Re: [PATCH 1/5] Handle WORD_REGISTER_OPERATIONS when reloading (subreg(reg))

[Christophe Lyon <christophe dot lyon at linaro dot org>]

Hi,

On 20 February 2017 at 13:08, Matthew Fortune
<Matthew.Fortune@imgtec.com> wrote:
> Vladimir Makarov <vmakarov@redhat.com> writes:
>> On 02/07/2017 09:08 AM, Matthew Fortune wrote:
>> > Hi,
>> >
>> > This change deals with reloading a subreg(reg) using the inner mode to
>> > prevent partial spilling of data like in the case described here:
>> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78660#c8
>> >
>> > No test case for now but I am investigating a targeted test using the
>> > RTL frontend for later submission.
>> >
>> >
>> > gcc/
>> >     PR target/78660
>> >     * lra-constraints.c (curr_insn_transform): Handle
>> >     WORD_REGISTER_OPERATIONS requirements when reloading SUBREGs.
>> >
>> The patch is OK.  So all 5 patches can be committed to the trunk.  I
>> hope Eric's test of the patches on SPARC will be successful.  Thank you
>> again for your efforts.
>
> Committed as r245598.
>

This patch is causing ICEs on arm-none-linux-gnueabi
FAIL: gcc.c-torture/execute/simd-2.c   -O1  (internal compiler error)
/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.c-torture/execute/simd-2.c:
In function 'main':
/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.c-torture/execute/simd-2.c:72:1:
error: unable to find a register to spill
/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.c-torture/execute/simd-2.c:72:1:
error: this is the insn:
(insn 276 2123 2129 2 (set (subreg:SI (reg:TI 886 [362]) 0)
        (and:SI (subreg:SI (reg:TI 567 [orig:111 j.1_2 ] [111]) 0)
            (subreg:SI (reg:TI 568 [orig:110 i.0_1 ] [110]) 0)))
"/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.c-torture/execute/simd-2.c":47
82 {*arm_andsi3_insn}
     (expr_list:REG_DEAD (reg:TI 568 [orig:110 i.0_1 ] [110])
        (expr_list:REG_DEAD (reg:TI 567 [orig:111 j.1_2 ] [111])
            (nil))))
/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.c-torture/execute/simd-2.c:72:1:
internal compiler error: in assign_by_spills, at lra-assigns.c:1457
0xacc9d3 _fatal_insn(char const*, rtx_def const*, char const*, int, char const*)
        /aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/rtl-error.c:108
0x9a6123 assign_by_spills
        /aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/lra-assigns.c:1457
0x9a7506 lra_assign()
        /aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/lra-assigns.c:1643
0x9a16d4 lra(_IO_FILE*)
        /aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/lra.c:2451
0x957669 do_reload
        /aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/ira.c:5452
0x957669 execute
        /aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/ira.c:5636



I've also noticed that --target arm-none-eabi with default --with-mode
and --with-cpu
fails to build libgcc (it may be easier to reproduce):
/tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/libgcc/fixed-bit.c:
In function '__gnu_mulhelperudq':
/tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/libgcc/fixed-bit.c:371:1:
error: unable to find a register to spill
 }
 ^
/tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/libgcc/fixed-bit.c:371:1:
error: this is the insn:
(insn 286 296 287 2 (set (reg:SI 232)
        (neg:SI (ltu:SI (subreg:SI (reg:DI 238 [orig:119 _10 ] [119]) 4)
                (subreg:SI (reg/v:DI 149 [ temp1 ]) 4))))
"/tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/libgcc/fixed-bit.c":314
849 {cstor
esi_nltu_thumb1}
     (expr_list:REG_DEAD (reg:DI 238 [orig:119 _10 ] [119])
        (nil)))
/tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/libgcc/fixed-bit.c:371:1:
internal compiler error: in assign_by_spills, at lra-assigns.c:1457
0xacc9c3 _fatal_insn(char const*, rtx_def const*, char const*, int, char const*)
        /tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/rtl-error.c:108
0x9a6113 assign_by_spills
        /tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/lra-assigns.c:1457
0x9a74f6 lra_assign()
        /tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/lra-assigns.c:1643
0x9a16c4 lra(_IO_FILE*)
        /tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/lra.c:2451
0x957659 do_reload
        /tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/ira.c:5452
0x957659 execute
        /tmp/870921_29.tmpdir/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/ira.c:5636
make[4]: *** [_mulhelperUDQ.o] Error 1

Thanks,

Christophe


> Thanks,
> Matthew