Merge branch develop into master

CHANGELOG.md

@@ -2,6 +2,15 @@
 
 This is a list of notable changes to Hyperscan, in reverse chronological order.
 
+## [4.5.1] 2017-06-16
+- Bugfix for issue #56: workaround for gcc-4.8 C++11 defect.
+- Bugfix for literal matching table generation, reversing a regression in
+  performance for some literal matching cases.
+- Bugfixes for hsbench, related to multicore benchmarking, portability fixes
+  for FreeBSD, and clarifying output results.
+- CMake: removed a duplicate else branch that causes very recent (v3.9) builds
+  of CMake to fail.
+
 ## [4.5.0] 2017-06-09
 - New API feature: approximate matching using the "edit distance" extended
   parameter. This allows the user to request all matches that are a given edit

CMakeLists.txt

@@ -3,7 +3,7 @@ project (hyperscan C CXX)
 
 set (HS_MAJOR_VERSION 4)
 set (HS_MINOR_VERSION 5)
-set (HS_PATCH_VERSION 0)
+set (HS_PATCH_VERSION 1)
 set (HS_VERSION ${HS_MAJOR_VERSION}.${HS_MINOR_VERSION}.${HS_PATCH_VERSION})
 
 set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake)

cmake/arch.cmake

@@ -75,7 +75,6 @@ else (NOT FAT_RUNTIME)
     if (NOT HAVE_AVX512)
         message(STATUS "Building without AVX512 support")
     endif ()
-else (NOT FAT_RUNTIME)
     if (NOT HAVE_SSSE3)
         message(FATAL_ERROR "A minimum of SSSE3 compiler support is required")
     endif ()

src/fdr/fdr_confirm_compile.cpp

@@ -163,7 +163,7 @@ bytecode_ptr<FDRConfirm> getFDRConfirm(const vector<hwlmLiteral> &lits,
     if (make_small) {
         nBits = min(10U, lg2(lits.size()) + 1);
     } else {
-        nBits = lg2(lits.size() + 4);
+        nBits = lg2(lits.size()) + 4;
     }
 
     CONF_TYPE mult = (CONF_TYPE)0x0b4e0ef37bc32127ULL;

src/util/small_vector.h

@@ -55,6 +55,9 @@ using small_vector = boost::container::small_vector<T, N, Allocator>;
 template <class T, std::size_t N, typename Allocator = std::allocator<T>>
 using small_vector = std::vector<T, Allocator>;
 
+// Support workarounds for flat_set/flat_map and GCC 4.8.
+#define SMALL_VECTOR_IS_STL_VECTOR 1
+
 #endif // HAVE_BOOST_CONTAINER_SMALL_VECTOR
 
 } // namespace ue2

src/util/ue2_containers.h

@@ -162,9 +162,19 @@ class flat_set
       public totally_ordered<flat_set<T, Compare, Allocator>> {
     using base_type = flat_detail::flat_base<T, Compare, Allocator>;
     using storage_type = typename base_type::storage_type;
+    using storage_iterator = typename storage_type::iterator;
+    using storage_const_iterator = typename storage_type::const_iterator;
     using base_type::data;
     using base_type::comp;
 
+#if defined(SMALL_VECTOR_IS_STL_VECTOR)
+    // Construct a non-const iterator from a const iterator. Used in flat_map
+    // and flat_set erase() calls to work around g++-4.8 compatibility issues.
+    storage_iterator mutable_iterator(storage_const_iterator it) {
+        return data().begin() + std::distance(data().cbegin(), it);
+    }
+#endif
+
 public:
     // Member types.
     using key_type = T;
@@ -282,11 +292,27 @@ public:
     }
 
     void erase(const_iterator pos) {
-        data().erase(pos.get());
+#if defined(SMALL_VECTOR_IS_STL_VECTOR)
+        // Cope with libstdc++ 4.8's incomplete STL (it's missing C++11
+        // vector::erase(const_iterator)) by explicitly using a non-const
+        // iterator.
+        auto pos_it = mutable_iterator(pos.get());
+#else
+        auto pos_it = pos.get();
+#endif
+        data().erase(pos_it);
     }
 
     void erase(const_iterator first, const_iterator last) {
-        data().erase(first.get(), last.get());
+#if defined(SMALL_VECTOR_IS_STL_VECTOR)
+        // As above, work around libstdc++ 4.8's incomplete C++11 support.
+        auto first_it = mutable_iterator(first.get());
+        auto last_it = mutable_iterator(last.get());
+#else
+        auto first_it = first.get();
+        auto last_it = last.get();
+#endif
+        data().erase(first_it, last_it);
     }
 
     void erase(const key_type &key) {
@@ -374,9 +400,19 @@ private:
         flat_detail::flat_base<std::pair<Key, T>, Compare, Allocator>;
     using keyval_storage_type = std::pair<key_type, mapped_type>;
     using storage_type = typename base_type::storage_type;
+    using storage_iterator = typename storage_type::iterator;
+    using storage_const_iterator = typename storage_type::const_iterator;
     using base_type::data;
     using base_type::comp;
 
+#if defined(SMALL_VECTOR_IS_STL_VECTOR)
+    // Construct a non-const iterator from a const iterator. Used in flat_map
+    // and flat_set erase() calls to work around g++-4.8 compatibility issues.
+    storage_iterator mutable_iterator(storage_const_iterator it) {
+        return data().begin() + std::distance(data().cbegin(), it);
+    }
+#endif
+
 public:
     // More Member types.
     using size_type = typename storage_type::size_type;
@@ -444,9 +480,6 @@ public:
     const_reverse_iterator rend() const { return crend(); }
 
 private:
-    using storage_iterator = typename storage_type::iterator;
-    using storage_const_iterator = typename storage_type::const_iterator;
-
     storage_iterator data_lower_bound(const key_type &key) {
         return std::lower_bound(
             data().begin(), data().end(), key,
@@ -526,11 +559,27 @@ public:
     }
 
     void erase(const_iterator pos) {
-        data().erase(pos.get());
+#if defined(SMALL_VECTOR_IS_STL_VECTOR)
+        // Cope with libstdc++ 4.8's incomplete STL (it's missing C++11
+        // vector::erase(const_iterator)) by explicitly using a non-const
+        // iterator.
+        auto pos_it = mutable_iterator(pos.get());
+#else
+        auto pos_it = pos.get();
+#endif
+        data().erase(pos_it);
     }
 
     void erase(const_iterator first, const_iterator last) {
-        data().erase(first.get(), last.get());
+#if defined(SMALL_VECTOR_IS_STL_VECTOR)
+        // As above, work around libstdc++ 4.8's incomplete C++11 support.
+        auto first_it = mutable_iterator(first.get());
+        auto last_it = mutable_iterator(last.get());
+#else
+        auto first_it = first.get();
+        auto last_it = last.get();
+#endif
+        data().erase(first_it, last_it);
     }
 
     void erase(const key_type &key) {

tools/hsbench/main.cpp

@@ -125,10 +125,10 @@ public:
     // Apply processor affinity (if available) to this thread.
     bool affine(UNUSED int cpu) {
 #ifdef HAVE_DECL_PTHREAD_SETAFFINITY_NP
-#if defined(__linux__)
+#if defined(__FreeBSD__)
-        cpu_set_t cpuset;
-#else // BSD
         cpuset_t cpuset;
+#else
+        cpu_set_t cpuset;
 #endif
         CPU_ZERO(&cpuset);
         assert(cpu >= 0 && cpu < CPU_SETSIZE);
@@ -205,7 +205,7 @@ static
 void processArgs(int argc, char *argv[], vector<BenchmarkSigs> &sigSets,
                  UNUSED unique_ptr<Grey> &grey) {
     const char options[] = "-b:c:Cd:e:E:G:hi:n:No:p:sVw:z:"
-#if HAVE_DECL_PTHREAD_SETAFFINITY_N
+#ifdef HAVE_DECL_PTHREAD_SETAFFINITY_NP
         "T:" // add the thread flag
 #endif
         ;
@@ -287,6 +287,7 @@ void processArgs(int argc, char *argv[], vector<BenchmarkSigs> &sigSets,
         case 'V':
             scan_mode = ScanMode::VECTORED;
             break;
+#ifdef HAVE_DECL_PTHREAD_SETAFFINITY_NP
         case 'T':
             if (!strToList(optarg, threadCores)) {
                 usage("Couldn't parse argument to -T flag, should be"
@@ -294,6 +295,7 @@ void processArgs(int argc, char *argv[], vector<BenchmarkSigs> &sigSets,
                 exit(1);
             }
             break;
+#endif
         case 'z': {
             unsigned int sinumber;
             if (!fromString(optarg, sinumber)) {
@@ -649,8 +651,8 @@ void displayResults(const vector<unique_ptr<ThreadContext>> &threads,
         }
     }
 
-    printf("Time spent scanning:     %'0.3f seconds\n", totalSecs);
+    printf("Time spent scanning:       %'0.3f seconds\n", totalSecs);
-    printf("Corpus size:             %'llu bytes ", bytesPerRun);
+    printf("Corpus size:               %'llu bytes ", bytesPerRun);
     switch (scan_mode) {
     case ScanMode::STREAMING:
         printf("(%'zu blocks in %'llu streams)\n", corpus_blocks.size(),
@@ -669,16 +671,16 @@ void displayResults(const vector<unique_ptr<ThreadContext>> &threads,
     u64a totalBlocks = corpus_blocks.size() * repeats * threads.size();
 
     double matchRate = ((double)matchesPerRun * 1024) / bytesPerRun;
-    printf("Matches per iteration:   %'llu (%'0.3f matches/kilobyte)\n",
+    printf("Matches per iteration:     %'llu (%'0.3f matches/kilobyte)\n",
            matchesPerRun, matchRate);
 
     double blockRate = (double)totalBlocks / (double)totalSecs;
-    printf("Overall block rate:      %'0.2f blocks/sec\n", blockRate);
+    printf("Overall block rate:        %'0.2f blocks/sec\n", blockRate);
-    printf("Mean throughput:         %'0.2Lf Mbit/sec\n",
+    printf("Mean throughput (overall): %'0.2Lf Mbit/sec\n",
            calc_mbps(totalSecs, totalBytes));
 
     double lowestScanTime = fastestResult(threads);
-    printf("Maximum throughput:      %'0.2Lf Mbit/sec\n",
+    printf("Max throughput (per core): %'0.2Lf Mbit/sec\n",
            calc_mbps(lowestScanTime, bytesPerRun));
     printf("\n");