Re: target/2273: [Alpha/Tru64 UNIX V5.1] SEGV in long double code

[Rainer Orth <ro@TechFak.Uni-Bielefeld.DE>]

The following reply was made to PR target/2273; it has been noted by GNATS.

From: Rainer Orth <ro@TechFak.Uni-Bielefeld.DE>
To: gcc-gnats@gcc.gnu.org
Cc: gcc-bugs@gcc.gnu.org, Richard Henderson <rth@redhat.com>
Subject: Re: target/2273: [Alpha/Tru64 UNIX V5.1] SEGV in long double code
Date: Tue, 3 Apr 2001 19:09:23 +0200 (MEST)

 I've single-stepped the test program and found the exact location where it
 SEGVs:
 
 int
 main (void)
 {
   long double ld = 1.0;
 }
 
 Here's an excerpt from the resulting assembler code:
 
 	.rdata
 	.quad 0
 	.align 4
 $LC0:
 	.quad 0x000000000,0x3fff000000000000
 
 and here's the relevant code for the assignment to ld, just after the
 function prologue:
 
 	lda $1,$LC0	; load address of ld into $1
 	ldq $1,0($1)	; store high quad of ld into $1, overwriting &ld!
 	ldq $2,8($1)	; the first quad of 1.0 happens to be 0, so 8($1)
 			; dereferences a NULL pointer and SEGVs
 
 Unfortunately, I cannot yet read the RTL dumps to understand why this is
 happening.
 
 	Rainer
 
 -----------------------------------------------------------------------------
 Rainer Orth, Faculty of Technology, Bielefeld University
 
 Email: ro@TechFak.Uni-Bielefeld.DE