Major performance regression in Math.log on aarch64

ADDITIONAL SYSTEM INFORMATION :
aarch64, Apple M1 Max, macOS 13.2.1

A DESCRIPTION OF THE PROBLEM :
Math.log using the generic dlog intrinsic is much slower than StrictMath.log on aarch64.

Caused by JDK-8215133
Related to JDK-8210858

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
javac Main.java
java Main
java -XX:+UnlockDiagnosticVMOptions -XX:DisableIntrinsic=_dlog Main

EXPECTED VERSUS ACTUAL BEHAVIOR :
EXPECTED -
The elapsed time without additional options is less than or equal to the time with the _dlog intrinsic disabled.
ACTUAL -
java Main
6200ms

java -XX:+UnlockDiagnosticVMOptions -XX:DisableIntrinsic=_dlog Main
860ms

---------- BEGIN SOURCE ----------
import java.util.Random;
import java.util.concurrent.TimeUnit;

public class Main {
    public static void main(String[] args) throws Exception {
        while (true) {
            final Random random = new Random();
            final double[] values = new double[100_000_000];
            for (int i = 0; i < values.length; i++)
                values[i] = random.nextDouble();

            System.gc();

            final long start = System.nanoTime();

            double blackhole = 0;
            for (int i = 0; i < values.length; i++)
                blackhole += Math.log(values[i]);

            final long elapsed = TimeUnit.NANOSECONDS.toMillis(System.nanoTime() - start);

            System.out.println(elapsed + "ms (" + blackhole + ")");
        }
    }
}
---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
Disable the _dlog intrinsic on aarch64 like -XX:DisableIntrinsic=_dlog do and use the StrictMath implementation.

FREQUENCY : always