portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Error message added when attempting to sum object which is too large for

Browse Source 
the shared memory
This commit is contained in:
Fabian Joswig 2021-10-13 20:45:46 +01:00
parent 6c66b8d997
commit 1f9688417a
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
Grid/lattice/Lattice_reduction_gpu.h
View File

@ -42,7 +42,6 @@ void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator
std::cout << GridLogDebug << "\twarpSize = " << warpSize << std::endl; std::cout << GridLogDebug << "\twarpSize = " << warpSize << std::endl;
std::cout << GridLogDebug << "\tsharedMemPerBlock = " << sharedMemPerBlock << std::endl; std::cout << GridLogDebug << "\tsharedMemPerBlock = " << sharedMemPerBlock << std::endl;
std::cout << GridLogDebug << "\tmaxThreadsPerBlock = " << maxThreadsPerBlock << std::endl; std::cout << GridLogDebug << "\tmaxThreadsPerBlock = " << maxThreadsPerBlock << std::endl;
std::cout << GridLogDebug << "\tmaxThreadsPerBlock = " << warpSize << std::endl;
std::cout << GridLogDebug << "\tmultiProcessorCount = " << multiProcessorCount << std::endl; std::cout << GridLogDebug << "\tmultiProcessorCount = " << multiProcessorCount << std::endl;
if (warpSize != WARP_SIZE) { if (warpSize != WARP_SIZE) {
@ -52,6 +51,10 @@ void getNumBlocksAndThreads(const Iterator n, const size_t sizeofsobj, Iterator
// let the number of threads in a block be a multiple of 2, starting from warpSize // let the number of threads in a block be a multiple of 2, starting from warpSize
threads = warpSize; threads = warpSize;
if ( threads*sizeofsobj > sharedMemPerBlock ) {
std::cout << GridLogError << "The object is too large for the shared memory." << std::endl;
exit(EXIT_FAILURE);
}
while( 2*threads*sizeofsobj < sharedMemPerBlock && 2*threads <= maxThreadsPerBlock ) threads *= 2; while( 2*threads*sizeofsobj < sharedMemPerBlock && 2*threads <= maxThreadsPerBlock ) threads *= 2;
// keep all the streaming multiprocessors busy // keep all the streaming multiprocessors busy
blocks = nextPow2(multiProcessorCount); blocks = nextPow2(multiProcessorCount);