推力减少和重载运算符 - (const float3&,const float3&)不会编译

时间:2017-06-06 10:14:08

标签: cuda operator-overloading thrust

我重载运算符,使float3中的vectorspace.cuh(和类似结构)上的向量空间:

// Boilerplate vector space over data type Pt
#pragma once

#include <type_traits>


// float3
__device__ __host__ float3 operator+=(float3& a, const float3& b) {
    a.x += b.x; a.y += b.y; a.z += b.z;
    return a;
}

__device__ __host__ float3 operator*=(float3& a, const float b) {
    a.x *= b; a.y *= b; a.z *= b;
    return a;
}

// float4
__device__ __host__ float4 operator+=(float4& a, const float4& b) {
    a.x += b.x; a.y += b.y; a.z += b.z; a.w += b.w;
    return a;
}

__device__ __host__ float4 operator*=(float4& a, const float b) {
    a.x *= b; a.y *= b; a.z *= b; a.w *= b;
    return a;
}


// Generalize += and *= to +, -=, -, *, /= and /
template<typename Pt> __device__ __host__
typename std::enable_if<std::is_class<Pt>::value || std::is_enum<Pt>::value, Pt>::type
operator+(const Pt& a, const Pt& b) {
    auto sum = a;
    sum += b;
    return sum;
}

template<typename Pt> __device__ __host__
typename std::enable_if<std::is_class<Pt>::value || std::is_enum<Pt>::value, Pt>::type
operator-=(Pt& a, const Pt& b) {
    a += -1*b;
    return a;
}

template<typename Pt> __device__ __host__
typename std::enable_if<std::is_class<Pt>::value || std::is_enum<Pt>::value, Pt>::type
operator-(const Pt& a, const Pt& b) {
    auto diff = a;
    diff -= b;
    return diff;
}

template<typename Pt> __device__ __host__
typename std::enable_if<std::is_class<Pt>::value || std::is_enum<Pt>::value, Pt>::type
operator-(const Pt& a) {
    return -1*a;
}

template<typename Pt> __device__ __host__
typename std::enable_if<std::is_class<Pt>::value || std::is_enum<Pt>::value, Pt>::type
operator*(const Pt& a, const float b) {
    auto prod = a;
    prod *= b;
    return prod;
}

template<typename Pt> __device__ __host__
typename std::enable_if<std::is_class<Pt>::value || std::is_enum<Pt>::value, Pt>::type
operator*(const float b, const Pt& a) {
    auto prod = a;
    prod *= b;
    return prod;
}

template<typename Pt> __device__ __host__
typename std::enable_if<std::is_class<Pt>::value || std::is_enum<Pt>::value, Pt>::type
operator/=(Pt& a, const float b) {
    a *= 1./b;
    return a;
}

template<typename Pt> __device__ __host__
typename std::enable_if<std::is_class<Pt>::value || std::is_enum<Pt>::value, Pt>::type
operator/(const Pt& a, const float b) {
    auto quot = a;
    quot /= b;
    return quot;
}

这些重载会破坏thrust::reduce的编译,这里有一个例子:

#include <thrust/reduce.h>
#include <thrust/execution_policy.h>

#include "vectorspace.cuh"


int main(int argc, char const *argv[]) {
    int n = 10;
    float3* d_arr;
    cudaMalloc(&d_arr, n*sizeof(float3));

    auto sum = thrust::reduce(thrust::device, d_arr, d_arr + n, float3 {0});

    return 0;
}

在Ubuntu 16.04上使用nvcc -std=c++11 -arch=sm_52会导致200多行编译错误:

$ nvcc -std=c++11 -arch=sm_52 sandbox/mean.cu 
sandbox/mean.cu(26): error: no operator "*" matches these operands
            operand types are: int * const thrust::zip_iterator<thrust::tuple<const float3 *, thrust::pointer<float3, thrust::system::cuda::detail::par_t, thrust::use_default, thrust::use_default>, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type>>
          detected during:
            instantiation of "std::enable_if<<expression>, Pt>::type operator-=(Pt &, const Pt &) [with Pt=thrust::zip_iterator<thrust::tuple<const float3 *, thrust::pointer<float3, thrust::system::cuda::detail::par_t, thrust::use_default, thrust::use_default>, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type>>]" 
(35): here
            instantiation of "std::enable_if<<expression>, Pt>::type operator-(const Pt &, const Pt &) [with Pt=thrust::zip_iterator<thrust::tuple<const float3 *, thrust::pointer<float3, thrust::system::cuda::detail::par_t, thrust::use_default, thrust::use_default>, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type, thrust::null_type>>]" 

...

如何在不破坏thrust的情况下重载运算符?

1 个答案:

答案 0 :(得分:2)

(在OP的编辑后编辑。)

问题在于运算符重载的“范围”:您不仅仅为您感兴趣的类重载,而是适合您enable_if条件的所有类的重载 - 这很放松。即使事情会编译,这已经是一个严重的错误。

更具体地说,推力使用算术运算,例如关于“zip iterators”(不管它们是什么),并且这些迭代器的操作编译失败,这是可以理解的。

所以你必须:

  • 确切指定与重载相关的类(例如,使用std::is_sameenable_if的分离)或
  • 使用trait class

    template<class T> struct needs_qivs_arithmetic_operators : public std::false_type {};
    
    template<> struct needs_qivs_arithmetic_operators<float3> : public std::true_type {};
    template<> struct needs_qivs_arithmetic_operators<float4> : public std::true_type {};
    /* ... etc. You can also add specializations elsewhere in the translation unit. */