Texture Memory in Matrix Multiplication Did anyone tried to use Texture Memory for Matrix Multiplica

[fcsc]

I was just wondering if anyone had ever tried that (I’m almost sure somebody tried, there is people for everything in this world) and if I could get some input.

I’m starting in this CUDA world and I just wanna use my graphic card as an amazing group of multiprocessors but I’m not so much into images. This means that if It is not possible to use the texture memory to store a matrix then I will put memory texture very down in my “To learn” list.

I’ve been reading and it looks like it could be done if you could avoid the interpolation step with the neighbours (actually when you ask for the value at i,j you want that value not some sort of interpolation with the adjacent cells).

If you read this, thanks and sorry for the trouble. If you answer this: Thank you so much!
FM.

[fcsc]

I just find out a very good example in the SDK, the dct8x8 where this is done.

If somebody had the same question I did just go and looked yourself is very self-explanatory!!

Thanks anyway!

I was just wondering if anyone had ever tried that (I’m almost sure somebody tried, there is people for everything in this world) and if I could get some input.

I’m starting in this CUDA world and I just wanna use my graphic card as an amazing group of multiprocessors but I’m not so much into images. This means that if It is not possible to use the texture memory to store a matrix then I will put memory texture very down in my “To learn” list.

I’ve been reading and it looks like it could be done if you could avoid the interpolation step with the neighbours (actually when you ask for the value at i,j you want that value not some sort of interpolation with the adjacent cells).

If you read this, thanks and sorry for the trouble. If you answer this: Thank you so much!

FM.