func (this *buffer) EnqueueRead(queue CommandQueue,
blocking_read 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.CLEnqueueReadBuffer(queue.GetID(),
this.memory_id,
blocking_read,
offset,
cb,
ptr,
numEvents,
events,
&event_id); errCode != cl.CL_SUCCESS {
return nil, fmt.Errorf("EnqueueRead failure with errcode_ret %d: %s", errCode, cl.ERROR_CODES_STRINGS[-errCode])
} else {
return &event{event_id}, nil
}
}
func main() {
// 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)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLGetPlatformIDs")
//-----------------------------------------------------
// STEP 2: Discover and initialize the GPU 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_GPU,
0,
nil,
&numDevices)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLGetDeviceIDs")
// 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_GPU,
numDevices,
devices,
nil)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLGetDeviceIDs")
//-----------------------------------------------------
// 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)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateContext")
defer cl.CLReleaseContext(context)
//-----------------------------------------------------
// STEP 4: Create a command queue
//-----------------------------------------------------
var commandQueue [MAX_COMMAND_QUEUE]cl.CL_command_queue
// Create a command queue using clCreateCommandQueueWithProperties(),
// and associate it with the device you want to execute
for i := 0; i < MAX_COMMAND_QUEUE; i++ {
commandQueue[i] = cl.CLCreateCommandQueueWithProperties(context,
devices[0],
nil,
&status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateCommandQueueWithProperties")
defer cl.CLReleaseCommandQueue(commandQueue[i])
}
//-----------------------------------------------------
// STEP 5: Create device buffers
//-----------------------------------------------------
producerGroupSize := cl.CL_size_t(PRODUCER_GROUP_SIZE)
producerGlobalSize := cl.CL_size_t(PRODUCER_GLOBAL_SIZE)
consumerGroupSize := cl.CL_size_t(CONSUMER_GROUP_SIZE)
consumerGlobalSize := cl.CL_size_t(CONSUMER_GLOBAL_SIZE)
var samplePipePkt [2]cl.CL_float
szPipe := cl.CL_uint(PIPE_SIZE)
szPipePkt := cl.CL_uint(unsafe.Sizeof(samplePipePkt))
if szPipe%PRNG_CHANNELS != 0 {
szPipe = (szPipe/PRNG_CHANNELS)*PRNG_CHANNELS + PRNG_CHANNELS
}
consumerGlobalSize = cl.CL_size_t(szPipe)
pipePktPerThread := cl.CL_int(szPipe) / PRNG_CHANNELS
seed := cl.CL_int(SEED)
rngType := cl.CL_int(RV_GAUSSIAN)
var histMin cl.CL_float
var histMax cl.CL_float
if rngType == cl.CL_int(RV_UNIFORM) {
histMin = 0.0
histMax = 1.0
} else {
histMin = -10.0
histMax = 10.0
}
localDevHist := make([]cl.CL_int, MAX_HIST_BINS)
cpuHist := make([]cl.CL_int, MAX_HIST_BINS)
//Create and initialize memory objects
rngPipe := cl.CLCreatePipe(context,
cl.CL_MEM_READ_WRITE,
szPipePkt,
szPipe,
nil,
&status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clCreatePipe")
devHist := cl.CLCreateBuffer(context,
cl.CL_MEM_READ_WRITE|cl.CL_MEM_COPY_HOST_PTR,
MAX_HIST_BINS*cl.CL_size_t(unsafe.Sizeof(localDevHist[0])),
unsafe.Pointer(&localDevHist[0]),
&status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clCreateBuffer")
//-----------------------------------------------------
// STEP 6: Create and compile the program
//-----------------------------------------------------
programSource, programeSize := utils.Load_programsource("pipe.cl")
// Create a program using clCreateProgramWithSource()
program := cl.CLCreateProgramWithSource(context,
1,
programSource[:],
programeSize[:],
&status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateProgramWithSource")
defer cl.CLReleaseProgram(program)
// Build (compile) the program for the devices with
// clBuildProgram()
options := "-cl-std=CL2.0"
status = cl.CLBuildProgram(program,
numDevices,
devices,
[]byte(options),
nil,
nil)
if status != cl.CL_SUCCESS {
var program_log interface{}
var log_size cl.CL_size_t
/* Find size of log and print to std output */
cl.CLGetProgramBuildInfo(program, devices[0], cl.CL_PROGRAM_BUILD_LOG,
0, nil, &log_size)
cl.CLGetProgramBuildInfo(program, devices[0], cl.CL_PROGRAM_BUILD_LOG,
log_size, &program_log, nil)
fmt.Printf("%s\n", program_log)
return
}
//utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLBuildProgram")
//-----------------------------------------------------
// STEP 7: Create the kernel
//-----------------------------------------------------
// Use clCreateKernel() to create a kernel
produceKernel := cl.CLCreateKernel(program, []byte("pipe_producer"), &status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateKernel")
defer cl.CLReleaseKernel(produceKernel)
consumeKernel := cl.CLCreateKernel(program, []byte("pipe_consumer"), &status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateKernel")
defer cl.CLReleaseKernel(consumeKernel)
//-----------------------------------------------------
// STEP 8: Set the kernel arguments
//-----------------------------------------------------
// Associate the input and output buffers with the
// kernel
// using clSetKernelArg()
// Set appropriate arguments to the kernel
status = cl.CLSetKernelArg(produceKernel,
0,
cl.CL_size_t(unsafe.Sizeof(rngPipe)),
unsafe.Pointer(&rngPipe))
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArg(rngPipe)")
status = cl.CLSetKernelArg(produceKernel,
1,
cl.CL_size_t(unsafe.Sizeof(pipePktPerThread)),
unsafe.Pointer(&pipePktPerThread))
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArg(pipePktPerThread)")
status = cl.CLSetKernelArg(produceKernel,
2,
cl.CL_size_t(unsafe.Sizeof(seed)),
unsafe.Pointer(&seed))
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArg(seed)")
status = cl.CLSetKernelArg(produceKernel,
3,
cl.CL_size_t(unsafe.Sizeof(rngType)),
unsafe.Pointer(&rngType))
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArg(rngType)")
//-----------------------------------------------------
// STEP 9: 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.
// Enqueue both the kernels.
var globalThreads = []cl.CL_size_t{producerGlobalSize}
var localThreads = []cl.CL_size_t{producerGroupSize}
//-----------------------------------------------------
// STEP 10: Enqueue the kernel for execution
//-----------------------------------------------------
// Execute the kernel by using
// clEnqueueNDRangeKernel().
// 'globalWorkSize' is the 1D dimension of the
// work-items
var produceEvt [1]cl.CL_event
status = cl.CLEnqueueNDRangeKernel(commandQueue[0],
produceKernel,
1,
nil,
globalThreads,
localThreads,
0,
nil,
&produceEvt[0])
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clEnqueueNDRangeKernel")
/*
launch consumer kernel only after producer has finished.
This is done to avoid concurrent kernels execution as the
memory consistency of pipe is guaranteed only across
synchronization points.
*/
status = cl.CLWaitForEvents(1, produceEvt[:])
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clWaitForEvents(produceEvt)")
//-----------------------------------------------------
// STEP 8: Set the kernel arguments
//-----------------------------------------------------
// Associate the input and output buffers with the
// kernel
// using clSetKernelArg()
// Set appropriate arguments to the kernel
status = cl.CLSetKernelArg(consumeKernel,
0,
cl.CL_size_t(unsafe.Sizeof(rngPipe)),
unsafe.Pointer(&rngPipe))
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArg(rngPipe)")
status = cl.CLSetKernelArg(consumeKernel,
1,
cl.CL_size_t(unsafe.Sizeof(devHist)),
unsafe.Pointer(&devHist))
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArg(devHist)")
status = cl.CLSetKernelArg(consumeKernel,
2,
cl.CL_size_t(unsafe.Sizeof(histMin)),
unsafe.Pointer(&histMin))
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArg(histMin)")
status = cl.CLSetKernelArg(consumeKernel,
3,
cl.CL_size_t(unsafe.Sizeof(histMax)),
unsafe.Pointer(&histMax))
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArg(histMax)")
//-----------------------------------------------------
// STEP 9: 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.
globalThreads[0] = consumerGlobalSize
localThreads[0] = consumerGroupSize
//-----------------------------------------------------
// STEP 10: Enqueue the kernel for execution
//-----------------------------------------------------
// Execute the kernel by using
// clEnqueueNDRangeKernel().
// 'globalWorkSize' is the 1D dimension of the
// work-items
var consumeEvt [1]cl.CL_event
status = cl.CLEnqueueNDRangeKernel(
commandQueue[1],
consumeKernel,
1,
nil,
globalThreads,
localThreads,
0,
nil,
&consumeEvt[0])
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clEnqueueNDRangeKernel")
status = cl.CLFlush(commandQueue[0])
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clFlush(0)")
status = cl.CLFlush(commandQueue[1])
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clFlush(1)")
//wait for kernels to finish
status = cl.CLFinish(commandQueue[0])
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clFinish(0)")
status = cl.CLFinish(commandQueue[1])
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clFinish(1)")
//-----------------------------------------------------
// STEP 11: Read the output buffer back to the host
//-----------------------------------------------------
// Use clEnqueueReadBuffer() to read the OpenCL output
// buffer (bufferC)
// to the host output array (C)
//copy the data back to host buffer
var readEvt cl.CL_event
status = cl.CLEnqueueReadBuffer(commandQueue[1],
devHist,
cl.CL_TRUE,
0,
(MAX_HIST_BINS)*cl.CL_size_t(unsafe.Sizeof(localDevHist[0])),
unsafe.Pointer(&localDevHist[0]),
0,
nil,
&readEvt)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clEnqueueReadBuffer")
//-----------------------------------------------------
// STEP 12: Verify the results
//-----------------------------------------------------
//Find the tolerance limit
fTol := (float32)(CONSUMER_GLOBAL_SIZE) * (float32)(COMP_TOL) / (float32)(100.0)
iTol := (int)(fTol)
if iTol == 0 {
iTol = 1
}
//CPU side histogram computation
CPUReference(seed, pipePktPerThread, rngType, cpuHist, histMax, histMin)
//Compare
for bin := 0; bin < MAX_HIST_BINS; bin++ {
diff := int(localDevHist[bin] - cpuHist[bin])
if diff < 0 {
diff = -diff
}
if diff > iTol {
println("Failed!")
return
}
}
println("Passed!")
}
func TestMatvec(t *testing.T) {
/* Host/device data structures */
var platform [1]cl.CL_platform_id
var device [1]cl.CL_device_id
var context cl.CL_context
var queue cl.CL_command_queue
var i, err cl.CL_int
/* Program/kernel data structures */
var program cl.CL_program
var program_buffer [1][]byte
var program_log interface{}
var program_size [1]cl.CL_size_t
var log_size cl.CL_size_t
var kernel cl.CL_kernel
/* Data and buffers */
var mat [16]float32
var vec, result [4]float32
var correct = [4]float32{0.0, 0.0, 0.0, 0.0}
var mat_buff, vec_buff, res_buff cl.CL_mem
/* Initialize data to be processed by the kernel */
for i = 0; i < 16; i++ {
mat[i] = float32(i) * 2.0
}
for i = 0; i < 4; i++ {
vec[i] = float32(i) * 3.0
correct[0] += mat[i] * vec[i]
correct[1] += mat[i+4] * vec[i]
correct[2] += mat[i+8] * vec[i]
correct[3] += mat[i+12] * vec[i]
}
/* Identify a platform */
err = cl.CLGetPlatformIDs(1, platform[:], nil)
if err < 0 {
t.Errorf("Couldn't find any platforms")
}
/* Access a device */
err = cl.CLGetDeviceIDs(platform[0], cl.CL_DEVICE_TYPE_GPU, 1, device[:], nil)
if err < 0 {
err = cl.CLGetDeviceIDs(platform[0], cl.CL_DEVICE_TYPE_CPU, 1, device[:], nil)
if err < 0 {
t.Errorf("Couldn't find any devices")
}
}
/* Create the context */
context = cl.CLCreateContext(nil, 1, device[:], nil, nil, &err)
if err < 0 {
t.Errorf("Couldn't create a context")
}
/* Read program file and place content into buffer */
program_handle, err1 := os.Open("matvec.cl")
if err1 != nil {
t.Errorf("Couldn't find the program file")
}
defer program_handle.Close()
fi, err2 := program_handle.Stat()
if err2 != nil {
t.Errorf("Couldn't find the program stat")
}
program_size[0] = cl.CL_size_t(fi.Size())
program_buffer[0] = make([]byte, program_size[0])
read_size, err3 := program_handle.Read(program_buffer[0])
if err3 != nil || cl.CL_size_t(read_size) != program_size[0] {
t.Errorf("read file error or file size wrong")
}
/* Create a program containing all program content */
program = cl.CLCreateProgramWithSource(context, 1,
program_buffer[:], program_size[:], &err)
if err < 0 {
t.Errorf("Couldn't create the program")
}
/* Build program */
err = cl.CLBuildProgram(program, 1, device[:], nil, nil, nil)
if err < 0 {
/* Find size of log and print to std output */
cl.CLGetProgramBuildInfo(program, device[0], cl.CL_PROGRAM_BUILD_LOG,
0, nil, &log_size)
//program_log = (char*) malloc(log_size+1);
//program_log[log_size] = '\0';
cl.CLGetProgramBuildInfo(program, device[0], cl.CL_PROGRAM_BUILD_LOG,
log_size, &program_log, nil)
t.Errorf("%s\n", program_log)
//free(program_log);
}
/* Create kernel for the mat_vec_mult function */
kernel = cl.CLCreateKernel(program, []byte("matvec_mult"), &err)
if err < 0 {
t.Errorf("Couldn't create the kernel")
return
}
/* Create CL buffers to hold input and output data */
mat_buff = cl.CLCreateBuffer(context, cl.CL_MEM_READ_ONLY|
cl.CL_MEM_COPY_HOST_PTR, cl.CL_size_t(unsafe.Sizeof(mat)), unsafe.Pointer(&mat[0]), &err)
if err < 0 {
t.Errorf("Couldn't create a buffer object")
return
}
vec_buff = cl.CLCreateBuffer(context, cl.CL_MEM_READ_ONLY|
cl.CL_MEM_COPY_HOST_PTR, cl.CL_size_t(unsafe.Sizeof(vec)), unsafe.Pointer(&vec[0]), nil)
res_buff = cl.CLCreateBuffer(context, cl.CL_MEM_WRITE_ONLY,
cl.CL_size_t(unsafe.Sizeof(result)), nil, nil)
/* Create kernel arguments from the CL buffers */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(mat_buff)), unsafe.Pointer(&mat_buff))
if err < 0 {
t.Errorf("Couldn't set the kernel argument")
return
}
cl.CLSetKernelArg(kernel, 1, cl.CL_size_t(unsafe.Sizeof(vec_buff)), unsafe.Pointer(&vec_buff))
cl.CLSetKernelArg(kernel, 2, cl.CL_size_t(unsafe.Sizeof(res_buff)), unsafe.Pointer(&res_buff))
/* Create a CL command queue for the device*/
queue = cl.CLCreateCommandQueue(context, device[0], 0, &err)
if err < 0 {
t.Errorf("Couldn't create the command queue, errcode=%d\n", err)
return
}
/* Enqueue the command queue to the device */
var work_units_per_kernel = [1]cl.CL_size_t{4} /* 4 work-units per kernel */
err = cl.CLEnqueueNDRangeKernel(queue, kernel, 1, nil, work_units_per_kernel[:],
nil, 0, nil, nil)
if err < 0 {
t.Errorf("Couldn't enqueue the kernel execution command, errcode=%d\n", err)
return
}
/* Read the result */
err = cl.CLEnqueueReadBuffer(queue, res_buff, cl.CL_TRUE, 0, cl.CL_size_t(unsafe.Sizeof(result)),
unsafe.Pointer(&result[0]), 0, nil, nil)
if err < 0 {
t.Errorf("Couldn't enqueue the read buffer command")
return
}
/* Test the result */
if (result[0] == correct[0]) && (result[1] == correct[1]) &&
(result[2] == correct[2]) && (result[3] == correct[3]) {
t.Logf("Matrix-vector multiplication successful.")
} else {
t.Errorf("Matrix-vector multiplication unsuccessful.")
}
/* Deallocate resources */
cl.CLReleaseMemObject(mat_buff)
cl.CLReleaseMemObject(vec_buff)
cl.CLReleaseMemObject(res_buff)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(program)
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 select1 [4]float32
var select2 [2]cl.CL_uchar
var select1_buffer, select2_buffer cl.CL_mem
/* Create a context */
device = utils.Create_device()
context = cl.CLCreateContext(nil, 1, device[:], nil, nil, &err)
if err < 0 {
println("Couldn't create a context")
return
}
/* Create a kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, nil)
kernel = cl.CLCreateKernel(*program, KERNEL_FUNC, &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create a write-only buffer to hold the output data */
select1_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_WRITE_ONLY,
cl.CL_size_t(unsafe.Sizeof(select1)), nil, &err)
if err < 0 {
println("Couldn't create a buffer")
return
}
select2_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_WRITE_ONLY,
cl.CL_size_t(unsafe.Sizeof(select2)), nil, &err)
/* Create kernel argument */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(select1_buffer)), unsafe.Pointer(&select1_buffer))
if err < 0 {
println("Couldn't set a kernel argument")
return
}
cl.CLSetKernelArg(kernel, 1, cl.CL_size_t(unsafe.Sizeof(select2_buffer)), unsafe.Pointer(&select2_buffer))
/* 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
}
/* Read and print the result */
err = cl.CLEnqueueReadBuffer(queue, select1_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(select1)), unsafe.Pointer(&select1), 0, nil, nil)
if err < 0 {
println("Couldn't read the buffer")
return
}
cl.CLEnqueueReadBuffer(queue, select2_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(select2)), unsafe.Pointer(&select2), 0, nil, nil)
fmt.Printf("select: ")
for i := 0; i < 3; i++ {
fmt.Printf("%.2f, ", select1[i])
}
fmt.Printf("%.2f\n", select1[3])
fmt.Printf("bitselect: %X, %X\n", select2[0], select2[1])
/* Deallocate resources */
cl.CLReleaseMemObject(select1_buffer)
cl.CLReleaseMemObject(select2_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
cl.CLReleaseContext(context)
}
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() {
/* Host/device 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 test [16]byte
var test_buffer cl.CL_mem
/* Create a 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 a kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, nil)
kernel = cl.CLCreateKernel(*program, KERNEL_FUNC, &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create a write-only buffer to hold the output data */
test_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_WRITE_ONLY,
cl.CL_size_t(unsafe.Sizeof(test)), nil, &err)
if err < 0 {
println("Couldn't create a buffer")
return
}
/* Create kernel argument */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(test_buffer)), unsafe.Pointer(&test_buffer))
if err < 0 {
println("Couldn't set a 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
}
/* Read and print the result */
err = cl.CLEnqueueReadBuffer(queue, test_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(test)), unsafe.Pointer(&test), 0, nil, nil)
if err < 0 {
println("Couldn't read the buffer")
return
}
for i := 0; i < 15; i++ {
fmt.Printf("0x%X, ", test[i])
}
fmt.Printf("0x%X\n", test[15])
/* Deallocate resources */
cl.CLReleaseMemObject(test_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
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 */
dim := cl.CL_uint(2)
var global_offset = [2]cl.CL_size_t{3, 5}
var global_size = [2]cl.CL_size_t{6, 4}
var local_size = [2]cl.CL_size_t{3, 2}
var test [24]float32
var test_buffer cl.CL_mem
/* 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 a kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, nil)
kernel = cl.CLCreateKernel(*program, KERNEL_FUNC, &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create a write-only buffer to hold the output data */
test_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_WRITE_ONLY,
cl.CL_size_t(unsafe.Sizeof(test)), nil, &err)
if err < 0 {
println("Couldn't create a buffer")
return
}
/* Create kernel argument */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(test_buffer)), unsafe.Pointer(&test_buffer))
if err < 0 {
println("Couldn't set a 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.CLEnqueueNDRangeKernel(queue, kernel, dim, global_offset[:],
global_size[:], local_size[:], 0, nil, nil)
if err < 0 {
println("Couldn't enqueue the kernel")
return
}
/* Read and print the result */
err = cl.CLEnqueueReadBuffer(queue, test_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(test)), unsafe.Pointer(&test), 0, nil, nil)
if err < 0 {
println("Couldn't read the buffer")
return
}
for i := 0; i < 24; i += 6 {
fmt.Printf("%.2f %.2f %.2f %.2f %.2f %.2f\n",
test[i], test[i+1], test[i+2], test[i+3], test[i+4], test[i+5])
}
/* Deallocate resources */
cl.CLReleaseMemObject(test_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
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 events */
var data []float32
var data_buffer cl.CL_mem
var user_event, kernel_event, read_event [1]cl.CL_event
/* Initialize data */
data = make([]float32, 4)
for i := 0; i < 4; i++ {
data[i] = float32(i) * 1.0
}
/* 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 a kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, nil)
kernel = cl.CLCreateKernel(*program, KERNEL_FUNC, &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create a buffer to hold data */
data_buffer = cl.CLCreateBuffer(context,
cl.CL_MEM_READ_WRITE|cl.CL_MEM_COPY_HOST_PTR,
cl.CL_size_t(unsafe.Sizeof(data[0]))*4, unsafe.Pointer(&data[0]), &err)
if err < 0 {
println("Couldn't create a buffer")
return
}
/* Create kernel argument */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(data_buffer)), unsafe.Pointer(&data_buffer))
if err < 0 {
println("Couldn't set a kernel argument")
return
}
/* Create a command queue */
queue = cl.CLCreateCommandQueue(context, device[0],
cl.CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE, &err)
if err < 0 {
println("Couldn't create a command queue")
return
}
/* Configure events */
user_event[0] = cl.CLCreateUserEvent(context, &err)
if err < 0 {
println("Couldn't enqueue the kernel")
return
}
/* Enqueue kernel */
err = cl.CLEnqueueTask(queue, kernel, 1, user_event[:], &kernel_event[0])
if err < 0 {
println("Couldn't enqueue the kernel")
return
}
/* Read the buffer */
err = cl.CLEnqueueReadBuffer(queue, data_buffer, cl.CL_FALSE, 0,
cl.CL_size_t(unsafe.Sizeof(data[0]))*4, unsafe.Pointer(&data[0]), 1, kernel_event[:], &read_event[0])
if err < 0 {
println("Couldn't read the buffer")
return
}
/* Set callback for event */
err = cl.CLSetEventCallback(read_event[0], cl.CL_COMPLETE,
read_complete, unsafe.Pointer(&data))
if err < 0 {
println("Couldn't set callback for event")
return
}
/* Sleep for a second to demonstrate the that commands haven't
started executing. Then prompt user */
time.Sleep(1)
fmt.Printf("Old data: %4.2f, %4.2f, %4.2f, %4.2f\n",
data[0], data[1], data[2], data[3])
fmt.Printf("Press ENTER to continue.\n")
//getchar();
reader := bufio.NewReader(os.Stdin)
reader.ReadString('\n')
/* Set user event to success */
cl.CLSetUserEventStatus(user_event[0], cl.CL_SUCCESS)
/* Deallocate resources */
cl.CLReleaseEvent(read_event[0])
cl.CLReleaseEvent(kernel_event[0])
cl.CLReleaseEvent(user_event[0])
cl.CLReleaseMemObject(data_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
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
var offset, global_size, local_size [1]cl.CL_size_t
/* Data and events */
var data [2]cl.CL_int
var data_buffer cl.CL_mem
/* 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 a kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, nil)
kernel = cl.CLCreateKernel(*program, KERNEL_FUNC, &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create a buffer to hold data */
data_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_WRITE_ONLY,
cl.CL_size_t(unsafe.Sizeof(data[0]))*2, nil, &err)
if err < 0 {
println("Couldn't create a buffer")
return
}
/* Create kernel argument */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(data_buffer)), unsafe.Pointer(&data_buffer))
if err < 0 {
println("Couldn't set a 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 */
offset[0] = 0
global_size[0] = 8
local_size[0] = 4
err = cl.CLEnqueueNDRangeKernel(queue, kernel, 1, offset[:], global_size[:], local_size[:], 0, nil, nil)
if err < 0 {
println("Couldn't enqueue the kernel")
return
}
/* Read the buffer */
err = cl.CLEnqueueReadBuffer(queue, data_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(data[0]))*2, unsafe.Pointer(&data[0]), 0, nil, nil)
if err < 0 {
println("Couldn't read the buffer")
return
}
fmt.Printf("Increment: %d\n", data[0])
fmt.Printf("Atomic increment: %d\n", data[1])
/* Deallocate resources */
cl.CLReleaseMemObject(data_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
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 a float32 = 6.0
var b float32 = 2.0
var result float32
var a_buffer, b_buffer, output_buffer cl.CL_mem
/* Extension data */
var sizeofuint cl.CL_uint
var addr_data interface{}
var ext_data interface{}
fp64_ext := "cl_khr_fp64"
var ext_size cl.CL_size_t
var options []byte
/* 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
}
/* Obtain the device data */
if cl.CLGetDeviceInfo(device[0], cl.CL_DEVICE_ADDRESS_BITS,
cl.CL_size_t(unsafe.Sizeof(sizeofuint)), &addr_data, nil) < 0 {
println("Couldn't read extension data")
return
}
fmt.Printf("Address width: %v\n", addr_data.(cl.CL_uint))
/* Define "FP_64" option if doubles are supported */
cl.CLGetDeviceInfo(device[0], cl.CL_DEVICE_EXTENSIONS,
0, nil, &ext_size)
// ext_data = (char*)malloc(ext_size + 1);
// ext_data[ext_size] = '\0';
cl.CLGetDeviceInfo(device[0], cl.CL_DEVICE_EXTENSIONS,
ext_size, &ext_data, nil)
if strings.Contains(ext_data.(string), fp64_ext) {
fmt.Printf("The %s extension is supported.\n", fp64_ext)
options = []byte("-DFP_64 ")
} else {
fmt.Printf("The %s extension is not supported. %s\n", fp64_ext, ext_data.(string))
}
/* Build the program and create the kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, options)
kernel = cl.CLCreateKernel(*program, KERNEL_FUNC, &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create CL buffers to hold input and output data */
a_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_READ_ONLY|
cl.CL_MEM_COPY_HOST_PTR, cl.CL_size_t(unsafe.Sizeof(a)), unsafe.Pointer(&a), &err)
if err < 0 {
println("Couldn't create a memory object")
return
}
b_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_READ_ONLY|
cl.CL_MEM_COPY_HOST_PTR, cl.CL_size_t(unsafe.Sizeof(b)), unsafe.Pointer(&b), nil)
output_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_WRITE_ONLY,
cl.CL_size_t(unsafe.Sizeof(b)), nil, nil)
/* Create kernel arguments */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(a_buffer)), unsafe.Pointer(&a_buffer))
if err < 0 {
println("Couldn't set a kernel argument")
return
}
cl.CLSetKernelArg(kernel, 1, cl.CL_size_t(unsafe.Sizeof(b_buffer)), unsafe.Pointer(&b_buffer))
cl.CLSetKernelArg(kernel, 2, cl.CL_size_t(unsafe.Sizeof(output_buffer)), unsafe.Pointer(&output_buffer))
/* 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
}
/* Read and print the result */
err = cl.CLEnqueueReadBuffer(queue, output_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(result)), unsafe.Pointer(&result), 0, nil, nil)
if err < 0 {
println("Couldn't read the output buffer")
return
}
fmt.Printf("The kernel result is %f\n", result)
/* Deallocate resources */
cl.CLReleaseMemObject(a_buffer)
cl.CLReleaseMemObject(b_buffer)
cl.CLReleaseMemObject(output_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
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 events */
var num_vectors cl.CL_int
var data [NUM_BYTES]byte
var data_buffer cl.CL_mem
var prof_event cl.CL_event
var total_time cl.CL_ulong
var time_start, time_end interface{}
var mapped_memory unsafe.Pointer
/* 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 a kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, nil)
kernel = cl.CLCreateKernel(*program, KERNEL_FUNC, &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create a buffer to hold data */
data_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_WRITE_ONLY,
cl.CL_size_t(unsafe.Sizeof(data[0]))*NUM_BYTES, nil, &err)
if err < 0 {
println("Couldn't create a buffer")
return
}
/* Create kernel argument */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(data_buffer)), unsafe.Pointer(&data_buffer))
if err < 0 {
println("Couldn't set a kernel argument")
return
}
/* Tell kernel number of char16 vectors */
num_vectors = NUM_BYTES / 16
cl.CLSetKernelArg(kernel, 1, cl.CL_size_t(unsafe.Sizeof(num_vectors)), unsafe.Pointer(&num_vectors))
/* Create a command queue */
queue = cl.CLCreateCommandQueue(context, device[0],
cl.CL_QUEUE_PROFILING_ENABLE, &err)
if err < 0 {
println("Couldn't create a command queue")
return
}
total_time = 0.0
for i := 0; i < NUM_ITERATIONS; i++ {
/* Enqueue kernel */
err = cl.CLEnqueueTask(queue, kernel, 0, nil, nil)
if err < 0 {
println("Couldn't enqueue the kernel")
return
}
if PROFILE_READ == 1 {
/* Read the buffer */
err = cl.CLEnqueueReadBuffer(queue, data_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(data[0]))*NUM_BYTES, unsafe.Pointer(&data[0]), 0, nil, &prof_event)
if err < 0 {
println("Couldn't read the buffer")
return
}
} else {
/* Create memory map */
mapped_memory = cl.CLEnqueueMapBuffer(queue, data_buffer, cl.CL_TRUE,
cl.CL_MAP_READ, 0, cl.CL_size_t(unsafe.Sizeof(data[0]))*NUM_BYTES, 0, nil, &prof_event, &err)
if err < 0 {
println("Couldn't map the buffer to host memory")
return
}
}
/* Get profiling information */
cl.CLGetEventProfilingInfo(prof_event, cl.CL_PROFILING_COMMAND_START,
cl.CL_size_t(unsafe.Sizeof(total_time)), &time_start, nil)
cl.CLGetEventProfilingInfo(prof_event, cl.CL_PROFILING_COMMAND_END,
cl.CL_size_t(unsafe.Sizeof(total_time)), &time_end, nil)
total_time += time_end.(cl.CL_ulong) - time_start.(cl.CL_ulong)
if PROFILE_READ == 0 {
/* Unmap the buffer */
err = cl.CLEnqueueUnmapMemObject(queue, data_buffer, mapped_memory,
0, nil, nil)
if err < 0 {
println("Couldn't unmap the buffer")
return
}
}
}
if PROFILE_READ == 1 {
fmt.Printf("Average read time: %v\n", total_time/NUM_ITERATIONS)
} else {
fmt.Printf("Average map time: %v\n", total_time/NUM_ITERATIONS)
}
/* Deallocate resources */
cl.CLReleaseEvent(prof_event)
cl.CLReleaseMemObject(data_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
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 r_coords = [4]float32{2, 1, 3, 4}
var angles = [4]float32{3 * M_PI / 8, 3 * M_PI / 4, 4 * M_PI / 3, 11 * M_PI / 6}
var x_coords, y_coords [4]float32
var r_coords_buffer, angles_buffer,
x_coords_buffer, y_coords_buffer cl.CL_mem
/* 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
}
/* Create a kernel */
program = utils.Build_program(context, device[:], PROGRAM_FILE, nil)
kernel = cl.CLCreateKernel(*program, KERNEL_FUNC, &err)
if err < 0 {
println("Couldn't create a kernel")
return
}
/* Create a write-only buffer to hold the output data */
r_coords_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_READ_ONLY|cl.CL_MEM_COPY_HOST_PTR,
cl.CL_size_t(unsafe.Sizeof(r_coords)), unsafe.Pointer(&r_coords[0]), &err)
if err < 0 {
println("Couldn't create a buffer")
return
}
angles_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_READ_ONLY|cl.CL_MEM_COPY_HOST_PTR,
cl.CL_size_t(unsafe.Sizeof(angles)), unsafe.Pointer(&angles[0]), &err)
x_coords_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_READ_WRITE,
cl.CL_size_t(unsafe.Sizeof(x_coords)), nil, &err)
y_coords_buffer = cl.CLCreateBuffer(context, cl.CL_MEM_READ_WRITE,
cl.CL_size_t(unsafe.Sizeof(y_coords)), nil, &err)
/* Create kernel argument */
err = cl.CLSetKernelArg(kernel, 0, cl.CL_size_t(unsafe.Sizeof(r_coords_buffer)), unsafe.Pointer(&r_coords_buffer))
if err < 0 {
println("Couldn't set a kernel argument")
return
}
cl.CLSetKernelArg(kernel, 1, cl.CL_size_t(unsafe.Sizeof(angles_buffer)), unsafe.Pointer(&angles_buffer))
cl.CLSetKernelArg(kernel, 2, cl.CL_size_t(unsafe.Sizeof(x_coords_buffer)), unsafe.Pointer(&x_coords_buffer))
cl.CLSetKernelArg(kernel, 3, cl.CL_size_t(unsafe.Sizeof(y_coords_buffer)), unsafe.Pointer(&y_coords_buffer))
/* 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
}
/* Read and print the result */
err = cl.CLEnqueueReadBuffer(queue, x_coords_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(x_coords)), unsafe.Pointer(&x_coords), 0, nil, nil)
if err < 0 {
println("Couldn't read the buffer")
return
}
cl.CLEnqueueReadBuffer(queue, y_coords_buffer, cl.CL_TRUE, 0,
cl.CL_size_t(unsafe.Sizeof(y_coords)), unsafe.Pointer(&y_coords), 0, nil, nil)
/* Display the results */
for i := 0; i < 4; i++ {
fmt.Printf("(%6.3f, %6.3f)\n", x_coords[i], y_coords[i])
}
/* Deallocate resources */
cl.CLReleaseMemObject(r_coords_buffer)
cl.CLReleaseMemObject(angles_buffer)
cl.CLReleaseMemObject(x_coords_buffer)
cl.CLReleaseMemObject(y_coords_buffer)
cl.CLReleaseKernel(kernel)
cl.CLReleaseCommandQueue(queue)
cl.CLReleaseProgram(*program)
cl.CLReleaseContext(context)
}