This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 09/10] i386: Replace assembly versions of e_log2f with generic e_log2f.c
- From: Joseph Myers <joseph at codesourcery dot com>
- To: Szabolcs Nagy <szabolcs dot nagy at arm dot com>
- Cc: "H.J. Lu" <hjl dot tools at gmail dot com>, GNU C Library <libc-alpha at sourceware dot org>, <nd at arm dot com>
- Date: Fri, 20 Oct 2017 16:04:01 +0000
- Subject: Re: [PATCH 09/10] i386: Replace assembly versions of e_log2f with generic e_log2f.c
- Authentication-results: sourceware.org; auth=none
- References: <20171019173159.21402-1-hjl.tools@gmail.com> <20171019173159.21402-10-hjl.tools@gmail.com> <CAMe9rOqwmBj8zFd1FiVqQR=+sOz8qQwz3RhnDpnmPEZxiw290Q@mail.gmail.com> <59EA0E54.5080204@arm.com>
On Fri, 20 Oct 2017, Szabolcs Nagy wrote:
> is this comparing x87 c code with x87 asm?
> or did the toolchain have -fpmath=sse?
>
> i would not expect that much speedup on i386
My guess would be: the processor's implementation of the fyl2xp1 and fyl2x
instructions may use a polynomial approximation internally, but one
producing a result accurate to 64-bit precision, so it's plausible to get
significant speedups by using an approximation that only needs to be
accurate to 24 bits.
--
Joseph S. Myers
joseph@codesourcery.com