[HashRecognize] Strip ValueEvolution (#148620)

The ValueEvolution logic is deeply flawed, and checking that zero-bits
are shifted can be exploited for miscompiles. In an effort to redo
HashRecognize with a pattern-matching based approach, extract and fix
the core logic of ValueEvolution, and strip it completely, showing that
none of the tests rely on the KnownBits computation of ValueEvolution.

Co-authored-by: Piotr Fusik <[email protected]>

Author: artagnon

llvm/include/llvm/Analysis/HashRecognize.h

@@ -20,18 +20,12 @@
 #include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/IR/Value.h"
-#include "llvm/Support/KnownBits.h"
 #include <variant>
 
 namespace llvm {
 
 class LPMUpdater;
 
-/// A tuple of bits that are expected to be zero, number N of them expected to
-/// be zero, with a boolean indicating whether it's the top or bottom N bits
-/// expected to be zero.
-using ErrBits = std::tuple<KnownBits, unsigned, bool>;
-
 /// A custom std::array with 256 entries, that also has a print function.
 struct CRCTable : public std::array<APInt, 256> {
   void print(raw_ostream &OS) const;
@@ -85,7 +79,7 @@ class HashRecognize {
   HashRecognize(const Loop &L, ScalarEvolution &SE);
 
   // The main analysis entry points.
-  std::variant<PolynomialInfo, ErrBits, StringRef> recognizeCRC() const;
+  std::variant<PolynomialInfo, StringRef> recognizeCRC() const;
   std::optional<PolynomialInfo> getResult() const;
 
   // Auxilary entry point after analysis to interleave the generating polynomial

llvm/include/llvm/IR/PatternMatch.h

@@ -2283,6 +2283,14 @@ m_ZExtOrSExtOrSelf(const OpTy &Op) {
   return m_CombineOr(m_ZExtOrSExt(Op), Op);
 }
 
+template <typename OpTy>
+inline match_combine_or<match_combine_or<CastInst_match<OpTy, ZExtInst>,
+                                         CastInst_match<OpTy, TruncInst>>,
+                        OpTy>
+m_ZExtOrTruncOrSelf(const OpTy &Op) {
+  return m_CombineOr(m_CombineOr(m_ZExt(Op), m_Trunc(Op)), Op);
+}
+
 template <typename OpTy>
 inline CastInst_match<OpTy, UIToFPInst> m_UIToFP(const OpTy &Op) {
   return CastInst_match<OpTy, UIToFPInst>(Op);