[Bug c++/64110] New: ICE: Max. number of generated reload insns per insn is achieved (90)

[Bug c++/64110] New: ICE: Max. number of generated reload insns per insn is achieved (90)

[thiago at kde dot org]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="UTF-8", Size: 11180 bytes --]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

            Bug ID: 64110
           Summary: ICE: Max. number of generated reload insns per insn is
                    achieved (90)
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: thiago at kde dot org

When compiling the attached preprocessed sources, GCC 5 (trunk r218173) prints
the following:

/home/thiago/src/qt/qt5/qtbase/src/corelib/tools/qstringargbuilder.cpp: In
function ‘replaceArgEscapes(QString const&, ArgEscapeData const&, int, QString
const&, QString const&, QChar)’:
/home/thiago/src/qt/qt5/qtbase/src/corelib/tools/qstringargbuilder.cpp:236:1:
internal compiler error: Max. number of generated reload insns per insn is
achieved (90)

 }
 ^
0xb58ff0 lra_constraints(bool)
        /home/thiago/src/gcc/gcc/lra-constraints.c:4225
0xb48f01 lra(_IO_FILE*)
        /home/thiago/src/gcc/gcc/lra.c:2277
0xb08109 do_reload
        /home/thiago/src/gcc/gcc/ira.c:5391
0xb08109 execute
        /home/thiago/src/gcc/gcc/ira.c:5561

Original command-line:
g++ -c -pipe -pipe -march=core-avx2 -mno-rtm -mno-hle -std=c++0x -pthread -O3
-g -maccumulate-outgoing-args -O3 -fvisibility=hidden
-fvisibility-inlines-hidden -std=c++0x -flto -ffat-lto-objects -Wall -W -Werror
-Wno-error=cpp -Wno-error=deprecated-declarations -Wno-error=strict-overflow
-D_REENTRANT -fPIC -DQT_NO_USING_NAMESPACE -DQT_BUILD_CORE_LIB -DQT_BUILDING_QT
-DQT_NO_CAST_TO_ASCII -DQT_ASCII_CAST_WARNINGS -DQT_MOC_COMPAT
-DQT_USE_QSTRINGBUILDER -DQT_DEPRECATED_WARNINGS
-DQT_DISABLE_DEPRECATED_BEFORE=0x050000
-DELF_INTERPRETER=\"/lib64/ld-linux-x86-64.so.2\" -DQT_USE_ICU -DQT_CORE_LIB
-D_LARGEFILE64_SOURCE -D_LARGEFILE_SOURCE -DQT_NO_DEBUG
-I/home/thiago/src/qt/qt5/qtbase/mkspecs/linux-g++-optimised
-I/home/thiago/src/qt/qt5/qtbase/src/corelib -I../../include
-I../../include/QtCore -I../../include/QtCore/5.4.0
-I../../include/QtCore/5.4.0/QtCore -Iglobal
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/harfbuzz/src
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/md5
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/md4
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/sha3
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/forkfd -isystem
/usr/include/glib-2.0 -I/usr/lib64/glib-2.0/include -I.moc -I.
/home/thiago/src/qt/qt5/qtbase/src/corelib/tools/qstringargbuilder.cpp -o
.obj/qstringargbuilder.o

Minimal command-line for the .ii source:
g++ -c -march=core-avx2 -mno-rtm -mno-hle -std=c++0x -pthread -O3 -D_REENTRANT
-fPIC qstringargbuilder.ii -o .obj/qstringargbuilder.o

If I remove the -march=core-avx2 option and compile the original .cpp file,
there is no error.
>From gcc-bugs-return-468899-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org Sat Nov 29 04:39:39 2014
Return-Path: <gcc-bugs-return-468899-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Delivered-To: listarch-gcc-bugs@gcc.gnu.org
Received: (qmail 27225 invoked by alias); 29 Nov 2014 04:39:39 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Delivered-To: mailing list gcc-bugs@gcc.gnu.org
Received: (qmail 27197 invoked by uid 48); 29 Nov 2014 04:39:35 -0000
From: "thiago at kde dot org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug c++/64111] New: [5.0 regression] ICE: conversion of register to a different size
Date: Sat, 29 Nov 2014 04:39:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: new
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: c++
X-Bugzilla-Version: 5.0
X-Bugzilla-Keywords:
X-Bugzilla-Severity: normal
X-Bugzilla-Who: thiago at kde dot org
X-Bugzilla-Status: UNCONFIRMED
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields: bug_id short_desc product version bug_status bug_severity priority component assigned_to reporter
Message-ID: <bug-64111-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2014-11/txt/msg03371.txt.bz2
Content-length: 4513

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64111

            Bug ID: 64111
           Summary: [5.0 regression] ICE: conversion of register to a
                    different size
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: thiago at kde dot org

When compiling the attached preprocessed source, GCC 5 (trunk 218173) prints
the following ICE:

/home/thiago/src/qt/qt5/qtbase/src/corelib/io/qresource.cpp: In member function
‘QResourceRoot::findNode(QString const&, QLocale const&) const’:
/home/thiago/src/qt/qt5/qtbase/src/corelib/io/qresource.cpp:616:5: error:
conversion of register to a different size
 int QResourceRoot::findNode(const QString &_path, const QLocale &locale) const
     ^
VIEW_CONVERT_EXPR<vector(16) short unsigned int>(vect__655.937_1129);

vect__656.938_1127 = VIEW_CONVERT_EXPR<vector(16) short unsigned
int>(vect__655.937_1129);
/home/thiago/src/qt/qt5/qtbase/src/corelib/io/qresource.cpp:616:5: error:
conversion of register to a different size
VIEW_CONVERT_EXPR<vector(16) short unsigned int>(vect__655.937_1128);

vect__656.938_1126 = VIEW_CONVERT_EXPR<vector(16) short unsigned
int>(vect__655.937_1128);
/home/thiago/src/qt/qt5/qtbase/src/corelib/io/qresource.cpp:616:5: internal
compiler error: verify_gimple failed
0xcc5cd1 verify_gimple_in_cfg(function*, bool)
        /home/thiago/src/gcc/gcc/tree-cfg.c:5049
0xbcbe3f execute_function_todo
        /home/thiago/src/gcc/gcc/passes.c:1940
0xbcc713 execute_todo
        /home/thiago/src/gcc/gcc/passes.c:1997

Original command-line:
g++ -c -include .pch/Qt5Core -pipe -pipe -march=core-avx2 -mno-rtm -mno-hle
-std=c++0x -pthread -O3 -g -maccumulate-outgoing-args -O3 -fvisibility=hidden
-fvisibility-inlines-hidden -std=c++0x -flto -ffat-lto-objects -Wall -W -Werror
-Wno-error=cpp -Wno-error=deprecated-declarations -Wno-error=strict-overflow
-D_REENTRANT -fPIC -DQT_NO_USING_NAMESPACE -DQT_BUILD_CORE_LIB -DQT_BUILDING_QT
-DQT_NO_CAST_TO_ASCII -DQT_ASCII_CAST_WARNINGS -DQT_MOC_COMPAT
-DQT_USE_QSTRINGBUILDER -DQT_DEPRECATED_WARNINGS
-DQT_DISABLE_DEPRECATED_BEFORE=0x050000
-DELF_INTERPRETER=\"/lib64/ld-linux-x86-64.so.2\" -DQT_USE_ICU -DQT_CORE_LIB
-D_LARGEFILE64_SOURCE -D_LARGEFILE_SOURCE -DQT_NO_DEBUG
-I/home/thiago/src/qt/qt5/qtbase/mkspecs/linux-g++-optimised
-I/home/thiago/src/qt/qt5/qtbase/src/corelib -I../../include
-I../../include/QtCore -I../../include/QtCore/5.4.0
-I../../include/QtCore/5.4.0/QtCore -Iglobal
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/harfbuzz/src
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/md5
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/md4
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/sha3
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/forkfd -isystem
/usr/include/glib-2.0 -I/usr/lib64/glib-2.0/include -I.moc -I. -o
.obj/qresource.o /home/thiago/src/qt/qt5/qtbase/src/corelib/io/qresource.cpp

This error disappears when the -include $(PCH) option is removed. Here's the
command-line for the precompilation:
g++ -pipe -pipe -march=core-avx2 -mno-rtm -mno-hle -std=c++0x -pthread -O3 -g
-maccumulate-outgoing-args -O3 -fvisibility=hidden -fvisibility-inlines-hidden
-std=c++0x -flto -ffat-lto-objects -Wall -W -Werror -Wno-error=cpp
-Wno-error=deprecated-declarations -Wno-error=strict-overflow -D_REENTRANT
-fPIC -DQT_NO_USING_NAMESPACE -DQT_BUILD_CORE_LIB -DQT_BUILDING_QT
-DQT_NO_CAST_TO_ASCII -DQT_ASCII_CAST_WARNINGS -DQT_MOC_COMPAT
-DQT_USE_QSTRINGBUILDER -DQT_DEPRECATED_WARNINGS
-DQT_DISABLE_DEPRECATED_BEFORE=0x050000
-DELF_INTERPRETER=\"/lib64/ld-linux-x86-64.so.2\" -DQT_USE_ICU -DQT_CORE_LIB
-D_LARGEFILE64_SOURCE -D_LARGEFILE_SOURCE -DQT_NO_DEBUG
-I/home/thiago/src/qt/qt5/qtbase/mkspecs/linux-g++-optimised
-I/home/thiago/src/qt/qt5/qtbase/src/corelib -I../../include
-I../../include/QtCore -I../../include/QtCore/5.4.0
-I../../include/QtCore/5.4.0/QtCore -Iglobal
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/harfbuzz/src
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/md5
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/md4
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/sha3
-I/home/thiago/src/qt/qt5/qtbase/src/3rdparty/forkfd -isystem
/usr/include/glib-2.0 -I/usr/lib64/glib-2.0/include -I.moc -I. -x c++-header -c
/home/thiago/src/qt/qt5/qtbase/src/corelib/global/qt_pch.h -o
.pch/Qt5Core.gch/c++
>From gcc-bugs-return-468900-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org Sat Nov 29 04:40:28 2014
Return-Path: <gcc-bugs-return-468900-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Delivered-To: listarch-gcc-bugs@gcc.gnu.org
Received: (qmail 28020 invoked by alias); 29 Nov 2014 04:40:28 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Delivered-To: mailing list gcc-bugs@gcc.gnu.org
Received: (qmail 27984 invoked by uid 48); 29 Nov 2014 04:40:24 -0000
From: "thiago at kde dot org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug c++/64111] [5.0 regression] ICE: conversion of register to a different size
Date: Sat, 29 Nov 2014 04:40:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: c++
X-Bugzilla-Version: 5.0
X-Bugzilla-Keywords:
X-Bugzilla-Severity: normal
X-Bugzilla-Who: thiago at kde dot org
X-Bugzilla-Status: UNCONFIRMED
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields: attachments.created
Message-ID: <bug-64111-4-DqY5aBfIYW@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-64111-4@http.gcc.gnu.org/bugzilla/>
References: <bug-64111-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2014-11/txt/msg03372.txt.bz2
Content-length: 270

https://gcc.gnu.org/bugzilla/show_bug.cgi?idd111

--- Comment #1 from Thiago Macieira <thiago at kde dot org> ---
Created attachment 34136
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id4136&actioníit
Preprocessed output for the source file with the problem

[Bug c++/64110] ICE: Max. number of generated reload insns per insn is achieved (90)

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #1 from Thiago Macieira <thiago at kde dot org> ---
Created attachment 34138
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34138&action=edit
Preprocessed sources for the build error

[Bug c++/64110] ICE: Max. number of generated reload insns per insn is achieved (90)

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #2 from Thiago Macieira <thiago at kde dot org> ---
creduce reduced to the following testcase:

====
class A {
public:
  short &m_fn1() { return ucs; }
  short ucs;
};

class C {
public:
  int m_fn2();
} a;
A *b;
C c;
void fn1(A p1) {
  while (&a) {
    int d = c.m_fn2();
    for (int i; i < d; ++i)
      b++->m_fn1() = p1.m_fn1();
  }
}
====

[Bug c++/64110] ICE: Max. number of generated reload insns per insn is achieved (90)

[thiago at kde dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #3 from Thiago Macieira <thiago at kde dot org> ---
This reduced testcase starts (apparently) with an uninitialised int i, but the
same problem happens if you initialise to 0 (for (int i = 0; i < d; ++i))

[Bug c++/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-11-29
                 CC|                            |jakub at redhat dot com
   Target Milestone|---                         |5.0
            Summary|ICE: Max. number of         |[5 Regression] ICE: Max.
                   |generated reload insns per  |number of generated reload
                   |insn is achieved (90)       |insns per insn is achieved
                   |                            |(90)
     Ever confirmed|0                           |1

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> ---
It was caused by r216541.

[Bug c++/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Simplified testcase (-O3 -march=core-avx2):

int foo (void);
int a;
short *b;

void
bar (short x)
{
  while (a--)
    {
      int i, j = foo ();
      for (i = 0; i < j; ++i)
    *b++ = x;
    }
}

[Bug c++/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|jakub at redhat dot com            |vmakarov at gcc dot gnu.org

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, before LRA we have:
(insn 27 36 28 2 (set (reg:SI 178 [ x ])
        (reg:SI 5 di [ x ])) pr64110.C:7 90 {*movsi_internal}
     (expr_list:REG_DEAD (reg:SI 5 di [ x ])
        (nil)))
(insn 28 27 29 2 (set (reg/v:HI 177 [ x ])
        (subreg:HI (reg:SI 178 [ x ]) 0)) pr64110.C:7 92 {*movhi_internal}
     (expr_list:REG_DEAD (reg:SI 178 [ x ])
        (nil)))
(note 29 28 89 2 NOTE_INSN_FUNCTION_BEG)
(insn 89 29 247 2 (set (reg:V16HI 214 [ vect_cst_.21 ])
        (vec_duplicate:V16HI (reg/v:HI 177 [ x ]))) 4254 {*vec_dupv16hi}
     (nil))

and *vec_dupv16hi is:
(define_insn "*vec_dup<mode>"
  [(set (match_operand:AVX2_VEC_DUP_MODE 0 "register_operand" "=x,x,x")
        (vec_duplicate:AVX2_VEC_DUP_MODE
          (match_operand:<ssescalarmode> 1 "nonimmediate_operand" "m,x,!r")))]
  "TARGET_AVX2"

Thus, I don't see a bug on the md side, I would have expected LRA to just
use (vec_duplicate:V16HI (reg/v:HI 5 di)) on the rhs (yes, it will be split
after reload).
But, during the ICE I see:
(insn 27 36 28 2 (set (reg:SI 178 [ x ])
        (reg:SI 5 di [ x ])) pr64110.C:7 90 {*movsi_internal}
     (expr_list:REG_DEAD (reg:SI 5 di [ x ])
        (nil)))

(insn 28 27 29 2 (set (reg/v:HI 177 [ x ])
        (subreg:HI (reg:SI 178 [ x ]) 0)) pr64110.C:7 92 {*movhi_internal}
     (expr_list:REG_DEAD (reg:SI 178 [ x ])
        (nil)))

(note 29 28 262 2 NOTE_INSN_FUNCTION_BEG)

(insn 262 29 263 2 (set (reg/v:HI 218 [orig:177 x ] [177])
        (reg/v:HI 177 [ x ])) 92 {*movhi_internal}
     (nil))

(insn 263 262 264 2 (set (reg/v:HI 219 [orig:177 x ] [177])
        (reg/v:HI 218 [orig:177 x ] [177])) 92 {*movhi_internal}
     (nil))

(insn 264 263 265 2 (set (reg/v:HI 220 [orig:177 x ] [177])
        (reg/v:HI 219 [orig:177 x ] [177])) 92 {*movhi_internal}
     (nil))

...
(repeated many times for pseudos 221, 222, ... 305)
(insn 350 349 351 2 (set (reg/v:HI 306 [orig:177 x ] [177])
        (reg/v:HI 305 [orig:177 x ] [177])) 92 {*movhi_internal}
     (nil))

(insn 351 350 89 2 (set (reg/v:HI 217 [orig:177 x ] [177])
        (reg/v:HI 306 [orig:177 x ] [177])) 92 {*movhi_internal}
     (nil))

(insn 89 351 261 2 (set (reg:V16HI 216 [orig:214 vect_cst_.21 ] [214])
        (vec_duplicate:V16HI (reg/v:HI 217 [orig:177 x ] [177]))) 4254
{*vec_dupv16hi}
     (nil))

Vlad, can you please have a look?  Thanks.

[Bug c++/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #7 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #6)
> 
> Vlad, can you please have a look?  Thanks.

I'll work on this.  But the related code in LRA is very tricky.  It can take a
few days to fix it and a lot of testing too.

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mpolacek at gcc dot gnu.org
          Component|c++                         |rtl-optimization

--- Comment #8 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Recateogrizing.

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ra
             Target|                            |x86_64-*-*
           Priority|P3                          |P1

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #9 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
Author: vmakarov
Date: Fri Dec 12 20:11:10 2014
New Revision: 218688

URL: https://gcc.gnu.org/viewcvs?rev=218688&root=gcc&view=rev
Log:
2014-12-12  Vladimir Makarov  <vmakarov@redhat.com>

    PR target/64110
    * lra-constraints.c (process_alt_operands): Refuse alternative
    when reload pseudo of given class can not hold value of given
    mode.

2014-12-12  Vladimir Makarov  <vmakarov@redhat.com>

    PR target/64110
    * gcc.target/i386/pr64110.c: New.


Added:
    trunk/gcc/testsuite/gcc.target/i386/pr64110.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/lra-constraints.c
    trunk/gcc/testsuite/ChangeLog

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #10 from H.J. Lu <hjl.tools at gmail dot com> ---
It fails with -m32:

[hjl@gnu-6 gcc]$ /export/build/gnu/gcc/build-x86_64-linux/gcc/xgcc
-B/export/build/gnu/gcc/build-x86_64-linux/gcc/
/export/gnu/import/git/gcc/gcc/testsuite/gcc.target/i386/pr64110.c 
-fno-diagnostics-show-caret -fdiagnostics-color=never   -O3 -march=core-avx2 -S
 -m32  -o pr64110.s
/export/gnu/import/git/gcc/gcc/testsuite/gcc.target/i386/pr64110.c: In function
\u2018bar\u2019:
/export/gnu/import/git/gcc/gcc/testsuite/gcc.target/i386/pr64110.c:17:1:
internal compiler error: Max. number of generated reload insns per insn is
achieved (90)

0xb65d03 lra_constraints(bool)
    /export/gnu/import/git/gcc/gcc/lra-constraints.c:4256
0xb52ceb lra(_IO_FILE*)
    /export/gnu/import/git/gcc/gcc/lra.c:2277
0xb03357 do_reload
    /export/gnu/import/git/gcc/gcc/ira.c:5402
0xb03704 execute
    /export/gnu/import/git/gcc/gcc/ira.c:5573
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <http://gcc.gnu.org/bugs.html> for instructions.
[hjl@gnu-6 gcc]$

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Looking at the generated assembly, I see there:
        movw    %di, -80(%rbp)
        vpbroadcastw    -80(%rbp), %ymm1
        vmovdqa %ymm1, -80(%rbp)
I'd have expected
        vmovd   %edi, %xmm1
        vpbroadcastw    %xmm1, %ymm1
        vmovdqa %ymm1, -80(%rbp)
Is there some bug in the *vec_dupv16hi pattern that LRA can't use the !r
alternative?  %di should be valid nonimmediate_operand in HImode.  For e.g.
typedef unsigned short V __attribute__((vector_size (32)));
V foo (unsigned short x)
{
  return (V) { x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x };
}
we emit what I expect.  In the latter case, at *.ira we have:
(insn 2 5 3 2 (set (reg:SI 90 [ x ])
        (reg:SI 5 di [ x ])) pr64110-4.c:3 90 {*movsi_internal}
     (expr_list:REG_DEAD (reg:SI 5 di [ x ])
        (nil)))
(note 3 2 4 2 NOTE_INSN_DELETED)
(note 4 3 7 2 NOTE_INSN_FUNCTION_BEG)
(insn 7 4 12 2 (set (reg:V16HI 92 [ D.2291 ])
        (vec_duplicate:V16HI (subreg:HI (reg:SI 90 [ x ]) 0))) pr64110-4.c:4
4233 {*vec_dupv16hi}
     (expr_list:REG_DEAD (reg:SI 90 [ x ])
        (nil)))
and in the former case we have:
(insn 27 36 28 2 (set (reg:SI 178 [ x ])
        (reg:SI 5 di [ x ])) pr64110.c:10 90 {*movsi_internal}
     (expr_list:REG_DEAD (reg:SI 5 di [ x ])
        (nil)))
(insn 28 27 29 2 (set (reg/v:HI 177 [ x ])
        (subreg:HI (reg:SI 178 [ x ]) 0)) pr64110.c:10 92 {*movhi_internal}
     (expr_list:REG_DEAD (reg:SI 178 [ x ])
        (nil)))
(note 29 28 89 2 NOTE_INSN_FUNCTION_BEG)
(insn 89 29 247 2 (set (reg:V16HI 214 [ vect_cst_.21 ])
        (vec_duplicate:V16HI (reg/v:HI 177 [ x ]))) 4233 {*vec_dupv16hi}
     (nil))
which isn't that much different...

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #12 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
I've just fixed the problem reported by H.J.

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #13 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #11)
> Looking at the generated assembly, I see there:
>         movw    %di, -80(%rbp)
>         vpbroadcastw    -80(%rbp), %ymm1
>         vmovdqa %ymm1, -80(%rbp)
> I'd have expected
>         vmovd   %edi, %xmm1
>         vpbroadcastw    %xmm1, %ymm1
>         vmovdqa %ymm1, -80(%rbp)
> Is there some bug in the *vec_dupv16hi pattern that LRA can't use the !r
> alternative?  %di should be valid nonimmediate_operand in HImode.  For e.g.

I know about the problem.  It is coming from a different treatment of ! in LRA
and reload.  I am going to work on this but as the fix would be in very
sensitive LRA part. I am afraid to fix it before my vacation.  So probably it
will be solved only in January.

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #14 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Vladimir Makarov from comment #13)
> (In reply to Jakub Jelinek from comment #11)
> > Looking at the generated assembly, I see there:
> >         movw    %di, -80(%rbp)
> >         vpbroadcastw    -80(%rbp), %ymm1
> >         vmovdqa %ymm1, -80(%rbp)
> > I'd have expected
> >         vmovd   %edi, %xmm1
> >         vpbroadcastw    %xmm1, %ymm1
> >         vmovdqa %ymm1, -80(%rbp)
> > Is there some bug in the *vec_dupv16hi pattern that LRA can't use the !r
> > alternative?  %di should be valid nonimmediate_operand in HImode.  For e.g.
> 
> I know about the problem.  It is coming from a different treatment of ! in
> LRA and reload.  I am going to work on this but as the fix would be in very
> sensitive LRA part. I am afraid to fix it before my vacation.  So probably
> it will be solved only in January.

Np, and thanks a lot.

The ! has been added by me for PR63594, so it isn't there from the era when
i?86 backend was using reload.  If there is a better way to express that RA
should prefer to use memory or xmm register and only use r constraint if it
already is in a r register and doesn't need to be reloaded, I can use that. 
Whether it is ?, ??? or something else.  ! description in gcc docs just fitted
most what I wanted...

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #15 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #14)
> (In reply to Vladimir Makarov from comment #13)
> > (In reply to Jakub Jelinek from comment #11)
> > > Looking at the generated assembly, I see there:
> > >         movw    %di, -80(%rbp)
> > >         vpbroadcastw    -80(%rbp), %ymm1
> > >         vmovdqa %ymm1, -80(%rbp)
> > > I'd have expected
> > >         vmovd   %edi, %xmm1
> > >         vpbroadcastw    %xmm1, %ymm1
> > >         vmovdqa %ymm1, -80(%rbp)
> > > Is there some bug in the *vec_dupv16hi pattern that LRA can't use the !r
> > > alternative?  %di should be valid nonimmediate_operand in HImode.  For e.g.
> > 
> > I know about the problem.  It is coming from a different treatment of ! in
> > LRA and reload.  I am going to work on this but as the fix would be in very
> > sensitive LRA part. I am afraid to fix it before my vacation.  So probably
> > it will be solved only in January.
> 
> Np, and thanks a lot.
> 
> The ! has been added by me for PR63594, so it isn't there from the era when
> i?86 backend was using reload.  If there is a better way to express that RA
> should prefer to use memory or xmm register and only use r constraint if it
> already is in a r register and doesn't need to be reloaded, I can use that. 
> Whether it is ?, ??? or something else.  ! description in gcc docs just
> fitted most what I wanted...

Jakub, I've checked this particular test and LRA works in the same way as
reload and as specified in the GCC documentation which mainly says that '!' is
ignored only when *all* operands satisfy the alternative constraints.  In this
case there is one reload therefore alternative with '!r' is rejected.  I
understand that for this insn the desirable behaviour is to reject alternative
only when operand with '!r' needs a reload.

I could implement such functionality.  The following unreserved special
characters can be used as such modifier with the new functionality:
   @ $ ^ -

What character would you prefer, Jakub?

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #16 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Vladimir Makarov from comment #15)
> Jakub, I've checked this particular test and LRA works in the same way as
> reload and as specified in the GCC documentation which mainly says that '!'
> is ignored only when *all* operands satisfy the alternative constraints.  In
> this case there is one reload therefore alternative with '!r' is rejected. 
> I understand that for this insn the desirable behaviour is to reject
> alternative only when operand with '!r' needs a reload.
> 
> I could implement such functionality.  The following unreserved special
> characters can be used as such modifier with the new functionality:
>    @ $ ^ -
> 
> What character would you prefer, Jakub?

No strong opinion, perhaps @ ?  Jeff, what would you prefer?

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com

--- Comment #17 from Jeffrey A. Law <law at redhat dot com> ---
No strong opinions.  I wouldn't be surprised if we find other uses for this
capability, so pick one and run with it.

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

--- Comment #18 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
Author: vmakarov
Date: Thu Jan 15 20:26:19 2015
New Revision: 219683

URL: https://gcc.gnu.org/viewcvs?rev=219683&root=gcc&view=rev
Log:
2015-01-15  Vladimir Makarov  <vmakarov@redhat.com>

    PR rtl-optimization/64110
    * stmt.c (parse_output_constraint): Process '^' and '$'.
    (parse_input_constraint): Ditto.
    * lra-constraints.c (process_alt_operands): Process the new
    constraints.
    * ira-costs.c (record_reg_classes): Process the new constraint
    '^'.
    * genoutput.c (indep_constraints): Add '^' and '$'.
    * config/i386/sse.md (*vec_dup<mode>): Use '$' instead of '!'.
    * doc/md.texi: Add description of the new constraints.

2015-01-15  Vladimir Makarov  <vmakarov@redhat.com>

    PR rtl-optimization/64110
    * gcc.target/i386/pr64110.c: Add scan-assembler.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/sse.md
    trunk/gcc/doc/md.texi
    trunk/gcc/genoutput.c
    trunk/gcc/ira-costs.c
    trunk/gcc/lra-constraints.c
    trunk/gcc/stmt.c
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.target/i386/pr64110.c

[Bug rtl-optimization/64110] [5 Regression] ICE: Max. number of generated reload insns per insn is achieved (90)

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64110

Jeffrey A. Law <law at redhat dot com> changed:

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

--- Comment #19 from Jeffrey A. Law <law at redhat dot com> ---
Vlad fixed this on the trunk.