CLI improvement to TensorParallelism/02_matmul_tp/matmul_tp_big.cu

[dkennetzoracle]

The code could benefit from the following changes:

A CLI with the following abstractions:

#define M 4096  // Rows of A, Rows of C
#define N 8192  // Columns of B, Columns of C
#define K 1024  // Columns of A, Rows of B
#define NGPUS 2 // Number of GPUs to use for computation

• variable sizes of M, N, K
• number of GPUs to use
• error checking for matrix dimensions from CLI values
• device detection to determine if number of GPUs selected is supported (IE 4 is picked, but you only have 2)
• boolean of whether to run host matmul or not, and if not - don't assert correctness or measure cpu timing. Should be pulled into a function.