Github user kpu commented on a diff in the pull request: https://github.com/apache/incubator-joshua/pull/65#discussion_r79491699 --- Diff: src/main/java/org/apache/joshua/decoder/ff/lm/StateMinimizingLanguageModel.java --- @@ -132,30 +138,32 @@ public DPState compute(Rule rule, List<HGNode> tailNodes, int i, int j, SourcePa * Maps given array of word/class ids to KenLM ids. For estimating cost and computing, * state retrieval differs slightly. */ - private long[] mapToKenLmIds(int[] ids, List<HGNode> tailNodes, boolean isOnlyEstimate) { + private void writeKenLmIds(int[] ids, List<HGNode> tailNodes, boolean isOnlyEstimate, + KenLMPool poolWrapper) { + + poolWrapper.getNgramBuffer().putLong(0, ids.length); + // The IDs we will to KenLM - long[] kenIds = new long[ids.length]; for (int x = 0; x < ids.length; x++) { int id = ids[x]; if (isNonterminal(id)) { if (isOnlyEstimate) { // For the estimate, we can just mark negative values - kenIds[x] = -1; + poolWrapper.getNgramBuffer().putLong((x + 1) * LONG_SIZE_IN_BYTES, -1); } else { // Nonterminal: retrieve the KenLM long that records the state int index = -(id + 1); final KenLMState state = (KenLMState) tailNodes.get(index).getDPState(stateIndex); - kenIds[x] = -state.getState(); + poolWrapper.getNgramBuffer().putLong((x + 1) * LONG_SIZE_IN_BYTES, -state.getState()); --- End diff -- Looks like this is crying out for an abstraction.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---