diff --git a/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp b/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp
--- a/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp
+++ b/llvm/lib/Transforms/IPO/ArgumentPromotion.cpp
@@ -312,6 +312,12 @@
           AAMDNodes AAInfo;
           OrigLoad->getAAMetadata(AAInfo);
           newLoad->setAAMetadata(AAInfo);
+          // And other metadata.
+          newLoad->copyMetadata(
+              *OrigLoad,
+              {LLVMContext::MD_nontemporal, LLVMContext::MD_nonnull,
+               LLVMContext::MD_dereferenceable, LLVMContext::MD_align,
+               LLVMContext::MD_noundef, LLVMContext::MD_range});
 
           Args.push_back(newLoad);
           ArgAttrVec.push_back(AttributeSet());
diff --git a/llvm/test/Transforms/ArgumentPromotion/metadata.ll b/llvm/test/Transforms/ArgumentPromotion/metadata.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Transforms/ArgumentPromotion/metadata.ll
@@ -0,0 +1,44 @@
+; RUN: opt < %s -argpromotion -S | FileCheck %s
+
+define i32 @should_copy_range({ i32, i32 }* %x) {
+; CHECK-LABEL: define i32 @should_copy_range(
+; CHECK:         %x.idx = getelementptr { i32, i32 }, { i32, i32 }* %x, i64 0, i32 0
+; CHECK-NEXT:    %x.idx.val = load i32, i32* %x.idx, align 4, !range !0
+; CHECK-NEXT:    %1 = call i32 @f_load_range(i32 %x.idx.val)
+; CHECK-NEXT:    ret i32 %1
+;
+  %1 = call i32 @f_load_range({ i32, i32 }* %x)
+  ret i32 %1
+}
+
+define internal i32 @f_load_range({ i32, i32 }* %v) {
+; CHECK-LABEL: define internal i32 @f_load_range(
+; CHECK:         ret i32
+;
+  %1 = getelementptr inbounds { i32, i32 }, { i32, i32 }* %v, i64 0, i32 0
+  %2 = load i32, i32* %1, align 4, !range !0
+  ret i32 %2
+}
+
+define i32* @should_copy_nonnull({ i32*, i32* }* %x) {
+; CHECK-LABEL: define i32* @should_copy_nonnull(
+; CHECK:         %x.idx = getelementptr { i32*, i32* }, { i32*, i32* }* %x, i64 0, i32 0
+; CHECK-NEXT:    %x.idx.val = load i32*, i32** %x.idx, align 4, !nonnull !1
+; CHECK-NEXT:    %1 = call i32* @f_load_nonnull(i32* %x.idx.val)
+; CHECK-NEXT:    ret i32* %1
+;
+  %1 = call i32* @f_load_nonnull({ i32*, i32* }* %x)
+  ret i32* %1
+}
+
+define internal i32* @f_load_nonnull({ i32*, i32* }* %v) {
+; CHECK-LABEL: define internal i32* @f_load_nonnull(
+; CHECK:         ret i32*
+;
+  %1 = getelementptr inbounds { i32*, i32* }, { i32*, i32* }* %v, i64 0, i32 0
+  %2 = load i32*, i32** %1, align 4, !nonnull !1
+  ret i32* %2
+}
+
+!0 = !{i32 0, i32 4}
+!1 = !{}