public unsafe void ComputeMandelbrotSet(
uint[,] iterations,
int width, int height,
double offsetX, double offsetY, double zoom,
uint maxIterations,
bool doublePrecision,
ref bool cancel)
{
Resize(width, height);
// Precompute the pixels coordinates.
double startX = -(width / (height * zoom)) + offsetX;
double startY = -1.0 / zoom + offsetY;
double scale = 2 / (height * zoom);
if (doublePrecision)
{
_doubleKernel(new Index2(width, height), _deviceMemory.View, width, height, startX, startY, scale, maxIterations);
}
else
{
_singleKernel(new Index2(width, height), _deviceMemory.View, width, height, (float)startX, (float)startY, (float)scale, maxIterations);
}
fixed(uint *ptr = iterations)
{
using var viewArrayWrapper = ViewPointerWrapper.Create(new IntPtr(ptr));
var arrayView = new ArrayView <uint>(viewArrayWrapper, 0, iterations.Length);
_deviceMemory.CopyTo(arrayView, 0);
_gpu.Synchronize();
}
}
public unsafe void ComputeMandelbrotSet(
uint[,] iterations,
int width, int height,
double offsetX, double offsetY, double zoom,
uint maxIterations,
bool doublePrecision,
ref bool cancel)
{
Resize(width, height);
var doubleOverhead = doublePrecision ? 32 : 1;
var cores = _gpu.NumMultiprocessors * 128; // TODO ILGPU proper number of CUDA cores per multi processor.
var maxPixelPerBatch = 200L * _gpu.ClockRate * cores / (doubleOverhead * maxIterations);
var blockSizeX = 16;
var blockSizeY = 8;
var gridX = DivUp(width, blockSizeX);
var gridY = DivUp(height, blockSizeY);
var maxBlocks = DivUp(checked ((int)maxPixelPerBatch), blockSizeX * blockSizeY);
var maxBlocksY = DivUp(maxBlocks, gridX);
// Precompute the pixels coordinates.
double startX = -(width / (height * zoom)) + offsetX;
double startY = -1.0 / zoom + offsetY;
double scale = 2 / (height * zoom);
// Batch the mandelbrot computation into sets of block so that we avoid hitting the Windows watchdog timer for CUDA (5 seconds on Windows 10).
for (int blockOffsetY = 0; blockOffsetY < gridY; blockOffsetY += maxBlocksY)
{
if (cancel)
{
break;
}
var lp = ((gridX, Math.Min(gridY, maxBlocksY)), (blockSizeX, blockSizeY));
if (doublePrecision)
{
_gpu.Launch(DoubleKernel, lp, _deviceMemory.View, width, height, blockOffsetY * blockSizeY, startX, startY, scale, maxIterations);
}
else
{
_gpu.Launch(SingleKernel, lp, _deviceMemory.View, width, height, blockOffsetY * blockSizeY, startX, startY, scale, maxIterations);
}
}
fixed(uint *ptr = iterations)
{
using var viewArrayWrapper = ViewPointerWrapper.Create(new IntPtr(ptr));
var arrayView = new ArrayView <uint>(viewArrayWrapper, 0, iterations.Length);
_deviceMemory.CopyTo(arrayView, 0);
}
}
