https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128606
Bug ID: 128606
Summary: ICE in c_parser_gimple_postfix_expression caused by
_Literal of scalable vector type
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: Chris.Bazley at arm dot com
Target Milestone: ---
GCC encounters an internal compiler error when parsing a nonuniform scalable
vector constant in a `__GIMPLE` function.
The failure occurs in `c_parser_gimple_postfix_expression`, where
`TYPE_VECTOR_SUBPARTS (type).to_constant()` is used to determine the initial
allocation size of a vector constructor.
For scalable vector types, `TYPE_VECTOR_SUBPARTS` is polynomial, so
`to_constant()` fails.
**Reproducer:**
```c
#include <arm_sve.h>
svuint8_t __GIMPLE (startwith ("expand"))
foo (void)
{
svuint8_t v;
v = _Literal (svuint8_t) {
1, 2, 1, 2, 1, 2, 1, 0,
0, 0, 0, 0, 0, 0, 0, 0
};
return v;
}
```
**Compilation options:**
```
-O2 -fgimple -march=armv8.2-a+sve -msve-vector-bits=scalable
```
**Observed result:**
```
internal compiler error: in to_constant, at poly-int.h:592
```
**Expected result:**
The compiler should accept the scalable vector constant or issue a diagnostic
if the constructor is unsupported, rather than terminate with an internal
compiler error.
**Possible fix:**
Replacing `TYPE_VECTOR_SUBPARTS (type).to_constant()` with
`constant_lower_bound (TYPE_VECTOR_SUBPARTS (type)` avoids the ICE and allows
the reproducer to compile successfully and produce the following expected
output:
```
svuint8_t foo ()
{
svuint8_t v;
v = { 1, 2, 1, 2, 1, 2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, ... };
return v;
}
```
Further testing is needed to establish whether this is sufficient for all
scalable vector constructors.
