Daniel Lemire
Daniel Lemire
Woohoo ! It is fixed.
I am going to close this issue for now.
Thanks @MarkPflug
Just so we are clear, I am thinking about working at the bit level. So this is prior to the `flatten` call. Yes, it is exotic.
The use case for `__declspec(noinline)` are functions that are small (so likely inlined) but rarely used. It does not make intuitive sense that it would help performance. In simdjson, we...
> something's wrong in their inlining. Though we have not worked much on simdcsv, we have worked much with simdjson and Visual Studio and I have found that, indeed, something...
> it is find_quote_mask + cmp_mask_against_input that I need to un-inline Or try manually inlining them!
> Those people seem to have the same kind of problems: https://github.com/xtensor-stack/xsimd/issues/617 I think their problem is too little inlining.
> fix manually unrolled loops in flatten_bits() which generate suboptimal code I think you are saying that the compiler under some setting unrolls better than manually unrolled code. I do...
@kelas Assuming that you have access to Linux or an ARM-based Apple device, can you do the following and report back on the output ? (The `sudo` may not be...