internal void CalculateAsync(int value)
{
int threadsPerBlock, blockCount;
if (count <= defaultThreadsPerBlock)
{
blockCount = 1;
threadsPerBlock = RoundUp(count, warpSize);
}
else if (count >= defaultThreadsPerBlock * defaultBlockCount)
{
threadsPerBlock = defaultThreadsPerBlock;
blockCount = defaultBlockCount;
}
else
{
threadsPerBlock = defaultThreadsPerBlock;
blockCount = (count + threadsPerBlock - 1) / threadsPerBlock;
}
multiply.BlockDimensions = new ManagedCuda.VectorTypes.dim3(threadsPerBlock, 1, 1);
multiply.GridDimensions = new ManagedCuda.VectorTypes.dim3(blockCount, 1, 1);
multiply.RunAsync(defaultStream.Stream, new object[] {
count,
deviceBuffer.DevicePointer,
secondDeviceBuffer.DevicePointer,
value
});
}
public static void Invoke(TSCudaContext context, CudaContext cudaContext, byte[] ptx, string baseName, params object[] args)
{
ThrowIfAnyTensorInvalid(args);
cudaContext.SetCurrent();
CudaDeviceProperties deviceInfo = context.DeviceInfoForContext(cudaContext);
IEnumerable <Tensor> allTensors = args.OfType <Tensor>();
Tensor firstTensor = allTensors.First();
long elementCount = firstTensor.ElementCount();
ApplySpecialization spec = new ApplySpecialization(allTensors.ToArray());
ConvertTensorArgs.Convert(cudaContext, spec.Use32BitIndices, args);
ManagedCuda.VectorTypes.dim3 block = ApplyUtils.GetApplyBlock();
ManagedCuda.VectorTypes.dim3 grid = ApplyUtils.GetApplyGrid(deviceInfo, elementCount);
string fullKernelName = PermutationGenerator.GetMangledName(baseName, spec);
CudaKernel kernel = context.KernelCache.Get(cudaContext, ptx, fullKernelName);
kernel.GridDimensions = grid;
kernel.BlockDimensions = block;
kernel.RunAsync(CUstream.NullStream, args);
}
private void RunKernel(CudaKernel kernel, int count, IEnumerable <object> parameters)
{
// configure the dimensions; note, usually this is a lot more dynamic based
// on input data, but we'll still go through the motions
int threadsPerBlock, blockCount;
if (count <= this._context.DefaultThreadsPerBlock) // a single block
{
blockCount = 1;
threadsPerBlock = RoundUp(count, this._context.WarpSize); // slight caveat here; if you are using "shuffle" operations, you
// need to use entire "warp"s - otherwise the result is undefined
}
else if (count <= this._context.DefaultThreadsPerBlock * this._context.DefaultBlockCount)
{
// more than enough work to keep us busy; just use that
threadsPerBlock = this._context.DefaultThreadsPerBlock;
blockCount = this._context.DefaultBlockCount;
}
else
{
// do the math to figure out how many blocks we need
threadsPerBlock = this._context.DefaultThreadsPerBlock;
blockCount = (count + threadsPerBlock - 1) / threadsPerBlock;
}
// we're using 1-D math, but actually CUDA supports blocks and grids that span 3 dimensions
kernel.BlockDimensions = new dim3(threadsPerBlock, 1, 1);
kernel.GridDimensions = new dim3(blockCount, 1, 1);
// invoke the kernel
var withCount = parameters.ToList();
withCount.Insert(0, count);
kernel.RunAsync(this._context.DefaultStream.Stream, withCount.ToArray());
}
internal void MultiplyAsync(int value)
{
// configure the dimensions; note, usually this is a lot more dynamic based
// on input data, but we'll still go through the motions
int threadsPerBlock, blockCount;
if (count <= defaultThreadsPerBlock) // a single block
{
blockCount = 1;
threadsPerBlock = RoundUp(count, warpSize); // slight caveat here; if you are using "shuffle" operations, you
// need to use entire "warp"s - otherwise the result is undefined
}
else if (count >= defaultThreadsPerBlock * defaultBlockCount)
{
// more than enough work to keep us busy; just use that
threadsPerBlock = defaultThreadsPerBlock;
blockCount = defaultBlockCount;
}
else
{
// do the math to figure out how many blocks we need
threadsPerBlock = defaultThreadsPerBlock;
blockCount = (count + threadsPerBlock - 1) / threadsPerBlock;
}
// we're using 1-D math, but actually CUDA supports blocks and grids that span 3 dimensions
multiply.BlockDimensions = new ManagedCuda.VectorTypes.dim3(threadsPerBlock, 1, 1);
multiply.GridDimensions = new ManagedCuda.VectorTypes.dim3(blockCount, 1, 1);
// invoke the kernel
multiply.RunAsync(defaultStream.Stream, new object[] {
// note the signature is (N, data, factor)
count, deviceBuffer.DevicePointer, value
});
}
/// <summary> Runs the kernel in asynchronous mode. </summary>
/// <param name="stream">If the stream is null, the default per-thread stream is used.</param>
/// <param name="args">MyMemoryBlock arguments are automatically converted to device pointers.</param>
public void RunAsync(CudaStream stream, params object[] args)
{
CheckExecutionSetup();
ConvertMemoryBlocksToDevicePtrs(args);
m_kernel.RunAsync(MyKernelFactory.GetCuStreamOrDefault(stream), args);
}
internal void MultiplyAsync(int value)
{
// invoke the kernel
multiply.RunAsync(defaultStream.Stream, new object[] {
// note the signature is (N, data, factor)
count, deviceBuffer.DevicePointer, value
});
}
private void RunKernel(Volume <T> input, Volume <T> output, CudaKernel kernel)
{
if (!Equals(input.Shape, output.Shape))
{
throw new ArgumentException($"{nameof(input)} and {nameof(output)} should have the same shape.");
}
var inputStorage = input.Storage as IVolumeStorage <T>;
if (inputStorage == null)
{
throw new ArgumentException($"{nameof(input)} storage should be VolumeStorage", nameof(input));
}
var outputStorage = output.Storage as IVolumeStorage <T>;
if (outputStorage == null)
{
throw new ArgumentException($"{nameof(output)} storage should be VolumeStorage", nameof(output));
}
inputStorage.CopyToDevice();
outputStorage.CopyToDevice();
var count = (int)input.Shape.TotalLength;
// configure the dimensions; note, usually this is a lot more dynamic based
// on input data, but we'll still go through the motions
int threadsPerBlock, blockCount;
if (count <= this._context.DefaultThreadsPerBlock) // a single block
{
blockCount = 1;
threadsPerBlock = RoundUp(count, this._context.WarpSize); // slight caveat here; if you are using "shuffle" operations, you
// need to use entire "warp"s - otherwise the result is undefined
}
else if (count >= this._context.DefaultThreadsPerBlock * this._context.DefaultBlockCount)
{
// more than enough work to keep us busy; just use that
threadsPerBlock = this._context.DefaultThreadsPerBlock;
blockCount = this._context.DefaultBlockCount;
}
else
{
// do the math to figure out how many blocks we need
threadsPerBlock = this._context.DefaultThreadsPerBlock;
blockCount = (count + threadsPerBlock - 1) / threadsPerBlock;
}
// we're using 1-D math, but actually CUDA supports blocks and grids that span 3 dimensions
kernel.BlockDimensions = new dim3(threadsPerBlock, 1, 1);
kernel.GridDimensions = new dim3(blockCount, 1, 1);
// invoke the kernel
kernel.RunAsync(this._context.DefaultStream.Stream, count, inputStorage.DeviceBuffer.DevicePointer, outputStorage.DeviceBuffer.DevicePointer);
}
private void Invoke(TSCudaContext context, CudaContext cudaContext, string kernelName, dim3 grid, dim3 block, uint smemSize, CUstream stream, params object[] args)
{
byte[] ptx = GetPtx(context.Compiler);
CudaKernel kernel = context.KernelCache.Get(cudaContext, ptx, kernelName);
kernel.GridDimensions = grid;
kernel.BlockDimensions = block;
kernel.DynamicSharedMemory = smemSize;
kernel.RunAsync(stream, args);
}
/// <summary>
/// Runs a kernel.
/// </summary>
/// <param name="threadCount">The number of instances of the kernel to run.</param>
/// <param name="method">The method to compiled and run as a kernel.</param>
/// <param name="target">An optional first argument to pass to the kernel.</param>
/// <param name="takesThreadId">Tells if the kernel's argument list is to be prefixed by a thread-unique ID.</param>
/// <param name="args">The list of arguments to feed to the kernel.</param>
/// <returns>A task that completes when the kernel does.</returns>
private static Task ForAsync(
int threadCount,
MethodInfo method,
object target,
bool takesThreadId,
params object[] args)
{
int paramCount = method.GetParameters().Length;
if (takesThreadId)
{
paramCount--;
}
bool hasThisArg = !method.IsStatic || args.Length != paramCount;
return(manager.RunAsync(
new KernelDescription <bool>(
method,
target,
takesThreadId ? (hasThisArg ? 1 : 0) : -1,
(module, stream, encodedTarget) =>
{
// TODO: create blocks, grids to better spread workload.
var kernelInstance = new CudaKernel(
module.EntryPointName,
module.CompiledModule,
module.Context,
threadCount);
if (hasThisArg)
{
kernelInstance.RunAsync(stream.Stream, new object[] { encodedTarget }.Concat(args).ToArray());
}
else
{
kernelInstance.RunAsync(stream.Stream, args);
}
return () => true;
})));
}
public void AssembleElemsAndCZs()
{
AllocateMemoryForLinearSystem();
sw.Restart();
cf.nElems = mc.elasticElements.Length;
cf.nCZ = mc.nonFailedCZs.Length;
// set kernel configurations
kelElementElasticityForce.GridDimensions = new dim3(grid(mc.elasticElements.Length), 1, 1);
// run kernels
kelElementElasticityForce.RunAsync(CUstream.NullStream,
g_ie_pcsr.DevicePointer, g_dn.DevicePointer, cf.TimeStep,
g_dvals.DevicePointer, g_drhs.DevicePointer,
mc.elasticElements.Length, el_elastic_stride, nd_stride);
ctx.Synchronize(); // this is for benchmarkng only - does not affect functionality
sw.Stop();
cf.KerElemForce += sw.ElapsedMilliseconds;
if (cz_stride != 0)
{
sw.Restart();
kczCZForce.GridDimensions = new dim3(grid(mc.nonFailedCZs.Length), 1, 1);
kczCZForce.RunAsync(CUstream.NullStream, g_dcz.DevicePointer, g_icz.DevicePointer,
g_dn.DevicePointer,
g_dvals.DevicePointer, g_drhs.DevicePointer,
cf.TimeStep,
mc.nonFailedCZs.Length, cz_stride, nd_stride);
// pointers to specific portions of the arrays
CUdeviceptr ptr_failed = g_icz.DevicePointer + sizeof(int) * cz_stride * TENTATIVE_FAILED_OFFSET_CZ;
CUdeviceptr ptr_damaged = g_icz.DevicePointer + sizeof(int) * cz_stride * TENTATIVE_DAMAGED_OFFSET_CZ;
cf.nCZDamaged = Sum(ptr_damaged, mc.nonFailedCZs.Length);
cf.nCZFailedThisStep = Sum(ptr_failed, mc.nonFailedCZs.Length);
sw.Stop();
cf.KerCZForce += sw.ElapsedMilliseconds;
}
}
public void RunAsync(CudaStream stream, params object[] args)
{
for (int i = 0; i < args.Length; i++)
{
if (args[i] is MyAbstractMemoryBlock)
{
args[i] = (args[i] as MyAbstractMemoryBlock).GetDevicePtr(m_GPU);
if (((CUdeviceptr)args[i]).Pointer == 0)
{
// TODO(Premek): this is now handled in observers, should be also handled in the simulation
throw new InvalidOperationException("Memory block resolved to null device ptr (not allocated on device?).");
}
}
}
CUstream cuStream = CUstream.NullStream;
if (stream != null)
{
cuStream = stream.Stream;
}
m_kernel.RunAsync(cuStream, args);
}
static void Main(string[] args)
{
int cuda_device = 0;
int nstreams = 4; // number of streams for CUDA calls
int nreps = 10; // number of times each experiment is repeated
int n = 16 * 1024 * 1024; // number of ints in the data set
int nbytes = n * sizeof(int); // number of data bytes
dim3 threads, blocks; // kernel launch configuration
float elapsed_time, time_memcpy, time_kernel; // timing variables
float scale_factor = 1.0f;
// allocate generic memory and pin it laster instead of using cudaHostAlloc()
// Untested in C#, so stick to cudaHostAlloc().
bool bPinGenericMemory = false; // we want this to be the default behavior
CUCtxFlags device_sync_method = CUCtxFlags.BlockingSync; // by default we use BlockingSync
int niterations; // number of iterations for the loop inside the kernel
ShrQATest.shrQAStart(args);
Console.WriteLine("[ simpleStreams ]");
foreach (var item in args)
{
if (item.Contains("help"))
{
printHelp();
ShrQATest.shrQAFinishExit(args, ShrQATest.eQAstatus.QA_PASSED);
}
}
bPinGenericMemory = false;
foreach (var item in args)
{
if (item.Contains("use_generic_memory"))
{
bPinGenericMemory = true;
}
}
for (int i = 0; i < args.Length; i++)
{
if (args[i].Contains("sync_method"))
{
int temp = -1;
bool error = false;
if (i < args.Length - 1)
{
error = int.TryParse(args[i + 1], out temp);
switch (temp)
{
case 0:
device_sync_method = CUCtxFlags.SchedAuto;
break;
case 1:
device_sync_method = CUCtxFlags.SchedSpin;
break;
case 2:
device_sync_method = CUCtxFlags.SchedYield;
break;
case 4:
device_sync_method = CUCtxFlags.BlockingSync;
break;
default:
error = true;
break;
}
}
if (!error)
{
Console.Write("Specifying device_sync_method = {0}, setting reps to 100 to demonstrate steady state\n", sDeviceSyncMethod[(int)device_sync_method]);
nreps = 100;
}
else
{
Console.Write("Invalid command line option sync_method=\"{0}\"\n", temp);
ShrQATest.shrQAFinishExit(args, ShrQATest.eQAstatus.QA_FAILED);
}
}
}
int num_devices = CudaContext.GetDeviceCount();
if (0 == num_devices)
{
Console.Write("your system does not have a CUDA capable device, waiving test...\n");
ShrQATest.shrQAFinishExit(args, ShrQATest.eQAstatus.QA_FAILED);
}
cuda_device = CudaContext.GetMaxGflopsDeviceId();
CudaDeviceProperties deviceProp = CudaContext.GetDeviceInfo(cuda_device);
if ((1 == deviceProp.ComputeCapability.Major) && (deviceProp.ComputeCapability.Minor < 1))
{
Console.Write("{0} does not have Compute Capability 1.1 or newer. Reducing workload.\n", deviceProp.DeviceName);
}
if (deviceProp.ComputeCapability.Major >= 2)
{
niterations = 100;
}
else
{
if (deviceProp.ComputeCapability.Minor > 1)
{
niterations = 5;
}
else
{
niterations = 1; // reduced workload for compute capability 1.0 and 1.1
}
}
// Check if GPU can map host memory (Generic Method), if not then we override bPinGenericMemory to be false
// In .net we cannot allocate easily generic aligned memory, so <bPinGenericMemory> is always false in our case...
if (bPinGenericMemory)
{
Console.Write("Device: <{0}> canMapHostMemory: {1}\n", deviceProp.DeviceName, deviceProp.CanMapHostMemory ? "Yes" : "No");
if (deviceProp.CanMapHostMemory == false)
{
Console.Write("Using cudaMallocHost, CUDA device does not support mapping of generic host memory\n");
bPinGenericMemory = false;
}
}
// Anything that is less than 32 Cores will have scaled down workload
scale_factor = Math.Max((32.0f / (ConvertSMVer2Cores(deviceProp.ComputeCapability.Major, deviceProp.ComputeCapability.Minor) * (float)deviceProp.MultiProcessorCount)), 1.0f);
n = (int)Math.Round((float)n / scale_factor);
Console.Write("> CUDA Capable: SM {0}.{1} hardware\n", deviceProp.ComputeCapability.Major, deviceProp.ComputeCapability.Minor);
Console.Write("> {0} Multiprocessor(s) x {1} (Cores/Multiprocessor) = {2} (Cores)\n",
deviceProp.MultiProcessorCount,
ConvertSMVer2Cores(deviceProp.ComputeCapability.Major, deviceProp.ComputeCapability.Minor),
ConvertSMVer2Cores(deviceProp.ComputeCapability.Major, deviceProp.ComputeCapability.Minor) * deviceProp.MultiProcessorCount);
Console.Write("> scale_factor = {0:0.0000}\n", 1.0f / scale_factor);
Console.Write("> array_size = {0}\n\n", n);
// enable use of blocking sync, to reduce CPU usage
Console.Write("> Using CPU/GPU Device Synchronization method ({0})\n", sDeviceSyncMethod[(int)device_sync_method]);
CudaContext ctx;
if (bPinGenericMemory)
{
ctx = new CudaContext(cuda_device, device_sync_method | CUCtxFlags.MapHost);
}
else
{
ctx = new CudaContext(cuda_device, device_sync_method);
}
//Load Kernel image from resources
string resName;
if (IntPtr.Size == 8)
{
resName = "simpleStreams_x64.ptx";
}
else
{
resName = "simpleStreams.ptx";
}
string resNamespace = "simpleStreams";
string resource = resNamespace + "." + resName;
Stream stream = Assembly.GetExecutingAssembly().GetManifestResourceStream(resource);
if (stream == null)
{
throw new ArgumentException("Kernel not found in resources.");
}
CudaKernel init_array = ctx.LoadKernelPTX(stream, "init_array");
// allocate host memory
int c = 5; // value to which the array will be initialized
int[] h_a = null; // pointer to the array data in host memory
CudaPageLockedHostMemory <int> hAligned_a = null; // pointer to the array data in host memory (aligned to MEMORY_ALIGNMENT)
//Note: In .net we have two seperated arrays: One is in managed memory (h_a), the other one in unmanaged memory (hAligned_a).
//In C++ hAligned_a would point somewhere inside the h_a array.
AllocateHostMemory(bPinGenericMemory, ref h_a, ref hAligned_a, nbytes);
Console.Write("\nStarting Test\n");
// allocate device memory
CudaDeviceVariable <int> d_c = c; //using new implicit cast to allocate memory and asign value
CudaDeviceVariable <int> d_a = new CudaDeviceVariable <int>(nbytes / sizeof(int));
CudaStream[] streams = new CudaStream[nstreams];
for (int i = 0; i < nstreams; i++)
{
streams[i] = new CudaStream();
}
// create CUDA event handles
// use blocking sync
CudaEvent start_event, stop_event;
CUEventFlags eventflags = ((device_sync_method == CUCtxFlags.BlockingSync) ? CUEventFlags.BlockingSync : CUEventFlags.Default);
start_event = new CudaEvent(eventflags);
stop_event = new CudaEvent(eventflags);
// time memcopy from device
start_event.Record(); // record in stream-0, to ensure that all previous CUDA calls have completed
hAligned_a.AsyncCopyToDevice(d_a, streams[0].Stream);
stop_event.Record();
stop_event.Synchronize(); // block until the event is actually recorded
time_memcpy = CudaEvent.ElapsedTime(start_event, stop_event);
Console.Write("memcopy:\t{0:0.00}\n", time_memcpy);
// time kernel
threads = new dim3(512, 1);
blocks = new dim3(n / (int)threads.x, 1);
start_event.Record();
init_array.BlockDimensions = threads;
init_array.GridDimensions = blocks;
init_array.RunAsync(streams[0].Stream, d_a.DevicePointer, d_c.DevicePointer, niterations);
stop_event.Record();
stop_event.Synchronize();
time_kernel = CudaEvent.ElapsedTime(start_event, stop_event);
Console.Write("kernel:\t\t{0:0.00}\n", time_kernel);
//////////////////////////////////////////////////////////////////////
// time non-streamed execution for reference
threads = new dim3(512, 1);
blocks = new dim3(n / (int)threads.x, 1);
start_event.Record();
for (int k = 0; k < nreps; k++)
{
init_array.BlockDimensions = threads;
init_array.GridDimensions = blocks;
init_array.Run(d_a.DevicePointer, d_c.DevicePointer, niterations);
hAligned_a.SynchronCopyToHost(d_a);
}
stop_event.Record();
stop_event.Synchronize();
elapsed_time = CudaEvent.ElapsedTime(start_event, stop_event);
Console.Write("non-streamed:\t{0:0.00} ({1:00} expected)\n", elapsed_time / nreps, time_kernel + time_memcpy);
//////////////////////////////////////////////////////////////////////
// time execution with nstreams streams
threads = new dim3(512, 1);
blocks = new dim3(n / (int)(nstreams * threads.x), 1);
byte[] memset = new byte[nbytes]; // set host memory bits to all 1s, for testing correctness
for (int i = 0; i < nbytes; i++)
{
memset[i] = 255;
}
System.Runtime.InteropServices.Marshal.Copy(memset, 0, hAligned_a.PinnedHostPointer, nbytes);
d_a.Memset(0); // set device memory to all 0s, for testing correctness
start_event.Record();
for (int k = 0; k < nreps; k++)
{
init_array.BlockDimensions = threads;
init_array.GridDimensions = blocks;
// asynchronously launch nstreams kernels, each operating on its own portion of data
for (int i = 0; i < nstreams; i++)
{
init_array.RunAsync(streams[i].Stream, d_a.DevicePointer + i * n / nstreams * sizeof(int), d_c.DevicePointer, niterations);
}
// asynchronously launch nstreams memcopies. Note that memcopy in stream x will only
// commence executing when all previous CUDA calls in stream x have completed
for (int i = 0; i < nstreams; i++)
{
hAligned_a.AsyncCopyFromDevice(d_a, i * n / nstreams * sizeof(int), i * n / nstreams * sizeof(int), nbytes / nstreams, streams[i].Stream);
}
}
stop_event.Record();
stop_event.Synchronize();
elapsed_time = CudaEvent.ElapsedTime(start_event, stop_event);
Console.Write("{0} streams:\t{1:0.00} ({2:0.00} expected with compute capability 1.1 or later)\n", nstreams, elapsed_time / nreps, time_kernel + time_memcpy / nstreams);
// check whether the output is correct
Console.Write("-------------------------------\n");
//We can directly access data in hAligned_a using the [] operator, but copying
//data first to h_a is faster.
System.Runtime.InteropServices.Marshal.Copy(hAligned_a.PinnedHostPointer, h_a, 0, nbytes / sizeof(int));
bool bResults = correct_data(h_a, n, c * nreps * niterations);
// release resources
for (int i = 0; i < nstreams; i++)
{
streams[i].Dispose();
}
start_event.Dispose();
stop_event.Dispose();
hAligned_a.Dispose();
d_a.Dispose();
d_c.Dispose();
CudaContext.ProfilerStop();
ctx.Dispose();
Console.ReadKey();
ShrQATest.shrQAFinishExit(args, bResults ? ShrQATest.eQAstatus.QA_PASSED : ShrQATest.eQAstatus.QA_FAILED);
}
static void Main(string[] args)
{
try
{
if (args.Length > 0)
{
deviceID = int.Parse(args[0]);
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Device ID parse error");
}
try
{
if (args.Length > 1)
{
port = int.Parse(args[1]);
Comms.ConnectToMaster(port);
}
else
{
TEST = true;
Logger.CopyToConsole = true;
CGraph.ShowCycles = true;
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Master connection error");
}
try
{
if (args.Length > 3)
{
gpuCount = int.Parse(args[3]);
fastCuda = gpuCount <= (Environment.ProcessorCount / 2);
if (fastCuda)
{
Logger.Log(LogLevel.Info, "Using single GPU blocking mode");
}
}
}
catch
{
}
if (TEST)
{
currentJob = nextJob = new Job()
{
jobID = 0,
k0 = 0xf4956dc403730b01L,
k1 = 0xe6d45de39c2a5a3eL,
k2 = 0xcbf626a8afee35f6L,
k3 = 0x4307b94b1a0c9980L,
pre_pow = TestPrePow,
timestamp = DateTime.Now
};
}
else
{
currentJob = nextJob = new Job()
{
jobID = 0,
k0 = 0xf4956dc403730b01L,
k1 = 0xe6d45de39c2a5a3eL,
k2 = 0xcbf626a8afee35f6L,
k3 = 0x4307b94b1a0c9980L,
pre_pow = TestPrePow,
timestamp = DateTime.Now
};
if (!Comms.IsConnected())
{
Console.WriteLine("Master connection failed, aborting");
Logger.Log(LogLevel.Error, "No master connection, exitting!");
return;
}
if (deviceID < 0)
{
int devCnt = CudaContext.GetDeviceCount();
GpuDevicesMessage gpum = new GpuDevicesMessage()
{
devices = new List <GpuDevice>(devCnt)
};
for (int i = 0; i < devCnt; i++)
{
string name = CudaContext.GetDeviceName(i);
var info = CudaContext.GetDeviceInfo(i);
gpum.devices.Add(new GpuDevice()
{
deviceID = i, name = name, memory = info.TotalGlobalMemory
});
}
//Console.WriteLine(devCnt);
Comms.gpuMsg = gpum;
Comms.SetEvent();
//Console.WriteLine("event fired");
Task.Delay(1000).Wait();
//Console.WriteLine("closing");
Comms.Close();
return;
}
}
try
{
var assembly = Assembly.GetEntryAssembly();
var resourceStream = assembly.GetManifestResourceStream("CudaSolver.kernel_x64.ptx");
ctx = new CudaContext(deviceID, !fastCuda ? (CUCtxFlags.BlockingSync | CUCtxFlags.MapHost) : CUCtxFlags.MapHost);
meanSeedA = ctx.LoadKernelPTX(resourceStream, "FluffySeed2A");
meanSeedA.BlockDimensions = 128;
meanSeedA.GridDimensions = 2048;
meanSeedA.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanSeedB = ctx.LoadKernelPTX(resourceStream, "FluffySeed2B");
meanSeedB.BlockDimensions = 128;
meanSeedB.GridDimensions = 2048;
meanSeedB.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanSeedB_4 = ctx.LoadKernelPTX(resourceStream, "FluffySeed2B");
meanSeedB_4.BlockDimensions = 128;
meanSeedB_4.GridDimensions = 1024;
meanSeedB_4.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRound = ctx.LoadKernelPTX(resourceStream, "FluffyRound");
meanRound.BlockDimensions = 512;
meanRound.GridDimensions = 4096;
meanRound.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRound_2 = ctx.LoadKernelPTX(resourceStream, "FluffyRound");
meanRound_2.BlockDimensions = 512;
meanRound_2.GridDimensions = 2048;
meanRound_2.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRoundJoin = ctx.LoadKernelPTX(resourceStream, "FluffyRound_J");
meanRoundJoin.BlockDimensions = 512;
meanRoundJoin.GridDimensions = 4096;
meanRoundJoin.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanTail = ctx.LoadKernelPTX(resourceStream, "FluffyTail");
meanTail.BlockDimensions = 1024;
meanTail.GridDimensions = 4096;
meanTail.PreferredSharedMemoryCarveout = CUshared_carveout.MaxL1;
meanRecover = ctx.LoadKernelPTX(resourceStream, "FluffyRecovery");
meanRecover.BlockDimensions = 256;
meanRecover.GridDimensions = 2048;
meanRecover.PreferredSharedMemoryCarveout = CUshared_carveout.MaxL1;
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Unable to create kernels: " + ex.Message);
Task.Delay(500).Wait();
Comms.Close();
return;
}
try
{
d_buffer = new CudaDeviceVariable <ulong>(BUFFER_SIZE_U32);
d_bufferMid = new CudaDeviceVariable <ulong>(d_buffer.DevicePointer + (BUFFER_SIZE_B * 8));
d_bufferB = new CudaDeviceVariable <ulong>(d_buffer.DevicePointer + (BUFFER_SIZE_A * 8));
d_indexesA = new CudaDeviceVariable <uint>(INDEX_SIZE * 2);
d_indexesB = new CudaDeviceVariable <uint>(INDEX_SIZE * 2);
Array.Clear(h_indexesA, 0, h_indexesA.Length);
Array.Clear(h_indexesB, 0, h_indexesA.Length);
d_indexesA = h_indexesA;
d_indexesB = h_indexesB;
streamPrimary = new CudaStream(CUStreamFlags.NonBlocking);
streamSecondary = new CudaStream(CUStreamFlags.NonBlocking);
}
catch (Exception ex)
{
Task.Delay(200).Wait();
Logger.Log(LogLevel.Error, $"Out of video memory! Only {ctx.GetFreeDeviceMemorySize()} free");
Task.Delay(500).Wait();
Comms.Close();
return;
}
try
{
AllocateHostMemory(true, ref h_a, ref hAligned_a, 1024 * 1024 * 32);
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Unable to create pinned memory.");
Task.Delay(500).Wait();
Comms.Close();
return;
}
int loopCnt = 0;
while (!Comms.IsTerminated)
{
try
{
if (!TEST && (Comms.nextJob.pre_pow == null || Comms.nextJob.pre_pow == "" || Comms.nextJob.pre_pow == TestPrePow))
{
Logger.Log(LogLevel.Info, string.Format("Waiting for job...."));
Task.Delay(1000).Wait();
continue;
}
if (!TEST && ((currentJob.pre_pow != Comms.nextJob.pre_pow) || (currentJob.origin != Comms.nextJob.origin)))
{
currentJob = Comms.nextJob;
currentJob.timestamp = DateTime.Now;
}
if (!TEST && (currentJob.timestamp.AddMinutes(30) < DateTime.Now) && Comms.lastIncoming.AddMinutes(30) < DateTime.Now)
{
Logger.Log(LogLevel.Info, string.Format("Job too old..."));
Task.Delay(1000).Wait();
continue;
}
// test runs only once
if (TEST && loopCnt++ > 100)
{
Comms.IsTerminated = true;
}
Solution s;
while (graphSolutions.TryDequeue(out s))
{
meanRecover.SetConstantVariable <ulong>("recovery", s.GetUlongEdges());
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRecover.RunAsync(streamPrimary.Stream, s.job.k0, s.job.k1, s.job.k2, s.job.k3, d_indexesB.DevicePointer);
streamPrimary.Synchronize();
s.nonces = new uint[40];
d_indexesB.CopyToHost(s.nonces, 0, 0, 40 * 4);
s.nonces = s.nonces.OrderBy(n => n).ToArray();
lock (Comms.graphSolutionsOut)
{
Comms.graphSolutionsOut.Enqueue(s);
}
Comms.SetEvent();
}
uint[] count;
do
{
if (!TEST && ((currentJob.pre_pow != Comms.nextJob.pre_pow) || (currentJob.origin != Comms.nextJob.origin)))
{
currentJob = Comms.nextJob;
currentJob.timestamp = DateTime.Now;
}
currentJob = currentJob.Next();
Logger.Log(LogLevel.Debug, string.Format("GPU NV{4}:Trimming #{4}: {0} {1} {2} {3}", currentJob.k0, currentJob.k1, currentJob.k2, currentJob.k3, currentJob.jobID, deviceID));
timer.Restart();
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanSeedA.RunAsync(streamPrimary.Stream, currentJob.k0, currentJob.k1, currentJob.k2, currentJob.k3, d_bufferMid.DevicePointer, d_indexesB.DevicePointer);
meanSeedB_4.RunAsync(streamPrimary.Stream, d_bufferMid.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, 0);
meanSeedB_4.RunAsync(streamPrimary.Stream, d_bufferMid.DevicePointer, d_buffer.DevicePointer + ((BUFFER_SIZE_A * 8) / 4) * 1, d_indexesB.DevicePointer, d_indexesA.DevicePointer, 16);
meanSeedB_4.RunAsync(streamPrimary.Stream, d_bufferMid.DevicePointer, d_buffer.DevicePointer + ((BUFFER_SIZE_A * 8) / 4) * 2, d_indexesB.DevicePointer, d_indexesA.DevicePointer, 32);
meanSeedB_4.RunAsync(streamPrimary.Stream, d_bufferMid.DevicePointer, d_buffer.DevicePointer + ((BUFFER_SIZE_A * 8) / 4) * 3, d_indexesB.DevicePointer, d_indexesA.DevicePointer, 48);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRound_2.RunAsync(streamPrimary.Stream, d_buffer.DevicePointer + ((BUFFER_SIZE_A * 8) / 4) * 2, d_bufferB.DevicePointer, d_indexesA.DevicePointer + (2048 * 4), d_indexesB.DevicePointer + (4096 * 4), DUCK_EDGES_A, DUCK_EDGES_B / 2);
meanRound_2.RunAsync(streamPrimary.Stream, d_buffer.DevicePointer, d_bufferB.DevicePointer - (BUFFER_SIZE_B * 8), d_indexesA.DevicePointer, d_indexesB.DevicePointer, DUCK_EDGES_A, DUCK_EDGES_B / 2);
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
meanRoundJoin.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer - (BUFFER_SIZE_B * 8), d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_B / 2, DUCK_EDGES_B / 2);
//d_indexesA.MemsetAsync(0, streamPrimary.Stream);
//meanRound.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_B, DUCK_EDGES_B / 2);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_buffer.DevicePointer, d_bufferB.DevicePointer, d_indexesA.DevicePointer, d_indexesB.DevicePointer, DUCK_EDGES_B / 2, DUCK_EDGES_B / 2);
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_B / 2, DUCK_EDGES_B / 2);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_buffer.DevicePointer, d_bufferB.DevicePointer, d_indexesA.DevicePointer, d_indexesB.DevicePointer, DUCK_EDGES_B / 2, DUCK_EDGES_B / 4);
for (int i = 0; i < trimRounds; i++)
{
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_B / 4, DUCK_EDGES_B / 4);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_buffer.DevicePointer, d_bufferB.DevicePointer, d_indexesA.DevicePointer, d_indexesB.DevicePointer, DUCK_EDGES_B / 4, DUCK_EDGES_B / 4);
}
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
meanTail.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer);
ctx.Synchronize();
streamPrimary.Synchronize();
count = new uint[2];
d_indexesA.CopyToHost(count, 0, 0, 8);
if (count[0] > 4194304)
{
// trouble
count[0] = 4194304;
// log
}
hAligned_a.AsyncCopyFromDevice(d_buffer.DevicePointer, 0, 0, count[0] * 8, streamPrimary.Stream);
streamPrimary.Synchronize();
System.Runtime.InteropServices.Marshal.Copy(hAligned_a.PinnedHostPointer, h_a, 0, ((int)count[0] * 8) / sizeof(int));
timer.Stop();
currentJob.solvedAt = DateTime.Now;
currentJob.trimTime = timer.ElapsedMilliseconds;
//Console.WriteLine("Trimmed in {0}ms to {1} edges", timer.ElapsedMilliseconds, count[0]);
Logger.Log(LogLevel.Info, string.Format("GPU NV{2}: Trimmed in {0}ms to {1} edges, h {3}", timer.ElapsedMilliseconds, count[0], deviceID, currentJob.height));
}while((currentJob.height != Comms.nextJob.height) && (!Comms.IsTerminated) && (!TEST));
if (TEST)
{
//Console.WriteLine("Trimmed in {0}ms to {1} edges", timer.ElapsedMilliseconds, count[0]);
CGraph cg = FinderBag.GetFinder();
if (cg == null)
{
continue;
}
cg.SetEdges(h_a, (int)count[0]);
cg.SetHeader(currentJob);
//currentJob = currentJob.Next();
Task.Factory.StartNew(() =>
{
Stopwatch sw = new Stopwatch();
sw.Start();
if (count[0] < 200000)
{
try
{
if (findersInFlight++ < 3)
{
Stopwatch cycleTime = new Stopwatch();
cycleTime.Start();
cg.FindSolutions(graphSolutions);
cycleTime.Stop();
AdjustTrims(cycleTime.ElapsedMilliseconds);
if (graphSolutions.Count > 0)
{
solutions++;
}
}
else
{
Logger.Log(LogLevel.Warning, "CPU overloaded!");
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Cycle finder error" + ex.Message);
}
finally
{
findersInFlight--;
FinderBag.ReturnFinder(cg);
}
}
sw.Stop();
if (++trims % 50 == 0)
{
Console.ForegroundColor = ConsoleColor.Green;
Console.WriteLine("SOLS: {0}/{1} - RATE: {2:F1}", solutions, trims, (float)trims / solutions);
Console.ResetColor();
}
//Console.WriteLine("Finder completed in {0}ms on {1} edges with {2} solution(s)", sw.ElapsedMilliseconds, count[0], graphSolutions.Count);
//Console.WriteLine("Duped edges: {0}", cg.dupes);
Logger.Log(LogLevel.Info, string.Format("Finder completed in {0}ms on {1} edges with {2} solution(s) and {3} dupes", sw.ElapsedMilliseconds, count[0], graphSolutions.Count, cg.dupes));
});
//h_indexesA = d_indexesA;
//h_indexesB = d_indexesB;
//var sumA = h_indexesA.Sum(e => e);
//var sumB = h_indexesB.Sum(e => e);
;
}
else
{
CGraph cg = FinderBag.GetFinder();
cg.SetEdges(h_a, (int)count[0]);
cg.SetHeader(currentJob);
Task.Factory.StartNew(() =>
{
if (count[0] < 200000)
{
try
{
if (findersInFlight++ < 3)
{
Stopwatch cycleTime = new Stopwatch();
cycleTime.Start();
cg.FindSolutions(graphSolutions);
cycleTime.Stop();
AdjustTrims(cycleTime.ElapsedMilliseconds);
if (graphSolutions.Count > 0)
{
solutions++;
}
}
else
{
Logger.Log(LogLevel.Warning, "CPU overloaded!");
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Cycle finder crashed: " + ex.Message);
}
finally
{
findersInFlight--;
FinderBag.ReturnFinder(cg);
}
}
});
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Critical error in main cuda loop " + ex.Message);
Task.Delay(5000).Wait();
}
}
// clean up
try
{
Task.Delay(500).Wait();
Comms.Close();
d_buffer.Dispose();
d_indexesA.Dispose();
d_indexesB.Dispose();
streamPrimary.Dispose();
streamSecondary.Dispose();
hAligned_a.Dispose();
if (ctx != null)
{
ctx.Dispose();
}
}
catch { }
}
static void Main(string[] args)
{
try
{
if (args.Length == 1 && args[0].ToLower().Contains("fidelity"))
{
string[] fseg = args[0].Split(':');
deviceID = int.Parse(fseg[1]);
nonce = Int64.Parse(fseg[2]) - 1;
range = int.Parse(fseg[3]);
QTEST = true;
}
else
{
if (args.Length > 0)
{
deviceID = int.Parse(args[0]);
}
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Device ID parse error: " + ex.Message);
}
try
{
if (args.Length > 0)
{
deviceID = int.Parse(args[0]);
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Device ID parse error");
}
try
{
if (args.Length > 1)
{
port = int.Parse(args[1]);
Comms.ConnectToMaster(port);
}
else
{
TEST = true;
Logger.CopyToConsole = true;
CGraph.ShowCycles = true;
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Master connection error");
}
try
{
if (args.Length > 3)
{
gpuCount = int.Parse(args[3]);
fastCuda = gpuCount <= (Environment.ProcessorCount / 2);
if (fastCuda)
{
Logger.Log(LogLevel.Info, "Using single GPU blocking mode");
}
}
}
catch
{
}
if (TEST)
{
currentJob = nextJob = new Job()
{
jobID = 0,
k0 = 0xf4956dc403730b01L,
k1 = 0xe6d45de39c2a5a3eL,
k2 = 0xcbf626a8afee35f6L,
k3 = 0x4307b94b1a0c9980L,
pre_pow = TestPrePow,
timestamp = DateTime.Now
};
}
else
{
currentJob = nextJob = new Job()
{
jobID = 0,
k0 = 0xf4956dc403730b01L,
k1 = 0xe6d45de39c2a5a3eL,
k2 = 0xcbf626a8afee35f6L,
k3 = 0x4307b94b1a0c9980L,
pre_pow = TestPrePow,
timestamp = DateTime.Now
};
if (!Comms.IsConnected())
{
Console.WriteLine("Master connection failed, aborting");
Logger.Log(LogLevel.Error, "No master connection, exitting!");
return;
}
if (deviceID < 0)
{
int devCnt = CudaContext.GetDeviceCount();
GpuDevicesMessage gpum = new GpuDevicesMessage()
{
devices = new List <GpuDevice>(devCnt)
};
for (int i = 0; i < devCnt; i++)
{
string name = CudaContext.GetDeviceName(i);
var info = CudaContext.GetDeviceInfo(i);
gpum.devices.Add(new GpuDevice()
{
deviceID = i, name = name, memory = info.TotalGlobalMemory
});
}
//Console.WriteLine(devCnt);
Comms.gpuMsg = gpum;
Comms.SetEvent();
//Console.WriteLine("event fired");
Task.Delay(1000).Wait();
//Console.WriteLine("closing");
Comms.Close();
return;
}
}
try
{
var assembly = Assembly.GetEntryAssembly();
var resourceStream = assembly.GetManifestResourceStream("CudaSolver.kernel_x64.ptx");
ctx = new CudaContext(deviceID, /*!fastCuda ? (CUCtxFlags.BlockingSync | CUCtxFlags.MapHost) :*/ CUCtxFlags.MapHost);
string pow = new StreamReader(resourceStream).ReadToEnd();
//pow = File.ReadAllText(@"kernel_x64.ptx");
Turing = ctx.GetDeviceInfo().MaxSharedMemoryPerMultiprocessor == 65536;
using (var s = GenerateStreamFromString(pow))
{
if (!Turing)
{
meanSeedA = ctx.LoadKernelPTX(s, "FluffySeed4K", new CUJITOption[] { CUJITOption.MaxRegisters }, new object[] { (uint)40 });
meanSeedA.BlockDimensions = 512;
meanSeedA.GridDimensions = 1024;
meanSeedA.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRound = ctx.LoadKernelPTX(s, "FluffyRound_A2", new CUJITOption[] { CUJITOption.MaxRegisters }, new object[] { (uint)40 });
meanRound.BlockDimensions = 512;
meanRound.GridDimensions = 4096;
meanRound.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRound_4 = ctx.LoadKernelPTX(s, "FluffyRound_A1", new CUJITOption[] { CUJITOption.MaxRegisters }, new object[] { (uint)32 });
meanRound_4.BlockDimensions = 1024;
meanRound_4.GridDimensions = 1024;
meanRound_4.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRoundJoin = ctx.LoadKernelPTX(s, "FluffyRound_A3", new CUJITOption[] { CUJITOption.MaxRegisters }, new object[] { (uint)32 });
meanRoundJoin.BlockDimensions = 1024;
meanRoundJoin.GridDimensions = 4096;
meanRoundJoin.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanTail = ctx.LoadKernelPTX(s, "FluffyTail");
meanTail.BlockDimensions = 1024;
meanTail.GridDimensions = 4096;
meanTail.PreferredSharedMemoryCarveout = CUshared_carveout.MaxL1;
meanRecover = ctx.LoadKernelPTX(s, "FluffyRecovery");
meanRecover.BlockDimensions = 256;
meanRecover.GridDimensions = 2048;
meanRecover.PreferredSharedMemoryCarveout = CUshared_carveout.MaxL1;
}
else
{
meanSeedA = ctx.LoadKernelPTX(s, "FluffySeed4K", new CUJITOption[] { CUJITOption.MaxRegisters }, new object[] { (uint)64 });
meanSeedA.BlockDimensions = 512;
meanSeedA.GridDimensions = 1024;
meanSeedA.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRound = ctx.LoadKernelPTX(s, "FluffyRound_C2", new CUJITOption[] { CUJITOption.MaxRegisters }, new object[] { (uint)32 });
meanRound.BlockDimensions = 1024;
meanRound.GridDimensions = 4096;
meanRound.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRound_4 = ctx.LoadKernelPTX(s, "FluffyRound_C1", new CUJITOption[] { CUJITOption.MaxRegisters }, new object[] { (uint)64 });
meanRound_4.BlockDimensions = 1024;
meanRound_4.GridDimensions = 1024;
meanRound_4.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanRoundJoin = ctx.LoadKernelPTX(s, "FluffyRound_C3", new CUJITOption[] { CUJITOption.MaxRegisters }, new object[] { (uint)32 });
meanRoundJoin.BlockDimensions = 1024;
meanRoundJoin.GridDimensions = 4096;
meanRoundJoin.PreferredSharedMemoryCarveout = CUshared_carveout.MaxShared;
meanTail = ctx.LoadKernelPTX(s, "FluffyTail");
meanTail.BlockDimensions = 1024;
meanTail.GridDimensions = 4096;
meanTail.PreferredSharedMemoryCarveout = CUshared_carveout.MaxL1;
meanRecover = ctx.LoadKernelPTX(s, "FluffyRecovery");
meanRecover.BlockDimensions = 256;
meanRecover.GridDimensions = 2048;
meanRecover.PreferredSharedMemoryCarveout = CUshared_carveout.MaxL1;
}
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Unable to create kernels: " + ex.Message);
Task.Delay(500).Wait();
Comms.Close();
return;
}
try
{
d_buffer = new CudaDeviceVariable <ulong>(BUFFER_SIZE_U32 * (temp ? 8 : 1));
d_bufferMid = new CudaDeviceVariable <ulong>(d_buffer.DevicePointer + (BUFFER_SIZE_B * 2));
d_bufferB = new CudaDeviceVariable <ulong>(d_buffer.DevicePointer + (BUFFER_SIZE_B * 8));
d_indexesA = new CudaDeviceVariable <uint>(INDEX_SIZE);
d_indexesB = new CudaDeviceVariable <uint>(INDEX_SIZE);
d_aux = new CudaDeviceVariable <uint>(512);
Array.Clear(h_indexesA, 0, h_indexesA.Length);
Array.Clear(h_indexesB, 0, h_indexesA.Length);
d_indexesA = h_indexesA;
d_indexesB = h_indexesB;
streamPrimary = new CudaStream(CUStreamFlags.NonBlocking);
}
catch (Exception ex)
{
Task.Delay(200).Wait();
Logger.Log(LogLevel.Error, $"Mem alloc exception. Out of video memory? {ctx.GetFreeDeviceMemorySize()} free");
Task.Delay(500).Wait();
Comms.Close();
return;
}
try
{
AllocateHostMemory(true, ref h_a, ref hAligned_a, 1024 * 1024 * 32);
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Unable to create pinned memory.");
Task.Delay(500).Wait();
Comms.Close();
return;
}
int loopCnt = 0;
while (!Comms.IsTerminated)
{
try
{
if (!TEST && (Comms.nextJob.pre_pow == null || Comms.nextJob.pre_pow == "" || Comms.nextJob.pre_pow == TestPrePow))
{
Logger.Log(LogLevel.Info, string.Format("Waiting for job...."));
Task.Delay(1000).Wait();
continue;
}
if (!TEST && ((currentJob.pre_pow != Comms.nextJob.pre_pow) || (currentJob.origin != Comms.nextJob.origin)))
{
currentJob = Comms.nextJob;
currentJob.timestamp = DateTime.Now;
}
if (!TEST && (currentJob.timestamp.AddMinutes(30) < DateTime.Now) && Comms.lastIncoming.AddMinutes(30) < DateTime.Now)
{
Logger.Log(LogLevel.Info, string.Format("Job too old..."));
Task.Delay(1000).Wait();
continue;
}
// test runs only once
if (TEST && ++loopCnt >= range)
{
Comms.IsTerminated = true;
}
Solution s;
while (graphSolutions.TryDequeue(out s))
{
meanRecover.SetConstantVariable <ulong>("recovery", s.GetUlongEdges());
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRecover.RunAsync(streamPrimary.Stream, s.job.k0, s.job.k1, s.job.k2, s.job.k3, d_indexesB.DevicePointer);
streamPrimary.Synchronize();
s.nonces = new uint[32];
d_indexesB.CopyToHost(s.nonces, 0, 0, 32 * 4);
s.nonces = s.nonces.OrderBy(n => n).ToArray();
//fidelity = (32-cycles_found / graphs_searched) * 32
solutions++;
s.fidelity = ((double)solutions / (double)trims) * 32.0;
//Console.WriteLine(s.fidelity.ToString("0.000"));
if (Comms.IsConnected())
{
Comms.graphSolutionsOut.Enqueue(s);
Comms.SetEvent();
}
if (QTEST)
{
Console.ForegroundColor = ConsoleColor.Red;
Console.WriteLine($"Solution for nonce {s.job.nonce}: {string.Join(' ', s.nonces)}");
Console.ResetColor();
}
}
if (QTEST)
{
currentJob = currentJob.NextSequential(ref nonce);
Console.WriteLine($"Nonce: {nonce} K0: {currentJob.k0:X} K1: {currentJob.k1:X} K2: {currentJob.k2:X} K3: {currentJob.k3:X}");
}
else
{
currentJob = currentJob.Next();
}
Logger.Log(LogLevel.Debug, string.Format("GPU NV{4}:Trimming #{4}: {0} {1} {2} {3}", currentJob.k0, currentJob.k1, currentJob.k2, currentJob.k3, currentJob.jobID, deviceID));
timer.Restart();
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
d_aux.MemsetAsync(0, streamPrimary.Stream);
meanSeedA.RunAsync(streamPrimary.Stream, currentJob.k0, currentJob.k1, currentJob.k2, currentJob.k3, d_bufferMid.DevicePointer, d_indexesB.DevicePointer, 0);
meanSeedA.RunAsync(streamPrimary.Stream, currentJob.k0, currentJob.k1, currentJob.k2, currentJob.k3, d_bufferMid.DevicePointer + ((BUFFER_SIZE_A * 8) / 4 / 4) * 1, d_indexesB.DevicePointer + (4096 * 4), EDGE_SEG);
meanSeedA.RunAsync(streamPrimary.Stream, currentJob.k0, currentJob.k1, currentJob.k2, currentJob.k3, d_bufferMid.DevicePointer + ((BUFFER_SIZE_A * 8) / 4 / 4) * 2, d_indexesB.DevicePointer + (4096 * 8), EDGE_SEG * 2);
meanSeedA.RunAsync(streamPrimary.Stream, currentJob.k0, currentJob.k1, currentJob.k2, currentJob.k3, d_bufferMid.DevicePointer + ((BUFFER_SIZE_A * 8) / 4 / 4) * 3, d_indexesB.DevicePointer + (4096 * 12), EDGE_SEG * 3);
meanRound_4.RunAsync(streamPrimary.Stream, d_bufferMid.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_A / 4, DUCK_EDGES_B / 4, 0);
meanRound_4.RunAsync(streamPrimary.Stream, d_bufferMid.DevicePointer, d_buffer.DevicePointer + ((BUFFER_SIZE_B * 8) / 4) * 1, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_A / 4, DUCK_EDGES_B / 4, 1024);
meanRound_4.RunAsync(streamPrimary.Stream, d_bufferMid.DevicePointer, d_buffer.DevicePointer + ((BUFFER_SIZE_B * 8) / 4) * 2, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_A / 4, DUCK_EDGES_B / 4, 2048);
meanRound_4.RunAsync(streamPrimary.Stream, d_bufferMid.DevicePointer, d_buffer.DevicePointer + ((BUFFER_SIZE_B * 8) / 4) * 3, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_A / 4, DUCK_EDGES_B / 4, 3072);
//streamPrimary.Synchronize();
//h_indexesA = d_indexesA;
//h_indexesB = d_indexesB;
//var sumA = h_indexesA.Sum(e => e);
//var sumB = h_indexesB.Sum(e => e);
//streamPrimary.Synchronize();
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRoundJoin.RunAsync(streamPrimary.Stream,
d_buffer.DevicePointer,
d_buffer.DevicePointer + ((BUFFER_SIZE_B * 8) / 4) * 1,
d_buffer.DevicePointer + ((BUFFER_SIZE_B * 8) / 4) * 2,
d_buffer.DevicePointer + ((BUFFER_SIZE_B * 8) / 4) * 3,
d_bufferB.DevicePointer,
d_indexesA.DevicePointer,
d_indexesB.DevicePointer, DUCK_EDGES_B / 4, DUCK_EDGES_B / 2);
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_B / 2, DUCK_EDGES_B / 2, 0, d_aux.DevicePointer);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_buffer.DevicePointer, d_bufferB.DevicePointer, d_indexesA.DevicePointer, d_indexesB.DevicePointer, DUCK_EDGES_B / 2, DUCK_EDGES_B / 2, 1, d_aux.DevicePointer);
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_B / 2, DUCK_EDGES_B / 2, 2, d_aux.DevicePointer);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_buffer.DevicePointer, d_bufferB.DevicePointer, d_indexesA.DevicePointer, d_indexesB.DevicePointer, DUCK_EDGES_B / 2, DUCK_EDGES_B / 4, 3, d_aux.DevicePointer);
for (int i = 0; i < (TEST ? 80 : trimRounds); i++)
//for (int i = 0; i < 85; i++)
{
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer, DUCK_EDGES_B / 4, DUCK_EDGES_B / 4, i * 2 + 4, d_aux.DevicePointer);
d_indexesB.MemsetAsync(0, streamPrimary.Stream);
meanRound.RunAsync(streamPrimary.Stream, d_buffer.DevicePointer, d_bufferB.DevicePointer, d_indexesA.DevicePointer, d_indexesB.DevicePointer, DUCK_EDGES_B / 4, DUCK_EDGES_B / 4, i * 2 + 5, d_aux.DevicePointer);
}
d_indexesA.MemsetAsync(0, streamPrimary.Stream);
meanTail.RunAsync(streamPrimary.Stream, d_bufferB.DevicePointer, d_buffer.DevicePointer, d_indexesB.DevicePointer, d_indexesA.DevicePointer);
Task.Delay((int)lastTrimMs).Wait();
streamPrimary.Synchronize();
uint[] count = new uint[2];
d_indexesA.CopyToHost(count, 0, 0, 8);
if (count[0] > 131071)
{
// trouble
count[0] = 131071;
// log
}
hAligned_a.AsyncCopyFromDevice(d_buffer.DevicePointer, 0, 0, count[0] * 8, streamPrimary.Stream);
streamPrimary.Synchronize();
System.Runtime.InteropServices.Marshal.Copy(hAligned_a.PinnedHostPointer, h_a, 0, ((int)count[0] * 8) / sizeof(int));
trims++;
timer.Stop();
lastTrimMs = (long)Math.Min(Math.Max((float)timer.ElapsedMilliseconds * 0.9f, 50), 500);
currentJob.solvedAt = DateTime.Now;
currentJob.trimTime = timer.ElapsedMilliseconds;
//Console.WriteLine("Trimmed in {0}ms to {1} edges", timer.ElapsedMilliseconds, count[0]);
Logger.Log(LogLevel.Info, string.Format("GPU NV{2}: Trimmed in {0}ms to {1} edges", timer.ElapsedMilliseconds, count[0], deviceID));
FinderBag.RunFinder(TEST, ref trims, count[0], h_a, currentJob, graphSolutions, timer);
if (trims % 50 == 0 && TEST)
{
Console.ForegroundColor = ConsoleColor.Green;
Console.WriteLine("SOLS: {0}/{1} - RATE: {2:F1}", solutions, trims, (float)trims / solutions);
Console.ResetColor();
}
/*
* if (TEST)
* {
* //Console.WriteLine("Trimmed in {0}ms to {1} edges", timer.ElapsedMilliseconds, count[0]);
*
* CGraph cg = FinderBag.GetFinder();
* cg.SetEdges(h_a, (int)count[0]);
* cg.SetHeader(currentJob);
*
* //currentJob = currentJob.Next();
*
* Task.Factory.StartNew(() =>
* {
* Stopwatch sw = new Stopwatch();
* sw.Start();
*
* if (count[0] < 131071)
* {
* try
* {
* if (findersInFlight++ < 3)
* {
* Stopwatch cycleTime = new Stopwatch();
* cycleTime.Start();
* cg.FindSolutions(graphSolutions);
* cycleTime.Stop();
* AdjustTrims(cycleTime.ElapsedMilliseconds);
* //if (graphSolutions.Count > 0) solutions++;
* }
* else
* Logger.Log(LogLevel.Warning, "CPU overloaded!");
* }
* catch (Exception ex)
* {
* Logger.Log(LogLevel.Error, "Cycle finder error" + ex.Message);
* }
* finally
* {
* FinderBag.ReturnFinder(cg);
* findersInFlight--;
* }
* }
*
* sw.Stop();
*
* if (trims % 50 == 0)
* {
* Console.ForegroundColor = ConsoleColor.Green;
* Console.WriteLine("SOLS: {0}/{1} - RATE: {2:F1}", solutions, trims, (float)trims/solutions );
* Console.ResetColor();
* }
* //Console.WriteLine("Finder completed in {0}ms on {1} edges with {2} solution(s)", sw.ElapsedMilliseconds, count[0], graphSolutions.Count);
* //Console.WriteLine("Duped edges: {0}", cg.dupes);
* if (!QTEST)
* Logger.Log(LogLevel.Info, string.Format("Finder completed in {0}ms on {1} edges with {2} solution(s) and {3} dupes", sw.ElapsedMilliseconds, count[0], graphSolutions.Count, cg.dupes));
* });
*
* //h_indexesA = d_indexesA;
* //h_indexesB = d_indexesB;
*
* //var sumA = h_indexesA.Sum(e => e);
* //var sumB = h_indexesB.Sum(e => e);
*
* ;
* }
* else
* {
* CGraph cg = FinderBag.GetFinder();
* cg.SetEdges(h_a, (int)count[0]);
* cg.SetHeader(currentJob);
*
* Task.Factory.StartNew(() =>
* {
* if (count[0] < 131071)
* {
* try
* {
* if (findersInFlight++ < 3)
* {
* Stopwatch cycleTime = new Stopwatch();
* cycleTime.Start();
* cg.FindSolutions(graphSolutions);
* cycleTime.Stop();
* AdjustTrims(cycleTime.ElapsedMilliseconds);
* }
* else
* Logger.Log(LogLevel.Warning, "CPU overloaded!");
* }
* catch (Exception ex)
* {
* Logger.Log(LogLevel.Warning, "Cycle finder crashed: " + ex.Message);
* }
* finally
* {
* FinderBag.ReturnFinder(cg);
* findersInFlight--;
* }
* }
* });
* }
*
*/
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Critical error in main cuda loop " + ex.Message);
Task.Delay(500).Wait();
break;
}
}
// clean up
try
{
Task.Delay(500).Wait();
Comms.Close();
d_buffer.Dispose();
d_indexesA.Dispose();
d_indexesB.Dispose();
d_aux.Dispose();
streamPrimary.Dispose();
streamSecondary.Dispose();
hAligned_a.Dispose();
if (ctx != null)
{
ctx.Dispose();
}
}
catch { }
}
