[llvm] [InstSimplify] Support ptrtoaddr in pointer subtraction fold (PR #162672)

[Nikita Popov via llvm-commits]

https://github.com/nikic created https://github.com/llvm/llvm-project/pull/162672

Add a new m_PtrToIntOrAddr() matcher which matches both ptrtoint and ptrtoaddr. Pointer arithmetic only works on the address bits, so supporting ptrtoaddr is always fine here.

>From eb7b9d596cb33411eeadf77eec1bc334bd37bdb3 Mon Sep 17 00:00:00 2001
From: Nikita Popov <npopov at redhat.com>
Date: Thu, 9 Oct 2025 16:50:42 +0200
Subject: [PATCH] [InstSimplify] Support ptrtoaddr in pointer subtraction fold

Add a new m_PtrToIntOrAddr() matcher which matches both ptrtoint
and ptrtoaddr. Pointer arithmetic only works on the address bits,
so supporting ptrtoaddr is always fine here.
---
 llvm/include/llvm/IR/PatternMatch.h           | 12 ++++++++++
 llvm/lib/Analysis/InstructionSimplify.cpp     |  4 +++-
 llvm/test/Transforms/InstCombine/ptrtoaddr.ll | 24 +++++++++++++++++++
 3 files changed, 39 insertions(+), 1 deletion(-)

diff --git a/llvm/include/llvm/IR/PatternMatch.h b/llvm/include/llvm/IR/PatternMatch.h
index 2e31fe5bb431e..79342777aba4b 100644
--- a/llvm/include/llvm/IR/PatternMatch.h
+++ b/llvm/include/llvm/IR/PatternMatch.h
@@ -2184,6 +2184,18 @@ inline PtrToIntSameSize_match<OpTy> m_PtrToIntSameSize(const DataLayout &DL,
   return PtrToIntSameSize_match<OpTy>(DL, Op);
 }
 
+/// Matches PtrToAddr.
+template <typename OpTy>
+inline CastOperator_match<OpTy, Instruction::PtrToAddr>
+m_PtrToAddr(const OpTy &Op) {
+  return CastOperator_match<OpTy, Instruction::PtrToAddr>(Op);
+}
+
+/// Matches PtrToInt or PtrToAddr.
+template <typename OpTy> inline auto m_PtrToIntOrAddr(const OpTy &Op) {
+  return m_CombineOr(m_PtrToInt(Op), m_PtrToAddr(Op));
+}
+
 /// Matches IntToPtr.
 template <typename OpTy>
 inline CastOperator_match<OpTy, Instruction::IntToPtr>
diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index d1977f0b0daf9..71050a4117d5a 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -853,10 +853,12 @@ static Value *simplifySubInst(Value *Op0, Value *Op1, bool IsNSW, bool IsNUW,
           return W;
 
   // Variations on GEP(base, I, ...) - GEP(base, i, ...) -> GEP(null, I-i, ...).
-  if (match(Op0, m_PtrToInt(m_Value(X))) && match(Op1, m_PtrToInt(m_Value(Y))))
+  if (match(Op0, m_PtrToIntOrAddr(m_Value(X))) &&
+      match(Op1, m_PtrToIntOrAddr(m_Value(Y)))) {
     if (Constant *Result = computePointerDifference(Q.DL, X, Y))
       return ConstantFoldIntegerCast(Result, Op0->getType(), /*IsSigned*/ true,
                                      Q.DL);
+  }
 
   // i1 sub -> xor.
   if (MaxRecurse && Op0->getType()->isIntOrIntVectorTy(1))
diff --git a/llvm/test/Transforms/InstCombine/ptrtoaddr.ll b/llvm/test/Transforms/InstCombine/ptrtoaddr.ll
index 5211fbd8b897a..1de8ea7a662c2 100644
--- a/llvm/test/Transforms/InstCombine/ptrtoaddr.ll
+++ b/llvm/test/Transforms/InstCombine/ptrtoaddr.ll
@@ -130,3 +130,27 @@ define i32 @ptrtoaddr_sub_consts_offset_addrsize() {
 ;
   ret i32 sub (i32 ptrtoaddr (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) @g.as1, i32 42) to i32), i32 ptrtoaddr (ptr addrspace(1) @g.as1 to i32))
 }
+
+define i64 @ptrtoaddr_sub_known_offset(ptr %p) {
+; CHECK-LABEL: define i64 @ptrtoaddr_sub_known_offset(
+; CHECK-SAME: ptr [[P:%.*]]) {
+; CHECK-NEXT:    ret i64 42
+;
+  %p2 = getelementptr inbounds i8, ptr %p, i64 42
+  %p.addr = ptrtoaddr ptr %p to i64
+  %p2.addr = ptrtoaddr ptr %p2 to i64
+  %sub = sub i64 %p2.addr, %p.addr
+  ret i64 %sub
+}
+
+define i32 @ptrtoaddr_sub_known_offset_addrsize(ptr addrspace(1) %p) {
+; CHECK-LABEL: define i32 @ptrtoaddr_sub_known_offset_addrsize(
+; CHECK-SAME: ptr addrspace(1) [[P:%.*]]) {
+; CHECK-NEXT:    ret i32 42
+;
+  %p2 = getelementptr inbounds i8, ptr addrspace(1) %p, i32 42
+  %p.addr = ptrtoaddr ptr addrspace(1) %p to i32
+  %p2.addr = ptrtoaddr ptr addrspace(1) %p2 to i32
+  %sub = sub i32 %p2.addr, %p.addr
+  ret i32 %sub
+}