Closed mrakgr closed 8 years ago
I'm currently very busy with other projects... Once I'm done with the Cuda 8 update I will have a look at it! Thanks for the contribution, Michael
I merged it to master and also changed the CudaBLAS comments for AMIN/AMAX. Please check it... If everything is fine, I'll update the NuGet package.
Thanks, Michael
It might be partly my fault since I had difficulties with cloning the repo, but it would be good if you could undo the changes on the comments in the latest commit as now it says Fortran notation instead of the C notation. I fixed that 4 months ago and it does not seem to have been applied and then I lost the changes in the latest commit it seems. I had it right in the old one. For comparison, this is what you have currently.
Edit: Looking at previous commits, I see that it was correct before the latest. I have no idea why I thought the changes weren't accepted. My bad.
Now I'm confused: you say the actual indexing is C-style 0 based, right?
I'm wondering, as e.g. on page 22 in the CUBLAS manual (chapter 2.5.2. cublasI
CudaBlas blas = new CudaBlas(PointerMode.Host);
float[] data = new float[] { 2, 20, 5, 1, 10 };
CudaDeviceVariable<float> d_data = data;
int res = blas.Min(d_data, 1 ); //res = 4
MessageBox.Show(res.ToString());
blas.Dispose();
//again but with device pointers...
blas = new CudaBlas(PointerMode.Device);
CudaDeviceVariable<int> d_res = new CudaDeviceVariable<int>(1);
blas.Min(d_data, 1, d_res);
res = d_res; //res = 4
MessageBox.Show(res.ToString());
reveals: FORTRAN-indexing starting at 1, isn't it?
Er, yes it does. Memory is a fickle thing. Once again, sorry about that.
I made the OpTensor functions for the cuDNN wrapper. I should have done this 4 month ago, but I missed it by accident it seems. Well, here it is now.
Also I've noticed you did not accept the changes to the descriptions of amin, amax and such. The ones about the C vs Fortran notation. It should be C indexing notation by default. You can find the correct description in one of the previous commits on my fork.