pbdMPI (version 0.5-2)

send-method: A Rank Send (blocking) an Object to the Other Rank

Description

This method lets a rank send (blocking) an object to the other rank in the same communicator. The default return is NULL.

Usage

send(x, rank.dest = .pbd_env$SPMD.CT$rank.dest,
     tag = .pbd_env$SPMD.CT$tag,
     comm = .pbd_env$SPMD.CT$comm,
     check.type = .pbd_env$SPMD.CT$check.type)

Value

A NULL is returned by default.

Arguments

x

an object to be sent from a rank.

rank.dest

a rank of destination where x send to.

tag

a tag number.

comm

a communicator number.

check.type

if checking data type first for handshaking.

Methods

For calling spmd.send.*():

signature(x = "ANY")

signature(x = "integer")

signature(x = "numeric")

signature(x = "raw")

Author

Wei-Chen Chen , George Ostrouchov, Drew Schmidt, Pragneshkumar Patel, and Hao Yu.

Details

A corresponding recv() should be evoked at the corresponding rank rank.dest.

These are high level S4 methods. By default, check.type is TRUE and an additional send()/recv() will make a handshaking call first, then deliver the data next. i.e. an integer vector of length two (type and length) will be deliver first between send() and recv() to ensure a buffer (of right type and right size/length) is properly allocated at the rank.dest side.

Currently, four data types are considered: integer, double, raw/byte, and default/raw.object. The default method will make a serialize() call first to convert the general R object into a raw vector before sending it away. After the raw vector is received at the rank.dest side, the vector will be unserialize() back to the R object format.

check.type set as FALSE will stop the additional handhsaking call, but the buffer should be prepared carefully by the user self. This is typically for the advanced users and more specifically calls are needed. i.e. calling those spmd.send.integer with spmd.recv.integer correspondingly.

check.type also needs to be set as FALSE for more efficient calls such as isend()/recv() or send()/irecv(). Currently, no check types are implemented in those mixed calls.

References

Programming with Big Data in R Website: https://pbdr.org/

See Also

isend(), recv(), irecv().

Examples

### Save code in a file "demo.r" and run with 2 processors by
### SHELL> mpiexec -np 2 Rscript demo.r

spmd.code <- "
### Initialize
suppressMessages(library(pbdMPI, quietly = TRUE))
.comm.size <- comm.size()
.comm.rank <- comm.rank()

### Examples.
N <- 5
x <- (1:N) + N * .comm.rank
if(.comm.rank == 0){
  y <- send(matrix(x, nrow = 1))
} else if(.comm.rank == 1){
  y <- recv()
}
comm.print(y, rank.print = 1)

### Finish.
finalize()
"
pbdMPI::execmpi(spmd.code, nranks = 2L)