void gemm(int M, int N, int K, const float* a, const float* b, float* c) {
    for (int i = 0; i < M; ++i) {
        for (int j = 0; j < N; ++j) {
            float sum = 0;
            for (int kk = 0; kk < K; ++k) {
                sum += a[i * K + kk] * b[kk * N + j];
            }
            c[i * N + j] = sum;
        }
    }
}