About the texture

[Quoc_Vinh]

This is the first time I use texture, I read the “Programming_guide” from 1.0 to now 2.0 beta2.
but I can not found any answer for my question is
“what is the difference between low-level and high-level texture runtime API”.
if anyone knows, please tell me.
thank for advance. :)

[Quoc_Vinh]

when I read the example code in the programming guide

the different between high-level and low-level in texture is

In low-level we can configure detail of the texture, but in high-level we can not.

for example bind a texture reference to linear memory in low-level

texture<float, 1, cudaReadModeElementType> texRef; 

textureReference* texRefPtr; 

cudaGetTextureReference(&texRefPtr, “texRef”); 

cudaChannelFormatDesc channelDesc = 

                             cudaCreateChannelDesc<float>(); 

cudaBindTexture(0, texRefPtr, devPtr, &channelDesc, size);

for example bind a texture reference to linear memory in high-level

texture<float, 1, cudaReadModeElementType> texRef; 

cudaBindTexture(0, texRef, devPtr, size);

I don’t think that is is the unique different. External Image

does anybody help me.

[MisterAnderson42]

The high level API simply automatically uses the textures channel descriptor. You can see the code for yourself at line 261 of cuda_runtime.h

template<class T, int dim, enum cudaTextureReadMode readMode>

__inline__ __host__ cudaError_t cudaBindTexture(

        size_t                           *offset,

  const struct texture<T, dim, readMode> &tex,

  const void                             *devPtr,

        size_t                            size = UINT_MAX

)

{

  return cudaBindTexture(offset, tex, devPtr, tex.channelDesc, size);

}

[Quoc_Vinh]

Thank for your advice. :D
I am trying to understand it clearly.

[anon76797866]

Hello,

i have a question regarding the code mentioned above.

One of the other differences between the low level  and high level runtime apis is that the low level runtime api does not necessarily need to be compiled  with nvcc.

But parts of the High level Api demand the usage of nvcc.

Now,coming to my question…

For my application i am implementing an Image class thru which i plan to allocate memory on the Texture memory.Since it does not require nvcc for the texture management  i chose to use the low level api(ie, “cuda_runtime_api.h”).

But when i use the following code as mentioned in the guide i get a compilation error as the 'texture" datatype  is not declared. But there is no problem with the “textureReference”.

texture<float, 1, cudaReadModeElementType> texRef; 

textureReference* texRefPtr;

These are my observations regarding the problem:

1.The struct “textureReference” is defined in “texture_types.h”.But “texture” is commented out in the following way…

#if defined(__cplusplus) && defined(__CUDACC__)

/*******************************************************************************

*                                                                              *

*                                                                              *

*                                                                              *

*******************************************************************************/

#include "host_defines.h"

#include "channel_descriptor.h"

/*******************************************************************************

*                                                                              *

*                                                                              *

*                                                                              *

*******************************************************************************/

/*TEXTURE_TYPE*/

template<class T, int dim = 1, enum cudaTextureReadMode = cudaReadModeElementType>

struct texture : public textureReference

{

  __host__ texture(int                         norm  = 0,

                   enum cudaTextureFilterMode  fMode = cudaFilterModePoint,

                   enum cudaTextureAddressMode aMode = cudaAddressModeClamp)

  {

    normalized     = norm;

    filterMode     = fMode;

    addressMode[0] = aMode;

    addressMode[1] = aMode;

    addressMode[2] = aMode;

    channelDesc    = cudaCreateChannelDesc<T>();

  }

 __host__ texture(int                          norm,

                   enum cudaTextureFilterMode   fMode,

                   enum cudaTextureAddressMode  aMode,

                   struct cudaChannelFormatDesc desc)

  {

    normalized     = norm;

    filterMode     = fMode;

    addressMode[0] = aMode;

    addressMode[1] = aMode;

    addressMode[2] = aMode;

    channelDesc    = desc;

  }

};

#endif /* __cplusplus && __CUDACC__ */

2.Even in the programming guide (Section 4.5.2.6) it has been mentioned that “texture” is a High level datatype.So i might have to  include “cuda_runtime.h” rt?.But apparently that too doesnt work.

I am developing on windows using the “cl” compiler and the SDK version i am using is the cuda 2.0 beta .

I would be extremely thankful for any suggestions to a possible work around .

Sunil

[snapback]437904[/snapback]