2x different performance with clang

main.c

#include <time.h>
#include <stdio.h>

double apple(double *arr, int size) ;
double orange(double *arr, int size) ;

int main() {
    const int size = 1000;
    double arr[size];

    for (int i=0; i<size; ++i) {
        arr[i] = i;
    }
    arr[size / 2] = -0.123213;


    double ra = apple(arr, size);
    double rb = orange(arr, size);
    printf("ra = %f | rb = %f\n", ra, rb);

    // benchmark
    clock_t ts, te;
    double dur;

    const int repeat = 100000;

    ts = clock();
    for (int i=0; i<repeat; ++i) apple(arr, size);
    te = clock();
    dur = te - ts;
    printf("apple %f\n", dur/CLOCKS_PER_SEC);

    ts = clock();
    for (int i=0; i<repeat; ++i) orange(arr, size);
    te = clock();
    dur = te - ts;

    printf("orange %f\n", dur/CLOCKS_PER_SEC);


    return 0;

}

nanmin.c

#include <math.h>


double apple(double *arr, int size) {
    double amin = INFINITY;
    int all_missing = 1;
    int i;
    double ai;
    for (i=0; i<size; ++i) { // increment i here
        ai = arr[i];

        if ( ai <= amin ) {
            amin = ai;
            all_missing = 0;
        }
    }
    if (all_missing) {
        amin = NAN;
    }
    return amin;
}


double orange(double *arr, int size) {
    double amin = INFINITY;
    int all_missing = 1;
    int i;
    double ai;
    for (i=0; i<size;) {
        ai = arr[i];
        ++i; // increment i here
        if ( ai <= amin ) {
            amin = ai;
            all_missing = 0;
        }
    }
    if (all_missing) {
        amin = NAN;
    }
    return amin;
}

michael@nyx:~/w/apples# ./test-3.8 
ra = -0.123213 | rb = -0.123213
apple 0.242737
orange 0.053176
michael@nyx:~/w/apples# ./test-3.6
ra = -0.123213 | rb = -0.123213
apple 0.235650
orange 0.054918

x86_64 Linux
Looks like the speed impact is even more drastic here, and that it's really not a pure 3.9 issue.