https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111573
Bug ID: 111573
Summary: lambda functions often not inlined and optimized out
Product: gcc
Version: 14.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: hubicka at gcc dot gnu.org
Target Milestone: ---
#include <functional>
using namespace std;
static int dosum(std::function<int(int,int)> fn)
{
return fn(5,6);
}
int test()
{
auto sum = [](int a, int b) {
return a + b;
};
int s;
for (s = 0; s < 10; s++)
s+= dosum(sum);
return s;
}
Gets optimized well only with early inlining. compiled with -fno-early-inlining
yields to:
_Z4testv:
.LFB2166:
.cfi_startproc
subq $56, %rsp
.cfi_def_cfa_offset 64
xorl %ecx, %ecx
.p2align 4,,10
.p2align 3
.L8:
leaq 12(%rsp), %rdx
leaq 8(%rsp), %rsi
movl $5, 8(%rsp)
leaq 16(%rsp), %rdi
movl $6, 12(%rsp)
call
_ZNSt17_Function_handlerIFiiiEZ4testvEUliiE_E9_M_invokeERKSt9_Any_dataOiS6_
leal 1(%rcx,%rax), %ecx
cmpl $9, %ecx
jle .L8
movl %ecx, %eax
addq $56, %rsp
.cfi_def_cfa_offset 8
ret
So we fail to inline since ipa-prop fails to track the constant function
address. I think this is really common in typical lambda function usage
