This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Fix an --enable-checking=rtl issue in var-tracking (PR debug/45849)
- From: Jakub Jelinek <jakub at redhat dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: Mon, 4 Oct 2010 17:42:11 +0200
- Subject: [PATCH] Fix an --enable-checking=rtl issue in var-tracking (PR debug/45849)
- Reply-to: Jakub Jelinek <jakub at redhat dot com>
Hi!
As swap_commutative_operands_p depends on REG_POINTER, and whether some
reg is REG_POINTER might change during var-tracking, simplification of
expressions that are rtx_equal_p might lead into expressions that aren't
rtx_equal_p any longer. This breaks ENABLE_RTL_CHECKING var-tracking
verification that cur_loc_changed flag works correctly.
As I think that checking is still useful, and I think it is unlikely to
ensure that there won't be REG_POINTER changes through the function on one
REG, either we'd need to modify swap_commutative_operand_p etc.
to disregard REG_POINTER/MEM_POINTER in debug insns/var-tracking (debug info
certainly doesn't care about that stuff), or, as done in this patch, simply
try harder in the var-tracking --enable-checking=rtl check if the simple
check fails. As it didn't hit until now, I'm not worried too much it will
need to be performed very often.
Bootstrapped/regtested on x86_64-linux and i686-linux. Ok for trunk?
2010-10-04 Jakub Jelinek <jakub@redhat.com>
PR debug/45849
* var-tracking.c (strip_pointer_flags): New function.
(emit_note_insn_var_location): If rtx_equal_p check failed,
retry on locations simplified with simplify_replace_fn_rtx
and strip_pointer_flags as its callback.
* gcc.dg/debug/pr45849.c: New test.
--- gcc/var-tracking.c.jj 2010-09-30 15:22:13.000000000 +0200
+++ gcc/var-tracking.c 2010-10-04 12:21:39.549377631 +0200
@@ -7133,6 +7133,19 @@ vt_expand_loc_dummy (rtx loc, htab_t var
#ifdef ENABLE_RTL_CHECKING
/* Used to verify that cur_loc_changed updating is safe. */
static struct pointer_map_t *emitted_notes;
+
+/* Strip REG_POINTER from REGs and MEM_POINTER from MEMs in order to
+ avoid differences in commutative operand simplification. */
+static rtx
+strip_pointer_flags (rtx x, const_rtx old_rtx ATTRIBUTE_UNUSED,
+ void *data ATTRIBUTE_UNUSED)
+{
+ if (REG_P (x) && REG_POINTER (x))
+ return gen_rtx_REG (GET_MODE (x), REGNO (x));
+ if (MEM_P (x) && MEM_POINTER (x))
+ return gen_rtx_MEM (GET_MODE (x), XEXP (x, 0));
+ return NULL_RTX;
+}
#endif
/* Emit the NOTE_INSN_VAR_LOCATION for variable *VARP. DATA contains
@@ -7332,8 +7345,21 @@ emit_note_insn_var_location (void **varp
if (!var->cur_loc_changed && (pnote || PAT_VAR_LOCATION_LOC (note_vl)))
{
gcc_assert (pnote);
- gcc_assert (rtx_equal_p (PAT_VAR_LOCATION_LOC (pnote),
- PAT_VAR_LOCATION_LOC (note_vl)));
+ if (!rtx_equal_p (PAT_VAR_LOCATION_LOC (pnote),
+ PAT_VAR_LOCATION_LOC (note_vl)))
+ {
+ /* There might be differences caused by REG_POINTER
+ differences. REG_POINTER affects
+ swap_commutative_operands_p. */
+ rtx old_vl = PAT_VAR_LOCATION_LOC (pnote);
+ rtx new_vl = PAT_VAR_LOCATION_LOC (note_vl);
+ old_vl = simplify_replace_fn_rtx (old_vl, NULL_RTX,
+ strip_pointer_flags, NULL);
+ new_vl = simplify_replace_fn_rtx (new_vl, NULL_RTX,
+ strip_pointer_flags, NULL);
+ gcc_assert (rtx_equal_p (old_vl, new_vl));
+ PAT_VAR_LOCATION_LOC (note_vl) = new_vl;
+ }
}
*note_slot = (void *) note_vl;
}
--- gcc/testsuite/gcc.dg/debug/pr45849.c.jj 2010-10-04 12:28:33.255364734 +0200
+++ gcc/testsuite/gcc.dg/debug/pr45849.c 2010-10-04 12:30:27.966658821 +0200
@@ -0,0 +1,31 @@
+/* PR debug/45849 */
+/* { dg-do compile } */
+/* { dg-options "-g -Wno-uninitialized" } */
+
+extern void bar (void);
+
+void
+foo (long repllen, char *rp)
+{
+ char *matchend;
+ char *scan;
+ long len;
+ char *matchstart;
+ char *text;
+ char *t;
+
+ repllen--;
+
+ for (;;)
+ {
+ matchstart = t + rp[0];
+ matchend = rp;
+ len = matchstart - text + repllen * (matchend - matchstart);
+ while (len)
+ ;
+ for (scan = text; scan != rp; scan++)
+ bar ();
+ if (matchstart)
+ text = matchend;
+ }
+}
Jakub