Merge pull request #44 from atom-ide-community/performance

Author: aminya

binding.gyp

@@ -11,9 +11,9 @@
       'configurations': {
         # Release Settings
         'Release': {
-          'defines': [ 'NAPI_DISABLE_CPP_EXCEPTIONS' ],
-          "cflags": [ "-fno-exceptions", "-O3" ],
-          "cflags_cc": [ "-fno-exceptions", "-O3", "-std=c++2a" ],
+          'defines': [ 'NAPI_DISABLE_CPP_EXCEPTIONS', 'NDEBUG' ],
+          "cflags": [ "-fno-exceptions", "-Ofast" ],
+          "cflags_cc": [ "-fno-exceptions", "-Ofast", "-std=c++2a" ],
           "xcode_settings": {
             'GCC_OPTIMIZATION_LEVEL': '3', # stop gyp from defaulting to -Os
             "CLANG_CXX_LIBRARY": "libc++",
@@ -28,6 +28,11 @@
               "AdditionalOptions": [
                 # C++ standard
                 "/std:c++latest",
+                "/O2",      # optimizations
+                "/Ob3",     # agressive inline
+                "/Oi",      # intrinsic functions
+                "/Ot",      # favor speed
+                "/DNDEBUG"  # turn off asserts
               ],
               'EnableFunctionLevelLinking': 'true',
               'EnableIntrinsicFunctions': 'true',
@@ -41,7 +46,7 @@
         },
         # Debug Settings
         'Debug': {
-          'defines': [ 'DEBUG', 'NAPI_CPP_EXCEPTIONS', 'ENABLE_DEBUG' ],
+          'defines': [ 'DEBUG', 'NAPI_CPP_EXCEPTIONS' ],
           'cflags': [ '-g', '-O0' ],
           "cflags_cc": [
             '-fexceptions', # enable exceptions

src/common.h

@@ -9,6 +9,7 @@
 #include <cstring>
 #include <iostream>
 #include <utility>
+#include <cassert>
 
 #include <napi.h>
 
@@ -17,6 +18,7 @@ using namespace std;
 constexpr size_t kMaxThreads = 16;
 
 #ifdef ENABLE_DEBUG
+// TODO does not work anymore because we added explicit to constructors
 // Safe string class that logs error when index is accessed outside the string.
 class SafeString : public std::string {
   public:
@@ -75,9 +77,9 @@ struct Options {
 struct AcronymResult {
     Score score;
     float pos;
-    int count;
+    size_t count;
 
-    explicit AcronymResult(Score s, float p, int c) noexcept : score(s), pos(p), count(c) {}
+    explicit AcronymResult(Score s, float p, size_t c) noexcept : score(s), pos(p), count(c) {}
 };
 
 extern Element ToLower(const Element &s);

src/filter.cc

@@ -24,12 +24,12 @@ void filter_internal(const std::vector<CandidateString> &candidates,
   const Options &options,
   size_t max_results,
   CandidateScorePriorityQueue &results) {
+    const auto scoreProvider = options.usePathScoring ? path_scorer_score : scorer_score;
     for (size_t i = 0, len = candidates.size(); i < len; i++) {
         const auto &candidate = candidates[i];
         if (candidate.empty()) {
             continue;
         }
-        const auto scoreProvider = options.usePathScoring ? path_scorer_score : scorer_score;
         auto score = scoreProvider(candidate, query, options);
         if (score > 0) {
             results.emplace(score, start_index + i);
@@ -40,20 +40,14 @@ void filter_internal(const std::vector<CandidateString> &candidates,
     }
 }
 
-void thread_worker_filter(const std::vector<CandidateString> &candidates,
-  size_t start_index,
-  const Element &query,
-  const Options &options,
-  size_t max_results,
-  CandidateScorePriorityQueue &results) {
-    filter_internal(candidates, start_index, query, options, max_results, results);
-}
-
 std::vector<CandidateIndex> sort_priority_queue(CandidateScorePriorityQueue &&candidates) {
     vector<CandidateScore> sorted;
     std::vector<CandidateIndex> ret;
-    sorted.reserve(candidates.size());
-    ret.reserve(candidates.size());
+
+    const auto initial_candidates_size = candidates.size();
+    sorted.reserve(initial_candidates_size);
+    ret.reserve(initial_candidates_size);
+
     while (!candidates.empty()) {
         sorted.emplace_back(candidates.top());
         candidates.pop();
@@ -66,6 +60,10 @@ std::vector<CandidateIndex> sort_priority_queue(CandidateScorePriorityQueue &&ca
 }
 
 std::vector<CandidateIndex> filter(const vector<std::vector<CandidateString>> &candidates, const Element &query, const Options &options) {
+    const auto candidates_size = candidates.size();
+
+    assert(1 <= candidates_size);// TODO handled outside
+
     CandidateScorePriorityQueue top_k;
     auto max_results = options.max_results;
     if (max_results == 0u) {
@@ -74,17 +72,24 @@ std::vector<CandidateIndex> filter(const vector<std::vector<CandidateString>> &c
 
     // Split the dataset and pass down to multiple threads.
     vector<thread> threads;
-    vector<CandidateScorePriorityQueue> results(candidates.size());
+    threads.reserve(candidates.size());
+
+    auto results = vector<CandidateScorePriorityQueue>(candidates.size());
+
     size_t start_index = 0;
-    for (size_t i = 1, len = candidates.size(); i < len; i++) {
-        start_index += candidates[i - 1].size();
-        threads.emplace_back(thread_worker_filter, ref(candidates[i]), start_index, ref(query), ref(options), max_results, ref(results[i]));
+    for (size_t i = 1; i < candidates_size; i++) {
+        assert(1 < i && i < candidates.size() && i < results.size());
+        start_index += candidates[i - 1].size();//inbounds
+        threads.emplace_back(filter_internal, ref(candidates[i]), start_index, ref(query), ref(options), max_results, ref(results[i]));// inbounds
     }
+
+    assert(threads.size() == candidates.size() && results.size() == candidates.size());
+
     // Do the work for first thread.
-    filter_internal(candidates[0], 0, query, options, max_results, top_k);
-    // Wait for threads to complete and merge the restuls.
-    for (size_t i = 1, len = candidates.size(); i < len; i++) {
-        threads[i - 1].join();
+    filter_internal(candidates[0], 0, query, options, max_results, top_k);//inbounds (candidate_size >= 1)
+    // Wait for threads to complete and merge the results.
+    for (size_t i = 1; i < candidates_size; i++) {
+        threads[i - 1].join();//inbounds
         while (!results[i].empty()) {
             top_k.emplace(results[i].top());
             results[i].pop();

src/matcher.cc

@@ -20,16 +20,16 @@ std::vector<size_t> computeMatch(const CandidateString &subject, const Candidate
     const auto &query_lw = preparedQuery.query_lw;
 
     // TODO making these two auto breaks the code. There are a lot of narrowing conversions in this file
-    const int m = subject.size();
-    const int n = query.size();
+    const int subject_size = subject.size();
+    const int query_size = query.size();
 
     // this is like the consecutive bonus, but for camelCase / snake_case initials
     const auto acro = scoreAcronyms(subject, subject_lw, query, query_lw);
     const auto acro_score = acro.score;
 
     // Init
-    vector<Score> score_row(n, 0);
-    vector<Score> csc_row(n, 0);
+    vector<Score> score_row(query_size, 0);
+    vector<Score> csc_row(query_size, 0);
 
     // Directions constants
     enum class Direction {
@@ -40,18 +40,18 @@ std::vector<size_t> computeMatch(const CandidateString &subject, const Candidate
     };
 
     // Traceback matrix
-    std::vector<Direction> trace(m * n, Direction::STOP);
+    std::vector<Direction> trace(subject_size * query_size, Direction::STOP);
     auto pos = -1;
 
     auto i = -1;
-    while (++i < m) {//foreach char is of subject
+    while (++i < subject_size) {//foreach char is of subject
         Score score = 0;
         Score score_up = 0;
         Score csc_diag = 0;
         const auto si_lw = subject_lw[i];
 
         auto j = -1;//0..n-1
-        while (++j < n) {//foreach char qj of query
+        while (++j < query_size) {//foreach char qj of query
             // reset score
             Score csc_score = 0;
             Score align = 0;
@@ -100,17 +100,17 @@ std::vector<size_t> computeMatch(const CandidateString &subject, const Candidate
     // Go back in the trace matrix
     // and collect matches (diagonals)
 
-    i = m - 1;
-    auto j = n - 1;
-    pos = i * n + j;
+    i = subject_size - 1;
+    auto j = query_size - 1;
+    pos = i * query_size + j;
     auto backtrack = true;
     std::vector<size_t> matches;
 
     while (backtrack && i >= 0 && j >= 0) {
         switch (trace[pos]) {
         case Direction::UP:
             i--;
-            pos -= n;
+            pos -= query_size;
             break;
         case Direction::LEFT:
             j--;
@@ -120,7 +120,7 @@ std::vector<size_t> computeMatch(const CandidateString &subject, const Candidate
             matches.emplace_back(i + offset);
             j--;
             i--;
-            pos -= n + 1;
+            pos -= query_size + 1;
             break;
         default:
             backtrack = false;
@@ -172,13 +172,13 @@ std::vector<size_t> basenameMatch(const CandidateString &subject, const Candidat
 // (Assume sequences are sorted, matches are sorted by construction.)
 //
 std::vector<size_t> mergeMatches(const std::vector<size_t> &a, const std::vector<size_t> &b) {
-    const auto m = a.size();
-    const auto n = b.size();
+    const auto a_size = a.size();
+    const auto b_size = b.size();
 
-    if (n == 0) {
+    if (b_size == 0) {
         return a;
     }
-    if (m == 0) {
+    if (a_size == 0) {
         return b;
     }
 
@@ -187,18 +187,18 @@ std::vector<size_t> mergeMatches(const std::vector<size_t> &a, const std::vector
     auto bj = b[j];
     std::vector<size_t> out;
 
-    while (++i < m) {
+    while (++i < a_size) {
         auto ai = a[i];
 
-        while (bj <= ai && ++j < n) {
+        while (bj <= ai && ++j < b_size) {
             if (bj < ai) {
                 out.emplace_back(bj);
             }
             bj = b[j];
         }
         out.emplace_back(ai);
     }
-    while (j < n) {
+    while (j < b_size) {
         out.emplace_back(b[j++]);
     }
     return out;

src/path_scorer.cc

@@ -51,7 +51,7 @@ Score scorePath(const CandidateString &subject, const CandidateString &subject_l
     // {preparedQuery, useExtensionBonus, pathSeparator} = options
 
     // Skip trailing slashes
-    auto end = subject.size() - 1;
+    int end = subject.size() - 1;
     while (subject[end] == options.pathSeparator) {
         end--;
     }
@@ -110,17 +110,19 @@ int countDir(const CandidateString &path, const size_t end, const char pathSepar
 
     //skip slash at the start so `foo/bar` and `/foo/bar` have the same depth.
     while ((i < end) && (path[i] == pathSeparator)) {//inbounds
-        // assert(i>=0); fuzz: if end==0, it does not enter while and i==0
+        assert(0 <= i);// fuzz: if end==0, it does not enter while and i==0
         ++i;
     }
+    assert(0 <= i);
 
     while (++i < end) {
-        // assert(i>=0); fuzz: if end==0, it does not enter while and i==0
+        assert(0 <= i && i < path.size());// fuzz: if end==0, it does not enter while and i==0
         if (path[i] == pathSeparator) {//inbounds
             count++;//record first slash, but then skip consecutive ones
             while ((++i < end) && (path[i] == pathSeparator)) {}
         }
     }
+    assert(0 <= i && i < path.size());
 
     return count;
 }
@@ -146,31 +148,32 @@ Score getExtensionScore(const CandidateString &candidate, const CandidateString
 
     // Check that (a) extension exist, (b) it is after the start of the basename
     int pos = candidate.rfind('.', endPos);
-    // assert(pos >= 0u);
+    assert(pos >= 0u);
     if (pos <= startPos) {
         return 0;// (note that startPos >= -1)
     }
 
-    int n = ext.size();
+    int ext_size = ext.size();
     auto m = endPos - pos;
 
     // n contain the smallest of both extension length, m the largest.
-    if (m < n) {
-        n = m;
+    if (m < ext_size) {
+        ext_size = m;
         m = ext.size();
     }
 
     //place cursor after dot & count number of matching characters in extension
     pos++;
-    // assert(pos >= 1u);
+    assert(pos >= 1u);
     auto matched = 0;
-    while (matched < n) {
-        // assert(matched >=0); // fuzz: if n==0, does not enter while and matched==0
+    while (matched < ext_size) {
+        assert(matched >= 0);// fuzz: if n==0, does not enter while and matched==0
         if (candidate[pos + matched] != ext[matched]) {// TODO candidate upper bound
             break;
         }
         ++matched;
     }
+    assert(matched >= 0);
 
     // if nothing found, try deeper for multiple extensions, with some penalty for depth
     if (matched == 0u && maxDepth > 0) {

src/query.cc

@@ -28,9 +28,10 @@ std::set<char> getCharCodes(const Element &str) {
 
     // create map
     while (i < len) {
-        // assert(i>=0); // fuzz: if len==0, does not enter while and i==0
+        assert(0 <= i && i < str.size());// fuzz: if len==0, does not enter while and i==0
         charCodes.insert(str[i]);//inbounds
         ++i;
     }
+    assert(0 <= i && i < str.size());
     return charCodes;
 }