AArch64: Compressed pointer decoding optimization when base is zero.

[XiaohongGong]

Currently decoding a compressed pointer with zero base will generate
the following code:
add x0, xzr, x0, lsl #3

This patch will optimize it to:
lsl x0, x0, #3

Here is a jmh benchmark to decode a compressed pointer:

public class UncompressPointer {
    private static final int field0 = 1000000;
    private int field1;

    static class A {
        B b;

        A(B b) {
            this.b = b;
        }
    }

    static class B {
        int field;

        B(int i) {
            this.field = i;
        }
    }

    static A[] arr;

    static {
        arr = new A[field0];
        for (int i = 0; i < field0; i++)
           arr[i] = new A(new B(i));
    }

    public static int func() {
        int result = 0;
        for (int i = 0; i < field0; i++) {
            result += arr[i].b.field;
        }
        return result;
    }

    @Benchmark
    public void jmhUncompressPointer() {
        field1 = func();
    }
}

And the performance results:

                         Score      Error    Units
Without this patch     21598.534 ? 2188.242  us/op
With this patch        19322.791 ? 1219.022  us/op

Change-Id: I5af6799eb82470d5598f5991e9210f99874f458a

[XiaohongGong]

Hi there,
Can anyone please take a look at this patch? Thank you!

[sanzinger]

Thank you for your contribution. I'll take a look as soon as possible. (And the other AArch64 related patches too)

[XiaohongGong]

Thank you for your contribution. I'll take a look as soon as possible. (And the other AArch64 related patches too)

Thanks so much!

[sanzinger]

I like this kind of test. Nice work.

[XiaohongGong]

Thank you!