[llvm] [InstCombine] Add limit for expansion of gep chains (PR #147065)
Nikita Popov via llvm-commits
llvm-commits at lists.llvm.org
Mon Jul 21 02:48:39 PDT 2025
https://github.com/nikic updated https://github.com/llvm/llvm-project/pull/147065
>From 90ddc853afc19bc6d5eb3416e7a1ba0d244124d1 Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov at redhat.com>
Date: Fri, 4 Jul 2025 16:00:21 +0200
Subject: [PATCH 1/4] [InstCombine] Add limit for expansion of gep chains
When converting gep subtraction / comparison to offset
subtraction / comparison, avoid expanding very long multi-use
gep chains.
Another improvement we could make is to expand one-use followed
by multi-use gep differently, by rewriting the multi-use gep to
include the one-use offsets. But I think we want to have some
kind of complexity cut-off in any case.
---
.../InstCombine/InstCombineAddSub.cpp | 24 ++++-
.../InstCombine/InstCombineCompares.cpp | 2 +-
.../InstCombine/InstCombineInternal.h | 3 +
llvm/test/Transforms/InstCombine/icmp-gep.ll | 88 +++++++++++++++++++
llvm/test/Transforms/InstCombine/sub-gep.ll | 52 +++++++++++
5 files changed, 167 insertions(+), 2 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index 981c5271fb3f6..a2e7ca4004174 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -2146,13 +2146,35 @@ CommonPointerBase CommonPointerBase::compute(Value *LHS, Value *RHS) {
return Base;
}
+bool CommonPointerBase::isExpensive() const {
+ bool SeenConst = false;
+ unsigned NumGEPs = 0;
+ auto ProcessGEPs = [&SeenConst, &NumGEPs](ArrayRef<GEPOperator *> GEPs) {
+ bool SeenMultiUse = false;
+ for (GEPOperator *GEP : GEPs) {
+ // Only count GEPs after the first multi-use GEP. For the first one,
+ // we will directly reuse the offset.
+ if (SeenMultiUse) {
+ bool IsConst = GEP->hasAllConstantIndices();
+ SeenConst |= IsConst;
+ NumGEPs += !IsConst;
+ }
+ SeenMultiUse |= !GEP->hasOneUse();
+ }
+ };
+ ProcessGEPs(LHSGEPs);
+ ProcessGEPs(RHSGEPs);
+ NumGEPs += SeenConst;
+ return NumGEPs > 2;
+}
+
/// Optimize pointer differences into the same array into a size. Consider:
/// &A[10] - &A[0]: we should compile this to "10". LHS/RHS are the pointer
/// operands to the ptrtoint instructions for the LHS/RHS of the subtract.
Value *InstCombinerImpl::OptimizePointerDifference(Value *LHS, Value *RHS,
Type *Ty, bool IsNUW) {
CommonPointerBase Base = CommonPointerBase::compute(LHS, RHS);
- if (!Base.Ptr)
+ if (!Base.Ptr || Base.isExpensive())
return nullptr;
// To avoid duplicating the offset arithmetic, rewrite the GEP to use the
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index c90ff2a868d4c..c5e1b04002545 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -712,7 +712,7 @@ Instruction *InstCombinerImpl::foldGEPICmp(GEPOperator *GEPLHS, Value *RHS,
};
CommonPointerBase Base = CommonPointerBase::compute(GEPLHS, RHS);
- if (Base.Ptr == RHS && CanFold(Base.LHSNW)) {
+ if (Base.Ptr == RHS && CanFold(Base.LHSNW) && !Base.isExpensive()) {
// ((gep Ptr, OFFSET) cmp Ptr) ---> (OFFSET cmp 0).
Type *IdxTy = DL.getIndexType(GEPLHS->getType());
Value *Offset =
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineInternal.h b/llvm/lib/Transforms/InstCombine/InstCombineInternal.h
index f7fbf0815df03..c67e27e5b3e7c 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineInternal.h
+++ b/llvm/lib/Transforms/InstCombine/InstCombineInternal.h
@@ -910,6 +910,9 @@ struct CommonPointerBase {
GEPNoWrapFlags RHSNW = GEPNoWrapFlags::all();
static CommonPointerBase compute(Value *LHS, Value *RHS);
+
+ /// Whether expanding the GEP chains is expensive.
+ bool isExpensive() const;
};
} // end namespace llvm
diff --git a/llvm/test/Transforms/InstCombine/icmp-gep.ll b/llvm/test/Transforms/InstCombine/icmp-gep.ll
index 3f104056fb1f2..6c7572089532a 100644
--- a/llvm/test/Transforms/InstCombine/icmp-gep.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-gep.ll
@@ -849,3 +849,91 @@ define i1 @gep_mugtiple_ugt_inbounds_nusw(ptr %base, i64 %idx, i64 %idx2) {
%cmp = icmp ugt ptr %gep2, %base
ret i1 %cmp
}
+
+define i1 @gep_multiple_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3) {
+; CHECK-LABEL: @gep_multiple_multi_use_below_limit(
+; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl i64 [[IDX1:%.*]], 2
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[GEP1_IDX]]
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP2_IDX]]
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 [[GEP3_IDX]]
+; CHECK-NEXT: call void @use(ptr [[GEP3]])
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP1_IDX]], [[GEP2_IDX]]
+; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[GEP3_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr i32, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr i32, ptr %gep2, i64 %idx3
+ call void @use(ptr %gep3)
+ %cmp = icmp eq ptr %gep3, %base
+ ret i1 %cmp
+}
+
+define i1 @gep_multiple_multi_use_below_limit_extra_one_use_gep(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_multiple_multi_use_below_limit_extra_one_use_gep(
+; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl i64 [[IDX1:%.*]], 2
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[GEP1_IDX]]
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP2_IDX]]
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 [[GEP3_IDX]]
+; CHECK-NEXT: [[GEP4_IDX_NEG:%.*]] = mul i64 [[IDX4:%.*]], -4
+; CHECK-NEXT: call void @use(ptr [[GEP3]])
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP1_IDX]], [[GEP2_IDX]]
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[GEP3_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP2]], [[GEP4_IDX_NEG]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr i32, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr i32, ptr %gep2, i64 %idx3
+ %gep4 = getelementptr i32, ptr %gep3, i64 %idx4
+ call void @use(ptr %gep3)
+ %cmp = icmp eq ptr %gep4, %base
+ ret i1 %cmp
+}
+
+define i1 @gep_multiple_multi_use_below_limit_consts(ptr %base, i64 %idx1, i64 %idx2) {
+; CHECK-LABEL: @gep_multiple_multi_use_below_limit_consts(
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 16
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl i64 [[IDX1:%.*]], 2
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP2_IDX]]
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 16
+; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[GEP3]], i64 [[GEP4_IDX]]
+; CHECK-NEXT: call void @use(ptr [[GEP4]])
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP2_IDX]], 32
+; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[GEP4_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr i32, ptr %base, i64 4
+ %gep2 = getelementptr i32, ptr %gep1, i64 %idx1
+ %gep3 = getelementptr i32, ptr %gep2, i64 4
+ %gep4 = getelementptr i32, ptr %gep3, i64 %idx2
+ call void @use(ptr %gep4)
+ %cmp = icmp eq ptr %gep4, %base
+ ret i1 %cmp
+}
+
+define i1 @gep_multiple_multi_use_above_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_multiple_multi_use_above_limit(
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i32, ptr [[BASE:%.*]], i64 [[IDX1:%.*]]
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i32, ptr [[GEP1]], i64 [[IDX2:%.*]]
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i32, ptr [[GEP2]], i64 [[IDX3:%.*]]
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i32, ptr [[GEP3]], i64 [[IDX4:%.*]]
+; CHECK-NEXT: call void @use(ptr [[GEP4]])
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq ptr [[GEP4]], [[BASE]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr i32, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr i32, ptr %gep2, i64 %idx3
+ %gep4 = getelementptr i32, ptr %gep3, i64 %idx4
+ call void @use(ptr %gep4)
+ %cmp = icmp eq ptr %gep4, %base
+ ret i1 %cmp
+}
diff --git a/llvm/test/Transforms/InstCombine/sub-gep.ll b/llvm/test/Transforms/InstCombine/sub-gep.ll
index 84e570395e03b..ef12b26746729 100644
--- a/llvm/test/Transforms/InstCombine/sub-gep.ll
+++ b/llvm/test/Transforms/InstCombine/sub-gep.ll
@@ -1172,3 +1172,55 @@ define i64 @nuw_ptrdiff_mul_nsw_nneg_scale_multiuse(ptr %base, i64 %idx) {
%diff = sub nuw i64 %lhs, %rhs
ret i64 %diff
}
+
+define i64 @multiple_geps_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @multiple_geps_multi_use_below_limit(
+; CHECK-NEXT: [[P1:%.*]] = getelementptr inbounds nuw i8, ptr [[BASE:%.*]], i64 [[IDX1:%.*]]
+; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[IDX2:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P2]])
+; CHECK-NEXT: [[P3:%.*]] = getelementptr inbounds nuw i8, ptr [[BASE]], i64 [[IDX3:%.*]]
+; CHECK-NEXT: [[P4:%.*]] = getelementptr inbounds nuw i8, ptr [[P3]], i64 [[IDX4:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P4]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[IDX1]], [[IDX2]]
+; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[IDX3]], [[IDX4]]
+; CHECK-NEXT: [[GEPDIFF:%.*]] = sub nsw i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret i64 [[GEPDIFF]]
+;
+ %p1 = getelementptr inbounds nuw i8, ptr %base, i64 %idx1
+ %p2 = getelementptr inbounds nuw i8, ptr %p1, i64 %idx2
+ call void @use(ptr %p2)
+ %p3 = getelementptr inbounds nuw i8, ptr %base, i64 %idx3
+ %p4 = getelementptr inbounds nuw i8, ptr %p3, i64 %idx4
+ call void @use(ptr %p4)
+ %i1 = ptrtoint ptr %p4 to i64
+ %i2 = ptrtoint ptr %p2 to i64
+ %d = sub i64 %i2, %i1
+ ret i64 %d
+}
+
+define i64 @multiple_geps_multi_use_above_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4, i64 %idx5) {
+; CHECK-LABEL: @multiple_geps_multi_use_above_limit(
+; CHECK-NEXT: [[P1:%.*]] = getelementptr inbounds nuw i8, ptr [[BASE:%.*]], i64 [[IDX1:%.*]]
+; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[IDX2:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P2]])
+; CHECK-NEXT: [[P3:%.*]] = getelementptr inbounds nuw i8, ptr [[BASE]], i64 [[IDX3:%.*]]
+; CHECK-NEXT: [[P4:%.*]] = getelementptr inbounds nuw i8, ptr [[P3]], i64 [[IDX4:%.*]]
+; CHECK-NEXT: [[P5:%.*]] = getelementptr inbounds nuw i8, ptr [[P4]], i64 [[IDX5:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P5]])
+; CHECK-NEXT: [[I1:%.*]] = ptrtoint ptr [[P5]] to i64
+; CHECK-NEXT: [[I2:%.*]] = ptrtoint ptr [[P2]] to i64
+; CHECK-NEXT: [[D:%.*]] = sub i64 [[I2]], [[I1]]
+; CHECK-NEXT: ret i64 [[D]]
+;
+ %p1 = getelementptr inbounds nuw i8, ptr %base, i64 %idx1
+ %p2 = getelementptr inbounds nuw i8, ptr %p1, i64 %idx2
+ call void @use(ptr %p2)
+ %p3 = getelementptr inbounds nuw i8, ptr %base, i64 %idx3
+ %p4 = getelementptr inbounds nuw i8, ptr %p3, i64 %idx4
+ %p5 = getelementptr inbounds nuw i8, ptr %p4, i64 %idx5
+ call void @use(ptr %p5)
+ %i1 = ptrtoint ptr %p5 to i64
+ %i2 = ptrtoint ptr %p2 to i64
+ %d = sub i64 %i2, %i1
+ ret i64 %d
+}
>From f2da011b8ed03e9ea9d0ba415ae783e07dd9e9ba Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov at redhat.com>
Date: Mon, 21 Jul 2025 11:37:17 +0200
Subject: [PATCH 2/4] Update for new one-use GEP expansion
---
.../InstCombine/InstCombineAddSub.cpp | 17 +++---
llvm/test/Transforms/InstCombine/icmp-gep.ll | 56 ++++++++-----------
llvm/test/Transforms/InstCombine/sub-gep.ll | 26 ++++-----
3 files changed, 44 insertions(+), 55 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index a2e7ca4004174..027e35a239f2b 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -2152,14 +2152,17 @@ bool CommonPointerBase::isExpensive() const {
auto ProcessGEPs = [&SeenConst, &NumGEPs](ArrayRef<GEPOperator *> GEPs) {
bool SeenMultiUse = false;
for (GEPOperator *GEP : GEPs) {
- // Only count GEPs after the first multi-use GEP. For the first one,
- // we will directly reuse the offset.
- if (SeenMultiUse) {
- bool IsConst = GEP->hasAllConstantIndices();
- SeenConst |= IsConst;
- NumGEPs += !IsConst;
+ // Only count multi-use GEPs, excluding the first one. For the first one,
+ // we will directly reuse the offset. For one-use GEPs, their offset will
+ // be folded into a multi-use GEP.
+ if (!GEP->hasOneUse()) {
+ if (SeenMultiUse) {
+ bool IsConst = GEP->hasAllConstantIndices();
+ SeenConst |= IsConst;
+ NumGEPs += !IsConst;
+ }
+ SeenMultiUse = true;
}
- SeenMultiUse |= !GEP->hasOneUse();
}
};
ProcessGEPs(LHSGEPs);
diff --git a/llvm/test/Transforms/InstCombine/icmp-gep.ll b/llvm/test/Transforms/InstCombine/icmp-gep.ll
index 6c7572089532a..54fcccdd4dead 100644
--- a/llvm/test/Transforms/InstCombine/icmp-gep.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-gep.ll
@@ -852,16 +852,12 @@ define i1 @gep_mugtiple_ugt_inbounds_nusw(ptr %base, i64 %idx, i64 %idx2) {
define i1 @gep_multiple_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3) {
; CHECK-LABEL: @gep_multiple_multi_use_below_limit(
-; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl i64 [[IDX1:%.*]], 2
-; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[GEP1_IDX]]
-; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
-; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP2_IDX]]
-; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3:%.*]], 2
-; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 [[GEP3_IDX]]
+; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
+; CHECK-NEXT: [[IDX3:%.*]] = add i64 [[GEP1_IDX1]], [[IDX4:%.*]]
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3]], 2
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2:%.*]], i64 [[GEP3_IDX]]
; CHECK-NEXT: call void @use(ptr [[GEP3]])
-; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP1_IDX]], [[GEP2_IDX]]
-; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[GEP3_IDX]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[GEP3_IDX]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 %idx1
@@ -874,17 +870,14 @@ define i1 @gep_multiple_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2, i
define i1 @gep_multiple_multi_use_below_limit_extra_one_use_gep(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
; CHECK-LABEL: @gep_multiple_multi_use_below_limit_extra_one_use_gep(
-; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl i64 [[IDX1:%.*]], 2
-; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[GEP1_IDX]]
-; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
-; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP2_IDX]]
-; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3:%.*]], 2
-; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 [[GEP3_IDX]]
-; CHECK-NEXT: [[GEP4_IDX_NEG:%.*]] = mul i64 [[IDX4:%.*]], -4
+; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
+; CHECK-NEXT: [[IDX3:%.*]] = add i64 [[GEP1_IDX1]], [[IDX5:%.*]]
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3]], 2
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2:%.*]], i64 [[GEP3_IDX]]
; CHECK-NEXT: call void @use(ptr [[GEP3]])
-; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP1_IDX]], [[GEP2_IDX]]
-; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[GEP3_IDX]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP2]], [[GEP4_IDX_NEG]]
+; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[IDX3]], [[IDX4:%.*]]
+; CHECK-NEXT: [[DOTMASK:%.*]] = and i64 [[TMP3]], 4611686018427387903
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[DOTMASK]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 %idx1
@@ -898,16 +891,12 @@ define i1 @gep_multiple_multi_use_below_limit_extra_one_use_gep(ptr %base, i64 %
define i1 @gep_multiple_multi_use_below_limit_consts(ptr %base, i64 %idx1, i64 %idx2) {
; CHECK-LABEL: @gep_multiple_multi_use_below_limit_consts(
-; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 16
-; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl i64 [[IDX1:%.*]], 2
-; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP2_IDX]]
-; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 16
-; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
-; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[GEP3]], i64 [[GEP4_IDX]]
+; CHECK-NEXT: [[IDX2:%.*]] = add i64 [[IDX1:%.*]], [[IDX3:%.*]]
+; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX2]], 2
+; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[GEP4_IDX]], 32
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[TMP3]]
; CHECK-NEXT: call void @use(ptr [[GEP4]])
-; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP2_IDX]], 32
-; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[GEP4_IDX]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP3]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 4
@@ -921,12 +910,13 @@ define i1 @gep_multiple_multi_use_below_limit_consts(ptr %base, i64 %idx1, i64 %
define i1 @gep_multiple_multi_use_above_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
; CHECK-LABEL: @gep_multiple_multi_use_above_limit(
-; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i32, ptr [[BASE:%.*]], i64 [[IDX1:%.*]]
-; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i32, ptr [[GEP1]], i64 [[IDX2:%.*]]
-; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i32, ptr [[GEP2]], i64 [[IDX3:%.*]]
-; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i32, ptr [[GEP3]], i64 [[IDX4:%.*]]
+; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP1_IDX1]], [[IDX3:%.*]]
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[IDX4:%.*]]
+; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 2
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[TMP3]]
; CHECK-NEXT: call void @use(ptr [[GEP4]])
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq ptr [[GEP4]], [[BASE]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP3]], 0
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 %idx1
diff --git a/llvm/test/Transforms/InstCombine/sub-gep.ll b/llvm/test/Transforms/InstCombine/sub-gep.ll
index ef12b26746729..4f7c6c1c216fd 100644
--- a/llvm/test/Transforms/InstCombine/sub-gep.ll
+++ b/llvm/test/Transforms/InstCombine/sub-gep.ll
@@ -1175,15 +1175,13 @@ define i64 @nuw_ptrdiff_mul_nsw_nneg_scale_multiuse(ptr %base, i64 %idx) {
define i64 @multiple_geps_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
; CHECK-LABEL: @multiple_geps_multi_use_below_limit(
-; CHECK-NEXT: [[P1:%.*]] = getelementptr inbounds nuw i8, ptr [[BASE:%.*]], i64 [[IDX1:%.*]]
-; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[IDX2:%.*]]
+; CHECK-NEXT: [[IDX2:%.*]] = add nuw nsw i64 [[IDX1:%.*]], [[IDX5:%.*]]
+; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1:%.*]], i64 [[IDX2]]
; CHECK-NEXT: call void @use(ptr [[P2]])
-; CHECK-NEXT: [[P3:%.*]] = getelementptr inbounds nuw i8, ptr [[BASE]], i64 [[IDX3:%.*]]
-; CHECK-NEXT: [[P4:%.*]] = getelementptr inbounds nuw i8, ptr [[P3]], i64 [[IDX4:%.*]]
+; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[IDX3:%.*]], [[IDX4:%.*]]
+; CHECK-NEXT: [[P4:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[TMP2]]
; CHECK-NEXT: call void @use(ptr [[P4]])
-; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[IDX1]], [[IDX2]]
-; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[IDX3]], [[IDX4]]
-; CHECK-NEXT: [[GEPDIFF:%.*]] = sub nsw i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: [[GEPDIFF:%.*]] = sub nsw i64 [[IDX2]], [[TMP2]]
; CHECK-NEXT: ret i64 [[GEPDIFF]]
;
%p1 = getelementptr inbounds nuw i8, ptr %base, i64 %idx1
@@ -1200,16 +1198,14 @@ define i64 @multiple_geps_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2,
define i64 @multiple_geps_multi_use_above_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4, i64 %idx5) {
; CHECK-LABEL: @multiple_geps_multi_use_above_limit(
-; CHECK-NEXT: [[P1:%.*]] = getelementptr inbounds nuw i8, ptr [[BASE:%.*]], i64 [[IDX1:%.*]]
-; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[IDX2:%.*]]
+; CHECK-NEXT: [[IDX2:%.*]] = add nuw nsw i64 [[IDX1:%.*]], [[IDX6:%.*]]
+; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1:%.*]], i64 [[IDX2]]
; CHECK-NEXT: call void @use(ptr [[P2]])
-; CHECK-NEXT: [[P3:%.*]] = getelementptr inbounds nuw i8, ptr [[BASE]], i64 [[IDX3:%.*]]
-; CHECK-NEXT: [[P4:%.*]] = getelementptr inbounds nuw i8, ptr [[P3]], i64 [[IDX4:%.*]]
-; CHECK-NEXT: [[P5:%.*]] = getelementptr inbounds nuw i8, ptr [[P4]], i64 [[IDX5:%.*]]
+; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[IDX3:%.*]], [[IDX4:%.*]]
+; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], [[IDX5:%.*]]
+; CHECK-NEXT: [[P5:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[TMP3]]
; CHECK-NEXT: call void @use(ptr [[P5]])
-; CHECK-NEXT: [[I1:%.*]] = ptrtoint ptr [[P5]] to i64
-; CHECK-NEXT: [[I2:%.*]] = ptrtoint ptr [[P2]] to i64
-; CHECK-NEXT: [[D:%.*]] = sub i64 [[I2]], [[I1]]
+; CHECK-NEXT: [[D:%.*]] = sub nsw i64 [[IDX2]], [[TMP3]]
; CHECK-NEXT: ret i64 [[D]]
;
%p1 = getelementptr inbounds nuw i8, ptr %base, i64 %idx1
>From b8ca1c6080d512e850c34c57469f0b06335116a4 Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov at redhat.com>
Date: Mon, 21 Jul 2025 11:45:28 +0200
Subject: [PATCH 3/4] Adjust tests for new behavior
---
llvm/test/Transforms/InstCombine/icmp-gep.ll | 106 ++++++++++++++-----
llvm/test/Transforms/InstCombine/sub-gep.ll | 36 +++++--
2 files changed, 107 insertions(+), 35 deletions(-)
diff --git a/llvm/test/Transforms/InstCombine/icmp-gep.ll b/llvm/test/Transforms/InstCombine/icmp-gep.ll
index 54fcccdd4dead..7f18de74b7aa8 100644
--- a/llvm/test/Transforms/InstCombine/icmp-gep.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-gep.ll
@@ -852,56 +852,109 @@ define i1 @gep_mugtiple_ugt_inbounds_nusw(ptr %base, i64 %idx, i64 %idx2) {
define i1 @gep_multiple_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3) {
; CHECK-LABEL: @gep_multiple_multi_use_below_limit(
-; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
-; CHECK-NEXT: [[IDX3:%.*]] = add i64 [[GEP1_IDX1]], [[IDX4:%.*]]
-; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3]], 2
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3:%.*]], 2
; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2:%.*]], i64 [[GEP3_IDX]]
; CHECK-NEXT: call void @use(ptr [[GEP3]])
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[GEP3_IDX]], 0
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[GEP3]], i64 [[GEP2_IDX]]
+; CHECK-NEXT: call void @use(ptr [[GEP4]])
+; CHECK-NEXT: [[GEP3_IDX1:%.*]] = shl i64 [[IDX4:%.*]], 2
+; CHECK-NEXT: [[GEP5:%.*]] = getelementptr i8, ptr [[GEP4]], i64 [[GEP3_IDX1]]
+; CHECK-NEXT: call void @use(ptr [[GEP5]])
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP3_IDX]], [[GEP2_IDX]]
+; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[GEP3_IDX1]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 %idx1
+ call void @use(ptr %gep1)
%gep2 = getelementptr i32, ptr %gep1, i64 %idx2
+ call void @use(ptr %gep2)
%gep3 = getelementptr i32, ptr %gep2, i64 %idx3
call void @use(ptr %gep3)
%cmp = icmp eq ptr %gep3, %base
ret i1 %cmp
}
-define i1 @gep_multiple_multi_use_below_limit_extra_one_use_gep(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
-; CHECK-LABEL: @gep_multiple_multi_use_below_limit_extra_one_use_gep(
-; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
-; CHECK-NEXT: [[IDX3:%.*]] = add i64 [[GEP1_IDX1]], [[IDX5:%.*]]
-; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3]], 2
-; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2:%.*]], i64 [[GEP3_IDX]]
+define i1 @gep_multiple_multi_use_below_limit_extra_one_use_gep1(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_multiple_multi_use_below_limit_extra_one_use_gep1(
+; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl i64 [[IDX1:%.*]], 2
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[GEP1_IDX]]
+; CHECK-NEXT: call void @use(ptr [[GEP1]])
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP2_IDX]]
+; CHECK-NEXT: call void @use(ptr [[GEP2]])
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 [[GEP3_IDX]]
; CHECK-NEXT: call void @use(ptr [[GEP3]])
-; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[IDX3]], [[IDX4:%.*]]
-; CHECK-NEXT: [[DOTMASK:%.*]] = and i64 [[TMP3]], 4611686018427387903
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[DOTMASK]], 0
+; CHECK-NEXT: [[GEP4_IDX_NEG:%.*]] = mul i64 [[IDX4:%.*]], -4
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP1_IDX]], [[GEP2_IDX]]
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[GEP3_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP2]], [[GEP4_IDX_NEG]]
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 %idx1
+ call void @use(ptr %gep1)
%gep2 = getelementptr i32, ptr %gep1, i64 %idx2
+ call void @use(ptr %gep2)
%gep3 = getelementptr i32, ptr %gep2, i64 %idx3
+ call void @use(ptr %gep3)
%gep4 = getelementptr i32, ptr %gep3, i64 %idx4
+ %cmp = icmp eq ptr %gep4, %base
+ ret i1 %cmp
+}
+
+define i1 @gep_multiple_multi_use_below_limit_extra_one_use_gep2(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_multiple_multi_use_below_limit_extra_one_use_gep2(
+; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[GEP1_IDX1]], 2
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[TMP1]]
+; CHECK-NEXT: call void @use(ptr [[GEP2]])
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 [[GEP3_IDX]]
+; CHECK-NEXT: call void @use(ptr [[GEP3]])
+; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX4:%.*]], 2
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[GEP3]], i64 [[GEP4_IDX]]
+; CHECK-NEXT: call void @use(ptr [[GEP4]])
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[GEP3_IDX]]
+; CHECK-NEXT: [[GEP4_IDX_NEG:%.*]] = sub i64 0, [[GEP4_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP2]], [[GEP4_IDX_NEG]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr i32, ptr %gep1, i64 %idx2
+ call void @use(ptr %gep2)
+ %gep3 = getelementptr i32, ptr %gep2, i64 %idx3
call void @use(ptr %gep3)
+ %gep4 = getelementptr i32, ptr %gep3, i64 %idx4
+ call void @use(ptr %gep4)
%cmp = icmp eq ptr %gep4, %base
ret i1 %cmp
}
define i1 @gep_multiple_multi_use_below_limit_consts(ptr %base, i64 %idx1, i64 %idx2) {
; CHECK-LABEL: @gep_multiple_multi_use_below_limit_consts(
-; CHECK-NEXT: [[IDX2:%.*]] = add i64 [[IDX1:%.*]], [[IDX3:%.*]]
-; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX2]], 2
-; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[GEP4_IDX]], 32
-; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[TMP3]]
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 16
+; CHECK-NEXT: call void @use(ptr [[GEP1]])
+; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP4_IDX]]
; CHECK-NEXT: call void @use(ptr [[GEP4]])
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP3]], 0
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP4]], i64 16
+; CHECK-NEXT: call void @use(ptr [[GEP3]])
+; CHECK-NEXT: [[GEP4_IDX1:%.*]] = shl i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP5:%.*]] = getelementptr i8, ptr [[GEP3]], i64 [[GEP4_IDX1]]
+; CHECK-NEXT: call void @use(ptr [[GEP5]])
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP4_IDX]], 32
+; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[GEP4_IDX1]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 4
+ call void @use(ptr %gep1)
%gep2 = getelementptr i32, ptr %gep1, i64 %idx1
+ call void @use(ptr %gep2)
%gep3 = getelementptr i32, ptr %gep2, i64 4
+ call void @use(ptr %gep3)
%gep4 = getelementptr i32, ptr %gep3, i64 %idx2
call void @use(ptr %gep4)
%cmp = icmp eq ptr %gep4, %base
@@ -910,18 +963,23 @@ define i1 @gep_multiple_multi_use_below_limit_consts(ptr %base, i64 %idx1, i64 %
define i1 @gep_multiple_multi_use_above_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
; CHECK-LABEL: @gep_multiple_multi_use_above_limit(
-; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
-; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP1_IDX1]], [[IDX3:%.*]]
-; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[IDX4:%.*]]
-; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 2
-; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 [[TMP3]]
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i32, ptr [[BASE:%.*]], i64 [[IDX1:%.*]]
; CHECK-NEXT: call void @use(ptr [[GEP4]])
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP3]], 0
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i32, ptr [[GEP4]], i64 [[IDX2:%.*]]
+; CHECK-NEXT: call void @use(ptr [[GEP3]])
+; CHECK-NEXT: [[GEP5:%.*]] = getelementptr i32, ptr [[GEP3]], i64 [[IDX3:%.*]]
+; CHECK-NEXT: call void @use(ptr [[GEP5]])
+; CHECK-NEXT: [[GEP6:%.*]] = getelementptr i32, ptr [[GEP5]], i64 [[IDX4:%.*]]
+; CHECK-NEXT: call void @use(ptr [[GEP6]])
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq ptr [[GEP6]], [[BASE]]
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 %idx1
+ call void @use(ptr %gep1)
%gep2 = getelementptr i32, ptr %gep1, i64 %idx2
+ call void @use(ptr %gep2)
%gep3 = getelementptr i32, ptr %gep2, i64 %idx3
+ call void @use(ptr %gep3)
%gep4 = getelementptr i32, ptr %gep3, i64 %idx4
call void @use(ptr %gep4)
%cmp = icmp eq ptr %gep4, %base
diff --git a/llvm/test/Transforms/InstCombine/sub-gep.ll b/llvm/test/Transforms/InstCombine/sub-gep.ll
index 4f7c6c1c216fd..45e5686ad70e7 100644
--- a/llvm/test/Transforms/InstCombine/sub-gep.ll
+++ b/llvm/test/Transforms/InstCombine/sub-gep.ll
@@ -1175,19 +1175,25 @@ define i64 @nuw_ptrdiff_mul_nsw_nneg_scale_multiuse(ptr %base, i64 %idx) {
define i64 @multiple_geps_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
; CHECK-LABEL: @multiple_geps_multi_use_below_limit(
-; CHECK-NEXT: [[IDX2:%.*]] = add nuw nsw i64 [[IDX1:%.*]], [[IDX5:%.*]]
-; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1:%.*]], i64 [[IDX2]]
+; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1:%.*]], i64 [[IDX2:%.*]]
; CHECK-NEXT: call void @use(ptr [[P2]])
-; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[IDX3:%.*]], [[IDX4:%.*]]
-; CHECK-NEXT: [[P4:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[TMP2]]
+; CHECK-NEXT: [[P4:%.*]] = getelementptr inbounds nuw i8, ptr [[P2]], i64 [[IDX5:%.*]]
; CHECK-NEXT: call void @use(ptr [[P4]])
-; CHECK-NEXT: [[GEPDIFF:%.*]] = sub nsw i64 [[IDX2]], [[TMP2]]
+; CHECK-NEXT: [[P3:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[IDX3:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P3]])
+; CHECK-NEXT: [[P5:%.*]] = getelementptr inbounds nuw i8, ptr [[P3]], i64 [[IDX4:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P5]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw nsw i64 [[IDX2]], [[IDX5]]
+; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[IDX3]], [[IDX4]]
+; CHECK-NEXT: [[GEPDIFF:%.*]] = sub nsw i64 [[TMP1]], [[TMP2]]
; CHECK-NEXT: ret i64 [[GEPDIFF]]
;
%p1 = getelementptr inbounds nuw i8, ptr %base, i64 %idx1
+ call void @use(ptr %p1)
%p2 = getelementptr inbounds nuw i8, ptr %p1, i64 %idx2
call void @use(ptr %p2)
%p3 = getelementptr inbounds nuw i8, ptr %base, i64 %idx3
+ call void @use(ptr %p3)
%p4 = getelementptr inbounds nuw i8, ptr %p3, i64 %idx4
call void @use(ptr %p4)
%i1 = ptrtoint ptr %p4 to i64
@@ -1198,21 +1204,29 @@ define i64 @multiple_geps_multi_use_below_limit(ptr %base, i64 %idx1, i64 %idx2,
define i64 @multiple_geps_multi_use_above_limit(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4, i64 %idx5) {
; CHECK-LABEL: @multiple_geps_multi_use_above_limit(
-; CHECK-NEXT: [[IDX2:%.*]] = add nuw nsw i64 [[IDX1:%.*]], [[IDX6:%.*]]
-; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1:%.*]], i64 [[IDX2]]
+; CHECK-NEXT: [[P2:%.*]] = getelementptr inbounds nuw i8, ptr [[P1:%.*]], i64 [[IDX2:%.*]]
; CHECK-NEXT: call void @use(ptr [[P2]])
-; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[IDX3:%.*]], [[IDX4:%.*]]
-; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], [[IDX5:%.*]]
-; CHECK-NEXT: [[P5:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[TMP3]]
+; CHECK-NEXT: [[P3:%.*]] = getelementptr inbounds nuw i8, ptr [[P2]], i64 [[IDX6:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P3]])
+; CHECK-NEXT: [[P5:%.*]] = getelementptr inbounds nuw i8, ptr [[P1]], i64 [[TMP3:%.*]]
; CHECK-NEXT: call void @use(ptr [[P5]])
-; CHECK-NEXT: [[D:%.*]] = sub nsw i64 [[IDX2]], [[TMP3]]
+; CHECK-NEXT: [[P6:%.*]] = getelementptr inbounds nuw i8, ptr [[P5]], i64 [[IDX7:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P6]])
+; CHECK-NEXT: [[P7:%.*]] = getelementptr inbounds nuw i8, ptr [[P6]], i64 [[IDX5:%.*]]
+; CHECK-NEXT: call void @use(ptr [[P7]])
+; CHECK-NEXT: [[I1:%.*]] = ptrtoint ptr [[P7]] to i64
+; CHECK-NEXT: [[I2:%.*]] = ptrtoint ptr [[P3]] to i64
+; CHECK-NEXT: [[D:%.*]] = sub i64 [[I2]], [[I1]]
; CHECK-NEXT: ret i64 [[D]]
;
%p1 = getelementptr inbounds nuw i8, ptr %base, i64 %idx1
+ call void @use(ptr %p1)
%p2 = getelementptr inbounds nuw i8, ptr %p1, i64 %idx2
call void @use(ptr %p2)
%p3 = getelementptr inbounds nuw i8, ptr %base, i64 %idx3
+ call void @use(ptr %p3)
%p4 = getelementptr inbounds nuw i8, ptr %p3, i64 %idx4
+ call void @use(ptr %p4)
%p5 = getelementptr inbounds nuw i8, ptr %p4, i64 %idx5
call void @use(ptr %p5)
%i1 = ptrtoint ptr %p5 to i64
>From 5709fc1f15f4de9b6ba6074205bda69fe9745877 Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov at redhat.com>
Date: Mon, 21 Jul 2025 11:46:50 +0200
Subject: [PATCH 4/4] Remove constant special case
With the new expansion this would have to be more careful about
a constant on multi-use GEP and non-constant on merged one-use
GEPs. Just drop this heuristic for now.
---
.../InstCombine/InstCombineAddSub.cpp | 11 +++-------
llvm/test/Transforms/InstCombine/icmp-gep.ll | 20 ++++++++-----------
2 files changed, 11 insertions(+), 20 deletions(-)
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index 027e35a239f2b..7f605be976549 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -2147,27 +2147,22 @@ CommonPointerBase CommonPointerBase::compute(Value *LHS, Value *RHS) {
}
bool CommonPointerBase::isExpensive() const {
- bool SeenConst = false;
unsigned NumGEPs = 0;
- auto ProcessGEPs = [&SeenConst, &NumGEPs](ArrayRef<GEPOperator *> GEPs) {
+ auto ProcessGEPs = [&NumGEPs](ArrayRef<GEPOperator *> GEPs) {
bool SeenMultiUse = false;
for (GEPOperator *GEP : GEPs) {
// Only count multi-use GEPs, excluding the first one. For the first one,
// we will directly reuse the offset. For one-use GEPs, their offset will
// be folded into a multi-use GEP.
if (!GEP->hasOneUse()) {
- if (SeenMultiUse) {
- bool IsConst = GEP->hasAllConstantIndices();
- SeenConst |= IsConst;
- NumGEPs += !IsConst;
- }
+ if (SeenMultiUse)
+ ++NumGEPs;
SeenMultiUse = true;
}
}
};
ProcessGEPs(LHSGEPs);
ProcessGEPs(RHSGEPs);
- NumGEPs += SeenConst;
return NumGEPs > 2;
}
diff --git a/llvm/test/Transforms/InstCombine/icmp-gep.ll b/llvm/test/Transforms/InstCombine/icmp-gep.ll
index 7f18de74b7aa8..938ec64c2bcad 100644
--- a/llvm/test/Transforms/InstCombine/icmp-gep.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-gep.ll
@@ -932,21 +932,17 @@ define i1 @gep_multiple_multi_use_below_limit_extra_one_use_gep2(ptr %base, i64
ret i1 %cmp
}
-define i1 @gep_multiple_multi_use_below_limit_consts(ptr %base, i64 %idx1, i64 %idx2) {
-; CHECK-LABEL: @gep_multiple_multi_use_below_limit_consts(
+define i1 @gep_multiple_multi_above_below_limit_consts(ptr %base, i64 %idx1, i64 %idx2) {
+; CHECK-LABEL: @gep_multiple_multi_above_below_limit_consts(
; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i8, ptr [[BASE:%.*]], i64 16
; CHECK-NEXT: call void @use(ptr [[GEP1]])
-; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX2:%.*]], 2
-; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i8, ptr [[GEP1]], i64 [[GEP4_IDX]]
-; CHECK-NEXT: call void @use(ptr [[GEP4]])
-; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP4]], i64 16
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr i32, ptr [[GEP1]], i64 [[IDX1:%.*]]
+; CHECK-NEXT: call void @use(ptr [[GEP2]])
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr i8, ptr [[GEP2]], i64 16
; CHECK-NEXT: call void @use(ptr [[GEP3]])
-; CHECK-NEXT: [[GEP4_IDX1:%.*]] = shl i64 [[IDX3:%.*]], 2
-; CHECK-NEXT: [[GEP5:%.*]] = getelementptr i8, ptr [[GEP3]], i64 [[GEP4_IDX1]]
-; CHECK-NEXT: call void @use(ptr [[GEP5]])
-; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[GEP4_IDX]], 32
-; CHECK-NEXT: [[TMP2:%.*]] = sub i64 0, [[GEP4_IDX1]]
-; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i32, ptr [[GEP3]], i64 [[IDX2:%.*]]
+; CHECK-NEXT: call void @use(ptr [[GEP4]])
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq ptr [[GEP4]], [[BASE]]
; CHECK-NEXT: ret i1 [[CMP]]
;
%gep1 = getelementptr i32, ptr %base, i64 4
More information about the llvm-commits
mailing list