[InstCombine] Fold icmp of gep chains #146714
Open
+159
−13
Conversation
This file contains hidden or 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
llvmbot
added
llvm:instcombine
|
@llvm/pr-subscribers-llvm-transforms Author: Nikita Popov (nikic) ChangesThis extends #144065 to the general case of an icmp between two GEP chains that have a common base. Full diff: https://github.com/llvm/llvm-project/pull/146714.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 6737b50405ee2..1b13a11e4652f 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -752,8 +752,7 @@ Instruction *InstCombinerImpl::foldGEPICmp(GEPOperator *GEPLHS, Value *RHS,
// If the base pointers are different, but the indices are the same, just
// compare the base pointer.
- Value *PtrBase = GEPLHS->getOperand(0);
- if (PtrBase != GEPRHS->getOperand(0)) {
+ if (GEPLHS->getOperand(0) != GEPRHS->getOperand(0)) {
bool IndicesTheSame =
GEPLHS->getNumOperands() == GEPRHS->getNumOperands() &&
GEPLHS->getPointerOperand()->getType() ==
@@ -779,7 +778,7 @@ Instruction *InstCombinerImpl::foldGEPICmp(GEPOperator *GEPLHS, Value *RHS,
if (GEPLHS->isInBounds() && GEPRHS->isInBounds() &&
(GEPLHS->hasAllConstantIndices() || GEPLHS->hasOneUse()) &&
(GEPRHS->hasAllConstantIndices() || GEPRHS->hasOneUse()) &&
- PtrBase->stripPointerCasts() ==
+ GEPLHS->getOperand(0)->stripPointerCasts() ==
GEPRHS->getOperand(0)->stripPointerCasts() &&
!GEPLHS->getType()->isVectorTy()) {
Value *LOffset = EmitGEPOffset(GEPLHS);
@@ -802,14 +801,10 @@ Instruction *InstCombinerImpl::foldGEPICmp(GEPOperator *GEPLHS, Value *RHS,
LOffset, ROffset);
return replaceInstUsesWith(I, Cmp);
}
-
- // Otherwise, the base pointers are different and the indices are
- // different. Try convert this to an indexed compare by looking through
- // PHIs/casts.
- return transformToIndexedCompare(GEPLHS, RHS, Cond, DL, *this);
}
- if (GEPLHS->getNumOperands() == GEPRHS->getNumOperands() &&
+ if (GEPLHS->getOperand(0) == GEPRHS->getOperand(0) &&
+ GEPLHS->getNumOperands() == GEPRHS->getNumOperands() &&
GEPLHS->getSourceElementType() == GEPRHS->getSourceElementType()) {
// If the GEPs only differ by one index, compare it.
unsigned NumDifferences = 0; // Keep track of # differences.
@@ -846,11 +841,14 @@ Instruction *InstCombinerImpl::foldGEPICmp(GEPOperator *GEPLHS, Value *RHS,
}
}
- if (CanFold(NW)) {
+ if (Base.Ptr && CanFold(Base.LHSNW & Base.RHSNW)) {
// ((gep Ptr, OFFSET1) cmp (gep Ptr, OFFSET2) ---> (OFFSET1 cmp OFFSET2)
- Value *L = EmitGEPOffset(GEPLHS, /*RewriteGEP=*/true);
- Value *R = EmitGEPOffset(GEPRHS, /*RewriteGEP=*/true);
- return NewICmp(NW, L, R);
+ Type *IdxTy = DL.getIndexType(GEPLHS->getType());
+ Value *L =
+ EmitGEPOffsets(Base.LHSGEPs, Base.LHSNW, IdxTy, /*RewriteGEP=*/true);
+ Value *R =
+ EmitGEPOffsets(Base.RHSGEPs, Base.RHSNW, IdxTy, /*RewriteGEP=*/true);
+ return NewICmp(Base.LHSNW & Base.RHSNW, L, R);
}
}
diff --git a/llvm/test/Transforms/InstCombine/icmp-gep.ll b/llvm/test/Transforms/InstCombine/icmp-gep.ll
index 3f104056fb1f2..6fd1eb112b450 100644
--- a/llvm/test/Transforms/InstCombine/icmp-gep.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-gep.ll
@@ -849,3 +849,151 @@ 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_gep_multiple_eq(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_gep_multiple_eq(
+; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
+; CHECK-NEXT: [[GEP3_IDX2:%.*]] = add i64 [[IDX3:%.*]], [[IDX4:%.*]]
+; CHECK-NEXT: [[CMP_UNSHIFTED:%.*]] = xor i64 [[GEP1_IDX1]], [[GEP3_IDX2]]
+; CHECK-NEXT: [[CMP_MASK:%.*]] = and i64 [[CMP_UNSHIFTED]], 4611686018427387903
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[CMP_MASK]], 0
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr i32, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr i32, ptr %base, i64 %idx3
+ %gep4 = getelementptr i32, ptr %gep3, i64 %idx4
+ %cmp = icmp eq ptr %gep2, %gep4
+ ret i1 %cmp
+}
+
+define i1 @gep_gep_multiple_eq_nuw(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_gep_multiple_eq_nuw(
+; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
+; CHECK-NEXT: [[GEP3_IDX2:%.*]] = add i64 [[IDX3:%.*]], [[IDX4:%.*]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[GEP1_IDX1]], [[GEP3_IDX2]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr nuw i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr nuw i32, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr nuw i32, ptr %base, i64 %idx3
+ %gep4 = getelementptr nuw i32, ptr %gep3, i64 %idx4
+ %cmp = icmp eq ptr %gep2, %gep4
+ ret i1 %cmp
+}
+
+define i1 @gep_gep_multiple_eq_nuw_different_scales(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_gep_multiple_eq_nuw_different_scales(
+; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl nuw i64 [[IDX1:%.*]], 2
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl nuw i64 [[IDX2:%.*]], 3
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl nuw i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl nuw i64 [[IDX4:%.*]], 3
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i64 [[GEP1_IDX]], [[GEP2_IDX]]
+; CHECK-NEXT: [[TMP2:%.*]] = add nuw i64 [[GEP3_IDX]], [[GEP4_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr nuw i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr nuw i64, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr nuw i32, ptr %base, i64 %idx3
+ %gep4 = getelementptr nuw i64, ptr %gep3, i64 %idx4
+ %cmp = icmp eq ptr %gep2, %gep4
+ ret i1 %cmp
+}
+
+define i1 @gep_gep_multiple_eq_partial_nuw_different_scales(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_gep_multiple_eq_partial_nuw_different_scales(
+; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl nuw i64 [[IDX1:%.*]], 2
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl nuw i64 [[IDX2:%.*]], 3
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl nuw i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX4:%.*]], 3
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i64 [[GEP1_IDX]], [[GEP2_IDX]]
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[GEP3_IDX]], [[GEP4_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr nuw i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr nuw i64, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr nuw i32, ptr %base, i64 %idx3
+ %gep4 = getelementptr i64, ptr %gep3, i64 %idx4
+ %cmp = icmp eq ptr %gep2, %gep4
+ ret i1 %cmp
+}
+
+define i1 @gep_gep_multiple_eq_partial_inbounds_different_scales(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_gep_multiple_eq_partial_inbounds_different_scales(
+; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl nsw i64 [[IDX1:%.*]], 2
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl nsw i64 [[IDX2:%.*]], 3
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl nsw i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl i64 [[IDX4:%.*]], 3
+; CHECK-NEXT: [[TMP1:%.*]] = add nsw i64 [[GEP1_IDX]], [[GEP2_IDX]]
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[GEP3_IDX]], [[GEP4_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr inbounds i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr inbounds i64, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr inbounds i32, ptr %base, i64 %idx3
+ %gep4 = getelementptr i64, ptr %gep3, i64 %idx4
+ %cmp = icmp eq ptr %gep2, %gep4
+ ret i1 %cmp
+}
+
+define i1 @gep_gep_multiple_ult_nuw(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_gep_multiple_ult_nuw(
+; CHECK-NEXT: [[GEP1_IDX1:%.*]] = add i64 [[IDX1:%.*]], [[IDX2:%.*]]
+; CHECK-NEXT: [[GEP3_IDX2:%.*]] = add i64 [[IDX3:%.*]], [[IDX4:%.*]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[GEP1_IDX1]], [[GEP3_IDX2]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr nuw i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr nuw i32, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr nuw i32, ptr %base, i64 %idx3
+ %gep4 = getelementptr nuw i32, ptr %gep3, i64 %idx4
+ %cmp = icmp ult ptr %gep2, %gep4
+ ret i1 %cmp
+}
+
+define i1 @gep_gep_multiple_ult_missing_nuw(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_gep_multiple_ult_missing_nuw(
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr nuw i32, ptr [[BASE:%.*]], i64 [[IDX1:%.*]]
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr nuw i32, ptr [[GEP1]], i64 [[IDX2:%.*]]
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr nuw i32, ptr [[BASE]], i64 [[IDX3:%.*]]
+; CHECK-NEXT: [[GEP4:%.*]] = getelementptr i32, ptr [[GEP3]], i64 [[IDX4:%.*]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult ptr [[GEP2]], [[GEP4]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr nuw i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr nuw i32, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr nuw i32, ptr %base, i64 %idx3
+ %gep4 = getelementptr i32, ptr %gep3, i64 %idx4
+ %cmp = icmp ult ptr %gep2, %gep4
+ ret i1 %cmp
+}
+
+define i1 @gep_gep_multiple_ult_nuw_multi_use(ptr %base, i64 %idx1, i64 %idx2, i64 %idx3, i64 %idx4) {
+; CHECK-LABEL: @gep_gep_multiple_ult_nuw_multi_use(
+; CHECK-NEXT: [[GEP1_IDX:%.*]] = shl nuw i64 [[IDX1:%.*]], 2
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr nuw i8, ptr [[BASE:%.*]], i64 [[GEP1_IDX]]
+; CHECK-NEXT: [[GEP2_IDX:%.*]] = shl nuw i64 [[IDX2:%.*]], 2
+; CHECK-NEXT: [[GEP2:%.*]] = getelementptr nuw i8, ptr [[GEP1]], i64 [[GEP2_IDX]]
+; CHECK-NEXT: [[GEP3_IDX:%.*]] = shl nuw i64 [[IDX3:%.*]], 2
+; CHECK-NEXT: [[GEP3:%.*]] = getelementptr nuw i8, ptr [[BASE]], i64 [[GEP3_IDX]]
+; CHECK-NEXT: [[GEP4_IDX:%.*]] = shl nuw i64 [[IDX4:%.*]], 2
+; CHECK-NEXT: [[GEP5:%.*]] = getelementptr nuw i8, ptr [[GEP3]], i64 [[GEP4_IDX]]
+; CHECK-NEXT: call void @use(ptr [[GEP2]])
+; CHECK-NEXT: call void @use(ptr [[GEP5]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i64 [[GEP1_IDX]], [[GEP2_IDX]]
+; CHECK-NEXT: [[TMP2:%.*]] = add nuw i64 [[GEP3_IDX]], [[GEP4_IDX]]
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[TMP1]], [[TMP2]]
+; CHECK-NEXT: ret i1 [[CMP]]
+;
+ %gep1 = getelementptr nuw i32, ptr %base, i64 %idx1
+ %gep2 = getelementptr nuw i32, ptr %gep1, i64 %idx2
+ %gep3 = getelementptr nuw i32, ptr %base, i64 %idx3
+ %gep4 = getelementptr nuw i32, ptr %gep3, i64 %idx4
+ call void @use(ptr %gep2)
+ call void @use(ptr %gep4)
+ %cmp = icmp ult ptr %gep2, %gep4
+ ret i1 %cmp
+}
|
|
We hit a tricky regression when the code contains both pointer comparison and difference sharing the same operands. It is hard to reassociate operands and reuse the offset... |
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.
This extends #144065 to the general case of an icmp between two GEP chains that have a common base.