[IndVarSimply] Fix assert/release build difference.

[Meinersbur]

Summary

In builds with assertions enabled (!NDEBUG), IndVarSimplify has an additional query to ScalarEvolution which may change future SCEV queries since it fills the internal cache differently. The result is actually only used when -verify-indvars. We fix the issue by only calling SE->getBackedgeTakenCount(L) if -verify-indvars is enabled such that only -verify-indvars shows the behavior, but not debug builds themselves. Also add a remark to the description of -verify-indvars about this behavior.

Fixes llvm.org/PR44815

[Meinersbur]

• Add test case

[jeroen.dobbelaere]

looks good to me, except for one minor issue in the testcase.

llvm/test/Transforms/IndVarSimplify/deterministic-scev-verify.ll

15

Shouldn't this be 'IR-LABEL:' ?

[Meinersbur]

• Fix mislabeling in test case

[reames]

Also LGTM

llvm/lib/Transforms/Scalar/IndVarSimplify.cpp

 STATISTIC(NumLFTR        , "Number of loop exit tests replaced");
 STATISTIC(NumElimExt     , "Number of IV sign/zero extends eliminated");
 STATISTIC(NumElimIV      , "Number of congruent IVs eliminated");
 
 // Trip count verification can be enabled by default under NDEBUG if we
 // implement a strong expression equivalence checker in SCEV. Until then, we
 // use the verify-indvars flag, which may assert in some cases.
 static cl::opt<bool> VerifyIndvars(
     "verify-indvars", cl::Hidden,
-  cl::desc("Verify the ScalarEvolution result after running indvars"));
+    cl::desc("Verify the ScalarEvolution result after running indvars. Has no "
+             "effect in release builds. (Note: this adds additional SCEV "
+             "queries potentially changing the analysis result)"));
 
 static cl::opt<ReplaceExitVal> ReplaceExitValue(
     "replexitval", cl::Hidden, cl::init(OnlyCheapRepl),
     cl::desc("Choose the strategy to replace exit value in IndVarSimplify"),
     cl::values(clEnumValN(NeverRepl, "never", "never replace exit value"),
                clEnumValN(OnlyCheapRepl, "cheap",
                           "only replace exit value when the cost is cheap"),
                clEnumValN(NoHardUse, "noharduse",
bool IndVarSimplify::run(Loop *L) {
   //    and we're in trouble if we can't find the induction variable even when
   //    we've manually inserted one.
   //  - LFTR relies on having a single backedge.
   if (!L->isLoopSimplifyForm())
     return false;
 
 #ifndef NDEBUG
   // Used below for a consistency check only
-  const SCEV *BackedgeTakenCount = SE->getBackedgeTakenCount(L);
+  // Note: Since the result returned by ScalarEvolution may depend on the order
+  // in which previous results are added to its cache, the call to
+  // getBackedgeTakenCount() may change following SCEV queries.
+  const SCEV *BackedgeTakenCount;
+  if (VerifyIndvars)
+    BackedgeTakenCount = SE->getBackedgeTakenCount(L);
 #endif
 
   bool Changed = false;
   // If there are any floating-point recurrences, attempt to
   // transform them to use integer recurrences.
   Changed |= rewriteNonIntegerIVs(L);
 
   // Create a rewriter object which we'll use to transform the code with.

llvm/test/Transforms/IndVarSimplify/deterministic-scev-verify.ll

+; RUN: opt -indvars -stats -disable-output < %s 2>&1 | FileCheck %s --check-prefix=STATS
+; RUN: opt -indvars -S < %s | FileCheck %s --check-prefix=IR
+; REQUIRES: asserts
+
+; Check that IndVarSimplify's result is not influenced by stray calls to
+; ScalarEvolution in debug builds. However, -verify-indvars may still do
+; such calls.
+; llvm.org/PR44815
+
+; STATS: 1 scalar-evolution - Number of loops with trip counts computed by force
+; STATS: 2 scalar-evolution - Number of loops with predictable loop counts
+
+; In this test, adding -verify-indvars causes %tmp13 to not be optimized away.
+; IR-LABEL: @foo
+; IR-NOT:   phi i32
+
+target triple = "x86_64-unknown-linux-gnu"
+
+@b = external dso_local local_unnamed_addr global i32
+
+define dso_local void @foo() {
+tmp0:
+  br label %tmp12
+
+tmp7:
+  %tmp8 = add nuw nsw i32 %tmp13, 1
+  store i32 undef, i32* @b
+  br label %tmp12
+
+tmp12:
+  %tmp13 = phi i32 [ 2, %tmp0 ], [ %tmp8, %tmp7 ]
+  %tmp14 = phi i32 [ 1, %tmp0 ], [ %tmp13, %tmp7 ]
+  %tmp15 = icmp ult i32 %tmp14, undef
+  br i1 %tmp15, label %tmp7, label %tmp16
+
+tmp16:
+  ret void
+}