public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var monteCarloAlgorithm = await hastlayer
.GenerateProxy(hardwareRepresentation, new MonteCarloAlgorithm());
var monteCarloResult = monteCarloAlgorithm.CalculateTorusSectionValues(5000000);
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var recursiveAlgorithms = await hastlayer.GenerateProxy(hardwareRepresentation, new RecursiveAlgorithms());
var fibonacci = recursiveAlgorithms.CalculateFibonacchiSeries(13); // 233
var factorial = recursiveAlgorithms.CalculateFactorial(6); // 720
}
public async Task <T> GenerateProxy <T>(
IHardwareRepresentation hardwareRepresentation,
T hardwareObject,
IProxyGenerationConfiguration configuration) where T : class
{
if (!hardwareRepresentation.SoftAssemblyPaths.Contains(hardwareObject.GetType().Assembly.Location))
{
throw new InvalidOperationException(
"The supplied type is not part of any assembly that this hardware representation was generated from.");
}
try
{
return(await
_host
.RunGet(scope => Task.Run(() => scope.Resolve <IProxyGenerator>().CreateCommunicationProxy(hardwareRepresentation, hardwareObject, configuration))));
}
catch (Exception ex) when(!ex.IsFatal())
{
var message =
"An error happened during generating the Hastlayer proxy for an object of the following type: " +
hardwareObject.GetType().FullName;
await _host.Run <ILoggerService>(logger => Task.Run(() => logger.Error(ex, message)));
throw new HastlayerException(message, ex);
}
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var ooShowcase = await hastlayer
.GenerateProxy(hardwareRepresentation, new ObjectOrientedShowcase());
var sum = ooShowcase.Run(93); // 293
}
public ExecutedOnHardwareEventArgs(
IHardwareRepresentation hardwareRepresentation,
string memberFullName,
IHardwareExecutionInformation hardwareExecutionInformation)
{
_hardwareRepresentation = hardwareRepresentation;
_memberFullName = memberFullName;
_hardwareExecutionInformation = hardwareExecutionInformation;
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var loopback = await hastlayer.GenerateProxy(hardwareRepresentation, new Loopback());
var output1 = loopback.Run(123);
var output2 = loopback.Run(1234);
var output3 = loopback.Run(-9);
var output4 = loopback.Run(0);
var output5 = loopback.Run(-19);
var output6 = loopback.Run(1);
}
public T CreateCommunicationProxy <T>(IHardwareRepresentation hardwareRepresentation, T target, IProxyGenerationConfiguration configuration) where T : class
{
var memberInvocationHandler = _memberInvocationHandlerFactory.CreateMemberInvocationHandler(hardwareRepresentation, target, configuration);
if (typeof(T).IsInterface)
{
return(_proxyGenerator.CreateInterfaceProxyWithTarget(target, new MemberInvocationInterceptor(memberInvocationHandler)));
}
return(_proxyGenerator.CreateClassProxyWithTarget(target, new MemberInvocationInterceptor(memberInvocationHandler)));
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var memoryTest = await hastlayer.GenerateProxy(hardwareRepresentation, new MemoryTest());
var output1 = memoryTest.Run(0, 1);
var output2 = memoryTest.Run(0, 3);
var output3 = memoryTest.Run(0, 7);
var output4 = memoryTest.Run(0, 50);
var output5 = memoryTest.Run(1, 1);
var output6 = memoryTest.Run(3, 7);
var output7 = memoryTest.Run(47, 100);
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
// Starting with 1 not to have a divide by zero.
var vector = Enumerable.Range(1, SimdCalculator.MaxDegreeOfParallelism * 4).ToArray();
var simdCalculator = await hastlayer.GenerateProxy(hardwareRepresentation, new SimdCalculator());
var sumVector = ThrowIfNotCorrect(simdCalculator, calculator => calculator.AddVectors(vector, vector));
var differenceVector = ThrowIfNotCorrect(simdCalculator, calculator => calculator.SubtractVectors(vector, vector));
var productVector = ThrowIfNotCorrect(simdCalculator, calculator => calculator.MultiplyVectors(vector, vector));
var quotientVector = ThrowIfNotCorrect(simdCalculator, calculator => calculator.DivideVectors(vector, vector));
}
public BasicExecutionContext(IHastlayer hastlayer,
string deviceName,
string communicationChannelName,
Dictionary <string, object> customConfiguration = null)
{
HardwareRepresentation = hastlayer.GenerateHardware(new Assembly[] { Assembly.GetExecutingAssembly() },
new HardwareGenerationConfiguration(deviceName)).Result;
ProxyGenerationConfiguration = new ProxyGenerationConfiguration()
{
CommunicationChannelName = communicationChannelName,
CustomConfiguration = customConfiguration ?? new Dictionary <string, object>(),
VerifyHardwareResults = false
};
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var cpuOjutput = new FSharpParallelAlgorithmContainer.FSharpParallelAlgorithm().Run(234234);
var parallelAlgorithm = await hastlayer.GenerateProxy(hardwareRepresentation, new FSharpParallelAlgorithmContainer.FSharpParallelAlgorithm());
var output1 = parallelAlgorithm.Run(234234);
var output2 = parallelAlgorithm.Run(123);
var output3 = parallelAlgorithm.Run(9999);
var sw = System.Diagnostics.Stopwatch.StartNew();
var cpuOutput = new FSharpParallelAlgorithmContainer.FSharpParallelAlgorithm().Run(234234);
sw.Stop();
System.Console.WriteLine("On CPU it took " + sw.ElapsedMilliseconds + "ms.");
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
using (var bitmap = new Bitmap("fpga.jpg"))
{
var imageContrastModifier = await hastlayer
.GenerateProxy(hardwareRepresentation, new ImageContrastModifier());
// This takes about 160ms on an i7 CPU and net 150ms on an FPGA.
var modifiedImage = imageContrastModifier.ChangeImageContrast(bitmap, -50);
modifiedImage.Save("contrast.bmp");
// ImageFilter disabled until it's improved.
//var imageFilter = await hastlayer.GenerateProxy(hardwareRepresentation, new ImageFilter());
//var filteredImage = imageFilter.DetectHorizontalEdges(bitmap);
}
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var hastlayerOptimizedAlgorithm = await hastlayer.GenerateProxy(hardwareRepresentation, new ParallelAlgorithm());
// This takes about 1900ms on an i7 processor with 4 physical (8 logical) cores and 300ms on an FPGA (with
// a MaxDegreeOfParallelism of 280 while the device is about 80% utilized). With a higher degree of
// parallelism it won't fit on the Nexys 4 DDR board's FPGA.
var output1 = hastlayerOptimizedAlgorithm.Run(234234);
var output2 = hastlayerOptimizedAlgorithm.Run(123);
var output3 = hastlayerOptimizedAlgorithm.Run(9999);
var sw = System.Diagnostics.Stopwatch.StartNew();
var cpuOutput = new ParallelAlgorithm().Run(234234);
sw.Stop();
System.Console.WriteLine("On CPU it took " + sw.ElapsedMilliseconds + "ms.");
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var primeCalculator = await hastlayer.GenerateProxy(hardwareRepresentation, new PrimeCalculator());
var isPrime = primeCalculator.IsPrimeNumber(15);
var isPrime2 = primeCalculator.IsPrimeNumber(13);
var isPrime3 = await primeCalculator.IsPrimeNumberAsync(21);
// Only 2341 is prime.
var arePrimes = primeCalculator.ArePrimeNumbers(new uint[] { 15, 493, 2341, 99237 });
var arePrimes2 = primeCalculator.ArePrimeNumbers(new uint[] { 13, 493 });
// You can also launch hardware-executed method calls in parallel. If there are multiple boards
// connected then all of them will be utilized. If the whole device pool is utilized calls will
// wait for their turn.
// Uncomment if you have multiple boards connected.
//var parallelLaunchedIsPrimeTasks = new List<Task<bool>>();
//for (uint i = 100; i < 110; i++)
//{
// parallelLaunchedIsPrimeTasks
// .Add(Task.Factory.StartNew(indexObject => primeCalculator.IsPrimeNumber((uint)indexObject), i));
//}
//var parallelLaunchedArePrimes = await Task.WhenAll(parallelLaunchedIsPrimeTasks);
// In-algorithm parallelization:
// Note that if the amount of numbers used here can't be divided by PrimeCalculator.MaxDegreeOfParallelism
// then for ParallelizedArePrimeNumbers the input and output will be padded to a divisible amount (see
// comments in the method). Thus the communication round-trip will be slower for ParallelizedArePrimeNumbers.
// Because of this since PrimeCalculator.MaxDegreeOfParallelism is 30 we use 30 numbers here.
// All of these numbers except for 9999 are primes.
var numbers = new uint[]
{
9749, 9999, 902119, 907469, 915851,
9749, 9973, 902119, 907469, 915851,
9749, 9999, 902119, 907469, 915851,
9749, 9973, 902119, 907469, 915851,
9749, 9999, 902119, 907469, 915851,
9749, 9973, 902119, 907469, 915851
};
var arePrimes3 = primeCalculator.ArePrimeNumbers(numbers);
var arePrimes4 = primeCalculator.ParallelizedArePrimeNumbers(numbers);
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var genomeMatcher = await hastlayer.GenerateProxy(hardwareRepresentation, new GenomeMatcher());
// Sample from IBM.
var inputOne = "GCCCTAGCG";
var inputTwo = "GCGCAATG";
var result = genomeMatcher.CalculateLongestCommonSubsequence(inputOne, inputTwo);
// Sample from Wikipedia.
inputOne = "ACACACTA";
inputTwo = "AGCACACA";
result = genomeMatcher.CalculateLongestCommonSubsequence(inputOne, inputTwo);
inputOne = "lombiqtech";
inputTwo = "coulombtech";
result = genomeMatcher.CalculateLongestCommonSubsequence(inputOne, inputTwo);
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var parallelAlgorithm = await hastlayer.GenerateProxy(hardwareRepresentation, new ParallelAlgorithm());
// This takes about 1900ms on an i7 processor with 4 physical (8 logical) cores and 300ms on an FPGA (with
// a MaxDegreeOfParallelism of 280 while the device is about 80% utilized). With a higher degree of
// parallelism it won't fit on the Nexys A7 board's FPGA.
// On Catapult a MaxDegreeOfParallelism of 700 will fit as well (70% resource utilization) and run in about
// 200ms (including communication latency) vs about 5s on the previous reference PC. Compiling that
// hardware design will take about 14.5 hours though (with MaxDegreeOfParallelism of 600 it'll take about
// 4).
var output1 = parallelAlgorithm.Run(234234);
var output2 = parallelAlgorithm.Run(123);
var output3 = parallelAlgorithm.Run(9999);
var sw = System.Diagnostics.Stopwatch.StartNew();
var cpuOutput = new ParallelAlgorithm().Run(234234);
sw.Stop();
System.Console.WriteLine("On CPU it took " + sw.ElapsedMilliseconds + "ms.");
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
var fixed64Calculator = await hastlayer.GenerateProxy(hardwareRepresentation, new Fix64Calculator());
var sum = fixed64Calculator.CalculateIntegerSumUpToNumber(10000000);
// This takes about 274ms on an i7 processor with 4 physical (8 logical) cores and 1300ms on an FPGA (with
// a MaxDegreeOfParallelism of 13 while the device is about 76% utilized).
// Since this basically does what the single-threaded sample but in multiple copies on multiple threads
// the single-threaded sample takes the same amount of time on the FPGA.
// Creating an array of numbers alternating between 9999999 and 10000001 so we can also see that threads
// don't step on each other's feet.
var numbers = new int[Fix64Calculator.MaxDegreeOfParallelism];
for (int i = 0; i < Fix64Calculator.MaxDegreeOfParallelism; i++)
{
numbers[i] = 10000000 + (i % 2 == 0 ? -1 : 1);
}
var sums = fixed64Calculator.ParallelizedCalculateIntegerSumUpToNumbers(numbers);
}
public static async Task Run(IHastlayer hastlayer, IHardwareRepresentation hardwareRepresentation)
{
uint iterationsCount = MonteCarloPiEstimator.MaxDegreeOfParallelism * 500000;
// On a Nexys A7 this takes about 34ms with a 76 degree of parallelism and method inlining. It takes
// about 120ms on an i7 processor with 4 physical (8 logical) cores.
// On Catapult with a 350 degree of parallelism it takes 26ms on hardware and 160ms on the (2*32 logical
// core) CPU.
var monteCarloPiEstimator = await hastlayer.GenerateProxy(hardwareRepresentation, new MonteCarloPiEstimator());
var piEstimateHardware = monteCarloPiEstimator.EstimatePi(iterationsCount);
Console.WriteLine("Estimate for Pi on hardware: " + piEstimateHardware);
var sw = System.Diagnostics.Stopwatch.StartNew();
var piEstimateSoftware = new MonteCarloPiEstimator().EstimatePi(iterationsCount);
sw.Stop();
Console.WriteLine("Estimate for Pi on software: " + piEstimateSoftware);
Console.WriteLine("On CPU it took " + sw.ElapsedMilliseconds + "ms.");
}
/// <summary>
/// Generates a proxy for the given object that will transfer suitable calls to the hardware implementation using the default proxy generation configuration.
/// </summary>
/// <typeparam name="T">Type of the object to generate a proxy for.</typeparam>
/// <param name="hardwareRepresentation">The representation of the assemblies implemented as hardware.</param>
/// <param name="hardwareObject">The object to generate the proxy for.</param>
/// <returns>The generated proxy object.</returns>
/// <exception cref="HastlayerException">
/// Thrown if any lower-level exception or other error happens during proxy generation.
/// </exception>
public static Task <T> GenerateProxy <T>(
this IHastlayer hastlayer,
IHardwareRepresentation hardwareRepresentation,
T hardwareObject) where T : class =>
hastlayer.GenerateProxy(hardwareRepresentation, hardwareObject, ProxyGenerationConfiguration.Default);
public MemberInvocationHandler CreateMemberInvocationHandler(
IHardwareRepresentation hardwareRepresentation,
object target,
IProxyGenerationConfiguration configuration)
{
return(invocation =>
{
var methodAsynchronicity = GetMethodAsynchronicity(invocation);
if (methodAsynchronicity == MethodAsynchronicity.AsyncFunction)
{
throw new NotSupportedException("Only async methods that return a Task, not Task<T>, are supported.");
}
async Task invocationHandler()
{
using (var workContext = _wca.CreateWorkContextScope())
{
// Although it says Method it can also be a property.
var memberFullName = invocation.Method.GetFullName();
var invocationContext = new MemberInvocationContext
{
Invocation = invocation,
MemberFullName = memberFullName,
HardwareRepresentation = hardwareRepresentation
};
var eventHandler = workContext.Resolve <IMemberInvocationEventHandler>();
eventHandler.MemberInvoking(invocationContext);
workContext.Resolve <IEnumerable <IMemberInvocationPipelineStep> >().InvokePipelineSteps(step =>
{
invocationContext.HardwareExecutionIsCancelled = step.CanContinueHardwareExecution(invocationContext);
});
if (!invocationContext.HardwareExecutionIsCancelled)
{
var hardwareMembers = hardwareRepresentation.HardwareDescription.HardwareEntryPointNamesToMemberIdMappings;
var memberNameAlternates = new HashSet <string>(hardwareMembers.Keys.SelectMany(member => member.GetMemberNameAlternates()));
if (!hardwareMembers.ContainsKey(memberFullName) && !memberNameAlternates.Contains(memberFullName))
{
invocationContext.HardwareExecutionIsCancelled = true;
}
}
if (invocationContext.HardwareExecutionIsCancelled)
{
invocation.Proceed();
if (methodAsynchronicity == MethodAsynchronicity.AsyncAction)
{
await(Task) invocation.ReturnValue;
}
return;
}
var communicationChannelName = configuration.CommunicationChannelName;
var deviceManifest = hardwareRepresentation.DeviceManifest;
if (string.IsNullOrEmpty(communicationChannelName))
{
communicationChannelName = deviceManifest.DefaultCommunicationChannelName;
}
if (!deviceManifest.SupportedCommunicationChannelNames.Contains(communicationChannelName))
{
throw new NotSupportedException(
"The configured communication channel \"" + communicationChannelName +
"\" is not supported by the current device.");
}
var memory = (SimpleMemory)invocation.Arguments.SingleOrDefault(argument => argument is SimpleMemory);
if (memory != null)
{
var memoryByteCount = (ulong)memory.CellCount * SimpleMemory.MemoryCellSizeBytes;
if (memoryByteCount > deviceManifest.AvailableMemoryBytes)
{
throw new InvalidOperationException(
"The input is too large to fit into the device's memory: the input is " +
memoryByteCount + " bytes, the available memory is " +
deviceManifest.AvailableMemoryBytes + " bytes.");
}
SimpleMemory softMemory = null;
if (configuration.VerifyHardwareResults)
{
softMemory = new SimpleMemory(memory.CellCount);
var memoryBytes = new SimpleMemoryAccessor(memory).Get();
memoryBytes.CopyTo(new SimpleMemoryAccessor(softMemory).Get());
var memoryArgumentIndex = invocation.Arguments
.Select((argument, index) => new { Argument = argument, Index = index })
.Single(argument => argument.Argument is SimpleMemory)
.Index;
invocation.SetArgumentValue(memoryArgumentIndex, softMemory);
// This needs to happen before the awaited Execute() call below, otherwise the Task
// in ReturnValue wouldn't be the original one any more.
invocation.Proceed();
if (methodAsynchronicity == MethodAsynchronicity.AsyncAction)
{
await(Task) invocation.ReturnValue;
}
}
// At this point we checked that the hardware entry point does have a mapping.
var memberId = hardwareRepresentation
.HardwareDescription
.HardwareEntryPointNamesToMemberIdMappings[memberFullName];
invocationContext.ExecutionInformation = await workContext
.Resolve <ICommunicationServiceSelector>()
.GetCommunicationService(communicationChannelName)
.Execute(
memory,
memberId,
new HardwareExecutionContext {
ProxyGenerationConfiguration = configuration, HardwareRepresentation = hardwareRepresentation
});
if (configuration.VerifyHardwareResults)
{
var mismatches = new List <HardwareExecutionResultMismatchException.Mismatch>();
if (memory.CellCount != softMemory.CellCount)
{
int overflowIndex = Math.Min(memory.CellCount, softMemory.CellCount);
mismatches.Add(new HardwareExecutionResultMismatchException.LengthMismatch(
memory.CellCount,
softMemory.CellCount,
overflowIndex,
memory.CellCount > softMemory.CellCount ? memory.Read4Bytes(overflowIndex) : new byte[0],
softMemory.CellCount > memory.CellCount ? softMemory.Read4Bytes(overflowIndex) : new byte[0]));
}
else
{
for (int i = 0; i < memory.CellCount; i++)
{
if (!memory.Read4Bytes(i).SequenceEqual(softMemory.Read4Bytes(i)))
{
mismatches.Add(new HardwareExecutionResultMismatchException.Mismatch(
i, memory.Read4Bytes(i), softMemory.Read4Bytes(i)));
}
}
}
if (mismatches.Any())
{
throw new HardwareExecutionResultMismatchException(mismatches);
}
}
}
else
{
throw new NotSupportedException(
"Only SimpleMemory-using implementations are supported for hardware execution. The invocation didn't include a SimpleMemory argument.");
}
eventHandler.MemberExecutedOnHardware(invocationContext);
}
}
if (methodAsynchronicity == MethodAsynchronicity.AsyncAction)
{
invocation.ReturnValue = invocationHandler();
}
else
{
invocationHandler().Wait();
}
});
}
