[libcxx-commits] [PATCH] D125958: Use __libcpp_clz for a tighter __log2i function

[Nikolas Klauser via Phabricator via libcxx-commits]

philnik requested changes to this revision.
philnik added a comment.

The CI failue look unrelated, but I can't recall any problems with the AArch64 CI. Just rebase it and try again. I don't think there are any benchmarks, but it would be nice if you could provide some in `libcxx/benchmarks`. I'd also like to see benchmark results before approving.



================
Comment at: libcxx/include/__algorithm/sort.h:505
+  if (sizeof(__n) <= sizeof(unsigned))
+    return sizeof(unsigned) * __CHAR_BIT__ - 1 - __libcpp_clz(static_cast<unsigned>(__n));
+  if (sizeof(__n) <= sizeof(unsigned long))
----------------
nilayvaish wrote:
> I am wondering whether __CHAR_BIT__ is preferred over using CHAR_BIT (available from climit header file).  I don't know what the convention is for writing library code.
I think `CHAR_BIT` is the preferred way. At least that's what I'm using.


================
Comment at: libcxx/include/__algorithm/sort.h:510-516
+
   _Number __log2 = 0;
   while (__n > 1) {
     __log2++;
     __n >>= 1;
   }
   return __log2;
----------------
nilayvaish wrote:
> Instead of retaining this, should we just have a static_assert() on the size of the type being less or equal to that of unsigned long long?
We support `__int128`, so this needs to stay.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D125958/new/

https://reviews.llvm.org/D125958