CUBLAS GEMM Example

A CUBLAS Matrix Multiply (GEMM) example.

Requirements:

- Nvidia GPU supporting CUDA
- CUDA v11.0 or greater
- CUBLAS v11.0 (should come with CUDA)
- openblas (max-perf CPU test)

Install and Compile:

a) Clone Repo:
    git clone https://github.com/Linus-Voss/cublas-gemm.git

b) Compile:
    cd cublas-gemm
    make

c) Data Types:
    You can specify the data type (half, float) for each matrix
    Example:
    make ATYPE=half BTYPE=half CTYPE=half CPTYPE=half

Run:

a) Run:
        run as ./prog dev nt n comptype mode
        
        dev:      Device ID
        nt:       Number of CPU threads (accelerates data init and CPU mode)
        m:        Matrix size of M
        n:        Matrix size of N
        k:        Matrix size of K
        comptype: GPU CUBLAS mode (should be same witch CPTYPE)
        mode:     CPU=0,  GPU=1
        loop:     GPU CUBLAS loop kernel

b) CUBLAS Compute Types:
        0  = CUBLAS_COMPUTE_16F
        1  = CUBLAS_COMPUTE_16F_PEDANTIC
        2  = CUBLAS_COMPUTE_32F
        3  = CUBLAS_COMPUTE_32F_PEDANTIC
        4  = CUBLAS_COMPUTE_32F_FAST_16F
        5  = CUBLAS_COMPUTE_32F_FAST_16BF
        6  = CUBLAS_COMPUTE_32F_FAST_TF32
        7  = CUBLAS_COMPUTE_64F
        8  = CUBLAS_COMPUTE_64F_PEDANTIC
        9  = CUBLAS_COMPUTE_32I
        10 = CUBLAS_COMPUTE_32I_PEDANTIC

Example executions:

a) make ATYPE=float BTYPE=float CTYPE=float CPTYPE=float
    ./prog 0 4 8192 8192 8192 2 1 10000

b) make ATYPE=half BTYPE=half CTYPE=half CPTYPE=half
    ./prog 0 4 8192 8192 8192 0 1 10000

c) make ATYPE=half BTYPE=half CTYPE=half CPTYPE=float
    ./prog 0 4 8192 8192 8192 0 2 10000


d) [CPU CBLAS] FP32 Using 8 CPU threads 
    make
./prog 0 8 8192 8192 8192 0 0 

e) [CPU CBLAS] FP64 Using 8 CPU threads 
    make CPUFP64=CPUFP64
   ./prog 0 8 8192 8192 8192 0 0