SWAP gate implementation using cuStateVec. The example prepares $\lvert 10\rangle$ and applies SWAP to produce $\lvert 01\rangle$.

// Compile: nvcc main.cu -o main -lcustatevec
// Run:     ./main
 
#include <stdio.h>
#include <cuda_runtime.h>
#include <custatevec.h>
 
int main() {
    const int nQubits = 2;
    const int dim = 1 << nQubits;
 
    // Prepare |10>: q0=1, q1=0 -> index 2 (binary 10)
    cuDoubleComplex h_sv[4] = {0};
    h_sv[2] = make_cuDoubleComplex(1.0, 0.0);
    cuDoubleComplex *d_sv;
    cudaMalloc(&d_sv, dim * sizeof(cuDoubleComplex));
    cudaMemcpy(d_sv, h_sv, dim * sizeof(cuDoubleComplex), cudaMemcpyHostToDevice);
 
    custatevecHandle_t handle;
    custatevecCreate(&handle);
 
    // SWAP gate matrix (4x4, row-major)
    cuDoubleComplex gate[16] = {
        {1,0},{0,0},{0,0},{0,0},
        {0,0},{0,0},{1,0},{0,0},
        {0,0},{1,0},{0,0},{0,0},
        {0,0},{0,0},{0,0},{1,0}
    };
    int32_t targets[] = {0, 1};
    custatevecApplyMatrix(
        handle, d_sv, CUDA_C_64F, nQubits,
        gate, CUDA_C_64F, CUSTATEVEC_MATRIX_LAYOUT_ROW, 0,
        targets, 2, NULL, NULL, 0,
        CUSTATEVEC_COMPUTE_64F, NULL, 0);
 
    cudaMemcpy(h_sv, d_sv, dim * sizeof(cuDoubleComplex), cudaMemcpyDeviceToHost);
    printf("|01>: (%.4f, %.4f)\n", cuCreal(h_sv[1]), cuCimag(h_sv[1]));
    printf("|10>: (%.4f, %.4f)\n", cuCreal(h_sv[2]), cuCimag(h_sv[2]));
    // |01>: (1.0000, 0.0000)  <- qubit states exchanged
    // |10>: (0.0000, 0.0000)
 
    custatevecDestroy(handle);
    cudaFree(d_sv);
    return 0;
}