NerdSHA256Plus blockheader format

[invpe]

What's the required blockheader format for NerdSHA256Plus ?

When given the block header in LE format (as it comes from the pool), with reversed merkle (post double sha256) to keep it LE

  std::string blockheader = "20000000dec6d0b18efc502ceb1c4df64793bbbcb291a6450002ac290000000000000000fbc46586cc5fc60ed3a55b2572abb4c4096f79b76878cc7c59f6d5ebb0d4e228671aa7731702f12800000000";
  to_byte_array(blockheader.c_str(), 80 * 2, u8BlockHeaderBytes);

BLOCKHEADER: 20000000dec6d0b18efc502ceb1c4df64793bbbcb291a6450002ac290000000000000000fbc46586cc5fc60ed3a55b2572abb4c4096f79b76878cc7c59f6d5ebb0d4e228671aa7731702f12800000000

Then passed through the hashing

nerdSHA256_context midstate0;   
  nerd_mids(&midstate0, u8BlockHeaderBytes);
  unsigned char *pHeader64 = u8BlockHeaderBytes + 64;
  nerd_sha256d(&midstate0, pHeader64, block_hash0);
  Serial.printf("NERDPLUS: ");
  for (size_t x = 0; x < sizeof(block_hash0); x++) {
    Serial.printf("%02x", block_hash0[x]);
  }
  Serial.println("");

The output returned by NerdSHA256Plus is

NERDPLUS: a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5a5|d34efd79

Where the NerdSHA256 (no plus) implementation returns correct results:

NERD: f44e3665ba9047ecf8259e0c598821f60df77aa1a5b680145f810644d34efd79
SHA256D: f44e3665ba9047ecf8259e0c598821f60df77aa1a5b680145f810644d34efd79
SHA256MID: f44e3665ba9047ecf8259e0c598821f60df77aa1a5b680145f810644d34efd79

Is there any expectation on the endianness for NerdSHA256Plus?
I noticed the NerdMiner reversing all blockheader fields coming from the pool but not merkle (post hashing its BE)

// reverse version
    uint8_t buff;
    size_t bword, bsize, boffset;
    boffset = 0;
    bsize = 4;
    for (size_t j = boffset; j < boffset + (bsize/2); j++) {
        buff = mMiner.bytearray_blockheader[j];
        mMiner.bytearray_blockheader[j] = mMiner.bytearray_blockheader[2 * boffset + bsize - 1 - j];
        mMiner.bytearray_blockheader[2 * boffset + bsize - 1 - j] = buff;
    }

    // reverse prev hash (4-byte word swap)
    boffset = 4;
    bword = 4;
    bsize = 32;
    for (size_t i = 1; i <= bsize / bword; i++) {
        for (size_t j = boffset; j < boffset + bword / 2; j++) {
            buff = mMiner.bytearray_blockheader[j];
            mMiner.bytearray_blockheader[j] = mMiner.bytearray_blockheader[2 * boffset + bword - 1 - j];
            mMiner.bytearray_blockheader[2 * boffset + bword - 1 - j] = buff;
        }
        boffset += bword;
    }

/*
    // reverse merkle (4-byte word swap)
    boffset = 36;
    bword = 4;
    bsize = 32;
    for (size_t i = 1; i <= bsize / bword; i++) {
        for (size_t j = boffset; j < boffset + bword / 2; j++) {
            buff = mMiner.bytearray_blockheader[j];
            mMiner.bytearray_blockheader[j] = mMiner.bytearray_blockheader[2 * boffset + bword - 1 - j];
            mMiner.bytearray_blockheader[2 * boffset + bword - 1 - j] = buff;
        }
        boffset += bword;
    }
*/
    // reverse ntime
    boffset = 68;
    bsize = 4;
    for (size_t j = boffset; j < boffset + (bsize/2); j++) {
        buff = mMiner.bytearray_blockheader[j];
        mMiner.bytearray_blockheader[j] = mMiner.bytearray_blockheader[2 * boffset + bsize - 1 - j];
        mMiner.bytearray_blockheader[2 * boffset + bsize - 1 - j] = buff;
    }

    // reverse difficulty
    boffset = 72;
    bsize = 4;
    for (size_t j = boffset; j < boffset + (bsize/2); j++) {
        buff = mMiner.bytearray_blockheader[j];
        mMiner.bytearray_blockheader[j] = mMiner.bytearray_blockheader[2 * boffset + bsize - 1 - j];
        mMiner.bytearray_blockheader[2 * boffset + bsize - 1 - j] = buff;
    }