2

我在推力 cuda 中使用 float4 时遇到了内存问题

将 float4“buggyVariable”作为成员添加到仿函数似乎会导致浮点数据左移 1 个浮点数。

在 CUDA_animateParticles 我清楚地将 Y 设置为 0 和 Z 设置为 1

然而，当运行函子并在 OpenGL 中绘制它时。我得到 Xposition=1 的粒子，这表明 Y 在函子内为 1。

我还用 float2 和 float3 进行了测试，它们似乎工作正常。

所以这似乎是内存对齐问题或错误。

任何人都可以对此有所了解吗？谢谢您的帮助。

#include <thrust/sort.h>
#include <thrust/random.h>
#include <thrust/device_vector.h>
#include "cutil_math.h"

struct animateParticles_functor
{
    float4 buggyVariable; //why does adding this variable cause following floats to get wrong values???
    float pex, pey, pez, pew;

    __host__ __device__
    animateParticles_functor( float x, float y, float z, float w) :
        pex(x), pey(y), pez(z), pew(w)
    {
    }

    template <typename Tuple>
    __host__ __device__
    void operator()(Tuple t)
    {
        if(pey > 0)
            thrust::get<0>(t) = make_float4(1, 0, 0, 0); //true if y is bugged
        else
            thrust::get<0>(t) = make_float4(0, 0, 0, 0); //false if its not bugged

        return;
    }
}

void CUDA_animateParticles(float4* cuda_devicePointer_vboPosition, float3* cuda_devicePointer_particleVelocitys, unsigned int numParticles, float4 particleEmitter)
{
    thrust::device_ptr<float4> d_pos(cuda_devicePointer_vboPosition);
    thrust::device_ptr<float3> d_vel(cuda_devicePointer_particleVelocitys);

    thrust::for_each(
        thrust::make_zip_iterator(thrust::make_tuple(d_pos, d_vel)),
        thrust::make_zip_iterator(thrust::make_tuple(d_pos + numParticles, d_vel + numParticles)),
        animateParticles_functor(0, 0, 1, 0) //notice that i set Z to 1 and not Y to 0
    );
}
4

1 回答 1

2

I think there are some flavors of MSVC where nvcc and cl.exe can't agree on sizeof(float4).

Try replacing your use of float4 with my_float4:

struct my_float4
{
  float x, y, z, w;
};
于 2013-03-19T18:22:26.193 回答