See how fast we can make direct-threaded code in C (using computed goto)

repeat-while.diff

     static void* prog[] = {
-        // loop:
         &&i_pushvar0,                       // pushvar i
         &&i_pushnum, (void*)100000000,      // pushnum 100000000
         &&i_jge, (void*)5,                  // jge end
+        // loop:
         &&i_pushvar0,                       // push i
         &&i_addvar1,                        // addvar s
         &&i_incvar0,                        // incvar i
-        &&i_jmp, (void*)((long long)-10),   // jmp loop
+        &&i_pushvar0,                       // pushvar i
+        &&i_pushnum, (void*)100000000,      // pushnum 100000000
+        &&i_jl, (void*)((long long)-8),     // jl loop
         // end:

vm-direct.c

/*
$ time awk 'BEGIN { for (; i<100000000; i++) s+=i; print s }'
4999999950000000

real    0m6.141s
user    0m6.140s
sys 0m0.000s

gcc -O2 vm-direct.c && time ./a.out

original: 0.635s
superinstruction "jge" instead of "less jz": 0.596s
pushvar0, pushvar1, addvar1, incvar0: 0.546s
    (moving literals to nums array: same as above -- reverted!)
    (tos variable: 0.756 -- reverted!)
repeat while condition: 0.455s

*/

#include <stdio.h>
#include <stdlib.h>

// #define NEXT printf("ip=%ld i=%lld s=%lld stack=[%lld %lld %lld %lld] sp=%ld\n", ip-prog, vars[0], vars[1], stack[0], stack[1], stack[2], stack[3], sp-stack); goto **ip++
#define NEXT goto **ip++
#define FETCH() (long long)(unsigned long long)(*ip++)
#define POP() (*sp++)
#define PUSH(x) *--sp = x

int main() {
    static void* prog[] = {
        &&i_pushvar0,                       // pushvar i
        &&i_pushnum, (void*)100000000,      // pushnum 100000000
        &&i_jge, (void*)5,                  // jge end
        // loop:
        &&i_pushvar0,                       // push i
        &&i_addvar1,                        // addvar s
        &&i_incvar0,                        // incvar i
        &&i_pushvar0,                       // pushvar i
        &&i_pushnum, (void*)100000000,      // pushnum 100000000
        &&i_jl, (void*)((long long)-8),     // jl loop
        // end:
        &&i_pushvar1,                       // pushvar s
        &&i_printnum,                       // printnum
        &&i_exit,                           // exit
    };
    void** ip = prog;
    long long vars[2] = {0, 0};
    long long stack[4] = {0, 0, 0, 0};
    long long* sp = &stack[4];

    NEXT;

i_pushvar0: {
    PUSH(vars[0]);
    NEXT;
}

i_pushvar1: {
    PUSH(vars[1]);
    NEXT;
}

i_pushnum: {
    long long a = FETCH();
    PUSH(a);
    NEXT;
}

i_jge: { // ( a b -- )
    long long b = POP();
    long long a = POP();
    long long offset = FETCH();
    if (a >= b) {
        ip += offset;
    }
    NEXT;
}
i_jl: { // ( a b -- )
    long long b = POP();
    long long a = POP();
    long long offset = FETCH();
    if (a < b) {
        ip += offset;
    }
    NEXT;
}

i_jmp: {
    long long offset = FETCH();
    ip += offset;
    NEXT;
}

i_addvar1: {
    vars[1] += POP();
    NEXT;
}

i_incvar0: {
    vars[0]++;
    NEXT;
}

i_printnum: {
    long long x = *sp++;
    printf("%lld\n", x);
    NEXT;
}

i_exit: {
    exit(0);
}

}