Vladislav Shchapov
Vladislav Shchapov
@Dead2 1) PR #1684 - small fix for #1662. 2) PR #1685 - split headers.
~~Rebase required.~~ ~~Waiting PR #1673 (NMake)~~ ~~Waiting #1681, #1682 (deflate optimizations).~~
Ready for review.
Rebased onto develop
Architecture specific functions from https://github.com/zlib-ng/zlib-ng/blob/develop/insert_string_tpl.h is very small. The overhead of calling them (``update_hash``) is greater than the benefits of optimizations. I think makes sense to create new architecture specific...
> I agree that we should make sure compiler can inline calls to `update_hash` variants, so the speed gain is not lost to overhead of calling the function. Call by...
Does ``memcpy`` from ``pointer`` to ``uintptr_t`` also triggered this GCC error or not?
Try removing ``Z_TLS`` from the ``functable`` definition to avoid tls overhead. This should be safe to execute on x86(_64), but may generate ThreadSanitizer warnings.
Which is greater: TLS overhead or overhead of calling a function by pointer?
If the main overhead is in TLS, then I have one idea on how to remove it for x86 and arm.