inline __device__ void store()

in kernels/fmha/smem_tile.h [1616:1626]

• 11 lines of code
• 4 McCabe index (conditional complexity)

    inline __device__ void store(const float (&sum)[LDGS]) {
        if (tidx_ % THREADS_PER_ROW == 0) {
            int row = tidx_ / THREADS_PER_ROW;
            #pragma unroll
            for (int i = 0; i < LDGS; ++i) {
                if (row + i * ROWS_PER_LDG < ROWS) {
                    smem_write_buffer_[row + i * ROWS_PER_LDG] = sum[i];
                }
            }
        }
    }