real_zerop thinks -0 == 0

real_zerop thinks -0 == 0

[Mark Wielaard]

Hi,

The following seems to fix a couple of the remaining Mauve failures (the
mauve tests are first compiled to byte-code).

--- jcf-write.c	12 Mar 2002 10:33:57 -0000	1.99.2.1
+++ jcf-write.c	15 Apr 2002 23:37:46 -0000
@@ -1536,9 +1536,7 @@
       {
 	int prec = TYPE_PRECISION (type) >> 5;
 	RESERVE(1);
-	if (real_zerop (exp))
-	  OP1 (prec == 1 ? OPCODE_fconst_0 : OPCODE_dconst_0);
-	else if (real_onep (exp))
+	if (real_onep (exp))
 	  OP1 (prec == 1 ? OPCODE_fconst_1 : OPCODE_dconst_1);
 	/* FIXME Should also use fconst_2 for 2.0f.
 	   Also, should use iconst_2/ldc followed by i2f/i2d

This isn't a real solution, but I don't know how to correctly test for
positive zero and I really must get some sleep now. Maybe someone more
familiar with this code can suggest the proper solution.

The example code to try is:

public class Z
{
  public static void main(String[] args)
  {
    System.out.println(-0.0f);
  }
}

Compile with gcj -C and run with gij. Output should be -0.0.

Good night,

Mark

Re: real_zerop thinks -0 == 0

[Tom Tromey]

>>>>> "Mark" == Mark Wielaard <mark@klomp.org> writes:

Mark> -	if (real_zerop (exp))
Mark> -	  OP1 (prec == 1 ? OPCODE_fconst_0 : OPCODE_dconst_0);
Mark> -	else if (real_onep (exp))
Mark> +	if (real_onep (exp))

Mark> This isn't a real solution, but I don't know how to correctly
Mark> test for positive zero and I really must get some sleep
Mark> now. Maybe someone more familiar with this code can suggest the
Mark> proper solution.

I don't know if it is the proper solution but in gcc/real.h I see
REAL_VALUE_MINUS_ZERO.  Would it be appropriate to use this?

Tom

Re: real_zerop thinks -0 == 0

[Mark Wielaard]

On Tue, 2002-04-16 at 17:05, Tom Tromey wrote:
> >>>>> "Mark" == Mark Wielaard <mark@klomp.org> writes:
> 
> Mark> -	if (real_zerop (exp))
> Mark> -	  OP1 (prec == 1 ? OPCODE_fconst_0 : OPCODE_dconst_0);
> Mark> -	else if (real_onep (exp))
> Mark> +	if (real_onep (exp))
> 
> Mark> This isn't a real solution, but I don't know how to correctly
> Mark> test for positive zero and I really must get some sleep
> Mark> now. Maybe someone more familiar with this code can suggest the
> Mark> proper solution.
> 
> I don't know if it is the proper solution but in gcc/real.h I see
> REAL_VALUE_MINUS_ZERO.  Would it be appropriate to use this?

I am now testing this patch which seems to do exactly what is needed.

--- gcc/java/jcf-write.c	12 Mar 2002 10:33:57 -0000	1.99.2.1
+++ gcc/java/jcf-write.c	16 Apr 2002 15:53:27 -0000
@@ -25,6 +25,7 @@
 #include "system.h"
 #include "jcf.h"
 #include "tree.h"
+#include "real.h"
 #include "java-tree.h"
 #include "obstack.h"
 #undef AND
@@ -1536,7 +1537,7 @@
       {
 	int prec = TYPE_PRECISION (type) >> 5;
 	RESERVE(1);
-	if (real_zerop (exp))
+	if (real_zerop (exp) && ! REAL_VALUE_MINUS_ZERO (TREE_REAL_CST (exp)))
 	  OP1 (prec == 1 ? OPCODE_fconst_0 : OPCODE_dconst_0);
 	else if (real_onep (exp))
 	  OP1 (prec == 1 ? OPCODE_fconst_1 : OPCODE_dconst_1);

Cheers,

Mark