Re: FYI: Latest gcc-8 snapshot gives ICE with later isl's

[Donald Parsons <dparsons at brightdsl dot net>]

On Mon, 2018-03-05 at 19:04 +0100, Marek Polacek wrote:
> On Mon, Mar 05, 2018 at 12:58:23PM -0500, Donald Parsons wrote:
> > 
> > I am getting ICE bootstrapping gcc-8-20180304.tar.xz when using
> > either
> > isl-0.18 or isl-0.19.   I had never had a problem using latest isl
> > over
> > the past couple of years, so a change in gcc last week introduced
> > the
> > problem.
> > 
> > Normally I disable-bootstrap and use last weeks gcc-8 to build this
> > weeks gcc-8 and avoid the extra long bootstrap time.  This actually
> > worked, at least with isl-0.18 but the resulting gcc-8 gave ICE
> > errors
> > building latest "pari-gp" and the latest-rc4 Linux kernel.  So went
> > back to bootstrapping with (Redhat gcc-7.3.1-2) and discovered the
> > above.
> > 
> > Here is failure (using isl-0.18; 0.19 looks same):
> > ...
> > libtool: compile:  /home/don/src/gcc-8-20180304-obj/./prev-gcc/xgcc
> > -B/home/don/src/gcc-8-20180304-obj/./prev-gcc/ -B/opt/gcc/gcc-8-
> > 0304/x86_64-redhat-linux/bin/ -B/opt/gcc/gcc-8-0304/x86_64-redhat-
> > linux/bin/ -B/opt/gcc/gcc-8-0304/x86_64-redhat-linux/lib/ -isystem
> > /opt/gcc/gcc-8-0304/x86_64-redhat-linux/include -isystem
> > /opt/gcc/gcc-
> > 8-0304/x86_64-redhat-linux/sys-include -DHAVE_CONFIG_H -I.
> > -I../../gcc-
> > 8-20180304/isl -I../../gcc-8-20180304/isl/include -Iinclude/ -g -O2
> > -gtoggle -MT isl_factorization.lo -MD -MP -MF
> > .deps/isl_factorization.Tpo -c ../../gcc-8-
> > 20180304/isl/isl_factorization.c -o isl_factorization.o
> > during GIMPLE pass: pre
> > ../../gcc-8-20180304/isl/isl_factorization.c: In function
> > 'isl_basic_set_factorizer':
> > ../../gcc-8-20180304/isl/isl_factorization.c:256:28: internal
> > compiler
> > error: in compute_antic_aux, at tree-ssa-pre.c:2148
> >  __isl_give isl_factorizer *isl_basic_set_factorizer(
> >                             ^~~~~~~~~~~~~~~~~~~~~~~~
> > 0x11e6790 compute_antic_aux
> > 	../../gcc-8-20180304/gcc/tree-ssa-pre.c:2148
> > 0x11e713b compute_antic
> > 	../../gcc-8-20180304/gcc/tree-ssa-pre.c:2364
> > 0x11ec1b1 execute
> > 	../../gcc-8-20180304/gcc/tree-ssa-pre.c:4131
> > Please submit a full bug report,
> > with preprocessed source if appropriate.
> > Please include the complete backtrace with any bug report.
> > See <https://gcc.gnu.org/bugs/> for instructions.
> > make[5]: *** [Makefile:1463: isl_factorization.lo] Error 1
> > make[5]: Leaving directory '/home/don/src/gcc-8-20180304-obj/isl'
> > make[4]: *** [Makefile:1571: all-recursive] Error 1
> > make[4]: Leaving directory '/home/don/src/gcc-8-20180304-obj/isl'
> > make[3]: *** [Makefile:1167: all] Error 2
> > make[3]: Leaving directory '/home/don/src/gcc-8-20180304-obj/isl'
> > make[2]: *** [Makefile:6155: all-stage2-isl] Error 2
> > make[2]: Leaving directory '/home/don/src/gcc-8-20180304-obj'
> > make[1]: *** [Makefile:25157: stage2-bubble] Error 2
> > make[1]: Leaving directory '/home/don/src/gcc-8-20180304-obj'
> > make: *** [Makefile:947: all] Error 2
> 
> I bet this is https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84670
> which is already
> fixed.
> 
> 	Marek

Adding Richard Biener to cc.

Marek, thanks for the heads up.  I applied the patch from the bugzilla
but it still fails to bootstrap with same ICE error, only line numbers
changes by about 4.  (isl-0.19 was used in this case)

So I tried using the system's (Fedora 26) isl-0.16 and this makes
everything build correctly.  So the latest isl-0.18 or 0.19 are more
sensitive to the changes made to gcc last week.

So it might be easiest to fix gcc to work with isl-0.18/19 now while
the changes being made to gcc are fresh in mind.

Don


PS: I had to manually apply the patch, so in case I misunderstood, here
     is resulting diff:

 --- tree-ssa-pre.c.orig	2018-03-05 13:48:02.446954477 -0500
+++ tree-ssa-pre.c	2018-03-05 13:52:51.851014394 -0500
@@ -1393,8 +1393,12 @@
 		    else
 		      {
 			unsigned value_id = get_expr_value_id
(constant);
-			constant = find_leader_in_sets (value_id,
set1, set2,
-							AVAIL_OUT
(pred));
+			/* We want a leader in ANTIC_OUT to be
consistent
+			   with the rest of phi_translate but we don't
have
+			   that available (and it isn't complete yet)
so
+			   we resort to AVAIL_OUT here.  */
+			constant = bitmap_find_leader (AVAIL_OUT
(pred),
+						       value_id);
 			if (constant)
 			  return constant;
 		      }
(sorry, evolution wrapped it)