[BUG] GCC optimizes away the write with `benchmark::DoNotOptimize`

[SaiCharanMarrivada]

GCC16.1 optimizes away the write when we use benchmark::DoNotOptimize.

Minimal reproducible example

void f() {
    int *a = new int[1 << 16];
    for (int i = 0; i < 1 << 16; i++) {
        benchmark::DoNotOptimize(a[i] = 10); // this is optimized away by the compiler
    }
    delete[] a;
}

Compiler explorer(CE) link: https://godbolt.org/z/j538536ss. The CE link contains a possible fix for the issue.

[LebedevRI]

@SaiCharanMarrivada could you please file an issue with GCC?
They should just provide a builtin that would do the right thing,
or at worst expand to the correct ASM sequence.

[SaiCharanMarrivada]

@LebedevRI Sure I will raise it. What do you think about the fix given in the compiler explorer link?

[ramanayakakumara01-ship-it]

void f() {
int *a = new int[1 << 16];
for (int i = 0; i < 1 << 16; i++) {
benchmark::DoNotOptimize(a[i] = 10); // this is optimized away by the compiler
}
delete[] a;
}