[SCEV] Rewrite more SCEVAddExpr when applying guards.

fhahn · fhahn · commit ad4d8202ae9c · 2025-10-09T20:19:43.000+01:00
When re-writing SCEVAddExprs to apply information from guards, check if we have information for the expression itself. If so, apply it. When we have an expression of the form (Const + A), check if we have have guard info for (Const + 1 + A) and use it. This is needed to avoid regressions in a few cases, where we have BTCs with a subtracted constant. Rewriting expressions could cause regressions, e.g. when comparing 2 SCEV expressions where we are only able to rewrite one side, but I could not find any cases where this happens more with this patch in practice. Depends on llvm#160012 (included in PR) Proofs for some of the test changes: https://alive2.llvm.org/ce/z/RPX6t_
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -16080,16 +16080,32 @@ const SCEV *ScalarEvolution::LoopGuards::rewrite(const SCEV *Expr) const {
     }
 
     const SCEV *visitAddExpr(const SCEVAddExpr *Expr) {
-      // Trip count expressions sometimes consist of adding 3 operands, i.e.
-      // (Const + A + B). There may be guard info for A + B, and if so, apply
-      // it.
-      // TODO: Could more generally apply guards to Add sub-expressions.
-      if (isa<SCEVConstant>(Expr->getOperand(0)) &&
-          Expr->getNumOperands() == 3) {
-        if (const SCEV *S = Map.lookup(
-                SE.getAddExpr(Expr->getOperand(1), Expr->getOperand(2))))
-          return SE.getAddExpr(Expr->getOperand(0), S);
+      if (const SCEV *S = Map.lookup(Expr))
+        return S;
+      if (isa<SCEVConstant>(Expr->getOperand(0))) {
+        // Trip count expressions sometimes consist of adding 3 operands, i.e.
+        // (Const + A + B). There may be guard info for A + B, and if so, apply
+        // it.
+        // TODO: Could more generally apply guards to Add sub-expressions.
+        if (Expr->getNumOperands() == 3) {
+          if (const SCEV *S = Map.lookup(
+                  SE.getAddExpr(Expr->getOperand(1), Expr->getOperand(2))))
+            return SE.getAddExpr(Expr->getOperand(0), S);
+        }
+
+        // For expressions of the form (Const + A), check if we have guard info
+        // for (Const + 1 + A), and rewrite to ((Const + 1 + A) - 1). This makes
+        // sure we don't loose information when rewriting expressions based on
+        // back-edge taken counts in some cases..
+        if (Expr->getNumOperands() == 2) {
+          auto *NewC =
+              SE.getAddExpr(Expr->getOperand(0), SE.getOne(Expr->getType()));
+          if (const SCEV *S =
+                  Map.lookup(SE.getAddExpr(NewC, Expr->getOperand(1))))
+            return SE.getMinusSCEV(S, SE.getOne(Expr->getType()));
+        }
       }
+
       SmallVector<const SCEV *, 2> Operands;
       bool Changed = false;
       for (const auto *Op : Expr->operands()) {
diff --git a/llvm/test/Analysis/ScalarEvolution/backedge-taken-count-guard-info-apply-to-adds.ll b/llvm/test/Analysis/ScalarEvolution/backedge-taken-count-guard-info-apply-to-adds.ll
@@ -33,9 +33,9 @@ declare void @clobber()
 define void @test_add_sub_1_guard(ptr %src, i32 %n) {
 ; CHECK-LABEL: 'test_add_sub_1_guard'
 ; CHECK-NEXT:  Determining loop execution counts for: @test_add_sub_1_guard
-; CHECK-NEXT:  Loop %loop: backedge-taken count is (zext i32 (-1 + (%n /u 2))<nsw> to i64)
-; CHECK-NEXT:  Loop %loop: constant max backedge-taken count is i64 4294967295
-; CHECK-NEXT:  Loop %loop: symbolic max backedge-taken count is (zext i32 (-1 + (%n /u 2))<nsw> to i64)
+; CHECK-NEXT:  Loop %loop: backedge-taken count is i64 0
+; CHECK-NEXT:  Loop %loop: constant max backedge-taken count is i64 0
+; CHECK-NEXT:  Loop %loop: symbolic max backedge-taken count is i64 0
 ; CHECK-NEXT:  Loop %loop: Trip multiple is 1
 ;
 entry:
diff --git a/llvm/test/Analysis/ScalarEvolution/trip-count-minmax.ll b/llvm/test/Analysis/ScalarEvolution/trip-count-minmax.ll
@@ -102,12 +102,12 @@ define void @umax(i32 noundef %a, i32 noundef %b) {
 ; CHECK-NEXT:    %cond = select i1 %cmp, i32 %mul, i32 %mul1
 ; CHECK-NEXT:    --> ((2 * %a) umax (4 * %b)) U: [0,-1) S: [-2147483648,2147483647)
 ; CHECK-NEXT:    %i.011 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
-; CHECK-NEXT:    --> {0,+,1}<nuw><nsw><%for.body> U: [0,-2147483648) S: [0,-2147483648) Exits: (-1 + ((2 * %a) umax (4 * %b))) LoopDispositions: { %for.body: Computable }
+; CHECK-NEXT:    --> {0,+,1}<nuw><nsw><%for.body> U: [0,2147483647) S: [0,2147483647) Exits: (-1 + ((2 * %a) umax (4 * %b))) LoopDispositions: { %for.body: Computable }
 ; CHECK-NEXT:    %inc = add nuw nsw i32 %i.011, 1
-; CHECK-NEXT:    --> {1,+,1}<nuw><%for.body> U: [1,-1) S: [1,-1) Exits: ((2 * %a) umax (4 * %b)) LoopDispositions: { %for.body: Computable }
+; CHECK-NEXT:    --> {1,+,1}<nuw><nsw><%for.body> U: [1,-2147483648) S: [1,-2147483648) Exits: ((2 * %a) umax (4 * %b)) LoopDispositions: { %for.body: Computable }
 ; CHECK-NEXT:  Determining loop execution counts for: @umax
 ; CHECK-NEXT:  Loop %for.body: backedge-taken count is (-1 + ((2 * %a) umax (4 * %b)))
-; CHECK-NEXT:  Loop %for.body: constant max backedge-taken count is i32 -3
+; CHECK-NEXT:  Loop %for.body: constant max backedge-taken count is i32 2147483646
 ; CHECK-NEXT:  Loop %for.body: symbolic max backedge-taken count is (-1 + ((2 * %a) umax (4 * %b)))
 ; CHECK-NEXT:  Loop %for.body: Trip multiple is 2
 ;
@@ -197,12 +197,12 @@ define void @smax(i32 noundef %a, i32 noundef %b) {
 ; CHECK-NEXT:    %cond = select i1 %cmp, i32 %mul, i32 %mul1
 ; CHECK-NEXT:    --> ((2 * %a)<nsw> smax (4 * %b)<nsw>) U: [0,-1) S: [-2147483648,2147483647)
 ; CHECK-NEXT:    %i.011 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
-; CHECK-NEXT:    --> {0,+,1}<nuw><nsw><%for.body> U: [0,-2147483648) S: [0,-2147483648) Exits: (-1 + ((2 * %a)<nsw> smax (4 * %b)<nsw>)) LoopDispositions: { %for.body: Computable }
+; CHECK-NEXT:    --> {0,+,1}<nuw><nsw><%for.body> U: [0,2147483647) S: [0,2147483647) Exits: (-1 + ((2 * %a)<nsw> smax (4 * %b)<nsw>)) LoopDispositions: { %for.body: Computable }
 ; CHECK-NEXT:    %inc = add nuw nsw i32 %i.011, 1
-; CHECK-NEXT:    --> {1,+,1}<nuw><%for.body> U: [1,-1) S: [1,-1) Exits: ((2 * %a)<nsw> smax (4 * %b)<nsw>) LoopDispositions: { %for.body: Computable }
+; CHECK-NEXT:    --> {1,+,1}<nuw><nsw><%for.body> U: [1,-2147483648) S: [1,-2147483648) Exits: ((2 * %a)<nsw> smax (4 * %b)<nsw>) LoopDispositions: { %for.body: Computable }
 ; CHECK-NEXT:  Determining loop execution counts for: @smax
 ; CHECK-NEXT:  Loop %for.body: backedge-taken count is (-1 + ((2 * %a)<nsw> smax (4 * %b)<nsw>))
-; CHECK-NEXT:  Loop %for.body: constant max backedge-taken count is i32 -3
+; CHECK-NEXT:  Loop %for.body: constant max backedge-taken count is i32 2147483646
 ; CHECK-NEXT:  Loop %for.body: symbolic max backedge-taken count is (-1 + ((2 * %a)<nsw> smax (4 * %b)<nsw>))
 ; CHECK-NEXT:  Loop %for.body: Trip multiple is 2
 ;
diff --git a/llvm/test/Transforms/IndVarSimplify/canonicalize-cmp.ll b/llvm/test/Transforms/IndVarSimplify/canonicalize-cmp.ll
@@ -343,14 +343,13 @@ define void @slt_no_smax_needed(i64 %n, ptr %dst) {
 ; CHECK-NEXT:    [[PRE:%.*]] = icmp ult i32 [[ADD_1]], 8
 ; CHECK-NEXT:    br i1 [[PRE]], label [[EXIT:%.*]], label [[LOOP_PREHEADER:%.*]]
 ; CHECK:       loop.preheader:
-; CHECK-NEXT:    [[SMAX:%.*]] = call i32 @llvm.smax.i32(i32 [[SHR]], i32 1)
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[LOOP_PREHEADER]] ]
 ; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr [[DST:%.*]], i32 [[IV]]
 ; CHECK-NEXT:    store i8 0, ptr [[GEP]], align 1
 ; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
-; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[IV_NEXT]], [[SMAX]]
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[IV_NEXT]], [[SHR]]
 ; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]]
 ; CHECK:       exit.loopexit:
 ; CHECK-NEXT:    br label [[EXIT]]
@@ -385,14 +384,13 @@ define void @ult_no_umax_needed(i64 %n, ptr %dst) {
 ; CHECK-NEXT:    [[PRE:%.*]] = icmp ult i32 [[ADD_1]], 8
 ; CHECK-NEXT:    br i1 [[PRE]], label [[EXIT:%.*]], label [[LOOP_PREHEADER:%.*]]
 ; CHECK:       loop.preheader:
-; CHECK-NEXT:    [[UMAX:%.*]] = call i32 @llvm.umax.i32(i32 [[SHR]], i32 1)
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
 ; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[LOOP_PREHEADER]] ]
 ; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds i8, ptr [[DST:%.*]], i32 [[IV]]
 ; CHECK-NEXT:    store i8 0, ptr [[GEP]], align 1
 ; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1
-; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[IV_NEXT]], [[UMAX]]
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[IV_NEXT]], [[SHR]]
 ; CHECK-NEXT:    br i1 [[EXITCOND]], label [[LOOP]], label [[EXIT_LOOPEXIT:%.*]]
 ; CHECK:       exit.loopexit:
 ; CHECK-NEXT:    br label [[EXIT]]
diff --git a/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll b/llvm/test/Transforms/LoopIdiom/add-nsw-zext-fold.ll
@@ -61,9 +61,9 @@ define void @test_memset_size_can_use_info_from_guards(i32 %x, ptr %dst) {
 ; CHECK:       [[LOOP1_BACKEDGE]]:
 ; CHECK-NEXT:    br label %[[LOOP1]]
 ; CHECK:       [[LOOP2_PREHEADER]]:
-; CHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[SUB]] to i64
-; CHECK-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 1
-; CHECK-NEXT:    [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[TMP1]], i64 1)
+; CHECK-NEXT:    [[TMP0:%.*]] = add nsw i32 [[SHR]], -1
+; CHECK-NEXT:    [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
+; CHECK-NEXT:    [[UMAX:%.*]] = add nuw nsw i64 [[TMP1]], 1
 ; CHECK-NEXT:    call void @llvm.memset.p0.i64(ptr align 1 [[DST]], i8 0, i64 [[UMAX]], i1 false)
 ; CHECK-NEXT:    br label %[[LOOP2:.*]]
 ; CHECK:       [[LOOP2]]:
diff --git a/llvm/test/Transforms/LoopUnroll/runtime-unroll-assume-no-remainder.ll b/llvm/test/Transforms/LoopUnroll/runtime-unroll-assume-no-remainder.ll
@@ -19,9 +19,16 @@ define dso_local void @assumeDivisibleTC(ptr noalias nocapture %a, ptr noalias n
 ; CHECK-NEXT:    [[CMP110:%.*]] = icmp sgt i32 [[N]], 0
 ; CHECK-NEXT:    br i1 [[CMP110]], label [[FOR_BODY_PREHEADER:%.*]], label [[EXIT]]
 ; CHECK:       for.body.preheader:
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[N]], -1
+; CHECK-NEXT:    [[XTRAITER:%.*]] = and i32 [[N]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 1
+; CHECK-NEXT:    br i1 [[TMP3]], label [[FOR_BODY_EPIL_PREHEADER:%.*]], label [[FOR_BODY_PREHEADER_NEW:%.*]]
+; CHECK:       for.body.preheader.new:
+; CHECK-NEXT:    [[UNROLL_ITER:%.*]] = sub i32 [[N]], [[XTRAITER]]
 ; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
 ; CHECK:       for.body:
-; CHECK-NEXT:    [[I_011:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INC_1:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[I_011:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INC_1:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[NITER:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NEXT_1:%.*]], [[FOR_BODY]] ]
 ; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i32 [[I_011]]
 ; CHECK-NEXT:    [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
 ; CHECK-NEXT:    [[ADD:%.*]] = add i8 [[TMP0]], 3
@@ -34,8 +41,25 @@ define dso_local void @assumeDivisibleTC(ptr noalias nocapture %a, ptr noalias n
 ; CHECK-NEXT:    [[ARRAYIDX4_1:%.*]] = getelementptr inbounds i8, ptr [[A]], i32 [[INC]]
 ; CHECK-NEXT:    store i8 [[ADD_1]], ptr [[ARRAYIDX4_1]], align 1
 ; CHECK-NEXT:    [[INC_1]] = add nuw nsw i32 [[I_011]], 2
-; CHECK-NEXT:    [[CMP1_1:%.*]] = icmp slt i32 [[INC_1]], [[N]]
-; CHECK-NEXT:    br i1 [[CMP1_1]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT:    [[NITER_NEXT_1]] = add i32 [[NITER]], 2
+; CHECK-NEXT:    [[NITER_NCMP_1:%.*]] = icmp ne i32 [[NITER_NEXT_1]], [[UNROLL_ITER]]
+; CHECK-NEXT:    br i1 [[NITER_NCMP_1]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT_UNR_LCSSA:%.*]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       exit.loopexit.unr-lcssa:
+; CHECK-NEXT:    [[I_011_UNR:%.*]] = phi i32 [ [[INC_1]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[LCMP_MOD:%.*]] = icmp ne i32 [[XTRAITER]], 0
+; CHECK-NEXT:    br i1 [[LCMP_MOD]], label [[FOR_BODY_EPIL_PREHEADER]], label [[EXIT_LOOPEXIT:%.*]]
+; CHECK:       for.body.epil.preheader:
+; CHECK-NEXT:    [[I_011_EPIL_INIT:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[I_011_UNR]], [[EXIT_LOOPEXIT_UNR_LCSSA]] ]
+; CHECK-NEXT:    [[LCMP_MOD1:%.*]] = icmp ne i32 [[XTRAITER]], 0
+; CHECK-NEXT:    call void @llvm.assume(i1 [[LCMP_MOD1]])
+; CHECK-NEXT:    br label [[FOR_BODY_EPIL:%.*]]
+; CHECK:       for.body.epil:
+; CHECK-NEXT:    [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds i8, ptr [[B]], i32 [[I_011_EPIL_INIT]]
+; CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr [[ARRAYIDX_EPIL]], align 1
+; CHECK-NEXT:    [[ADD_EPIL:%.*]] = add i8 [[TMP4]], 3
+; CHECK-NEXT:    [[ARRAYIDX4_EPIL:%.*]] = getelementptr inbounds i8, ptr [[A]], i32 [[I_011_EPIL_INIT]]
+; CHECK-NEXT:    store i8 [[ADD_EPIL]], ptr [[ARRAYIDX4_EPIL]], align 1
+; CHECK-NEXT:    br label [[EXIT_LOOPEXIT]]
 ; CHECK:       exit.loopexit:
 ; CHECK-NEXT:    br label [[EXIT]]
 ; CHECK:       exit:
diff --git a/llvm/test/Transforms/LoopVectorize/dont-fold-tail-for-divisible-TC.ll b/llvm/test/Transforms/LoopVectorize/dont-fold-tail-for-divisible-TC.ll
@@ -193,7 +193,7 @@ define dso_local void @cannotProveAlignedTC(ptr noalias nocapture %A, i32 %p, i3
 ; CHECK-NEXT:    store i32 13, ptr [[TMP12]], align 1
 ; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE6]]
 ; CHECK:       pred.store.continue6:
-; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 4)
 ; CHECK-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
 ; CHECK-NEXT:    br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-checks-difference.ll b/llvm/test/Transforms/LoopVectorize/runtime-checks-difference.ll
@@ -465,12 +465,7 @@ define void @remove_diff_checks_via_guards(i32 %x, i32 %y, ptr %A) {
 ; CHECK-NEXT:    [[TMP13:%.*]] = icmp ugt i64 [[SMAX]], 4294967295
 ; CHECK-NEXT:    [[TMP14:%.*]] = or i1 [[TMP12]], [[TMP13]]
 ; CHECK-NEXT:    [[TMP15:%.*]] = or i1 [[TMP9]], [[TMP14]]
-; CHECK-NEXT:    br i1 [[TMP15]], [[SCALAR_PH]], label %[[VECTOR_MEMCHECK:.*]]
-; CHECK:       [[VECTOR_MEMCHECK]]:
-; CHECK-NEXT:    [[TMP16:%.*]] = sext i32 [[OFFSET]] to i64
-; CHECK-NEXT:    [[TMP17:%.*]] = shl nsw i64 [[TMP16]], 2
-; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP17]], 16
-; CHECK-NEXT:    br i1 [[DIFF_CHECK]], [[SCALAR_PH]], [[VECTOR_PH1:label %.*]]
+; CHECK-NEXT:    br i1 [[TMP15]], [[SCALAR_PH]], [[VECTOR_PH:label %.*]]
 ;
 entry:
   %offset = sub i32 %x, %y
