Function main

examples/neon_gemm_s8_f32.cpp:80–239

}

int main(int argc, char **argv)
{
    size_t M = 4;
    size_t N = 4;
    size_t K = 4;

    // Parse args
    if (argc < 3) /* case default matrix sizes */
    {
        // Print help
        std::cout << "Usage: ./build/neon_gemm_qasymm8 M N K\n";
        std::cout << "Too few or no inputs provided. Using default M=4, N=4, K=4\n\n";
    }
    else /* case M N K arguments provided */
    {
        M = strtol(argv[1], nullptr, 10);
        N = strtol(argv[2], nullptr, 10);
        K = strtol(argv[3], nullptr, 10);
    }

    /*** Floating point matrix multiplication ***/

    // Initialise input matrices
    NEGEMM fgemm{};

    Tensor src1;
    Tensor src2;
    Tensor dst;
    src1.allocator()->init(TensorInfo(TensorShape(K, M), 1, DataType::F32));
    src2.allocator()->init(TensorInfo(TensorShape(N, K), 1, DataType::F32));
    dst.allocator()->init(TensorInfo(TensorShape(N, M), 1, DataType::F32));
    fgemm.configure(&src1, &src2, nullptr, &dst, 1, 0);

    // Allocate matrices
    src1.allocator()->allocate();
    src2.allocator()->allocate();
    dst.allocator()->allocate();

    float min1 = 0.0f;
    float max1 = 1.0f;
    fill_random_tensor(src1, 0, min1, max1);

    float min2 = -1.0f;
    float max2 = 2.0f;
    fill_random_tensor(src2, 1, min2, max2);

    // Run single precision gemm and print result
    fgemm.run();

#if ARM_COMPUTE_DEBUG_ENABLED
    std::cout << "# F32 GEMM result:\n";
    std::cout << "src1=[ \n";
    src1.print(std::cout);
    std::cout << "] \n";
    std::cout << "src2=[ \n";
    src2.print(std::cout);
    std::cout << "] \n";
    std::cout << "dst=[ \n";