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 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
// Use clGetPlatformIDs() to retrieve the number of
// platforms
status = cl.CLGetPlatformIDs(0, nil, &numPlatforms)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLGetPlatformIDs")
// 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
// 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")
var caps cl.CL_device_svm_capabilities
var caps_value interface{}
status = cl.CLGetDeviceInfo(
devices[0],
cl.CL_DEVICE_SVM_CAPABILITIES,
cl.CL_size_t(unsafe.Sizeof(caps)),
&caps_value,
nil)
caps = caps_value.(cl.CL_device_svm_capabilities)
// Coarse-grained buffer SVM should be available on any OpenCL 2.0 device.
// So it is either not an OpenCL 2.0 device or it must support coarse-grained buffer SVM:
if !(status == cl.CL_SUCCESS && (caps&cl.CL_DEVICE_SVM_FINE_GRAIN_BUFFER) != 0) {
fmt.Printf("Cannot detect fine-grained buffer SVM capabilities on the device. The device seemingly doesn't support fine-grained buffer SVM. caps=%x\n", caps)
println("")
return
}
//-----------------------------------------------------
// STEP 3: Create a 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
//-----------------------------------------------------
// Create a command queue using clCreateCommandQueueWithProperties(),
// and associate it with the device you want to execute
queue := cl.CLCreateCommandQueueWithProperties(context,
devices[0],
nil,
&status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateCommandQueueWithProperties")
defer cl.CLReleaseCommandQueue(queue)
//-----------------------------------------------------
// STEP 5: Create and compile the program
//-----------------------------------------------------
programSource, programeSize := utils.Load_programsource("svmfg.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 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, &log, nil)
fmt.Printf("%s\n", log)
return
}
//utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLBuildProgram")
//-----------------------------------------------------
// STEP 7: Create the kernel
//-----------------------------------------------------
// Use clCreateKernel() to create a kernel
kernel := cl.CLCreateKernel(program,
[]byte("svmbasic"),
&status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateKernel")
defer cl.CLReleaseKernel(kernel)
// Then call the main sample routine - resource allocations, OpenCL kernel
// execution, and so on.
svmbasic(1024*1024, context, queue, kernel)
// All resource deallocations happen in defer.
}
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 cmdQueue cl.CL_command_queue
// Create a command queue using clCreateCommandQueueWithProperties(),
// and associate it with the device you want to execute
cmdQueue = cl.CLCreateCommandQueueWithProperties(context,
devices[0],
nil,
&status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateCommandQueueWithProperties")
defer cl.CLReleaseCommandQueue(cmdQueue)
//-----------------------------------------------------
// STEP 5: Create device buffers
//-----------------------------------------------------
// initialize any device/SVM memory here.
/* svm buffer for binary tree */
svmTreeBuf := cl.CLSVMAlloc(context,
cl.CL_MEM_READ_WRITE,
cl.CL_size_t(NUMBER_OF_NODES*unsafe.Sizeof(sampleNode)),
0)
if nil == svmTreeBuf {
println("clSVMAlloc(svmTreeBuf) failed.")
return
}
defer cl.CLSVMFree(context, svmTreeBuf)
/* svm buffer for search keys */
svmSearchBuf := cl.CLSVMAlloc(context,
cl.CL_MEM_READ_WRITE,
cl.CL_size_t(NUMBER_OF_SEARCH_KEY*unsafe.Sizeof(sampleKey)),
0)
if nil == svmSearchBuf {
println("clSVMAlloc(svmSearchBuf) failed.")
return
}
defer cl.CLSVMFree(context, svmSearchBuf)
//create the binary tree and set the root
/* root node of the binary tree */
svmRoot := cpuCreateBinaryTree(cmdQueue, svmTreeBuf)
//initialize search keys
cpuInitSearchKeys(cmdQueue, svmSearchBuf)
/* if voice is not deliberately muzzled, shout parameters */
fmt.Printf("-------------------------------------------------------------------------\n")
fmt.Printf("Searching %d keys in a BST having %d Nodes...\n", NUMBER_OF_SEARCH_KEY, NUMBER_OF_NODES)
fmt.Printf("-------------------------------------------------------------------------\n")
//-----------------------------------------------------
// STEP 6: Create and compile the program
//-----------------------------------------------------
programSource, programeSize := utils.Load_programsource("bst.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
//-----------------------------------------------------
var kernel cl.CL_kernel
// Use clCreateKernel() to create a kernel
kernel = cl.CLCreateKernel(program, []byte("bst_kernel"), &status)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLCreateKernel")
defer cl.CLReleaseKernel(kernel)
//-----------------------------------------------------
// 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.CLSetKernelArgSVMPointer(kernel,
0,
svmTreeBuf)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArgSVMPointer(svmTreeBuf)")
status = cl.CLSetKernelArgSVMPointer(kernel,
1,
svmSearchBuf)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clSetKernelArgSVMPointer(svmSearchBuf)")
//-----------------------------------------------------
// 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.
var localWorkSize [1]cl.CL_size_t
var kernelWorkGroupSize interface{}
status = cl.CLGetKernelWorkGroupInfo(kernel,
devices[0],
cl.CL_KERNEL_WORK_GROUP_SIZE,
cl.CL_size_t(unsafe.Sizeof(localWorkSize[0])),
&kernelWorkGroupSize,
nil)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLGetKernelWorkGroupInfo")
localWorkSize[0] = kernelWorkGroupSize.(cl.CL_size_t)
var globalWorkSize [1]cl.CL_size_t
globalWorkSize[0] = NUMBER_OF_SEARCH_KEY
//-----------------------------------------------------
// STEP 10: 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[:],
localWorkSize[:],
0,
nil,
nil)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "CLEnqueueNDRangeKernel")
status = cl.CLFlush(cmdQueue)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clFlush")
status = cl.CLFinish(cmdQueue)
utils.CHECK_STATUS(status, cl.CL_SUCCESS, "clFinish")
//-----------------------------------------------------
// 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
//this demo doesn't need clEnqueueReadBuffer due to SVM
//-----------------------------------------------------
// STEP 12: Verify the results
//-----------------------------------------------------
// reference implementation
svmBinaryTreeCPUReference(cmdQueue,
svmRoot,
svmTreeBuf,
svmSearchBuf)
// compare the results and see if they match
pass := svmCompareResults(cmdQueue, svmSearchBuf)
if pass {
println("Passed!")
} else {
println("Failed!")
}
}
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)
}