Paste: opencl

STRING: kernel-source2
__kernel square(
    __global float* input,
    __global float* output,
    const unsigned int count,
    const float addit)
{
    int i = get_global_id(0);
    if (i < count)
        output[i] = input[i] * input[i] + addit ;
}
;

:: opencl-demo2 ( -- result )
    [
        platforms first devices>> first                                         :> device
        device 1array <context> &dispose                                        :> context
        context device f f <command-queue> &dispose                             :> queue
        device "" context kernel-source2 1array sources>program &dispose        :> program
        program "square" <kernel> &dispose                                      :> kernel
        context read-access 16 float-array{ 1.0 2.0 3.0 4.0 } <buffer> &dispose :> in-data
        context write-access 16 f <buffer> &dispose                             :> out-data
        kernel 0 in-data set-kernel-arg-buffer
        kernel 1 out-data set-kernel-arg-buffer
        kernel 2 4 <uint> set-kernel-arg-data
        kernel 3 9.5 <float> set-kernel-arg-data
        queue kernel { 4 } [ ] enqueue-kernel &dispose                          :> event
        queue flush-queue
        event wait-event
        queue out-data 0 buffer-ptr boa 16 read-buffer 4 <direct-float-array>
    ] with-destructors ;