public void LoadKernel(string filepath)
{
this.Log("Loading File: " + filepath, DebugChannel.Log, 8);
string path = Path.GetFullPath(filepath);
string source = PPAPI.PreprocessSource(path, null);
string[] kernelNames = FindKernelNames(source);
foreach (var kernelName in kernelNames)
{
this.Log("Trying to Load Kernel: " + kernelName, DebugChannel.Log, 8);
if (_kernels.ContainsKey(kernelName))
{
this.Log("Kernel " + kernelName + " was already cached.", DebugChannel.Log, 7);
return;
}
this.Log("Building Program from source..", DebugChannel.Log, 8);
CLProgram p = c.CreateAndBuildProgramFromString(source);
int kernelNameIndex = source.IndexOf(" " + kernelName + " ");
kernelNameIndex = (kernelNameIndex == -1) ? source.IndexOf(" " + kernelName + "(") : kernelNameIndex;
_kernels.Add(kernelName,
new CLFilter(kernelName, p.CreateKernel(kernelName),
KernelParameter.CreateKernelParametersFromKernelCode(source, kernelNameIndex,
source.Substring(kernelNameIndex, source.Length - kernelNameIndex).IndexOf(')') + 1)));
}
}
public void LoadKernel(string filepath, string kernel)
{
string path = Path.GetFullPath(filepath);
if (_kernels.ContainsKey(path))
{
return;
}
CLProgram p = c.CreateAndBuildProgramFromFile(filepath);
_kernels.Add(path + "." + kernel, p.CreateKernel(kernel));
}
public void LoadKernel(string filepath, string kernel)
{
this.Log("Loading File: " + filepath, DebugChannel.Log, 8);
string path = Path.GetFullPath(filepath);
this.Log("Trying to Load Kernel:" + kernel, DebugChannel.Log, 8);
if (_kernels.ContainsKey(kernel))
{
this.Log("Kernel " + kernel + " was already cached.", DebugChannel.Log, 7);
return;
}
this.Log("Building Program from source..", DebugChannel.Log, 8);
string source = File.ReadAllText(filepath);
CLProgram p = c.CreateAndBuildProgramFromString(source);
int kernelNameIndex = source.IndexOf(kernel);
_kernels.Add(kernel, new CLFilter(kernel, p.CreateKernel(kernel),
KernelParameter.CreateKernelParametersFromKernelCode(source, kernelNameIndex,
source.Substring(kernelNameIndex, source.Length - kernelNameIndex).IndexOf(')') + 1)));
}
static void Main(string[] args)
{
try
{
if (args.Length > 0)
{
deviceID = int.Parse(args[0]);
}
if (args.Length > 2)
{
platformID = int.Parse(args[2]);
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Device ID parse error", ex);
}
try
{
if (args.Length > 1)
{
port = int.Parse(args[1]);
Comms.ConnectToMaster(port);
}
else
{
TEST = true;
CGraph.ShowCycles = true;
Logger.CopyToConsole = true;
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Master connection error");
}
// Gets all available platforms and their corresponding devices, and prints them out in a table
List <Platform> platforms = null;
try
{
platforms = Platform.GetPlatforms().ToList();
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Failed to get OpenCL platform list");
return;
}
if (TEST)
{
currentJob = nextJob = new Job()
{
jobID = 0,
k0 = 0xf4956dc403730b01L,
k1 = 0xe6d45de39c2a5a3eL,
k2 = 0xcbf626a8afee35f6L,
k3 = 0x4307b94b1a0c9980L,
//k0 = 0x10ef16eadd6aa061L,
//k1 = 0x563f07e7a3c788b3L,
//k2 = 0xe8d7c8db1518f29aL,
//k3 = 0xc0ab7d1b4ca1adffL,
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!");
Task.Delay(500).Wait();
return;
}
if (deviceID < 0)
{
try
{
//Environment.SetEnvironmentVariable("GPU_FORCE_64BIT_PTR", "1", EnvironmentVariableTarget.Machine);
Environment.SetEnvironmentVariable("GPU_MAX_HEAP_SIZE", "100", EnvironmentVariableTarget.User);
Environment.SetEnvironmentVariable("GPU_USE_SYNC_OBJECTS", "1", EnvironmentVariableTarget.User);
Environment.SetEnvironmentVariable("GPU_MAX_ALLOC_PERCENT", "100", EnvironmentVariableTarget.User);
Environment.SetEnvironmentVariable("GPU_SINGLE_ALLOC_PERCENT", "100", EnvironmentVariableTarget.User);
Environment.SetEnvironmentVariable("GPU_64BIT_ATOMICS", "1", EnvironmentVariableTarget.User);
Environment.SetEnvironmentVariable("GPU_MAX_WORKGROUP_SIZE", "1024", EnvironmentVariableTarget.User);
//Environment.SetEnvironmentVariable("AMD_OCL_BUILD_OPTIONS_APPEND", "-cl-std=CL2.0", EnvironmentVariableTarget.Machine);
GpuDevicesMessage gpum = new GpuDevicesMessage()
{
devices = new List <GpuDevice>()
};
//foreach (Platform platform in platforms)
for (int p = 0; p < platforms.Count(); p++)
{
Platform platform = platforms[p];
var devices = platform.GetDevices(DeviceType.Gpu).ToList();
//foreach (Device device in platform.GetDevices(DeviceType.All))
for (int d = 0; d < devices.Count(); d++)
{
Device device = devices[d];
string name = device.Name;
string pName = platform.Name;
//Console.WriteLine(device.Name + " " + platform.Version.VersionString);
gpum.devices.Add(new GpuDevice()
{
deviceID = d, platformID = p, platformName = pName, name = name, memory = device.GlobalMemorySize
});
}
}
Comms.gpuMsg = gpum;
Comms.SetEvent();
Task.Delay(1000).Wait();
Comms.Close();
return;
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Unable to enumerate OpenCL devices");
Task.Delay(500).Wait();
Comms.Close();
return;
}
}
}
try
{
Device chosenDevice = null;
try
{
chosenDevice = platforms[platformID].GetDevices(DeviceType.Gpu).ToList()[deviceID];
Console.WriteLine($"Using OpenCL device: {chosenDevice.Name} ({chosenDevice.Vendor})");
Console.WriteLine();
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, $"Unable to select OpenCL device {deviceID} on platform {platformID} ");
Task.Delay(500).Wait();
Comms.Close();
return;
}
var assembly = Assembly.GetEntryAssembly();
var resourceStream = assembly.GetManifestResourceStream("OclSolver.kernel.cl");
using (StreamReader reader = new StreamReader(resourceStream))
{
using (Context context = Context.CreateContext(chosenDevice))
{
/*
* Once the program has been created you can use clGetProgramInfo with CL_PROGRAM_BINARY_SIZES and then CL_PROGRAM_BINARIES, storing the resulting binary programs (one for each device of the context) into a buffer you supply. You can then save this binary data to disk for use in later runs.
* Not all devices might support binaries, so you will need to check the CL_PROGRAM_BINARY_SIZES result (it returns a zero size for that device if binaries are not supported).
*/
using (OpenCl.DotNetCore.Programs.Program program = context.CreateAndBuildProgramFromString(reader.ReadToEnd()))
{
using (CommandQueue commandQueue = CommandQueue.CreateCommandQueue(context, chosenDevice))
{
IntPtr clearPattern = IntPtr.Zero;
uint[] edgesCount;
int[] edgesLeft;
int trims = 0;
try
{
clearPattern = Marshal.AllocHGlobal(4);
Marshal.Copy(new byte[4] {
0, 0, 0, 0
}, 0, clearPattern, 4);
try
{
bufferA1 = context.CreateBuffer <uint>(MemoryFlag.ReadWrite, BUFFER_SIZE_A1);
bufferA2 = context.CreateBuffer <uint>(MemoryFlag.ReadWrite, BUFFER_SIZE_A2);
bufferB = context.CreateBuffer <uint>(MemoryFlag.ReadWrite, BUFFER_SIZE_B);
bufferI1 = context.CreateBuffer <uint>(MemoryFlag.ReadWrite, INDEX_SIZE);
bufferI2 = context.CreateBuffer <uint>(MemoryFlag.ReadWrite, INDEX_SIZE);
bufferR = context.CreateBuffer <uint>(MemoryFlag.ReadOnly, 42 * 2);
}
catch (Exception ex)
{
Task.Delay(500).Wait();
Logger.Log(LogLevel.Error, "Unable to allocate buffers, out of memory?");
Task.Delay(500).Wait();
Comms.Close();
return;
}
using (Kernel kernelSeedA = program.CreateKernel("FluffySeed2A"))
using (Kernel kernelSeedB1 = program.CreateKernel("FluffySeed2B"))
using (Kernel kernelSeedB2 = program.CreateKernel("FluffySeed2B"))
using (Kernel kernelRound1 = program.CreateKernel("FluffyRound1"))
using (Kernel kernelRoundO = program.CreateKernel("FluffyRoundNO1"))
using (Kernel kernelRoundNA = program.CreateKernel("FluffyRoundNON"))
using (Kernel kernelRoundNB = program.CreateKernel("FluffyRoundNON"))
using (Kernel kernelTail = program.CreateKernel("FluffyTailO"))
using (Kernel kernelRecovery = program.CreateKernel("FluffyRecovery"))
{
Stopwatch sw = new Stopwatch();
kernelSeedA.SetKernelArgumentGeneric(0, currentJob.k0);
kernelSeedA.SetKernelArgumentGeneric(1, currentJob.k1);
kernelSeedA.SetKernelArgumentGeneric(2, currentJob.k2);
kernelSeedA.SetKernelArgumentGeneric(3, currentJob.k3);
kernelSeedA.SetKernelArgument(4, bufferB);
kernelSeedA.SetKernelArgument(5, bufferA1);
kernelSeedA.SetKernelArgument(6, bufferI1);
kernelSeedB1.SetKernelArgument(0, bufferA1);
kernelSeedB1.SetKernelArgument(1, bufferA1);
kernelSeedB1.SetKernelArgument(2, bufferA2);
kernelSeedB1.SetKernelArgument(3, bufferI1);
kernelSeedB1.SetKernelArgument(4, bufferI2);
kernelSeedB1.SetKernelArgumentGeneric(5, (uint)32);
kernelSeedB2.SetKernelArgument(0, bufferB);
kernelSeedB2.SetKernelArgument(1, bufferA1);
kernelSeedB2.SetKernelArgument(2, bufferA2);
kernelSeedB2.SetKernelArgument(3, bufferI1);
kernelSeedB2.SetKernelArgument(4, bufferI2);
kernelSeedB2.SetKernelArgumentGeneric(5, (uint)0);
kernelRound1.SetKernelArgument(0, bufferA1);
kernelRound1.SetKernelArgument(1, bufferA2);
kernelRound1.SetKernelArgument(2, bufferB);
kernelRound1.SetKernelArgument(3, bufferI2);
kernelRound1.SetKernelArgument(4, bufferI1);
kernelRound1.SetKernelArgumentGeneric(5, (uint)DUCK_SIZE_A * 1024);
kernelRound1.SetKernelArgumentGeneric(6, (uint)DUCK_SIZE_B * 1024);
kernelRoundO.SetKernelArgument(0, bufferB);
kernelRoundO.SetKernelArgument(1, bufferA1);
kernelRoundO.SetKernelArgument(2, bufferI1);
kernelRoundO.SetKernelArgument(3, bufferI2);
kernelRoundNA.SetKernelArgument(0, bufferB);
kernelRoundNA.SetKernelArgument(1, bufferA1);
kernelRoundNA.SetKernelArgument(2, bufferI1);
kernelRoundNA.SetKernelArgument(3, bufferI2);
kernelRoundNB.SetKernelArgument(0, bufferA1);
kernelRoundNB.SetKernelArgument(1, bufferB);
kernelRoundNB.SetKernelArgument(2, bufferI2);
kernelRoundNB.SetKernelArgument(3, bufferI1);
kernelTail.SetKernelArgument(0, bufferB);
kernelTail.SetKernelArgument(1, bufferA1);
kernelTail.SetKernelArgument(2, bufferI1);
kernelTail.SetKernelArgument(3, bufferI2);
kernelRecovery.SetKernelArgumentGeneric(0, currentJob.k0);
kernelRecovery.SetKernelArgumentGeneric(1, currentJob.k1);
kernelRecovery.SetKernelArgumentGeneric(2, currentJob.k2);
kernelRecovery.SetKernelArgumentGeneric(3, currentJob.k3);
kernelRecovery.SetKernelArgument(4, bufferR);
kernelRecovery.SetKernelArgument(5, bufferI2);
int loopCnt = 0;
//for (int i = 0; i < runs; i++)
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++ > 100000)
{
Comms.IsTerminated = true;
}
Logger.Log(LogLevel.Debug, string.Format("GPU AMD{4}:Trimming #{4}: {0} {1} {2} {3}", currentJob.k0, currentJob.k1, currentJob.k2, currentJob.k3, currentJob.jobID, deviceID));
//Stopwatch srw = new Stopwatch();
//srw.Start();
Solution s;
while (graphSolutions.TryDequeue(out s))
{
kernelRecovery.SetKernelArgumentGeneric(0, s.job.k0);
kernelRecovery.SetKernelArgumentGeneric(1, s.job.k1);
kernelRecovery.SetKernelArgumentGeneric(2, s.job.k2);
kernelRecovery.SetKernelArgumentGeneric(3, s.job.k3);
commandQueue.EnqueueWriteBufferEdges(bufferR, s.GetLongEdges());
commandQueue.EnqueueClearBuffer(bufferI2, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueNDRangeKernel(kernelRecovery, 1, 2048 * 256, 256, 0);
s.nonces = commandQueue.EnqueueReadBuffer <uint>(bufferI2, 42);
OpenCl.DotNetCore.Interop.CommandQueues.CommandQueuesNativeApi.Finish(commandQueue.Handle);
s.nonces = s.nonces.OrderBy(n => n).ToArray();
Comms.graphSolutionsOut.Enqueue(s);
Comms.SetEvent();
}
//srw.Stop();
//Console.WriteLine("RECOVERY " + srw.ElapsedMilliseconds);
currentJob = currentJob.Next();
kernelSeedA.SetKernelArgumentGeneric(0, currentJob.k0);
kernelSeedA.SetKernelArgumentGeneric(1, currentJob.k1);
kernelSeedA.SetKernelArgumentGeneric(2, currentJob.k2);
kernelSeedA.SetKernelArgumentGeneric(3, currentJob.k3);
sw.Restart();
commandQueue.EnqueueClearBuffer(bufferI2, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueClearBuffer(bufferI1, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueNDRangeKernel(kernelSeedA, 1, 2048 * 128, 128, 0);
commandQueue.EnqueueNDRangeKernel(kernelSeedB1, 1, 1024 * 128, 128, 0);
commandQueue.EnqueueNDRangeKernel(kernelSeedB2, 1, 1024 * 128, 128, 0);
commandQueue.EnqueueClearBuffer(bufferI1, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueNDRangeKernel(kernelRound1, 1, 4096 * 1024, 1024, 0);
commandQueue.EnqueueClearBuffer(bufferI2, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueNDRangeKernel(kernelRoundO, 1, 4096 * 1024, 1024, 0);
commandQueue.EnqueueClearBuffer(bufferI1, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueNDRangeKernel(kernelRoundNB, 1, 4096 * 1024, 1024, 0);
for (int r = 0; r < trimRounds; r++)
{
commandQueue.EnqueueClearBuffer(bufferI2, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueNDRangeKernel(kernelRoundNA, 1, 4096 * 1024, 1024, 0);
commandQueue.EnqueueClearBuffer(bufferI1, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueNDRangeKernel(kernelRoundNB, 1, 4096 * 1024, 1024, 0);
}
commandQueue.EnqueueClearBuffer(bufferI2, 64 * 64 * 4, clearPattern);
commandQueue.EnqueueNDRangeKernel(kernelTail, 1, 4096 * 1024, 1024, 0);
edgesCount = commandQueue.EnqueueReadBuffer <uint>(bufferI2, 1);
edgesCount[0] = edgesCount[0] > 1000000 ? 1000000 : edgesCount[0];
edgesLeft = commandQueue.EnqueueReadBuffer(bufferA1, (int)edgesCount[0] * 2);
OpenCl.DotNetCore.Interop.CommandQueues.CommandQueuesNativeApi.Flush(commandQueue.Handle);
OpenCl.DotNetCore.Interop.CommandQueues.CommandQueuesNativeApi.Finish(commandQueue.Handle);
sw.Stop();
currentJob.trimTime = sw.ElapsedMilliseconds;
currentJob.solvedAt = DateTime.Now;
Logger.Log(LogLevel.Info, string.Format("GPU AMD{2}: Trimmed in {0}ms to {1} edges", sw.ElapsedMilliseconds, edgesCount[0], deviceID));
CGraph cg = new CGraph();
cg.SetEdges(edgesLeft, (int)edgesCount[0]);
cg.SetHeader(currentJob);
Task.Factory.StartNew(() =>
{
if (edgesCount[0] < 200000)
{
try
{
if (findersInFlight++ < 3)
{
Stopwatch cycleTime = new Stopwatch();
cycleTime.Start();
cg.FindSolutions(graphSolutions);
cycleTime.Stop();
AdjustTrims(cycleTime.ElapsedMilliseconds);
if (TEST)
{
Logger.Log(LogLevel.Info, string.Format("Finder completed in {0}ms on {1} edges with {2} solution(s) and {3} dupes", sw.ElapsedMilliseconds, edgesCount[0], graphSolutions.Count, cg.dupes));
if (++trims % 50 == 0)
{
Console.ForegroundColor = ConsoleColor.Green;
Console.WriteLine("SOLS: {0}/{1} - RATE: {2:F1}", solutions, trims, (float)trims / solutions);
Console.ResetColor();
}
}
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--;
}
}
});
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Critical error in main ocl loop " + ex.Message);
Task.Delay(5000).Wait();
}
}
//uint[] resultArray = commandQueue.EnqueueReadBuffer<uint>(bufferI1, 64 * 64);
//uint[] resultArray2 = commandQueue.EnqueueReadBuffer<uint>(bufferI2, 64 * 64);
//Console.WriteLine("SeedA: " + resultArray.Sum(e => e) + " in " + sw.ElapsedMilliseconds / runs);
//Console.WriteLine("SeedB: " + resultArray2.Sum(e => e) + " in " + sw.ElapsedMilliseconds / runs);
//Task.Delay(1000).Wait();
//Console.WriteLine("");
}
}
finally
{
// clear pattern
if (clearPattern != IntPtr.Zero)
{
Marshal.FreeHGlobal(clearPattern);
}
}
}
}
}
}
}
catch (Exception ex)
{
Logger.Log(LogLevel.Error, "Critical error in OCL Init " + ex.Message);
Console.ForegroundColor = ConsoleColor.Yellow;
Console.WriteLine(ex.Message);
Console.ResetColor();
Task.Delay(500).Wait();
}
finally
{
Task.Delay(500).Wait();
try
{
Comms.Close();
bufferA1.Dispose();
bufferA2.Dispose();
bufferB.Dispose();
bufferI1.Dispose();
bufferI2.Dispose();
bufferR.Dispose();
if (OpenCl.DotNetCore.CommandQueues.CommandQueue.resultValuePointer != IntPtr.Zero)
{
Marshal.FreeHGlobal(OpenCl.DotNetCore.CommandQueues.CommandQueue.resultValuePointer);
}
}
catch { }
}
//Console.ReadKey();
}
