Fangrui Song

I've written a compression program benchmark program https://gist.github.com/MaskRay/74cdaa83c1f44ee105fcebcdff0ba9a7 To work around the lack of a release, I just download the whole repo as zip :) Since there are build issues...

Thanks! This gets rid of the conditions that are very difficult to get right: ``` /* * Detect if the architecture is fine with unaligned access. */ #if !defined(BLOSC_STRICT_ALIGN) #define...

`memcpy(dst, src, 4);` has been optimized by GCC for 10+ years (check godbolt 4.1 from 2007), if not 20+/30+ years. (I think MSVC should have had the optimization for 10+...

I can reproduce `bench/b2bench blosclz shuffle single 1 8388608 4 19` slowdown with this patch, but only with GCC, not Clang. I create two build directories for main (out/gcc0) and...

If it helps, creating two patches may be feasible. One is to replace things like `*(uint64_t*)out = *(uint64_t*)from;` with `memcpy`. This is always fine for optimizations and usually evens leads...

Do you have a language client that needs this?

Thanks for the explanation. > ccls is really usable by 2 because of the existing extension/package which does a restart (and only one is usable without having to write extra...

Feel free to force push to this PR and drop my commit after the change is merged into yours. I don't mind:) I am still not clear why the client...

I download prebuilt Clang packages on https://releases.llvm.org/download.html then set CMAKE_PREFIX_PATH to the root directory of the extracted tarball. ``Clang_EXECUTABLE` will respect that. `clang-${Clang_VERSION_MAJOR}` will not work if the distribution only...

> Is there a way to obtain a callgraph in standalone mode? If you implement a minimum client to calls initialize and didOpen, I think it is doable... You might...