/// <summary>
/// Calculates for multiple numbers whether they're primes, in a parallelized way.
/// </summary>
/// <remarks>
/// This demonstrates how you can write parallelized code that Hastlayer will process and turn into hardware-level
/// parallelization: the Tasks' bodies will be copied in hardware as many times as many Tasks you start; thus,
/// the actual level of parallelism you get on the hardware corresponds to the number of Tasks, not the number
/// of CPU cores.
/// </remarks>
public virtual void ParallelizedArePrimeNumbers(SimpleMemory memory)
{
// We need this information explicitly as we can't store arrays directly in memory.
uint numberCount = memory.ReadUInt32(ArePrimeNumbers_InputUInt32CountIndex);
// At the moment Hastlayer only supports a fixed degree of parallelism so we need to pad the input array
// if necessary, see PrimeCalculatorExtensions.
var tasks = new Task <bool> [MaxDegreeOfParallelism];
int i = 0;
while (i < numberCount)
{
for (int m = 0; m < MaxDegreeOfParallelism; m++)
{
var currentNumber = memory.ReadUInt32(ArePrimeNumbers_InputUInt32sStartIndex + i + m);
// Note that you can just call (thread-safe) methods from inside Tasks as usual. In hardware those
// invoked methods will be copied together with the Tasks' bodies too.
tasks[m] = Task.Factory.StartNew(
numberObject => IsPrimeNumberInternal((uint)numberObject),
currentNumber);
}
// Hastlayer doesn't support async code at the moment since ILSpy doesn't handle the new Roslyn-compiled
// code. See: https://github.com/icsharpcode/ILSpy/issues/502
Task.WhenAll(tasks).Wait();
for (int m = 0; m < MaxDegreeOfParallelism; m++)
{
memory.WriteBoolean(ArePrimeNumbers_OutputBooleansStartIndex + i + m, tasks[m].Result);
}
i += MaxDegreeOfParallelism;
}
}
public virtual void EstimatePi(SimpleMemory memory)
{
var iterationsCount = memory.ReadUInt32(EstimatePi_IteractionsCountUInt32Index);
var randomSeed = (ushort)memory.ReadUInt32(EstimatePi_RandomSeedUInt32Index);
var iterationsPerTask = iterationsCount / MaxDegreeOfParallelism;
var tasks = new Task <uint> [MaxDegreeOfParallelism];
for (uint i = 0; i < MaxDegreeOfParallelism; i++)
{
tasks[i] = Task.Factory.StartNew(
indexObject =>
{
var index = (uint)indexObject;
// A 16b PRNG is enough for this task and the xorshift one has suitable quality.
var random = new RandomXorshiftLfsr16 {
State = (ushort)(randomSeed + index)
};
uint inCircleCount = 0;
for (var j = 0; j < iterationsPerTask; j++)
{
uint a = random.NextUInt16();
uint b = random.NextUInt16();
// A bit of further parallelization can be exploited with SIMD to shave off some execution
// time. However, this needs so much resources on the hardware that the degree of
// parallelism needs to be lowered substantially (below 60).
//var randomNumbers = new uint[] { random.NextUInt16(), random.NextUInt16() };
//var products = Common.Numerics.SimdOperations.MultiplyVectors(randomNumbers, randomNumbers, 2);
if ((ulong)(a * a) + b * b <= ((uint)ushort.MaxValue * ushort.MaxValue))
//if ((ulong)products[0] + products[1] <= ((uint)ushort.MaxValue * ushort.MaxValue))
{
inCircleCount++;
}
}
return(inCircleCount);
},
i);
}
Task.WhenAll(tasks).Wait();
uint inCircleCountSum = 0;
for (int i = 0; i < MaxDegreeOfParallelism; i++)
{
inCircleCountSum += tasks[i].Result;
}
memory.WriteUInt32(EstimatePi_InCircleCountSumUInt32Index, inCircleCountSum);
}
public virtual void ArePrimeNumbers(SimpleMemory memory)
{
uint numberCount = memory.ReadUInt32(ArePrimeNumbers_InputUInt32CountIndex);
for (int i = 0; i < numberCount; i++)
{
uint number = memory.ReadUInt32(ArePrimeNumbers_InputUInt32sStartIndex + i);
var isPrime = IsPrimeNumberInternal(number);
memory.WriteBoolean(ArePrimeNumbers_OutputUInt32sStartIndex + i, isPrime);
}
}
public virtual void ArePrimeNumbers(SimpleMemory memory)
{
// We need this information explicitly as we can't store arrays directly in memory.
uint numberCount = memory.ReadUInt32(ArePrimeNumbers_InputUInt32CountIndex);
for (int i = 0; i < numberCount; i++)
{
uint number = memory.ReadUInt32(ArePrimeNumbers_InputUInt32sStartIndex + i);
memory.WriteBoolean(ArePrimeNumbers_OutputBooleansStartIndex + i, IsPrimeNumberInternal(number));
}
}
public virtual void MWC64X(SimpleMemory memory)
{
uint stateHighWord = memory.ReadUInt32(1);
uint stateLowWord = memory.ReadUInt32(0);;
ulong randomState = stateLowWord * 0xFFFEB81BUL + stateHighWord;
uint randomWord = stateLowWord ^ stateHighWord;
memory.WriteUInt32(0, (uint)randomState); //LE: 1 is high byte, 0 is low byte
memory.WriteUInt32(1, (uint)(randomState >> 32));
memory.WriteUInt32(2, randomWord);
}
public virtual void ParallelizedArePrimeNumbers(SimpleMemory memory)
{
// We need this information explicitly as we can't store arrays directly in memory.
uint numberCount = memory.ReadUInt32(ArePrimeNumbers_InputUInt32CountIndex);
// At the moment Hastlayer only supports a fixed degree of parallelism so we need to pad the input array
// if necessary, see PrimeCalculatorExtensions.
var tasks = new Task <bool> [MaxDegreeOfParallelism];
int i = 0;
while (i < numberCount)
{
for (int m = 0; m < MaxDegreeOfParallelism; m++)
{
var currentNumber = memory.ReadUInt32(ArePrimeNumbers_InputUInt32sStartIndex + i + m);
tasks[m] = Task.Factory.StartNew(
numberObject =>
{
// This is a copy of the body of IsPrimeNumberInternal(). We could also call that method
// from this lambda but it's more efficient to just do it directly, not adding indirection.
var number = (uint)numberObject;
uint factor = number / 2;
for (uint x = 2; x <= factor; x++)
{
if ((number % x) == 0)
{
return(false);
}
}
return(true);
},
currentNumber);
}
// Hastlayer doesn't support async code at the moment since ILSpy doesn't handle the new Roslyn-compiled
// code. See: https://github.com/icsharpcode/ILSpy/issues/502
Task.WhenAll(tasks).Wait();
for (int m = 0; m < MaxDegreeOfParallelism; m++)
{
memory.WriteBoolean(ArePrimeNumbers_OutputBooleansStartIndex + i + m, tasks[m].Result);
}
i += MaxDegreeOfParallelism;
}
}
public static uint Run(this ParallelAlgorithm algorithm, uint input)
{
var memory = new SimpleMemory(1);
memory.WriteUInt32(ParallelAlgorithm.Run_InputUInt32Index, input);
algorithm.Run(memory);
return memory.ReadUInt32(ParallelAlgorithm.Run_OutputUInt32Index);
}
public virtual void Run(SimpleMemory memory)
{
var inputNumber = memory.ReadUInt32(Run_InputUInt32Index);
// Or:
inputNumber = new MemoryContainer(memory).GetInput();
// Arrays can be initialized as usual, as well as objects.
var numberContainers1 = new[]
{
new NumberContainer {
Number = inputNumber
},
new NumberContainer {
Number = inputNumber + 4
},
new NumberContainer {
Number = 24
},
new NumberContainer(9)
};
// Array elements can be accessed and modified as usual.
numberContainers1[0].NumberPlusFive = inputNumber + 10;
numberContainers1[1].IncreaseNumber(5);
numberContainers1[2].IncreaseNumberBy10();
// Using ref and out.
uint increaseBy = 10;
numberContainers1[3].IncreaseNumberByParameterTimes10(ref increaseBy, out uint originalNumber);
public virtual void IsPrimeNumber(SimpleMemory memory)
{
var number = memory.ReadUInt32(IsPrimeNumber_InputUInt32Index);
var isPrime = IsPrimeNumberInternal(number);
memory.WriteBoolean(IsPrimeNumber_OutputBooleanIndex, isPrime);
}
public virtual void MWC64X(SimpleMemory memory)
{
uint stateHighWord = memory.ReadUInt32(1);
uint stateLowWord = memory.ReadUInt32(0);;
// Creating the value 0xFFFEB81BUL. This literal can't be directly used due to an ILSpy bug, see:
// https://github.com/icsharpcode/ILSpy/issues/807
uint constantHighShort = 0xFFFE;
uint constantLowShort = 0xB81B;
uint constantWord = (0 << 32) | (constantHighShort << 16) | constantLowShort;
ulong randomState = (ulong)stateLowWord * (ulong)constantWord + (ulong)stateHighWord;
uint randomWord = stateLowWord ^ stateHighWord;
memory.WriteUInt32(0, (uint)randomState); //LE: 1 is high byte, 0 is low byte
memory.WriteUInt32(1, (uint)(randomState >> 32));
memory.WriteUInt32(2, randomWord);
}
public uint CalculateFibonacchiSeries(short number)
{
var memory = new SimpleMemory(2);
memory.WriteInt32(CalculateFibonacchiSeries_InputShortIndex, number);
CalculateFibonacchiSeries(memory);
return(memory.ReadUInt32(CalculateFibonacchiSeries_OutputUInt32Index));
}
public static uint CalculateFactorial(this RecursiveAlgorithms recursiveAlgorithms, short number)
{
var memory = new SimpleMemory(2);
memory.WriteInt32(RecursiveAlgorithms.CalculateFactorial_InputShortIndex, number);
recursiveAlgorithms.CalculateFactorial(memory);
return(memory.ReadUInt32(RecursiveAlgorithms.CalculateFactorial_OutputUInt32Index));
}
public uint CalculateFactorial(short number)
{
var memory = new SimpleMemory(2);
memory.WriteInt32(CalculateFactorial_InputShortIndex, number);
CalculateFactorial(memory);
return(memory.ReadUInt32(CalculateFactorial_OutputUInt32Index));
}
public static uint Run(this ObjectOrientedShowcase algorithm, uint input)
{
var memory = new SimpleMemory(1);
memory.WriteUInt32(ObjectOrientedShowcase.Run_InputUInt32Index, input);
algorithm.Run(memory);
return(memory.ReadUInt32(ObjectOrientedShowcase.Run_OutputUInt32Index));
}
public uint Run(uint input)
{
var memory = new SimpleMemory(1);
memory.WriteUInt32(Run_InputUInt32Index, input);
Run(memory);
return(memory.ReadUInt32(Run_OutputUInt32Index));
}
/// <summary>
/// Calculates whether a number is prime.
/// </summary>
/// <remarks>
/// Note that the entry point of SimpleMemory-using algorithms should be void methods having a single
/// <see cref="SimpleMemory"/> argument.
/// </remarks>
/// <param name="memory">The <see cref="SimpleMemory"/> object representing the accessible memory space.</param>
public virtual void IsPrimeNumber(SimpleMemory memory)
{
// Reading out the input parameter.
var number = memory.ReadUInt32(IsPrimeNumber_InputUInt32Index);
// Writing back the output.
memory.WriteBoolean(IsPrimeNumber_OutputBooleanIndex, IsPrimeNumberInternal(number));
}
/// <summary>
/// This function adds two numbers on the FPGA using <see cref="KpzKernelsInterface.TestAdd(SimpleMemory)"/>.
/// </summary>
public static uint TestAddWrapper(this KpzKernelsInterface kernels, uint a, uint b)
{
var sm = new SimpleMemory(3);
sm.WriteUInt32(0, a);
sm.WriteUInt32(1, b);
kernels.TestAdd(sm);
return(sm.ReadUInt32(2));
}
/// <summary>Pull table from the FPGA.</summary>
public static void CopyFromSimpleMemoryToGrid(KpzNode[,] gridDst, SimpleMemory memorySrc)
{
for (int x = 0; x < KpzKernels.GridWidth; x++)
{
for (int y = 0; y < KpzKernels.GridHeight; y++)
{
gridDst[x, y] = KpzNode.DeserializeFromUInt32(memorySrc.ReadUInt32(KpzKernels.MemIndexGrid + y * KpzKernels.GridWidth + x));
}
}
}
/// <summary>
/// Copies the grid data to BRAM/LUT RAM from DDR.
/// </summary>
public void CopyFromSimpleMemoryToRawGrid(SimpleMemory memory)
{
for (int x = 0; x < GridWidth; x++)
{
for (int y = 0; y < GridHeight; y++)
{
int index = y * GridWidth + x;
_gridRaw[index] = memory.ReadUInt32(MemIndexGrid + index);
}
}
}
public static void WriteHexdump(TextWriter writer, SimpleMemory memory)
{
for (int i = 0; i < memory.CellCount; i += HexDumpBlocksPerLine)
{
for (int j = 0; j < HexDumpBlocksPerLine && i + j < memory.CellCount; j++)
{
writer.Write("{0}{1:X8}", j == 0 ? "" : " ", memory.ReadUInt32(i + j));
}
writer.WriteLine();
}
}
/// <summary>Pull table from the FPGA.</summary>
public static void CopyFromSimpleMemoryToGrid(KpzNode[,] gridDst, SimpleMemory memorySrc)
{
for (int x = 0; x < KpzKernelsParallelizedInterface.GridSize; x++)
{
for (int y = 0; y < KpzKernelsParallelizedInterface.GridSize; y++)
{
gridDst[x, y] = KpzNode.DeserializeFromUInt32(
memorySrc.ReadUInt32(KpzKernelsParallelizedInterface.MemIndexGrid + y * KpzKernelsParallelizedInterface.GridSize + x));
}
}
}
private uint RecursivelyCalculateFactorial(SimpleMemory memory, short number)
{
memory.WriteUInt32(
CalculateFactorial_InvocationCounterUInt32Index,
memory.ReadUInt32(CalculateFactorial_InvocationCounterUInt32Index) + 1);
if (number == 0)
{
return(1);
}
return((uint)(number * RecursivelyCalculateFactorial(memory, (short)(number - 1))));
}
// The return value should be a type with a bigger range than the input. Although we can use 64b numbers
// internally we can't write the to memory yet so the input needs to be a short.
private uint RecursivelyCalculateFibonacchiSeries(SimpleMemory memory, short number)
{
memory.WriteUInt32(
CalculateFibonacchiSeries_InvocationCounterUInt32Index,
memory.ReadUInt32(CalculateFibonacchiSeries_InvocationCounterUInt32Index) + 1);
if (number == 0 || number == 1)
{
return((uint)number);
}
return(RecursivelyCalculateFibonacchiSeries(memory, (short)(number - 2)) + RecursivelyCalculateFibonacchiSeries(memory, (short)(number - 1)));
}
/// <summary>
/// Calculates the weight and centre of mass of a section of torus with varying density from a <see cref="SimpleMemory"/> object.
/// </summary>
/// <param name="simpleMemory">The <see cref="SimpleMemory"/> object that contains the result.</param>
/// <returns>Returns the weight and centre of mass of a section of torus with varying density in the form of a <see cref="MonteCarloResult"/> object.</returns>
private static MonteCarloResult GetResult(SimpleMemory simpleMemory)
{
return(new MonteCarloResult
{
W = simpleMemory.ReadUInt32(MonteCarloAlgorithm.MonteCarloAlgorithm_WIndex),
X = simpleMemory.ReadUInt32(MonteCarloAlgorithm.MonteCarloAlgorithm_XIndex),
Y = simpleMemory.ReadUInt32(MonteCarloAlgorithm.MonteCarloAlgorithm_YIndex),
Z = simpleMemory.ReadUInt32(MonteCarloAlgorithm.MonteCarloAlgorithm_ZIndex),
DW = simpleMemory.ReadUInt32(MonteCarloAlgorithm.MonteCarloAlgorithm_DWIndex),
DX = simpleMemory.ReadUInt32(MonteCarloAlgorithm.MonteCarloAlgorithm_DXIndex),
DY = simpleMemory.ReadUInt32(MonteCarloAlgorithm.MonteCarloAlgorithm_DYIndex),
DZ = simpleMemory.ReadUInt32(MonteCarloAlgorithm.MonteCarloAlgorithm_DZIndex)
});
}
/// <summary>
/// It loads the TestMode, NumberOfIterations parameters and also the PRNG seed from the SimpleMemory at
/// the beginning.
/// </summary>
/// <param name="memory"></param>
public void InitializeParametersFromMemory(SimpleMemory memory)
{
Prng1 = new PrngMWC64X((((ulong)memory.ReadUInt32(MemIndexRandomStates)) << 32) |
memory.ReadUInt32(MemIndexRandomStates + 1));
Prng2 = new PrngMWC64X((((ulong)memory.ReadUInt32(MemIndexRandomStates + 2)) << 32) |
memory.ReadUInt32(MemIndexRandomStates + 3));
TestMode = (memory.ReadUInt32(MemIndexStepMode) & 1) == 1;
NumberOfIterations = memory.ReadUInt32(MemIndexNumberOfIterations);
}
public virtual void Run(SimpleMemory memory)
{
var inputNumber = memory.ReadUInt32(Run_InputUInt32Index);
// Or:
inputNumber = new MemoryContainer(memory).GetInput();
// Arrays can be initialized as usual, as well as objects.
var numberContainers1 = new[]
{
new NumberContainer {
Number = inputNumber
},
new NumberContainer {
Number = inputNumber + 4
},
new NumberContainer {
Number = 24
},
new NumberContainer(9)
};
// Array elements can be accessed and modified as usual.
numberContainers1[0].NumberPlusFive = inputNumber + 10;
numberContainers1[1].IncreaseNumber(5);
// Note that array dimensions need to be defined compile-time. They needn't bee constants directly used
// when instantiating the array but the size argument needs to be resolvable compile-time (so if it's a
// variable then its value should be computable from all other values at compile-time).
var numberContainers2 = new NumberContainer[1];
var numberContainer = new NumberContainer();
numberContainer.Number = 5;
numberContainer.Number = numberContainer.NumberPlusFive;
if (!numberContainer.WasIncreased)
{
numberContainer.IncreaseNumber(5);
}
numberContainers2[0] = numberContainer;
for (int i = 0; i < numberContainers1.Length; i++)
{
numberContainers1[i].IncreaseNumber(numberContainers2[0].Number);
}
// You can also pass arrays and other objects around to other methods.
memory.WriteUInt32(Run_OutputUInt32Index, SumNumberCointainers(numberContainers1));
}
public virtual void Run(SimpleMemory memory)
{
var input = memory.ReadUInt32(Run_InputUInt32Index);
var tasks = new Task <uint> [MaxDegreeOfParallelism];
// Hastlayer will figure out how many Tasks you want to start if you kick them off in a loop like this.
// If this is more involved then you'll need to tell Hastlayer the level of parallelism, see the comment in
// ParallelAlgorithmSampleRunner.
for (uint i = 0; i < MaxDegreeOfParallelism; i++)
{
tasks[i] = Task.Factory.StartNew(
indexObject =>
{
var index = (uint)indexObject;
uint result = input + index * 2;
var even = true;
for (int j = 2; j < 9999999; j++)
{
if (even)
{
result += index;
}
else
{
result -= index;
}
even = !even;
}
return(result);
},
i);
}
// Task.WhenAny() can be used too.
Task.WhenAll(tasks).Wait();
uint output = 0;
for (int i = 0; i < MaxDegreeOfParallelism; i++)
{
output += tasks[i].Result;
}
memory.WriteUInt32(Run_OutputUInt32Index, output);
}
/// <summary>
/// This function generates random numbers on the FPGA using
/// <see cref="KpzKernelsInterface.TestPrng(SimpleMemory)"/>.
/// </summary>
public static uint[] TestPrngWrapper(this KpzKernelsInterface kernels)
{
var numbers = new uint[KpzKernels.GridWidth * KpzKernels.GridHeight];
var sm = new SimpleMemory(KpzKernels.SizeOfSimpleMemory);
CopyParametersToMemory(sm, false, 0x5289a3b89ac5f211, 0x5289a3b89ac5f211, 0);
kernels.TestPrng(sm);
for (int i = 0; i < KpzKernels.GridWidth * KpzKernels.GridHeight; i++)
{
numbers[i] = sm.ReadUInt32(i);
}
return(numbers);
}
public virtual void Run(SimpleMemory memory)
{
var input = memory.ReadUInt32(Run_InputUInt32Index);
var tasks = new Task <uint> [MaxDegreeOfParallelism];
for (uint i = 0; i < MaxDegreeOfParallelism; i++)
{
tasks[i] = Task.Factory.StartNew(
indexObject =>
{
var index = (uint)indexObject;
uint result = input + index * 2;
var even = true;
for (int j = 2; j < 9999999; j++)
{
if (even)
{
result += index;
}
else
{
result -= index;
}
even = !even;
}
return(result);
},
i);
}
// Task.WhenAny() can be used too.
Task.WhenAll(tasks).Wait();
uint output = 0;
for (int i = 0; i < MaxDegreeOfParallelism; i++)
{
output += tasks[i].Result;
}
memory.WriteUInt32(Run_OutputUInt32Index, output);
}
/// <summary>
/// Extracts the longest common subsequence from the <see cref="SimpleMemory"/> object.
/// </summary>
/// <param name="simpleMemory">The <see cref="SimpleMemory"/> object that contains the result.</param>
/// <param name="inputOne">The first string to compare.</param>
/// <param name="inputTwo">The second string to compare.</param>
/// <returns>Returns the longest common subsequence.</returns>
private string GetResult(SimpleMemory simpleMemory, string inputOne, string inputTwo)
{
var maxInputLength = Math.Max(inputOne.Length, inputTwo.Length);
var result = "";
var startIndex = GetLCS_InputOneStartIndex + inputOne.Length + inputTwo.Length + (inputOne.Length * inputTwo.Length) * 2;
for (int i = 0; i < maxInputLength; i++)
{
var currentChar = simpleMemory.ReadUInt32(startIndex + i);
var currentCharBytes = BitConverter.GetBytes(currentChar);
var chars = Encoding.UTF8.GetChars(currentCharBytes);
result += chars[0];
}
return(result.Replace("\0", ""));
}
