smh_latency.cpp

template <typename T>
void match_multiple_smh_latency_test(T & smh, std::vector<u8 *> & buffers, std::vector<size_t> & lengths,
                       std::vector<u32> & results) {
        u32 i = 0;
        u32 tmp = 0;
#ifndef NO_UNROLL
        // NOTE: experimental code only. Note that the addition of 'tmp' - being the id of a possible
        // match - could take us RIGHT outside our buffer if we actually matched something. We aren't
        // in this particular run, but so it goes. Saner would be to build up an all-zero id vector
        for (; i+7 < buffers.size(); i+=8) {
            tmp = results[i+0] = smh.match(buffers[i+0 + tmp], lengths[i+0] + tmp); LFENCE
            tmp = results[i+1] = smh.match(buffers[i+1 + tmp], lengths[i+1] + tmp); LFENCE
            tmp = results[i+2] = smh.match(buffers[i+2 + tmp], lengths[i+2] + tmp); LFENCE
            tmp = results[i+3] = smh.match(buffers[i+3 + tmp], lengths[i+3] + tmp); LFENCE
            tmp = results[i+4] = smh.match(buffers[i+4 + tmp], lengths[i+4] + tmp); LFENCE
            tmp = results[i+5] = smh.match(buffers[i+5 + tmp], lengths[i+5] + tmp); LFENCE
            tmp = results[i+6] = smh.match(buffers[i+6 + tmp], lengths[i+6] + tmp); LFENCE
            tmp = results[i+7] = smh.match(buffers[i+7 + tmp], lengths[i+7] + tmp); LFENCE
        }
#endif
        for (; i < buffers.size(); ++i) {
            tmp = results[i] = smh.match(buffers[i + tmp], lengths[i + tmp]); LFENCE
        }
}