rose: use program offset, not final_id, in atable

This removes the need to look up the program offset in a table when
handling an anchored literal match.

diff --git a/src/rose/match.c b/src/rose/match.c
index 95cb141eafa50fced8f537bb549e166559e311cc..b641e39d842a6d2d2678ef204374ccd53eb3ebd4 100644 (file)
--- a/src/rose/match.c
+++ b/src/rose/match.c
@@ -220,10 +220,9 @@ int roseAnchoredCallback(u64a start, u64a end, u32 id, void *ctx) {
         tctxt->lastEndOffset = real_end;
     }
 
-    const u32 *programs = getByOffset(t, t->litProgramOffset);
-    assert(id < t->literalCount);
+    // Note that the "id" we have been handed is the program offset.
     const u8 flags = ROSE_PROG_FLAG_IN_ANCHORED;
-    if (roseRunProgram(t, scratch, programs[id], start, real_end, match_len,
+    if (roseRunProgram(t, scratch, id, start, real_end, match_len,
                        flags) == HWLM_TERMINATE_MATCHING) {
         assert(can_stop_matching(scratch));
         DEBUG_PRINTF("caller requested termination\n");

diff --git a/src/rose/rose_build_anchored.cpp b/src/rose/rose_build_anchored.cpp
index 286cc7ae7968efb24d3ae7a0ae180e071ed23e3e..befd0bad6e0133da49af83251ad5c2acc9b0a9ea 100644 (file)
--- a/src/rose/rose_build_anchored.cpp
+++ b/src/rose/rose_build_anchored.cpp
@@ -204,6 +204,28 @@ void remapAnchoredReports(RoseBuildImpl &tbi) {
     }
 }
 
+static
+void remapIds(flat_set<ReportID> &reports, const vector<u32> &litPrograms) {
+    flat_set<ReportID> new_reports;
+    for (auto id : reports) {
+        assert(id < litPrograms.size());
+        new_reports.insert(litPrograms.at(id));
+    }
+    reports = move(new_reports);
+}
+
+/**
+ * \brief Replace the reports (which are literal final_ids) in the given
+ * raw_dfa with program offsets.
+ */
+static
+void remapIdsToPrograms(raw_dfa &rdfa, const vector<u32> &litPrograms) {
+    for (dstate &ds : rdfa.states) {
+        remapIds(ds.reports, litPrograms);
+        remapIds(ds.reports_eod, litPrograms);
+    }
+}
+
 static
 void populate_holder(const simple_anchored_info &sai, const set<u32> &exit_ids,
                      NGHolder *h_in) {
@@ -826,7 +848,7 @@ vector<raw_dfa> buildAnchoredDfas(RoseBuildImpl &build) {
 
 aligned_unique_ptr<anchored_matcher_info>
 buildAnchoredMatcher(RoseBuildImpl &build, vector<raw_dfa> &dfas,
-                     size_t *asize) {
+                     const vector<u32> &litPrograms, size_t *asize) {
     const CompileContext &cc = build.cc;
 
     if (dfas.empty()) {
@@ -835,6 +857,10 @@ buildAnchoredMatcher(RoseBuildImpl &build, vector<raw_dfa> &dfas,
         return nullptr;
     }
 
+    for (auto &rdfa : dfas) {
+        remapIdsToPrograms(rdfa, litPrograms);
+    }
+
     vector<aligned_unique_ptr<NFA>> nfas;
     vector<u32> start_offset; // start offset for each dfa (dots removed)
     size_t total_size = buildNfas(dfas, &nfas, &start_offset, cc, build.rm);

diff --git a/src/rose/rose_build_anchored.h b/src/rose/rose_build_anchored.h
index a5317f8949ca12d8802532fafe6575c943502496..579b26d720657639318d331be16f07ef839982c7 100644 (file)
--- a/src/rose/rose_build_anchored.h
+++ b/src/rose/rose_build_anchored.h
@@ -56,10 +56,13 @@ std::vector<raw_dfa> buildAnchoredDfas(RoseBuildImpl &build);
 /**
  * \brief Construct an anchored_matcher_info runtime structure from the given
  * set of DFAs.
+ *
+ * Remap the literal final_ids used for raw_dfa reports to the program offsets
+ * given in litPrograms.
  */
 aligned_unique_ptr<anchored_matcher_info>
 buildAnchoredMatcher(RoseBuildImpl &build, std::vector<raw_dfa> &dfas,
-                     size_t *asize);
+                     const std::vector<u32> &litPrograms, size_t *asize);
 
 u32 anchoredStateSize(const anchored_matcher_info &atable);
 

diff --git a/src/rose/rose_build_bytecode.cpp b/src/rose/rose_build_bytecode.cpp
index 5cd8161b384250749815de90ecd8c2b2d798dabc..f451b8ea341d7e06edd2bebe202e1a840d737854 100644 (file)
--- a/src/rose/rose_build_bytecode.cpp
+++ b/src/rose/rose_build_bytecode.cpp
@@ -417,6 +417,10 @@ struct build_context : boost::noncopyable {
      * that have already been pushed into the engine_blob. */
     ue2::unordered_map<u32, u32> engineOffsets;
 
+    /** \brief Literal programs, indexed by final_id, after they have been
+     * written to the engine_blob. */
+    vector<u32> litPrograms;
+
     /** \brief Minimum offset of a match from the floating table. */
     u32 floatingMinLiteralMatchOffset = 0;
 
@@ -4736,20 +4740,20 @@ pair<u32, u32> buildLiteralPrograms(RoseBuildImpl &build, build_context &bc) {
     const u32 num_literals = build.final_id_to_literal.size();
     auto lit_edge_map = findEdgesByLiteral(build);
 
-    vector<u32> litPrograms(num_literals);
+    bc.litPrograms.resize(num_literals);
     vector<u32> delayRebuildPrograms(num_literals);
 
     for (u32 finalId = 0; finalId != num_literals; ++finalId) {
         const auto &lit_edges = lit_edge_map[finalId];
 
-        litPrograms[finalId] =
+        bc.litPrograms[finalId] =
             writeLiteralProgram(build, bc, finalId, lit_edges);
         delayRebuildPrograms[finalId] =
             buildDelayRebuildProgram(build, bc, finalId);
     }
 
     u32 litProgramsOffset =
-        add_to_engine_blob(bc, begin(litPrograms), end(litPrograms));
+        add_to_engine_blob(bc, begin(bc.litPrograms), end(bc.litPrograms));
     u32 delayRebuildProgramsOffset = add_to_engine_blob(
         bc, begin(delayRebuildPrograms), end(delayRebuildPrograms));
 
@@ -5206,7 +5210,8 @@ aligned_unique_ptr<RoseEngine> RoseBuildImpl::buildFinalEngine(u32 minWidth) {
     // Build anchored matcher.
     size_t asize = 0;
     u32 amatcherOffset = 0;
-    auto atable = buildAnchoredMatcher(*this, anchored_dfas, &asize);
+    auto atable = buildAnchoredMatcher(*this, anchored_dfas, bc.litPrograms,
+                                       &asize);
     if (atable) {
         currOffset = ROUNDUP_CL(currOffset);
         amatcherOffset = currOffset;