Description

This is a small collection of toy HPC problems, mainly focused as both self-exercise and practicing the use of high performance numerical libraries such as BLAS, LAPACK, and NVIDIA Performance Libraries (NVPL).

The source code is typed with experimentation in mind, thus the implementations will not be 100% correct but follows how the official NVPL sample code (https://github.com/NVIDIA/NVPLSamples) was created.

This repository does not accept any pull requests. Any requests would be ignored.

Dependencies and How to Use

Currently, this repository assumes that the user owns or have access to:

1. Nvidia Grace Superchip w/ 240GB memory, and
2. nvhpc/25.5 module (for nvc compiler).

The code does not have any other dependencies besides NVPL.

How to use:

1. Clone the repository:

   https://github.com/accable/hpc_toy.git

2. Navigate to the project directory:

   cd hpc_toy
   cd src

3. To compile the .c files, first load the nvhpc module:

   module load nvhpc/25.5

   Then compile the files using the nvc compiler:

   nvc 1_1_heat_diffusion_dynmem.c -o foo

   If BLAS/LAPACK is required, simply add -lblas or -llapack on the compiler arguments:

   nvc 1_2_heat_diffusion_dynmem_lapack.c -o foolapack -llapack

   Since some of the files require nvplTENSOR, add -lnvpl_tensor on the compiler arguments:

   nvc 2_0_multi_head_attention_cutensor.c -o footensor -lnvpl_tensor

   Some of the files also require specific way of compiling, add -Mnvpl on the compiler arguments:

   nvc 2_1_multi_head_attention_blas.c -o fooblas -Mnvpl=blas

4. Run the application:

   ./foo