nvcc 生成无效错误编译 JNI 代码

Question

$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Fri_Sep__1_21:08:03_CDT_2017
Cuda compilation tools, release 9.0, V9.0.176

错误是：

MyFile.cu(231): error: expression must have pointer type

相关代码：

JNIEXPORT jboolean JNICALL Java_MyFile_convergeMatrixCuda (
    JNIEnv *env, jclass clazz, jfloatArray fxnMatrixJ, jfloatArray mulMatrixJ, jfloatArray addMatrixJ, 
    jfloatArray resultsJ, jint numRowsJ, jint numColsJ, jint maxIterations, jfloat epsilonJ)
{
    int     numRows = (int) numRowsJ;
    int     numCols = (int) numColsJ;
    int     maxIter = (int) maxIterations;
    float   epsilon = (float) epsilonJ;
    float   *fxnMatrixH = (*env)->GetFloatArrayElements (env, fxnMatrixJ, NULL);

GetFloatArrayElements returns 一个浮点数*。将“(*env)->GetFloatArrayElements”替换为 "env->GetFloatArrayElements" 会出现以下错误：

float   *fxnMatrixH = env->GetFloatArrayElements (env, fxnMatrixJ, NULL);

MyFile.cu(231): error: argument of type "JNIEnv *" is incompatible with parameter of type "jfloatArray"

MyFile.cu(231): error: argument of type "jfloatArray" is incompatible with parameter of type "jboolean *"

MyFile.cu(231): error: too many arguments in function call

nvcc 在编译非 JNI 代码时确实可以正常工作

Answer 1

NVidia's documentation 表示

Source files for CUDA applications consist of a mixture of conventional C++ host code, plus GPU device functions.

(*env)->GetFloatArrayElements (env, fxnMatrixJ, NULL); 是您在 C 中调用 JNI 函数的方式。但在 C++ 中它将是 env->GetFloatArrayElements(fxnMatrixJ, NULL);

nvcc 生成无效错误编译 JNI 代码

nvcc generating invalid error compiling JNI code

java-native-interface

nvcc