This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC 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: Still fldl/fstpl use with fpmath=sse?

From: Uros Bizjak <uros at kss-loka dot si>
To: rguenth at tat dot physik dot uni-tuebingen dot de
Cc: gcc at gcc dot gnu dot org
Date: Mon, 10 Jan 2005 17:20:05 +0100
Subject: Re: Still fldl/fstpl use with fpmath=sse?

Hello!

Use of x87 has improved considerably with -mfpmath=sse, but
I still see a lot of double moves going through the fp stack,
like:

.L8350:
       fldl    (%esi)
       fstpl   -40(%ebp)

These are generated by FP compares. Please read http://gcc.gnu.org/ml/gcc-patches/2005-01/msg00394.html

I tried to separate x87 and SSE compares, but they are so unbeliveably interwoven together that I lost my will to fix them... There is a mixture of x87 and SSE compare patterns, min/max patterns and especially SSE and x87 cmove patterns that somehow interfere and finally produce a mixture of SSE and x87 instructions. You can try to add -mno-80387, but this can (and will) produce an ICE (PR 19307).

Uros.

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