Hi all,
if I have a shared float var;
I cannot initialize a shared variable at declaration time like shared float var = 4; // not valid
one way what I think is to initialize it in the first line of my kernel, and then put __syncthreads();
is it a right approach or there is any other efficient way.
please let me know thanks
–
biebo
NUST Pakistan
This is the right way.
Look at it this way: if you ask the compiler to initialize int shared var = 4; - which thread should perform the initialization? One? All of them?
Even more difficult when you demand this from the compiler:
int shared myarray[4] = { 1,2,3,4 };
What should the compiler make of it? Your optimal code would need to look similar to this
// tid would be threadIdx.x
if (tid < 4)
{
myarray[tid] = tid+1;
}
__syncthreads();
That’s nothing a compiler could “guess” or “invent” on its own.
Christian
kernel(){
if(tid == 0){
var = 4;
}
__syncthreads();
… more …
} // kernel ends
–
biebo
Yup… that’s the correct and safest way to perform shared mem initialization…