* [PATCH] Enable fixing PR64715
@ 2015-03-20 11:38 Richard Biener
2015-03-20 12:30 ` Jakub Jelinek
0 siblings, 1 reply; 2+ messages in thread
From: Richard Biener @ 2015-03-20 11:38 UTC (permalink / raw)
To: gcc-patches; +Cc: Jakub Jelinek
This removes a road-block towards fixing PR64715 - the gimplifier
performing an undesired "simplification" of &X + CST to
&MEM[&X, CST]. IMHO this is premature there.
The fallout is small - two testcases are no longer simplified at -O0
(who cares), and devirt-40.C shows that SCEV adheres too closely
to the fold view of type compatibility (I had a fix for that in
my dev tree for quite some time it seeems). Of course the latter
means that there might be more fallout in code that builds
GENERIC and then re-gimplifies that, but I can't think of anything
that wouldn't be fixed with followup passes.
Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Ok?
Thanks,
Richard.
2015-03-20 Richard Biener <rguenther@suse.de>
PR middle-end/64715
* tree-chrec.c (chrec_fold_poly_cst): Use useless_type_conversion_p
for type comparison and gcc_checking_assert.
(chrec_fold_plus_poly_poly): Likewise.
(chrec_fold_multiply_poly_poly): Likewise.
(chrec_convert_1): Likewise.
* gimplify.c (gimplify_expr): Remove premature folding of
&X + CST to &MEM[&X, CST].
* gcc.dg/pr15347.c: Use -O.
* c-c++-common/pr19807-1.c: Likewise.
Index: gcc/tree-chrec.c
===================================================================
--- gcc/tree-chrec.c (revision 221515)
+++ gcc/tree-chrec.c (working copy)
@@ -78,8 +78,8 @@ chrec_fold_poly_cst (enum tree_code code
gcc_assert (poly);
gcc_assert (cst);
gcc_assert (TREE_CODE (poly) == POLYNOMIAL_CHREC);
- gcc_assert (!is_not_constant_evolution (cst));
- gcc_assert (type == chrec_type (poly));
+ gcc_checking_assert (!is_not_constant_evolution (cst));
+ gcc_checking_assert (useless_type_conversion_p (type, chrec_type (poly)));
switch (code)
{
@@ -124,10 +124,11 @@ chrec_fold_plus_poly_poly (enum tree_cod
gcc_assert (TREE_CODE (poly0) == POLYNOMIAL_CHREC);
gcc_assert (TREE_CODE (poly1) == POLYNOMIAL_CHREC);
if (POINTER_TYPE_P (chrec_type (poly0)))
- gcc_assert (ptrofftype_p (chrec_type (poly1)));
+ gcc_checking_assert (ptrofftype_p (chrec_type (poly1))
+ && useless_type_conversion_p (type, chrec_type (poly0)));
else
- gcc_assert (chrec_type (poly0) == chrec_type (poly1));
- gcc_assert (type == chrec_type (poly0));
+ gcc_checking_assert (useless_type_conversion_p (type, chrec_type (poly0))
+ && useless_type_conversion_p (type, chrec_type (poly1)));
/*
{a, +, b}_1 + {c, +, d}_2 -> {{a, +, b}_1 + c, +, d}_2,
@@ -208,8 +209,8 @@ chrec_fold_multiply_poly_poly (tree type
gcc_assert (poly1);
gcc_assert (TREE_CODE (poly0) == POLYNOMIAL_CHREC);
gcc_assert (TREE_CODE (poly1) == POLYNOMIAL_CHREC);
- gcc_assert (chrec_type (poly0) == chrec_type (poly1));
- gcc_assert (type == chrec_type (poly0));
+ gcc_checking_assert (useless_type_conversion_p (type, chrec_type (poly0))
+ && useless_type_conversion_p (type, chrec_type (poly1)));
/* {a, +, b}_1 * {c, +, d}_2 -> {c*{a, +, b}_1, +, d}_2,
{a, +, b}_2 * {c, +, d}_1 -> {a*{c, +, d}_1, +, b}_2,
@@ -1352,7 +1353,7 @@ chrec_convert_1 (tree type, tree chrec,
return chrec;
ct = chrec_type (chrec);
- if (ct == type)
+ if (useless_type_conversion_p (type, ct))
return chrec;
if (!evolution_function_is_affine_p (chrec))
Index: gcc/testsuite/gcc.dg/pr15347.c
===================================================================
--- gcc/testsuite/gcc.dg/pr15347.c (revision 221530)
+++ gcc/testsuite/gcc.dg/pr15347.c (working copy)
@@ -1,4 +1,5 @@
/* { dg-do link } */
+/* { dg-options "-O" } */
extern void link_error (void);
int
Index: gcc/testsuite/c-c++-common/pr19807-1.c
===================================================================
--- gcc/testsuite/c-c++-common/pr19807-1.c (revision 221530)
+++ gcc/testsuite/c-c++-common/pr19807-1.c (working copy)
@@ -1,4 +1,5 @@
/* { dg-do link } */
+/* { dg-options "-O" } */
extern void link_error(void);
int main()
Index: gcc/gimplify.c
===================================================================
*** gcc/gimplify.c (revision 221514)
--- gcc/gimplify.c (working copy)
*************** gimplify_expr (tree *expr_p, gimple_seq
*** 8524,8546 ****
post_p, is_gimple_val, fb_rvalue);
recalculate_side_effects (*expr_p);
ret = MIN (r0, r1);
- /* Convert &X + CST to invariant &MEM[&X, CST]. Do this
- after gimplifying operands - this is similar to how
- it would be folding all gimplified stmts on creation
- to have them canonicalized, which is what we eventually
- should do anyway. */
- if (TREE_CODE (TREE_OPERAND (*expr_p, 1)) == INTEGER_CST
- && is_gimple_min_invariant (TREE_OPERAND (*expr_p, 0)))
- {
- *expr_p = build_fold_addr_expr_with_type_loc
- (input_location,
- fold_build2 (MEM_REF, TREE_TYPE (TREE_TYPE (*expr_p)),
- TREE_OPERAND (*expr_p, 0),
- fold_convert (ptr_type_node,
- TREE_OPERAND (*expr_p, 1))),
- TREE_TYPE (*expr_p));
- ret = MIN (ret, GS_OK);
- }
break;
}
--- 8524,8529 ----
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Enable fixing PR64715
2015-03-20 11:38 [PATCH] Enable fixing PR64715 Richard Biener
@ 2015-03-20 12:30 ` Jakub Jelinek
0 siblings, 0 replies; 2+ messages in thread
From: Jakub Jelinek @ 2015-03-20 12:30 UTC (permalink / raw)
To: Richard Biener; +Cc: gcc-patches
On Fri, Mar 20, 2015 at 12:38:51PM +0100, Richard Biener wrote:
>
> This removes a road-block towards fixing PR64715 - the gimplifier
> performing an undesired "simplification" of &X + CST to
> &MEM[&X, CST]. IMHO this is premature there.
>
> The fallout is small - two testcases are no longer simplified at -O0
> (who cares), and devirt-40.C shows that SCEV adheres too closely
> to the fold view of type compatibility (I had a fix for that in
> my dev tree for quite some time it seeems). Of course the latter
> means that there might be more fallout in code that builds
> GENERIC and then re-gimplifies that, but I can't think of anything
> that wouldn't be fixed with followup passes.
>
> Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
>
> Ok?
LGTM.
> 2015-03-20 Richard Biener <rguenther@suse.de>
>
> PR middle-end/64715
> * tree-chrec.c (chrec_fold_poly_cst): Use useless_type_conversion_p
> for type comparison and gcc_checking_assert.
> (chrec_fold_plus_poly_poly): Likewise.
> (chrec_fold_multiply_poly_poly): Likewise.
> (chrec_convert_1): Likewise.
> * gimplify.c (gimplify_expr): Remove premature folding of
> &X + CST to &MEM[&X, CST].
>
> * gcc.dg/pr15347.c: Use -O.
> * c-c++-common/pr19807-1.c: Likewise.
Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2015-03-20 12:30 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-03-20 11:38 [PATCH] Enable fixing PR64715 Richard Biener
2015-03-20 12:30 ` Jakub Jelinek
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).