rose: parameterise CHECK_LIT_EARLY

diff --git a/src/rose/program_runtime.h b/src/rose/program_runtime.h
index 783970709edd5632272981dde995a55768d259e1..a913ae27c390e7b238884d58c0d92613b71aa952 100644 (file)
--- a/src/rose/program_runtime.h
+++ b/src/rose/program_runtime.h
@@ -880,9 +880,9 @@ hwlmcb_rv_t roseRunProgram(const struct RoseEngine *t,
             PROGRAM_NEXT_INSTRUCTION
 
             PROGRAM_CASE(CHECK_LIT_EARLY) {
-                if (end < t->floatingMinLiteralMatchOffset) {
-                    DEBUG_PRINTF("halt: too soon, min offset=%u\n",
-                                 t->floatingMinLiteralMatchOffset);
+                if (end < ri->min_offset) {
+                    DEBUG_PRINTF("halt: before min_offset=%u\n",
+                                 ri->min_offset);
                     return HWLM_CONTINUE_MATCHING;
                 }
             }

diff --git a/src/rose/rose_build_bytecode.cpp b/src/rose/rose_build_bytecode.cpp
index 3d1b5c6b2f0ccb22cf107cea783fcc6174d15518..b148155b1cec48dee5579c5be2d6cb829e4f12f4 100644 (file)
--- a/src/rose/rose_build_bytecode.cpp
+++ b/src/rose/rose_build_bytecode.cpp
@@ -3568,6 +3568,19 @@ void makeGroupSquashInstruction(const RoseBuildImpl &build, u32 final_id,
     program.push_back(move(ri));
 }
 
+static
+u32 findMinOffset(const RoseBuildImpl &build, u32 lit_id) {
+    const auto &lit_vertices = build.literal_info.at(lit_id).vertices;
+    assert(!lit_vertices.empty());
+
+    u32 min_offset = UINT32_MAX;
+    for (const auto &v : lit_vertices) {
+        min_offset = min(min_offset, build.g[v].min_offset);
+    }
+
+    return min_offset;
+}
+
 static
 void makeCheckLitEarlyInstruction(const RoseBuildImpl &build, build_context &bc,
                                   u32 final_id,
@@ -3591,22 +3604,36 @@ void makeCheckLitEarlyInstruction(const RoseBuildImpl &build, build_context &bc,
         return;
     }
 
-    size_t min_offset = SIZE_MAX;
+    size_t min_len = SIZE_MAX;
+    u32 min_offset = UINT32_MAX;
     for (u32 lit_id : lit_ids) {
         const auto &lit = build.literals.right.at(lit_id);
-        min_offset = min(min_offset, lit.elength());
+        size_t lit_min_len = lit.elength();
+        u32 lit_min_offset = findMinOffset(build, lit_id);
+        DEBUG_PRINTF("lit_id=%u has min_len=%zu, min_offset=%u\n", lit_id,
+                     lit_min_len, lit_min_offset);
+        min_len = min(min_len, lit_min_len);
+        min_offset = min(min_offset, lit_min_offset);
     }
 
-    DEBUG_PRINTF("%zu lits, min_offset=%zu\n", lit_ids.size(), min_offset);
+    DEBUG_PRINTF("final_id=%u has min_len=%zu, min_offset=%u, "
+                 "global min is %u\n", final_id, min_len, min_offset,
+                 bc.floatingMinLiteralMatchOffset);
 
     // If we can't match before the min offset, we don't need the check.
-    if (min_offset >= bc.floatingMinLiteralMatchOffset) {
+    if (min_len >= bc.floatingMinLiteralMatchOffset) {
         DEBUG_PRINTF("no need for check, min is %u\n",
-                      bc.floatingMinLiteralMatchOffset);
+                     bc.floatingMinLiteralMatchOffset);
         return;
     }
 
-    program.push_back(RoseInstruction(ROSE_INSTR_CHECK_LIT_EARLY));
+    assert(min_offset >= bc.floatingMinLiteralMatchOffset);
+    assert(min_offset < UINT32_MAX);
+
+    DEBUG_PRINTF("adding lit early check, min_offset=%u\n", min_offset);
+    auto ri = RoseInstruction(ROSE_INSTR_CHECK_LIT_EARLY);
+    ri.u.checkLitEarly.min_offset = min_offset;
+    program.push_back(move(ri));
 }
 
 static

diff --git a/src/rose/rose_dump.cpp b/src/rose/rose_dump.cpp
index f6badd1ba74cdc32d7e7100a3f5686135a7dab6c..ad776780139225a630bd9684bfd143369ed5f039 100644 (file)
--- a/src/rose/rose_dump.cpp
+++ b/src/rose/rose_dump.cpp
@@ -253,7 +253,9 @@ void dumpProgram(ofstream &os, const RoseEngine *t, const char *pc) {
             }
             PROGRAM_NEXT_INSTRUCTION
 
-            PROGRAM_CASE(CHECK_LIT_EARLY) {}
+            PROGRAM_CASE(CHECK_LIT_EARLY) {
+                os << "    min_offset " << ri->min_offset << endl;
+            }
             PROGRAM_NEXT_INSTRUCTION
 
             PROGRAM_CASE(CHECK_GROUPS) {

diff --git a/src/rose/rose_program.h b/src/rose/rose_program.h
index 01572dbd061cdf3c96b8a424045fce840ce8862d..5c57bf548cc9811c45c267041c9db4035044be28 100644 (file)
--- a/src/rose/rose_program.h
+++ b/src/rose/rose_program.h
@@ -120,6 +120,7 @@ struct ROSE_STRUCT_CHECK_LIT_MASK {
 /** Note: check failure will halt program. */
 struct ROSE_STRUCT_CHECK_LIT_EARLY {
     u8 code; //!< From enum RoseInstructionCode.
+    u32 min_offset; //!< Minimum offset for this literal.
 };
 
 /** Note: check failure will halt program. */