[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263 Martin Liška changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution|--- |FIXED --- Comment #10 from Martin Liška --- Should be fixed now.
[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263 --- Comment #9 from Martin Liška --- Author: marxin Date: Fri Jun 29 10:57:00 2018 New Revision: 262247 URL: https://gcc.gnu.org/viewcvs?rev=262247=gcc=rev Log: Fix bit-test expansion for single cluster (PR tree-optimization/86263). 2018-06-29 Martin Liska PR tree-optimization/86263 * tree-switch-conversion.c (switch_decision_tree::try_switch_expansion): Make edge redirection. 2018-06-29 Martin Liska PR tree-optimization/86263 * gcc.dg/tree-ssa/pr86263.c: New test. Added: trunk/gcc/testsuite/gcc.dg/tree-ssa/pr86263.c Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-switch-conversion.c
[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263 Martin Liška changed: What|Removed |Added Status|REOPENED|ASSIGNED --- Comment #8 from Martin Liška --- Confirmed, I've got patch candidate for it.
[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263
--- Comment #7 from Tom de Vries ---
The pass_lower_switch has no effect. It generates a switch expansion at
bb12, but the function still uses the switch at bb10:
...
(gdb) call debug_function (cfun.decl, 0)
_dcvt (void * ptr, char type, int opt, int val)
{
char D.1454;
int _1;
char _12;
unsigned int _13;
unsigned int _14;
_Bool _15;
long unsigned int _16;
long unsigned int _17;
_Bool _18;
long unsigned int _19;
_Bool _20;
long unsigned int _21;
_Bool _22;
:
_1 = (int) type_3(D);
:
goto ; [100.00%]
:
_13 = (unsigned int) _1;
_14 = _13 + 4294967227;
_15 = _14 > 34;
if (_15 != 0)
goto ; [INV]
else
goto ; [INV]
:
_16 = 1 << _14;
_17 = _16 & 4294967297;
_18 = _17 != 0;
if (_18 != 0)
goto ; [INV]
else
goto ; [INV]
:
_19 = _16 & 17179869188;
_20 = _19 != 0;
if (_20 != 0)
goto ; [INV]
else
goto ; [INV]
:
_21 = _16 & 8589934594;
_22 = _21 != 0;
if (_22 != 0)
goto ; [INV]
else
goto ; [INV]
:
goto ; [INV]
:
switch (_1) [INV], case 69: [INV], case 70: [INV],
case 71: [INV], case 101: [INV], case 102: [INV], case 103:
[INV]>
:
:
f1 (ptr_5(D));
goto ; [INV]
:
:
if (opt_7(D) == 0)
goto ; [INV]
else
goto ; [INV]
:
opt_8 = 1;
:
f2 (ptr_5(D));
goto ; [INV]
:
:
f3 (ptr_5(D));
:
:
_12 = (char) val_11(D);
:
:
return _12;
}
...
[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263 cesar at gcc dot gnu.org changed: What|Removed |Added CC||cesar at gcc dot gnu.org --- Comment #6 from cesar at gcc dot gnu.org --- Created attachment 44311 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44311=edit newlib test case I attached a reduced test case from ecvtbuf.c. At Thomas mentioned, it's failing with nvptx-gcc.
[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263 --- Comment #5 from Thomas Schwinge --- Looking at newlib's "newlib/libc/stdlib/ecvtbuf.c", the diff of "ecvtbuf.c.040t.switchconv" between before/after the r261793 "Switch other switch expansion methods into classes" changes is just: beginning to process the following SWITCH statement ([...]/newlib/libc/stdlib/ecvtbuf.c:431) : --- switch (type_5(D)) [INV], case 69: [INV], case 70: [INV], case 71: [INV], case 101: [INV], case 102: [INV], case 103: [INV]> -Bailing out - no common successor to all case label target blocks found +Bailing out - expanding as bit test is preferable _dcvt (struct _reent * ptr, char * buffer, double invalue, int precision, int width, char type, int dot) This doesn't get changed back by your later r261886 -- is that expected/correct?
[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263 Thomas Schwinge changed: What|Removed |Added Status|RESOLVED|REOPENED Last reconfirmed|2018-06-21 00:00:00 |2018-6-22 Resolution|FIXED |--- --- Comment #4 from Thomas Schwinge --- Thanks for having a look, but I'm still seeing the same "internal compiler error: in emit_case_dispatch_table, at stmt.c:851". :-| Hmm. How did you test that it's now fixed?
[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263 Martin Liška changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution|--- |FIXED --- Comment #3 from Martin Liška --- Fixed.
[Bug tree-optimization/86263] [9 Regression] [nvptx] casesi, tablejump
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86263 --- Comment #2 from Martin Liška --- Author: marxin Date: Fri Jun 22 11:32:20 2018 New Revision: 261886 URL: https://gcc.gnu.org/viewcvs?rev=261886=gcc=rev Log: Fix emission of jump tables (PR tree-optimization/86263). 2018-06-22 Martin Liska PR tree-optimization/86263 * tree-switch-conversion.c (jump_table_cluster::find_jump_tables): Bail out if is_enabled is false. * tree-switch-conversion.h (jump_table_cluster::is_enabled): New declaration. (jump_table_cluster::is_enabled): New function. Modified: trunk/gcc/ChangeLog trunk/gcc/tree-switch-conversion.c trunk/gcc/tree-switch-conversion.h
