RFR: 8267531: [x86] Assembler::andb(Address, Register) encoding is incorrect

[Aleksey Shipilev]

On Fri, 21 May 2021 15:36:51 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

> > Is it possible to add a jtreg test for this fix?
> 
> I guess it is possible to minimize the fuzzer test and/or create a fully synthetic regression test. But given how it looks like a copy-paste omission in the recent patch, I would prefer to invest time somewhere else.

I tried a few things, without success. Something like:


/**
 * @test
 * @bug 8267531
 * @summary [x86] Assembler::andb(Address,Register) encoding is incorrect
 *
 * @run main/othervm -XX:-TieredCompilation -XX:+PrintAssembly -XX:-UseNewCode
 *      -XX:CompileCommand=compileonly,compiler.c2.Test8267531::test
 *      compiler.c2.Test8267531
 * @run main/othervm -XX:-TieredCompilation -XX:+PrintAssembly -XX:+UseNewCode
 *      -XX:CompileCommand=compileonly,compiler.c2.Test8267531::test
 *      compiler.c2.Test8267531
 */

package compiler.c2;

import java.util.Arrays;

public class Test8267531 {
    static byte[] array = new byte[] { -1, -1, -1, -1, -1, -1, -1 };
    static int idx = 3;
    static int mask;

    public static void test() {
        //   match(Set dst (StoreB dst (AndI (LoadB dst) src)));
        array[idx] &= mask;
    }

    public static void main(String[] args) {
        for (int i = 0; i < 1000000; i++) {
            test();
            if (array[4] != -1 || array[3] != 0 || array[2] != -1) {
                throw new IllegalStateException("Error: " + Arrays.toString(array));
            }
        }
    }
}


It would seem we need to get lucky that both dst-addr-idx and src regs are fitting the branch path in `Assembler::prefix`. Let's mark the bug `noreg-hard` and move on?

-------------

PR: https://git.openjdk.java.net/jdk/pull/4145