Tried to build this code on Visual Studio, lots of compile problems, any suggestions?

[JacquesRoth]

a_frag, b_frag, and c_frag declarations give undefined errors, adding "using namespace nvcuda;" did not help. Adding "include <mma.h> did not help. Is there something that must be configured in Visual Studio to enable tensor core operations?
Invocation of naive_tensor_mat_mul_kernel<<<dim_grid... gives expected an expression error.
"int warpM = blockIdx.x;" gives blockIdx undefined error!

Quite a few other errors. Has anyone tried this code with Visual Studio for Cuda?

[tgautam03]

It could be a driver issue (assuming your GPU has tensor cores). Did you compile it using the make file I provided?

Try compiling using the following commands:

make 00_benchmark_cuBLAS.out

make 01_benchmark_naive.out

[JacquesRoth]

No, I am using Visual Studio and I don't know if there is some way to run your make file. I am having the same issues with other code that invokes the tensor core functionality, just will not build without many errors.

…

On 12/26/2024 11:50 PM PST Tushar Gautam ***@***.***> wrote: It could be a driver issue (assuming your GPU has tensor cores). Did you compile it using the make file I provided? — Reply to this email directly, view it on GitHub #1 (comment), or unsubscribe https://github.com/notifications/unsubscribe-auth/AGWPKZKMELZSB3ZEC2ZJLJL2HUBC7AVCNFSM6AAAAABUFKKFC2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNRTGQZDKMZUHA. You are receiving this because you authored the thread.Message ID: ***@***.***>

[tgautam03]

I left two commands in the previous comment. Please try them out in the terminal (ensuring that the terminal is opened in tgemm folder) and see if that works.