[Bug target/65153] New: sh: "insn does not satisfy its constraints" when compiling libmcrypt

[Bug target/65153] New: sh: "insn does not satisfy its constraints" when compiling libmcrypt

[glaubitz at physik dot fu-berlin.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

            Bug ID: 65153
           Summary: sh: "insn does not satisfy its constraints" when
                    compiling libmcrypt
           Product: gcc
           Version: 4.9.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: glaubitz at physik dot fu-berlin.de
            Target: sh*-*-*

Hi!

Another regression in gcc-4.9.2, this time when trying to compile libmcrypt
2.5.8:

libtool: link: ( cd ".libs" && rm -f "arcfour.la" && ln -s "../arcfour.la"
"arcfour.la" )
/bin/sh ../../libtool  --tag=CC   --mode=compile gcc -DHAVE_CONFIG_H  -I.
-I../.. -I. -I../..  -I../../lib    -g -O2 -c -o cast-256.lo cast-256.c
libtool: compile:  gcc -DHAVE_CONFIG_H -I. -I../.. -I. -I../.. -I../../lib -g
-O2 -c cast-256.c  -fPIC -DPIC -o .libs/cast-256.o
cast-256.c: In function 'cast_256_LTX__mcrypt_set_key':
cast-256.c:409:1: error: insn does not satisfy its constraints:
 }
 ^
(insn 143 142 144 12 (set (reg/v:SI 7 r7 [orig:227 t ] [227])
        (ior:SI (reg:SI 1 r1)
            (reg:SI 13 r13 [orig:718 D.4992 ] [718]))) cast-256.c:387 130
{*iorsi3_compact}
     (expr_list:REG_DEAD (reg:SI 13 r13 [orig:718 D.4992 ] [718])
        (nil)))
cast-256.c:409:1: internal compiler error: in copyprop_hardreg_forward_1, at
regcprop.c:775
Please submit a full bug report,
with preprocessed source if appropriate.
See <file:///usr/share/doc/gcc-4.9/README.Bugs> for instructions.
Preprocessed source stored into /tmp/ccwCPfub.out file, please attach this to
your bugreport.

Full build log can be found in [1]. Attaching the precompiled source file
ccwCPfub.out.

Adrian

> [1] http://buildd.debian-ports.org/status/fetch.php?pkg=libmcrypt&arch=sh4&ver=2.5.8-3.3&stamp=1424499784

[Bug target/65153] sh: "insn does not satisfy its constraints" when compiling libmcrypt

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |olegendo at gcc dot gnu.org

--- Comment #1 from Oleg Endo <olegendo at gcc dot gnu.org> ---
(In reply to John Paul Adrian Glaubitz from comment #0)
> 
> Full build log can be found in [1]. Attaching the precompiled source file
> ccwCPfub.out.

There are no attachments in this PR.  Could you please check again?

[Bug target/65153] sh: "insn does not satisfy its constraints" when compiling libmcrypt

[glaubitz at physik dot fu-berlin.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #2 from John Paul Adrian Glaubitz <glaubitz at physik dot fu-berlin.de> ---
Created attachment 34823
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34823&action=edit
Pre-compiled source file for libmcrypt 2.5.8

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|sh: "insn does not satisfy  |[SH][4.9 Regression] "insn
                   |its constraints" when       |does not satisfy its
                   |compiling libmcrypt         |constraints" when compiling
                   |                            |libmcrypt

--- Comment #3 from Oleg Endo <olegendo at gcc dot gnu.org> ---
I was able to reproduce the problem on 4.9 branch sh-elf with -m4 -ml -O2
-fPIC.
On trunk (GCC 5) it seems not to happen.

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-02-21
     Ever confirmed|0                           |1

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kkojima at gcc dot gnu.org

--- Comment #4 from Oleg Endo <olegendo at gcc dot gnu.org> ---
(insn 143 142 144 12 (set (reg/v:SI 7 r7 [orig:227 t ] [227])
        (ior:SI (reg:SI 1 r1)
            (reg:SI 13 r13 [orig:718 D.4992 ] [718]))) cast-256.c:387 130
{*iorsi3_compact}
     (expr_list:REG_DEAD (reg:SI 13 r13 [orig:718 D.4992 ] [718])
        (nil)))

This is an impossible register allocation for the insn.  It allows only 2
different regs but there are 3 different regs assigned.  I've seen this
somewhere before ... but I can't recall where.  Kaz do you remember anything
like that?

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[kkojima at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #5 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
Created attachment 34828
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34828&action=edit
reduced testcase

Looks an interesting "why didn't we see it?" target bug.
The problem happens at peephole2 phase.  The insn sequence

(insn 353 103 105 4 (set (reg:SI 1 r1) (reg:SI 6 r6)))
(insn 105 353 106 4 (set (reg:SI 7 r7) (ashift:SI (reg:SI 7 r7) (reg:SI 0
r0))))
(insn 106 105 107 4 (set (reg/v:SI 6 r6)
             (ior:SI (reg:SI 6 r6) (reg:SI 7 r7))))

is tested by the complex peephole

(define_peephole2
  [(set (match_operand 0 "any_register_operand" "")
    (match_operand 1 "any_register_operand" ""))
   (set (match_operand 2 "any_register_operand" "") (match_operand 3 "" ""))
   (set (match_operand 4 "" "") (match_operand 5 "" ""))]
  "(HARD_REGNO_NREGS (REGNO (operands[0]), GET_MODE (operands[2]))
    <= HARD_REGNO_NREGS (REGNO (operands[0]), GET_MODE (operands[0])))
   && peep2_reg_dead_p (3, operands[0]) && peep2_reg_dead_p (3, operands[2])
   && ! FIND_REG_INC_NOTE (peep2_next_insn (2), operands[0])
   && ! FIND_REG_INC_NOTE (peep2_next_insn (2), operands[2])
   && ! reg_overlap_mentioned_p (operands[0], operands[3])
   && ! reg_overlap_mentioned_p (operands[2], operands[0])
   && ! reg_overlap_mentioned_p (operands[0], operands[1])
   && (REGNO_REG_CLASS (REGNO (operands[0]))
       == REGNO_REG_CLASS (REGNO (operands[2])))
   && (REGNO_REG_CLASS (REGNO (operands[1]))
       == REGNO_REG_CLASS (REGNO (operands[0])))"
  [(set (match_dup 0) (match_dup 3))
   (set (match_dup 4) (match_dup 5))]
{
  ...

This peephole2 in sh.md tries to replace operands[0] with operands[1]
and operands[2] with operands[0] in operands[4]/operands[5].  In our
case, operands[5] is (ior:SI (reg:SI 6 r6) (reg:SI 7 r7)) and the peephole
tries to do term rewriting r1 -> r6, r7 -> r1.  This peephole fails
after all and then tries to restore operands[5] with term rewriting
r6 -> r1, r1 -> r7.  Unfortunately this can't reverse the first term
rewriting and yields (ior:SI (reg:SI 1 r1) (reg:SI 7 r7)).  Thus we
got a wrong insn
(set (reg/v:SI 6 r6) (ior:SI (reg:SI 1 r1) (reg:SI 7 r7))).

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[kkojima at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #6 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
Created attachment 34831
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34831&action=edit
a possible patch

I think the problem is latent on trunk.

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #8 from Oleg Endo <olegendo at gcc dot gnu.org> ---
It seems that the problematic peephole is trying to wallpaper some unlucky
register allocation/reload choices like:

before peephole:
    mov.w    .L26,r1
    add    r8,r1
    mov    r1,r8
    mov    #0,r1
    mov.l    r1,@(32,r8)

after peephole:
    mov.w    .L26,r1
    add    r8,r1
    mov    #0,r8
    mov.l    r8,@(32,r1)

I've tried to disable the peephole on trunk and compared CSiBE results.  It
seems the peephole doesn't hit very often:
sum:  3371887 -> 3371943    +56 / +0.001661 %

So we could probably remove the complex peephole without a big loss.

I have a set of similar but simpler peepholes with a higher hit rate.  Maybe we
can replace the complex peephole with some simpler ones.

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #9 from Oleg Endo <olegendo at gcc dot gnu.org> ---
(In reply to Oleg Endo from comment #8)
> 
> I've tried to disable the peephole on trunk and compared CSiBE results.  It
> seems the peephole doesn't hit very often:
> sum:  3371887 -> 3371943    +56 / +0.001661 %
> 
> So we could probably remove the complex peephole without a big loss.
> 
> I have a set of similar but simpler peepholes with a higher hit rate.  Maybe
> we can replace the complex peephole with some simpler ones.

The set of peepholes can be found in PR 61142, attachment 34839.
CSiBE numbers with that patch:

sum:  3371887 -> 3367967    -3920 / -0.116255 %
avg: -217.777778 / -0.167780 %
max: replaypc-0.4.0.preproc   57212 -> 57188          -24 / -0.041949 %
min: mpeg2dec-0.3.1           54348 -> 53984         -364 / -0.669758 %

and with the patch, but without the problematic peephole of this PR:

sum:  3371887 -> 3367983    -3904 / -0.115781 %
avg: -216.888889 / -0.167686 %
max: replaypc-0.4.0.preproc   57212 -> 57188          -24 / -0.041949 %
min: mpeg2dec-0.3.1           54348 -> 53984         -364 / -0.669758 %

It seems that the other set of peepholes covers most of the cases of the
complex peephole.

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[kkojima at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #10 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
The new peepholes are fine for trunk even in the stage4 but a bit
invasive to the release branch.  For 4.9 branch, I'd like to simply
remove the problematic peephole.

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #11 from Oleg Endo <olegendo at gcc dot gnu.org> ---
Sure.  I was actually referring to trunk all the time :)
I agree to remove the problematic peephole on 4.9 branch.

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[kkojima at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #12 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
Created attachment 34840
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34840&action=edit
patch for 4_9-branch

I've noticed that that peephole is the last user of sh.c:replace_n_hard_rtx.
I'm testing this and the similar patch for trunk on sh4-unknown-linux-gnu.

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P4
   Target Milestone|---                         |4.9.3

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[kkojima at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #14 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
Author: kkojima
Date: Wed Feb 25 01:00:29 2015
New Revision: 220957

URL: https://gcc.gnu.org/viewcvs?rev=220957&root=gcc&view=rev
Log:
PR target/65153
Backport from mainline: [SH] Fix PR target/65153 with removing movsicc_true+3
peephole2.


Modified:
    branches/gcc-4_9-branch/gcc/ChangeLog
    branches/gcc-4_9-branch/gcc/config/sh/sh-protos.h
    branches/gcc-4_9-branch/gcc/config/sh/sh.c
    branches/gcc-4_9-branch/gcc/config/sh/sh.md

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[kkojima at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

Kazumoto Kojima <kkojima at gcc dot gnu.org> changed:

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

--- Comment #15 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
Fixed.

[Bug target/65153] [SH][4.9 Regression] "insn does not satisfy its constraints" when compiling libmcrypt

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65153

--- Comment #16 from Oleg Endo <olegendo at gcc dot gnu.org> ---
Author: olegendo
Date: Sat Mar  7 19:35:22 2015
New Revision: 221257

URL: https://gcc.gnu.org/viewcvs?rev=221257&root=gcc&view=rev
Log:
gcc/testsuite/
    PR target/65153
    * gcc.c-torture/compile/pr65153.c: New.

Added:
    trunk/gcc/testsuite/gcc.c-torture/compile/pr65153.c
Modified:
    trunk/gcc/testsuite/ChangeLog