https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121357
Bug ID: 121357
Summary: Bad codegen when generically 1-filling a struct in C
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: pskocik at gmail dot com
Target Milestone: ---
I tried having
#define filled$(Tp_t) (*(typeof(Tp_t)*)memset(&(Tp_t){0},255,sizeof(Tp_t)))
for generically filling (to all bits 1) a struct in C. I tested it on
struct it { unsigned long x[1]; };
void takeit(struct it X);
void wanted_codegen(void){ takeit((struct it){-1ul}); }
void wanted_ccode(void){ takeit(filled$(struct it)); }
struct it2 { unsigned long x[2]; };
void takeit2(struct it2 X);
void wanted_codegen2(void){ takeit2((struct it2){-1ul,-1ul}); }
void wanted_ccode2(void){ takeit2(filled$(struct it2)); }
On clang, the wanted codegen matches the one rendered for the wanted c-code.
On gcc, however, the code generated for the wanted c-code spills the struct to
the stack (which also blocks tailcalling).
https://godbolt.org/z/xvTPEToh3
Could this be fixed please?
