Main Content


Atomically compare and swap the value of a variable in global or shared memory

Since R2021b



    [A,oldA] = gpucoder.atomicCAS(A,B,C) compares B to the value of A in global or shared memory and if the values are the same writes the value of C 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.


    collapse all

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

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

    function a = myAtomicCAS(a,b,c)
    for i =1:numel(a)
        [a(i),~] = gpucoder.atomicCAS(a(i), b, c);

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

    To generate a CUDA library, use the codegen function.

    cfg = coder.gpuConfig('lib');
    cfg.GenerateReport = true;
    codegen -config cfg -args inputArgs myAtomicCAS -d myAtomicCAS

    The generated CUDA code contains the myAtomicCAS_kernel1 kernel with calls to the atomicCAS() CUDA APIs.

    // File:
    static __global__ __launch_bounds__(1024, 1) void myAtomicCAS_kernel1(
        const uint32_T c, 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);
        atomicCAS(&a_data[b_i], b, c);
    void myAtomicCAS(uint32_T a_data[], int32_T a_size[2], uint32_T b, uint32_T c)
      dim3 block;
      dim3 grid;
      if (validLaunchParams) {
        cudaMemcpy(gpu_a_data, a_data, a_size[1] * sizeof(uint32_T),
        myAtomicCAS_kernel1<<<grid, block>>>(c, b, i, gpu_a_data);
        cudaMemcpy(a_data, gpu_a_data, a_size[1] * sizeof(uint32_T),

    Input Arguments

    collapse all

    Operands, specified as scalars, vectors, matrices, or multidimensional arrays. Inputs A, B, and C 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,C is a scalar or 1-by-N row vector.

    Data Types: int32 | uint32 | uint64

    Version History

    Introduced in R2021b