Display origin of template instantiations

CMakeLists.txt

@@ -3,18 +3,12 @@ project(ClangBuildAnalyzer)
 
 include(GNUInstallDirs)
 
-set(SRC
+set(EXTERNAL_SRC
 # C sources
     "src/external/cwalk/cwalk.c"
     "src/external/inih/ini.c"
     "src/external/xxHash/xxhash.c"
 # C++ sources
-    "src/Analysis.cpp"
-    "src/Arena.cpp"
-    "src/BuildEvents.cpp"
-    "src/Colors.cpp"
-    "src/main.cpp"
-    "src/Utils.cpp"
     "src/external/enkiTS/TaskScheduler.cpp"
     "src/external/inih/cpp/INIReader.cpp"
     "src/external/llvm-Demangle/lib/Demangle.cpp"
@@ -23,7 +17,34 @@ set(SRC
     "src/external/llvm-Demangle/lib/MicrosoftDemangleNodes.cpp"
     "src/external/simdjson/simdjson.cpp"
 )
-add_executable(ClangBuildAnalyzer "${SRC}")
+
+set(SRC
+# C++ sources
+    "src/Analysis.cpp"
+    "src/Arena.cpp"
+    "src/BuildEvents.cpp"
+    "src/Colors.cpp"
+    "src/main.cpp"
+    "src/Utils.cpp"
+)
+
+set(DISABLE_WARNINGS_FLAG "")
+
+if(CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang|AppleClang|Intel") # Check C++ compiler first
+    set(DISABLE_WARNINGS_FLAG "-w")
+elseif(CMAKE_C_COMPILER_ID MATCHES "GNU|Clang|AppleClang|Intel") # Fallback for C-only
+    set(DISABLE_WARNINGS_FLAG "-w")
+elseif(CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
+    set(DISABLE_WARNINGS_FLAG "/w") # MSVC uses /w
+elseif(CMAKE_C_COMPILER_ID MATCHES "MSVC")
+    set(DISABLE_WARNINGS_FLAG "/w")
+endif()
+
+set_source_files_properties(${EXTERNAL_SRC}
+    PROPERTIES COMPILE_FLAGS "${DISABLE_WARNINGS_FLAG}"
+)
+
+add_executable(ClangBuildAnalyzer "${EXTERNAL_SRC}" "${SRC}")
 target_compile_features(ClangBuildAnalyzer PRIVATE cxx_std_17)
 
 find_library(LIBRT rt)

ClangBuildAnalyzer.ini

@@ -6,17 +6,17 @@
 [counts]
 
 # files that took most time to parse
-fileParse = 10
+fileParse = 20
 # files that took most time to generate code for
-fileCodegen = 10
+fileCodegen = 20
 # functions that took most time to generate code for
 function = 30
 # header files that were most expensive to include
-header = 10
+header = 20
 # for each expensive header, this many include paths to it are shown
-headerChain = 5
+headerChain = 10
 # templates that took longest to instantiate
-template = 30
+template = 50
 
 
 # Minimum times (in ms) for things to be recorded into trace
@@ -29,7 +29,7 @@ file = 10
 [misc]
 
 # Maximum length of symbol names printed; longer names will get truncated
-maxNameLength = 70
+maxNameLength = 512
 
 # Only print "root" headers in expensive header report, i.e.
 # only headers that are directly included by at least one source file

src/Analysis.cpp

@@ -21,16 +21,16 @@ struct IUnknown; // workaround for old Win SDK header failures when using /permi
 
 struct Config
 {
-    int fileParseCount = 10;
-    int fileCodegenCount = 10;
-    int templateCount = 30;
+    int fileParseCount = 20;
+    int fileCodegenCount = 20;
+    int templateCount = 50;
     int functionCount = 30;
-    int headerCount = 10;
-    int headerChainCount = 5;
+    int headerCount = 20;
+    int headerChainCount = 10;
 
     int minFileTime = 10;
 
-    int maxName = 70;
+    int maxName = 512;
 
     bool onlyRootHeaders = true;
 };
@@ -87,6 +87,8 @@ struct Analysis
     {
         int count = 0;
         int64_t us = 0;
+        DetailIndex file;
+        std::unordered_map<int, int> templateCount;
     };
     struct FileEntry
     {
@@ -115,6 +117,8 @@ struct Analysis
         const ska::bytell_hash_map<std::string_view, InstantiateEntry> &collapsed,
         const char *header_string);
 
+    void PrintSortedTemplateCounts(const InstantiateEntry& e);
+
     // key is (name,objfile), value is milliseconds
     typedef std::pair<DetailIndex, DetailIndex> IndexPair;
     ska::bytell_hash_map<IndexPair, int64_t, pair_hash> functions;
@@ -136,7 +140,6 @@ DetailIndex Analysis::FindPath(EventIndex eventIndex) const
     while(eventIndex > EventIndex())
     {
         const BuildEvent& ev = events[eventIndex];
-        if (ev.type == BuildEventType::kCompiler || ev.type == BuildEventType::kFrontend || ev.type == BuildEventType::kBackend || ev.type == BuildEventType::kOptModule)
             if (ev.detailIndex != DetailIndex())
                 return ev.detailIndex;
         eventIndex = ev.parent;
@@ -160,6 +163,7 @@ void Analysis::ProcessEvent(EventIndex eventIndex)
         auto& e = instantiations[eventIndex];
         ++e.count;
         e.us += event.dur;
+        e.file = event.filenameDetailIndex;
     }
 
     if (event.type == BuildEventType::kFrontend)
@@ -282,6 +286,33 @@ std::string_view Analysis::GetCollapsedName(DetailIndex detail)
     return name;
 }
 
+void Analysis::PrintSortedTemplateCounts(const InstantiateEntry& e)
+{
+    std::vector<std::pair<int, int>> templateCounts;
+    templateCounts.reserve(e.templateCount.size());
+
+    for (const auto& t : e.templateCount)
+    {
+        if (t.second < 3) // TODO: make customizable
+            continue;
+
+        templateCounts.emplace_back(t.first, t.second);
+    }
+
+    std::sort(templateCounts.begin(), templateCounts.end(), [](const auto& a, const auto& b)
+    {
+        if (a.second != b.second)
+            return a.second > b.second;
+
+        return a.first < b.first;
+    });
+
+    for (const auto& t : templateCounts)
+    {
+        fprintf(out, "  %s%6i%s time(s): %s\n", col::kBold, t.second, col::kReset, GetBuildName(DetailIndex{t.first}).data());
+    }
+}
+
 void Analysis::EmitCollapsedInfo(
     const ska::bytell_hash_map<std::string_view, InstantiateEntry> &collapsed,
     const char *header_string)
@@ -305,6 +336,8 @@ void Analysis::EmitCollapsedInfo(
         int ms = int(elt.second.us / 1000);
         int avg = int(ms / elt.second.count);
         fprintf(out, "%s%6i%s ms: %s (%i times, avg %i ms)\n", col::kBold, ms, col::kReset, dname.c_str(), elt.second.count, avg);
+
+        PrintSortedTemplateCounts(elt.second);
     }
     fprintf(out, "\n");
 }
@@ -336,6 +369,7 @@ void Analysis::EmitCollapsedTemplates()
         {
             stats.us += inst.second.us;
             stats.count += inst.second.count;
+            stats.templateCount[inst.second.file.idx] += 1;
         }
     }
     EmitCollapsedInfo(collapsed, "Template sets that took longest to instantiate");
@@ -423,6 +457,7 @@ void Analysis::EndAnalysis()
             instArray[d.idx].first = d;
             instArray[d.idx].second.us += inst.second.us;
             instArray[d.idx].second.count += inst.second.count;
+            instArray[d.idx].second.templateCount[inst.second.file.idx] += 1;
         }
         size_t n = std::min<size_t>(config.templateCount, instArray.size());
         auto cmp = [&](const auto&a, const auto &b) {
@@ -441,6 +476,8 @@ void Analysis::EndAnalysis()
             int ms = int(e.second.us / 1000);
             int avg = int(ms / std::max(e.second.count,1));
             fprintf(out, "%s%6i%s ms: %s (%i times, avg %i ms)\n", col::kBold, ms, col::kReset, dname.c_str(), e.second.count, avg);
+
+            PrintSortedTemplateCounts(e.second);
         }
         fprintf(out, "\n");
 

src/BuildEvents.cpp

@@ -215,8 +215,10 @@ struct BuildEventsParser
             BuildEvent& ev = resultEvents[EventIndex(int(i))];
             if (ev.parent.idx >= 0)
                 ev.parent.idx += offset;
+
             for (auto& ch : ev.children)
                 ch.idx += offset;
+
             if (ev.detailIndex.idx != 0)
             {
                 assert(ev.detailIndex.idx >= 0);
@@ -225,6 +227,15 @@ struct BuildEventsParser
                 assert(ev.detailIndex.idx >= 0);
                 assert(ev.detailIndex.idx < static_cast<int>(resultNameToIndex.size()));
             }
+
+            if (ev.filenameDetailIndex.idx != 0)
+            {
+                assert(ev.filenameDetailIndex.idx >= 0);
+                assert(ev.filenameDetailIndex.idx < static_cast<int>(nameToIndex.size()));
+                ev.filenameDetailIndex = detailRemap[ev.filenameDetailIndex];
+                assert(ev.filenameDetailIndex.idx >= 0);
+                assert(ev.filenameDetailIndex.idx < static_cast<int>(resultNameToIndex.size()));
+            }
         }
 
         assert(resultNameToIndex.size() == resultNames.size());
@@ -321,6 +332,7 @@ struct BuildEventsParser
         BuildEvent event;
         bool valid = true;
         std::string_view detailPtr;
+        std::string_view filenameDetailPtr;
         for (const simdjson::dom::key_value_pair& kv : node)
         {
             std::string_view nodeKey = kv.key;
@@ -400,20 +412,20 @@ struct BuildEventsParser
         if (event.type== BuildEventType::kUnknown || !valid)
             return;
 
-        // if the "compiler" event has no detail name, use the current json file name
-        if (detailPtr.empty() && event.type == BuildEventType::kCompiler)
-            detailPtr = curFileName;
-        if (!detailPtr.empty())
+        const auto processDetail = [this, &event, &nameToIndexLocal](std::string_view& theDetailPtr) -> DetailIndex
         {
+            if (theDetailPtr.empty())
+                return DetailIndex{};
+
             std::string detailString;
             if (event.type == BuildEventType::kParseFile || event.type == BuildEventType::kOptModule)
             {
                 // do various cleanups/nice-ifications of the detail name:
                 // make paths shorter (i.e. relative to project) where possible
-                detailString = utils::GetNicePath(detailPtr);
+                detailString = utils::GetNicePath(theDetailPtr);
             }
             else
-                detailString = detailPtr;
+                detailString = theDetailPtr;
 
             // don't report the clang trace .json file, instead get the object file at the same location if it's there
             if (utils::EndsWith(detailString, ".json"))
@@ -435,8 +447,20 @@ struct BuildEventsParser
             if (event.type == BuildEventType::kOptFunction)
                 detailString = llvm::demangle(detailString);
 
-            event.detailIndex = NameToIndex(detailString.c_str(), nameToIndexLocal);
-        }
+            return NameToIndex(detailString.c_str(), nameToIndexLocal);
+        };
+
+        // if the "compiler" event has no detail name, use the current json file name
+        if (detailPtr.empty() && event.type == BuildEventType::kCompiler)
+            detailPtr = curFileName;
+
+        event.detailIndex = processDetail(detailPtr);
+
+        // use the current json file name for the "instantiateX" event filename detail
+        if (event.type == BuildEventType::kInstantiateClass || event.type == BuildEventType::kInstantiateFunction)
+            filenameDetailPtr = curFileName;
+
+        event.filenameDetailIndex = processDetail(filenameDetailPtr);
 
         // starting with clang 19, some Source events are pairs of "b" immediately followed
         // by "e" events. Handle merging of those if needed; do not support other cases of
@@ -610,6 +634,7 @@ bool SaveBuildEvents(BuildEventsParser* parser, const std::string& fileName)
         w.Write(e.ts);
         w.Write(e.dur);
         w.Write(e.detailIndex.idx);
+        w.Write(e.filenameDetailIndex.idx);
         w.Write(e.parent.idx);
         int64_t childCount = e.children.size();
         w.Write(childCount);
@@ -670,6 +695,7 @@ bool LoadBuildEvents(const std::string& fileName, BuildEvents& outEvents, BuildN
         r.Read(e.ts);
         r.Read(e.dur);
         r.Read(e.detailIndex.idx);
+        r.Read(e.filenameDetailIndex.idx);
         r.Read(e.parent.idx);
         int64_t childCount = 0;
         r.Read(childCount);

src/BuildEvents.h

@@ -80,6 +80,7 @@ struct BuildEvent
     int64_t ts = 0;
     int64_t dur = 0;
     DetailIndex detailIndex;
+    DetailIndex filenameDetailIndex;
     EventIndex parent{ -1 };
     char phase = 0;
     std::vector<EventIndex> children;