undefined | Better HN

0 pointsbjourne2y ago0 comments

The compilation command errors out for me:

/home/bjourne/p/laser/benchmarks/gemm/gemm_bench_float32.nim(77, 8) Warning: use `std/os` instead; ospaths is deprecated [Deprecated] /home/bjourne/p/laser/benchmarks/gemm/gemm_bench_float32.nim(101, 8) template/generic instantiation of `bench` from here /home/bjourne/p/laser/benchmarks/gemm/gemm_bench_float32.nim(106, 21) template/generic instantiation of `gemm_nn_fallback` from here /home/bjourne/p/laser/benchmarks/gemm/arraymancer/blas_l3_gemm.nim(85, 34) template/generic instantiation of `newBlasBuffer` from here /home/bjourne/p/laser/benchmarks/gemm/arraymancer/blas_l3_gemm_data_structure.nim(30, 6) Error: signature for '=destroy' must be proc[T: object](x: var T) or proc[T: object](x: T)

Anyway the reason for your competitive performance is likely that you are benchmarking with very small matrices. OpenBLAS spends some time preprocessing the tiles which doesn't really pay off until they become really huge.

0 comments

2 comments · 1 top-level

mratsim2y ago· 1 in thread

Ah,

It was from an older implementation that wasn't compatible with Nim v2. I've commented it out.

If you pull again it should work.

> Anyway the reason for your competitive performance is likely that you are benchmarking with very small matrices. OpenBLAS spends some time preprocessing the tiles which doesn't really pay off until they become really huge.

I don't get why you think it's impossible to reach BLAS speed. The matrix sizes are configured here: https://github.com/mratsim/laser/blob/master/benchmarks/gemm...

It defaults to 1920x1920 * 1920x1920. Note, if you activate the benchmarks versus PyTorch Glow, in the past it didn't support non-multiple of 16 or something, not sure today.

Packing is done here: https://github.com/mratsim/laser/blob/master/laser/primitive...

And it also support pre-packing which is useful to reimplement batch_matmul like what CuBLAS provides and is quite useful for convolution via matmul.

bjourneOP2y ago

Ok, I will benchmark this more when I have time. My gut feeling is that it is impossible to reach OpenBLAS-like performance without carefully tuned kernels and without explicit SIMD code. Clearly, it's not impossible to be as fast as OpenBLAS otherwise OpenBLAS itself woudln't be that fast, but it is very difficult and takes a lot of work. There is a reason much of OpenBLAS is implemented in assembly and not C.

j / k navigate · click thread line to collapse