generators" with:

"The pseudorandom number generator on each image has a state independent
of the state of that on any other image.  References to RANDOM_INIT,
RANDOM_NUMBER, and RANDOM_SEED may update this state and no other image's
pseudorandom number generator's state.  The state is the sole determiner,
per image, of the sequence of pseudorandom numbers generated."

16.9.155p3, page 416, RANDOM_INIT (Arguments)

In the description of REPEATABLE, replace the text "If it has the value
... on each call." with:

"If it has the value true, subsequent calls to RANDOM_NUMBER in the
current image will generate the same sequence of pseudorandom numbers
each time RANDOM_INIT is called for a given number of images in the
current team. If it has the value false, subsequent calls to
RANDOM_NUMBER will select from an unpredictably different sequence
after each call to RANDOM_INIT."

In the description of IMAGE_DISTINCT, replace the text "If it has the
value true ... calls RANDOM_INIT."

with:

"If it has the value true, subsequent calls to RANDOM_NUMBER in the
current image will produce a sequence of pseudorandom numbers that
is dependent on the image index of the current image.
If it has the value false, subsequent calls to RANDOM_NUMBER in the
current image will select from a sequence of pseudorandom numbers that
does not depend directly on the image index of the current image."

16.9.155p4, page 416, RANDOM_INIT (Example)

In paragraph 4 (Example) replace the sentence by "The following statement
initializes the pseudorandom number generator of the current image so that
the pseudorandom number sequence is different from those established by
earlier calls to RANDOM_INIT and depends on the image index of the current
image."

After paragraph 4, add a note:

"Changes in a program's execution environment may affect the
repeatability of pseudorandom number sequences established by
a call to RANDOM_INIT."

16.9.156p3, Page 416, RANDOM_NUMBER (Arguments)

In the description of HARVEST delete the last sentence
"If images use . . . processor dependent."

[535]
A.2p1, page 535, Processor Dependencies

Delete the bullet "on images that use . . . unordered segments (16.7)."

==END==