func (this *buffer) EnqueueWrite(queue CommandQueue,
blocking_write cl.CL_bool,
offset cl.CL_size_t,
cb cl.CL_size_t,
ptr unsafe.Pointer,
event_wait_list []Event) (Event, error) {
var errCode cl.CL_int
var event_id cl.CL_event
numEvents := cl.CL_uint(len(event_wait_list))
var events []cl.CL_event
if numEvents > 0 {
events = make([]cl.CL_event, numEvents)
for i := cl.CL_uint(0); i < numEvents; i++ {
events[i] = event_wait_list[i].GetID()
}
}
if errCode = cl.CLEnqueueWriteBuffer(queue.GetID(),
this.memory_id,
blocking_write,
offset,
cb,
ptr,
numEvents,
events,
&event_id); errCode != cl.CL_SUCCESS {
return nil, fmt.Errorf("EnqueueWrite failure with errcode_ret %d: %s", errCode, cl.ERROR_CODES_STRINGS[-errCode])
} else {
return &event{event_id}, nil
}
}
func main() {
// This code executes on the OpenCL host
// Host data
var size cl.CL_int
var A []cl.CL_int //input array
var B []cl.CL_int //input array
var C []cl.CL_int //output array
// Elements in each array
const elements = cl.CL_size_t(2048)
// Compute the size of the data
datasize := cl.CL_size_t(unsafe.Sizeof(size)) * elements
// Allocate space for input/output data
A = make([]cl.CL_int, datasize)
B = make([]cl.CL_int, datasize)
C = make([]cl.CL_int, datasize)
// Initialize the input data
for i := cl.CL_int(0); i < cl.CL_int(elements); i++ {
A[i] = i
B[i] = i
}
// Use this to check the output of each API call
var status cl.CL_int
//-----------------------------------------------------
// STEP 1: Discover and initialize the platforms
//-----------------------------------------------------
var numPlatforms cl.CL_uint
var platforms []cl.CL_platform_id
// Use clGetPlatformIDs() to retrieve the number of
// platforms
status = cl.CLGetPlatformIDs(0, nil, &numPlatforms)
// Allocate enough space for each platform
platforms = make([]cl.CL_platform_id, numPlatforms)
// Fill in platforms with clGetPlatformIDs()
status = cl.CLGetPlatformIDs(numPlatforms, platforms, nil)
if status != cl.CL_SUCCESS {
println("CLGetPlatformIDs status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 2: Discover and initialize the devices
//-----------------------------------------------------
var numDevices cl.CL_uint
var devices []cl.CL_device_id
// Use clGetDeviceIDs() to retrieve the number of
// devices present
status = cl.CLGetDeviceIDs(platforms[0],
cl.CL_DEVICE_TYPE_ALL,
0,
nil,
&numDevices)
if status != cl.CL_SUCCESS {
println("CLGetDeviceIDs status!=cl.CL_SUCCESS")
return
}
// Allocate enough space for each device
devices = make([]cl.CL_device_id, numDevices)
// Fill in devices with clGetDeviceIDs()
status = cl.CLGetDeviceIDs(platforms[0],
cl.CL_DEVICE_TYPE_ALL,
numDevices,
devices,
nil)
if status != cl.CL_SUCCESS {
println("CLGetDeviceIDs status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 3: Create a context
//-----------------------------------------------------
var context cl.CL_context
// Create a context using clCreateContext() and
// associate it with the devices
context = cl.CLCreateContext(nil,
numDevices,
devices,
nil,
nil,
&status)
if status != cl.CL_SUCCESS {
println("CLCreateContext status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 4: Create a command queue
//-----------------------------------------------------
var cmdQueue cl.CL_command_queue
// Create a command queue using clCreateCommandQueue(),
// and associate it with the device you want to execute
// on
cmdQueue = cl.CLCreateCommandQueue(context,
devices[0],
0,
&status)
if status != cl.CL_SUCCESS {
println("CLCreateCommandQueue status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 5: Create device buffers
//-----------------------------------------------------
var bufferA cl.CL_mem // Input array on the device
var bufferB cl.CL_mem // Input array on the device
var bufferC cl.CL_mem // Output array on the device
// Use clCreateBuffer() to create a buffer object (d_A)
// that will contain the data from the host array A
bufferA = cl.CLCreateBuffer(context,
cl.CL_MEM_READ_ONLY,
datasize,
nil,
&status)
if status != cl.CL_SUCCESS {
println("CLCreateBuffer status!=cl.CL_SUCCESS")
return
}
// Use clCreateBuffer() to create a buffer object (d_B)
// that will contain the data from the host array B
bufferB = cl.CLCreateBuffer(context,
cl.CL_MEM_READ_ONLY,
datasize,
nil,
&status)
if status != cl.CL_SUCCESS {
println("CLCreateBuffer status!=cl.CL_SUCCESS")
return
}
// Use clCreateBuffer() to create a buffer object (d_C)
// with enough space to hold the output data
bufferC = cl.CLCreateBuffer(context,
cl.CL_MEM_WRITE_ONLY,
datasize,
nil,
&status)
if status != cl.CL_SUCCESS {
println("CLCreateBuffer status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 6: Write host data to device buffers
//-----------------------------------------------------
// Use clEnqueueWriteBuffer() to write input array A to
// the device buffer bufferA
status = cl.CLEnqueueWriteBuffer(cmdQueue,
bufferA,
cl.CL_FALSE,
0,
datasize,
unsafe.Pointer(&A[0]),
0,
nil,
nil)
if status != cl.CL_SUCCESS {
println("CLEnqueueWriteBuffer status!=cl.CL_SUCCESS")
return
}
// Use clEnqueueWriteBuffer() to write input array B to
// the device buffer bufferB
status = cl.CLEnqueueWriteBuffer(cmdQueue,
bufferB,
cl.CL_FALSE,
0,
datasize,
unsafe.Pointer(&B[0]),
0,
nil,
nil)
if status != cl.CL_SUCCESS {
println("CLEnqueueWriteBuffer status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 7: Create and compile the program
//-----------------------------------------------------
programSource, programeSize := utils.Load_programsource("chapter2.cl")
// Create a program using clCreateProgramWithSource()
program := cl.CLCreateProgramWithSource(context,
1,
programSource[:],
programeSize[:],
&status)
if status != cl.CL_SUCCESS {
println("CLCreateProgramWithSource status!=cl.CL_SUCCESS")
return
}
// Build (compile) the program for the devices with
// clBuildProgram()
status = cl.CLBuildProgram(program,
numDevices,
devices,
nil,
nil,
nil)
if status != cl.CL_SUCCESS {
println("CLBuildProgram status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 8: Create the kernel
//-----------------------------------------------------
var kernel cl.CL_kernel
// Use clCreateKernel() to create a kernel from the
// vector addition function (named "vecadd")
kernel = cl.CLCreateKernel(program, []byte("vecadd"), &status)
if status != cl.CL_SUCCESS {
println("CLCreateKernel status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 9: Set the kernel arguments
//-----------------------------------------------------
// Associate the input and output buffers with the
// kernel
// using clSetKernelArg()
status = cl.CLSetKernelArg(kernel,
0,
cl.CL_size_t(unsafe.Sizeof(bufferA)),
unsafe.Pointer(&bufferA))
status |= cl.CLSetKernelArg(kernel,
1,
cl.CL_size_t(unsafe.Sizeof(bufferB)),
unsafe.Pointer(&bufferB))
status |= cl.CLSetKernelArg(kernel,
2,
cl.CL_size_t(unsafe.Sizeof(bufferC)),
unsafe.Pointer(&bufferC))
if status != cl.CL_SUCCESS {
println("CLSetKernelArg status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 10: Configure the work-item structure
//-----------------------------------------------------
// Define an index space (global work size) of work
// items for
// execution. A workgroup size (local work size) is not
// required,
// but can be used.
var globalWorkSize [1]cl.CL_size_t
// There are 'elements' work-items
globalWorkSize[0] = elements
//-----------------------------------------------------
// STEP 11: Enqueue the kernel for execution
//-----------------------------------------------------
// Execute the kernel by using
// clEnqueueNDRangeKernel().
// 'globalWorkSize' is the 1D dimension of the
// work-items
status = cl.CLEnqueueNDRangeKernel(cmdQueue,
kernel,
1,
nil,
globalWorkSize[:],
nil,
0,
nil,
nil)
if status != cl.CL_SUCCESS {
println("CLEnqueueNDRangeKernel status!=cl.CL_SUCCESS")
return
}
//-----------------------------------------------------
// STEP 12: Read the output buffer back to the host
//-----------------------------------------------------
// Use clEnqueueReadBuffer() to read the OpenCL output
// buffer (bufferC)
// to the host output array (C)
cl.CLEnqueueReadBuffer(cmdQueue,
bufferC,
cl.CL_TRUE,
0,
datasize,
unsafe.Pointer(&C[0]),
0,
nil,
nil)
if status != cl.CL_SUCCESS {
println("CLEnqueueReadBuffer status!=cl.CL_SUCCESS")
return
}
// Verify the output
result := true
for i := cl.CL_int(0); i < cl.CL_int(elements); i++ {
if C[i] != i+i {
result = false
break
}
}
if result {
println("Output is correct\n")
} else {
println("Output is incorrect\n")
}
//-----------------------------------------------------
// STEP 13: Release OpenCL resources
//-----------------------------------------------------
// Free OpenCL resources
cl.CLReleaseKernel(kernel)
cl.CLReleaseProgram(program)
cl.CLReleaseCommandQueue(cmdQueue)
cl.CLReleaseMemObject(bufferA)
cl.CLReleaseMemObject(bufferB)
cl.CLReleaseMemObject(bufferC)
cl.CLReleaseContext(context)
}
func main() {
var i, j cl.CL_size_t
// Rows and columns in the input image
inputFile := "test.png"
outputFile := "output.png"
refFile := "ref.png"
// Homegrown function to read a BMP from file
inputpixels, imageWidth, imageHeight, err1 := utils.Read_image_data(inputFile)
if err1 != nil {
log.Fatal(err1)
return
} else {
fmt.Printf("width=%d, height=%d (%d)\n", imageWidth, imageHeight, inputpixels[0])
}
// Output image on the host
outputpixels := make([]uint16, imageHeight*imageWidth)
inputImage := make([]float32, imageHeight*imageWidth)
outputImage := make([]float32, imageHeight*imageWidth)
refImage := make([]float32, imageHeight*imageWidth)
for i = 0; i < imageHeight*imageWidth; i++ {
inputImage[i] = float32(inputpixels[i])
}
// 45 degree motion blur
var filter = [49]float32{0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0,
0, 0, -1, 0, 1, 0, 0,
0, 0, -2, 0, 2, 0, 0,
0, 0, -1, 0, 1, 0, 0,
0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0}
// The convolution filter is 7x7
filterWidth := cl.CL_size_t(7)
filterSize := cl.CL_size_t(filterWidth * filterWidth) // Assume a square kernel
// Set up the OpenCL environment
var status cl.CL_int
// Discovery platform
var platform [1]cl.CL_platform_id
status = cl.CLGetPlatformIDs(1, platform[:], nil)
chk(status, "clGetPlatformIDs")
// Discover device
var device [1]cl.CL_device_id
cl.CLGetDeviceIDs(platform[0], cl.CL_DEVICE_TYPE_ALL, 1, device[:], nil)
chk(status, "clGetDeviceIDs")
// Create context
//var props =[3]cl.CL_context_properties{cl.CL_CONTEXT_PLATFORM,
// (cl.CL_context_properties)(unsafe.Pointer(&platform[0])), 0};
var context cl.CL_context
context = cl.CLCreateContext(nil, 1, device[:], nil, nil, &status)
chk(status, "clCreateContext")
// Create command queue
var queue cl.CL_command_queue
queue = cl.CLCreateCommandQueue(context, device[0], 0, &status)
chk(status, "clCreateCommandQueue")
// The image format describes how the data will be stored in memory
var format cl.CL_image_format
format.Image_channel_order = cl.CL_R // single channel
format.Image_channel_data_type = cl.CL_FLOAT // float data type
var desc cl.CL_image_desc
desc.Image_type = cl.CL_MEM_OBJECT_IMAGE2D
desc.Image_width = imageWidth
desc.Image_height = imageHeight
desc.Image_depth = 0
desc.Image_array_size = 0
desc.Image_row_pitch = 0
desc.Image_slice_pitch = 0
desc.Num_mip_levels = 0
desc.Num_samples = 0
desc.Buffer = cl.CL_mem{}
// Create space for the source image on the device
d_inputImage := cl.CLCreateImage(context, cl.CL_MEM_READ_ONLY, &format, &desc,
nil, &status)
chk(status, "clCreateImage")
// Create space for the output image on the device
d_outputImage := cl.CLCreateImage(context, cl.CL_MEM_WRITE_ONLY, &format, &desc,
nil, &status)
chk(status, "clCreateImage")
// Create space for the 7x7 filter on the device
d_filter := cl.CLCreateBuffer(context, 0, filterSize*cl.CL_size_t(unsafe.Sizeof(filter[0])),
nil, &status)
chk(status, "clCreateBuffer")
// Copy the source image to the device
var origin = [3]cl.CL_size_t{0, 0, 0} // Offset within the image to copy from
var region = [3]cl.CL_size_t{cl.CL_size_t(imageWidth), cl.CL_size_t(imageHeight), 1} // Elements to per dimension
status = cl.CLEnqueueWriteImage(queue, d_inputImage, cl.CL_FALSE, origin, region,
0, 0, unsafe.Pointer(&inputImage[0]), 0, nil, nil)
chk(status, "clEnqueueWriteImage")
// Copy the 7x7 filter to the device
status = cl.CLEnqueueWriteBuffer(queue, d_filter, cl.CL_FALSE, 0,
filterSize*cl.CL_size_t(unsafe.Sizeof(filter[0])), unsafe.Pointer(&filter[0]), 0, nil, nil)
chk(status, "clEnqueueWriteBuffer")
// Create the image sampler
sampler := cl.CLCreateSampler(context, cl.CL_FALSE,
cl.CL_ADDRESS_CLAMP_TO_EDGE, cl.CL_FILTER_NEAREST, &status)
chk(status, "clCreateSampler")
// Create a program object with source and build it
program := utils.Build_program(context, device[:], "convolution.cl", nil)
kernel := cl.CLCreateKernel(*program, []byte("convolution"), &status)
chk(status, "clCreateKernel")
// Set the kernel arguments
var w, h, f cl.CL_int
w = cl.CL_int(imageWidth)
h = cl.CL_int(imageHeight)
f = cl.CL_int(filterWidth)
status = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(d_inputImage)), unsafe.Pointer(&d_inputImage))
status |= cl.CLSetKernelArg(kernel, 1, cl.CL_size_t(unsafe.Sizeof(d_outputImage)), unsafe.Pointer(&d_outputImage))
status |= cl.CLSetKernelArg(kernel, 2, cl.CL_size_t(unsafe.Sizeof(h)), unsafe.Pointer(&h))
status |= cl.CLSetKernelArg(kernel, 3, cl.CL_size_t(unsafe.Sizeof(w)), unsafe.Pointer(&w))
status |= cl.CLSetKernelArg(kernel, 4, cl.CL_size_t(unsafe.Sizeof(d_filter)), unsafe.Pointer(&d_filter))
status |= cl.CLSetKernelArg(kernel, 5, cl.CL_size_t(unsafe.Sizeof(f)), unsafe.Pointer(&f))
status |= cl.CLSetKernelArg(kernel, 6, cl.CL_size_t(unsafe.Sizeof(sampler)), unsafe.Pointer(&sampler))
chk(status, "clSetKernelArg")
// Set the work item dimensions
var globalSize = [2]cl.CL_size_t{imageWidth, imageHeight}
status = cl.CLEnqueueNDRangeKernel(queue, kernel, 2, nil, globalSize[:], nil, 0,
nil, nil)
chk(status, "clEnqueueNDRange")
// Read the image back to the host
status = cl.CLEnqueueReadImage(queue, d_outputImage, cl.CL_TRUE, origin,
region, 0, 0, unsafe.Pointer(&outputImage[0]), 0, nil, nil)
chk(status, "clEnqueueReadImage")
// Write the output image to file
for i = 0; i < imageHeight*imageWidth; i++ {
outputpixels[i] = uint16(outputImage[i])
}
utils.Write_image_data(outputFile, outputpixels, imageWidth, imageHeight)
// Compute the reference image
for i = 0; i < imageHeight; i++ {
for j = 0; j < imageWidth; j++ {
refImage[i*imageWidth+j] = 0
}
}
// Iterate over the rows of the source image
halfFilterWidth := filterWidth / 2
var sum float32
for i = 0; i < imageHeight; i++ {
// Iterate over the columns of the source image
for j = 0; j < imageWidth; j++ {
sum = 0 // Reset sum for new source pixel
// Apply the filter to the neighborhood
for k := -halfFilterWidth; k <= halfFilterWidth; k++ {
for l := -halfFilterWidth; l <= halfFilterWidth; l++ {
if i+k >= 0 && i+k < imageHeight &&
j+l >= 0 && j+l < imageWidth {
sum += inputImage[(i+k)*imageWidth+j+l] *
filter[(k+halfFilterWidth)*filterWidth+
l+halfFilterWidth]
} else {
i_k := i + k
j_l := j + l
if i+k < 0 {
i_k = 0
} else if i+k >= imageHeight {
i_k = imageHeight - 1
}
if j+l < 0 {
j_l = 0
} else if j+l >= imageWidth {
j_l = imageWidth - 1
}
sum += inputImage[(i_k)*imageWidth+j_l] *
filter[(k+halfFilterWidth)*filterWidth+
l+halfFilterWidth]
}
}
}
refImage[i*imageWidth+j] = sum
}
}
// Write the ref image to file
for i = 0; i < imageHeight*imageWidth; i++ {
outputpixels[i] = uint16(refImage[i])
}
utils.Write_image_data(refFile, outputpixels, imageWidth, imageHeight)
failed := 0
for i = 0; i < imageHeight; i++ {
for j = 0; j < imageWidth; j++ {
if math.Abs(float64(outputImage[i*imageWidth+j]-refImage[i*imageWidth+j])) > 0.01 {
//fmt.Printf("Results are INCORRECT\n");
//fmt.Printf("Pixel mismatch at <%d,%d> (%f vs. %f) %f\n", i, j,
// outputImage[i*imageWidth+j], refImage[i*imageWidth+j], inputImage[i*imageWidth+j]);
failed++
}
}
}
fmt.Printf("Mismatch Pixel number/Total pixel number = %d/%d\n", failed, imageWidth*imageHeight)
// Free OpenCL resources
cl.CLReleaseKernel(kernel)
cl.CLReleaseProgram(*program)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseMemObject(d_inputImage)
cl.CLReleaseMemObject(d_outputImage)
cl.CLReleaseMemObject(d_filter)
cl.CLReleaseSampler(sampler)
cl.CLReleaseContext(context)
}
func main() {
/* OpenCL data structures */
var device []cl.CL_device_id
var context cl.CL_context
var queue cl.CL_command_queue
var program *cl.CL_program
var kernel cl.CL_kernel
var err cl.CL_int
/* Data and buffers */
var full_matrix, zero_matrix [80]float32
var sizeoffloat32 = cl.CL_size_t(unsafe.Sizeof(full_matrix[0]))
var buffer_origin = [3]cl.CL_size_t{5 * sizeoffloat32, 3, 0}
var host_origin = [3]cl.CL_size_t{1 * sizeoffloat32, 1, 0}
var region = [3]cl.CL_size_t{4 * sizeoffloat32, 4, 1}
var matrix_buffer cl.CL_mem
/* Initialize data */
for i := 0; i < 80; i++ {
full_matrix[i] = float32(i) * 1.0
zero_matrix[i] = 0.0
}
/* Create a device and context */
device = utils.Create_device()
context = cl.CLCreateContext(nil, 1, device[:], nil, nil, &err)
if err < 0 {
println("Couldn't create a context")
return
}
/* Build the program and create the kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, nil)
if program == nil {
println("Couldn't build program")
return
}
kernel = cl.CLCreateKernel(*program, []byte(KERNEL_FUNC), &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create a buffer to hold 80 floats */
matrix_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_READ_WRITE|
cl.CL_MEM_COPY_HOST_PTR, cl.CL_size_t(unsafe.Sizeof(full_matrix)), unsafe.Pointer(&full_matrix[0]), &err)
if err < 0 {
println("Couldn't create a buffer object")
return
}
/* Set buffer as argument to the kernel */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(matrix_buffer)), unsafe.Pointer(&matrix_buffer))
if err < 0 {
println("Couldn't set the buffer as the kernel argument")
return
}
/* Create a command queue */
queue = cl.CLCreateCommandQueue(context, device[0], 0, &err)
if err < 0 {
println("Couldn't create a command queue")
return
}
/* Enqueue kernel */
err = cl.CLEnqueueTask(queue, kernel, 0, nil, nil)
if err < 0 {
println("Couldn't enqueue the kernel")
return
}
/* Enqueue command to write to buffer */
err = cl.CLEnqueueWriteBuffer(queue, matrix_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(full_matrix)), unsafe.Pointer(&full_matrix[0]), 0, nil, nil)
if err < 0 {
println("Couldn't write to the buffer object")
return
}
/* Enqueue command to read rectangle of data */
err = cl.CLEnqueueReadBufferRect(queue, matrix_buffer, cl.CL_TRUE,
buffer_origin, host_origin, region, 10*sizeoffloat32, 0,
10*sizeoffloat32, 0, unsafe.Pointer(&zero_matrix[0]), 0, nil, nil)
if err < 0 {
println("Couldn't read the rectangle from the buffer object")
return
}
/* Display updated buffer */
for i := 0; i < 8; i++ {
for j := 0; j < 10; j++ {
fmt.Printf("%6.1f", zero_matrix[j+i*10])
}
fmt.Printf("\n")
}
/* Deallocate resources */
cl.CLReleaseMemObject(matrix_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
cl.CLReleaseContext(context)
}
