Julia interface for the OpenCL parallel computation API
This package aims to be a complete solution for OpenCL programming in Julia, similar in scope to PyOpenCL for Python. It provides a high level API for OpenCL to make programing hardware accelerators, such as GPUs, FPGAs, and DSPs, as well as multicore CPUs much less onerous.
OpenCL.jl needs your help! If you can help maintaining this package, please reach out on the JuliaLang Slack #gpu channel
- Install an OpenCL driver. You can install one system-wide, i.e., using your package manager, or use
pocl_jll.jl
for a CPU back-end. - Add OpenCL to your Julia environment:
using Pkg
Pkg.add("OpenCL")
Note: We use cl.create_compute_context()
here which only considers GPUs and CPUs.
using LinearAlgebra
using OpenCL, pocl_jll
const sum_kernel = "
__kernel void sum(__global const float *a,
__global const float *b,
__global float *c)
{
int gid = get_global_id(0);
c[gid] = a[gid] + b[gid];
}
"
a = rand(Float32, 50_000)
b = rand(Float32, 50_000)
device, ctx, queue = cl.create_compute_context()
a_buff = cl.Buffer(Float32, ctx, length(a), (:r, :copy), hostbuf=a)
b_buff = cl.Buffer(Float32, ctx, length(b), (:r, :copy), hostbuf=b)
c_buff = cl.Buffer(Float32, ctx, length(a), :w)
p = cl.Program(ctx, source=sum_kernel) |> cl.build!
k = cl.Kernel(p, "sum")
queue(k, size(a), nothing, a_buff, b_buff, c_buff)
r = cl.read(queue, c_buff)
if isapprox(norm(r - (a+b)), zero(Float32))
@info "Success!"
else
@error "Norm should be 0.0f"
end
You may want to check out the examples
folder. Either git clone
the repository to your local machine or navigate to the OpenCL.jl install directory via:
using OpenCL
cd(joinpath(dirname(pathof(OpenCL)), ".."))
Otherwise, feel free to take a look at the Jupyter notebooks below:
This package is heavily influenced by the work of others:
- PyOpenCL by Andreas Klockner
- oclpb by Sean Ross
- Boost.Compute by Kyle Lutz
- rust-opencl
Here's a rough translation between the OpenCL API in C to this Julia version. Optional arguments are indicated by [name?]
(see clCreateBuffer
, for example). For a quick reference to the C version, see the Khronos quick reference card.
C | Julia | Notes |
---|---|---|
clGetPlatformIDs |
cl.platforms() |
|
clGetPlatformInfo |
cl.info(platform, :symbol) |
Platform info: :profile , :version , :name , :vendor , :extensions |
clGetDeviceIDs |
cl.devices() , cl.devices(platform) , cl.devices(:type) |
Device types: :all , :cpu , :gpu , :accelerator , :custom , :default |
clGetDeviceInfo |
cl.info(device, :symbol) |
Device info: :driver_version , :version , :profile , :extensions , :platform , :name , :device_type , :has_image_support , :queue_properties , :has_queue_out_of_order_exec , :has_queue_profiling , :has_native_kernel , :vendor_id , :max_compute_units , :max_work_item_size , :max_clock_frequency , :address_bits , :max_read_image_args , :max_write_image_args , :global_mem_size , :max_mem_alloc_size , :max_const_buffer_size , :local_mem_size , :has_local_mem , :host_unified_memory , :available , :compiler_available , :max_work_group_size , :max_work_item_dims , :max_parameter_size , :profiling_timer_resolution , :max_image2d_shape , :max_image3d_shape |
clCreateContext |
cl.context(queue) , cl.context(CLMemObject), cl.context(CLArray)` |
|
clReleaeContext |
cl.release! |
C | Julia | Notes |
---|---|---|
clCreateBuffer |
cl.Buffer(type, context, [length?]; [hostbuf?]) , cl.Buffer(type, context, flags, [length?]; [hostbuf?]) |
Memory flags: :rw , :r , :w , :use , :alloc , :copy |
clEnqueueCopyBuffer |
cl.copy!(queue, destination, source) |
|
clEnqueueFillBuffer |
cl.enqueue_fill_buffer(queue, buffer, pattern, offset, nbytesm wait_for) |
|
clEnqueueReadBuffer |
cl.enqueue_read_buffer(queue, buffer, hostbuf, dev_offset, wait_for, is_blocking) |
|
clEnqueueWriteBuffer |
cl.enqueue_write_buffer(queue, buffer, hostbuf, byte_count, offset, wait_for, is_blocking) |
C | Julia | Notes |
---|---|---|
clCreateProgramWithSource |
cl.Program(ctx; source) |
|
clCreateProgramWithBinaries |
cl.Program(ctx; binaries) |
|
clReleaseProgram |
cl.release! |
|
clBuildProgram |
cl.build!(progrm, options) |
|
clGetProgramInfo |
cl.info(program, :symbol) |
Program info: :reference_count , :devices , :context , :num_devices , :source , :binaries , :build_log , :build_status |
C | Julia | Notes |
---|---|---|
clCreateKernel |
cl.Kernel(program, "kernel_name") |
|
clGetKernelInfo |
cl.info(kernel, :symbol) |
Kernel info: :name , :num_args , :reference_count , :program , :attributes |
clEnqueueNDRangeKernel |
cl.enqueue_kernel(queue, kernel, global_work_size) , cl.enqueue_kernel(queue, kernel, global_work_size, local_work_size; global_work_offset, wait_on) |
|
clSetKernelArg |
cl.set_arg!(kernel, idx, arg) |
idx starts at 1 |
clCreateUserEvent |
cl.UserEvent(ctx; retain) |
|
clGetEventInfo |
cl.info(event, :symbol) |
Event info: :context , :command_queue , :reference_count , :command_type , :status , :profile_start , :profile_end , :profile_queued , :profile_submit , :profile_duration |
clWaitForEvents |
cl.wait(event) , cl.wait(events) |
|
clEnqueueMarkerWithWaitList |
cl.enqueue_marker_with_wait_list(queue, wait_for) |
|
clEnqueueBarrierWithWaitList |
cl.enqueue_barrier_with_wait_list(queue, wait_for) |