* [PATCH] Fix PRs 33693, 33695 and 33697
@ 2007-10-08 15:22 Richard Guenther
2007-10-08 16:43 ` Andrew Pinski
0 siblings, 1 reply; 6+ messages in thread
From: Richard Guenther @ 2007-10-08 15:22 UTC (permalink / raw)
To: gcc-patches
More fold-const.c fallout wrt enabled type checking.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to mainline.
Richard.
2007-10-08 Richard Guenther <rguenther@suse.de>
PR middle-end/33693
PR middle-end/33695
PR middle-end/33697
* fold-const.c (fold_binary): Use correct types in folding
of a * (1 << b) to (a << b). Likewise for ~A & ~B to ~(A | B)
and building of RROTATE_EXPR.
* gcc.dg/pr33693.c: New testcase.
* gcc.dg/pr33695.c: Likewise.
* gcc.dg/pr33697.c: Likewise.
Index: fold-const.c
===================================================================
*** fold-const.c (revision 129036)
--- fold-const.c (working copy)
*************** fold_binary (enum tree_code code, tree t
*** 10344,10359 ****
&& (tem = negate_expr (arg1)) != arg1
&& !TREE_OVERFLOW (tem))
return fold_build2 (MULT_EXPR, type,
! negate_expr (arg0), tem);
/* (a * (1 << b)) is (a << b) */
if (TREE_CODE (arg1) == LSHIFT_EXPR
&& integer_onep (TREE_OPERAND (arg1, 0)))
! return fold_build2 (LSHIFT_EXPR, type, arg0,
TREE_OPERAND (arg1, 1));
if (TREE_CODE (arg0) == LSHIFT_EXPR
&& integer_onep (TREE_OPERAND (arg0, 0)))
! return fold_build2 (LSHIFT_EXPR, type, arg1,
TREE_OPERAND (arg0, 1));
strict_overflow_p = false;
--- 10344,10359 ----
&& (tem = negate_expr (arg1)) != arg1
&& !TREE_OVERFLOW (tem))
return fold_build2 (MULT_EXPR, type,
! fold_convert (type, negate_expr (arg0)), tem);
/* (a * (1 << b)) is (a << b) */
if (TREE_CODE (arg1) == LSHIFT_EXPR
&& integer_onep (TREE_OPERAND (arg1, 0)))
! return fold_build2 (LSHIFT_EXPR, type, op0,
TREE_OPERAND (arg1, 1));
if (TREE_CODE (arg0) == LSHIFT_EXPR
&& integer_onep (TREE_OPERAND (arg0, 0)))
! return fold_build2 (LSHIFT_EXPR, type, op1,
TREE_OPERAND (arg0, 1));
strict_overflow_p = false;
*************** fold_binary (enum tree_code code, tree t
*** 11003,11010 ****
{
return fold_build1 (BIT_NOT_EXPR, type,
build2 (BIT_IOR_EXPR, type,
! TREE_OPERAND (arg0, 0),
! TREE_OPERAND (arg1, 0)));
}
/* If arg0 is derived from the address of an object or function, we may
--- 11003,11012 ----
{
return fold_build1 (BIT_NOT_EXPR, type,
build2 (BIT_IOR_EXPR, type,
! fold_convert (type,
! TREE_OPERAND (arg0, 0)),
! fold_convert (type,
! TREE_OPERAND (arg1, 0))));
}
/* If arg0 is derived from the address of an object or function, we may
*************** fold_binary (enum tree_code code, tree t
*** 11540,11546 ****
tree tem = build_int_cst (TREE_TYPE (arg1),
GET_MODE_BITSIZE (TYPE_MODE (type)));
tem = const_binop (MINUS_EXPR, tem, arg1, 0);
! return fold_build2 (RROTATE_EXPR, type, arg0, tem);
}
/* If we have a rotate of a bit operation with the rotate count and
--- 11542,11548 ----
tree tem = build_int_cst (TREE_TYPE (arg1),
GET_MODE_BITSIZE (TYPE_MODE (type)));
tem = const_binop (MINUS_EXPR, tem, arg1, 0);
! return fold_build2 (RROTATE_EXPR, type, op0, tem);
}
/* If we have a rotate of a bit operation with the rotate count and
Index: testsuite/gcc.dg/pr33693.c
===================================================================
*** testsuite/gcc.dg/pr33693.c (revision 0)
--- testsuite/gcc.dg/pr33693.c (revision 0)
***************
*** 0 ****
--- 1,8 ----
+ /* { dg-do compile } */
+
+ /* This used to ICE with type-checking enabled. */
+
+ unsigned long modify_field (unsigned long mask, long fieldval)
+ {
+ return (~fieldval & ~mask);
+ }
Index: testsuite/gcc.dg/pr33695.c
===================================================================
*** testsuite/gcc.dg/pr33695.c (revision 0)
--- testsuite/gcc.dg/pr33695.c (revision 0)
***************
*** 0 ****
--- 1,8 ----
+ /* { dg-do compile } */
+
+ /* We used to ICE with type-checking enabled. */
+
+ unsigned int bfstages(int M, float *Utbl, int Ustride)
+ {
+ return ((unsigned int) 1 << M) * Ustride;
+ }
Index: testsuite/gcc.dg/pr33697.c
===================================================================
*** testsuite/gcc.dg/pr33697.c (revision 0)
--- testsuite/gcc.dg/pr33697.c (revision 0)
***************
*** 0 ****
--- 1,10 ----
+ /* { dg-do compile } */
+
+ /* We used to ICE for this with type-checking enabled. */
+
+ typedef signed short gint16;
+ typedef unsigned short guint16;
+ gint16 dissect_old_pflog(gint16 rnr)
+ {
+ return (guint16) ((guint16) ((guint16)rnr >> 8) | (guint16) ((guint16)rnr << 8));
+ }
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix PRs 33693, 33695 and 33697
2007-10-08 15:22 [PATCH] Fix PRs 33693, 33695 and 33697 Richard Guenther
@ 2007-10-08 16:43 ` Andrew Pinski
2007-10-09 7:57 ` Richard Guenther
0 siblings, 1 reply; 6+ messages in thread
From: Andrew Pinski @ 2007-10-08 16:43 UTC (permalink / raw)
To: Richard Guenther; +Cc: gcc-patches
On 10/8/07, Richard Guenther <rguenther@suse.de> wrote:
> * gcc.dg/pr33693.c: New testcase.
> * gcc.dg/pr33695.c: Likewise.
> * gcc.dg/pr33697.c: Likewise.
I think these tests should have been in torture.
Thanks,
Andrew Pinski
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix PRs 33693, 33695 and 33697
2007-10-08 16:43 ` Andrew Pinski
@ 2007-10-09 7:57 ` Richard Guenther
2007-10-09 10:35 ` Andrew Pinski
0 siblings, 1 reply; 6+ messages in thread
From: Richard Guenther @ 2007-10-09 7:57 UTC (permalink / raw)
To: Andrew Pinski; +Cc: gcc-patches
On Mon, 8 Oct 2007, Andrew Pinski wrote:
> On 10/8/07, Richard Guenther <rguenther@suse.de> wrote:
> > * gcc.dg/pr33693.c: New testcase.
> > * gcc.dg/pr33695.c: Likewise.
> > * gcc.dg/pr33697.c: Likewise.
>
> I think these tests should have been in torture.
I thought about this, and types are verified only after gimplification,
so optimization doesn't matter.
Richard.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix PRs 33693, 33695 and 33697
2007-10-09 7:57 ` Richard Guenther
@ 2007-10-09 10:35 ` Andrew Pinski
2007-10-09 11:04 ` Richard Guenther
0 siblings, 1 reply; 6+ messages in thread
From: Andrew Pinski @ 2007-10-09 10:35 UTC (permalink / raw)
To: Richard Guenther; +Cc: gcc-patches
On 10/9/07, Richard Guenther <rguenther@suse.de> wrote:
> I thought about this, and types are verified only after gimplification,
> so optimization doesn't matter.
But we should be doing more than just after gimplification anyways. I
don't know why you did not add it as a normal verify pass.
-- Pinski
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix PRs 33693, 33695 and 33697
2007-10-09 10:35 ` Andrew Pinski
@ 2007-10-09 11:04 ` Richard Guenther
2007-10-09 18:15 ` Andrew Pinski
0 siblings, 1 reply; 6+ messages in thread
From: Richard Guenther @ 2007-10-09 11:04 UTC (permalink / raw)
To: Andrew Pinski; +Cc: gcc-patches
On Tue, 9 Oct 2007, Andrew Pinski wrote:
> On 10/9/07, Richard Guenther <rguenther@suse.de> wrote:
> > I thought about this, and types are verified only after gimplification,
> > so optimization doesn't matter.
>
> But we should be doing more than just after gimplification anyways. I
> don't know why you did not add it as a normal verify pass.
Because it falls all over ;) And the immediate goal was to get the
interface between FEs and the middle-end clean, not to fix every error
in the middle-end itself (ok, we catch a lot of fold bugs this way).
But yes, this is future work -- which I delayed until after the tuples
hickup.
Richard.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] Fix PRs 33693, 33695 and 33697
2007-10-09 11:04 ` Richard Guenther
@ 2007-10-09 18:15 ` Andrew Pinski
0 siblings, 0 replies; 6+ messages in thread
From: Andrew Pinski @ 2007-10-09 18:15 UTC (permalink / raw)
To: Richard Guenther; +Cc: gcc-patches
On 10/9/07, Richard Guenther <rguenther@suse.de> wrote:
> But yes, this is future work -- which I delayed until after the tuples
> hickup.
One I doubt tuples will be ready any time soon anyways. I am willing
to say it won't be ready until at least 4.5 because of expand changes
and getting expand back where it was before tuples.
--Pinski
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2007-10-09 18:15 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2007-10-08 15:22 [PATCH] Fix PRs 33693, 33695 and 33697 Richard Guenther
2007-10-08 16:43 ` Andrew Pinski
2007-10-09 7:57 ` Richard Guenther
2007-10-09 10:35 ` Andrew Pinski
2007-10-09 11:04 ` Richard Guenther
2007-10-09 18:15 ` Andrew Pinski
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).