https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
--- Comment #8 from Martin Liška ---
Simplified a bit more:
$ cat /tmp/vect.c
unsigned int m;
#define N 128
unsigned int a[N];
unsigned int
__attribute__((noipa))
df_count_refs (bool include_defs)
{
int size = 0;
for (unsigned int regno =
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
--- Comment #7 from Martin Liška ---
Even easier, use the following self-contained test-case:
$ cat /tmp/vect.c
unsigned int m;
struct df_reg_info
{
unsigned int n_regs;
};
#define N 128
struct df_reg_info a[N];
unsigned int
__attribute__(
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
--- Comment #6 from Martin Liška ---
Easier way how to bisect that is to use only these dbg-cnt options:
-O3 -march=haswell -fdbg-cnt=vect_slp:0 -fdbg-cnt=vect_loop:3
which is first bad, vect_loop:2 is fine.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
--- Comment #5 from Martin Liška ---
> 3) build df-scan.o with -O3 -march=skylake
same happens for -march=haswell.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
--- Comment #4 from Martin Liška ---
Created attachment 47145
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47145&action=edit
Reproduce patch
Ok, steps to reproduce that:
1) install latest GCC trunk and use it as compiler
2) apply the pa
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
--- Comment #3 from Martin Liška ---
One can reproduce that by using df-scan.o from stage2 and libgcc ICEs with -g.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
Martin Liška changed:
What|Removed |Added
Status|ASSIGNED|NEW
Assignee|marxin at gcc dot
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
Martin Liška changed:
What|Removed |Added
Keywords|needs-bisection |
CC|
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
Martin Liška changed:
What|Removed |Added
Status|NEW |ASSIGNED
Assignee|unassigned at
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92301
Martin Liška changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Last reconfirmed|
10 matches
Mail list logo