[Bug bootstrap/53427] [4.8 Regression] Bootstrap failure

["ubizjak at gmail dot com" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53427

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2012-05-20 19:14:20 UTC ---
(gdb) bt
#0  save_call_clobbered_regs () at ../../gcc-svn/trunk/gcc/caller-save.c:877
#1  0x0000000000873bd7 in reload (first=0x7ffff1b100c0, global=1) at
../../gcc-svn/trunk/gcc/reload1.c:939
#2  0x00000000007b6488 in do_reload () at ../../gcc-svn/trunk/gcc/ira.c:4261
#3  rest_of_handle_reload () at ../../gcc-svn/trunk/gcc/ira.c:4352
#4  0x0000000000827645 in execute_one_pass (pass=0x1427920) at
../../gcc-svn/trunk/gcc/passes.c:2183
#5  0x0000000000827a05 in execute_pass_list (pass=0x1427920) at
../../gcc-svn/trunk/gcc/passes.c:2238
#6  0x0000000000827a17 in execute_pass_list (pass=0x1428060) at
../../gcc-svn/trunk/gcc/passes.c:2239
#7  0x0000000000616688 in expand_function (node=0x7ffff19b1750) at
../../gcc-svn/trunk/gcc/cgraphunit.c:1603
#8  0x000000000061822b in expand_all_functions () at
../../gcc-svn/trunk/gcc/cgraphunit.c:1708
#9  compile () at ../../gcc-svn/trunk/gcc/cgraphunit.c:2007

(gdb) list
872                       && HARD_REGISTER_P (cheap)
873                       && TEST_HARD_REG_BIT (call_used_reg_set, REGNO
(cheap)))
874                     {
875                       rtx call_set = single_set (insn);
876                       rtx dest = SET_DEST (call_set);
877                       rtx pat = gen_rtx_SET (VOIDmode, cheap,
878                                              copy_rtx (dest));
879                       chain = insert_one_insn (chain, 0, -1, pat);
880                     }
881                 }

(gdb) p call_set
$7 = (rtx_def *) 0x0

(gdb) p debug_rtx (insn)
(call_insn 22 21 25 2 (parallel [
            (set (reg:DI 0 ax)
                (call (mem:QI (symbol_ref:DI ("memcpy") [flags 0x41]
<function_decl 0x7ffff1aee600 memcpy>) [0 memcpy S1 A8])
                    (const_int 0 [0])))
            (unspec [
                    (const_int 2 [0x2])
                ] UNSPEC_CALL_NEEDS_VZEROUPPER)
        ]) pr53427.c:12 632 {*call_value_vzeroupper}
     (expr_list:REG_RETURNED (reg/v/f:DI 2 cx [orig:64 arg ] [64])
        (expr_list:REG_DEAD (reg:DI 5 di)
            (expr_list:REG_DEAD (reg:DI 4 si)
                (expr_list:REG_DEAD (reg:DI 1 dx)
                    (expr_list:REG_UNUSED (reg:DI 0 ax)
                        (expr_list:REG_EH_REGION (const_int 0 [0])
                            (nil)))))))
    (expr_list:REG_FRAME_RELATED_EXPR (set (reg:DI 0 ax)
            (reg:DI 5 di))
        (expr_list:REG_FRAME_RELATED_EXPR (use (reg:DI 5 di))
            (expr_list:REG_FRAME_RELATED_EXPR (use (reg:DI 4 si))
                (expr_list:REG_FRAME_RELATED_EXPR (use (reg:DI 1 dx))
                    (nil))))))

As said elsewhere, tagging call insns with vzeroupper unspecs was not a good
idea. While it worked, it is very fragile and not future proof.

Please look at PR 47440. The fix for that PR will magically avoid failures of
this type.