[Bug c/17297] New: ICE with FP vector constructor containing qnan calculation

[Bug c/17297] New: ICE with FP vector constructor containing qnan calculation

[wilson at gcc dot gnu dot org]

Trying to initialize a floating point vector with an FP expression that is
constant, but is not evaluated at compile time, leads to an ICE.  With checking
enabled, it dies in build_vector as called from digest_init.  Without checking
enabled, it dies much later, in an immed_double_const call.

The immediate problem is that digest_init calls build_vector if it has a vector
with a constant value, when it should call build_vector only if it has a vector
of actual constants.  However, the real problem might be elsewhere.  I haven't
fully analyzed the problem.

This testcase reproduces the bug on x86_64 with -O and on mipsisa64 with -O
-mpaired-single.

I can also reproduce the bug with gcc-3.4.x, so it has been around for a while.

-- 
           Summary: ICE with FP vector constructor containing qnan
                    calculation
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: wilson at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
  GCC host triplet: x86_64-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] ICE with FP vector constructor containing qnan calculation

[wilson at gcc dot gnu dot org]

------- Additional Comments From wilson at gcc dot gnu dot org  2004-09-02 19:00 -------
Created an attachment (id=7035)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7035&action=view)
Vector testcase with qnan constructor that generates an ICE.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.5 Regression] ICE with FP vector constructor containing qnan calculation

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-09 03:46 -------
: Search converges between 2004-09-02-trunk (#532) and 2004-09-04-trunk (#533).
Confirmed, a regression.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |ice-on-valid-code
   Last reconfirmed|0000-00-00 00:00:00         |2004-09-09 03:46:43
               date|                            |
            Summary|ICE with FP vector          |[3.5 Regression] ICE with FP
                   |constructor containing qnan |vector constructor
                   |calculation                 |containing qnan calculation
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.5 Regression] ICE with FP vector constructor containing qnan calculation

[wilson at tuliptree dot org]

------- Additional Comments From wilson at tuliptree dot org  2004-09-09 18:18 -------
Subject: Re:  [3.5 Regression] ICE with FP vector constructor
	containing qnan calculation

On Wed, 2004-09-08 at 20:46, pinskia at gcc dot gnu dot org wrote:
> ------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-09 03:46 -------
> : Search converges between 2004-09-02-trunk (#532) and 2004-09-04-trunk (#533).
> Confirmed, a regression.

This is incorrect.  If you check, you will see that this is broken in
gcc-3.4.  I believe this has always been broken.

Unfortunately, 5 months ago, there was a typo in a patch from Zdenek
that accidentally broke the treatment of FP.  This patch accidentally
turned off flag_trapping_math.  When that happened, this testcasse
started working by accident.  When I fixed Zdenek's typo last week, this
testcase started failing again.  This inconsistency is incidentally how
I managed to find Zdenek's typo.  Reverting my patch is definitely not
the correct solution to this problem.  This testcase is supposed to
fail.

At least, that is, until someone fixes the underlying problem, which I
believe if a C front end bug in the vector initializer support.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.5 Regression] ICE with FP vector constructor containing qnan calculation

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-09 18:21 -------
It worked for me in 3.4.0 so ....

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/3.5 Regression] ICE with FP vector constructor containing qnan calculation

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-09 18:27 -------
With 3.4 branch we do get an ICE so this is also a 3.4 regression.
: Search converges between 2004-06-01-3.4 (#8) and 2004-06-15-3.4 (#9).

input.c: In function `main':
input.c:7: internal compiler error: in immed_double_const, at emit-rtl.c:466
Please submit a full bug report,
with preprocessed source if appropriate.
See  for instructions.

So with -ftraping-math turned on, we get another date on the mainline:
: Search converges between 2004-02-01-trunk (#445) and 2004-03-01-trunk (#446).
And a different ICE:
input.c: In function `main':
input.c:13: internal compiler error: in subreg_hard_regno, at emit-rtl.c:955
Please submit a full bug report,
with preprocessed source if appropriate.
See  for instructions.

So this is still a regression.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[3.5 Regression] ICE with FP|[3.4/3.5 Regression] ICE
                   |vector constructor          |with FP vector constructor
                   |containing qnan calculation |containing qnan calculation
   Target Milestone|3.5.0                       |3.4.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/3.5 Regression] ICE with FP vector constructor containing qnan calculation

[wilson at tuliptree dot org]

------- Additional Comments From wilson at tuliptree dot org  2004-09-09 19:56 -------
Subject: Re:  [3.5 Regression] ICE with FP vector constructor
	containing qnan calculation

On Thu, 2004-09-09 at 11:21, pinskia at gcc dot gnu dot org wrote:
> ------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-09 18:21 -------
> It worked for me in 3.4.0 so ....

You are right, this does not fail with a gcc-3.4.0 tree.  It does
however fail with gcc-3.4.1.  This was "fixed" by a C front end patch by
Roger Sayle in June.  See PR 14649.

I am not sure whether the regression tester is giving us any useful info
in this case.  There are too many different bugs interacting in
different ways to be sure.  Since mainline in February did not have
Roger's June C front end patch, the failure found by the regression
tester might be a different one than what we have now.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-09 20:30 -------
It does not matter if the problem is a latent bug or not because it is an user visible regression.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[coyote at coyotegulch dot com]

------- Additional Comments From coyote at coyotegulch dot com  2004-09-21 12:49 -------
So what is the resolution on this? Is the 3.4 patch going to be moved to
mainline, or is this being handled some other way?

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |coyote at coyotegulch dot
                   |                            |com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[wilson at tuliptree dot org]

------- Additional Comments From wilson at tuliptree dot org  2004-09-21 21:33 -------
Subject: Re:  [3.4/4.0 Regression] ICE with FP vector
	constructor containing qnan calculation

On Tue, 2004-09-21 at 05:49, coyote at coyotegulch dot com wrote:
> ------- Additional Comments From coyote at coyotegulch dot com  2004-09-21 12:49 -------
> So what is the resolution on this? Is the 3.4 patch going to be moved to
> mainline, or is this being handled some other way?

No patch has been written yet.

There is a complex set of circumstances that cause this to fail with
some gcc versions but not with others.  None of these are patches that
fix the problem.  They are just patches that alternatively hide or
expose the problem.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[mmitchel at gcc dot gnu dot org]

------- Additional Comments From mmitchel at gcc dot gnu dot org  2004-11-01 00:44 -------
Postponed until GCC 3.4.4.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4.3                       |3.4.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal
           Keywords|                            |ice-checking


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-13 18:17 -------
Patch here: <http://gcc.gnu.org/ml/gcc-patches/2005-01/msg00739.html>.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-01-19 09:27 -------
Subject: Bug 17297

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	jakub@gcc.gnu.org	2005-01-19 09:27:24

Modified files:
	gcc            : ChangeLog c-typeck.c gimplify.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.c-torture/compile: 20050113-1.c 
	gcc/testsuite/gcc.dg: 20050113-1.c 

Log message:
	PR c/17297
	* c-typeck.c (digest_init): Only call build_vector if all constructor
	elements are *_CST nodes.
	* gimplify.c (gimplify_init_constructor): Likewise.
	
	* gcc.c-torture/compile/20050113-1.c: New testcase.
	
	PR middle-end/19164
	* c-typeck.c (digest_init): Only call build_vector if inside_init
	is a CONSTRUCTOR.
	
	* gcc.dg/20050113-1.c: New testcase.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7183&r2=2.7184
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/c-typeck.c.diff?cvsroot=gcc&r1=1.410&r2=1.411
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/gimplify.c.diff?cvsroot=gcc&r1=2.103&r2=2.104
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.4907&r2=1.4908
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.c-torture/compile/20050113-1.c.diff?cvsroot=gcc&r1=NONE&r2=1.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/20050113-1.c.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-01-19 09:45 -------
Subject: Bug 17297

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	gcc-3_4-branch
Changes by:	jakub@gcc.gnu.org	2005-01-19 09:44:49

Modified files:
	gcc            : ChangeLog c-typeck.c 
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gcc.c-torture/compile: 20050113-1.c 
	gcc/testsuite/gcc.dg: 20050113-1.c 

Log message:
	PR c/17297
	* c-typeck.c (digest_init): Only call build_vector if all constructor
	elements are *_CST nodes.
	
	* gcc.c-torture/compile/20050113-1.c: New testcase.
	
	PR middle-end/19164
	* c-typeck.c (digest_init): Only call build_vector if inside_init
	is a CONSTRUCTOR.
	
	* gcc.dg/20050113-1.c: New testcase.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=2.2326.2.777&r2=2.2326.2.778
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/c-typeck.c.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.272.2.12&r2=1.272.2.13
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=1.3389.2.346&r2=1.3389.2.347
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.c-torture/compile/20050113-1.c.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=NONE&r2=1.1.2.1
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gcc.dg/20050113-1.c.diff?cvsroot=gcc&only_with_tag=gcc-3_4-branch&r1=NONE&r2=1.1.2.1



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297

[Bug c/17297] [3.4/4.0 Regression] ICE with FP vector constructor containing qnan calculation

[jakub at gcc dot gnu dot org]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-01-19 09:49 -------
Fixed in CVS.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17297