[llvm] [MemCpyOpt] Use `dyn_cast` to fix assertion failure in `processMemCpyMemCpyDependence` (PR #98686)

[Yingwei Zheng via llvm-commits]

https://github.com/dtcxzyw updated https://github.com/llvm/llvm-project/pull/98686

>From 709de67f188e0c4a3389144abaebd25e56cd196c Mon Sep 17 00:00:00 2001
From: Yingwei Zheng <dtcxzyw2333 at gmail.com>
Date: Sat, 13 Jul 2024 03:33:07 +0800
Subject: [PATCH 1/2] [MemCpyOpt] Always create gep instructions in
 `processMemCpyMemCpyDependence`

---
 llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp |  5 +++--
 .../MemCpyOpt/memcpy-memcpy-offset.ll          | 18 ++++++++++++++++++
 2 files changed, 21 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 1c65219585e5a..c542084c1ad2f 100644
--- a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -1188,8 +1188,9 @@ bool MemCpyOptPass::processMemCpyMemCpyDependence(MemCpyInst *M,
     if (MDestOffset == MForwardOffset)
       CopySource = M->getDest();
     else {
-      NewCopySource = cast<Instruction>(Builder.CreateInBoundsPtrAdd(
-          CopySource, Builder.getInt64(MForwardOffset)));
+      NewCopySource = GetElementPtrInst::CreateInBounds(
+          Builder.getInt8Ty(), CopySource, Builder.getInt64(MForwardOffset), "",
+          Builder.GetInsertPoint());
       CopySource = NewCopySource;
     }
     // We need to update `MCopyLoc` if an offset exists.
diff --git a/llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll b/llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll
index 6abb0da827990..455a06045ab8d 100644
--- a/llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll
+++ b/llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll
@@ -197,6 +197,24 @@ define void @do_not_forward_offset_and_store(ptr %src, ptr %dest) {
   ret void
 }
 
+; Make sure we don't crash when the copy source is a constant.
+ at buf = external global [32 x i8]
+
+define void @pr98675(ptr noalias %p1, ptr noalias %p2) {
+; CHECK-LABEL: define void @pr98675(
+; CHECK-SAME: ptr noalias [[P1:%.*]], ptr noalias [[P2:%.*]]) {
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[P1]], ptr @buf, i64 26, i1 false)
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i8, ptr [[P1]], i64 10
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i8, ptr @buf, i64 10
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[P2]], ptr [[TMP1]], i64 1, i1 false)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p0.p0.i64(ptr %p1, ptr @buf, i64 26, i1 false)
+  %gep = getelementptr i8, ptr %p1, i64 10
+  call void @llvm.memmove.p0.p0.i64(ptr %p2, ptr %gep, i64 1, i1 false)
+  ret void
+}
+
 declare void @use(ptr)
 
 declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture, i64, i1)

>From b9a2d08a1a4c82c8d58bee712a114dffd7dd38f8 Mon Sep 17 00:00:00 2001
From: Yingwei Zheng <dtcxzyw2333 at gmail.com>
Date: Sat, 13 Jul 2024 03:55:17 +0800
Subject: [PATCH 2/2] [MemCpyOpt] Address review comments.

---
 llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp         | 7 +++----
 llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll | 3 +--
 2 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index c542084c1ad2f..4063762c88a2e 100644
--- a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -1188,10 +1188,9 @@ bool MemCpyOptPass::processMemCpyMemCpyDependence(MemCpyInst *M,
     if (MDestOffset == MForwardOffset)
       CopySource = M->getDest();
     else {
-      NewCopySource = GetElementPtrInst::CreateInBounds(
-          Builder.getInt8Ty(), CopySource, Builder.getInt64(MForwardOffset), "",
-          Builder.GetInsertPoint());
-      CopySource = NewCopySource;
+      CopySource = Builder.CreateInBoundsPtrAdd(
+          CopySource, Builder.getInt64(MForwardOffset));
+      NewCopySource = dyn_cast<Instruction>(CopySource);
     }
     // We need to update `MCopyLoc` if an offset exists.
     MCopyLoc = MCopyLoc.getWithNewPtr(CopySource);
diff --git a/llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll b/llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll
index 455a06045ab8d..da654438d7bd6 100644
--- a/llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll
+++ b/llvm/test/Transforms/MemCpyOpt/memcpy-memcpy-offset.ll
@@ -205,8 +205,7 @@ define void @pr98675(ptr noalias %p1, ptr noalias %p2) {
 ; CHECK-SAME: ptr noalias [[P1:%.*]], ptr noalias [[P2:%.*]]) {
 ; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[P1]], ptr @buf, i64 26, i1 false)
 ; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i8, ptr [[P1]], i64 10
-; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i8, ptr @buf, i64 10
-; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[P2]], ptr [[TMP1]], i64 1, i1 false)
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr [[P2]], ptr getelementptr inbounds (i8, ptr @buf, i64 10), i64 1, i1 false)
 ; CHECK-NEXT:    ret void
 ;
   call void @llvm.memcpy.p0.p0.i64(ptr %p1, ptr @buf, i64 26, i1 false)