// Locks to an OS thread and initializes CUDA for that thread.
func Init(gpu int) {
if cudaCtx != 0 {
return // needed for tests
}
runtime.LockOSThread()
tryCuInit()
dev := cu.Device(gpu)
cudaCtx = cu.CtxCreate(cu.CTX_SCHED_YIELD, dev)
cudaCtx.SetCurrent()
M, m := dev.ComputeCapability()
cudaCC = 10*M + m
Version = cu.Version()
DevName = dev.Name()
TotalMem = dev.TotalMem()
GPUInfo = fmt.Sprint("CUDA ", Version, " ", DevName, "(", (TotalMem)/(1024*1024), "MB) ", "cc", M, ".", m)
if M < 2 {
log.Fatalln("GPU has insufficient compute capability, need 2.0 or higher.")
}
if Synchronous {
log.Println("DEBUG: synchronized CUDA calls")
}
// test PTX load so that we can catch CUDA_ERROR_NO_BINARY_FOR_GPU early
fatbinLoad(madd2_map, "madd2")
}
// needed for all other tests.
func init() {
cu.Init(0)
ctx := cu.CtxCreate(cu.CTX_SCHED_AUTO, 0)
cu.CtxSetCurrent(ctx)
fmt.Println("Created CUDA context")
}