Main Content

gpucoder.atomicOr

Atomically perform bit-wise OR between a specified value and a variable in global or shared memory

Since R2021b

collapse all in page

    Syntax

    [A,oldA] = gpucoder.atomicOr(A,B)

    Description

    [A,oldA] = gpucoder.atomicOr(A,B) perform bit-wise OR between B and the value of A in global or shared memory and writes the results back into A. The operation is atomic in a sense that the entire read-modify-write operation is guaranteed to be performed without interference from other threads. The order of the input and output arguments must match the syntax provided.

    example

    Examples

    collapse all

    Perform a simple atomic addition operation by using the gpucoder.atomicOr function and generate CUDA® code that calls corresponding CUDA atomicOr() APIs.

    In one file, write an entry-point function myAtomicOr that accepts matrix inputs a and b.

    function a = myAtomicOr(a,b)

coder.gpu.kernelfun;
for i =1:numel(a)
    [a(i),~] = gpucoder.atomicOr(a(i), b);
end

end

    To create a type for a matrix of doubles for use in code generation, use the coder.newtype function.

    A = coder.newtype('uint32', [1 30], [0 1]);
B = coder.newtype('uint32', [1 1], [0 0]);
inputArgs = {A,B};

    To generate a CUDA library, use the codegen function.

    cfg = coder.gpuConfig('lib');
cfg.GenerateReport = true;

codegen -config cfg -args inputArgs myAtomicOr -d myAtomicOr

    The generated CUDA code contains the myAtomicOr_kernel1 kernel with calls to the atomicOr() CUDA APIs.

    //
// File: myAtomicOr.cu
//
...

static __global__ __launch_bounds__(1024, 1) void myAtomicOr_kernel1(
    const uint32_T b, const int32_T i, uint32_T a_data[])
{
  uint64_T loopEnd;
  uint64_T threadId;
...

  for (uint64_T idx{threadId}; idx <= loopEnd; idx += threadStride) {
    int32_T b_i;
    b_i = static_cast<int32_T>(idx);
    atomicOr(&a_data[b_i], b);
  }
}
...

void myAtomicOr(uint32_T a_data[], int32_T a_size[2], uint32_T b)
{
  dim3 block;
  dim3 grid;
...

    cudaMemcpy(gpu_a_data, a_data, a_size[1] * sizeof(uint32_T),
               cudaMemcpyHostToDevice);
    myAtomicOr_kernel1<<<grid, block>>>(b, i, gpu_a_data);
    cudaMemcpy(a_data, gpu_a_data, a_size[1] * sizeof(uint32_T),
               cudaMemcpyDeviceToHost);
...

}

    Input Arguments

    collapse all

    Operands, specified as scalars, vectors, matrices, or multidimensional arrays. Inputs A and B must satisfy the following requirements:

    • Have the same data type.

    • Have the same size or have sizes that are compatible. For example, A is an M-by-N matrix and B is a scalar or 1-by-N row vector.

    Data Types: int32 | uint32 | uint64

    Limitations

    • Function handle input to the gpucoder.stencilKernel pragma cannot contain calls to atomic functions. For example,

      out1 = gpucoder.stencilKernel(@myAtomicOr,A,[3 3],'same',B);

    Version History

    Introduced in R2021b

    See Also

    Functions

    Topics