[Mono-list] Integer division problems

[Martin Baulig]

--=-=-=

Dietmar Maurer <dietmar@ximian.com> writes:

> No, there is something wrong! I use CDQ to sign extend to 64bit in EAX/EDX,
> so I assume your code is right for DIV_UN (althought you should remove that
> unnecessary CDQ then), but DIV should work the way it is?

Yes, I was confused. For DIV_UN, the CDQ is wrong since it sets EDX to -1 if
EAX is "negative", but we're performing an unsigned division.

DIV works the way it is and does not any longer with my patch.

Here's the corrected patch:

====
2002-03-09  Martin Baulig  <martin@gnome.org>

	* x86.brg (DIV_UN): Clear EDX before performing the division and remove
	the unnecessary CDQ instruction.

=====


--=-=-=
Content-Disposition: attachment; filename=dietmar-diff.txt

Index: x86.brg
===================================================================
RCS file: /cvs/public/mono/mono/jit/x86.brg,v
retrieving revision 1.92
diff -u -u -r1.92 x86.brg
--- x86.brg	2002/03/08 06:08:46	1.92
+++ x86.brg	2002/03/09 21:13:31
@@ -816,7 +816,7 @@
 	if (tree->left->reg1 != X86_EAX)
 		x86_mov_reg_reg (s->code, X86_EAX, tree->left->reg1, 4);
 
-	x86_cdq (s->code);
+	x86_mov_reg_imm (s->code, X86_EDX, 0);
 	x86_div_reg (s->code, tree->right->reg1, FALSE);
 
 	mono_assert (tree->reg1 == X86_EAX &&
Index: ChangeLog
===================================================================
RCS file: /cvs/public/mono/mono/jit/ChangeLog,v
retrieving revision 1.121
diff -u -u -r1.121 ChangeLog
--- ChangeLog	2002/03/08 06:08:46	1.121
+++ ChangeLog	2002/03/09 21:13:31
@@ -1,3 +1,7 @@
+2002-03-09  Martin Baulig  <martin@gnome.org>
+
+	* x86.brg (DIV_UN): Clear EDX before performing the division.
+
 2002-03-07  Dietmar Maurer  <dietmar@ximian.com>
 
 	* jit.c (mono_analyze_stack): only load addresses of value types

--=-=-=


Ok this time ?


-- 
Martin Baulig
martin@gnome.org

--=-=-=--