AVF performance

[rybasd]

I've compiled and run very basic performance test.
Virtualization penalty under AVF for simple computation is about 1.2 times slower compared to run in Termux (i.e. android kernel).

System that I have is Samsung Gwlaxy Tab S11 (Dimensity 9400+)

Is it expected? Doesn't seem like "near native" performance.

What i get in VM (that's average in multiple runs so no throttling interference)
podroid:# /root/bench 3000000000
INT: 3.211 sec, 934.17 Mops/s (sum=8317209874376305593)
DBL: 9.582 sec, 156.55 Mops/s (sum=inf)
podroid:#
What I get in Termux:
~ $ ~/bench 3000000000
INT: 2.559 sec, 1172.41 Mops/s (sum=8317209874376305593)
DBL: 8.491 sec, 176.66 Mops/s (sum=inf)
~ $

Code that I run (in both podroid's alpine VM and Termux)
#include <stdio.h>
#include <time.h>
#include <stdlib.h>

static void bench_int(long n) {
long long a = 1, b = 2;
clock_t start = clock();
for (long i = 0; i < n; i++) {
a = a * 3 + b;
b = b * 3 + a;
a ^= (a >> 7);
b ^= (b >> 7);
}
clock_t end = clock();
double t = (double)(end - start) / CLOCKS_PER_SEC;
printf("INT: %8.3f sec, %7.2f Mops/s (sum=%lld)\n", t, n / t / 1e6, a + b);
}

static void bench_double(long n) {
double a = 1.0, b = 2.0;
clock_t start = clock();
for (long i = 0; i < n; i++) {
a = a * 1.0000001 + b;
b = b * 1.0000001 + a;
a = a / 1.0000001;
b = b / 1.0000001;
}
clock_t end = clock();
double t = (double)(end - start) / CLOCKS_PER_SEC;
printf("DBL: %8.3f sec, %7.2f Mops/s (sum=%f)\n", t, n / t / 1e6, a + b);
}

int main(int argc, char** argv) {
long n = 100000000;
if (argc > 1) n = atol(argv[1]);
bench_int(n);
bench_double(n / 2);
return 0;
}