public inbox for gas2@sourceware.org
 help / color / mirror / Atom feed
From: Alan Modra <alan@spri.levels.unisa.edu.au>
To: gas2@cygnus.com, egcs@cygnus.com, gcc2@cygnus.com
Subject: gcc generates invalid i386 fmulp
Date: Tue, 14 Apr 1998 00:15:00 -0000	[thread overview]
Message-ID: <199804131501.AAA05109@mullet.Levels.UniSA.Edu.Au> (raw)

Hello all,
  I've been doing a little work on gas for i386, tidying up a few
things and adding reasonably strict operand/suffix checking.  The new
gas has already found a couple of errors in the linux kernel asm, one
in glibc1, and a problem with gcc.

For anyone who's interested, the gas patches can be found at
ftp://mullet.levels.unisa.edu.au/private/teabag/gas-980407-alan.diff.gz
They should apply to binutils-2.9 without problems too.

The gcc problem (found by mikeg@weiden.de (Michael L. Galbraith)),
when compiling glibc2 is that gcc will generate instructions like
	`fsubp %st(1),%st'

Now there's no instruction to do %st <- %st - %st(1) and pop, and it
doesn't make sense to pop the result anyway.  Older versions of gas
assemble this instruction to 0xdee9, which does
%st(1) <- %st(1) - %st, pop, ie. `fsubp %st,%st(1)'  (Note!! Because
of the broken UnixWare compatibility requirement for gas, you need to
write `fsubrp %st,%st(1)'  See the comments I put at the top of
binutils/include/opcode/i386.h)


	* config/i386/i386.c (output_387_binary_op): Swap operands
	when popping if the result is st(0), to prevent invalid
	instructions like `fmulp %st(1),%st' and `fsubp %st(1),%st'

--- gcc-2.8.1/config/i386/i386.c	Thu Dec  4 00:31:23 1997
+++ gcc/config/i386/i386.c	Tue Apr 14 00:00:43 1998
@@ -3827,9 +3827,12 @@ output_387_binary_op (insn, operands)
 	  return "";
 	}
 
       if (find_regno_note (insn, REG_DEAD, REGNO (operands[2])))
-	return strcat (buf, AS2 (p,%2,%0));
+	if (STACK_TOP_P (operands[0]))
+	  return strcat (buf, AS2 (p,%0,%2));
+	else
+	  return strcat (buf, AS2 (p,%2,%0));
 
       if (STACK_TOP_P (operands[0]))
 	return strcat (buf, AS2C (%y2,%0));
       else
@@ -3858,12 +3861,18 @@ output_387_binary_op (insn, operands)
       if (! STACK_REG_P (operands[1]) || ! STACK_REG_P (operands[2]))
 	abort ();
 
       if (find_regno_note (insn, REG_DEAD, REGNO (operands[2])))
-	return strcat (buf, AS2 (rp,%2,%0));
+	if (STACK_TOP_P (operands[0]))
+	  return strcat (buf, AS2 (p,%0,%2));
+	else
+	  return strcat (buf, AS2 (rp,%2,%0));
 
       if (find_regno_note (insn, REG_DEAD, REGNO (operands[1])))
-	return strcat (buf, AS2 (p,%1,%0));
+	if (STACK_TOP_P (operands[0]))
+	  return strcat (buf, AS2 (rp,%0,%1));
+	else
+	  return strcat (buf, AS2 (p,%1,%0));
 
       if (STACK_TOP_P (operands[0]))
 	{
 	  if (STACK_TOP_P (operands[1]))

             reply	other threads:[~1998-04-14  0:15 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
1998-04-14  0:15 Alan Modra [this message]
1998-04-14  8:19 ` H.J. Lu
1998-04-14 16:45   ` Alan Modra
1998-04-14 20:18     ` Michael L. Galbraith
1998-04-14 20:30       ` H.J. Lu
1998-04-14 21:35         ` Michael L. Galbraith
1998-04-14 20:44       ` H.J. Lu
1998-04-15  7:18   ` Alan Modra
1998-04-15  9:01     ` Robert Lipe
1998-04-15 10:05       ` Ulrich Drepper

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=199804131501.AAA05109@mullet.Levels.UniSA.Edu.Au \
    --to=alan@spri.levels.unisa.edu.au \
    --cc=egcs@cygnus.com \
    --cc=gas2@cygnus.com \
    --cc=gcc2@cygnus.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).