c++ - openmp omp declare uniform 这在 GCC 中不受支持吗？

Question

我有一个简单的矩阵类，我想对其加运算符进行矢量化。但是，统一这似乎不受 GCC 支持(使用英特尔 C++ 编译器工作正常)。如果有任何解决方法，我很好奇。 (下面是编译命令的代码)

如果您有任何意见，请告诉我。

#include <iostream>
#include <cassert>
#include <omp.h>


#define printvar(a) std::cerr << (#a) << " = " << (a) << std::endl;
#define printline std::cerr << "FILE " << __FILE__ << " LINE " << __LINE__ << " FUNC: " << __func__ << std::endl;
#define NUMEL 10000

#pragma omp declare simd simdlen(2)
#pragma omp declare simd simdlen(4)
#pragma omp declare simd simdlen(8)
#pragma omp declare simd simdlen(16)
int myadd(int a, int b){
    return a+b;
}

template <typename V>
class Matrix {
public:
    V* data;
    int nrows, ncols;
    bool istemp = false;
    std::string name;
    Matrix(){
        data = NULL;
    }
    Matrix(int nrows, int ncols, V val, std::string name = "none"){

        this->nrows = nrows;
        this->ncols = ncols;
        this->name = name;


        data = (V*) malloc(sizeof(V)*nrows*ncols);
        for(int i = 0; i < numel(); i++){
            csi0(i) = val;
        }

    }

    Matrix(const Matrix& m){

        this->nrows = m.nrows;
        this->ncols = m.ncols;
        this->data = (V*) malloc(sizeof(V)*nrows*ncols);
        for(int i = 0; i < numel(); i++){
            csi0(i) = m.csi0(i);
        }

    }

    inline void swap(Matrix<V>& a, Matrix<V>& b){
        V* data = a.data; a.data = b.data; b.data = data; 
        int nrows = a.nrows; a.nrows = b.nrows; b.nrows = nrows; 
        int ncols = a.ncols; a.ncols = b.ncols; b.ncols = ncols; 
    }

    Matrix<V>& operator=(const Matrix<V>& m){
        Matrix<V> tmp(m);
        swap(*this,tmp);        
    }

    #pragma omp declare simd 
    #pragma omp declare simd simdlen(4) uniform(this)
    #pragma omp declare simd simdlen(8) uniform(this)
    #pragma omp declare simd simdlen(16) uniform(this)
    V& csi0 (int i) const {
        return this->data[i];
    }

    V& cij0 (int i, int j) const {
        return data[i+nrows*j];
    }

    int numel() const {     
        return nrows*ncols;
    }

    friend Matrix<V> operator+(const Matrix<V>& a, const Matrix<V>& b){

        assert(a.nrows == b.nrows && a.ncols == b.ncols);
        Matrix<V> retmat(a.nrows, a.ncols, 0, "retmat");
        #pragma omp parallel for simd 
        for(int i = 0; i < a.numel(); i++){
            retmat.csi0(i) = a.csi0(i) + b.csi0(i);
        }

        // retmat.istemp = true;
        return retmat;
    }

    ~Matrix(){
        if(data != NULL && !istemp){
    
            free(data);
        }

    }


    friend std::ostream& operator<<(std::ostream& os, Matrix<V> m){
        os << "{{" << std::endl;
        for(int i = 0; i < m.nrows; i++){
            for(int j = 0; j < m.ncols; j++){
                os << m.csi0(i) << ", ";
            }
            os << std::endl;
        }
        os << "}}" << std::endl;
        return os;
    }
};

int main(int argc, char const *argv[])
{
    Matrix<int> a(4,5,1), b(4,5,2), c;
    c = a + b;
    printvar(c)
}



int main_1(int argc, char const *argv[])
{
    int a[NUMEL], b[NUMEL], c[NUMEL];

    #pragma omp parallel for 
    for(int i = 0; i < NUMEL; i++){
        c[i] = myadd(a[i], b[i]);
    }
    return 0;
}

编译命令

g++.exe -fopt-info-all=all.optrpt -O3  -fopenmp -fopenmp-simd -mthreads -mavx -fmax-errors=5   -c gcc_ompsimd_test.cpp -Fo:gcc_ompsimd_test.o
gcc_ompsimd_test.cpp:65:46: error: expected unqualified-id before 'this'
  #pragma omp declare simd simdlen(4) uniform(this)
                                              ^~~~
gcc_ompsimd_test.cpp:66:46: error: expected unqualified-id before 'this'
  #pragma omp declare simd simdlen(8) uniform(this)
                                              ^~~~
gcc_ompsimd_test.cpp:67:47: error: expected unqualified-id before 'this'
  #pragma omp declare simd simdlen(16) uniform(this)

Answer 1

在 gcc 中，您似乎不能在类声明中使用 uniform(this)。在 clang 和 Intel 编译器中是可以的。所以，成员函数的定义不应该在类声明中:

#pragma omp declare simd simdlen(16) uniform(this)
template <typename V>
V& Matrix<V>::csi0 (int i) const 
{
    return data[i];
}

从 OpenMP 4.5 开始，规范包含以下行 ( 2.8.2)。

The special this pointer can be used as if it was one of the argumentsto the function in any of the linear, aligned, or uniform clauses.

因此任何 OpenMP 4.5 兼容编译器都应该支持它。