I am trying to implement the LU factorization of the vandermonde matrix with OpenMPI.
Therefore, the matrix is split onto the processors in a cycle-wise manner, e.g. processor 0 gets rows 0, 3, 6, 9 and so on.
In iteration k, every process needs to use row k. The corresponding processor should send it already at the end of iteration k-1 with a non-blocking broadcast, and any other processor should receive it at the beginning of iteration k.
What I did:
Row 0 is send before iterating over k by Process 0.
At the beginning of iteration k:
int source = k%size;
if (source != rank)
{
printf("Process %d waits for row %d from process %d\n", rank, k, source);
MPI_Irecv(bcast_buffer, n, MPI_DOUBLE, source, k, MPI_COMM_WORLD, &request);
MPI_Wait(&request, &status);
printf("Process %d received row %d from process %d and continues\n", rank, k, source);
}
And the end of iteration k (l2gl is local to global for row indices)
// if current row (i) is row k+1
if((l2gl[i]==k+1) && (l2gl[i]!=n-1))
{
// Safe row to broadcast buffer and send it to any other process
for(tmp=0; tmp<n; tmp++)
{
bcast_buffer[tmp] = A[i][tmp];
}
for(int r=0; r<size; r++)
{
printf("Process %d sends row %d to process %d\n", rank, k+1, r);
MPI_Isend(bcast_buffer, n, MPI_DOUBLE, r, k+1, MPI_COMM_WORLD, &request);
}
}
In the output I find the following:
Process 3 received row 6 from Process 2 and continues
Process 2 received row 5 from Process 1 and continues
Process 2 sends row 6 to Process 0
Process 2 sends row 6 to Process 1
Process 2 sends row 6 to Process 2
Process 2 sends row 6 to Process 3
Does it definitely mean that I have an error in communication or may the prints be in a 'wrong' order due to delays of the print output or whatever?
As the result of the factorization is wrong, there has to be a mistake, either in the communication or the calculation.
I appreciate any help.
Related
Suppose that I have a code block that attempts to calculate the first 15 numbers in the Fibonacci sequence and distributes each unique number among 3 processes (MPI_Send) using a for loop as shown in the code block below.
int main(int argc, char* argv[]) {
int rank, size, recieve_data1, recieve_data2;
MPI_Init(NULL, NULL);
MPI_Status status;
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &size);
printf("Available ranks are: %d \n \n", rank); // first rank rollcall
fflush(stdout);
int num1 = 1; int num2 = 1;
int RecieveNum; int SumNum;
for
(int n = 0; n < 16; n++) {
if
(rank == 0) {
// perform the fibb sequence algorithim
SumNum = num1 + num2;
num1 = num2;
num2 = SumNum;
// define the sorting algorithim
int DeliverTo = (n % 3) + 1;
// send calculated result
MPI_Send(&SumNum, 1, MPI_INT, DeliverTo, 1, MPI_COMM_WORLD);
}
else {
// recieve the element integer
MPI_Recv(&RecieveNum, 1, MPI_INT, 0, 1, MPI_COMM_WORLD, &status);
// print and flush the buffer
printf("I am process rank %d, and I recieved the number %d. \n", rank, RecieveNum);
fflush(stdout);
}
}
printf("Available ranks are: %d \n \n", rank); // second rank roll call
fflush(stdout);
/*
more code that I run...
*/
MPI_Finalize();
return 0;
}
Before the first for loop is called, processes 0,1,2,3 all respond to the printf(Available ranks are: %d \n \n", rank); on line 25. However, after executing the first for loop and using a second printf, only process 0 responds. I was expecting all 4 processes 0 - 3 to respond again after the execution of the for loop. To solve this problem, I isolated this section of code and attempted to debug for several hours with no sucess. This particular issue proves to be problematic, as I have additional code (not shown here for the sake of being concise), that will access the numbers generated from this sequence.
Finally, I am running the code by building the solution, running the VS terminal as an administrator, and typing mpiexec -n 4 my_file_name.exe. No build errors or complication mistakes occurred. From what I can see (correct me if I'm wrong), all processes hang after completing the for loop, but I am unsure why or how to fix it.
After searching the website, I did not see anything that answered this question (from my point of view). I am a bit of an MPI (and Stack Overflow) newbie, so any code pointers are also welcomed. Thanks
You have process zero compute who to send to. And then everyone does a receive. That means that all processes that are not the computed receiver will hang.
This scenario where you send to a dynamically computed receiver is not easy to do in MPI. You either need to
send a message to all other processes "no, I have nothing for you", or
send a message to all, and all-but-one process ignores the data, or
use one sided operations where you MPI_Put the data on the computer receiver.
I'm implementing the Chan and Dehne sorting algorithm using MPI and the CGM realistic parallel model. So far each process receives N/p numbers from the original vector, each process then order their numbers sequentially using quick sort, each process then creates a sample from it's local vector (the sample has size p), each process then sends their sample over to P0; P0 should receive all samples in a bigger vector of size p*p so it can accommodate data from all processors. This is where I'm stuck, it seems to be working but for some reason after P0 receives all the data it exits with Signal: Segmentation fault (11). Thank you.
Here is the relevant part of the code:
// Step 2. Each process calculates it's local sample with size comm_sz
local_sample = create_local_sample(sub_vec, n_over_p, comm_sz);
// Step 3. Each process sends it's local sample to P0
if (my_rank == 0) {
global_sample_receiver = (int*)malloc(pow(comm_sz,2)*sizeof(int));
global_sample_receiver = local_sample;
for (i = 1; i < comm_sz; i++) {
MPI_Recv(global_sample_receiver+(i*comm_sz), comm_sz, MPI_INT,
i, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
}
} else {
MPI_Send(local_sample, comm_sz, MPI_INT, 0, 0, MPI_COMM_WORLD);
}
printf("P%d got here\n", my_rank);
MPI_Finalize();
What is funny is that every process reachs the command printf("P%d got here\n", my_rank); and therefor prints to the terminal. Also global_sample_receiver does contain the data it is supposed to contain at the end, but the program still finished with a segmentation fault.
Here is the output:
P2 got here
P0 got here
P3 got here
P1 got here
[Krabbe-Ubuntu:05969] *** Process received signal ***
[Krabbe-Ubuntu:05969] Signal: Segmentation fault (11)
[Krabbe-Ubuntu:05969] Signal code: Address not mapped (1)
[Krabbe-Ubuntu:05969] Failing at address: 0x18000003e7
--------------------------------------------------------------------------
mpiexec noticed that process rank 0 with PID 5969 on node Krabbe-Ubuntu
exited on signal 11 (Segmentation fault).
--------------------------------------------------------------------------
Edit: I found the problem, turns out local_sample also needed a malloc.
The issue is you overwrite global_sample_receiver (which is a pointer) with local_sample (which is an other pointer) on rank zero.
If you want to set the first comm_sz elements of global_sample_receiver with the first comm_sz elements from local_sample, then you have to copy the data (e.g. not the pointer) manually.
memcpy(global_sample_receiver, local_sample, comm_sz * sizeof(int));
That being said, the natural MPI way of doing this is via MPI_Gather().
Here is what step 3 would look like :
// Step 3. Each process sends it's local sample to P0
if (my_rank == 0) {
global_sample_receiver = (int*)malloc(pow(comm_sz,2)*sizeof(int));
}
MPI_Gather(global_sample_receiver,comm_sz, MPI_INT, local_sample, comm_sz, MPI_INT, 0, MPI_COMM_WORLD);
This is what I am trying to achieve.
Blue is the message.
Yellow is when the specific node changes the leader known to it.
Green is the final election of each node.
The code seems correct to me but it's always stuck inside the while loop no matter what I tried. For a small number of nodes during runtime it returns a segmentation fault after a while.
election_status=0;
firstmsg[0]=world_rank; // self rank
firstmsg[1]=0; // counter for hops
chief=world_rank; // each node declares himself as leader
counter=0; // message counter for each node
// each node sends the first message to the next one
MPI_Send(&firstmsg, 2, MPI_INT, (world_rank+1)%world_size, 1, MPI_COMM_WORLD);
printf("Sent ID with counter to the right node [%d -> %d]\n",world_rank, (world_rank+1)%world_size);
while (election_status==0){
// EDIT: Split MPI_Recv for rank 0 and rest
if (world_rank==0){
MPI_Recv(&incoming, 2, MPI_INT, world_size-1, 1, MPI_COMM_WORLD, &status);
}
else {
MPI_Recv(&incoming, 2, MPI_INT, (world_rank-1)%world_size, 1, MPI_COMM_WORLD, &status);
}
counter=counter+1;
if (incoming[0]<chief){
chief=incoming[0];
}
incoming[1]=incoming[1]+1;
// if message is my own and hopped same times as counter
if (incoming[0]==world_rank && incoming[1]==counter) {
printf("Node %d declares node %d a leader.\n", world_rank, chief);
election_status=1;
}
//sends the incremented message to the next node
MPI_Send(&incoming, 2, MPI_INT, (world_rank+1)%world_size, 1, MPI_COMM_WORLD);
}
MPI_Finalize();
In order to determine some minimum number among a number of ranks for all ranks, use MPI_Allreduce!
MPI_Send is blocking. It can block forever until a matching receive is posted. Your program deadlocks on the first call to MPI_Send - and any successive once should it complete by coincidence. To avoid that specifically use MPI_Sendrecv.
(world_rank-1)%world_size will produce -1 for world_rank == 0. Using -1 as rank number is not valid. It might coincidentially be MPI_ANY_SOURCE.
#include <mpi.h>
#include <stdlib.h>
#include <stdio.h>
int main(int argc, char *argv[])
{
int myid, numprocs, number_of_completed_operation;
char message = 'a';
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &myid);
MPI_Request* requests = (MPI_Request*)malloc((numprocs - 1)*sizeof(MPI_Request));
MPI_Status* statuses = (MPI_Status*)malloc(sizeof(MPI_Status)*(numprocs - 1));
int* indices = (int *)malloc((numprocs - 1)*sizeof(int));
char* buf = (char *)malloc((numprocs - 1)*sizeof(char));
if (myid != numprocs - 1)
{//worker
printf("***this is sender %d\n", myid);
MPI_Send(&message, 1, MPI_CHAR, numprocs - 1, 110, MPI_COMM_WORLD);
printf("*.*sender %d is done\n", myid);
}
else if (myid == numprocs - 1)
{
//master
int number_of_left_messages = numprocs - 1;//有numprocs-1个信息到来
int i;
for (i = 0; i < numprocs - 1; i++)
{
MPI_Irecv(&buf+i, 1, MPI_CHAR,i, 110, MPI_COMM_WORLD, &requests[i]);
}
MPI_Waitsome(numprocs - 1, requests, &number_of_completed_operation, indices, statuses);
number_of_left_messages = number_of_left_messages - number_of_completed_operation;
printf("number of completed operation is %d\n", number_of_left_messages);
printf("left message amount is %d\n", number_of_left_messages);
int j;
for (j = 0; j <numprocs - 1; j++)
{
printf("-------------\n");
printf("index is %d\n",indices[j]);
printf("source is %d\n", statuses[j].MPI_SOURCE);
//printf("good\n");
printf("--------====\n");
}
while (number_of_left_messages > 0)
{
MPI_Waitsome(numprocs - 1, requests, &number_of_completed_operation, indices, statuses);
printf("number of completed operation is %d\n", number_of_completed_operation);
for (j = 0; j <numprocs - 1; j++)
{
printf("-------------\n");
printf("index is %d\n", indices[j]);
printf("source is %d\n", statuses[j].MPI_SOURCE);
printf("--------====\n");
}
number_of_left_messages = number_of_left_messages - number_of_completed_operation;
printf("left message amount is %d\n", number_of_left_messages);
The logic is simple,I set the final process as the master process,all the other process are worker process,the workers send a message to the master,the master use the waitsome function to receive.
When I set the number of processes as 4 or larger, the system shown me the error as following:
[soit-mpi-pro-1:12197] *** An error occurred in MPI_Waitsome
[soit-mpi-pro-1:12197] *** reported by process [140533176729601,140531329925123]
[soit-mpi-pro-1:12197] *** on communicator MPI_COMM_WORLD
[soit-mpi-pro-1:12197] *** MPI_ERR_REQUEST: invalid request
[soit-mpi-pro-1:12197] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
[soit-mpi-pro-1:12197] *** and potentially your MPI job)
It looks like your call to MPI_Irecv might be a problem. Remove the extra & before the buf (you have a double pointer instead of a pointer).
MPI_Irecv(buf+i, 1, MPI_CHAR,i, 110, MPI_COMM_WORLD, &requests[i]);
When I fix that, add closing braces and a call to MPI_Finalize(), and remove a bunch of extra output, I don't have any issues running your program:
$ mpiexec -n 8 ./a.out
***this is sender 3
*.*sender 3 is done
***this is sender 4
*.*sender 4 is done
***this is sender 5
*.*sender 5 is done
***this is sender 6
*.*sender 6 is done
***this is sender 0
*.*sender 0 is done
***this is sender 1
*.*sender 1 is done
***this is sender 2
*.*sender 2 is done
number of completed operation is 1
left message amount is 6
number of completed operation is 1
left message amount is 5
number of completed operation is 1
left message amount is 4
number of completed operation is 1
left message amount is 3
number of completed operation is 1
left message amount is 2
number of completed operation is 1
left message amount is 1
number of completed operation is 1
left message amount is 0
I have no idea if it gets the right answer or not, but that's a different question.
You are passing MPI_Irecv the address of the pointer buf itself plus an offset instead of its value. When the message is received, it overwrites the last byte (on little endian systems like x86/x64) of the value of one or more nearby stack variables, which, depending on the stack layout, might include requests and statuses. Therefore MPI_Waitsome receives a pointer that doesn't point to beginning of the array of requests but rather somewhere before it, after it or in the middle of it, hence some of the request handles are invalid and MPI_Waitsome complains. On a big endian system, this would overwite the highest byte of the address and will much rather result in an invalid address and a segmentation fault.
Either use buf+i (as per Wesley Bland's answer) or use &buf[i]. I usually find it a matter of personal taste whether one uses the first of the second form.
What I have
I have a C-program using MPI, and it uses 4 processes: 1 vehicle(taskid=0) and 3 passengers.
Vehicle can accommodate 2 passengers at a time.
3 customers keep coming back to get a ride.
For Vehicle, I have:
int passengers[C] = {0};
while(1)
MPI_Recv(&pid, 1, MPI_INT, MPI_ANY_SOURCE, 1, MPI_COMM_WORLD, &status);
//put pid in passengers[totalNumberArrived]
if(totalNumberArrived == 2)
printf("vehicle left...");
sleep(5);
printf("vehicle came back...");
for (i=0; i<2; i++)
MPI_Send(&passengers[i], 1, MPI_INT, passengers[i], 1, MPI_COMM_WORLD);
totalNumberArrived = 0;
if(done)//omitting the details here
break;
And, for each passengers, I have:
for (i to NumOfRound)
sleep(X);
printf("%d is sending a msg", tasked)
MPI_Send(&taskid, 1, MPI_INT, 0, 1, MPI_COMM_WORLD);
MPI_Recv(&pid, 1, MPI_INT, 0, 1, MPI_COMM_WORLD, &pstatus);
printf("%d received from %d\n", tasked, pid, pstatus.MPI_SOURCE);
issue
If the vehicle left for a ride with taskid 1 and 3, I expect to see this kind of output:
vehicle left...
vehicle came back...
1 is sending a msg
3 is sending a msg (this could be before 1's msg though)
but I sometimes get
vehicle left...
1 is sending a msg
3 is sending a msg
vehicle came back...
which looks like the passenger is not blocked until the vehicle comes back.
I thought that MPI_Recv blocks the task until it gets a msg from the vehicle, so I researched and read that MPI_Recv does block but this kind of issues occur because the printf is not necessarily printing in order. I also read that some recommends to use flush but in some cases flush doesn't work.
I'm not sure what I should do in my case. Is it really just the matter of printf order?
I've also read this: Ordering Output in MPI
and wonder if I should add a master thread and let it be the central controller for vehicle and passengers??
You can't rely on printing output to be ordered between processes. The only thing you can count on is that output will be ordered per processes. Therefore, if for some reason it's critical that you can print things to STDOUT/STDERR in order, you need to aggregate it to one process first.