On Wed, 20 May 2026 07:17:07 GMT, Shawn Emery <[email protected]> wrote:
>> ooh.. this one. this really tripped me up, because the way this intrinsic
>> loops over user data is very non-intuitive (though its probably more
>> intrinsic-friendly). Best I can describe: "limit is the start of the last
>> block". From DigestBase.java
>>
>>
>> // compress complete blocks
>> if (len >= blockSize) {
>> int limit = ofs + len;
>> ofs = implCompressMultiBlock(b, ofs, limit - blockSize); //
>> <<<<< HERE!!!
>> len = limit - ofs;
>> }
>> // copy remainder to buffer
>> if (len > 0) {
>> System.arraycopy(b, ofs, buffer, 0, len);
>> bufOfs = len;
>> }
>> }
>> // compress complete blocks
>> private int implCompressMultiBlock(byte[] b, int ofs, int limit) {
>> implCompressMultiBlockCheck(b, ofs, limit);
>> return implCompressMultiBlock0(b, ofs, limit);
>> }
>>
>> @IntrinsicCandidate
>> private int implCompressMultiBlock0(byte[] b, int ofs, int limit) {
>> for (; ofs <= limit; ofs += blockSize) {
>> implCompress(b, ofs);
>> }
>> return ofs;
>> }
>>
>> private void implCompressMultiBlockCheck(byte[] b, int ofs, int limit) {
>> if (limit < 0) {
>> return; // not an error because implCompressMultiBlockImpl
>> won't execute if limit < 0
>> // and an exception is thrown if ofs < 0.
>> }
>>
>> Objects.requireNonNull(b);
>> Preconditions.checkIndex(ofs, b.length,
>> Preconditions.AIOOBE_FORMATTER);
>>
>> int endIndex = (limit / blockSize) * blockSize + blockSize - 1; //
>> <<<<< HERE
>> if (endIndex >= b.length) {
>> throw new ArrayIndexOutOfBoundsException(endIndex);
>> }
>> }
>
> But `SHA3::implCompressCheck` et al. only checks up to `ofs + blockSize - 1`
> which is why I believe the old code had an array of masks to round out the
> various blocks sizes in order to prevent accessing memory outside of the
> array in Java.
I did indeed have a problem here; with the mask! wow.. I looked at that number
a hundred times. `__ movl(rax, 0x1F);` is wrong (i.e. the value of `k1`). It
should be `0x1` (we are reading one 64bit value at a time for the non-parallel
keccaks). (I just kept on seeing an `L` instead of `F`.. thanks. that was close)
Reran my fuzz tests.. works.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/31125#discussion_r3277833165
