[llvm] [InstCombine] Fold dependent IVs (PR #81151)
Nikita Popov via llvm-commits
llvm-commits at lists.llvm.org
Mon Feb 19 07:59:40 PST 2024
https://github.com/nikic updated https://github.com/llvm/llvm-project/pull/81151
>From 75daf2271b80f4e5eb9b6f39cfb8ae868f6c765d Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov at redhat.com>
Date: Thu, 8 Feb 2024 14:33:38 +0100
Subject: [PATCH] [InstCombine] Fold dependent IVs
Fold `iv = phi(start, iv.next = iv2.next + start)`
where `iv2 = phi(iv2.start, iv2.next = iv2 + iv2.step)`
to `iv = iv2 + start`
removing one induction variable from the loop.
Proof: https://alive2.llvm.org/ce/z/hfmwgf
---
.../Transforms/InstCombine/InstCombinePHI.cpp | 48 +++++++++++++++
.../Transforms/InstCombine/dependent-ivs.ll | 60 ++++++++-----------
2 files changed, 72 insertions(+), 36 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp b/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp
index 20b34c1379d575..694908072ace8d 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp
@@ -1378,6 +1378,51 @@ static Value *simplifyUsingControlFlow(InstCombiner &Self, PHINode &PN,
return nullptr;
}
+// Fold iv = phi(start, iv.next = iv2.next op start)
+// where iv2 = phi(iv2.start, iv2.next = iv2 + iv2.step)
+// and iv2.start op start = start
+// to iv = iv2 op start
+// Currently limited to op == add/gep and iv2.start == 0.
+static Value *foldDependentIVs(PHINode &PN, IRBuilderBase &Builder) {
+ BasicBlock *BB = PN.getParent();
+ if (PN.getNumIncomingValues() != 2)
+ return nullptr;
+
+ Value *Start;
+ Instruction *IvNext;
+ BinaryOperator *Iv2Next;
+ auto MatchOuterIV = [&](Value *V1, Value *V2) {
+ if (match(V2, m_c_Add(m_Specific(V1), m_BinOp(Iv2Next))) ||
+ match(V2, m_GEP(m_Specific(V1), m_BinOp(Iv2Next)))) {
+ Start = V1;
+ IvNext = cast<Instruction>(V2);
+ return true;
+ }
+ return false;
+ };
+
+ if (!MatchOuterIV(PN.getIncomingValue(0), PN.getIncomingValue(1)) &&
+ !MatchOuterIV(PN.getIncomingValue(1), PN.getIncomingValue(0)))
+ return nullptr;
+
+ PHINode *Iv2;
+ Value *Iv2Start, *Iv2Step;
+ if (!matchSimpleRecurrence(Iv2Next, Iv2, Iv2Start, Iv2Step) ||
+ Iv2->getParent() != BB || !match(Iv2Start, m_Zero()))
+ return nullptr;
+
+ Builder.SetInsertPoint(&*BB, BB->getFirstInsertionPt());
+ if (Start->getType()->isPtrOrPtrVectorTy()) {
+ auto *GEP = cast<GEPOperator>(IvNext);
+ return Builder.CreateGEP(GEP->getSourceElementType(), Start, Iv2, "",
+ cast<GEPOperator>(IvNext)->isInBounds());
+ }
+
+ auto *OBO = cast<OverflowingBinaryOperator>(IvNext);
+ return Builder.CreateAdd(Start, Iv2, "", OBO->hasNoUnsignedWrap(),
+ OBO->hasNoSignedWrap());
+}
+
// PHINode simplification
//
Instruction *InstCombinerImpl::visitPHINode(PHINode &PN) {
@@ -1595,5 +1640,8 @@ Instruction *InstCombinerImpl::visitPHINode(PHINode &PN) {
if (auto *V = simplifyUsingControlFlow(*this, PN, DT))
return replaceInstUsesWith(PN, V);
+ if (Value *Res = foldDependentIVs(PN, Builder))
+ return replaceInstUsesWith(PN, Res);
+
return nullptr;
}
diff --git a/llvm/test/Transforms/InstCombine/dependent-ivs.ll b/llvm/test/Transforms/InstCombine/dependent-ivs.ll
index 0ee11a13592557..73dd451dc2a016 100644
--- a/llvm/test/Transforms/InstCombine/dependent-ivs.ll
+++ b/llvm/test/Transforms/InstCombine/dependent-ivs.ll
@@ -7,11 +7,10 @@ define void @int_iv_nuw(i64 %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV2:%.*]] = phi i64 [ [[IV2_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV2:%.*]] = add nuw i64 [[IV]], [[BASE]]
; CHECK-NEXT: call void @use.i64(i64 [[IV2]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV2_NEXT]] = add nuw i64 [[IV_NEXT]], [[BASE]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -39,11 +38,10 @@ define void @int_iv_nsw(i64 %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV2:%.*]] = phi i64 [ [[IV2_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV2:%.*]] = add nsw i64 [[IV]], [[BASE]]
; CHECK-NEXT: call void @use.i64(i64 [[IV2]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV2_NEXT]] = add nsw i64 [[IV_NEXT]], [[BASE]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -72,11 +70,10 @@ define void @int_iv_commuted_add(i64 %base, i64 %end) {
; CHECK-NEXT: [[BASE2:%.*]] = mul i64 [[BASE]], 42
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV2:%.*]] = phi i64 [ [[IV2_NEXT:%.*]], [[LOOP]] ], [ [[BASE2]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV2:%.*]] = add i64 [[BASE2]], [[IV]]
; CHECK-NEXT: call void @use.i64(i64 [[IV2]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV2_NEXT]] = add i64 [[BASE2]], [[IV_NEXT]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -105,11 +102,10 @@ define void @int_iv_commuted_phi1(i64 %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV2:%.*]] = phi i64 [ [[BASE]], [[ENTRY:%.*]] ], [ [[IV2_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[IV2:%.*]] = add i64 [[IV]], [[BASE]]
; CHECK-NEXT: call void @use.i64(i64 [[IV2]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV2_NEXT]] = add i64 [[IV_NEXT]], [[BASE]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -137,11 +133,10 @@ define void @int_iv_commuted_phi2(i64 %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV2:%.*]] = phi i64 [ [[IV2_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV2:%.*]] = add i64 [[IV]], [[BASE]]
; CHECK-NEXT: call void @use.i64(i64 [[IV2]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV2_NEXT]] = add i64 [[IV_NEXT]], [[BASE]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -169,11 +164,10 @@ define void @int_iv_vector(<2 x i64> %base) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV2:%.*]] = phi <2 x i64> [ [[IV2_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi <2 x i64> [ [[IV_NEXT:%.*]], [[LOOP]] ], [ zeroinitializer, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi <2 x i64> [ [[IV_NEXT:%.*]], [[LOOP]] ], [ zeroinitializer, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV2:%.*]] = add <2 x i64> [[IV]], [[BASE]]
; CHECK-NEXT: call void @use.v2i64(<2 x i64> [[IV2]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw <2 x i64> [[IV]], <i64 4, i64 4>
-; CHECK-NEXT: [[IV2_NEXT]] = add <2 x i64> [[IV_NEXT]], [[BASE]]
; CHECK-NEXT: [[CMP:%.*]] = call i1 @get.i1()
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -201,12 +195,11 @@ define void @int_iv_loop_variant_step(i64 %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV2:%.*]] = phi i64 [ [[IV2_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV2:%.*]] = add nuw i64 [[IV]], [[BASE]]
; CHECK-NEXT: call void @use.i64(i64 [[IV2]])
; CHECK-NEXT: [[STEP:%.*]] = call i64 @get.i64()
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], [[STEP]]
-; CHECK-NEXT: [[IV2_NEXT]] = add nuw i64 [[IV_NEXT]], [[BASE]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -235,11 +228,10 @@ define void @ptr_iv_inbounds(ptr %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV_PTR:%.*]] = phi ptr [ [[IV_PTR_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV_PTR:%.*]] = getelementptr inbounds i8, ptr [[BASE]], i64 [[IV]]
; CHECK-NEXT: call void @use.p0(ptr [[IV_PTR]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV_PTR_NEXT]] = getelementptr inbounds i8, ptr [[BASE]], i64 [[IV_NEXT]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -267,11 +259,10 @@ define void @ptr_iv_no_inbounds(ptr %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV_PTR:%.*]] = phi ptr [ [[IV_PTR_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV_PTR:%.*]] = getelementptr i8, ptr [[BASE]], i64 [[IV]]
; CHECK-NEXT: call void @use.p0(ptr [[IV_PTR]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV_PTR_NEXT]] = getelementptr i8, ptr [[BASE]], i64 [[IV_NEXT]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -299,11 +290,10 @@ define void @ptr_iv_non_i8_type(ptr %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV_PTR:%.*]] = phi ptr [ [[IV_PTR_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV_PTR:%.*]] = getelementptr i32, ptr [[BASE]], i64 [[IV]]
; CHECK-NEXT: call void @use.p0(ptr [[IV_PTR]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV_PTR_NEXT]] = getelementptr i32, ptr [[BASE]], i64 [[IV_NEXT]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -331,11 +321,10 @@ define void @ptr_iv_vector(<2 x ptr> %base, i64 %end) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV_PTR:%.*]] = phi <2 x ptr> [ [[IV_PTR_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV_PTR:%.*]] = getelementptr inbounds i8, <2 x ptr> [[BASE]], i64 [[IV]]
; CHECK-NEXT: call void @use.v2p0(<2 x ptr> [[IV_PTR]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 4
-; CHECK-NEXT: [[IV_PTR_NEXT]] = getelementptr inbounds i8, <2 x ptr> [[BASE]], i64 [[IV_NEXT]]
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[END]]
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
@@ -363,11 +352,10 @@ define void @ptr_iv_vector2(<2 x ptr> %base) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV_PTR:%.*]] = phi <2 x ptr> [ [[IV_PTR_NEXT:%.*]], [[LOOP]] ], [ [[BASE]], [[ENTRY:%.*]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi <2 x i64> [ [[IV_NEXT:%.*]], [[LOOP]] ], [ zeroinitializer, [[ENTRY]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi <2 x i64> [ [[IV_NEXT:%.*]], [[LOOP]] ], [ zeroinitializer, [[ENTRY:%.*]] ]
+; CHECK-NEXT: [[IV_PTR:%.*]] = getelementptr i8, <2 x ptr> [[BASE]], <2 x i64> [[IV]]
; CHECK-NEXT: call void @use.v2p0(<2 x ptr> [[IV_PTR]])
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw <2 x i64> [[IV]], <i64 4, i64 4>
-; CHECK-NEXT: [[IV_PTR_NEXT]] = getelementptr i8, <2 x ptr> [[BASE]], <2 x i64> [[IV_NEXT]]
; CHECK-NEXT: [[CMP:%.*]] = call i1 @get.i1()
; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
; CHECK: exit:
More information about the llvm-commits
mailing list