RFR: 8336274: MutableBigInteger.leftShift(int) optimization [v9]

[Raffaello Giulietti]

On Thu, 12 Sep 2024 12:27:01 GMT, fabioromano1 <duke at openjdk.org> wrote:

>> I mean something like
>> 
>>     private void primitiveRightShift(int n, int[] result, int resPos) {
>>         int[] val = value;
>>         int n2 = 32 - n;
>>         int c = 0;
>>         for (int i = 0; i < intLen; i++) {
>>             int b = val[offset + i];
>>             result[resPos + i] = c << n2 | b >>> n;
>>             c = b;
>>         }
>>     }
>> 
>> and
>> 
>>     private void primitiveLeftShift(int n, int[] result, int resPos) {
>>         int[] val = value;
>>         int n2 = 32 - n;
>>         int c = 0;
>>         for (int i = intLen - 1; i >= 0; i--) {
>>             int b = val[offset + i];
>>             result[resPos + i] = c >>> n2 | b << n;
>>             c = b;
>>         }
>> 
>> They are compatible with the precondition, right?
>
> But in the second version, the precondition must be `resPos <= offset - intLen || resPos >= offset` to make sure that the result is correct whether `result == value`, and the condition `resPos <= offset - intLen` is stronger than `resPos <= offset`.

Yes, I agree.

This one should correctly account for the precondition, and IMO is slightly simpler to read.

    private void primitiveLeftShift(int n, int[] result, int resPos) {
        int[] val = value;
        int n2 = 32 - n;
        final int m = intLen - 1;
        int b = val[offset];
        for (int i = 0; i < m; i++) {
            int c = val[offset + i + 1];
            result[resPos + i] = (b << n) | (c >>> n2);
            b = c;
        }
        result[resPos + m] = b << n;
    }

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

PR Review Comment: https://git.openjdk.org/jdk/pull/20008#discussion_r1756855722