[llvm] [llvm, Pass, SCCP] Support constant structure in PhiNode (PR #163713)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Oct 15 23:48:16 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-transforms
Author: None (aokblast)
<details>
<summary>Changes</summary>
This patch adds support for constant propagation of individual structure members through Phi nodes. Each member is handled independently, allowing optimization opportunities when specific members become constant.
Also, update the testcase since we are able to optimize the call parameter now.
---
Full diff: https://github.com/llvm/llvm-project/pull/163713.diff
2 Files Affected:
- (modified) llvm/lib/Transforms/Utils/SCCPSolver.cpp (+53-29)
- (modified) llvm/test/Transforms/SCCP/constant-range-struct.ll (+51-9)
``````````diff
diff --git a/llvm/lib/Transforms/Utils/SCCPSolver.cpp b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
index 9693ae6b8ceb5..f030e2e1391f4 100644
--- a/llvm/lib/Transforms/Utils/SCCPSolver.cpp
+++ b/llvm/lib/Transforms/Utils/SCCPSolver.cpp
@@ -20,6 +20,7 @@
#include "llvm/Analysis/ValueLatticeUtils.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/ConstantRange.h"
+#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/InstVisitor.h"
#include "llvm/IR/Instructions.h"
@@ -1383,49 +1384,72 @@ bool SCCPInstVisitor::isEdgeFeasible(BasicBlock *From, BasicBlock *To) const {
// 7. If a conditional branch has a value that is overdefined, make all
// successors executable.
void SCCPInstVisitor::visitPHINode(PHINode &PN) {
- // If this PN returns a struct, just mark the result overdefined.
- // TODO: We could do a lot better than this if code actually uses this.
- if (PN.getType()->isStructTy())
- return (void)markOverdefined(&PN);
-
- if (getValueState(&PN).isOverdefined())
- return; // Quick exit
-
// Super-extra-high-degree PHI nodes are unlikely to ever be marked constant,
// and slow us down a lot. Just mark them overdefined.
if (PN.getNumIncomingValues() > 64)
return (void)markOverdefined(&PN);
- unsigned NumActiveIncoming = 0;
-
// Look at all of the executable operands of the PHI node. If any of them
// are overdefined, the PHI becomes overdefined as well. If they are all
// constant, and they agree with each other, the PHI becomes the identical
// constant. If they are constant and don't agree, the PHI is a constant
// range. If there are no executable operands, the PHI remains unknown.
- ValueLatticeElement PhiState = getValueState(&PN);
+ if (StructType *STy = dyn_cast<StructType>(PN.getType())) {
+ for (unsigned i = 0, e = STy->getNumElements(); i < e; ++i) {
+ ValueLatticeElement PNState = getStructValueState(&PN, i);
+ if (PNState.isOverdefined())
+ return;
+ }
+ } else {
+ ValueLatticeElement PhiState = getValueState(&PN);
+ if (PhiState.isOverdefined())
+ return;
+ }
+ std::vector<unsigned> FeasibleIncomingIndices;
for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) {
if (!isEdgeFeasible(PN.getIncomingBlock(i), PN.getParent()))
continue;
+ FeasibleIncomingIndices.push_back(i);
+ }
- ValueLatticeElement IV = getValueState(PN.getIncomingValue(i));
- PhiState.mergeIn(IV);
- NumActiveIncoming++;
- if (PhiState.isOverdefined())
- break;
- }
-
- // We allow up to 1 range extension per active incoming value and one
- // additional extension. Note that we manually adjust the number of range
- // extensions to match the number of active incoming values. This helps to
- // limit multiple extensions caused by the same incoming value, if other
- // incoming values are equal.
- mergeInValue(&PN, PhiState,
- ValueLatticeElement::MergeOptions().setMaxWidenSteps(
- NumActiveIncoming + 1));
- ValueLatticeElement &PhiStateRef = getValueState(&PN);
- PhiStateRef.setNumRangeExtensions(
- std::max(NumActiveIncoming, PhiStateRef.getNumRangeExtensions()));
+ if (StructType *STy = dyn_cast<StructType>(PN.getType())) {
+ for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
+ ValueLatticeElement PhiState = getStructValueState(&PN, i);
+ for (unsigned j : FeasibleIncomingIndices) {
+ ValueLatticeElement IV = getStructValueState(PN.getIncomingValue(j), i);
+ PhiState.mergeIn(IV);
+ if (PhiState.isOverdefined())
+ break;
+ }
+ mergeInValue(getStructValueState(&PN, i), &PN, PhiState,
+ ValueLatticeElement::MergeOptions().setMaxWidenSteps(
+ FeasibleIncomingIndices.size() + 1));
+ ValueLatticeElement &PhiStateRef = getStructValueState(&PN, i);
+ PhiStateRef.setNumRangeExtensions(
+ std::max((unsigned)FeasibleIncomingIndices.size(),
+ PhiStateRef.getNumRangeExtensions()));
+ }
+ } else {
+ ValueLatticeElement PhiState = getValueState(&PN);
+ for (unsigned i : FeasibleIncomingIndices) {
+ ValueLatticeElement IV = getValueState(PN.getIncomingValue(i));
+ PhiState.mergeIn(IV);
+ if (PhiState.isOverdefined())
+ break;
+ }
+ // We allow up to 1 range extension per active incoming value and one
+ // additional extension. Note that we manually adjust the number of range
+ // extensions to match the number of active incoming values. This helps to
+ // limit multiple extensions caused by the same incoming value, if other
+ // incoming values are equal.
+ mergeInValue(&PN, PhiState,
+ ValueLatticeElement::MergeOptions().setMaxWidenSteps(
+ FeasibleIncomingIndices.size() + 1));
+ ValueLatticeElement &PhiStateRef = getValueState(&PN);
+ PhiStateRef.setNumRangeExtensions(
+ std::max((unsigned)FeasibleIncomingIndices.size(),
+ PhiStateRef.getNumRangeExtensions()));
+ }
}
void SCCPInstVisitor::visitReturnInst(ReturnInst &I) {
diff --git a/llvm/test/Transforms/SCCP/constant-range-struct.ll b/llvm/test/Transforms/SCCP/constant-range-struct.ll
index 7a399df11fb13..ac7abfe6d3a56 100644
--- a/llvm/test/Transforms/SCCP/constant-range-struct.ll
+++ b/llvm/test/Transforms/SCCP/constant-range-struct.ll
@@ -39,14 +39,14 @@ define void @struct1_caller() {
; CHECK-NEXT: [[S:%.*]] = call { i64, i64 } @struct1()
; CHECK-NEXT: [[V1:%.*]] = extractvalue { i64, i64 } [[S]], 0
; CHECK-NEXT: [[V2:%.*]] = extractvalue { i64, i64 } [[S]], 1
-; CHECK-NEXT: [[T_1:%.*]] = icmp ne i64 [[V1]], 10
-; CHECK-NEXT: call void @use(i1 [[T_1]])
-; CHECK-NEXT: [[T_2:%.*]] = icmp ult i64 [[V1]], 100
-; CHECK-NEXT: call void @use(i1 [[T_2]])
-; CHECK-NEXT: [[T_3:%.*]] = icmp ne i64 [[V2]], 0
+; CHECK-NEXT: call void @use(i1 true)
+; CHECK-NEXT: call void @use(i1 true)
+; CHECK-NEXT: [[T_3:%.*]] = icmp eq i64 [[V1]], 20
; CHECK-NEXT: call void @use(i1 [[T_3]])
-; CHECK-NEXT: [[T_4:%.*]] = icmp ult i64 [[V2]], 301
-; CHECK-NEXT: call void @use(i1 [[T_4]])
+; CHECK-NEXT: call void @use(i1 true)
+; CHECK-NEXT: call void @use(i1 true)
+; CHECK-NEXT: [[T_6:%.*]] = icmp eq i64 [[V2]], 300
+; CHECK-NEXT: call void @use(i1 [[T_6]])
; CHECK-NEXT: ret void
;
%s = call {i64, i64} @struct1()
@@ -57,10 +57,14 @@ define void @struct1_caller() {
call void @use(i1 %t.1)
%t.2 = icmp ult i64 %v1, 100
call void @use(i1 %t.2)
- %t.3 = icmp ne i64 %v2, 0
+ %t.3 = icmp eq i64 %v1, 20
call void @use(i1 %t.3)
- %t.4 = icmp ult i64 %v2, 301
+ %t.4 = icmp ne i64 %v2, 0
call void @use(i1 %t.4)
+ %t.5 = icmp ult i64 %v2, 301
+ call void @use(i1 %t.5)
+ %t.6 = icmp eq i64 %v2, 300
+ call void @use(i1 %t.6)
ret void
}
@@ -153,3 +157,41 @@ define void @struct2_caller() {
ret void
}
+
+%"phi_type" = type {i64, i64}
+
+; Function Attrs: mustprogress
+define internal noundef %"phi_type" @test(i32 noundef %input) local_unnamed_addr readnone nounwind {
+; CHECK-LABEL: @test(
+; CHECK-NEXT: br label [[COND_TRUE_I:%.*]]
+; CHECK: cond.true.i:
+; CHECK-NEXT: br label [[COND_END_I:%.*]]
+; CHECK: cond.end.i:
+; CHECK-NEXT: ret [[PHI_TYPE:%.*]] poison
+;
+ %cmp.cond = icmp eq i32 %input, 1
+ br i1 %cmp.cond, label %cond.true.i, label %cond.false.i
+
+cond.true.i:
+ %r1.tmp = insertvalue %"phi_type" undef, i64 1, 0
+ %r1.tmp.2 = insertvalue %"phi_type" %r1.tmp, i64 2, 1
+ br label %cond.end.i
+
+cond.false.i:
+ %r2.tmp = insertvalue %"phi_type" undef, i64 3, 0
+ %r2.tmp.2 = insertvalue %"phi_type" %r2.tmp, i64 4, 1
+ br label %cond.end.i
+
+cond.end.i:
+ %retval = phi %"phi_type" [ %r1.tmp.2, %cond.true.i ], [ %r2.tmp.2, %cond.false.i ]
+ ret %"phi_type" %retval
+}
+
+define dso_local noundef %"phi_type" @test2() {
+; CHECK-LABEL: @test2(
+; CHECK-NEXT: [[CALL_1:%.*]] = tail call fastcc [[PHI_TYPE:%.*]] @[[TEST:[a-zA-Z0-9_$\"\\.-]*[a-zA-Z_$\"\\.-][a-zA-Z0-9_$\"\\.-]*]](i32 noundef 1)
+; CHECK-NEXT: ret [[PHI_TYPE]] { i64 1, i64 2 }
+;
+ %call.1 = tail call fastcc noundef %"phi_type" @test(i32 noundef 1)
+ ret %"phi_type" %call.1
+}
``````````
</details>
https://github.com/llvm/llvm-project/pull/163713
More information about the llvm-commits
mailing list