Archives Discussions

[ryta1203]

Ok, so built code using float4s in not using the GPU (no kernel calls) and it works completely. I then attempted to port the same code over to the GPU and I get absolutely nothing (just junk numbers, so the program ends early) in CAL mode and off results in CPU mode. I am hoping someone can lend me a hand, I'm not sure if it is the kernel that is bad or something in my wrapper function, so I am including them both (the entire .br file):

[ryta1203]

So, to more specific, do you think it's a problem that I am using a wrapper function and handling the arrays this way? In my main cpp file I am using the F arrays in further calculation.

Like I said, I have this same code in CPU (using float4s) except the kernel is just another function and uses two for loops to cover the domain.

[ryta1203]

So has anyone else experienced any problems with calling functions inside the .br file from outside the .br file and using arrays in this fashion?

[michael_chu]

Hi ryta1203, can you send a complete test program to streamdeveloper@amd.com and I'll pass it to the Brook+ team?

Thanks!

Michael.

[ryta1203]

Michael,

Thanks, this actually got corrected in another thread. It seems that you have to manually initialize every element in a stream, otherwise you just get garbage for uninitialized elements.

[michael_chu]

Thanks for letting my know ryta1203! 🙂

Sorry for being out of sync! There is quite a bit of activity on the forums and when you are like me trying to catchup and answer forum questions, its too easy to miss a thread.

Michael.