[llvm] 484c622 - [SCEV] Do not strengthen nuw/nsw flags during get[Zero,Sign]ExtendedExpr.

[Florian Hahn via llvm-commits]

Author: Florian Hahn
Date: 2023-03-15T20:57:34Z
New Revision: 484c622760e2a46aa1b52c7a791ee59543364225

URL: https://github.com/llvm/llvm-project/commit/484c622760e2a46aa1b52c7a791ee59543364225
DIFF: https://github.com/llvm/llvm-project/commit/484c622760e2a46aa1b52c7a791ee59543364225.diff

LOG: [SCEV] Do not strengthen nuw/nsw flags during get[Zero,Sign]ExtendedExpr.

Modifying AddRecs when constructing other expressions can lead to
surprising changes. It also seems like it is not really beneficial i
most cases.

At the moment, there's a single regression, but we still might be able
to improve the flags at AddRec construction.

Might help with the issue discussed in D143409.

Reviewed By: mkazantsev

Differential Revision: https://reviews.llvm.org/D144051

Added: 
    

Modified: 
    llvm/lib/Analysis/ScalarEvolution.cpp
    llvm/test/Transforms/IndVarSimplify/lftr-reuse.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 424ff1c26c68..cc9abcc7af96 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -1625,11 +1625,6 @@ const SCEV *ScalarEvolution::getZeroExtendExprImpl(const SCEV *Op, Type *Ty,
       unsigned BitWidth = getTypeSizeInBits(AR->getType());
       const Loop *L = AR->getLoop();
 
-      if (!AR->hasNoUnsignedWrap()) {
-        auto NewFlags = proveNoWrapViaConstantRanges(AR);
-        setNoWrapFlags(const_cast<SCEVAddRecExpr *>(AR), NewFlags);
-      }
-
       // If we have special knowledge that this addrec won't overflow,
       // we don't need to do any further analysis.
       if (AR->hasNoUnsignedWrap()) {
@@ -2009,11 +2004,6 @@ const SCEV *ScalarEvolution::getSignExtendExprImpl(const SCEV *Op, Type *Ty,
       unsigned BitWidth = getTypeSizeInBits(AR->getType());
       const Loop *L = AR->getLoop();
 
-      if (!AR->hasNoSignedWrap()) {
-        auto NewFlags = proveNoWrapViaConstantRanges(AR);
-        setNoWrapFlags(const_cast<SCEVAddRecExpr *>(AR), NewFlags);
-      }
-
       // If we have special knowledge that this addrec won't overflow,
       // we don't need to do any further analysis.
       if (AR->hasNoSignedWrap()) {

diff  --git a/llvm/test/Transforms/IndVarSimplify/lftr-reuse.ll b/llvm/test/Transforms/IndVarSimplify/lftr-reuse.ll
index 4a9f6a10a0f4..8aa698a4cb51 100644
--- a/llvm/test/Transforms/IndVarSimplify/lftr-reuse.ll
+++ b/llvm/test/Transforms/IndVarSimplify/lftr-reuse.ll
@@ -126,6 +126,8 @@ exit:
 
 ; Force SCEVExpander to look for an existing well-formed phi.
 ; Perform LFTR without generating extra preheader code.
+; TODO: Recover regression after not strengthening AddRec flags during
+; get[Sign,Zero]ExtendExpr for INDVARS_IV_NEXT.
 define void @guardedloop(ptr %matrix, ptr %vector,
 ;
 ; CHECK-LABEL: @guardedloop(
@@ -146,7 +148,7 @@ define void @guardedloop(ptr %matrix, ptr %vector,
 ; CHECK-NEXT:    [[VECTORP:%.*]] = getelementptr inbounds [0 x double], ptr [[VECTOR:%.*]], i32 0, i64 [[INDVARS_IV2]]
 ; CHECK-NEXT:    [[V2:%.*]] = load double, ptr [[VECTORP]], align 8
 ; CHECK-NEXT:    call void @use(double [[V2]])
-; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], [[TMP0]]
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], [[TMP0]]
 ; CHECK-NEXT:    [[INDVARS_IV_NEXT3]] = add nuw nsw i64 [[INDVARS_IV2]], 1
 ; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT3]], [[WIDE_TRIP_COUNT]]
 ; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[RETURN_LOOPEXIT:%.*]]