void GPUInterface::MemsetShort(GPUPtr dest,
unsigned short val,
size_t count) {
#ifdef BEAGLE_DEBUG_FLOW
fprintf(stderr, "\t\t\tEntering GPUInterface::MemsetShort\n");
#endif
SAFE_CUPP(cuMemsetD16(dest, val, count));
#ifdef BEAGLE_DEBUG_FLOW
fprintf(stderr, "\t\t\tLeaving GPUInterface::MemsetShort\n");
#endif
}
CUresult I_WRAP_SONAME_FNNAME_ZZ(libcudaZdsoZa, cuMemsetD16Async)(CUdeviceptr dstDevice, unsigned short us, size_t N, CUstream hStream) {
int error = 0;
long vgErrorAddress;
vgErrorAddress = VALGRIND_CHECK_MEM_IS_DEFINED(&hStream, sizeof(CUstream));
if (vgErrorAddress) {
error++;
VALGRIND_PRINTF("Error: 'hStream' in call to cuMemsetD16Async not defined.\n");
}
cgLock();
CUcontext ctx = NULL;
cgCtxListType *nodeCtx;
cgMemListType *nodeMemDst;
// Get current context ..
cgGetCtx(&ctx);
nodeCtx = cgFindCtx(ctx);
// .. and locate memory if we are handling device memory
nodeMemDst = cgFindMem(nodeCtx, dstDevice);
if (nodeMemDst && nodeMemDst->locked & 2 && nodeMemDst->stream != hStream) {
error++;
VALGRIND_PRINTF("Error: Concurrent write and read access by different streams.\n");
}
if (nodeMemDst) {
nodeMemDst->locked = nodeMemDst->locked | 2;
nodeMemDst->stream = hStream;
}
cgUnlock();
if (error) {
VALGRIND_PRINTF_BACKTRACE("");
}
return cuMemsetD16(dstDevice, us, N);
}
