[Bug tree-optimization/110280] internal compiler error: in const_unop, at fold-const.cc:1884
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110280 Andrew Pinski changed: What|Removed |Added See Also||https://gcc.gnu.org/bugzill ||a/show_bug.cgi?id=96463 --- Comment #9 from Andrew Pinski --- The creation of that VEC_PERM comes started at r13-1055-g494bec025002df422f2f (aka PR 96463). The VEC_PERM match-and-simplify was not updated for the new semantics it seems (it was added at r10-507-gebd733a78ccf ).
[Bug tree-optimization/110280] internal compiler error: in const_unop, at fold-const.cc:1884
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110280
--- Comment #8 from Andrew Pinski ---
In the original case FRE is trying to do the same thing as the reduced
testcase:
```
Successfully combined 4 partial definitions
Setting value number of _39 to { 0, 0, 0, 0 } (changed)
Value numbering stmt = _40 = VEC_PERM_EXPR <_39, _39, { 0, 1, 2, 3, ... }>;
Match-and-simplified VEC_PERM_EXPR <_39, _39, { 0, 1, 2, 3, ... }> to { 0, 0,
0, 0 }
RHS VEC_PERM_EXPR <_39, _39, { 0, 1, 2, 3, ... }> simplified to { 0, 0, 0, 0 }
Setting value number of _40 to { 0, 0, 0, 0 } (changed)
Value numbering stmt = lanes ={v} {CLOBBER(eol)};
Setting value number of .MEM_54 to .MEM_54 (changed)
Value numbering stmt = _41 = [vec_unpack_lo_expr] _40;
```
[Bug tree-optimization/110280] internal compiler error: in const_unop, at fold-const.cc:1884
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110280
--- Comment #7 from Andrew Pinski ---
I think this part of match is incorrect:
(if (sel.series_p (0, 1, 0, 1))
{ op0; }
(if (sel.series_p (0, 1, nelts, 1))
{ op1; }
This is only true if type == TREE_TYPE (op0) (well compatible) ...
[Bug tree-optimization/110280] internal compiler error: in const_unop, at fold-const.cc:1884
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110280
--- Comment #6 from Andrew Pinski ---
Match-and-simplified VEC_PERM_EXPR <_10, _10, { 0, 1, 2, 3, ... }> to { 0, 0,
0, 0 }
Ok, that is wrong ...
[Bug tree-optimization/110280] internal compiler error: in const_unop, at fold-const.cc:1884
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110280
Andrew Pinski changed:
What|Removed |Added
Target Milestone|--- |13.2
Known to fail||13.1.0
--- Comment #5 from Andrew Pinski ---
First reduced testcase with the same ICE:
```
#include "arm_sve.h"
auto l() {
svuint32_t ttt;
{
alignas(16) const unsigned int lanes[4] = {0, 0, 0, 0};
ttt = svld1rq_u32(svptrue_b8(), lanes);
}
return svunpklo_u64 (ttt);
}
```
Another reduced testcase with a related ICE:
```
#include "arm_sve.h"
svuint32_t l() {
alignas(16) const unsigned int lanes[4] = {0, 0, 0, 0};
return svld1rq_u32(svptrue_b8(), lanes);
}
```
For the first one, we have:
```
lanes[0] = 0;
lanes[1] = 0;
lanes[2] = 0;
lanes[3] = 0;
_10 = MEM [(unsigned int * {ref-all})&lanes];
ttt_7 = VEC_PERM_EXPR <_10, _10, { 0, 1, 2, 3, ... }>;
lanes ={v} {CLOBBER(eol)};
_9 = [vec_unpack_lo_expr] ttt_7;
```
The ICE is on simplifying _9.
The second one we have a similar thing except no vec_unpack_lo_expr there.
I suspect it is the simplification of the VEC_PERM which is causing the issue
rather than vec_unpack_lo_expr directly.
And dealing with how _10 is generated inside fre to a vector.
[Bug tree-optimization/110280] internal compiler error: in const_unop, at fold-const.cc:1884
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110280 --- Comment #4 from Andrew Pinski --- (gdb) p debug_gimple_stmt(stmt) _41 = [vec_unpack_lo_expr] _40;
[Bug tree-optimization/110280] internal compiler error: in const_unop, at fold-const.cc:1884
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110280 --- Comment #3 from Andrew Pinski --- #6 0x0087710b in const_unop (code=VEC_UNPACK_LO_EXPR, type=, arg0=) at ../../gcc/fold-const.cc:1954 1954gcc_assert (known_eq (out_nelts, TYPE_VECTOR_SUBPARTS (type)));
