[Mpi3-rma] MPI RMW operations

[Shipman, Galen M.]

Rajeev,


On Mar 27, 2008, at 6:05 PM, Rajeev Thakur wrote:

> Some comments:
>
> * Are the semantics supposed to be that the original value at the  
> target is
> read and stored at the result_addr on the origin (like a Get), and the
> target memory is updated with the result of the reduction operation  
> (like
> Accumulate)?
Exactly, the value prior to the update is "returned" in the  
result_addr and the target memory is updated with the result of the  
operation.

>
> * What are the semantics of the two-operand versions?
>
For MPI_RMW2_COMP_SWAP:
operand_addr is the comparand, operand2_addr is the operand.  
Comparand is compared to the value at the target, if these values are  
equal then operand is written to the value at the target.

For MPI_RMW2_MASK_SWAP:
operand_addr is the maskarand, operand2_addr is the operand.  
Maskarand identifies the bits within the value at the target to be  
updated with the corresponding bits from operand.


> * Why is there no count, datatype at the target like in Accumulate?

This is intentional, datatypes and count must match on passive and  
active sides and are therefore implicit.


>
> * When assert == MPI_MODE_PASSIVE_EPOCH_IMPLICIT, what is the lock  
> type for
> the Win_lock -- shared or exclusive?
As all MPI_RMW/2 operations have the potential of updating the remote  
value and may therefore cause conflicts I would expect this to be an  
MPI_LOCK_EXCLUSIVE

> Also, Win_unlock takes only two
> parameters.
>
Yes, the example needs to be corrected.



Thanks for the comments,

Galen

> * Regarding the interaction with explicit epochs of Rmw when assert ==
> MPI_MODE_PASSIVE_EPOCH_IMPLICIT, it should be the same as the  
> interaction of
> its equivalent sequence, namely, Lock - Rmw - Unlock. In other  
> words, there
> shouldn't be another epoch around it on the same process (nested  
> epochs not
> allowed), and the epochs on other processes must not conflict.
>
> Rajeev
>
>
>> -----Original Message-----
>> From: mpi3-rma-bounces at lists.mpi-forum.org
>> [mailto:mpi3-rma-bounces at lists.mpi-forum.org] On Behalf Of
>> Shipman, Galen M.
>> Sent: Monday, March 24, 2008 10:10 PM
>> To: mpi3-rma at lists.mpi-forum.org
>> Subject: [Mpi3-rma] MPI RMW operations
>>
>>
>> Howard and I have been working on the RMW operations listed
>> below for
>> consideration.
>>
>> - Galen Shipman - ORNL
>>
>>
>> Single operand RMW operations
>>
>> MPI_Rmw(void *operand_addr,
>>                      int count,
>> 	            MPI_Datatype datatype,
>> 	            void *result_addr,
>> 	            int target_rank,
>> 	            MPI_Aint target_disp,
>> 	            int assert,
>> 	            MPI_RMW_Op op,
>>                      MPI_Win win)
>>
>>
>> Two operand RMW operations
>> 	
>> MPI_Rmw2(void *operand_addr,
>>                        int count,
>>                        MPI_Datatype datatype,
>> 	              void *operand2_addr,
>> 	              void *result_addr,
>> 	              int target_rank,
>> 	              MPI_Aint target_disp,
>>                        int assert,
>> 	              MPI_RMW_2_Op op,
>>                        MPI_Win win)
>>
>>
>>
>> MPI_RMW_Op Predefined RMW single operand operations
>> The following predefined operations are supplied for MPI_RMW
>> and related
>> functions. These operations are invoked  by placing the
>> following in op.
>>
>> Name		 Meaning
>>
>> MPI_RMW_INC		increment
>> MPI_RMW_PROD	product
>> MPI_RMW_SUM		sum
>> MPI_RMW_LAND	logical and
>> MPI_RMW_LOR		logical or
>> MPI_RMW_LXOR	logical xor
>> MPI_RMW_BAND	binary and
>> MPI_RMW_BOR		binary or
>> MPI_RMW_BXOR	binary xor
>> MPI_RMW_SWAP	swap value
>>
>> MPI_RMW_2_Op Predefined RMW two operand operations
>>
>> Name						Meaning
>>
>> MPI_RMW2_MASK_SWAP	swap masked bits
>> MPI_RMW2_COMP_SWAP	compare and swap
>>
>> Note: MPI_RMW2_COMP_SWAP is an equality comparison, we may wish to
>> consider multiple comparison operations such as >, >=, <, <=, ==, <>.
>>
>> Both MPI_Rmw and MPI_Rmw2 take an assert argument.
>> When assert == MPI_MODE_PASSIVE_EPOCH_IMPLICIT MPI_Rmw/2 is
>> equivalent to:
>>
>> MPI_Win_lock(target_rank,0,win);
>> MPI_Rmw(.....);
>> MPI_Win_unlock(target_rank,0,win);
>>
>> When assert  == MPI_MODE_PASSIVE_EPOCH_IMPLICIT | MPI_MODE_NOCHECK
>> the MPI library knows that the window lock at the target need not be
>> acquired.
>>
>>
>>
>>
>> _______________________________________________
>> mpi3-rma mailing list
>> mpi3-rma at lists.mpi-forum.org
>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
>>
>>
>
> _______________________________________________
> mpi3-rma mailing list
> mpi3-rma at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-rma/attachments/20080327/5c8e42ec/attachment-0001.html>