diff options
author | Joseph Myers <joseph@codesourcery.com> | 2018-12-19 22:55:32 +0000 |
---|---|---|
committer | Joseph Myers <joseph@codesourcery.com> | 2018-12-19 22:55:32 +0000 |
commit | da75c1b180f9355a8b2b2584ecf1fcfe03f7107e (patch) | |
tree | ada1ed2161fc346c6b79e24d1fb8fece01cacedb /manual/examples/testopt.c | |
parent | 515f463f52b8e4f048e6818d0db2dd2ae45e5f36 (diff) | |
download | glibc-da75c1b180f9355a8b2b2584ecf1fcfe03f7107e.tar glibc-da75c1b180f9355a8b2b2584ecf1fcfe03f7107e.tar.gz glibc-da75c1b180f9355a8b2b2584ecf1fcfe03f7107e.tar.bz2 glibc-da75c1b180f9355a8b2b2584ecf1fcfe03f7107e.zip |
Remove x86 mathinline.h.
After previous cleanups, the only code in the x86 bits/mathinline.h
that is relevant with current compilers is the inline of
__ieee754_atan2l that is conditional on __LIBC_INTERNAL_MATH_INLINES
(i.e. for when libm itself is being built).
This inline is something that does belong in glibc not GCC, since
__ieee754_atan2l is a purely internal function name. This patch moves
that inline to a new sysdeps/x86/fpu/math_private.h, removing the
bits/mathinline.h header.
Note that previously the inline was only for non-SSE 32-bit x86. That
condition does not make sense, however, for a long double function; if
it's not inlined, exactly the same x87 instruction will end up getting
used by the out-of-line function, for both 32-bit and 64-bit. So that
condition is not retained in the new version.
Tested for x86_64 and x86. As expected, installed stripped shared
libraries are unchanged for 32-bit x86, but installed stripped libm.so
is changed for x86_64 because calls to __ieee754_atan2l start being
inlined where previously they were out of line calls. (The same
change to start inlining the function would presumably also apply for
32-bit built with -mfpmath=sse, but that's not a configuration I've
tested.)
* sysdeps/x86/fpu/math_private.h: New file.
* sysdeps/x86/fpu/bits/mathinline.h: Remove.
Diffstat (limited to 'manual/examples/testopt.c')
0 files changed, 0 insertions, 0 deletions