[Bug rtl-optimization/46387] [4.6 Regression] ICE in add_stores, at var-tracking.c:5415
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46387 --- Comment #1 from Richard Guenther rguenth at gcc dot gnu.org 2010-11-09 11:01:31 UTC --- Created attachment 22341 -- http://gcc.gnu.org/bugzilla/attachment.cgi?id=22341 required stack-protector support patch Required patch. Reducing.
[Bug rtl-optimization/46387] [4.6 Regression] ICE in add_stores, at var-tracking.c:5415
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46387
Richard Guenther rguenth at gcc dot gnu.org changed:
What|Removed |Added
Target Milestone|--- |4.6.0
--- Comment #2 from Richard Guenther rguenth at gcc dot gnu.org 2010-11-09
11:35:35 UTC ---
Reduced testcase, no longer requires stack-protector:
typedef signed short gint16;
typedef int gint;
typedef double gdouble;
typedef struct _GdkDrawable GdkPixmap;
typedef struct _GdkDrawable GdkWindow;
typedef enum { GDK_BUTTON1_MASK = 1 8 } GdkModifierType;
typedef struct _GdkEventMotion GdkEventMotion;
struct _GdkEventMotion {
GdkWindow *window;
gdouble x;
gdouble y;
gint16 is_hint;
};
typedef struct _GtkWidget GtkWidget;
typedef struct {
unsigned char width;
} gn_bmp;
GdkPixmap *drawingPixmap = ((void *)0);
int mouseButtonPushed = 0;
gn_bmp bitmap, oldBitmap;
int activeTool = 0;
gint
DrawingAreaMotionNotifyEvent(GtkWidget * widget, GdkEventMotion * event)
{
int x, y;
GdkModifierType state;
if (!mouseButtonPushed activeTool != 0)
return (!(0));
if (event-is_hint)
gdk_window_get_pointer(x, y, state);
else
{
x = event-x;
y = event-y;
}
x = x / (5 + 1);
y = y / (5 + 1);
if (x 0)
x = 0;
if (x bitmap.width - 1)
x = bitmap.width - 1;
switch (activeTool)
{
case 0:
ToolBrush(widget, x, y, 1);
}
return (!(0));
}
./cc1 -quiet xgnokii_logos.3.i -O2 -g
xgnokii_logos.3.i: In function 'DrawingAreaMotionNotifyEvent':
xgnokii_logos.3.i:48:1: internal compiler error: in add_stores, at
var-tracking.c:5415
Please submit a full bug report,
with preprocessed source if appropriate.
See http://bugs.opensuse.org/ for instructions.
reproducible with a cross to ia64-linux.
[Bug rtl-optimization/46387] [4.6 Regression] ICE in add_stores, at var-tracking.c:5415
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46387
Jakub Jelinek jakub at gcc dot gnu.org changed:
What|Removed |Added
CC||jakub at gcc dot gnu.org
--- Comment #3 from Jakub Jelinek jakub at gcc dot gnu.org 2010-11-09
13:19:45 UTC ---
Reduced testcase:
struct S { double x; double y; short z; };
int a = 0, b = 0, c;
void bar (int, int, int);
void baz (int *, int *, int *);
void
foo (struct S *v)
{
int x, y, z;
if (!a b != 0)
return;
if (v-z)
baz (x, y, z);
else
{
x = v-x;
y = v-y;
}
x = x / (5 + 1);
y = y / (5 + 1);
if (x 0)
x = 0;
if (x c - 1)
x = c - 1;
if (b == 0)
bar (x, y, 1);
return;
}
Looking into it.
[Bug rtl-optimization/46387] [4.6 Regression] ICE in add_stores, at var-tracking.c:5415
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46387
--- Comment #4 from Jakub Jelinek jakub at gcc dot gnu.org 2010-11-09
15:20:47 UTC ---
The interesting insns here are:
(insn:TI 70 272 60 5 (set (reg:BI 262 p6 [394])
(ge:BI (reg:SI 16 r16 [orig:347 prephitmp.13 ] [347])
(const_int 0 [0]))) pr46387.c:21 318 {*cmpsi_adjusted}
(nil))
(insn 60 70 271 5 (set (mem/c/i:SI (reg/f:DI 32 r33 [409]) [2 x+0 S4 A64])
(reg:SI 16 r16 [380])) pr46387.c:19 4 {movsi_internal}
(nil))
...
(insn:TI 75 270 251 5 (cond_exec (eq (reg:BI 262 p6 [394])
(const_int 0 [0]))
(set (mem/c/i:SI (reg/f:DI 32 r33 [409]) [2 x+0 S4 A64])
(const_int 0 [0]))) pr46387.c:22 813 {*p movsi_internal}
(nil))
...
(insn 57 65 253 5 (set (reg:SI 14 r14 [orig:347 prephitmp.13 ] [347])
(reg:SI 16 r16 [380])) pr46387.c:19 4 {movsi_internal}
(expr_list:REG_DEAD (reg:SI 16 r16 [380])
(nil)))
...
(insn 4 140 254 5 (cond_exec (eq (reg:BI 262 p6 [394])
(const_int 0 [0]))
(set (reg:SI 14 r14 [orig:347 prephitmp.13 ] [347])
(const_int 0 [0]))) pr46387.c:22 813 {*p movsi_internal}
(nil))
We use the same VALUE for r14 in insn 4 as for (if_then_else (eq (p6) (0)) (0)
(r14)).
[Bug rtl-optimization/46387] [4.6 Regression] ICE in add_stores, at var-tracking.c:5415
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46387
--- Comment #5 from Jakub Jelinek jakub at gcc dot gnu.org 2010-11-09
17:48:04 UTC ---
The problem is when processing
(insn:TI 75 270 251 5 (cond_exec (eq (reg:BI 262 p6 [394])
(const_int 0 [0]))
(set (mem/c/i:SI (reg/f:DI 32 r33 [409]) [2 x+0 S4 A64])
(const_int 0 [0]))) pr46387.c:22 813 {*p movsi_internal}
(nil))
we don't invalidate the old mem content. cselib_invalidate_mem is called with
mem_rtx:
(mem/c/i:SI (reg/f:DI 32 r33 [409]) [2 x+0 S4 A64])
and we see x:
(mem/c/i:SI (value/u:DI 23:3920 @0x16199f0/0x1623ec0) [2 x+0 S4 A64])
But
canon_true_dependence (mem_rtx, GET_MODE (mem_rtx), mem_addr, x, NULL_RTX,
cselib_rtx_varies_p)
returns 0, which means it is not invalidated.
debug_rtx (((cselib_val *)0x1623ec0)-locs-loc) shows:
(reg/f:DI 32 r33 [409])
debug_rtx (((cselib_val *)0x1623ec0)-locs-next-loc) shows:
(plus:DI (value/u:DI 1:1 @0x16197e0/0x166c490)
(const_int -8 [0xfff8]))
and that is the whole list. VALUE 1:1 has only one living location:
(reg/f:DI 0 ap)
find_base_term of r33 gives r12 as base though (as init_alias_analysis has been
run before vt_initialize and doesn't know about the munging vt_initialize
performs to replace sp or hfp (+ offset) with cfa_base_rtx (which is either
argp or fp).
I guess we want to (temporarily, for the rest of var-tracking) set
REG_BASE_VALUE (cfa_base_rtx) to REG_BASE_VALUE (frame_pointer_needed ?
hard_frame_pointer_rtx : stack_pointer_rtx).
