September 2018 - mpiwg-rma - mailman3.cels.anl.gov

Canceled: RMA WG Biweekly Telecon
by Balaji, Pavan 14 Sep '18

14 Sep '18

Biweekly telecon using Blue Jeans: http://argonne.bluejeans.com/balaji.anl.gov

1 0

RMA WG Biweekly Telecon
by Balaji, Pavan 14 Sep '18

14 Sep '18

Biweekly telecon using Blue Jeans: http://argonne.bluejeans.com/balaji.anl.gov

1 0

Re: [mpiwg-rma] Atomicity and Pair Types
by maik peterson 13 Sep '18

13 Sep '18

shit, who let you guys out again ? still playing around with rma. that has really no future... Am Do., 13. Sep. 2018 um 16:09 Uhr schrieb Mark Hoemmen via mpiwg-rma < mpiwg-rma(a)lists.mpi-forum.org>: > Gmail suggests “sounds good to me” as a standard response and i’m ok with > that ;-). I think they were more worried about the case of a struct with > padding bits never making it through a CAS, even though the padding bits > aren’t part of the value representation. > > On Wed, Sep 12, 2018 at 10:40 PM Jeff Hammond <jeff.science(a)gmail.com> > wrote: > >> >> >> > On Sep 12, 2018, at 9:11 PM, Mark Hoemmen via mpiwg-rma < >> mpiwg-rma(a)lists.mpi-forum.org> wrote: >> > >> > True, it wouldn't affect MPI_COMPARE_AND_SWAP, since it only works on >> > integer, byte, or logical types. Never mind then :-) >> > >> > MPI_COMPARE_AND_SWAP uses language "if the compare buffer and the >> > target buffer are identical," which sounds like "bitwise identical" -- >> > the C++ paper I cited is trying to move away from that definition >> > towards operator==. It doesn't matter now but it might if people want >> > to extend it to floating-point types, with -0.0 == +0.0 etc. >> >> We will never do that because -0.0 and +0.0 are different numbers and if >> anybody tries to make the MPI standard say otherwise, I’m going to bludgeon >> them with a pillowcase full of numerical analysis textbooks. >> >> Jeff >> >> > mfh >> >> On Wed, Sep 12, 2018 at 9:42 PM Nathan Hjelm <hjelmn(a)me.com> wrote: >> >> >> >> Interesting but not quite relevant here. The struct types are only >> valid with minloc and maxloc so we don't have the same issue. The value of >> the padding bits does not influence the results of the operation. >> >> >> >> -Nathan >> >> >> >>> On Sep 12, 2018, at 8:43 PM, Mark Hoemmen via mpiwg-rma < >> mpiwg-rma(a)lists.mpi-forum.org> wrote: >> >>> >> >>> On Wed, Sep 12, 2018 at 2:55 PM Balaji, Pavan via mpiwg-rma >> >>> <mpiwg-rma(a)lists.mpi-forum.org> wrote: >> >>>> Yes, correct. Although, Nathan had a ticket at some point where he >> wanted to allow MPI to overwrite the padding bytes in such cases. >> >>> >> >>> This issue of padding bytes in atomic updates to a struct came up in >> >>> the past few C++ meetings: >> >>> >> >>> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0528r3.html >> >>> >> >>> For more explanation, see >> >>> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0528r0.html >> . >> >>> The goal is that compare-and-exchange shouldn't always return false if >> >>> the padding bits don't participate in the value representation of a >> >>> struct. >> >>> >> >>> mfh >> >>> _______________________________________________ >> >>> mpiwg-rma mailing list >> >>> mpiwg-rma(a)lists.mpi-forum.org >> >>> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma >> > _______________________________________________ >> > mpiwg-rma mailing list >> > mpiwg-rma(a)lists.mpi-forum.org >> > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma >> > _______________________________________________ > mpiwg-rma mailing list > mpiwg-rma(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma >

1 0

Re: [mpiwg-rma] Atomicity and Pair Types
by Mark Hoemmen 13 Sep '18

13 Sep '18

Gmail suggests “sounds good to me” as a standard response and i’m ok with that ;-). I think they were more worried about the case of a struct with padding bits never making it through a CAS, even though the padding bits aren’t part of the value representation. On Wed, Sep 12, 2018 at 10:40 PM Jeff Hammond <jeff.science(a)gmail.com> wrote: > > > > On Sep 12, 2018, at 9:11 PM, Mark Hoemmen via mpiwg-rma < > mpiwg-rma(a)lists.mpi-forum.org> wrote: > > > > True, it wouldn't affect MPI_COMPARE_AND_SWAP, since it only works on > > integer, byte, or logical types. Never mind then :-) > > > > MPI_COMPARE_AND_SWAP uses language "if the compare buffer and the > > target buffer are identical," which sounds like "bitwise identical" -- > > the C++ paper I cited is trying to move away from that definition > > towards operator==. It doesn't matter now but it might if people want > > to extend it to floating-point types, with -0.0 == +0.0 etc. > > We will never do that because -0.0 and +0.0 are different numbers and if > anybody tries to make the MPI standard say otherwise, I’m going to bludgeon > them with a pillowcase full of numerical analysis textbooks. > > Jeff > > > mfh > >> On Wed, Sep 12, 2018 at 9:42 PM Nathan Hjelm <hjelmn(a)me.com> wrote: > >> > >> Interesting but not quite relevant here. The struct types are only > valid with minloc and maxloc so we don't have the same issue. The value of > the padding bits does not influence the results of the operation. > >> > >> -Nathan > >> > >>> On Sep 12, 2018, at 8:43 PM, Mark Hoemmen via mpiwg-rma < > mpiwg-rma(a)lists.mpi-forum.org> wrote: > >>> > >>> On Wed, Sep 12, 2018 at 2:55 PM Balaji, Pavan via mpiwg-rma > >>> <mpiwg-rma(a)lists.mpi-forum.org> wrote: > >>>> Yes, correct. Although, Nathan had a ticket at some point where he > wanted to allow MPI to overwrite the padding bytes in such cases. > >>> > >>> This issue of padding bytes in atomic updates to a struct came up in > >>> the past few C++ meetings: > >>> > >>> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0528r3.html > >>> > >>> For more explanation, see > >>> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0528r0.html . > >>> The goal is that compare-and-exchange shouldn't always return false if > >>> the padding bits don't participate in the value representation of a > >>> struct. > >>> > >>> mfh > >>> _______________________________________________ > >>> mpiwg-rma mailing list > >>> mpiwg-rma(a)lists.mpi-forum.org > >>> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma > > _______________________________________________ > > mpiwg-rma mailing list > > mpiwg-rma(a)lists.mpi-forum.org > > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma >

1 0

Re: [mpiwg-rma] Atomicity and Pair Types
by Jeff Hammond 12 Sep '18

12 Sep '18

> On Sep 12, 2018, at 9:11 PM, Mark Hoemmen via mpiwg-rma <mpiwg-rma(a)lists.mpi-forum.org> wrote: > > True, it wouldn't affect MPI_COMPARE_AND_SWAP, since it only works on > integer, byte, or logical types. Never mind then :-) > > MPI_COMPARE_AND_SWAP uses language "if the compare buffer and the > target buffer are identical," which sounds like "bitwise identical" -- > the C++ paper I cited is trying to move away from that definition > towards operator==. It doesn't matter now but it might if people want > to extend it to floating-point types, with -0.0 == +0.0 etc. We will never do that because -0.0 and +0.0 are different numbers and if anybody tries to make the MPI standard say otherwise, I’m going to bludgeon them with a pillowcase full of numerical analysis textbooks. Jeff > mfh >> On Wed, Sep 12, 2018 at 9:42 PM Nathan Hjelm <hjelmn(a)me.com> wrote: >> >> Interesting but not quite relevant here. The struct types are only valid with minloc and maxloc so we don't have the same issue. The value of the padding bits does not influence the results of the operation. >> >> -Nathan >> >>> On Sep 12, 2018, at 8:43 PM, Mark Hoemmen via mpiwg-rma <mpiwg-rma(a)lists.mpi-forum.org> wrote: >>> >>> On Wed, Sep 12, 2018 at 2:55 PM Balaji, Pavan via mpiwg-rma >>> <mpiwg-rma(a)lists.mpi-forum.org> wrote: >>>> Yes, correct. Although, Nathan had a ticket at some point where he wanted to allow MPI to overwrite the padding bytes in such cases. >>> >>> This issue of padding bytes in atomic updates to a struct came up in >>> the past few C++ meetings: >>> >>> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0528r3.html >>> >>> For more explanation, see >>> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0528r0.html . >>> The goal is that compare-and-exchange shouldn't always return false if >>> the padding bits don't participate in the value representation of a >>> struct. >>> >>> mfh >>> _______________________________________________ >>> mpiwg-rma mailing list >>> mpiwg-rma(a)lists.mpi-forum.org >>> https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma > _______________________________________________ > mpiwg-rma mailing list > mpiwg-rma(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma

1 0

Re: [mpiwg-rma] Atomicity and Pair Types
by Mark Hoemmen 12 Sep '18

12 Sep '18

True, it wouldn't affect MPI_COMPARE_AND_SWAP, since it only works on integer, byte, or logical types. Never mind then :-) MPI_COMPARE_AND_SWAP uses language "if the compare buffer and the target buffer are identical," which sounds like "bitwise identical" -- the C++ paper I cited is trying to move away from that definition towards operator==. It doesn't matter now but it might if people want to extend it to floating-point types, with -0.0 == +0.0 etc. mfh On Wed, Sep 12, 2018 at 9:42 PM Nathan Hjelm <hjelmn(a)me.com> wrote: > > Interesting but not quite relevant here. The struct types are only valid with minloc and maxloc so we don't have the same issue. The value of the padding bits does not influence the results of the operation. > > -Nathan > > > On Sep 12, 2018, at 8:43 PM, Mark Hoemmen via mpiwg-rma <mpiwg-rma(a)lists.mpi-forum.org> wrote: > > > > On Wed, Sep 12, 2018 at 2:55 PM Balaji, Pavan via mpiwg-rma > > <mpiwg-rma(a)lists.mpi-forum.org> wrote: > >> Yes, correct. Although, Nathan had a ticket at some point where he wanted to allow MPI to overwrite the padding bytes in such cases. > > > > This issue of padding bytes in atomic updates to a struct came up in > > the past few C++ meetings: > > > > http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0528r3.html > > > > For more explanation, see > > http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0528r0.html . > > The goal is that compare-and-exchange shouldn't always return false if > > the padding bits don't participate in the value representation of a > > struct. > > > > mfh > > _______________________________________________ > > mpiwg-rma mailing list > > mpiwg-rma(a)lists.mpi-forum.org > > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma

1 0

Re: [mpiwg-rma] Atomicity and Pair Types
by Mark Hoemmen 12 Sep '18

12 Sep '18

On Wed, Sep 12, 2018 at 2:55 PM Balaji, Pavan via mpiwg-rma <mpiwg-rma(a)lists.mpi-forum.org> wrote: > Yes, correct. Although, Nathan had a ticket at some point where he wanted to allow MPI to overwrite the padding bytes in such cases. This issue of padding bytes in atomic updates to a struct came up in the past few C++ meetings: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0528r3.html For more explanation, see http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0528r0.html . The goal is that compare-and-exchange shouldn't always return false if the padding bits don't participate in the value representation of a struct. mfh

1 0

Re: [mpiwg-rma] Atomicity and Pair Types
by Balaji, Pavan 12 Sep '18

12 Sep '18

Yes, correct. Although, Nathan had a ticket at some point where he wanted to allow MPI to overwrite the padding bytes in such cases. -- Pavan > On Sep 12, 2018, at 3:49 PM, Jim Dinan via mpiwg-rma <mpiwg-rma(a)lists.mpi-forum.org> wrote: > > Hi All, > > Sanity check --- my reading of the spec. indicates that accumulate operations performed on pair types guarantee atomicity at the level of the whole pair type. Here are the relevant passages: > > Section 11.7.1: The outcome of concurrent accumulate operations to the same location with the same predefined datatype is as if the accumulates were done at that location in some serial order. > > Section 5.9.4: In order to use MPI_MINLOC and MPI_MAXLOC in a reduce operation, one must provide a datatype argument that represents a pair (value and index). MPI provides nine such predefined datatypes. > > Is that right? > > Thanks, > ~Jim. > _______________________________________________ > mpiwg-rma mailing list > mpiwg-rma(a)lists.mpi-forum.org > https://lists.mpi-forum.org/mailman/listinfo/mpiwg-rma

1 0

Atomicity and Pair Types
by Jim Dinan 12 Sep '18

12 Sep '18

Hi All, Sanity check --- my reading of the spec. indicates that accumulate operations performed on pair types guarantee atomicity at the level of the whole pair type. Here are the relevant passages: Section 11.7.1: The outcome of concurrent accumulate operations to the same location with the same *predefined* datatype is as if the accumulates were done at that location in some serial order. Section 5.9.4: In order to use MPI_MINLOC and MPI_MAXLOC in a reduce operation, one must provide a datatype argument that represents a pair (value and index). MPI provides nine such *predefined* datatypes. Is that right? Thanks, ~Jim.

1 0

Re: [mpiwg-rma] RMA WG Telecon
by Sameh S Sharkawi 12 Sep '18

12 Sep '18

1 0