This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Polyhedron performance regression

Paul Richard Thomas wrote:

I bet we are missing an inline x86_64 variant.  But it should be not too
hard to implement.


'tis what I thought.

My view is that Steve is right about right and slow; however it is a
very visible right and slow.  Which is the right button to press to
have the inline version crafted for us?

fmod{d,s,x}f3 intriniscs are constrainted by:

 "TARGET_USE_FANCY_MATH_387
  && (!(TARGET_SSE2 && TARGET_SSE_MATH) || TARGET_MIX_SSE_I387)"

as they apply to x87 math only. The problem here are not-yet-implemented SSE versions (x86_64 defaults to SSE) of fmod{d,s}f3 named pattern. In this case, libcall is emitted, and calls to math library on x86_64 are quite non-optimal. When expander for SSE version is provided, the x86_64 version will fly again.

HTH,
Uros.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]