Fix O(n*2) analysis in const-bit-op-tree (#6791)

Note this might miss some cases where a sub-tree within an And/Or/Xor
tree is optimizeable, but not the whole tree, but in practice this seems
to work better than the alternative of keeping a set of failed nodes and
bail early.

src/V3Const.cpp

@@ -1297,23 +1297,32 @@ class ConstVisitor final : public VNVisitor {
         if (nodep->widthMin() != 1) return false;
         if (!v3Global.opt.fConstBitOpTree()) return false;
 
-        string debugPrefix;
+        const int width = nodep->width();
+        AstNodeExpr* rootp = nodep;
+        unsigned externalOps = 0;
+        // Reach past a plain making AND
+        if (const AstAnd* const andp = VN_CAST(nodep, And)) {
+            if (isConst(andp->lhsp(), 1)) {
+                rootp = andp->rhsp();
+                externalOps = 1;
+            }
+        }
+
+        // Only optimize if rootp is in fact the root of a tree of identical
+        // operations, otherwise analysis on an eventually unoptimizable unablanced
+        // tree can go O(N^2) as we would re-analyze every time as we move up the
+        // tree, repeatedly re-discover the sub-tree is not optimizable.
+        if (rootp->type() == nodep->backp()->type()) return false;
+
+        std::string debugPrefix;
         if (debug() >= 9) {  // LCOV_EXCL_START
             static int s_c = 0;
-            debugPrefix = "-  matchBitOpTree[";
-            debugPrefix += cvtToStr(++s_c);
-            debugPrefix += "] ";
+            debugPrefix = "-  matchBitOpTree[" + std::to_string(++s_c) + "] ";
             nodep->dumpTree(debugPrefix + "INPUT: ");
         }  // LCOV_EXCL_STOP
 
-        AstNode* newp = nullptr;
-        const AstAnd* const andp = VN_CAST(nodep, And);
-        const int width = nodep->width();
-        if (andp && isConst(andp->lhsp(), 1)) {  // 1 & BitOpTree
-            newp = ConstBitOpTreeVisitor::simplify(andp->rhsp(), width, 1, m_statBitOpReduction);
-        } else {  // BitOpTree
-            newp = ConstBitOpTreeVisitor::simplify(nodep, width, 0, m_statBitOpReduction);
-        }
+        AstNodeExpr* const newp
+            = ConstBitOpTreeVisitor::simplify(rootp, width, externalOps, m_statBitOpReduction);
 
         if (newp) {
             nodep->replaceWithKeepDType(newp);

test_regress/t/t_opt_const.py

@@ -16,7 +16,7 @@ test.compile(verilator_flags2=["-Wno-UNOPTTHREADS", "-fno-dfg", "--stats", test.
 test.execute()
 
 if test.vlt:
-    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 44)
+    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 49)
     test.file_grep(test.stats, r'SplitVar, packed variables split automatically\s+(\d+)', 1)
 
 test.passes()

test_regress/t/t_opt_const_cov.py

@@ -16,6 +16,6 @@ test.compile(verilator_flags2=["-Wno-UNOPTTHREADS", "--stats", "--coverage", "--
 test.execute()
 
 if test.vlt:
-    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 478)
+    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 550)
 
 test.passes()

test_regress/t/t_opt_const_dfg.py

@@ -18,7 +18,7 @@ test.compile(verilator_flags2=["-Wno-UNOPTTHREADS", "--stats", test.pli_filename
 test.execute()
 
 if test.vlt:
-    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 37)
+    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 43)
     test.file_grep(test.stats, r'SplitVar, packed variables split automatically\s+(\d+)', 1)
 
 test.passes()

test_regress/t/t_opt_const_no_expand.py

@@ -19,6 +19,6 @@ test.compile(verilator_flags2=[
 test.execute()
 
 if test.vlt:
-    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 1)
+    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 4)
 
 test.passes()

test_regress/t/t_opt_const_red.py

@@ -16,6 +16,6 @@ test.compile(verilator_flags2=["-Wno-UNOPTTHREADS", "--stats"])
 test.execute()
 
 if test.vlt:
-    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 158)
+    test.file_grep(test.stats, r'Optimizations, Const bit op reduction\s+(\d+)', 160)
 
 test.passes()