diff --git a/llvm/lib/Transforms/Scalar/LICM.cpp b/llvm/lib/Transforms/Scalar/LICM.cpp
--- a/llvm/lib/Transforms/Scalar/LICM.cpp
+++ b/llvm/lib/Transforms/Scalar/LICM.cpp
@@ -1176,8 +1176,37 @@
     if (isLoadInvariantInLoop(LI, DT, CurLoop))
       return true;
 
+    auto MU = cast<MemoryUse>(MSSA->getMemoryAccess(LI));
+
+    if (LI->hasMetadata(LLVMContext::MD_invariant_group)) {
+      if (!Flags.getIsSink()) {
+        // If hoisting, we only need to check that there is no store
+        // to the loaded pointer between the start of the loop, and
+        // the load (since all values must be the same).
+
+        MemoryAccess *Source;
+        // See declaration of SetLicmMssaOptCap for usage details.
+        if (Flags.tooManyClobberingCalls())
+          Source = MU->getDefiningAccess();
+        else {
+          Source = MSSA->getSkipSelfWalker()->getClobberingMemoryAccess(MU);
+          Flags.incrementClobberingCalls();
+        }
+
+        // This can be checked in two conditions:
+        // 1) if the memoryaccess is outside the loop
+        // 2) the earliest access is at the loop header.
+        // if the memory loaded is the phi node
+
+        return MSSA->isLiveOnEntryDef(Source) ||
+           !CurLoop->contains(Source->getBlock()) ||
+           (Source->getBlock() == CurLoop->getHeader() && isa<MemoryPhi>(Source));
+      }
+    }
+
+
     bool Invalidated = pointerInvalidatedByLoop(
-        MSSA, cast<MemoryUse>(MSSA->getMemoryAccess(LI)), CurLoop, I, Flags);
+        MSSA, MU, CurLoop, I, Flags);
     // Check loop-invariant address because this may also be a sinkable load
     // whose address is not necessarily loop-invariant.
     if (ORE && Invalidated && CurLoop->isLoopInvariant(LI->getPointerOperand()))
diff --git a/llvm/test/Transforms/LICM/invariant.group.ll b/llvm/test/Transforms/LICM/invariant.group.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Transforms/LICM/invariant.group.ll
@@ -0,0 +1,70 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=licm < %s -S | FileCheck %s
+; RUN: opt -aa-pipeline=basic-aa -passes='require<aa>,require<targetir>,require<scalar-evolution>,require<opt-remark-emit>,loop-mssa(licm)' < %s -S | FileCheck %s
+
+define void @test(ptr %arg, ptr %arg1) {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:  bb2:
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARG1:%.*]], align 4, !invariant.group !0
+; CHECK-NEXT:    br label [[BB5:%.*]]
+; CHECK:       bb5:
+; CHECK-NEXT:    [[TMP6:%.*]] = phi i64 [ 0, [[BB2:%.*]] ], [ [[TMP10:%.*]], [[BB5]] ]
+; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[ARG:%.*]], i64 [[TMP6]]
+; CHECK-NEXT:    store i32 [[TMP3]], ptr [[TMP7]], align 8
+; CHECK-NEXT:    [[TMP10]] = add nuw nsw i64 [[TMP6]], 1
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[TMP10]], 200
+; CHECK-NEXT:    br i1 [[TMP11]], label [[BB12:%.*]], label [[BB5]]
+; CHECK:       bb12:
+; CHECK-NEXT:    ret void
+;
+bb2:                                              ; preds = %bb
+  br label %bb5
+
+bb5:                                              ; preds = %bb5, %bb2
+  %tmp6 = phi i64 [ 0, %bb2 ], [ %tmp10, %bb5 ]
+  %tmp3 = load i32, i32* %arg1, align 4, !invariant.group !0
+  %tmp7 = getelementptr inbounds i32, ptr %arg, i64 %tmp6
+  store i32 %tmp3, ptr %tmp7, align 8
+  %tmp10 = add nuw nsw i64 %tmp6, 1
+  %tmp11 = icmp eq i64 %tmp10, 200
+  br i1 %tmp11, label %bb12, label %bb5
+
+bb12:                                             ; preds = %bb5, %bb
+  ret void
+}
+
+
+define void @test_fail(ptr %arg, ptr %arg1) {
+; CHECK-LABEL: @test_fail(
+; CHECK-NEXT:  bb2:
+; CHECK-NEXT:    br label [[BB5:%.*]]
+; CHECK:       bb5:
+; CHECK-NEXT:    [[TMP6:%.*]] = phi i64 [ 0, [[BB2:%.*]] ], [ [[TMP10:%.*]], [[BB5]] ]
+; CHECK-NEXT:    store i32 3, ptr [[ARG1:%.*]], align 4
+; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ARG1]], align 4, !invariant.group !0
+; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[ARG:%.*]], i64 [[TMP6]]
+; CHECK-NEXT:    store i32 [[TMP3]], ptr [[TMP7]], align 8
+; CHECK-NEXT:    [[TMP10]] = add nuw nsw i64 [[TMP6]], 1
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[TMP10]], 200
+; CHECK-NEXT:    br i1 [[TMP11]], label [[BB12:%.*]], label [[BB5]]
+; CHECK:       bb12:
+; CHECK-NEXT:    ret void
+;
+bb2:                                              ; preds = %bb
+  br label %bb5
+
+bb5:                                              ; preds = %bb5, %bb2
+  %tmp6 = phi i64 [ 0, %bb2 ], [ %tmp10, %bb5 ]
+  store i32 3, ptr %arg1
+  %tmp3 = load i32, i32* %arg1, align 4, !invariant.group !0
+  %tmp7 = getelementptr inbounds i32, ptr %arg, i64 %tmp6
+  store i32 %tmp3, ptr %tmp7, align 8
+  %tmp10 = add nuw nsw i64 %tmp6, 1
+  %tmp11 = icmp eq i64 %tmp10, 200
+  br i1 %tmp11, label %bb12, label %bb5
+
+bb12:                                             ; preds = %bb5, %bb
+  ret void
+}
+
+!0 = !{}