Accelerate SVE128 SBGEMM/BGEMM

[fadara01]

This accelerates SBGEMM/BGEMM by extending the existing 8x4 kernel to 8x8 (unrolling N by 8)

Not sure if it's a good idea to delete the previous 8x4 kernel?

Here are the speedups on single core Neoverse-V2 (SVE128) compared to prev state:

  M=N=K=64: SBGEMM 1.164x (16.42%), BGEMM 1.133x (13.30%)
  M=N=K=128: SBGEMM 1.220x (22.02%), BGEMM 1.186x (18.56%)
  M=N=K=256: SBGEMM 1.241x (24.08%), BGEMM 1.235x (23.54%)
  M=N=K=512: SBGEMM 1.240x (23.95%), BGEMM 1.227x (22.75%)
  M=N=K=1024: SBGEMM 1.251x (25.11%), BGEMM 1.232x (23.23%)
  M=N=K=2048: SBGEMM 1.235x (23.47%), BGEMM 1.246x (24.64%)

and here are the speedups for the same benchmark on Neoverse-N2:

  M=N=K=64: SBGEMM 1.019x (1.93%), BGEMM 1.055x (5.49%)
  M=N=K=128: SBGEMM 1.030x (3.02%), BGEMM 1.053x (5.31%)
  M=N=K=256: SBGEMM 1.129x (12.90%), BGEMM 1.121x (12.06%)
  M=N=K=512: SBGEMM 1.143x (14.28%), BGEMM 1.132x (13.25%)
  M=N=K=1024: SBGEMM 1.144x (14.41%), BGEMM 1.137x (13.69%)

[fadara01]

Hi @martin-frbg - could you please have a look?

(This currently copies the 8x4 kernel and extends it to 8x8 - please let me know if it's a good idea to remove the 8x4 kernel)

[martin-frbg]

Looks good to me, thanks. And I still think it makes sense to leave older kernels around, even if nothing uses them at the moment - they might still turn out to be adequate (or at least provide inspiration for kernels) for other cpus (or data types) that do not benefit from the latest enhancement or unrolling pattern

[fadara01]

thanks for reviewing!
it would be great if we could get this merged before the next release OpenBLAS release for us to pick it up in PyTorch