[Bug c/48090] New: gcc 4.5.2 miscompilation when building on arm

2011-03-12 Thread arnaud.pat...@rtp-net.org 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48090

   Summary: gcc 4.5.2 miscompilation when building on arm
   Product: gcc
   Version: unknown
Status: UNCONFIRMED
  Severity: normal
  Priority: P3
 Component: c
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: arnaud.pat...@rtp-net.org


With gcc-4.5.2, even with gcc-4.5-20110310, the following line of code gets
miscompiled (See attachment) :

size -= nblk * 512;

size and nblk are 64 bits.
gcc tries to compute -nblk. nblk is stored in r0,r1. gcc produces the following
asm code :

rsbsr1, r0, #0
rsc r2, r1, #0

So r1 gets corrupted by the rsbs insn, thus getting wrong value in r2.

I'm building the asm code with :
gcc -O2  -march=armv5t list2.i -o - -S

When I use :
gcc -O2 -fno-cse-follow-jumps  -march=armv5t list2.i -o - -S

I get :

rsbsr0, r0, #0
rsc r1, r1, #0

which is fine.

[Bug c/48090] gcc 4.5.2 miscompilation when building on arm

2011-03-12 Thread arnaud.pat...@rtp-net.org 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48090

--- Comment #1 from arnaud patard  2011-03-12 
09:02:30 UTC ---
Created attachment 23636
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=23636
test file

[Bug c/48090] gcc 4.5.2 miscompilation when building on arm

2011-03-12 Thread arnaud.pat...@rtp-net.org 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48090

--- Comment #3 from arnaud patard  2011-03-12 
12:16:12 UTC ---
(In reply to comment #2)
> I get correct-looking code on armv5tel-linux with vanilla gcc-4.6-20110305 and
> gcc-4.5-20110310, with 4.4 the code looks different but not obviously broken.

I said -march=armv5t (or -march=armv4t I guess). It's important as
-march=armv5te won't trigger the bug.

>
> Can you make the test case standalone and include a runtime check for the
> possible miscompilation?

My test case was to build tar and run the delete01.at check but it's not really
a standalone test.
So far, I've not managed to have a test case small than compiling the
preprocessed file.

> 
> How was your gcc configured?

I reproduce it with my gcc and debian gcc 4.5.2 so I believe this rules out
misconfiguration.

[Bug c/48090] gcc 4.5.2 miscompilation when building on arm

2011-03-13 Thread arnaud.pat...@rtp-net.org 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48090

--- Comment #9 from arnaud patard  2011-03-13 
20:40:27 UTC ---
I confirm that backporting r159644 and r159683 make things work. From comment
8, I guess that the bug is still there and that one can still hit it sooner or
later, right  ? (btw, amazing job)