MRF issueshttps://gitlab.inria.fr/pacanows/MRF/-/issues2020-11-04T20:45:28+01:00https://gitlab.inria.fr/pacanows/MRF/-/issues/77Slow loading of big assets2020-11-04T20:45:28+01:00MURRAY DavidSlow loading of big assetsMentioned by Arthur, loading huge asset is very slow.
It seems to come from an include in cuda material kernels. We need more info from him.Mentioned by Arthur, loading huge asset is very slow.
It seems to come from an include in cuda material kernels. We need more info from him.https://gitlab.inria.fr/pacanows/MRF/-/issues/129Cache mechanism for CUDA kernels2021-02-25T09:36:51+01:00PACANOWSKI RomainCache mechanism for CUDA kernelsMalia 1.2PACANOWSKI RomainPACANOWSKI Romainhttps://gitlab.inria.fr/pacanows/MRF/-/issues/148Faster ColorImage2020-11-04T20:48:26+01:00PACANOWSKI RomainFaster ColorImageBy using an Eigen version of Vec3f
like this is done for Array2D
provide an Eigen version of it if Eigen3 is availableBy using an Eigen version of Vec3f
like this is done for Array2D
provide an Eigen version of it if Eigen3 is availablehttps://gitlab.inria.fr/pacanows/MRF/-/issues/188refactor path_ray_data2021-02-22T15:31:49+01:00MURRAY Davidrefactor path_ray_dataCurrently the PRD contains some data that is unused/may be optimized:
```cpp
struct PerRayData_pathtrace
{
mrf::spectrum::COLOR attenuation;
mrf::spectrum::COLOR result;
optix::float3 origin;
optix::float3 direction;
#ifdef MRF...Currently the PRD contains some data that is unused/may be optimized:
```cpp
struct PerRayData_pathtrace
{
mrf::spectrum::COLOR attenuation;
mrf::spectrum::COLOR result;
optix::float3 origin;
optix::float3 direction;
#ifdef MRF_RENDERING_MODE_SPECTRAL_MULTIPLEXED
float wavelength_offset;
#endif
mrf::spectrum::COLOR_B ior; // **CAN BE REMOVED (replaced by an int) USING AN ALL-IOR BUFFER**
int id_wavelength;
int backface; // **Unused**
unsigned int seed;
unsigned int num_sample;
optix::float2 offset_sampling; //cranley patterson
int depth;
bool countEmitted;
bool done;
bool is_inside;
int shadow_catcher;
};
```
The `mrf::spectrum::COLOR_B ior;` is heavy and not really fully used. A better solution is to build a buffer containing all IORs (eta+kappa) and only store a indexed reference in the PRD. That implies additionnal "permanent" storage for the buffer but grealty reduces the memory bandwith usage for the PRD, which all thing considered, should significantly make the additionnal storage worth it.
Using such a buffer would also be useful for medium tracking, to store index of previous/futur medium in a memory friendly way.Malia 1.1.0MURRAY DavidMURRAY David