JAVA和JNI之间的速度差异

我想让我的Android Studio代码更快,所以我将一些Java代码翻译成原生C。令人惊讶的是,用C编写的函数执行速度几乎比Java函数慢三倍。此外,我还应用了ndk构建优化,但这并没有改变任何事情。我的代码:

1) Java函数:

private int[] convolve(short signal[], short kernel[]) {
    int ret[] = new int[signal.length + kernel.length - 1];

    for (int n = 0; n < ret.length; n++){
        ret[n] = 0;

        int kmin = (n >= kernel.length - 1) ? n - (kernel.length - 1) : 0;
        int kmax = (n < signal.length - 1) ? n : signal.length - 1;

        for (int k = kmin; k <= kmax; k++) {
            ret[n] += signal[k] * kernel[n - k];
        }
    }

    return ret;
}

2) C功能:

int convolve_max(int16_t Signal[/* SignalLen */], int32_t SignalLen,
             int16_t Kernel[/* KernelLen */], int32_t KernelLen)
{
int32_t n;
double Result;
double MaxVal = 0;
int32_t MaxInd = 0;
double abs_sum = 0;

for (n = 0; n < SignalLen + KernelLen - 1; n++) {

    int32_t kmin, kmax, k;
    Result = 0;
    kmin = (n >= KernelLen - 1) ? n - (KernelLen - 1) : 0;
    kmax = (n < SignalLen - 1) ? n : SignalLen - 1;

    for (k = kmin; k <= kmax; k++) {
        Result += Signal[k] * Kernel[n - k];
    }

    Result = abs(Result);

    abs_sum += Result;

    if (Result > MaxVal) {
        MaxVal = Result;
        MaxInd = n;
    }

}

double trs = abs_sum / 200;
int offset = 0;

if (MaxVal > trs) {
    offset = MaxInd - 10 * 512;
}
else {
    offset = 51;
}
if (abs(offset) < 25) {
    return 1;
}

return 0;

}

我知道它们并不完全相同,但最重要的是嵌套循环,这是相同的。

此外,我的Gradle ndk外观:

externalNativeBuild {
     ndkBuild {
         cppFlags "-std=c++11"
         arguments                                                                       "APP_STL=gnustl_static","NDK_TOOLCHAIN_VERSION=4.9","APP_OPTIM=release","NDKDEBUG=0", "V=1"
         abiFilters "armeabi"
        }
    }`

那么,我是否犯了一些错误,或者jni存在一些问题?