generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added results check into multi_gpu sample

Browse Source
This commit is contained in:
Alexey Spizhevoy
2011-01-28 09:42:45 +00:00
parent b582330b90
commit e6d17406af
1 changed files with 18 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
samples/gpu/multi_gpu.cpp
View File

@@ -62,13 +62,13 @@ int main()
cuSafeCall(cuCtxCreate(&contexts[0], 0, device)); cuSafeCall(cuCtxCreate(&contexts[0], 0, device));
CUcontext prev_context; CUcontext prev_context;
cuCtxPopCurrent(&prev_context); cuSafeCall(cuCtxPopCurrent(&prev_context));
// Create context for the second GPU // Create context for the second GPU
cuSafeCall(cuDeviceGet(&device, 1)); cuSafeCall(cuDeviceGet(&device, 1));
cuSafeCall(cuCtxCreate(&contexts[1], 1, device)); cuSafeCall(cuCtxCreate(&contexts[1], 1, device));
cuCtxPopCurrent(&prev_context); cuSafeCall(cuCtxPopCurrent(&prev_context));
// Execute calculation in two threads using two GPUs // Execute calculation in two threads using two GPUs
int devices[] = {0, 1}; int devices[] = {0, 1};
@@ -81,35 +81,42 @@ int main()
void Worker::operator()(int device_id) const void Worker::operator()(int device_id) const
{ {
cuCtxPushCurrent(contexts[device_id]); // Set proper context
cuSafeCall(cuCtxPushCurrent(contexts[device_id]));
// Generate random matrix
Mat src(1000, 1000, CV_32F); Mat src(1000, 1000, CV_32F);
Mat dst;
RNG rng(0); RNG rng(0);
rng.fill(src, RNG::UNIFORM, 0, 1); rng.fill(src, RNG::UNIFORM, 0, 1);
// Upload data on GPU // CPU works
transpose(src, dst);
GpuMat d_src(src); GpuMat d_src(src);
GpuMat d_dst; GpuMat d_dst;
// GPU works
transpose(d_src, d_dst); transpose(d_src, d_dst);
// Deallocate here, otherwise deallocation will be performed // Check results
bool passed = norm(dst - Mat(d_dst), NORM_INF) < 1e-3;
cout << "GPU #" << device_id << ": "<< (passed ? "passed" : "FAILED") << endl;
// Deallocate data here, otherwise deallocation will be performed
// after context is extracted from the stack // after context is extracted from the stack
d_src.release(); d_src.release();
d_dst.release(); d_dst.release();
CUcontext prev_context; CUcontext prev_context;
cuCtxPopCurrent(&prev_context); cuSafeCall(cuCtxPopCurrent(&prev_context));
cout << "Device " << device_id << " finished\n";
} }
void destroyContexts() void destroyContexts()
{ {
cuCtxDestroy(contexts[0]); cuSafeCall(cuCtxDestroy(contexts[0]));
cuCtxDestroy(contexts[1]); cuSafeCall(cuCtxDestroy(contexts[1]));
} }
#endif #endif