From aoyama@XXXXXXXX Thu Dec 7 02:43:01 2006 X-Mailer: QUALCOMM Windows Eudora Version 5.0.2-J Date: Thu, 07 Dec 2006 17:42:50 +0900 To: William Gropp From: Yukiya Aoyama Subject: Re: SEND COUNT and RECEIVE COUNT In-Reply-To: References: <5.0.2.5.2.20061207162618.04c34650@XXXXXXXXXXXXXXXX> <5.0.2.5.2.20061207162618.04c34650@XXXXXXXXXXXXXXXX>Mime-Version: 1.0 Content-Type: text/plain; charset="ISO-2022-JP"; format=flowed Content-Transfer-Encoding: 7bit X-Virus-Scanned: by amavisd-new-20030616-p10 (Debian) at mailgw.mcs.anl.gov Hi. Thanks for your quick answer. I understand. I have one comment. In MPI-1 document, I think the following 「」 part is a little ambiguous. It may seem ' receive count is greater than send count ' is OK. MPI_ALLTOALLV recvcounts integer array equal to the group size specifying the number of elements 「that can be received」 from each processor. Best Regards, Yukiya Aoyama At 02:30 06/12/07 -0600, you wrote: >In point-to-point operations, it is ok to provide a larger receive >buffer as in your example (assuming that the datatypes are the >same). Note that what is really important is the pair (count, >datatype), and the receive buffer specified by that must be at least >as large as the send buffer (with the necessary modifications for >heterogeneity). > >In the collective cases, the (sendcount,datatype) must describe a >buffer that matches the relevant (recvcount,datatype) exactly. Note >that if the datatypes are not the same, for example a basic type for >a send and a vector type for a receive, the send and receive counts >will be different. > >Bill > > >On Dec 7, 2006, at 1:42 AM, Yukiya Aoyama wrote: > >> >> >>Hi. >> >>I belong to RIKEN Japan. >> >>I have a question about send count and recv count. >> >>In MPI_SEND and MPI_RECV, the following is OK. >> In Rank 0, MPI_SEND to Rank1, sendcount is 3 >> In Rank 1, MPI_RECV from Rank 0 recvcount is 10 >> >>How about Collective communication ? >>(suppose send type and receive type are the same) >> >>For example, in MPI_ALLTOALLV, >> In Rank 0, sendcount to Rank 1 is 3 >> In Rank 1, recvcount from Rank 0 is 10 >>Is this specification correct? >> >>According to MPI-1 document Section 4,4, I think the above is wrong >>and sendcount and recvcount must be the same. >> > > ********************************** 青山 幸也 (Yukiya Aoyama) 理化学研究所 情報基盤センター 電子メール:aoyama@XXXXXXXX 住所:〒351-0198 埼玉県和光市広沢2-1 電話:(外線)048-467-4982 (内線)3822 FAX :048-462-4634 **********************************