[SpecialCaseList] Add RadixTree for substring matching

[vitalybuka]

This commit adds a new RadixTree to SpecialCaseList for handling
substring matches. Previously, SpecialCaseList only supported prefix
and suffix matching. With this change, patterns that have neither
prefixes nor suffixes can now be efficiently filtered.

According to SpecialCaseListBM:

Lookup benchmarks (significant improvements):

OVERALL_GEOMEAN                       -0.7809

Lookup *test* like benchmarks (huge improvements):

OVERALL_GEOMEAN                       -0.9947

https://gist.github.com/vitalybuka/ee7f681b448eb18974386ab35e2d4d27

[llvmbot]

@llvm/pr-subscribers-llvm-support

Author: Vitaly Buka (vitalybuka)

Changes

This commit adds a new RadixTree to SpecialCaseList for handling
substring matches. Previously, SpecialCaseList only supported prefix
and suffix matching. With this change, patterns that have neither
prefixes nor suffixes can now be efficiently filtered.

---

Full diff: https://github.com/llvm/llvm-project/pull/164545.diff

2 Files Affected:

• (modified) llvm/include/llvm/Support/SpecialCaseList.h (+4)
• (modified) llvm/lib/Support/SpecialCaseList.cpp (+28)

diff --git a/llvm/include/llvm/Support/SpecialCaseList.h b/llvm/include/llvm/Support/SpecialCaseList.h
index c077f8857c9c8..f66cd6fe733a7 100644
--- a/llvm/include/llvm/Support/SpecialCaseList.h
+++ b/llvm/include/llvm/Support/SpecialCaseList.h
@@ -170,6 +170,10 @@ class SpecialCaseList {
               RadixTree<iterator_range<StringRef::const_iterator>,
                         SmallVector<const GlobMatcher::Glob *, 1>>>
         SuffixPrefixToGlob;
+
+    RadixTree<iterator_range<StringRef::const_iterator>,
+              SmallVector<const GlobMatcher::Glob *, 1>>
+        SubstrToGlob;
   };
 
   /// Represents a set of patterns and their line numbers
diff --git a/llvm/lib/Support/SpecialCaseList.cpp b/llvm/lib/Support/SpecialCaseList.cpp
index 15367afd91e72..37fd5bfad750d 100644
--- a/llvm/lib/Support/SpecialCaseList.cpp
+++ b/llvm/lib/Support/SpecialCaseList.cpp
@@ -94,6 +94,19 @@ void SpecialCaseList::GlobMatcher::preprocess(bool BySize) {
     StringRef Prefix = G.Pattern.prefix();
     StringRef Suffix = G.Pattern.suffix();
 
+    if (Suffix.empty() && Prefix.empty()) {
+      // If both prefix and suffix are empty put into special tree to search by
+      // substring in a middle.
+      StringRef Substr = G.Pattern.longest_substr();
+      if (!Substr.empty()) {
+        // But only if substring is not empty. Searching this tree is more
+        // expensive.
+        auto &V = SubstrToGlob.emplace(Substr).first->second;
+        V.emplace_back(&G);
+        continue;
+      }
+    }
+
     auto &PToGlob = SuffixPrefixToGlob.emplace(reverse(Suffix)).first->second;
     auto &V = PToGlob.emplace(Prefix).first->second;
     V.emplace_back(&G);
@@ -116,6 +129,21 @@ void SpecialCaseList::GlobMatcher::match(
       }
     }
   }
+
+  if (!SubstrToGlob.empty()) {
+    // As we don't know when substring exactly starts, we will try all
+    // possibilities. In most cases search will fail on first characters.
+    for (StringRef Q = Query; !Q.empty(); Q = Q.drop_front()) {
+      for (const auto &[_, V] : SubstrToGlob.find_prefixes(Q)) {
+        for (const auto *G : reverse(V)) {
+          if (G->Pattern.match(Query)) {
+            Cb(G->Name, G->LineNo);
+            break;
+          }
+        }
+      }
+    }
+  }
 }
 
 SpecialCaseList::Matcher::Matcher(bool UseGlobs, bool RemoveDotSlash)

[fmayer]

nit: "Use RadixTree" in commit message?