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();
}
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, 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();
}
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();
}
}
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");
}
}
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");
}
}
}
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);
}
}
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;
}
}
