[Bug bootstrap/62301] New: PowerPC bootstrap broken since r214654

[Bug bootstrap/62301] New: PowerPC bootstrap broken since r214654

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

            Bug ID: 62301
           Summary: PowerPC bootstrap broken since r214654
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
          Assignee: unassigned at gcc dot gnu.org
          Reporter: wschmidt at gcc dot gnu.org
                CC: rdsandiford at googlemail dot com
              Host: powerpc64le-unknown-linux-gnu
            Target: powerpc64le-unknown-linux-gnu
             Build: powerpc64le-unknown-linux-gnu

Bootstrap is failing in stage 2 for powerpc64le-unknown-linux-gnu (and likely
in powerpc64-unknown-linux-gnu as well, though I have not yet confirmed this). 
I've bisected the start of the failure to r214654:

2014-08-28  Richard Sandiford  <rdsandiford@googlemail.com>                     

        * rtl.h (get_pool_constant, rtx_referenced_p): Replace rtx parameters   
        with const_rtx parameters.                                              
        * varasm.c (get_pool_constant): Likewise.                               
        * rtlanal.c (rtx_referenced_p_1): Delete.                               
        (rtx_referenced_p): Use FOR_EACH_SUBRTX instead of for_each_rtx.        
        Assert that the rtx we're looking for is nonnull.  Allow searches       
        for constant pool SYMBOL_REFs.                                          

Many compiles fail while building libgcc, all segfaulting:

/home/wschmidt/gcc/gcc-mainline-base/libgcc/libgcc2.c: In function '__ashlti3':
/home/wschmidt/gcc/gcc-mainline-base/libgcc/libgcc2.c:454:1: internal compiler
error: Segmentation fault
 }
 ^
xgcc: internal compiler error: Segmentation fault (program cc1)
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

I'll try to get more information in the morning.

[Bug bootstrap/62301] PowerPC bootstrap broken since r214654

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

Bill Schmidt <wschmidt at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|powerpc64le-unknown-linux-g |powerpc64*-unknown-linux-gn
                   |nu                          |u
               Host|powerpc64le-unknown-linux-g |powerpc64*-unknown-linux-gn
                   |nu                          |u
              Build|powerpc64le-unknown-linux-g |powerpc64*-unknown-linux-gn
                   |nu                          |u

--- Comment #1 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Confirmed that this fails for big-endian as well as little-endian (both Power8
boxes).

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |5.0
            Summary|PowerPC bootstrap broken    |[5 Regression] PowerPC
                   |since r214654               |bootstrap broken since
                   |                            |r214654

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

--- Comment #2 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Example backtrace for segv:

#0  0x000000004d5a28a0 in ?? ()
#1  0x0000000010370128 in mem_loc_descriptor(rtx_def*, machine_mode,
machine_mode, var_init_status) ()
#2  0x00000000103704f4 in mem_loc_descriptor(rtx_def*, machine_mode,
machine_mode, var_init_status) ()
#3  0x000000001037a59c in loc_descriptor(rtx_def*, machine_mode,
var_init_status) ()
#4  0x000000001037a7b8 in loc_descriptor(rtx_def*, machine_mode,
var_init_status) ()
#5  0x000000001037ad58 in dw_loc_list_1(tree_node*, rtx_def*, int,
var_init_status) ()
#6  0x0000000010375dac in loc_list_from_tree(tree_node*, int) ()
#7  0x000000001037d43c in add_location_or_const_value_attribute(die_struct*,
tree_node*, bool, dwarf_attribute) [clone .constprop.340] ()
#8  0x0000000010363e3c in gen_variable_die(tree_node*, tree_node*, die_struct*)
    ()
#9  0x0000000010365dd4 in gen_decl_die(tree_node*, tree_node*, die_struct*) ()
#10 0x0000000010380484 in decls_for_scope(tree_node*, die_struct*, int) ()
#11 0x000000001035fd70 in gen_subprogram_die(tree_node*, die_struct*) ()
#12 0x0000000010365b18 in gen_decl_die(tree_node*, tree_node*, die_struct*) ()
#13 0x0000000010367518 in dwarf2out_function_decl(tree_node*) ()
#14 0x00000000103e65a0 in (anonymous namespace)::pass_final::execute(function*)
    ()
#15 0x0000000010635a7c in execute_one_pass(opt_pass*) ()
#16 0x0000000010636184 in execute_pass_list_1(opt_pass*) ()
#17 0x000000001063619c in execute_pass_list_1(opt_pass*) ()
#18 0x000000001063619c in execute_pass_list_1(opt_pass*) ()
#19 0x0000000010636224 in execute_pass_list(function*, opt_pass*) ()
#20 0x00000000102d98a4 in cgraph_node::expand() ()
#21 0x00000000102db3b8 in symbol_table::compile() ()
#22 0x00000000102dd2f8 in symbol_table::finalize_compilation_unit() ()
#23 0x0000000010132738 in c_write_global_declarations() ()
#24 0x00000000107220c0 in compile_file() ()
#25 0x000000001072522c in toplev_main(int, char**) ()
#26 0x000000001010f288 in main ()

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

--- Comment #3 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Looks like a subtle logic change in the patch:

+  FOR_EACH_SUBRTX (iter, array, body, ALL)
+    if (const_rtx y = *iter)
+      {
+       /* Check if a label_ref Y refers to label X.  */
+       if (GET_CODE (y) == LABEL_REF && LABEL_P (y) && XEXP (y, 0) == x)
+         return true;

-  if (*body == NULL_RTX)
-    return y == NULL_RTX;
+       if (rtx_equal_p (x, y))
+         return true;

-  /* Return true if a label_ref *BODY refers to label Y.  */
-  if (GET_CODE (*body) == LABEL_REF && LABEL_P (y))
-    return XEXP (*body, 0) == y;

If XEXP (y, 0) != x, it seems we should return false to match what was done
before.  Testing a patch to fix this.

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

--- Comment #4 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Unfortunately that was not sufficient -- same SEGVs are still occurring.

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[rsandifo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2014-08-29
                 CC|                            |rsandifo at gcc dot gnu.org
           Assignee|unassigned at gcc dot gnu.org      |rsandifo at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #5 from rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> ---
(In reply to Bill Schmidt from comment #4)
> Unfortunately that was not sufficient -- same SEGVs are still occurring.

Sorry, it was a really dumb typo.  Instead of:

    if (GET_CODE (y) == LABEL_REF && LABEL_P (y) && XEXP (y, 0) == x)

it should be:

    if (GET_CODE (y) == LABEL_REF && LABEL_P (x) && XEXP (y, 0) == x)

I confirmed that that is enough to get identical expmed.c code before
and after the patch, am testing a full bootstrap now.

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

--- Comment #6 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Ah...I've been staring at the two versions for so long and that never leaped
out at me. :)  Thanks, Richard!

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[rsandifo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

--- Comment #7 from rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> ---
Author: rsandifo
Date: Fri Aug 29 15:51:13 2014
New Revision: 214733

URL: https://gcc.gnu.org/viewcvs?rev=214733&root=gcc&view=rev
Log:
gcc/
    PR bootstrap/62301
    * rtlanal.c (rtx_referenced_p): Fix typo in LABEL_P call.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/rtlanal.c

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[rsandifo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #8 from rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> ---
Patch applied.

[Bug bootstrap/62301] [5 Regression] PowerPC bootstrap broken since r214654

[wschmidt at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62301

--- Comment #9 from Bill Schmidt <wschmidt at gcc dot gnu.org> ---
Thanks, Richard!