[committed] libstdc++: Use double for unordered container load factors [PR 96958]

[Jonathan Wakely]

On 31/10/20 00:23 +0000, Jonathan Wakely wrote:
>These calculations were changed to use long double nearly ten years ago
>in order to get more precision than float:
>https://gcc.gnu.org/pipermail/libstdc++/2011-September/036420.html
>
>However, double should be sufficient, whlie being potentially faster
>than long double, and not requiring soft FP calculations for targets
>without native long double support.
>
>libstdc++-v3/ChangeLog:
>
>	PR libstdc++/96958
>	* include/bits/hashtable_policy.h (_Prime_rehash_policy)
>	(_Power2_rehash_policy): Use double instead of long double.
>
>Tested powerpc64le-linux. Committed to trunk.
>
>This doesn't fix the PR, because there are also long double
>calculations in src/c++11/hashtable_c++0x.cc, so another patch is
>needed.

Here's that other patch. This also fixes some failures I was seeing
when mixing -mabi=ieeelongdouble with -mabi=ibmlongdouble in a local
branch for the ieee128 transition work.

Tested powerpc64le-linux. Committed to trunk.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: patch.txt
Type: text/x-patch
Size: 4335 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/libstdc++/attachments/20201031/0983e2bb/attachment.bin>