luhn with simd

SIMDLuhn.java

package ghadi;

import jdk.incubator.vector.*;
import java.nio.charset.StandardCharsets;

public class SIMDLuhn {
  
    // assume that credit cards are 16 digits
    static final VectorSpecies<Byte> SPECIES = ByteVector.SPECIES_128;

    /*
     * calculates luhn check digit from 16 digit string
     * returns 0 when valid,
     * set last digit to 0 if calculating checksum
     */
    static int luhn(String s) {
        var bs = s.getBytes(StandardCharsets.US_ASCII);
        var b = ByteVector.fromArray(SPECIES, bs, 0);

        // decode ascii by subtracting 0x30, which is '0'
        b = b.sub((byte) 0x30);

        // double every other element by adding vector to a masked version of itself
        // 0x5555 = 0101010101010101
        b = b.add(b, VectorMask.fromLong(SPECIES, 0x5555));

        // subtract 9 from any elements that are > 9
        // has effect of adding carry digit 12 => 3, 16 => 7
        b = b.sub((byte) 9, b.compare(VectorOperators.GT, (byte) 9));

        // horizontal addition across all lanes
        // add the second half of the vector to the first half,
        // then the second quarter of the first half to the first quarter, etc
        // Intel intrinsics include _mm_sad_epu8 to do some of this
        // but nothing exactly like fold available across platforms
        b = b.add(b.rearrange(VectorShuffle.fromValues(SPECIES, 8, 9, 10, 11, 12, 13, 14, 15, 0, 0, 0, 0, 0, 0, 0, 0)));
        b = b.add(b.rearrange(VectorShuffle.fromValues(SPECIES, 4, 5, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)));
        b = b.add(b.rearrange(VectorShuffle.fromValues(SPECIES, 2, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)));
        b = b.add(b.rearrange(VectorShuffle.fromValues(SPECIES, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0)));
        // sum is in lane 0
        int digit = b.lane(0) % 10;
        return digit;
    }
}