diff --git a/llvm/lib/Linker/IRMover.cpp b/llvm/lib/Linker/IRMover.cpp
--- a/llvm/lib/Linker/IRMover.cpp
+++ b/llvm/lib/Linker/IRMover.cpp
@@ -20,6 +20,7 @@
 #include "llvm/IR/TypeFinder.h"
 #include "llvm/Object/ModuleSymbolTable.h"
 #include "llvm/Support/Error.h"
+#include "llvm/Support/Path.h"
 #include "llvm/Transforms/Utils/Cloning.h"
 #include <utility>
 using namespace llvm;
@@ -1443,20 +1444,34 @@
   if (DstM.getDataLayout().isDefault())
     DstM.setDataLayout(SrcM->getDataLayout());
 
-  if (SrcM->getDataLayout() != DstM.getDataLayout()) {
-    emitWarning("Linking two modules of different data layouts: '" +
-                SrcM->getModuleIdentifier() + "' is '" +
-                SrcM->getDataLayoutStr() + "' whereas '" +
-                DstM.getModuleIdentifier() + "' is '" +
-                DstM.getDataLayoutStr() + "'\n");
-  }
-
   // Copy the target triple from the source to dest if the dest's is empty.
   if (DstM.getTargetTriple().empty() && !SrcM->getTargetTriple().empty())
     DstM.setTargetTriple(SrcM->getTargetTriple());
 
   Triple SrcTriple(SrcM->getTargetTriple()), DstTriple(DstM.getTargetTriple());
 
+  // NVIDIA supplies libdevice bitcode without data layout set (pre-CUDA-11), or
+  // with the layout that is different from the one used by us (it does not
+  // include i128). Issuing a warning for that is not very helpful as there's
+  // not much the user can do about that.
+  bool SuppressDLWarning = false;
+  if (SrcTriple.isNVPTX()) {
+    std::string ModuleId = SrcM->getModuleIdentifier();
+    StringRef FileName = llvm::sys::path::filename(ModuleId);
+    SuppressDLWarning =
+        FileName.startswith("libdevice") && FileName.endswith(".10.bc") &&
+        (SrcM->getDataLayoutStr().empty() ||
+         SrcM->getDataLayoutStr() == "e-i64:64-v16:16-v32:32-n16:32:64");
+  }
+
+  if (!SuppressDLWarning && (SrcM->getDataLayout() != DstM.getDataLayout())) {
+    emitWarning("Linking two modules of different data layouts: '" +
+                SrcM->getModuleIdentifier() + "' is '" +
+                SrcM->getDataLayoutStr() + "' whereas '" +
+                DstM.getModuleIdentifier() + "' is '" +
+                DstM.getDataLayoutStr() + "'\n");
+  }
+
   if (!SrcM->getTargetTriple().empty()&&
       !SrcTriple.isCompatibleWith(DstTriple))
     emitWarning("Linking two modules of different target triples: '" +
diff --git a/llvm/lib/Support/Triple.cpp b/llvm/lib/Support/Triple.cpp
--- a/llvm/lib/Support/Triple.cpp
+++ b/llvm/lib/Support/Triple.cpp
@@ -16,6 +16,7 @@
 #include "llvm/Support/SwapByteOrder.h"
 #include "llvm/Support/TargetParser.h"
 #include "llvm/Support/VersionTuple.h"
+#include "llvm/Support/raw_ostream.h"
 #include <cassert>
 #include <cstring>
 using namespace llvm;
@@ -1639,6 +1640,25 @@
     return getArch() == Other.getArch() && getSubArch() == Other.getSubArch() &&
            getVendor() == Other.getVendor() && getOS() == Other.getOS();
 
+  if (isNVPTX()) {
+    // libdevice bitcode uses nvptx64-nvidia-gpulibs or just
+    // 'nvptx-unknown-unknown' triple (before CUDA-10.x) and is compatible with
+    // all NVPTX variants.
+    auto IsLibDevice = [](const Triple &T) {
+      if (T.getVendor() == Triple::NVIDIA && T.getOSName() == "gpulibs")
+        return true;
+      // Compare unknown-unknown as strings, so we would not consider something
+      // intentionally different, but unknown to LLVM, as compatible.
+      if (T.getVendorName() == "unknown" && T.getOSName() == "unknown")
+        return true;
+      return false;
+    };
+
+    // If one of the modules appears to be a libdevice, consider it to be
+    // compatible as long as it is an nvptx variant.
+    if (IsLibDevice(*this) || IsLibDevice(Other))
+      return Other.isNVPTX();
+  }
   return *this == Other;
 }