Path: utzoo!telly!philmtl!uunet!tut.cis.ohio-state.edu!CS.WISC.EDU!larus
From: larus@CS.WISC.EDU (James Larus)
Newsgroups: gnu.gcc.bug
Subject: Unnecessary floating point moves
Message-ID: <8912131621.AA19808@primost.cs.wisc.edu>
Date: 13 Dec 89 16:21:16 GMT
Sender: daemon@tut.cis.ohio-state.edu
Reply-To: larus@cs.wisc.edu
Distribution: gnu
Organization: GNUs Not Usenet
Lines: 25


gcc version 1.36 on DECstation 3100 (MIPS R2000) compiles the
following function:

float
foo (s)
     char *s;
{
  double atof();

  return atof(s);
}

into the following code:

	addiu	$29,$29,0xfff0	#subsi3	$29,16 -> $29
	jal	atof	# call  atof  regle 2-call (VOIDmode)
	mov.d	$f4,$f0	#movdf $f0 -> $f4 
	cvt.s.d	$f4,$f4	#truncdfsf2	 $f4 -> $f4
	mov.s $f0,$f4	#movsf $f4 -> $f0 
	addiu	$29,$29,0x10	#addsi3	$29,16 -> $29

The move to $f4 to truncate and then back to $f0 is unncessary.

/Jim