diff --git a/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp b/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
--- a/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
+++ b/openmp/libomptarget/plugins-nextgen/amdgpu/src/rtl.cpp
@@ -363,9 +363,10 @@
 /// generic kernel class.
 struct AMDGPUKernelTy : public GenericKernelTy {
   /// Create an AMDGPU kernel with a name and an execution mode.
-  AMDGPUKernelTy(const char *Name, OMPTgtExecModeFlags ExecutionMode)
+  AMDGPUKernelTy(const char *Name, OMPTgtExecModeFlags ExecutionMode,
+                 uint16_t CodeObjectVersion)
       : GenericKernelTy(Name, ExecutionMode),
-        ImplicitArgsSize(sizeof(utils::AMDGPUImplicitArgsTy)) {}
+        ImplicitArgsSize(utils::getImplicitArgsSize(CodeObjectVersion)) {}
 
   /// Initialize the AMDGPU kernel.
   Error initImpl(GenericDeviceTy &Device, DeviceImageTy &Image) override {
@@ -1652,9 +1653,17 @@
       return Plugin::error("Invalid execution mode %d for '%s'",
                            ExecModeGlobal.getValue(), KernelEntry.name);
 
+    uint16_t Cov = 0;
+    if (auto Err = GHandler.readCOVFromELFHeader(*this, Image, Cov)) {
+      return Plugin::error("Failed to read code object version from the ELF"
+                           "header for the kernel '%s'\n",
+                           KernelEntry.name);
+    }
+
     // Allocate and initialize the AMDGPU kernel.
     AMDGPUKernelTy *AMDKernel = Plugin::get().allocate<AMDGPUKernelTy>();
-    new (AMDKernel) AMDGPUKernelTy(KernelEntry.name, ExecModeGlobal.getValue());
+    new (AMDKernel)
+        AMDGPUKernelTy(KernelEntry.name, ExecModeGlobal.getValue(), Cov);
 
     return AMDKernel;
   }
@@ -2396,9 +2405,7 @@
     return Err;
 
   // Initialize implicit arguments.
-  utils::AMDGPUImplicitArgsTy *ImplArgs =
-      reinterpret_cast<utils::AMDGPUImplicitArgsTy *>(
-          static_cast<char *>(AllArgs) + KernelArgsSize);
+  uint8_t *ImplArgs = static_cast<uint8_t *>(AllArgs) + KernelArgsSize;
 
   // Initialize the implicit arguments to zero.
   std::memset(ImplArgs, 0, ImplicitArgsSize);
diff --git a/openmp/libomptarget/plugins-nextgen/amdgpu/utils/UtilitiesRTL.h b/openmp/libomptarget/plugins-nextgen/amdgpu/utils/UtilitiesRTL.h
--- a/openmp/libomptarget/plugins-nextgen/amdgpu/utils/UtilitiesRTL.h
+++ b/openmp/libomptarget/plugins-nextgen/amdgpu/utils/UtilitiesRTL.h
@@ -24,19 +24,9 @@
 namespace plugin {
 namespace utils {
 
-// The implicit arguments of AMDGPU kernels.
-struct AMDGPUImplicitArgsTy {
-  uint64_t OffsetX;
-  uint64_t OffsetY;
-  uint64_t OffsetZ;
-  uint64_t HostcallPtr;
-  uint64_t Unused0;
-  uint64_t Unused1;
-  uint64_t Unused2;
-};
-
-static_assert(sizeof(AMDGPUImplicitArgsTy) == 56,
-              "Unexpected size of implicit arguments");
+const uint16_t getImplicitArgsSize(uint16_t Version) {
+  return Version < ELF::ELFABIVERSION_AMDGPU_HSA_V5 ? 56 : 256;
+}
 
 /// Parse a TargetID to get processor arch and feature map.
 /// Returns processor subarch.
diff --git a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/GlobalHandler.h b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/GlobalHandler.h
--- a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/GlobalHandler.h
+++ b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/GlobalHandler.h
@@ -130,6 +130,9 @@
   Error readGlobalFromImage(GenericDeviceTy &Device, DeviceImageTy &Image,
                             const GlobalTy &HostGlobal);
 
+  Error readCOVFromELFHeader(GenericDeviceTy &Device, DeviceImageTy &Image,
+                             uint16_t &CodeObjectVersion);
+
   /// Get the address and size of a global from the device. Address is return in
   /// \p DeviceGlobal, the global name and expected size are passed in
   /// \p DeviceGlobal.
diff --git a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/GlobalHandler.cpp b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/GlobalHandler.cpp
--- a/openmp/libomptarget/plugins-nextgen/common/PluginInterface/GlobalHandler.cpp
+++ b/openmp/libomptarget/plugins-nextgen/common/PluginInterface/GlobalHandler.cpp
@@ -20,6 +20,7 @@
 using namespace omp;
 using namespace target;
 using namespace plugin;
+using namespace llvm::ELF;
 
 const ELF64LEObjectFile *
 GenericGlobalHandlerTy::getOrCreateELFObjectFile(const GenericDeviceTy &Device,
@@ -136,6 +137,27 @@
   return getGlobalMetadataFromELF(Image, **SymOrErr, **SecOrErr, ImageGlobal);
 }
 
+Error GenericGlobalHandlerTy::readCOVFromELFHeader(
+    GenericDeviceTy &Device, DeviceImageTy &Image,
+    uint16_t &CodeObjectVersion) {
+
+  // Get the ELF object file for the image. Notice the ELF object may already
+  // be created in previous calls, so we can reuse it.
+  const ELF64LEObjectFile *ELFObj = getOrCreateELFObjectFile(Device, Image);
+  if (!ELFObj)
+    return Plugin::error("Unable to create ELF object for image %p",
+                         Image.getStart());
+
+  // Get the header and lookup into e_ident.
+  CodeObjectVersion =
+      (uint8_t)ELFObj->getELFFile().getHeader().e_ident[EI_ABIVERSION];
+  if (CodeObjectVersion == 0)
+    return Plugin::error("Failed to get code object version from the ELF"
+                         "header\n");
+
+  return Plugin::success();
+}
+
 Error GenericGlobalHandlerTy::readGlobalFromImage(GenericDeviceTy &Device,
                                                   DeviceImageTy &Image,
                                                   const GlobalTy &HostGlobal) {
diff --git a/openmp/libomptarget/plugins/amdgpu/impl/get_elf_mach_gfx_name.h b/openmp/libomptarget/plugins/amdgpu/impl/get_elf_mach_gfx_name.h
--- a/openmp/libomptarget/plugins/amdgpu/impl/get_elf_mach_gfx_name.h
+++ b/openmp/libomptarget/plugins/amdgpu/impl/get_elf_mach_gfx_name.h
@@ -57,6 +57,4 @@
   COV5_HEAPV1_PTR_SIZE = 8
 };
 
-const uint16_t getImplicitArgsSize(uint16_t Version);
-
 #endif