I am trying to send a large array which half of its values are calculated by the child process and the other half is calculated by the parent process. I would like the parent process to obtain the calculated values for half of the array from the child process and merge/join both arrays from both processes together so that I have one final array which is the product of the child and parent array calculations added together. I am trying to use the piping method only. My current code uses a "dummy" array which is the determined array calculations for the child process. This array is then passed into the parent process through piping which allows for the Test->iterations array results from the parent process to be added all together. The array is defined in a structure called Test which is in the format of:
typedef struct{
int *iterations
int height;
int width;
int start;
int end;
} Test;
Since this is a 2D array, the size of iterations is Test->width * Test->height * sizeof(int)). I am also sending the array using chunk read and write functions since it is a very large array. My parallel program function (function where the two processes are spawned) follows the format of
int main(){
Test t;
t.height=1000;
t.width=1000;
if ((t->iterations = malloc(t->width * t->height * sizeof(int))) ==
NULL) {
perror("Cannot allocate memory (iterations)");
exit(EXIT_FAILURE);
}
processes(Test *);
}
void processes(Test *){
int p2c[2], c2p[2];
int i,j,k;
int half;
int *dummy = malloc(Test->width*Test->height*sizeof(int));
if(dummy==NULL){
perror("Malloc Error");
exit(EXIT_FAILURE);
}
for (i = 0; i < Test->height; i++) {
for (j = 0; j < Test->width; j++) {
dummy[i * Test->width + j] = 0.0;
}
}
half=Test->height >> 1;
pipe(p2c);
pipe(c2p);
if(fork()==0){
Test->start=0;
Test->end=half;
Compute(Test);
for(i=Test->start;i<Test->end;i++){
for(j=0;j<Test->width;j++){
dummy[i*Test->width+j]=Test->iterations[i*Test->width+j];
}
}
chwrite(c2p[WRITE], (char*)dummy, Test->width * half * sizeof(int), 2048);
printf("Sending Iterations\n");
close(p2c[READ]);
close(c2p[WRITE]);
exit(EXIT_SUCCESS);
}
else{
Test->start=half;
Test->end=Test->height;
Compute(Test);
sleep(1);
chread(c2p[READ], (char*)dummy, Test->width * half * sizeof(int), 2048);
printf("Reading Iterations\n");
wait(NULL);
for(i=0;i<Test->height;i++){
for(j=0;j<Test->width;j++){
Test->iterations[i*Test->width+j]=dummy[i*Test->width+j];
}
}
However, when i do this, the iterations array is full of zeros. What is the problem and how can i fix this?
My chunk read and write functions are as follows:
void chwrite(int fd, char *buf, int count, int chunksize){
int numChunks = (int)(count/chunksize);
int remsize = count-numChunks*chunksize;
int i, j;
for(i=0,j=0; i<numChunks; j+=chunksize,i++){
write(fd, &(buf[j]), chunksize);
}
write(fd, &(buf[j]),remsize);
}
void chread(int fd, char *buf, int count, int chunksize){
int numChunks = (int)(count/chunksize);
int remsize=count-numChunks*chunksize;
int i,j;
for(i=0,j=0; i<numChunks; j+=numChunks, i++){
read(fd, &(buf[j]),chunksize);
}
read(fd, &(buf[j]) ,remsize);
}
The chwrite and chread functions do not check for errors or incomplete writes or reads. Also, there is a bug in chread where j is being increased by numChunks instead of by chunksize:
for(i=0,j=0; i<numChunks; j+=numChunks, i++){
read(fd, &(buf[j]),chunksize);
}
Here is a possible replacement for chread that checks for errors and allows incomplete reads. It returns the amount read if anything has been read, returns 0 if nothing has been read and either the amount to be read was 0 or an end-of-file condition was encountered, or -1 if nothing was read and an error was encountered:
int chread(int fd, char *buf, int count, int chunksize){
int done = 0;
int rc = count ? -1 : 0;
while (done < count) {
if (chunksize > done - count) {
chunksize = done - count;
}
rc = read(fd, &buf[done], chunksize);
if (rc <= 0) {
break;
}
done += rc;
}
return done ? done : rc;
}
Here is a replacement chwrite function. It is similar to chread except that there is no check for an end-of-file condition. (An end-of-file condition doesn't make sense on write, and a properly functioning write call should not return 0 unless called with a count of 0.):
int chwrite(int fd, const char *buf, int count, int chunksize){
int done = 0;
int rc = count ? -1 : 0;
while (done < count) {
if (chunksize > done - count) {
chunksize = done - count;
}
rc = write(fd, &buf[done], chunksize);
if (rc < 0) {
break;
}
done += rc;
}
return done ? done : rc;
}
There is also a problem in the parent process overwriting the second half of the test results with zeros here:
for(i=0;i<Test->height;i++){
for(j=0;j<Test->width;j++){
Test->iterations[i*Test->width+j]=dummy[i*Test->width+j];
}
}
(Note: OP's incomplete code has a variable t of type Test so the the use of a pointer called Test is very probably incorrect and shouldn't even compile, but let's go with it for now. OP's incomplete code has other errors, such as the function call processes(Test *); and the function definition void processes(Test *){ … }.)
The outer for loop condition should be i<half or i<Test->start:
for(i=0;i<Test->start;i++){
for(j=0;j<Test->width;j++){
Test->iterations[i*Test->width+j]=dummy[i*Test->width+j];
}
}