This is the mail archive of the gcc-bugs@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug target/81291] [5/6/7/8 Regression] wrong code with -O2 -fno-rerun-cse-after-loop -fno-tree-ter -fno-tree-vrp -funroll-loops due to improper carry


https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81291

--- Comment #1 from Zdenek Sojka <zsojka at seznam dot cz> ---
Created attachment 41673
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=41673&action=edit
testcase needing less compiler flags

$ powerpc-unknown-linux-gnu-gcc -Og -fgcse -fweb testcase.c
$ qemu-ppc ./a.out 
qemu: uncaught target signal 6 (Aborted) - core dumped
Aborted

$ powerpc-unknown-linux-gnu-gcc -v
Using built-in specs.
COLLECT_GCC=/repo/gcc-trunk//binary-latest-powerpc/bin/powerpc-unknown-linux-gnu-gcc
COLLECT_LTO_WRAPPER=/repo/gcc-trunk/binary-trunk-249846-checking-yes-rtl-df-extra-nographite-powerpc/bin/../libexec/gcc/powerpc-unknown-linux-gnu/8.0.0/lto-wrapper
Target: powerpc-unknown-linux-gnu
Configured with: /repo/gcc-trunk//configure --enable-languages=c,c++
--enable-valgrind-annotations --disable-nls --enable-checking=yes,rtl,df,extra
--without-cloog --without-ppl --without-isl
--with-sysroot=/usr/powerpc-unknown-linux-gnu --build=x86_64-pc-linux-gnu
--host=x86_64-pc-linux-gnu --target=powerpc-unknown-linux-gnu
--with-ld=/usr/bin/powerpc-unknown-linux-gnu-ld
--with-as=/usr/bin/powerpc-unknown-linux-gnu-as --disable-libstdcxx-pch
--prefix=/repo/gcc-trunk//binary-trunk-249846-checking-yes-rtl-df-extra-nographite-powerpc
Thread model: posix
gcc version 8.0.0 20170630 (experimental) (GCC) 

The situations seems to be the same as in comment #0:

$ cat testcase.c.259.init-regs
...
(insn 73 99 75 5 (parallel [
            (set (reg:SI 185 [+4 ])
                (minus:SI (const_int 0 [0])
                    (reg:SI 179 [ _5+4 ])))
            (set (reg:SI 76 ca)
                (leu:SI (reg:SI 179 [ _5+4 ])
                    (const_int 0 [0])))
        ]) "testcase.c":12 104 {subfsi3_carry}
     (expr_list:REG_UNUSED (reg:SI 76 ca)
        (nil)))
...
(insn 79 78 80 5 (parallel [
            (set (reg:SI 187 [+4 ])
                (plus:SI (reg:SI 189 [+4 ])
                    (const_int 1 [0x1])))
            (set (reg:SI 76 ca)
                (geu:SI (reg:SI 189 [+4 ])
                    (const_int -1 [0xffffffffffffffff])))
        ]) "testcase.c":14 79 {*addsi3_imm_carry_pos}
     (expr_list:REG_DEAD (reg:SI 189 [+4 ])
        (nil)))
(insn 80 79 102 5 (parallel [
            (set (reg:SI 186)
                (plus:SI (reg:SI 185 [+4 ])
                    (reg:SI 76 ca)))
            (clobber (reg:SI 76 ca))
        ]) "testcase.c":14 85 {addsi3_carry_in_0}
     (expr_list:REG_DEAD (reg:SI 185 [+4 ])
        (expr_list:REG_DEAD (reg:SI 76 ca)
            (expr_list:REG_UNUSED (reg:SI 76 ca)
                (nil)))))
...

$ cat testcase.c.261.combine
...
Trying 73 -> 80:
Failed to match this instruction:
(parallel [
        (set (reg:SI 186)
            (const_int 1 [0x1]))
        (clobber (reg:SI 76 ca))
    ])
Successfully matched this instruction:
(set (reg:SI 186)
    (const_int 1 [0x1]))
allowing combination of insns 73 and 80
original costs 0 + 4 = 0
replacement cost 4
deferring deletion of insn with uid = 73.
deferring deletion of insn with uid = 73.
modifying insn i3    80: r186:SI=0x1
deferring rescan insn with uid = 80.
...

But the analysis I did in comment #0 is not correct, since I misunderstood some
of the semantics from the dumps.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]