Support memintrinsics

Author: gandhi56

llvm/lib/Transforms/IPO/GlobalOpt.cpp

@@ -211,6 +211,9 @@ CleanupPointerRootUsers(GlobalVariable *GV,
     } else if (auto *CE = dyn_cast<ConstantExpr>(U)) {
       if (isa<GEPOperator>(CE))
         append_range(Worklist, CE->users());
+    } else if (auto *MI = dyn_cast<MemIntrinsic>(U)) {
+      if (MI->getRawDest() == GV)
+        Writes.push_back(MI);
     }
   }
 

llvm/test/Transforms/GlobalOpt/cleanup-pointer-root-loads.ll

@@ -0,0 +1,42 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -passes=globalopt < %s -S | FileCheck %s
+
+@gv = internal unnamed_addr global [3 x ptr] zeroinitializer, align 16
+@gv2 = internal unnamed_addr global i32 0, align 4
+
+;; This test includes a load from @gv. No stores
+;; or memintrinsics with destination @gv should be removed.
+define i32 @main_with_load_from_b() local_unnamed_addr {
+; CHECK-LABEL: define i32 @main_with_load_from_b() local_unnamed_addr {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[E:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    store ptr [[E]], ptr getelementptr inbounds ([3 x ptr], ptr @gv, i64 0, i64 2), align 16
+; CHECK-NEXT:    [[LOAD_B:%.*]] = load ptr, ptr getelementptr inbounds ([3 x ptr], ptr @gv, i64 0, i64 2), align 16
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr @gv, ptr null, i64 8, i1 false)
+; CHECK-NEXT:    ret i32 0
+;
+entry:
+  %e = alloca i32, align 4
+  store ptr %e, ptr getelementptr inbounds ([3 x ptr], ptr @gv, i64 0, i64 2), align 16
+  %load.b = load ptr, ptr getelementptr inbounds ([3 x ptr], ptr @gv, i64 0, i64 2), align 16
+  call void @llvm.memcpy.p0i8.p0i8.i64(ptr getelementptr inbounds ([3 x ptr], ptr @gv, i64 0, i64 0), ptr null, i64 8, i1 false)
+  ret i32 0
+}
+
+;; This test includes a memcpy with @c as it's source and destination
+;; operands. CleanupPointerRootUsers is not called in this case.
+define i32 @main_with_load_store_c() local_unnamed_addr {
+; CHECK-LABEL: define i32 @main_with_load_store_c() local_unnamed_addr {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[E:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr @gv2, ptr @gv2, i64 4, i1 false)
+; CHECK-NEXT:    ret i32 0
+;
+entry:
+  %e = alloca i32, align 4
+  call void @llvm.memcpy.p0i8.p0i8.i64(ptr @gv2, ptr @gv2, i64 4, i1 false)
+  ret i32 0
+}
+
+declare void @llvm.memcpy.p0i8.p0i8.i64(ptr nocapture, ptr nocapture readonly, i64, i1) local_unnamed_addr
+declare void @llvm.memset.p0i8.i64(ptr nocapture, i8, i64, i1) local_unnamed_addr

llvm/test/Transforms/GlobalOpt/cleanup-pointer-root-stores.ll

@@ -4,6 +4,9 @@
 @a = internal unnamed_addr global i32 0, align 4
 @b = internal unnamed_addr global [3 x ptr] zeroinitializer, align 16
 
+;; This test is extracted from the issue reported in #64680, with an
+;; additional memcpy and a memset. Ensure all stores and memintrinsics with
+;; destination @b are removed as @b is dead.
 define i32 @main() local_unnamed_addr {
 ; CHECK-LABEL: define i32 @main() local_unnamed_addr {
 ; CHECK-NEXT:  entry:
@@ -52,11 +55,15 @@ if.end:                                           ; preds = %if.then, %for.body
   %inc = add nsw i32 %1, 1
   store i32 %inc, ptr @a, align 4
   %cmp = icmp slt i32 %1, 2
+  call void @llvm.memset.p0i8.i64(ptr getelementptr inbounds ([3 x ptr], ptr @b, i64 0, i64 0), i8 0, i64 8, i1 false)
   br i1 %cmp, label %for.body, label %for.end
 
 for.end:                                          ; preds = %if.end, %entry
+  call void @llvm.memcpy.p0i8.p0i8.i64(ptr getelementptr inbounds ([3 x ptr], ptr @b, i64 0, i64 0), ptr null, i64 8, i1 false)
   ret i32 0
 }
 
 declare void @bar20_() local_unnamed_addr
 declare void @foo() local_unnamed_addr
+declare void @llvm.memcpy.p0i8.p0i8.i64(ptr nocapture, ptr nocapture readonly, i64, i1) local_unnamed_addr
+declare void @llvm.memset.p0i8.i64(ptr nocapture, i8, i64, i1) local_unnamed_addr