small changes

Author: ieviev

src/libraries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/Symbolic/MatchingState.cs

@@ -18,12 +18,7 @@ internal MatchingState(SymbolicRegexNode<TSet> node, uint prevCharKind)
         }
 
         /// <summary>
-        /// TODO: The CLR assigns an entire field for this byte which is a waste,
-        /// and the much more preferred way to use this is in _nullabilityArray in the matcher
-        /// but the current design relies on interfaces/flags and
-        /// using the MatchingState directly so this byte is a quick solution to cheapen
-        /// it there by ~30% as well without having to breaking it all to pieces.
-        /// removing IsNullableFor in other places requires major redesign
+        /// TODO: This is only used to speed up the existing architecture, ideally should be removed along with IsNullableFor
         /// </summary>
         internal readonly int NullabilityInfo;
 

src/libraries/System.Text.RegularExpressions/src/System/Text/RegularExpressions/Symbolic/SymbolicRegexMatcher.cs

@@ -81,7 +81,7 @@ internal sealed partial class SymbolicRegexMatcher<TSet> : SymbolicRegexMatcher
         /// <summary>Data and routines for skipping ahead to the next place a match could potentially start.</summary>
         private readonly RegexFindOptimizations? _findOpts;
 
-        /// <summary>Dead end state to quickly return NoMatch</summary>
+        /// <summary>Dead end state to quickly return NoMatch, this could potentially be a constant</summary>
         private readonly int _deadStateId;
 
         /// <summary>Whether the pattern contains any anchor</summary>
@@ -663,6 +663,7 @@ private bool FindEndPositionDeltasDFANoSkipAscii(ReadOnlySpan<char> input, int l
         /// TODO: this is essentially a stripped down version when there's no good prefix optimizations
         /// i don't trust the compiler to optimize this and it makes a
         /// ~50% difference in performance with removing unnecessary checks alone
+        ///
         /// </summary>
         private bool FindEndPositionDeltasDFANoSkip(ReadOnlySpan<char> input, int lengthMinus1, RegexRunnerMode mode,
                 ref int posRef, int startStateId, ref int endPosRef, ref int endStateIdRef, ref int initialStatePosRef, ref int initialStatePosCandidateRef)
@@ -674,12 +675,19 @@ private bool FindEndPositionDeltasDFANoSkip(ReadOnlySpan<char> input, int length
             byte[] mtlookup = _mintermClassifier.ByteLookup()!;
             int endStateId = endStateIdRef;
             int currStateId = startStateId;
+            // ldfld only once
+            int deadStateId = _deadStateId;
             try
             {
                 // Loop through each character in the input, transitioning from state to state for each.
+                // The goal is to make this loop as fast as it can possible be,
+                // every single piece of overhead should be removed here
+                // there should be not a single callvirt instruction in the loop
+                // ldfld only if necessary (e.g. a reference changes)
+                // no memory writes unless necessary
                 while (true)
                 {
-                    if (currStateId == _deadStateId)
+                    if (currStateId == deadStateId)
                     {
                         return true;
                     }
@@ -709,9 +717,7 @@ private bool FindEndPositionDeltasDFANoSkip(ReadOnlySpan<char> input, int length
                         }
                         // one off check for the final position
                         // this is just to move it out of the hot loop
-                        if ((!_stateFlagsArray[currStateId].IsNullable() &&
-                             !_stateArray[currStateId]!.IsNullableFor(
-                                 GetPositionKind(-1))))
+                        if ((!_stateFlagsArray[currStateId].IsNullable() && IsNullableWithContext(currStateId, 0)))
                         {
                             return false;
                         }