1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
|
// =============================================================================
// === GPUQREngine/Source/GPUQREngine_UberKernel.cu ============================
// =============================================================================
//
// This is the actual concrete kernel invocation, transfering control flow to
// the GPU accelerator briefly. We actually launch kernels using alternating
// streams to overlap communication with computation, so the launch is actually
// asynchronous in nature. We use the CUDA events and streams model througout
// the Scheduler to coordinate asynchronous launch behavior.
//
// =============================================================================
#define CUDA_INCLUDE
#include "Kernel/uberKernel.cu"
void GPUQREngine_UberKernel
(
cudaStream_t kernelStream, // The stream on which to launch the kernel
TaskDescriptor *gpuWorkQueue, // The list of work items for the GPU
int numTasks // The # of items in the work list
)
{
/* Set the standard launch configuration. */
dim3 threads(NUMTHREADS, 1);
dim3 grid(numTasks, 1);
/* Launch the kernel */
qrKernel<<<grid, threads, 0, kernelStream>>>(gpuWorkQueue, numTasks);
}
|
