This massively improves the performance of tight loops relying on a type() call. (#721)

* This massively improves the performance of tight loops relying on a type() call.

* Adding a few more benchmarks

benchmark/bench_dom_api.cpp

@@ -202,8 +202,8 @@ static void numbers_exceptions_size_scan(State& state) {
     std::vector<double> container;
     container.resize(arr.size());
     size_t pos = 0;
-    for (double x : arr) {
+    for (auto e : arr) {
-      container[pos++] = x;
+      container[pos++] = double(e);
     }
     if(pos != container.size()) { cerr << "bad count" << endl; }
     benchmark::DoNotOptimize(container.data());
@@ -213,6 +213,48 @@ static void numbers_exceptions_size_scan(State& state) {
 BENCHMARK(numbers_exceptions_size_scan);
 
 
+
+static void numbers_type_exceptions_scan(State& state) {
+  // Prints the number of results in twitter.json
+  dom::parser parser;
+  dom::array arr = parser.load(NUMBERS_JSON);
+  for (auto _ : state) {
+    std::vector<double> container;
+    for (auto e : arr) {
+      dom::element_type actual_type = e.type();
+      if(actual_type != dom::element_type::DOUBLE) {
+        cerr << "found a node that is not an number?" << endl; break;
+      }
+      container.push_back(double(e));
+    }
+    benchmark::DoNotOptimize(container.data());
+    benchmark::ClobberMemory();
+  }
+}
+BENCHMARK(numbers_type_exceptions_scan);
+
+static void numbers_type_exceptions_size_scan(State& state) {
+  // Prints the number of results in twitter.json
+  dom::parser parser;
+  dom::array arr = parser.load(NUMBERS_JSON);
+  for (auto _ : state) {
+    std::vector<double> container;
+    container.resize(arr.size());
+    size_t pos = 0;
+    for (auto e : arr) {
+      dom::element_type actual_type = e.type();
+      if(actual_type != dom::element_type::DOUBLE) {
+        cerr << "found a node that is not an number?" << endl; break;
+      }
+      container[pos++] = double(e);
+    }
+    if(pos != container.size()) { cerr << "bad count" << endl; }
+    benchmark::DoNotOptimize(container.data());
+    benchmark::ClobberMemory();
+  }
+}
+BENCHMARK(numbers_type_exceptions_size_scan);
+
 static void numbers_exceptions_load_scan(State& state) {
   // Prints the number of results in twitter.json
   dom::parser parser;

include/simdjson/document.h

@@ -44,7 +44,7 @@ constexpr const uint64_t JSON_VALUE_MASK = 0x00FFFFFFFFFFFFFF;
 constexpr const uint32_t JSON_COUNT_MASK = 0xFFFFFF;
 
 /**
- * The possible types in the tape. Internal only.
+ * The possible types in the tape.
  */
 enum class tape_type {
   ROOT = 'r',
@@ -99,14 +99,14 @@ namespace simdjson::dom {
  * This is the type it is most easily cast to with get<>.
  */
 enum class element_type {
-  ARRAY,     ///< dom::array
+  ARRAY = '[',     ///< dom::array
-  OBJECT,    ///< dom::object
+  OBJECT = '{',    ///< dom::object
-  INT64,     ///< int64_t
+  INT64 = 'l',     ///< int64_t
-  UINT64,    ///< uint64_t: any integer that fits in uint64_t but *not* int64_t
+  UINT64 = 'u',    ///< uint64_t: any integer that fits in uint64_t but *not* int64_t
-  DOUBLE,    ///< double: Any number with a "." or "e" that fits in double.
+  DOUBLE = 'd',    ///< double: Any number with a "." or "e" that fits in double.
-  STRING,    ///< std::string_view
+  STRING = '"',    ///< std::string_view
-  BOOL,      ///< bool
+  BOOL = 't',      ///< bool
-  NULL_VALUE ///< null
+  NULL_VALUE = 'n' ///< null
 };
 
 /**

include/simdjson/inline/document.h

@@ -738,30 +738,8 @@ really_inline element::element(const document *_doc, size_t _json_index) noexcep
 
 
 inline element_type element::type() const noexcept {
-  switch (tape_ref_type()) {
+  auto tape_type = tape_ref_type();
-    case internal::tape_type::START_ARRAY:
+  return tape_type == internal::tape_type::FALSE_VALUE ? element_type::BOOL : static_cast<element_type>(tape_type);
-      return element_type::ARRAY;
-    case internal::tape_type::START_OBJECT:
-      return element_type::OBJECT;
-    case internal::tape_type::INT64:
-      return element_type::INT64;
-    case internal::tape_type::UINT64:
-      return element_type::UINT64;
-    case internal::tape_type::DOUBLE:
-      return element_type::DOUBLE;
-    case internal::tape_type::STRING:
-      return element_type::STRING;
-    case internal::tape_type::TRUE_VALUE:
-    case internal::tape_type::FALSE_VALUE:
-      return element_type::BOOL;
-    case internal::tape_type::NULL_VALUE:
-      return element_type::NULL_VALUE;
-    case internal::tape_type::ROOT:
-    case internal::tape_type::END_ARRAY:
-    case internal::tape_type::END_OBJECT:
-    default:
-      abort();
-  }
 }
 really_inline bool element::is_null() const noexcept {
   return is_null_on_tape();