public Mandelbrot( Platform platform, int width, int height )
{
openCLPlatform = platform;
openCLDevices = openCLPlatform.QueryDevices(DeviceType.ALL);
openCLContext = openCLPlatform.CreateDefaultContext();
openCLCQ = openCLContext.CreateCommandQueue(openCLDevices[0], CommandQueueProperties.PROFILING_ENABLE);
mandelBrotProgram = openCLContext.CreateProgramWithSource(File.ReadAllText("Mandelbrot.cl"));
try
{
mandelBrotProgram.Build();
}
catch (OpenCLException)
{
string buildLog = mandelBrotProgram.GetBuildLog(openCLDevices[0]);
MessageBox.Show(buildLog,"Build error(64 bit debug sessions in vs2008 always fail like this - debug in 32 bit or use vs2010)");
Application.Exit();
}
mandelbrotKernel = mandelBrotProgram.CreateKernel("Mandelbrot");
Left = -2.0f;
Top = 2.0f;
Right = 2.0f;
Bottom = -2.0f;
BitmapWidth = width;
BitmapHeight = height;
mandelbrotMemBuffer = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.WRITE_ONLY), width*height*4, IntPtr.Zero);
}
internal NativeKernelCallbackData(NativeKernel nk,CommandQueue cq, object o, Mem[] buffers)
{
NativeKernel = nk;
CQ = cq;
O = o;
Buffers = buffers;
}
// Dispose(bool disposing) executes in two distinct scenarios.
// If disposing equals true, the method has been called directly
// or indirectly by a user's code. Managed and unmanaged resources
// can be disposed.
// If disposing equals false, the method has been called by the
// runtime from inside the finalizer and you should not reference
// other objects. Only unmanaged resources can be disposed.
private void Dispose(bool disposing)
{
// Check to see if Dispose has already been called.
if (!this.disposed)
{
// If disposing equals true, dispose all managed
// and unmanaged resources.
if (disposing)
{
// Dispose managed resources.
Bitmap.Dispose();
Bitmap = null;
Image.Dispose();
Image = null;
CQ = null;
Context = null;
}
// Call the appropriate methods to clean up
// unmanaged resources here.
// If disposing is false,
// only the following code is executed.
// Note disposing has been done.
disposed = true;
}
}
public void ReleaseDeviceResources()
{
oclFullyInitialized = false;
if (OCLSampler != null)
{
OCLSampler.Dispose();
OCLSampler = null;
}
if (OCLInputImage != null)
{
OCLInputImage.Dispose();
OCLInputImage = null;
}
if (OCLOutputImage != null)
{
OCLOutputImage.Dispose();
OCLOutputImage = null;
}
if (FilterKernel != null)
{
FilterKernel.Dispose();
FilterKernel = null;
}
if (oclProgram != null)
{
oclProgram.Dispose();
oclProgram = null;
}
if (oclCQ != null)
{
oclCQ.Dispose();
oclCQ = null;
}
if (oclContext != null)
{
oclContext.Dispose();
oclContext = null;
}
}
public virtual void Write(CommandQueue cq, long dstOffset, double[] srcData, int srcStartIndex, int count)
{
IntPtr p = cq.EnqueueMapBuffer(this, true, MapFlags.WRITE, dstOffset, (long)count * sizeof(double));
double* pBlock = (double*)p.ToPointer();
for (long i = 0; i < count; i++)
pBlock[i] = srcData[i + srcStartIndex];
cq.EnqueueUnmapMemObject(this, p);
cq.Finish();
}
public virtual void MemSet(CommandQueue cq, byte value)
{
long offset = 0;
long count = MemSize.ToInt64();
IntPtr p = cq.EnqueueMapBuffer(this, true, MapFlags.WRITE, offset, count);
byte* pBlock = (byte*)p.ToPointer();
for (long i = 0; i < count; i++)
pBlock[i] = value;
cq.EnqueueUnmapMemObject(this, p);
cq.Finish();
}
/// <summary>
/// Test all versions of:
///
/// EnqueueMapBuffer
/// EnqueueMapImage
///
/// The test bounces an array from a managed byte buffer to a mapped buffer,
/// to an image. The image is then mapped and copied to a new managed buffer
/// where the result is compared to the original.
///
/// On error, the actual point of failure will have to be identified manually.
/// </summary>
/// <param name="c"></param>
/// <param name="cq"></param>
private void TestMapBuffer(Context c, CommandQueue cq)
{
if (!cq.Device.ImageSupport)
{
Output("Skipping EnqueueMapBuffer and EnqueueMapImage tests(not supported on this device)");
return;
}
Output("Testing MapBuffer");
OpenCLNet.Image img0 = null;
OpenCLNet.Mem mem0 = null;
int imgWidth = 1024;
int imgHeight = 1024;
int bufLen = imgWidth * 4 * imgHeight;
byte[] srcData = new byte[bufLen];
byte[] cmpData = new byte[bufLen];
Event event0;
Event event1;
for (int i = 0; i < srcData.Length; i++)
srcData[i] = (byte)(i);
Array.Clear(cmpData, 0, cmpData.Length);
try
{
img0 = c.CreateImage2D(MemFlags.READ_WRITE, ImageFormat.RGBA8U, imgWidth, imgHeight);
mem0 = c.CreateBuffer(MemFlags.READ_WRITE, bufLen, IntPtr.Zero);
Array.Clear(cmpData, 0, cmpData.Length);
fixed (byte* pSrc = srcData)
{
fixed (byte* pCmp = cmpData)
{
{
IntPtr[] origin = new IntPtr[3] { (IntPtr)0, (IntPtr)0, (IntPtr)0 };
IntPtr[] region = new IntPtr[3] { (IntPtr)imgWidth, (IntPtr)imgHeight, (IntPtr)1 };
IntPtr[] dstOrigin = new IntPtr[3] { (IntPtr)0, (IntPtr)0, (IntPtr)0 };
IntPtr[] dstRegion = new IntPtr[3] { (IntPtr)imgWidth, (IntPtr)imgHeight, (IntPtr)1 };
IntPtr mapPtr;
byte* pMapPtr;
IntPtr image_row_pitch;
IntPtr image_slice_pitch;
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
cq.EnqueueCopyBufferToImage(mem0, img0, (IntPtr)0, origin, region);
mapPtr = cq.EnqueueMapImage(img0, true, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * (int)image_row_pitch;
byte* pDstRowPtr = pCmp + y*imgWidth*4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (IntPtr version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
Event fdjk;
cq.EnqueueCopyBufferToImage(mem0, img0, (IntPtr)0, origin, region, 0, null, out fdjk);
cq.Finish();
mapPtr = cq.EnqueueMapImage(img0, false, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch, 0, null, out event0 );
cq.EnqueueWaitForEvent(event0);
cq.Finish();
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * (int)image_row_pitch;
byte* pDstRowPtr = pCmp + y * imgWidth * 4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (IntPtr version)Copy not identical to source when using event output and no wait list");
Event[] waitList = new Event[] { event0 };
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
cq.EnqueueCopyBufferToImage(mem0, img0, (IntPtr)0, origin, region);
mapPtr = cq.EnqueueMapImage(img0, false, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch, 1, waitList, out event1);
cq.EnqueueWaitForEvent(event1);
cq.Finish();
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * (int)image_row_pitch;
byte* pDstRowPtr = pCmp + y * imgWidth * 4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (IntPtr version)Copy not identical to source when using event output and no wait list");
event0.Dispose();
event1.Dispose();
}
{
int[] origin = new int[3] { (int)0, (int)0, (int)0 };
int[] region = new int[3] { (int)imgWidth, (int)imgHeight, (int)1 };
IntPtr mapPtr;
byte* pMapPtr;
int image_row_pitch;
int image_slice_pitch;
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
cq.EnqueueCopyBufferToImage(mem0, img0, 0, origin, region);
mapPtr = cq.EnqueueMapImage(img0, true, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * (int)image_row_pitch;
byte* pDstRowPtr = pCmp + y*imgWidth*4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (int version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
cq.EnqueueCopyBufferToImage(mem0, img0, 0, origin, region);
mapPtr = cq.EnqueueMapImage(img0, false, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch, 0, null, out event0 );
cq.EnqueueWaitForEvent(event0);
cq.Finish();
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * (int)image_row_pitch;
byte* pDstRowPtr = pCmp + y * imgWidth * 4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (int version)Copy not identical to source when using event output and no wait list");
Event[] waitList = new Event[] { event0 };
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
cq.EnqueueCopyBufferToImage(mem0, img0, 0, origin, region);
mapPtr = cq.EnqueueMapImage(img0, false, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch, 1, waitList, out event1);
cq.EnqueueWaitForEvent(event1);
cq.Finish();
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * (int)image_row_pitch;
byte* pDstRowPtr = pCmp + y * imgWidth * 4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (int version)Copy not identical to source when using event output and no wait list");
event0.Dispose();
event1.Dispose();
}
{
long[] origin = new long[3] { (long)0, (long)0, (long)0 };
long[] region = new long[3] { (long)imgWidth, (long)imgHeight, (long)1 };
IntPtr mapPtr;
byte* pMapPtr;
long image_row_pitch;
long image_slice_pitch;
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
cq.EnqueueCopyBufferToImage(mem0, img0, (long)0, origin, region);
mapPtr = cq.EnqueueMapImage(img0, true, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * image_row_pitch;
byte* pDstRowPtr = pCmp + y * imgWidth * 4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (long version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
cq.EnqueueCopyBufferToImage(mem0, img0, (long)0, origin, region);
mapPtr = cq.EnqueueMapImage(img0, false, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch, 0, null, out event0);
cq.EnqueueWaitForEvent(event0);
cq.Finish();
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * image_row_pitch;
byte* pDstRowPtr = pCmp + y * imgWidth * 4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (long version)Copy not identical to source when using event output and no wait list");
Event[] waitList = new Event[] { event0 };
Array.Clear(cmpData, 0, cmpData.Length);
mapPtr = cq.EnqueueMapBuffer(mem0, true, MapFlags.WRITE, 0, bufLen);
pMapPtr = (byte*)mapPtr.ToPointer();
for (int i = 0; i < bufLen; i++)
pMapPtr[i] = srcData[i];
cq.EnqueueUnmapMemObject(mem0, mapPtr);
cq.EnqueueCopyBufferToImage(mem0, img0, (long)0, origin, region);
mapPtr = cq.EnqueueMapImage(img0, false, MapFlags.READ, origin, region, out image_row_pitch, out image_slice_pitch, 1, waitList, out event1);
cq.EnqueueWaitForEvent(event1);
cq.Finish();
pMapPtr = (byte*)mapPtr.ToPointer();
for (int y = 0; y < imgHeight; y++)
{
byte* pSrcRowPtr = pMapPtr + y * image_row_pitch;
byte* pDstRowPtr = pCmp + y * imgWidth * 4;
for (int x = 0; x < imgWidth * 4; x++)
{
pDstRowPtr[x] = pSrcRowPtr[x];
}
}
cq.EnqueueUnmapMemObject(img0, mapPtr);
if (!CompareArray(cmpData, srcData))
Error("EnqueueEnqueueMapBuffer/EnqueueMapImage: (long version)Copy not identical to source when using event output and no wait list");
event0.Dispose();
event1.Dispose();
}
}
}
}
catch (Exception e)
{
Error("Exception during testing: " + e.ToString());
}
finally
{
if (img0 != null)
img0.Dispose();
if (mem0 != null)
mem0.Dispose();
}
}
/// <summary>
/// Test all versions of:
///
/// EnqueueWriteImage
/// EnqueueReadImage
/// EnqueueCopyImage
///
/// The test just copies the entirety of a buffer and checks if the result is equal to the original.
/// An error indicates that one of the above functions failed and further manual analysis is required
/// to pinpoint the error.
/// </summary>
/// <param name="c"></param>
/// <param name="cq"></param>
private void TestImageReadWriteCopyOps(Context c, CommandQueue cq)
{
if (!cq.Device.ImageSupport)
{
Output("Skipping image read/write/copy tests(not supported on this device)");
return;
}
Output("Testing image read/write/copy functions");
OpenCLNet.Image img0 = null;
OpenCLNet.Image img1 = null;
OpenCLNet.Image img2 = null;
int imgWidth = 1024;
int imgHeight = 1024;
int bufLen = imgWidth*4*imgHeight;
byte[] srcData = new byte[bufLen];
byte[] cmpData = new byte[bufLen];
Event event0;
Event event1;
Event event2;
Event event3;
Event event4;
Event event5;
for (int i = 0; i < srcData.Length; i++)
srcData[i] = (byte)(i);
Array.Clear(cmpData, 0, cmpData.Length);
try
{
img0 = c.CreateImage2D(MemFlags.READ_WRITE, ImageFormat.RGBA8U, imgWidth, imgHeight);
img1 = c.CreateImage2D(MemFlags.READ_WRITE, ImageFormat.RGBA8U, imgWidth, imgHeight);
img2 = c.CreateImage2D(MemFlags.READ_WRITE, ImageFormat.RGBA8U, imgWidth, imgHeight);
Array.Clear(cmpData, 0, cmpData.Length);
fixed (byte* pSrc = srcData)
{
fixed (byte* pCmp = cmpData)
{
{
IntPtr[] origin = new IntPtr[3] { (IntPtr)0, (IntPtr)0, (IntPtr)0 };
IntPtr[] region = new IntPtr[3] { (IntPtr)imgWidth, (IntPtr)imgHeight, (IntPtr)1 };
IntPtr[] dstOrigin = new IntPtr[3] { (IntPtr)0, (IntPtr)0, (IntPtr)0 };
IntPtr[] dstRegion = new IntPtr[3] { (IntPtr)imgWidth, (IntPtr)imgHeight, (IntPtr)1 };
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, IntPtr.Zero, IntPtr.Zero, (IntPtr)pSrc);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, IntPtr.Zero, IntPtr.Zero, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (IntPtr version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, IntPtr.Zero, IntPtr.Zero, (IntPtr)pSrc, 0, null, out event0);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 0, null, out event1);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, IntPtr.Zero, IntPtr.Zero, (IntPtr)pCmp, 0, null, out event2);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (IntPtr version)Copy not identical to source with event output and no wait list");
Array.Clear(cmpData, 0, cmpData.Length);
Event[] events = new Event[] { event0, event1, event2 };
cq.EnqueueWriteImage(img0, true, origin, region, IntPtr.Zero, IntPtr.Zero, (IntPtr)pSrc, 3, events);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 3, events);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, IntPtr.Zero, IntPtr.Zero, (IntPtr)pCmp, 3, events);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (IntPtr version)Copy not identical to source using no event output and a wait list");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, IntPtr.Zero, IntPtr.Zero, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 3, events, out event4);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, IntPtr.Zero, IntPtr.Zero, (IntPtr)pCmp, 3, events, out event5);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (IntPtr version)Copy not identical to source using event output and a wait list");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
{
int[] origin = new int[3] { 0, 0, 0 };
int[] region = new int[3] { imgWidth, imgHeight, 1 };
int[] dstOrigin = new int[3] { 0, 0, 0 };
int[] dstRegion = new int[3] { imgWidth, imgHeight, 1 };
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, 0, 0, (IntPtr)pSrc);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, 0, 0, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (int version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, 0, 0, (IntPtr)pSrc, 0, null, out event0);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 0, null, out event1);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, 0, 0, (IntPtr)pCmp, 0, null, out event2);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (int version)Copy not identical to source with event output and no wait list");
Array.Clear(cmpData, 0, cmpData.Length);
Event[] events = new Event[] { event0, event1, event2 };
cq.EnqueueWriteImage(img0, true, origin, region, 0, 0, (IntPtr)pSrc, 3, events);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 0, events);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, 0, 0, (IntPtr)pCmp, 3, events);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (int version)Copy not identical to source using no event output and a wait list");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, 0, 0, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 0, events, out event4);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, 0, 0, (IntPtr)pCmp, 3, events, out event5);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (int version)Copy not identical to source using no event output and a wait list");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
{
long[] origin = new long[3] { 0, 0, 0 };
long[] region = new long[3] { imgWidth, imgHeight, 1 };
long[] dstOrigin = new long[3] { 0, 0, 0 };
long[] dstRegion = new long[3] { imgWidth, imgHeight, 1 };
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, 0L, 0L, (IntPtr)pSrc);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, 0L, 0L, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (long version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, 0L, 0L, (IntPtr)pSrc, 0, null, out event0);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 0, null, out event1);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, 0L, 0L, (IntPtr)pCmp, 0, null, out event2);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (long version)Copy not identical to source with event output and no wait list");
Array.Clear(cmpData, 0, cmpData.Length);
Event[] events = new Event[] { event0, event1, event2 };
cq.EnqueueWriteImage(img0, true, origin, region, 0L, 0L, (IntPtr)pSrc, 3, events);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 0, events);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, 0L, 0L, (IntPtr)pCmp, 3, events);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (long version)Copy not identical to source using no event output and a wait list");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteImage(img0, true, origin, region, 0L, 0L, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueCopyImage(img0, img1, origin, dstOrigin, region, 0, events, out event4);
cq.EnqueueBarrier();
cq.EnqueueReadImage(img1, true, origin, region, 0L, 0L, (IntPtr)pCmp, 3, events, out event5);
if (!CompareArray(cmpData, srcData))
Error("TestImageReadWriteCopyOps: (long version)Copy not identical to source using no event output and a wait list");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
}
}
}
catch (Exception e)
{
Error("Exception during testing: "+e.ToString());
}
finally
{
if (img0 != null)
img0.Dispose();
if (img1 != null)
img1.Dispose();
if (img2 != null)
img2.Dispose();
}
}
private void TestMem(Context c, CommandQueue cq, Dictionary<string, Kernel> kernelDictionary)
{
int size = 8192;
byte[] testData = new byte[size];
Output( "Testing Mem class" );
//Output( "Allocating "+size+" bytes of READ_WRITE memory" );
using (Mem buffer = c.CreateBuffer(MemFlags.READ_WRITE, size, IntPtr.Zero))
{
for (int i = 0; i < size / 2; i++)
{
testData[i] = 0;
testData[size/2+i] = 1;
}
//Output("Mem.MemSize=" + size);
if (buffer.MemSize.ToInt64() != size)
Error("Mem.Size!=input size");
//Output("Mem.MemType=" + buffer.MemType);
if (buffer.MemType != MemObjectType.BUFFER)
Error("Mem.MemType!=MemObjectType.BUFFER");
//Output("Mem.MapCount=" + buffer.MapCount);
if (buffer.MapCount != 0)
Error("Mem.MapCount!=0");
buffer.Write(cq, 0L, testData, 0, size);
Kernel k = kernelDictionary["TestReadWriteMemory"];
k.SetArg(0, buffer);
k.SetArg(1, (long)size);
Event bleh;
cq.EnqueueTask(k,0,null,out bleh);
cq.EnqueueBarrier();
cq.Finish();
buffer.Read(cq, 0L, testData, 0, size);
for (int i = 0; i < size / 2; i++)
{
if (testData[i] != 1)
{
Error("TestReadWriteMemory failed");
break;
}
if( testData[size / 2 + i] != 0 )
{
Error("TestReadWriteMemory failed");
break;
}
}
}
//Output("Allocating " + size + " bytes of READ memory");
using (Mem buffer = c.CreateBuffer(MemFlags.READ_ONLY, size, IntPtr.Zero))
{
//Output("Mem.MemSize=" + size);
if (buffer.MemSize.ToInt64() != size)
Error("Mem.Size!=input size");
//Output("Mem.MemType=" + buffer.MemType);
if (buffer.MemType != MemObjectType.BUFFER)
Error("Mem.MemType!=MemObjectType.BUFFER");
//Output("Mem.MapCount=" + buffer.MapCount);
if (buffer.MapCount != 0)
Error("Mem.MapCount!=0");
Kernel k = kernelDictionary["TestReadMemory"];
k.SetArg(0, buffer);
k.SetArg(1, (long)size);
cq.EnqueueTask(k);
cq.Finish();
}
//Output("Allocating " + size + " bytes of WRITE memory");
using (Mem buffer = c.CreateBuffer(MemFlags.WRITE_ONLY, size, IntPtr.Zero))
{
Array.Clear(testData, 0, size);
//Output("Mem.MemSize=" + size);
if (buffer.MemSize.ToInt64() != size)
Error("Mem.Size!=input size");
//Output("Mem.MemType=" + buffer.MemType);
if (buffer.MemType != MemObjectType.BUFFER)
Error("Mem.MemType!=MemObjectType.BUFFER");
//Output("Mem.MapCount=" + buffer.MapCount);
if (buffer.MapCount != 0)
Error("Mem.MapCount!=0");
Kernel k = kernelDictionary["TestWriteMemory"];
k.SetArg(0, buffer);
k.SetArg(1, (long)size);
cq.EnqueueTask(k);
cq.Finish();
buffer.Read(cq, 0L, testData, 0, size);
for (int i = 0; i < size; i++)
{
if (testData[i] != 1)
{
Error("TestWriteMemory failed");
break;
}
}
}
TestReadWriteCopyOps(c, cq);
TestImageReadWriteCopyOps(c, cq);
TestTransfersBetweenImageAndBuffers(c, cq);
TestMapBuffer(c, cq);
TestEnqueueNDRangeKernel(c, cq, kernelDictionary["EmptyKernel"]);
TestBufferRectFunctions(c, cq);
}
internal Event( Context context, CommandQueue cq, IntPtr eventID )
{
Context = context;
CommandQueue = cq;
EventID = eventID;
}
unsafe public FLACCLTask(Program _openCLProgram, int channelsCount, int channels, uint bits_per_sample, int max_frame_size, FLACCLWriter writer, int groupSize, bool gpuOnly, bool gpuRice)
{
this.UseGPUOnly = gpuOnly;
this.UseGPURice = gpuOnly && gpuRice;
this.UseMappedMemory = writer._settings.MappedMemory || writer._settings.DeviceType == OpenCLDeviceType.CPU;
this.groupSize = groupSize;
this.channels = channels;
this.channelsCount = channelsCount;
this.writer = writer;
openCLProgram = _openCLProgram;
#if DEBUG
var prop = CommandQueueProperties.PROFILING_ENABLE;
#else
var prop = CommandQueueProperties.NONE;
#endif
openCLCQ = openCLProgram.Context.CreateCommandQueue(openCLProgram.Context.Devices[0], prop);
int MAX_ORDER = this.writer.eparams.max_prediction_order;
int MAX_FRAMES = this.writer.framesPerTask;
int MAX_CHANNELSIZE = MAX_FRAMES * ((writer.eparams.block_size + 3) & ~3);
residualTasksLen = sizeof(FLACCLSubframeTask) * 32 * channelsCount * MAX_FRAMES;
bestResidualTasksLen = sizeof(FLACCLSubframeTask) * channels * MAX_FRAMES;
int samplesBufferLen = writer.PCM.BlockAlign * MAX_CHANNELSIZE * channelsCount;
int residualBufferLen = sizeof(int) * MAX_CHANNELSIZE * channels; // need to adjust residualOffset?
int partitionsLen = sizeof(int) * ((writer.PCM.BitsPerSample > 16 ? 31 : 15) * 2 << 8) * channels * MAX_FRAMES;
int riceParamsLen = sizeof(int) * (4 << 8) * channels * MAX_FRAMES;
int autocorLen = sizeof(float) * (MAX_ORDER + 1) * lpc.MAX_LPC_WINDOWS * channelsCount * MAX_FRAMES;
int lpcDataLen = autocorLen * 32;
int resOutLen = sizeof(int) * channelsCount * (lpc.MAX_LPC_WINDOWS * lpc.MAX_LPC_ORDER + 8) * MAX_FRAMES;
int wndLen = sizeof(float) * MAX_CHANNELSIZE /** 2*/ * lpc.MAX_LPC_WINDOWS;
int selectedLen = sizeof(int) * 32 * channelsCount * MAX_FRAMES;
int riceLen = sizeof(int) * channels * MAX_CHANNELSIZE;
if (!this.UseMappedMemory)
{
clSamplesBytes = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, samplesBufferLen / 2);
clResidual = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, residualBufferLen);
clBestRiceParams = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, riceParamsLen / 4);
clResidualTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, residualTasksLen);
clBestResidualTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, bestResidualTasksLen);
clWindowFunctions = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, wndLen);
clSelectedTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, selectedLen);
clRiceOutput = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, riceLen);
clSamplesBytesPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, samplesBufferLen / 2);
clResidualPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, residualBufferLen);
clBestRiceParamsPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, riceParamsLen / 4);
clResidualTasksPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, residualTasksLen);
clBestResidualTasksPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, bestResidualTasksLen);
clWindowFunctionsPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, wndLen);
clSelectedTasksPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, selectedLen);
clRiceOutputPinned = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, riceLen);
clSamplesBytesPtr = openCLCQ.EnqueueMapBuffer(clSamplesBytesPinned, true, MapFlags.READ_WRITE, 0, samplesBufferLen / 2);
clResidualPtr = openCLCQ.EnqueueMapBuffer(clResidualPinned, true, MapFlags.READ_WRITE, 0, residualBufferLen);
clBestRiceParamsPtr = openCLCQ.EnqueueMapBuffer(clBestRiceParamsPinned, true, MapFlags.READ_WRITE, 0, riceParamsLen / 4);
clResidualTasksPtr = openCLCQ.EnqueueMapBuffer(clResidualTasksPinned, true, MapFlags.READ_WRITE, 0, residualTasksLen);
clBestResidualTasksPtr = openCLCQ.EnqueueMapBuffer(clBestResidualTasksPinned, true, MapFlags.READ_WRITE, 0, bestResidualTasksLen);
clWindowFunctionsPtr = openCLCQ.EnqueueMapBuffer(clWindowFunctionsPinned, true, MapFlags.READ_WRITE, 0, wndLen);
clSelectedTasksPtr = openCLCQ.EnqueueMapBuffer(clSelectedTasksPinned, true, MapFlags.READ_WRITE, 0, selectedLen);
clRiceOutputPtr = openCLCQ.EnqueueMapBuffer(clRiceOutputPinned, true, MapFlags.READ_WRITE, 0, riceLen);
}
else
{
clSamplesBytes = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, (uint)samplesBufferLen / 2);
clResidual = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, residualBufferLen);
clBestRiceParams = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, riceParamsLen / 4);
clResidualTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, residualTasksLen);
clBestResidualTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, bestResidualTasksLen);
clWindowFunctions = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, wndLen);
clSelectedTasks = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, selectedLen);
clRiceOutput = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE | MemFlags.ALLOC_HOST_PTR, riceLen);
clSamplesBytesPtr = openCLCQ.EnqueueMapBuffer(clSamplesBytes, true, MapFlags.READ_WRITE, 0, samplesBufferLen / 2);
clResidualPtr = openCLCQ.EnqueueMapBuffer(clResidual, true, MapFlags.READ_WRITE, 0, residualBufferLen);
clBestRiceParamsPtr = openCLCQ.EnqueueMapBuffer(clBestRiceParams, true, MapFlags.READ_WRITE, 0, riceParamsLen / 4);
clResidualTasksPtr = openCLCQ.EnqueueMapBuffer(clResidualTasks, true, MapFlags.READ_WRITE, 0, residualTasksLen);
clBestResidualTasksPtr = openCLCQ.EnqueueMapBuffer(clBestResidualTasks, true, MapFlags.READ_WRITE, 0, bestResidualTasksLen);
clWindowFunctionsPtr = openCLCQ.EnqueueMapBuffer(clWindowFunctions, true, MapFlags.READ_WRITE, 0, wndLen);
clSelectedTasksPtr = openCLCQ.EnqueueMapBuffer(clSelectedTasks, true, MapFlags.READ_WRITE, 0, selectedLen);
clRiceOutputPtr = openCLCQ.EnqueueMapBuffer(clRiceOutput, true, MapFlags.READ_WRITE, 0, riceLen);
//clSamplesBytesPtr = clSamplesBytes.HostPtr;
//clResidualPtr = clResidual.HostPtr;
//clBestRiceParamsPtr = clBestRiceParams.HostPtr;
//clResidualTasksPtr = clResidualTasks.HostPtr;
//clBestResidualTasksPtr = clBestResidualTasks.HostPtr;
//clWindowFunctionsPtr = clWindowFunctions.HostPtr;
}
clSamples = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, samplesBufferLen);
clLPCData = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, lpcDataLen);
clAutocorOutput = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, autocorLen);
clSelectedTasksSecondEstimate = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, selectedLen);
clSelectedTasksBestMethod = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, selectedLen);
if (UseGPUOnly)
{
clPartitions = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, partitionsLen);
clRiceParams = openCLProgram.Context.CreateBuffer(MemFlags.READ_WRITE, riceParamsLen);
}
//openCLCQ.EnqueueMapBuffer(clSamplesBytes, true, MapFlags.WRITE, 0, samplesBufferLen / 2);
clComputeAutocor = openCLProgram.CreateKernel("clComputeAutocor");
clStereoDecorr = openCLProgram.CreateKernel("clStereoDecorr");
//cudaChannelDecorr = openCLProgram.CreateKernel("clChannelDecorr");
clChannelDecorr2 = openCLProgram.CreateKernel("clChannelDecorr2");
clChannelDecorrX = openCLProgram.CreateKernel("clChannelDecorrX");
clFindWastedBits = openCLProgram.CreateKernel("clFindWastedBits");
clComputeLPC = openCLProgram.CreateKernel("clComputeLPC");
clQuantizeLPC = openCLProgram.CreateKernel("clQuantizeLPC");
//cudaComputeLPCLattice = openCLProgram.CreateKernel("clComputeLPCLattice");
clSelectStereoTasks = openCLProgram.CreateKernel("clSelectStereoTasks");
clEstimateResidual = openCLProgram.CreateKernel("clEstimateResidual");
clChooseBestMethod = openCLProgram.CreateKernel("clChooseBestMethod");
if (UseGPUOnly)
{
clEncodeResidual = openCLProgram.CreateKernel("clEncodeResidual");
if (openCLCQ.Device.DeviceType != DeviceType.CPU)
{
clCalcPartition = openCLProgram.CreateKernel("clCalcPartition");
clCalcPartition16 = openCLProgram.CreateKernel("clCalcPartition16");
}
clSumPartition = openCLProgram.CreateKernel("clSumPartition");
clFindRiceParameter = openCLProgram.CreateKernel("clFindRiceParameter");
clFindPartitionOrder = openCLProgram.CreateKernel("clFindPartitionOrder");
if (UseGPURice)
{
clCalcOutputOffsets = openCLProgram.CreateKernel("clCalcOutputOffsets");
clRiceEncoding = openCLProgram.CreateKernel("clRiceEncoding");
}
}
samplesBuffer = new int[MAX_CHANNELSIZE * channelsCount];
outputBuffer = new byte[max_frame_size * MAX_FRAMES + 1];
frame = new FlacFrame(channelsCount);
frame.writer = new BitWriter(outputBuffer, 0, outputBuffer.Length);
if (writer._settings.DoVerify)
{
verify = new FlakeReader(new AudioPCMConfig((int)bits_per_sample, channels, 44100));
verify.DoCRC = false;
}
}
protected virtual void Initialize(DeviceType deviceType, string source)
{
Devices = Platform.QueryDevices(deviceType);
if (Devices.Length == 0)
throw new OpenCLException("No devices of type "+deviceType+" present");
Context = Platform.CreateContext(null,Devices,null, IntPtr.Zero);
CQs = new CommandQueue[Devices.Length];
for( int i=0; i<CQs.Length; i++ )
CQs[i] = Context.CreateCommandQueue(Devices[i], CommandQueueProperties.PROFILING_ENABLE);
CQ = CQs[0];
Program = Context.CreateProgramWithSource(source);
Program.Build();
Kernels = Program.CreateKernelDictionary();
}
public Core(int Nxp,int Nyp, int Nzp, int Ntm, double Bbeta, double Flux)
{
Nx = Nxp; Ny = Nyp; Nz = Nzp; Nt = Ntm; betagauge = (floattype)Bbeta; flux = (floattype)Flux;
N = Nx * Ny * Nz * Nt; Nspace = Nx * Ny * Nz;
string strforcompiler = "-D Nt=" + Nt.ToString() + " -D Nxyz=" + (Nx * Ny * Nz).ToString() + " -D Nxy=" + (Nx*Ny).ToString() +
" -D Nx="+(Nx).ToString()+" -D Ny="+(Ny).ToString()+" -D Nz="+(Nz).ToString();
strforcompiler += typeof(floattype) == typeof(double) ? " -D floattype=double -D floattype2=double2 -D floattype4=double4" :
" -D floattype=float -D floattype2=float2 -D floattype4=float4";
strforcompiler += " -D phi=" + flux.ToString().Replace(',', '.') + " -D KAPPA=" + kappa.ToString().Replace(',', '.');
string fp64support = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n";
Plocalsize = AdjustLocalSize(Nspace);
Slocalsize = AdjustLocalSize(N / 2);
XhermYlocalsize = AdjustLocalSize(4 * N);
// Plocalsize = 16; Slocalsize = 16;
PNumGroups = Nx * Ny * Nz / Plocalsize;
SNumGroups = N/2 / Slocalsize;
XhermYNumGroups = 4*4*N / XhermYlocalsize;
BufferLength = N * 4 * 9 * 2 * sizeof(floattype);
SeedBufLen = N * sizeof(Int32)/2 * 4;
AllocBuffers();
openCLPlatform = OpenCL.GetPlatform(0);
openCLDevices = openCLPlatform.QueryDevices(DeviceType.ALL);
openCLContext = openCLPlatform.CreateDefaultContext();
openCLCQ = openCLContext.CreateCommandQueue(openCLDevices[0], CommandQueueProperties.PROFILING_ENABLE);
MyKernelProgram = openCLContext.CreateProgramWithSource(
(typeof(floattype)==typeof(double)?fp64support:"") + File.ReadAllText("MyKernel.cl")+File.ReadAllText("dirak_mul.cl"));
try
{
MyKernelProgram.Build(openCLDevices, strforcompiler, null, IntPtr.Zero);
}
catch (OpenCLException)
{
string buildLog = MyKernelProgram.GetBuildLog(openCLDevices[0]);
MessageBox.Show(buildLog, "Build error(64 bit debug sessions in vs2008 always fail like this - debug in 32 bit or use vs2010)");
// Application.Exit();
}
MyKernelKernel = MyKernelProgram.CreateKernel("MyKernel");
PReductionKernel = MyKernelProgram.CreateKernel("PLoop");
SReductionKernel = MyKernelProgram.CreateKernel("CalcS");
DiralMulKernel = MyKernelProgram.CreateKernel("dirakMatrMul");
FillWithKernel = MyKernelProgram.CreateKernel("FillWith");
FillLinkWithKernel = MyKernelProgram.CreateKernel("FillLinkWith");
FillWithRandomKernel = MyKernelProgram.CreateKernel("FillWithRandom");
AXPYKernel = MyKernelProgram.CreateKernel("AXPY");
XhermYKernel = MyKernelProgram.CreateKernel("XhermY");
BackupLinkKernel = MyKernelProgram.CreateKernel("BackupLink");
RestoreLinkKernel = MyKernelProgram.CreateKernel("RestoreLink");
SeedMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), SeedBufLen, IntPtr.Zero);
LinkMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), BufferLength, IntPtr.Zero);
PGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * PNumGroups, IntPtr.Zero);
PResMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
SGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * SNumGroups, IntPtr.Zero);
SResMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
XhermYGroupMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * 2*XhermYNumGroups, IntPtr.Zero);
XhermYresMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize * 2, IntPtr.Zero);
XhermYrespointer = System.Runtime.InteropServices.Marshal.AllocHGlobal(floatsize * 2);
SeedVectorMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), SeedVectorBuf.Length * sizeof(int), IntPtr.Zero);
StorageMem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), linksize, IntPtr.Zero);
dSmem = openCLContext.CreateBuffer((MemFlags)((long)MemFlags.READ_WRITE), floatsize, IntPtr.Zero);
dSpointer = System.Runtime.InteropServices.Marshal.AllocHGlobal(floatsize);
MyKernelKernel.SetArg(0, (byte)EvenOdd);
MyKernelKernel.SetArg(1, (floattype)betagauge);
MyKernelKernel.SetArg(2, (floattype)flux);
MyKernelKernel.SetArg(3, SeedMem);
MyKernelKernel.SetArg(4, LinkMem);
PReductionKernel.SetArg(0, LinkMem);
PReductionKernel.SetArg(1, PGroupMem);
PReductionKernel.SetArg(2, PResMem);
IntPtr ptr = new IntPtr(Plocalsize * floatsize);
PReductionKernel.SetArg(3, ptr, IntPtr.Zero);
SReductionKernel.SetArg(0, LinkMem);
SReductionKernel.SetArg(1, SGroupMem);
SReductionKernel.SetArg(2, SResMem);
IntPtr ptr1 = new IntPtr(Slocalsize * floatsize);
SReductionKernel.SetArg(3, ptr1, IntPtr.Zero);
XhermYKernel.SetArg(2, XhermYresMem);
XhermYKernel.SetArg(3, XhermYGroupMem);
XhermYKernel.SetArg(4, new IntPtr(XhermYlocalsize*floatsize*2),IntPtr.Zero);
openCLCQ.EnqueueWriteBuffer(SeedMem, true, 0, SeedBufLen, ipseed);
openCLCQ.EnqueueWriteBuffer(LinkMem, true, 0, BufferLength, ip);
openCLCQ.EnqueueWriteBuffer(SeedVectorMem, true, 0, SeedVectorBuf.Length*sizeof(int), ipseedvector);
rhat0 = new Vector();
//init BICGStab vectors
phi = new Vector();
r0 = new Vector();
//rprev = new Vector();
pi = new Vector();
vi = new Vector();
t = new Vector();
s = new Vector();
// xprev = new Vector();
// vprev = new Vector();
// pprev = new Vector();
temp = new Vector();
ri = new Vector();
x = new Vector();
//for fermion update
chi = new Vector();
CalculateS();
double s1 = S[0];
BackupLink(0, 0,1, 0, 1);
CalculateS();
double s2 = S[0];
RestoreLink(0, 0, 1, 0, 1);
CalculateS();
double s3 = S[0];
//MessageBox.Show(s1.ToString() + s2.ToString() + s3.ToString());
}
private void TestCommandQueueMemCopy(Context c, CommandQueue cq)
{
Output("Testing synchronous host memory->memory copy");
AlignedArrayFloat aafSrc = new AlignedArrayFloat(1024 * 1024, 64);
AlignedArrayFloat aafDst = new AlignedArrayFloat(1024 * 1024, 64);
SetAAF(aafSrc, 0.0f);
SetAAF(aafDst, 1.0f);
/// Test HOST_PTR -> HOST_PTR copy
/// The call to EnqueueMapBuffer synchronizes caches before testing the result
using (Mem memSrc = c.CreateBuffer((MemFlags)((ulong)MemFlags.READ_WRITE+(ulong)MemFlags.USE_HOST_PTR), aafSrc.ByteLength, aafSrc))
{
using (Mem memDst = c.CreateBuffer((MemFlags)((ulong)MemFlags.READ_WRITE+(ulong)MemFlags.USE_HOST_PTR), aafDst.ByteLength, aafDst))
{
cq.EnqueueCopyBuffer(memSrc, memDst, IntPtr.Zero, IntPtr.Zero, (IntPtr)aafSrc.ByteLength);
cq.EnqueueBarrier();
IntPtr mappedPtr = cq.EnqueueMapBuffer(memDst, true, MapFlags.READ_WRITE, (IntPtr)0, (IntPtr)aafDst.ByteLength);
if (!TestAAF(aafDst, 0.0f))
Error("EnqueueCopyBuffer failed, destination is invalid");
cq.EnqueueUnmapMemObject(memDst, mappedPtr);
cq.EnqueueBarrier();
}
}
/// Test COPY_HOST_PTR -> COPY_HOST_PTR copy
/// Verify that original source buffers are intact and that the copy was successful
SetAAF(aafSrc, 0.0f);
SetAAF(aafDst, 1.0f);
using (Mem memSrc = c.CreateBuffer(MemFlags.COPY_HOST_PTR, aafSrc.ByteLength, aafSrc))
{
using (Mem memDst = c.CreateBuffer(MemFlags.COPY_HOST_PTR, aafSrc.ByteLength, aafDst))
{
SetAAF(aafSrc, 2.0f);
SetAAF(aafDst, 3.0f);
cq.EnqueueCopyBuffer(memSrc, memDst, IntPtr.Zero, IntPtr.Zero, (IntPtr)aafSrc.ByteLength);
cq.Finish();
if (!TestAAF(aafSrc, 2.0f))
Error("Memory copy destroyed src buffer");
if (!TestAAF(aafDst, 3.0f))
Error("Memory copy destroyed dst buffer");
Event ev;
cq.EnqueueReadBuffer(memDst, false, IntPtr.Zero, (IntPtr)aafDst.ByteLength, aafDst,0, null, out ev);
cq.EnqueueWaitForEvents(1, new Event[] { ev });
ev.Dispose();
cq.Finish();
if (!TestAAF(aafDst, 0.0f))
Error("Memory copy failed");
}
}
/// Test ALLOC_HOST_PTR -> ALLOC_HOST_PTR copy
SetAAF(aafSrc, 0.0f);
SetAAF(aafDst, 1.0f);
using (Mem memSrc = c.CreateBuffer((MemFlags)((ulong)MemFlags.ALLOC_HOST_PTR + (ulong)MemFlags.READ_WRITE), aafSrc.ByteLength, IntPtr.Zero))
{
using (Mem memDst = c.CreateBuffer((MemFlags)((ulong)MemFlags.ALLOC_HOST_PTR + (ulong)MemFlags.WRITE_ONLY), aafSrc.ByteLength, IntPtr.Zero))
{
cq.EnqueueWriteBuffer(memSrc, false, (IntPtr)0, (IntPtr)aafSrc.ByteLength, aafSrc);
cq.EnqueueWriteBuffer(memDst, false, (IntPtr)0, (IntPtr)aafSrc.ByteLength, aafSrc);
cq.EnqueueBarrier();
cq.EnqueueCopyBuffer(memSrc, memDst, IntPtr.Zero, IntPtr.Zero, (IntPtr)aafSrc.ByteLength);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(memDst, true, IntPtr.Zero, (IntPtr)aafDst.ByteLength, aafDst);
if (!TestAAF(aafDst, 0.0f))
Error("Memory copy failed");
}
}
/// Test DEFAULT -> DEFAULT copy
SetAAF(aafSrc, 0.0f);
SetAAF(aafDst, 1.0f);
using (Mem memSrc = c.CreateBuffer((MemFlags)((ulong)MemFlags.ALLOC_HOST_PTR + (ulong)MemFlags.READ_ONLY), aafSrc.ByteLength, IntPtr.Zero))
{
using (Mem memDst = c.CreateBuffer((MemFlags)((ulong)MemFlags.ALLOC_HOST_PTR + (ulong)MemFlags.WRITE_ONLY), aafSrc.ByteLength, IntPtr.Zero))
{
cq.EnqueueWriteBuffer(memSrc, false, (IntPtr)0, (IntPtr)aafSrc.ByteLength, aafSrc);
cq.EnqueueWriteBuffer(memDst, false, (IntPtr)0, (IntPtr)aafSrc.ByteLength, aafSrc);
cq.EnqueueBarrier();
cq.EnqueueCopyBuffer(memSrc, memDst, IntPtr.Zero, IntPtr.Zero, (IntPtr)aafSrc.ByteLength);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(memDst, true, IntPtr.Zero, (IntPtr)aafDst.ByteLength, aafDst);
if (!TestAAF(aafDst, 0.0f))
Error("Memory copy failed");
}
}
}
/// <summary>
/// Test all versions of:
///
/// EnqueueReadBuffer
/// EnqueueWriteBuffer
/// EnqueueCopyBuffer
///
/// The test just copies the entirety of a buffer and checks if the result is equal to the original.
/// An error indicates that one of the above functions failed and further manual analysis is required
/// to pinpoint the error.
/// </summary>
/// <param name="c"></param>
/// <param name="cq"></param>
private void TestReadWriteCopyOps(Context c, CommandQueue cq)
{
Output("Testing read/write/copy functions");
Mem buf0 = null;
Mem buf1 = null;
Mem buf2 = null;
int bufLen = 1024 * 1024;
byte[] srcData = new byte[bufLen];
byte[] cmpData = new byte[bufLen];
Event event0;
Event event1;
Event event2;
Event event3;
Event event4;
Event event5;
for (int i = 0; i < srcData.Length; i++)
srcData[i] = (byte)(i);
Array.Clear(cmpData, 0, cmpData.Length);
try
{
buf0 = c.CreateBuffer(MemFlags.READ_WRITE, bufLen, IntPtr.Zero);
buf1 = c.CreateBuffer(MemFlags.READ_WRITE, bufLen, IntPtr.Zero);
buf2 = c.CreateBuffer(MemFlags.READ_WRITE, bufLen, IntPtr.Zero);
#region Test EnqueueReadBuffer EnqueueWriteBuffer EnqueueCopyBuffer
fixed (byte* pSrc = srcData)
{
fixed (byte* pCmp = cmpData)
{
{
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, (IntPtr)0, (IntPtr)bufLen, (IntPtr)pSrc);
cq.EnqueueCopyBuffer(buf0, buf1, (IntPtr)0, (IntPtr)0, (IntPtr)bufLen);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, (IntPtr)0, (IntPtr)bufLen, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(IntPtr version): Copy not identical to source");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, (IntPtr)0, (IntPtr)bufLen, (IntPtr)pSrc, 0, null, out event0 );
cq.EnqueueCopyBuffer(buf0, buf1, (IntPtr)0, (IntPtr)0, (IntPtr)bufLen, 0, null, out event1);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, (IntPtr)0, (IntPtr)bufLen, (IntPtr)pCmp, 0, null, out event2);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(IntPtr version): Copy not identical to source");
Event[] events = new Event[] { event0, event1, event2 };
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, (IntPtr)0, (IntPtr)bufLen, (IntPtr)pSrc, 3, events);
cq.EnqueueCopyBuffer(buf0, buf1, (IntPtr)0, (IntPtr)0, (IntPtr)bufLen, 3, events);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, (IntPtr)0, (IntPtr)bufLen, (IntPtr)pCmp, 3, events);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(IntPtr version): Copy not identical to source");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, (IntPtr)0, (IntPtr)bufLen, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueCopyBuffer(buf0, buf1, (IntPtr)0, (IntPtr)0, (IntPtr)bufLen, 3, events, out event4);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, (IntPtr)0, (IntPtr)bufLen, (IntPtr)pCmp, 3, events, out event5);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(IntPtr version): Copy not identical to source");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
{
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, 0, bufLen, (IntPtr)pSrc);
cq.EnqueueCopyBuffer(buf0, buf1, 0, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, 0, bufLen, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(int version): Copy not identical to source");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, 0, bufLen, (IntPtr)pSrc, 0, null, out event0);
cq.EnqueueCopyBuffer(buf0, buf1, 0, 0, bufLen, 0, null, out event1);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, 0, bufLen, (IntPtr)pCmp, 0, null, out event2);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(int version): Copy not identical to source");
Event[] events = new Event[] { event0, event1, event2 };
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, 0, bufLen, (IntPtr)pSrc, 3, events);
cq.EnqueueCopyBuffer(buf0, buf1, 0, 0, bufLen, 3, events);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, 0, bufLen, (IntPtr)pCmp, 3, events);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(int version): Copy not identical to source");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, 0, bufLen, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueCopyBuffer(buf0, buf1, 0, 0, bufLen, 3, events, out event4);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, 0, bufLen, (IntPtr)pCmp, 3, events, out event5);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(int version): Copy not identical to source");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
{
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, 0L, (long)bufLen, (IntPtr)pSrc);
cq.EnqueueCopyBuffer(buf0, buf1, 0L, 0L, (long)bufLen);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, 0L, (long)bufLen, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(long version): Copy not identical to source");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, 0L, (long)bufLen, (IntPtr)pSrc, 0, null, out event0);
cq.EnqueueCopyBuffer(buf0, buf1, 0L, 0L, (long)bufLen, 0, null, out event1);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, 0L, (long)bufLen, (IntPtr)pCmp, 0, null, out event2);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(long version): Copy not identical to source");
Event[] events = new Event[] { event0, event1, event2 };
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, 0L, (long)bufLen, (IntPtr)pSrc, 3, events);
cq.EnqueueCopyBuffer(buf0, buf1, 0L, 0L, (long)bufLen, 3, events);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, 0L, (long)bufLen, (IntPtr)pCmp, 3, events);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(long version): Copy not identical to source");
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueWriteBuffer(buf0, true, 0L, (long)bufLen, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueCopyBuffer(buf0, buf1, 0L, 0L, (long)bufLen, 3, events, out event4);
cq.EnqueueBarrier();
cq.EnqueueReadBuffer(buf1, true, 0L, (long)bufLen, (IntPtr)pCmp, 3, events, out event5);
if (!CompareArray(cmpData, srcData))
Error("TestReadWriteCopyOps(long version): Copy not identical to source");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
}
}
#endregion
}
catch (Exception e)
{
Error("Exception during testing: " + e.ToString());
}
finally
{
if (buf0 != null)
buf0.Dispose();
if (buf1 != null)
buf1.Dispose();
if (buf2 != null)
buf2.Dispose();
}
}
/// <summary>
/// Test all versions of:
///
/// EnqueueNDRangeKernel
///
/// The tests just issue a dummy kernel a bunch of times with the various overloads
/// </summary>
/// <param name="c"></param>
/// <param name="cq"></param>
private void TestEnqueueNDRangeKernel(Context c, CommandQueue cq, Kernel k )
{
Output("Testing EnqueueNDRangeKernel");
Event event0 = null;
Event event1 = null;
try
{
{
IntPtr[] globalWorkSize = new IntPtr[] { (IntPtr)10 };
IntPtr[] localWorkSize = new IntPtr[] { (IntPtr)1 };
cq.EnqueueNDRangeKernel(k, (uint)1, null, globalWorkSize, localWorkSize);
cq.EnqueueNDRangeKernel(k, (uint)1, null, globalWorkSize, localWorkSize, 0, null, out event0);
Event[] waitList = new Event[] { event0 };
cq.EnqueueNDRangeKernel(k, (uint)1, null, globalWorkSize, localWorkSize, 1, waitList, out event1);
cq.Finish();
event0.Dispose();
event1.Dispose();
}
{
int[] globalWorkSize = new int[] { (int)10 };
int[] localWorkSize = new int[] { (int)1 };
cq.EnqueueNDRangeKernel(k, 1, null, globalWorkSize, localWorkSize);
cq.EnqueueNDRangeKernel(k, 1, null, globalWorkSize, localWorkSize, 0, null, out event0);
Event[] waitList = new Event[] { event0 };
cq.EnqueueNDRangeKernel(k, 1, null, globalWorkSize, localWorkSize, 1, waitList, out event1);
cq.Finish();
event0.Dispose();
event1.Dispose();
}
{
long[] globalWorkSize = new long[] { (long)10 };
long[] localWorkSize = new long[] { (long)1 };
cq.EnqueueNDRangeKernel(k, 1, null, globalWorkSize, localWorkSize);
cq.EnqueueNDRangeKernel(k, 1, null, globalWorkSize, localWorkSize, 0, null, out event0);
Event[] waitList = new Event[] { event0 };
cq.EnqueueNDRangeKernel(k, 1, null, globalWorkSize, localWorkSize, 1, waitList, out event1);
cq.Finish();
event0.Dispose();
event1.Dispose();
}
}
catch (Exception e)
{
Error("Exception during testing: " + e.ToString());
}
finally
{
if (event0 != null)
event0.Dispose();
if (event1 != null)
event1.Dispose();
}
}
/// <summary>
/// Test all versions of:
///
/// EnqueueReadBufferRect
/// EnqueueWriteBufferRect
/// EnqueueCopyBufferRect
///
/// The test just copies the entirety of a buffer and checks if the result is equal to the original.
/// An error indicates that one of the above functions failed and further manual analysis is required
/// to pinpoint the error.
/// </summary>
/// <param name="c"></param>
/// <param name="cq"></param>
private void TestBufferRectFunctions(Context c, CommandQueue cq)
{
if (!(cq.Device.Platform.OpenCLMajorVersion >= 1 && cq.Device.Platform.OpenCLMinorVersion >= 1))
{
Output("Skipping EnqueueReadBufferRect, EnqueueWriteBufferRect and EnqueueCopyBufferRect tests(Requires OpenCL 1.1 or higher)");
return;
}
Output("Testing EnqueueReadBufferRect, EnqueueWriteBufferRect and EnqueueCopyBufferRect");
OpenCLNet.Mem mem0 = null;
OpenCLNet.Mem mem1 = null;
int bufWidth = 16;
int bufHeight = 16;
int bufLen = bufWidth * bufHeight;
byte[] srcData = new byte[bufLen];
byte[] cmpData = new byte[bufLen];
Event event0;
Event event1;
Event event2;
Event event3;
Event event4;
Event event5;
Array.Clear(srcData, 0, srcData.Length);
for (int i = 8; i < 12; i++)
for (int j = 8; j < 12; j++)
srcData[bufWidth * i + j] = (byte)1;
Array.Clear(cmpData, 0, cmpData.Length);
try
{
mem0 = c.CreateBuffer(MemFlags.READ_WRITE, bufLen, IntPtr.Zero);
mem1 = c.CreateBuffer(MemFlags.READ_WRITE, bufLen, IntPtr.Zero);
fixed (byte* pSrc = srcData)
{
fixed (byte* pCmp = cmpData)
{
{
IntPtr[] bufferOffset = new IntPtr[3] { (IntPtr)0, (IntPtr)0, (IntPtr)0 };
IntPtr[] hostOffset = new IntPtr[3] { (IntPtr)0, (IntPtr)0, (IntPtr)0 };
IntPtr[] region = new IntPtr[3] { (IntPtr)bufWidth, (IntPtr)bufHeight, (IntPtr)1 };
IntPtr bufferRowPitch = (IntPtr)bufWidth;
IntPtr bufferSlicePitch = (IntPtr)0;
IntPtr hostRowPitch = (IntPtr)bufWidth;
IntPtr hostSlicePitch = (IntPtr)0;
cq.EnqueueWriteBufferRect(mem0, true, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch);
cq.Finish();
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, true, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (IntPtr version)Copy not identical to source when using no event args");
cq.EnqueueWriteBufferRect(mem0, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc, 0, null, out event0);
cq.EnqueueWaitForEvent(event0);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, 0, null, out event1);
cq.EnqueueWaitForEvent(event1);
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp, 0, null, out event2);
cq.Finish();
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (IntPtr version)Copy not identical to source when using event output and no event args");
Event[] events = new Event[] { event0, event1, event2 };
cq.EnqueueWriteBufferRect(mem0, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueWaitForEvent(event3);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, 3, events, out event4);
cq.EnqueueWaitForEvent(event4);
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp, 3, events, out event5);
cq.Finish();
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (IntPtr version)Copy not identical to source when using event output and event args");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
{
int[] bufferOffset = new int[3] { 0, 0, 0 };
int[] hostOffset = new int[3] { 0, 0, 0 };
int[] region = new int[3] { bufWidth, bufHeight, 1 };
int bufferRowPitch = bufWidth;
int bufferSlicePitch = 0;
int hostRowPitch = bufWidth;
int hostSlicePitch = 0;
cq.EnqueueWriteBufferRect(mem0, true, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch);
cq.Finish();
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, true, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (int version)Copy not identical to source when using no event args");
cq.EnqueueWriteBufferRect(mem0, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc, 0, null, out event0);
cq.EnqueueWaitForEvent(event0);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, 0, null, out event1);
cq.EnqueueWaitForEvent(event1);
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp, 0, null, out event2);
cq.Finish();
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (int version)Copy not identical to source when using event output and no event args");
Event[] events = new Event[] { event0, event1, event2 };
cq.EnqueueWriteBufferRect(mem0, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueWaitForEvent(event3);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, 3, events, out event4);
cq.EnqueueWaitForEvent(event4);
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp, 3, events, out event5);
cq.Finish();
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (int version)Copy not identical to source when using event output and event args");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
{
long[] bufferOffset = new long[3] { 0L, 0L, 0L };
long[] hostOffset = new long[3] { 0L, 0L, 0L };
long[] region = new long[3] { (long)bufWidth, (long)bufHeight, (long)1 };
long bufferRowPitch = bufWidth;
long bufferSlicePitch = 0;
long hostRowPitch = bufWidth;
long hostSlicePitch = 0;
cq.EnqueueWriteBufferRect(mem0, true, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch);
cq.Finish();
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, true, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp);
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (long version)Copy not identical to source when using no event args");
cq.EnqueueWriteBufferRect(mem0, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc, 0, null, out event0);
cq.EnqueueWaitForEvent(event0);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, 0, null, out event1);
cq.EnqueueWaitForEvent(event1);
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp, 0, null, out event2);
cq.Finish();
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (long version)Copy not identical to source when using event output and no event args");
Event[] events = new Event[] { event0, event1, event2 };
cq.EnqueueWriteBufferRect(mem0, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pSrc, 3, events, out event3);
cq.EnqueueWaitForEvent(event3);
cq.EnqueueCopyBufferRect(mem0, mem1, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, 3, events, out event4);
cq.EnqueueWaitForEvent(event4);
Array.Clear(cmpData, 0, cmpData.Length);
cq.EnqueueReadBufferRect(mem1, false, bufferOffset, hostOffset, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch, (IntPtr)pCmp, 3, events, out event5);
cq.Finish();
if (!CompareArray(cmpData, srcData))
Error("Read-/Write-/CopyRect: (long version)Copy not identical to source when using event output and event args");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
event4.Dispose();
event5.Dispose();
}
}
}
}
catch (Exception e)
{
Error("Exception during testing: " + e.ToString());
}
finally
{
if (mem0 != null)
mem0.Dispose();
if (mem1 != null)
mem1.Dispose();
}
}
private unsafe void TestKernel(Context c, CommandQueue cq, Kernel argIOKernel)
{
Mem outArgBuffer = c.CreateBuffer((MemFlags)((ulong)MemFlags.ALLOC_HOST_PTR|(ulong)MemFlags.READ_WRITE), sizeof(IOKernelArgs), IntPtr.Zero);
byte[] data = new byte[sizeof(IOKernelArgs)];
Output("Testing kernel - Argument return");
argIOKernel.SetArg(0, 1);
argIOKernel.SetArg(1, 65L);
argIOKernel.SetArg(2, 38.4f);
argIOKernel.SetArg(3, outArgBuffer);
Event ev;
cq.EnqueueTask(argIOKernel,0,null,out ev);
cq.Finish();
if ((int)ev.ExecutionStatus < 0)
{
Error(cq.Device.Name + ": argIOKernel failed with error code " + (ErrorCode)ev.ExecutionStatus);
ev.Dispose();
}
else
{
outArgBuffer.Read(cq, 0L, data, 0, sizeof(IOKernelArgs));
IntPtr outArgPtr = cq.EnqueueMapBuffer(outArgBuffer, true, MapFlags.READ, IntPtr.Zero, (IntPtr)sizeof(IOKernelArgs));
IOKernelArgs args = (IOKernelArgs)Marshal.PtrToStructure(outArgPtr, typeof(IOKernelArgs));
cq.EnqueueUnmapMemObject(outArgBuffer, outArgPtr);
if (args.outInt != 1)
Error(cq.Device.Name + ": argIOKernel failed to return correct arguments");
if (args.outLong != 65)
Error(cq.Device.Name + ": argIOKernel failed to return correct arguments");
if (args.outSingle != 38.4f)
Error(cq.Device.Name + ": argIOKernel failed to return correct arguments");
}
}
/// <summary>
/// Test all versions of:
///
/// EnqueueCopyImageToBuffer
/// EnqueueCopyBufferToImage
///
/// The test just copies the entirety of a buffer and checks if the result is equal to the original.
/// An error indicates that one of the above functions failed and further manual analysis is required
/// to pinpoint the error.
/// </summary>
/// <param name="c"></param>
/// <param name="cq"></param>
private void TestTransfersBetweenImageAndBuffers(Context c, CommandQueue cq)
{
if (!cq.Device.ImageSupport)
{
Output("Skipping CopyImageToBuffer and CopyBufferToImage tests(not supported on this device)");
return;
}
Output("Testing CopyImageToBuffer and CopyBufferToImage");
OpenCLNet.Image img0 = null;
OpenCLNet.Mem mem0 = null;
int imgWidth = 1024;
int imgHeight = 1024;
int bufLen = imgWidth * 4 * imgHeight;
byte[] srcData = new byte[bufLen];
byte[] cmpData = new byte[bufLen];
Event event0;
Event event1;
Event event2;
Event event3;
for (int i = 0; i < srcData.Length; i++)
srcData[i] = (byte)(i);
Array.Clear(cmpData, 0, cmpData.Length);
try
{
img0 = c.CreateImage2D(MemFlags.READ_WRITE, ImageFormat.RGBA8U, imgWidth, imgHeight);
mem0 = c.CreateBuffer(MemFlags.READ_WRITE, bufLen, IntPtr.Zero);
Array.Clear(cmpData, 0, cmpData.Length);
fixed (byte* pSrc = srcData)
{
fixed (byte* pCmp = cmpData)
{
{
IntPtr[] origin = new IntPtr[3] { (IntPtr)0, (IntPtr)0, (IntPtr)0 };
IntPtr[] region = new IntPtr[3] { (IntPtr)imgWidth, (IntPtr)imgHeight, (IntPtr)1 };
IntPtr[] dstOrigin = new IntPtr[3] { (IntPtr)0, (IntPtr)0, (IntPtr)0 };
IntPtr[] dstRegion = new IntPtr[3] { (IntPtr)imgWidth, (IntPtr)imgHeight, (IntPtr)1 };
Array.Clear(cmpData, 0, cmpData.Length);
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, (IntPtr)0, origin, region);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, (IntPtr)0);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (IntPtr version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, (IntPtr)0, origin, region, 0, null, out event0);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, (IntPtr)0, 0, null, out event1);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (IntPtr version)Copy not identical to source when using event output and event args");
Array.Clear(cmpData, 0, cmpData.Length);
Event[] events = new Event[] { event0, event1 };
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, (IntPtr)0, origin, region, 2, events, out event2);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, (IntPtr)0, 2, events, out event3);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (IntPtr version)Copy not identical to source when using event output and a wait list");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
}
{
int[] origin = new int[3] { 0, 0, 0 };
int[] region = new int[3] { imgWidth, imgHeight, 1 };
int[] dstOrigin = new int[3] { 0, 0, 0 };
int[] dstRegion = new int[3] { imgWidth, imgHeight, 1 };
Array.Clear(cmpData, 0, cmpData.Length);
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, 0, origin, region);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, 0);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (int version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, 0, origin, region, 0, null, out event0);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, 0, 0, null, out event1);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (int version)Copy not identical to source when using event output no event args");
Array.Clear(cmpData, 0, cmpData.Length);
Event[] events = new Event[] { event0, event1 };
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, 0, origin, region, 2, events, out event2);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, 0, 2, events, out event3);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (int version)Copy not identical to source when using event output and a wait list");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
}
{
long[] origin = new long[3] { 0, 0, 0 };
long[] region = new long[3] { imgWidth, imgHeight, 1 };
long[] dstOrigin = new long[3] { 0, 0, 0 };
long[] dstRegion = new long[3] { imgWidth, imgHeight, 1 };
Array.Clear(cmpData, 0, cmpData.Length);
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, 0L, origin, region);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, 0L);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (long version)Copy not identical to source when using no event args");
Array.Clear(cmpData, 0, cmpData.Length);
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, 0L, origin, region, 0, null, out event0);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, 0L, 0, null, out event1);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (long version)Copy not identical to source when using event output and no event args");
Array.Clear(cmpData, 0, cmpData.Length);
Event[] events = new Event[] { event0, event1 };
mem0.Write(cq, 0L, srcData, 0, bufLen);
cq.EnqueueBarrier();
cq.EnqueueCopyBufferToImage(mem0, img0, 0L, origin, region, 2, events, out event2);
cq.EnqueueBarrier();
cq.EnqueueCopyImageToBuffer(img0, mem0, origin, region, 0L, 2, events, out event3);
cq.EnqueueBarrier();
mem0.Read(cq, 0L, cmpData, 0, bufLen);
if (!CompareArray(cmpData, srcData))
Error("EnqueueCopyBufferToImage/EnqueueCopyImageToBuffer: (long version)Copy not identical to source when using event output and a wait list");
event0.Dispose();
event1.Dispose();
event2.Dispose();
event3.Dispose();
}
}
}
}
catch (Exception e)
{
Error("Exception during testing: " + e.ToString());
}
finally
{
if (img0 != null)
img0.Dispose();
if (mem0 != null)
mem0.Dispose();
}
}
public virtual void MemSet(CommandQueue cq, long dstByteOffset, double value, long count)
{
IntPtr p = cq.EnqueueMapBuffer(this, true, MapFlags.WRITE, dstByteOffset, count * sizeof(double));
double* pBlock = (double*)p.ToPointer();
for (long i = 0; i < count; i++)
pBlock[i] = value;
cq.EnqueueUnmapMemObject(this, p);
cq.Finish();
}
private void TestUserEventCallbacks(Context c, CommandQueue cq)
{
if (!(cq.Device.Platform.OpenCLMajorVersion >= 1 && cq.Device.Platform.OpenCLMinorVersion >= 1))
{
Output("Skipping TestUserEventCallbacks tests(Requires OpenCL 1.1 or higher)");
return;
}
Output( "Testing User Events and event callbacks" );
TestUserEventCallbackCalled = false;
Event event0 = c.CreateUserEvent();
event0.SetCallback(ExecutionStatus.COMPLETE, TestUserEventCallback, this);
event0.SetUserEventStatus(ExecutionStatus.COMPLETE);
if (TestUserEventCallbackCalled != true)
{
Error("Event callback not called");
return;
}
c.WaitForEvent(event0);
}
public virtual void Read(CommandQueue cq, long srcOffset, double[] dstData, int dstStartIndex, int count)
{
IntPtr p = cq.EnqueueMapBuffer(this, true, MapFlags.READ, srcOffset, count * sizeof(double));
double* pBlock = (double*)p.ToPointer();
for (long i = 0; i < count; i++)
dstData[dstStartIndex+i] = pBlock[i];
cq.EnqueueUnmapMemObject(this, p);
cq.Finish();
}
private unsafe void TestVecKernel(Context c, CommandQueue cq, Kernel k)
{
Float2 f2 = new Float2(0.0f,1.0f);
float[] memory = new float[2];
fixed (float* pMemory = memory)
{
Mem mem = c.CreateBuffer((MemFlags)((ulong)MemFlags.READ_WRITE | (ulong)MemFlags.USE_HOST_PTR), 4 * 2, pMemory);
k.SetArg(0, f2);
k.SetArg(1, mem);
cq.EnqueueTask(k);
cq.EnqueueBarrier();
IntPtr pMap = cq.EnqueueMapBuffer(mem, true, MapFlags.READ, 0, 2 * 4);
cq.EnqueueUnmapMemObject(mem, pMap);
}
}
public void CreateContext( Platform platform, Device device )
{
IntPtr[] contextProperties = new IntPtr[]
{
(IntPtr)ContextProperties.PLATFORM, platform.PlatformID,
IntPtr.Zero, IntPtr.Zero
};
Device[] devices = new Device[]
{
device
};
oclContext = platform.CreateContext(contextProperties, devices, oclContextNotify, IntPtr.Zero);
oclCQ = oclContext.CreateCommandQueue(device, CommandQueueProperties.PROFILING_ENABLE);
}
private void TestCommandQueue(Context c, CommandQueue cq )
{
string programName = "OpenCL" + Path.DirectorySeparatorChar + "src" + Path.DirectorySeparatorChar + "MemoryTests.cl";
Output("Testing compilation of: " + programName);
OpenCLNet.Program p0 = c.CreateProgramWithSource(File.ReadAllLines(programName));
OpenCLNet.Program p = c.CreateProgramWithSource(File.ReadAllText(programName));
p0.Build();
p.Build();
Kernel k = p.CreateKernel(@"LoopAndDoNothing");
TestCommandQueueMemCopy(c, cq);
TestCommandQueueAsync(c, cq, k );
}
protected virtual void Initialize(Context context, CommandQueue cq, int width, int height)
{
BitmapData bd;
Bitmap = new Bitmap(width,height,PixelFormat.Format32bppArgb);
bd = LockEntireBitmap( Bitmap, ImageLockMode.ReadOnly );
Image = context.CreateImage2D(MemFlags.COPY_HOST_PTR, OpenCLNet.ImageFormat.ARGB8U, width, height, bd.Stride, bd.Scan0);
Context = context;
CQ = cq;
}
private void TestCommandQueueAsync(Context c, CommandQueue cq, Kernel kernel )
{
List<Event> events = new List<Event>();
Event clEvent;
Output("Testing asynchronous task issuing (clEnqueueTask) and waiting for events");
// Issue a bunch of slow operations
kernel.SetArg(0, 5000000);
for (int i = 0; i < 10; i++)
{
cq.EnqueueTask(kernel, 0, null, out clEvent);
events.Add(clEvent);
}
// Issue a bunch of fast operations
kernel.SetArg(0, 500);
for (int i = 0; i < 10; i++)
{
cq.EnqueueTask(kernel, 0, null, out clEvent);
events.Add(clEvent);
}
Event[] eventList = events.ToArray();
cq.EnqueueWaitForEvents(eventList.Length, eventList);
while (events.Count > 0)
{
if ((int)events[0].ExecutionStatus < 0)
{
Output(cq.Device.Name + ": TestCommandQueueAsync failed with error code " + (ErrorCode)events[0].ExecutionStatus);
}
events[0].Dispose();
events.RemoveAt(0);
}
}
internal OpenCLBackedBitmap(Context context, CommandQueue cq, int width, int height)
{
Initialize(context, cq, width, height);
}
private static int AddNativeKernelParams( NativeKernel nk, CommandQueue cq, object o, Mem[] buffers)
{
int callbackId;
NativeKernelCallbackData callbackData = new NativeKernelCallbackData(nk, cq, o, buffers);
bool gotMutex = false;
try
{
gotMutex = NativeKernelParamsMutex.WaitOne();
do
{
callbackId = NativeKernelParamsId++;
} while (NativeKernelDispatch.ContainsKey(callbackId));
NativeKernelDispatch.Add(callbackId, callbackData);
}
finally
{
if (gotMutex)
NativeKernelParamsMutex.ReleaseMutex();
}
return callbackId;
}