Started from here
http://gcc.gnu.org/ml/gcc-regression/2011-08/msg00051.html

It is caused by revision 177218:

http://gcc.gnu.org/ml/gcc-cvs/2011-08/msg00228.html

On Linux/ia32, adding -march=corei7-avx caused:

[hjl@gnu-snb-1 src]$ /export/gnu/import/rrs-bootstrap/177218/bld/./gcc/xgcc -shared-libgcc -B/export/gnu/import/rrs-bootstrap/177218/bld/./gcc -nostdinc++ -L/export/gnu/import/rrs-bootstrap/177218/bld/x86_64-unknown-linux-gnu/32/libstdc++-v3/src -L/export/gnu/import/rrs-bootstrap/177218/bld/x86_64-unknown-linux-gnu/32/libstdc++-v3/src/.libs -B/export/gnu/import/rrs-bootstrap/177218/usr/x86_64-unknown-linux-gnu/bin/ -B/export/gnu/import/rrs-bootstrap/177218/usr/x86_64-unknown-linux-gnu/lib/ -isystem /export/gnu/import/rrs-bootstrap/177218/usr/x86_64-unknown-linux-gnu/include -isystem /export/gnu/import/rrs-bootstrap/177218/usr/x86_64-unknown-linux-gnu/sys-include -m32 -I/export/gnu/import/rrs-bootstrap/177218/bld/x86_64-unknown-linux-gnu/32/libstdc++-v3/include/x86_64-unknown-linux-gnu -I/export/gnu/import/rrs-bootstrap/177218/bld/x86_64-unknown-linux-gnu/32/libstdc++-v3/include -I/export/gnu/import/rrs-bootstrap/177218/src/libstdc++-v3/libsupc++ -fno-implicit-templates -Wall -Wextra -Wwrite-strings -Wcast-qual -fdiagnostics-show-location=once -ffunction-sections -fdata-sections -frandom-seed=functexcept.lo -g -O2 -D_GNU_SOURCE -m32 -std=gnu++0x -c ../../../../../src/libstdc++-v3/src/functexcept.cc  -fPIC -DPIC -S -march=corei7-avx
../../../../../src/libstdc++-v3/src/functexcept.cc: In function ‘void std::__throw_system_error(int)’:
../../../../../src/libstdc++-v3/src/functexcept.cc:106:62: internal compiler error: in fixup_args_size_notes, at expr.c:3586
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
[hjl@gnu-snb-1 src]$

Mine.

It's a bug in the i386 backend:

(call_insn 28 27 136 2 (parallel [
            (parallel [
                    (call (mem:QI (mem/f:SI (plus:SI (reg/f:SI 73 [ __ec$_M_cat_2->_vptr.error_category ])
                                    (const_int 12 [0xc])) [3 MEM[(int (*__vtbl_ptr_type) () *)D.126512_12 + 12B]+0 S4 A32]) [0 *OBJ_TYPE_REF(D.126513_13;__ec$_M_cat_2->3) S1 A8])
                        (const_int 12 [0xc]))
                    (set (reg/f:SI 7 sp)
                        (plus:SI (reg/f:SI 7 sp)
                            (const_int 4 [0x4])))
                ])
            (unspec [
                    (const_int 2 [0x2])
                ] UNSPEC_CALL_NEEDS_VZEROUPPER)
        ]) /home/rth/work/gcc/bld/i686-linux/libstdc++-v3/include/system_error:158 480 {*call_pop_vzeroupper}

Note the nested PARALLEL.

Created attachment 24910 [details]
A patch

I am testing this.

Created attachment 24911 [details]
full patch

Your patch misses several patterns.

This bootstraps, but I don't have an avx cpu locally.

(In reply to comment #6)
> Created attachment 24911 [details]
> full patch
> 
> Your patch misses several patterns.
> 
> This bootstraps, but I don't have an avx cpu locally.

I am testing it now.  Thanks.

(In reply to comment #6)
> Created attachment 24911 [details]
> full patch
> 
> Your patch misses several patterns.
> 
> This bootstraps, but I don't have an avx cpu locally.

It works:

http://gcc.gnu.org/ml/gcc-testresults/2011-08/msg00355.html

Author: rth
Date: Thu Aug  4 15:47:42 2011
New Revision: 177408

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=177408
Log:
PR target/49964
        * config/i386/i386.c (ix86_expand_call): Don't create nested
        PARALLELs for TARGET_VZEROUPPER.
        (ix86_split_call_vzeroupper): Fix extraction of the original call.
        * config/i386/i386.md (*call_rex64_ms_sysv_vzeroupper): Don't
        recognize nested PARALLELs.
        (*call_pop_vzeroupper, *sibcall_pop_vzeroupper,
        *call_value_rex64_ms_sysv_vzeroupper, *call_value_pop_vzeroupper,
        *sibcall_value_pop_vzeroupper): Likewise.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386.c
    trunk/gcc/config/i386/i386.md

Fixed.