func getDevicesList() []deviceInfo {
var deviceIds = make([]cl.DeviceId, 100)
var numDevices uint32
err := cl.GetDeviceIDs(nil, cl.DEVICE_TYPE_ALL, uint32(len(deviceIds)),
&deviceIds[0], &numDevices)
if err != cl.SUCCESS {
log.Fatal("Failed to create device group")
}
var allDevices = make([]deviceInfo, numDevices)
for i := uint32(0); i < numDevices; i++ {
allDevices[i].deviceId = deviceIds[i]
var responseLength uint64
var responseData = make([]byte, BufferSize)
cl.GetDeviceInfo(deviceIds[i], cl.DEVICE_NAME, BufferSize,
unsafe.Pointer(&responseData[0]), &responseLength)
allDevices[i].name = string(responseData[:responseLength])
cl.GetDeviceInfo(deviceIds[i], cl.DEVICE_VENDOR, BufferSize,
unsafe.Pointer(&responseData[0]), &responseLength)
allDevices[i].vendor = string(responseData[:responseLength])
cl.GetDeviceInfo(deviceIds[i], cl.DEVICE_TYPE, BufferSize,
unsafe.Pointer(&responseData[0]), &responseLength)
allDevices[i].deviceType = string(responseData[:responseLength])
}
return allDevices
}
// StatInfo prints a bunch of usefull information about the currently available
// CL devices
func StatInfo() {
ids := make([]cl.PlatformID, 100)
actual := uint32(0)
cl.GetPlatformIDs(uint32(len(ids)), &ids[0], &actual)
for x := 0; x < int(actual); x++ {
data := make([]byte, DataSize)
size := uint64(0)
cl.GetPlatformInfo(ids[x], cl.PLATFORM_PROFILE, DataSize, unsafe.Pointer(&data[0]), &size)
profilestring := string(data[0:size])
cl.GetPlatformInfo(ids[x], cl.PLATFORM_VERSION, DataSize, unsafe.Pointer(&data[0]), &size)
versionstring := string(data[0:size])
cl.GetPlatformInfo(ids[x], cl.PLATFORM_NAME, DataSize, unsafe.Pointer(&data[0]), &size)
namestring := string(data[0:size])
cl.GetPlatformInfo(ids[x], cl.PLATFORM_VENDOR, DataSize, unsafe.Pointer(&data[0]), &size)
vendorstring := string(data[0:size])
cl.GetPlatformInfo(ids[x], cl.PLATFORM_EXTENSIONS, DataSize, unsafe.Pointer(&data[0]), &size)
extensionsstring := string(data[0:size])
log.Print("PLATFORM_PROFILE:\t\t", profilestring)
log.Print("PLATFORM_VERSION:\t\t", versionstring)
log.Print("PLATFORM_NAME:\t\t", namestring)
log.Print("PLATFORM_VENDOR:\t\t", vendorstring)
log.Print("PLATFORM_EXTENSIONS:\t", extensionsstring)
devices := make([]cl.DeviceId, 100)
actualDid := uint32(0)
cl.GetDeviceIDs(ids[x], cl.DEVICE_TYPE_ALL, uint32(len(devices)), &devices[0], &actualDid)
log.Println("Devices: ")
for y := 0; y < int(actualDid); y++ {
cl.GetDeviceInfo(devices[y], cl.DEVICE_NAME, DataSize, unsafe.Pointer(&data[0]), &size)
deviceName := string(data[0:size])
log.Print("\tname: "+deviceName+" @ ", &devices[y])
}
}
}
func main() {
data := make([]float32, DataSize)
for x := 0; x < len(data); x++ {
data[x] = rand.Float32()*99 + 1
}
//Get Device
var device cl.DeviceId
err := cl.GetDeviceIDs(nil, cl.DEVICE_TYPE_GPU, 1, &device, nil)
if err != cl.SUCCESS {
log.Fatal("Failed to create device group")
}
var errptr *cl.ErrorCode
//Create Computer Context
context := cl.CreateContext(nil, 1, &device, nil, nil, errptr)
if errptr != nil && cl.ErrorCode(*errptr) != cl.SUCCESS {
log.Fatal("couldnt create context")
}
defer cl.ReleaseContext(context)
//Create Command Queue
cq := cl.CreateCommandQueue(context, device, 0, errptr)
if errptr != nil && cl.ErrorCode(*errptr) != cl.SUCCESS {
log.Fatal("couldnt create command queue")
}
defer cl.ReleaseCommandQueue(cq)
//Create program
srcptr := cl.Str(KernelSource)
program := cl.CreateProgramWithSource(context, 1, &srcptr, nil, errptr)
if errptr != nil && cl.ErrorCode(*errptr) != cl.SUCCESS {
log.Fatal("couldnt create program")
}
defer cl.ReleaseProgram(program)
err = cl.BuildProgram(program, 1, &device, nil, nil, nil)
if err != cl.SUCCESS {
var length uint64
buffer := make([]byte, DataSize)
log.Println("Error: Failed to build program executable!")
cl.GetProgramBuildInfo(program, device, cl.PROGRAM_BUILD_LOG, uint64(len(buffer)), unsafe.Pointer(&buffer[0]), &length)
log.Fatal(string(buffer[0:length]))
}
//Get Kernel
kernel := cl.CreateKernel(program, cl.Str("square"+"\x00"), errptr)
if errptr != nil && cl.ErrorCode(*errptr) != cl.SUCCESS {
log.Fatal("couldnt create compute kernel")
}
defer cl.ReleaseKernel(kernel)
//Create buffers
input := cl.CreateBuffer(context, cl.MEM_READ_ONLY, 4*DataSize, nil, errptr)
if errptr != nil && cl.ErrorCode(*errptr) != cl.SUCCESS {
log.Fatal("couldnt create input buffer")
}
defer cl.ReleaseMemObject(input)
output := cl.CreateBuffer(context, cl.MEM_WRITE_ONLY, 4*DataSize, nil, errptr)
if errptr != nil && cl.ErrorCode(*errptr) != cl.SUCCESS {
log.Fatal("couldnt create output buffer")
}
defer cl.ReleaseMemObject(output)
//Write data
err = cl.EnqueueWriteBuffer(cq, input, cl.TRUE, 0, 4*DataSize, unsafe.Pointer(&data[0]), 0, nil, nil)
if err != cl.SUCCESS {
log.Fatal("Failed to write to source array")
}
//Set kernel args
count := uint32(DataSize)
err = cl.SetKernelArg(kernel, 0, 8, unsafe.Pointer(&input))
if err != cl.SUCCESS {
log.Fatal("Failed to write kernel arg 0")
}
err = cl.SetKernelArg(kernel, 1, 8, unsafe.Pointer(&output))
if err != cl.SUCCESS {
log.Fatal("Failed to write kernel arg 1")
}
err = cl.SetKernelArg(kernel, 2, 4, unsafe.Pointer(&count))
if err != cl.SUCCESS {
log.Fatal("Failed to write kernel arg 2")
}
local := uint64(0)
err = cl.GetKernelWorkGroupInfo(kernel, device, cl.KERNEL_WORK_GROUP_SIZE, 8, unsafe.Pointer(&local), nil)
if err != cl.SUCCESS {
log.Fatal("Failed to get kernel work group info")
}
global := local
err = cl.EnqueueNDRangeKernel(cq, kernel, 1, nil, &global, &local, 0, nil, nil)
if err != cl.SUCCESS {
log.Fatal("Failed to execute kernel!")
}
cl.Finish(cq)
results := make([]float32, DataSize)
err = cl.EnqueueReadBuffer(cq, output, cl.TRUE, 0, 4*1024, unsafe.Pointer(&results[0]), 0, nil, nil)
if err != cl.SUCCESS {
log.Fatal("Failed to read buffer!")
}
success := 0
notzero := 0
for i, x := range data {
if math.Abs(float64(x*x-results[i])) < 0.5 {
success++
}
if results[i] > 0 {
notzero++
}
log.Printf("I/O: %f\t%f", x, results[i])
}
log.Printf("%d/%d success", success, DataSize)
log.Printf("values not zero: %d", notzero)
}
