[indvars] Prove non-negative for widening IVs in count down loops #71214
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
If we have a count down loop with an unsigned backedge test, SCEV is currently failing to prove that the IV is non-negative. The observation we can use is that if the increment of a loop would have been a NUW subtract (i.e. not cross zero), then an IV which starts with a positive value must be positive on all iterations. We can re-infer the sub nuw fact from an add with negative offset by checking to see if the loop backedge is guarded by a condition which implies IV >=u abs(step). I took several attempts at fixing this in SCEV itself, but decided for the moment, having indvars do the expensive proof is a better tradeoff. Once this is in, I plan to investigate SCEV further. I think we probably can sink this into SCEV, I just haven't quite found the right place yet.
|
@llvm/pr-subscribers-llvm-transforms Author: Philip Reames (preames) ChangesIf we have a count down loop with an unsigned backedge test, SCEV is currently failing to prove that the IV is non-negative. The observation we can use is that if the increment of a loop would have been a NUW subtract (i.e. not cross zero), then an IV which starts with a positive value must be positive on all iterations. We can re-infer the sub nuw fact from an add with negative offset by checking to see if the loop backedge is guarded by a condition which implies IV >=u abs(step). I took several attempts at fixing this in SCEV itself, but decided for the moment, having indvars do the expensive proof is a better tradeoff. Once this is in, I plan to investigate SCEV further. I think we probably can sink this into SCEV, I just haven't quite found the right place yet. Full diff: https://github.com/llvm/llvm-project/pull/71214.diff 3 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp b/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
index a618d72b406b397..dd8bef69ad1707f 100644
--- a/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
@@ -1122,6 +1122,7 @@ class WidenIV {
bool widenLoopCompare(NarrowIVDefUse DU);
bool widenWithVariantUse(NarrowIVDefUse DU);
+ bool isKnownNonNegative(const SCEV *S);
void pushNarrowIVUsers(Instruction *NarrowDef, Instruction *WideDef);
private:
@@ -1885,12 +1886,45 @@ Instruction *WidenIV::widenIVUse(WidenIV::NarrowIVDefUse DU, SCEVExpander &Rewri
return WideUse;
}
+// A special version of isKnownNonNegative which additionally tries
+// to prove that an addrec with a negative step would be non-negative
+// because the start is non-negative, and the increment would have
+// "nuw" if using a sub-instruction.
+// TODO: All of this should be sunk into SCEV once we figure out how to
+// reasonable do so without exploding compile time.
+bool WidenIV::isKnownNonNegative(const SCEV *S) {
+ const SCEV *Zero = SE->getZero(S->getType());
+ if (SE->isKnownPredicate(ICmpInst::ICMP_SGE, S, Zero))
+ return true;
+ auto *AR = dyn_cast<SCEVAddRecExpr>(S);
+ if (!AR || !AR->isAffine())
+ return false;
+
+ const SCEV *Start = AR->getStart();
+ const SCEV *Step = AR->getStepRecurrence(*SE);
+ const SCEV *PostInc = AR->getPostIncExpr(*SE);
+ // For a negative step, we can prove the result non-negative if the addrec
+ // only traverses values in the range zext([0,UINT_MAX]).
+ // TODO: Consider extending this for unknown steps? Would need to
+ // handle the positive step bound
+ if (!SE->isKnownNegative(Step))
+ return false;
+ if (!SE->isLoopEntryGuardedByCond(L, ICmpInst::ICMP_SGE, Start, Zero))
+ return false;
+
+ // Check for the unsigned form of these comparisons. The signed form
+ // should have been handled recursively in the query above.
+ uint32_t BitWidth = cast<IntegerType>(AR->getType())->getBitWidth();
+ const SCEV *N = SE->getConstant(APInt::getMaxValue(BitWidth) -
+ SE->getSignedRangeMin(Step));
+ return SE->isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT, AR, N) ||
+ SE->isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT, PostInc, N);
+}
+
/// Add eligible users of NarrowDef to NarrowIVUsers.
void WidenIV::pushNarrowIVUsers(Instruction *NarrowDef, Instruction *WideDef) {
- const SCEV *NarrowSCEV = SE->getSCEV(NarrowDef);
- bool NonNegativeDef =
- SE->isKnownPredicate(ICmpInst::ICMP_SGE, NarrowSCEV,
- SE->getZero(NarrowSCEV->getType()));
+ assert(L->contains(NarrowDef));
+ const bool NonNegativeDef = isKnownNonNegative(SE->getSCEV(NarrowDef));
for (User *U : NarrowDef->users()) {
Instruction *NarrowUser = cast<Instruction>(U);
diff --git a/llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll b/llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll
index 6f659a88da2e2bf..b743e5480c7ee83 100644
--- a/llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll
+++ b/llvm/test/Transforms/IndVarSimplify/AArch64/widen-loop-comp.ll
@@ -507,15 +507,15 @@ declare void @consume.i1(i1)
define i32 @test10(i32 %v) {
; CHECK-LABEL: @test10(
; CHECK-NEXT: entry:
-; CHECK-NEXT: [[SEXT:%.*]] = sext i32 [[V:%.*]] to i64
+; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[V:%.*]] to i64
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
-; CHECK-NEXT: [[TMP0:%.*]] = mul nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[TMP0]], [[SEXT]]
-; CHECK-NEXT: call void @consume.i1(i1 [[TMP1]])
-; CHECK-NEXT: call void @consume.i64(i64 [[TMP0]])
+; CHECK-NEXT: [[TMP1:%.*]] = mul nsw i64 [[INDVARS_IV]], -1
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP0]]
+; CHECK-NEXT: call void @consume.i1(i1 [[CMP]])
+; CHECK-NEXT: call void @consume.i64(i64 [[TMP1]])
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], 11
; CHECK-NEXT: br i1 [[EXITCOND]], label [[LOOP]], label [[LEAVE:%.*]]
; CHECK: leave:
diff --git a/llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll b/llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll
index b0392cc6fe2c1ce..9591944fef3866d 100644
--- a/llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll
+++ b/llvm/test/Transforms/IndVarSimplify/widen-nonnegative-countdown.ll
@@ -82,12 +82,10 @@ define void @zext_postinc(ptr %A, i32 %start) {
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
-; CHECK-NEXT: [[J_016_US:%.*]] = phi i32 [ [[INC_US:%.*]], [[FOR_BODY]] ], [ [[START]], [[FOR_BODY_PREHEADER]] ]
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
-; CHECK-NEXT: [[INC_US]] = add nsw i32 [[J_016_US]], -1
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[INC_US]], 6
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV_NEXT]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -124,8 +122,8 @@ define void @zext_preinc(ptr %A, i32 %start) {
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -223,8 +221,7 @@ define void @sext_postinc(ptr %A, i32 %start) {
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[TMP1]], 6
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV_NEXT]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -262,8 +259,7 @@ define void @sext_preinc(ptr %A, i32 %start) {
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[TMP1]], 6
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -365,15 +361,13 @@ define void @zext_postinc_offset_constant_one(ptr %A, i32 %start) {
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
-; CHECK-NEXT: [[J_016_US:%.*]] = phi i32 [ [[INC_US:%.*]], [[FOR_BODY]] ], [ [[START]], [[FOR_BODY_PREHEADER]] ]
; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
; CHECK-NEXT: [[ADD_US:%.*]] = add i32 [[TMP1]], 1
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext i32 [[ADD_US]] to i64
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
-; CHECK-NEXT: [[INC_US]] = add nsw i32 [[J_016_US]], -1
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[INC_US]], 6
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV_NEXT]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -414,8 +408,8 @@ define void @zext_preinc_offset_constant_one(ptr %A, i32 %start) {
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext i32 [[ADD_US]] to i64
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -521,8 +515,7 @@ define void @sext_postinc_offset_constant_one(ptr %A, i32 %start) {
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[TMP2]], 6
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV_NEXT]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -564,8 +557,7 @@ define void @sext_preinc_offset_constant_one(ptr %A, i32 %start) {
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[INDVARS_IV]] to i32
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[TMP2]], 6
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -662,15 +654,18 @@ define void @zext_postinc_offset_constant_minus_one(ptr %A, i32 %start) {
; CHECK-NEXT: [[NONPOS:%.*]] = icmp slt i32 [[START:%.*]], 2
; CHECK-NEXT: br i1 [[NONPOS]], label [[EXIT:%.*]], label [[FOR_BODY_PREHEADER:%.*]]
; CHECK: for.body.preheader:
+; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[START]] to i64
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
+; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
; CHECK-NEXT: [[J_016_US:%.*]] = phi i32 [ [[INC_US:%.*]], [[FOR_BODY]] ], [ [[START]], [[FOR_BODY_PREHEADER]] ]
; CHECK-NEXT: [[ADD_US:%.*]] = add i32 [[J_016_US]], -1
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext i32 [[ADD_US]] to i64
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
+; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
; CHECK-NEXT: [[INC_US]] = add nsw i32 [[J_016_US]], -1
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[INC_US]], 6
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV_NEXT]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -711,9 +706,9 @@ define void @zext_preinc_offset_constant_minus_one(ptr %A, i32 %start) {
; CHECK-NEXT: [[IDXPROM_US:%.*]] = zext i32 [[ADD_US]] to i64
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
+; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
; CHECK-NEXT: [[INC_US]] = add nsw i32 [[J_016_US]], -1
; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
-; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -811,14 +806,14 @@ define void @sext_postinc_offset_constant_minus_one(ptr %A, i32 %start) {
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
-; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
-; CHECK-NEXT: [[ADD_US:%.*]] = add i32 [[TMP1]], -1
+; CHECK-NEXT: [[J_016_US:%.*]] = phi i32 [ [[INC_US:%.*]], [[FOR_BODY]] ], [ [[START]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[ADD_US:%.*]] = add i32 [[J_016_US]], -1
; CHECK-NEXT: [[IDXPROM_US:%.*]] = sext i32 [[ADD_US]] to i64
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[TMP2]], 6
+; CHECK-NEXT: [[INC_US]] = add nsw i32 [[J_016_US]], -1
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV_NEXT]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
@@ -854,14 +849,14 @@ define void @sext_preinc_offset_constant_minus_one(ptr %A, i32 %start) {
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.body:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
-; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
-; CHECK-NEXT: [[ADD_US:%.*]] = add i32 [[TMP1]], -1
+; CHECK-NEXT: [[J_016_US:%.*]] = phi i32 [ [[INC_US:%.*]], [[FOR_BODY]] ], [ [[START]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[ADD_US:%.*]] = add i32 [[J_016_US]], -1
; CHECK-NEXT: [[IDXPROM_US:%.*]] = sext i32 [[ADD_US]] to i64
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[IDXPROM_US]]
; CHECK-NEXT: tail call void @use_ptr(ptr [[ARRAYIDX_US]])
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], -1
-; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[INDVARS_IV]] to i32
-; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i32 [[TMP2]], 6
+; CHECK-NEXT: [[INC_US]] = add nsw i32 [[J_016_US]], -1
+; CHECK-NEXT: [[CMP2_US:%.*]] = icmp ugt i64 [[INDVARS_IV]], 6
; CHECK-NEXT: br i1 [[CMP2_US]], label [[FOR_BODY]], label [[EXIT_LOOPEXIT:%.*]]
; CHECK: exit.loopexit:
; CHECK-NEXT: br label [[EXIT]]
|
preames
commented
Nov 3, 2023
| @@ -811,14 +806,14 @@ define void @sext_postinc_offset_constant_minus_one(ptr %A, i32 %start) { | |||
| ; CHECK-NEXT: br label [[FOR_BODY:%.*]] | |||
| ; CHECK: for.body: | |||
| ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[TMP0]], [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ] | |||
| ; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32 | |||
| ; CHECK-NEXT: [[ADD_US:%.*]] = add i32 [[TMP1]], -1 | |||
| ; CHECK-NEXT: [[J_016_US:%.*]] = phi i32 [ [[INC_US:%.*]], [[FOR_BODY]] ], [ [[START]], [[FOR_BODY_PREHEADER]] ] | |||
There was a problem hiding this comment.
This regression should be resolved by #70967. I didn't make it a dependent review as all this one does is shift around where we have that problem. As you can see in some of the test changes above, it sometimes removes the fallback and sometimes adds it.
There was a problem hiding this comment.
The change above landed. I did realize we need one other change as well, to fully handle these cases: #71557
You can test this locally with the following command:git-clang-format --diff 49168b2512ef55e225e9b7cd0821daa5c8ae5a9b 7078504bdf3e69414d7d44332e794072c6509328 -- llvm/lib/Transforms/Utils/SimplifyIndVar.cppView the diff from clang-format here.diff --git a/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp b/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
index dd8bef69ad17..ce018f3eb11e 100644
--- a/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
+++ b/llvm/lib/Transforms/Utils/SimplifyIndVar.cpp
@@ -1918,7 +1918,7 @@ bool WidenIV::isKnownNonNegative(const SCEV *S) {
const SCEV *N = SE->getConstant(APInt::getMaxValue(BitWidth) -
SE->getSignedRangeMin(Step));
return SE->isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT, AR, N) ||
- SE->isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT, PostInc, N);
+ SE->isLoopBackedgeGuardedByCond(L, ICmpInst::ICMP_UGT, PostInc, N);
}
/// Add eligible users of NarrowDef to NarrowIVUsers.
|
|
I found another approach to these same cases over in #71557. My current thinking is that both are valuable, but that change is probably more obvious. I suggest reviewing that one first, and then I'll need to figure out some more extensive test coverage to justify the complexity of this one. |
|
No longer actively working on this. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
If we have a count down loop with an unsigned backedge test, SCEV is currently failing to prove that the IV is non-negative. The observation we can use is that if the increment of a loop would have been a NUW subtract (i.e. not cross zero), then an IV which starts with a positive value must be positive on all iterations. We can re-infer the sub nuw fact from an add with negative offset by checking to see if the loop backedge is guarded by a condition which implies IV >=u abs(step).
I took several attempts at fixing this in SCEV itself, but decided for the moment, having indvars do the expensive proof is a better tradeoff. Once this is in, I plan to investigate SCEV further. I think we probably can sink this into SCEV, I just haven't quite found the right place yet.