glm 0.9.5.2 CUDA compilation errors

[andrewcorrigan]

I am unable to compile a simple code using glm 0.9.5.2 with CUDA 5.5 on Mac 10.9 (using Apple LLVM version 5.0 (clang-500.2.79) (based on LLVM 3.3svn)) or CUDA 6.0 (CUDA 5.0 also) on Ubuntu 13.10 (g++ Ubuntu/Linaro 4.8.1-10ubuntu9) 4.8.1.

Note: I am not even trying to call glm from device code, I am simply compiling host code with the nvcc compiler.

Compilation command:

nvcc -I${GLM_INCLUDE_PATH} test.cu

Simple reproducer:

#ifdef __CUDACC__
#  include <cuda.h>
#  define GLM_FORCE_CUDA
#endif
#define GLM_FORCE_RADIANS
#include <glm/glm.hpp>
#include <glm/gtc/type_ptr.hpp>
#include <glm/gtc/matrix_transform.hpp>
#include <glm/gtx/transform.hpp>

typedef float GLfloat;

int main(int argc, char** argv)
{
    GLfloat angle_x = 0.0f;
    GLfloat angle_y = 0.0f;
    GLfloat distance = 50.0f;
    glm::vec3 position (0.0f,  0.0f, distance);

    const GLfloat degrees_to_radians_f = 3.14159265359f / 180.0f;

    glm::mat4 view;
    {
        glm::vec3 direction(0.0f,  0.0f, -1.0f);
        glm::vec3 right    (1.0f,  0.0f,  0.0f);

        view = glm::lookAt(position, position+direction, glm::cross(right, direction));
    }

    glm::mat4 model;
    {
        model =   glm::rotate(glm::mat4(1.0f), angle_x * degrees_to_radians_f, glm::vec3(1.0f, 0.0f, 0.0f))
        /**/    * glm::rotate(glm::mat4(1.0f), angle_y * degrees_to_radians_f, glm::vec3(0.0f, 1.0f, 0.0f));
    }

    return 0;
}

The warnings are of the form:

glm/glm/detail/type_mat2x3.inl(375): warning: a __host__ function("glm::detail::operator+") redeclared with __host__ __device__, hence treated as a __host__ __device__ function

These warnings have existed for many versions of glm, but now that I have upgraded to glm 0.9.5.2, errors have begun to appear:

glm/glm/gtc/matrix_transform.inl(431): error: expected a field name
glm/glm/gtc/matrix_transform.inl(432): error: expected a field name
glm/glm/gtc/matrix_transform.inl(85): error: expected a field name

[Groovounet]

I found some issues and I removed all errors and warnings from GLM 0.9.5 branch when using CUDA (tessted with 6rc).

This will be available with GLM 0.9.5.3 release.

By the way with your code sample you probably don't need:
"#ifdef CUDACC

include <cuda.h>

define GLM_FORCE_CUDA

#endif"

Just including GLM should be enough.

Thanks,
Christophe