diff --git a/Makefile b/Makefile
index 73f8f1f63..41702ae50 100644
--- a/Makefile
+++ b/Makefile
@@ -583,8 +583,8 @@ $(HWLOC_BASE):
 external_alpaka: $(ALPAKA_BASE)
 
 $(ALPAKA_BASE):
-	git clone git@github.com:alpaka-group/alpaka.git -b 0.9.0-rc1 $@
-	cd $@ && git checkout ebc1171feac21f1e21c49bcd9f053e7b01b584d0
+	git clone git@github.com:alpaka-group/alpaka.git -b develop $@
+	cd $@ && git checkout 879b95ffce2da499c9cc6e12d4cfd5545effa701
 
 # Kokkos
 external_kokkos: $(KOKKOS_LIB)
diff --git a/src/alpaka/AlpakaCore/CachingAllocator.h b/src/alpaka/AlpakaCore/CachingAllocator.h
index 5411ecdc5..20b26ebe0 100644
--- a/src/alpaka/AlpakaCore/CachingAllocator.h
+++ b/src/alpaka/AlpakaCore/CachingAllocator.h
@@ -90,6 +90,11 @@ namespace cms::alpakatools {
     using Event = alpaka::Event<Queue>;  // the events used to synchronise the operations
     using Buffer = alpaka::Buf<Device, std::byte, alpaka::DimInt<1u>, size_t>;
 
+    // The "memory device" type can either be the same as the "synchronisation device" type, or be the host CPU.
+    static_assert(std::is_same_v<Device, alpaka::Dev<Queue>> or std::is_same_v<Device, alpaka::DevCpu>,
+                  "The \"memory device\" type can either be the same as the \"synchronisation device\" type, or be the "
+                  "host CPU.");
+
     struct CachedBytes {
       size_t free = 0;       // total bytes freed and cached on this device
       size_t live = 0;       // total bytes currently in use oin this device
@@ -311,11 +316,24 @@ namespace cms::alpakatools {
       return false;
     }
 
+    Buffer allocateBuffer(size_t bytes, Queue const& queue) {
+      if constexpr (std::is_same_v<Device, alpaka::Dev<Queue>>) {
+        // allocate device memory
+        return alpaka::allocBuf<std::byte, size_t>(device_, bytes);
+      } else if constexpr (std::is_same_v<Device, alpaka::DevCpu>) {
+        // allocate pinned host memory
+        return alpaka::allocMappedBuf<std::byte, size_t>(device_, alpaka::getDev(queue), bytes);
+      } else {
+        // unsupported combination
+        static_assert(std::is_same_v<Device, alpaka::Dev<Queue>> or std::is_same_v<Device, alpaka::DevCpu>,
+                      "The \"memory device\" type can either be the same as the \"synchronisation device\" type, or be "
+                      "the host CPU.");
+      }
+    }
+
     void allocateNewBlock(BlockDescriptor& block) {
       try {
-        // FIXME simplify alpaka::Vec<alpaka::DimInt<1u>, size_t>{block.bytes} to block.bytes ?
-        block.buffer =
-            alpaka::allocBuf<std::byte, size_t>(device_, alpaka::Vec<alpaka::DimInt<1u>, size_t>{block.bytes});
+        block.buffer = allocateBuffer(block.bytes, *block.queue);
       } catch (std::runtime_error const& e) {
         // the allocation attempt failed: free all cached blocks on the device and retry
         if (debug_) {
@@ -329,25 +347,8 @@ namespace cms::alpakatools {
         freeAllCached();
 
         // throw an exception if it fails again
-        block.buffer =
-            alpaka::allocBuf<std::byte, size_t>(device_, alpaka::Vec<alpaka::DimInt<1u>, size_t>{block.bytes});
-      }
-
-      // for host memory, pin the newly allocated block
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
-      if (not cms::alpakatools::devices<alpaka::PltfCudaRt>.empty()) {
-        // it is possible to initialise the CUDA runtime and call cudaHostRegister
-        // only if the system has at least one supported GPU
-        alpaka::prepareForAsyncCopy(*block.buffer);
-      }
-#endif  // ALPAKA_ACC_GPU_CUDA_ENABLED
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-      if (not cms::alpakatools::devices<alpaka::PltfHipRt>.empty()) {
-        // it is possible to initialise the ROCm runtime and call hipHostRegister
-        // only if the system has at least one supported GPU
-        alpaka::prepareForAsyncCopy(*block.buffer);
+        block.buffer = allocateBuffer(block.bytes, *block.queue);
       }
-#endif  // ALPAKA_ACC_GPU_HIP_ENABLED
 
       // create a new event associated to the "synchronisation device"
       block.event = Event{block.device()};
diff --git a/src/alpaka/AlpakaCore/HostOnlyTask.h b/src/alpaka/AlpakaCore/HostOnlyTask.h
index dc9c5b7af..d010dc3eb 100644
--- a/src/alpaka/AlpakaCore/HostOnlyTask.h
+++ b/src/alpaka/AlpakaCore/HostOnlyTask.h
@@ -24,6 +24,8 @@ namespace alpaka {
     //! The CUDA async queue enqueue trait specialization for "safe tasks"
     template <>
     struct Enqueue<QueueCudaRtNonBlocking, HostOnlyTask> {
+      using TApi = ApiCudaRt;
+
       static void CUDART_CB callback(cudaStream_t /*queue*/, cudaError_t /*status*/, void* arg) {
         //ALPAKA_UNIFORM_CUDA_HIP_RT_CHECK(status);
         std::unique_ptr<HostOnlyTask> pTask(static_cast<HostOnlyTask*>(arg));
@@ -42,7 +44,9 @@ namespace alpaka {
     //! The HIP async queue enqueue trait specialization for "safe tasks"
     template <>
     struct Enqueue<QueueHipRtNonBlocking, HostOnlyTask> {
-      static void HIPRT_CB callback(hipStream_t /*queue*/, hipError_t /*status*/, void* arg) {
+      using TApi = ApiHipRt;
+
+      static void callback(hipStream_t /*queue*/, hipError_t /*status*/, void* arg) {
         //ALPAKA_UNIFORM_CUDA_HIP_RT_CHECK(status);
         std::unique_ptr<HostOnlyTask> pTask(static_cast<HostOnlyTask*>(arg));
         (*pTask)();
diff --git a/src/alpaka/AlpakaCore/alpaka/initialise.cc b/src/alpaka/AlpakaCore/alpaka/initialise.cc
new file mode 100644
index 000000000..690975af8
--- /dev/null
+++ b/src/alpaka/AlpakaCore/alpaka/initialise.cc
@@ -0,0 +1,32 @@
+#include <iostream>
+
+#include <alpaka/alpaka.hpp>
+
+#include "AlpakaCore/alpakaConfig.h"
+#include "AlpakaCore/alpakaDevices.h"
+#include "AlpakaCore/initialise.h"
+#include "Framework/demangle.h"
+
+namespace cms::alpakatools {
+
+  template <typename TPlatform>
+  void initialise() {
+    constexpr const char* suffix[] = {"devices.", "device:", "devices:"};
+
+    if (devices<TPlatform>.empty()) {
+      devices<TPlatform> = enumerate<TPlatform>();
+      auto size = devices<TPlatform>.size();
+      //std::cout << edm::demangle<TPlatform> << " platform succesfully initialised." << std::endl;
+      std::cout << "Found " << size << " " << suffix[size < 2 ? size : 2] << std::endl;
+      for (auto const& device : devices<TPlatform>) {
+        std::cout << "  - " << alpaka::getName(device) << std::endl;
+      }
+    } else {
+      //std::cout << edm::demangle<TPlatform> << " platform already initialised." << std::endl;
+    }
+  }
+
+  // explicit template instantiation definition
+  template void initialise<ALPAKA_ACCELERATOR_NAMESPACE::Platform>();
+
+}  // namespace cms::alpakatools
diff --git a/src/alpaka/AlpakaCore/alpakaConfig.h b/src/alpaka/AlpakaCore/alpakaConfig.h
index 447ced874..ae4d6c058 100644
--- a/src/alpaka/AlpakaCore/alpakaConfig.h
+++ b/src/alpaka/AlpakaCore/alpakaConfig.h
@@ -1,9 +1,7 @@
 #ifndef AlpakaCore_alpakaConfig_h
 #define AlpakaCore_alpakaConfig_h
 
-#include <type_traits>
-
-#include <alpaka/alpaka.hpp>
+#include "AlpakaCore/alpakaFwd.h"
 
 namespace alpaka_common {
 
@@ -32,7 +30,7 @@ namespace alpaka_common {
 
   // host types
   using DevHost = alpaka::DevCpu;
-  using PltfHost = alpaka::Pltf<DevHost>;
+  using PltfHost = alpaka::PltfCpu;
 
 }  // namespace alpaka_common
 
@@ -44,7 +42,7 @@ namespace alpaka_common {
 #define DEFINE_FWK_ALPAKA_EVENTSETUP_MODULE(name) \
   DEFINE_FWK_ALPAKA_EVENTSETUP_MODULE2(ALPAKA_ACCELERATOR_NAMESPACE::name)
 
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
 namespace alpaka_cuda_async {
   using namespace alpaka_common;
 
@@ -61,13 +59,13 @@ namespace alpaka_cuda_async {
 
 }  // namespace alpaka_cuda_async
 
-#endif  // ALPAKA_ACC_GPU_CUDA_ENABLED
+#endif  // ALPAKA_ACC_GPU_CUDA_PRESENT
 
 #ifdef ALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_cuda_async
 #endif  // ALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND
 
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
 namespace alpaka_rocm_async {
   using namespace alpaka_common;
 
@@ -84,13 +82,13 @@ namespace alpaka_rocm_async {
 
 }  // namespace alpaka_rocm_async
 
-#endif  // ALPAKA_ACC_GPU_HIP_ENABLED
+#endif  // ALPAKA_ACC_GPU_HIP_PRESENT
 
 #ifdef ALPAKA_ACC_GPU_HIP_ASYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_rocm_async
 #endif  // ALPAKA_ACC_GPU_HIP_ASYNC_BACKEND
 
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
 namespace alpaka_serial_sync {
   using namespace alpaka_common;
 
@@ -107,13 +105,13 @@ namespace alpaka_serial_sync {
 
 }  // namespace alpaka_serial_sync
 
-#endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
 
 #ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_serial_sync
 #endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND
 
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
 namespace alpaka_tbb_async {
   using namespace alpaka_common;
 
@@ -130,13 +128,13 @@ namespace alpaka_tbb_async {
 
 }  // namespace alpaka_tbb_async
 
-#endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
 
 #ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_tbb_async
 #endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND
 
-#ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_PRESENT
 namespace alpaka_omp2_async {
   using namespace alpaka_common;
 
@@ -153,7 +151,7 @@ namespace alpaka_omp2_async {
 
 }  // namespace alpaka_omp2_async
 
-#endif  // ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ENABLED
+#endif  // ALPAKA_ACC_CPU_B_OMP2_T_SEQ_PRESENT
 
 #ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ASYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_omp2_async
diff --git a/src/alpaka/AlpakaCore/alpakaFwd.h b/src/alpaka/AlpakaCore/alpakaFwd.h
new file mode 100644
index 000000000..2234e440b
--- /dev/null
+++ b/src/alpaka/AlpakaCore/alpakaFwd.h
@@ -0,0 +1,96 @@
+#ifndef AlpakaCore_alpakaFwd_h
+#define AlpakaCore_alpakaFwd_h
+
+#include <cstddef>
+#include <cstdint>
+#include <type_traits>
+
+/**
+ * This file forward declares specific types defined in Alpaka
+ * (depending on the backend-enabling macros) so that these types
+ * would be available throughout CMSSW without a direct dependence on
+ * Alpaka in order to avoid the constraints that would impose
+ * (primarily the device compiler)
+ *
+ * This is a little bit brittle, but let's see how it goes.
+ */
+namespace alpaka {
+
+  // miscellanea
+  template <std::size_t N>
+  using DimInt = std::integral_constant<std::size_t, N>;
+
+  template <typename TDim, typename TVal>
+  class Vec;
+
+  template <typename TDim, typename TIdx>
+  class WorkDivMembers;
+
+  // API
+  struct ApiCudaRt;
+  struct ApiHipRt;
+
+  // Platforms
+  class PltfCpu;
+  template <typename TApi>
+  class PltfUniformCudaHipRt;
+  using PltfCudaRt = PltfUniformCudaHipRt<ApiCudaRt>;
+  using PltfHipRt = PltfUniformCudaHipRt<ApiHipRt>;
+
+  // Devices
+  class DevCpu;
+  template <typename TApi>
+  class DevUniformCudaHipRt;
+  using DevCudaRt = DevUniformCudaHipRt<ApiCudaRt>;
+  using DevHipRt = DevUniformCudaHipRt<ApiHipRt>;
+
+  // Queues
+  template <typename TDev>
+  class QueueGenericThreadsBlocking;
+  using QueueCpuBlocking = QueueGenericThreadsBlocking<DevCpu>;
+
+  template <typename TDev>
+  class QueueGenericThreadsNonBlocking;
+  using QueueCpuNonBlocking = QueueGenericThreadsNonBlocking<DevCpu>;
+
+  namespace uniform_cuda_hip::detail {
+    template <typename TApi, bool TBlocking>
+    class QueueUniformCudaHipRt;
+  }
+  using QueueCudaRtBlocking = uniform_cuda_hip::detail::QueueUniformCudaHipRt<ApiCudaRt, true>;
+  using QueueCudaRtNonBlocking = uniform_cuda_hip::detail::QueueUniformCudaHipRt<ApiCudaRt, false>;
+  using QueueHipRtBlocking = uniform_cuda_hip::detail::QueueUniformCudaHipRt<ApiHipRt, true>;
+  using QueueHipRtNonBlocking = uniform_cuda_hip::detail::QueueUniformCudaHipRt<ApiHipRt, false>;
+
+  // Events
+  template <typename TDev>
+  class EventGenericThreads;
+  using EventCpu = EventGenericThreads<DevCpu>;
+
+  template <typename TApi>
+  class EventUniformCudaHipRt;
+  using EventCudaRt = EventUniformCudaHipRt<ApiCudaRt>;
+  using EventHipRt = EventUniformCudaHipRt<ApiHipRt>;
+
+  // Accelerators
+  template <typename TApi, typename TDim, typename TIdx>
+  class AccGpuUniformCudaHipRt;
+
+  template <typename TDim, typename TIdx>
+  using AccGpuCudaRt = AccGpuUniformCudaHipRt<ApiCudaRt, TDim, TIdx>;
+
+  template <typename TDim, typename TIdx>
+  using AccGpuHipRt = AccGpuUniformCudaHipRt<ApiHipRt, TDim, TIdx>;
+
+  template <typename TDim, typename TIdx>
+  class AccCpuSerial;
+
+  template <typename TDim, typename TIdx>
+  class AccCpuTbbBlocks;
+
+  template <typename TDim, typename TIdx>
+  class AccCpuOmp2Blocks;
+
+}  // namespace alpaka
+
+#endif  // AlpakaCore_alpakaFwd_h
diff --git a/src/alpaka/AlpakaCore/alpakaMemory.h b/src/alpaka/AlpakaCore/alpakaMemory.h
index 0ec3167c3..5ab53bd59 100644
--- a/src/alpaka/AlpakaCore/alpakaMemory.h
+++ b/src/alpaka/AlpakaCore/alpakaMemory.h
@@ -72,86 +72,59 @@ namespace cms::alpakatools {
 
   }  // namespace detail
 
-  // helper function for pinning memory buffers
-
-  template <typename TBuf>
-  void pin_buffer(TBuf& buffer) {
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
-    if (not cms::alpakatools::devices<alpaka::PltfCudaRt>.empty()) {
-      // it is possible to initialise the CUDA runtime and call cudaHostRegister
-      // only if the system has at least one supported GPU
-      alpaka::prepareForAsyncCopy(buffer);
-    }
-#endif  // ALPAKA_ACC_GPU_CUDA_ENABLED
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-    if (not cms::alpakatools::devices<alpaka::PltfHipRt>.empty()) {
-      // it is possible to initialise the ROCm runtime and call hipHostRegister
-      // only if the system has at least one supported GPU
-      alpaka::prepareForAsyncCopy(buffer);
-    }
-#endif  // ALPAKA_ACC_GPU_HIP_ENABLED
-  }
-
   // scalar and 1-dimensional host buffers
 
   template <typename T>
   using host_buffer = typename detail::buffer_type<DevHost, T>::type;
 
-  // non-cached, scalar and 1-dimensional host buffers
-  // the memory is pinned explicitly
+  // non-cached, non-pinned, scalar and 1-dimensional host buffers
 
   template <typename T>
   std::enable_if_t<not std::is_array_v<T>, host_buffer<T>> make_host_buffer() {
-    auto buffer = alpaka::allocBuf<T, Idx>(host, Scalar{});
-    pin_buffer(buffer);
-    return buffer;
+    return alpaka::allocBuf<T, Idx>(host, Scalar{});
   }
 
   template <typename T>
   std::enable_if_t<cms::is_unbounded_array_v<T> and not std::is_array_v<std::remove_extent_t<T>>, host_buffer<T>>
   make_host_buffer(Extent extent) {
-    auto buffer = alpaka::allocBuf<std::remove_extent_t<T>, Idx>(host, Vec1D{extent});
-    pin_buffer(buffer);
-    return buffer;
+    return alpaka::allocBuf<std::remove_extent_t<T>, Idx>(host, Vec1D{extent});
   }
 
   template <typename T>
   std::enable_if_t<cms::is_bounded_array_v<T> and not std::is_array_v<std::remove_extent_t<T>>, host_buffer<T>>
   make_host_buffer() {
-    auto buffer = alpaka::allocBuf<std::remove_extent_t<T>, Idx>(host, Vec1D{std::extent_v<T>});
-    pin_buffer(buffer);
-    return buffer;
+    return alpaka::allocBuf<std::remove_extent_t<T>, Idx>(host, Vec1D{std::extent_v<T>});
   }
 
-  // potentially cached, scalar and 1-dimensional host buffers, associated to a work queue
-  // the memory is pinned by the caching allocator, or explicitly if it is not used
+  // potentially cached, pinned, scalar and 1-dimensional host buffers, associated to a work queue
+  // the memory is pinned according to the device associated to the queue
 
   template <typename T, typename TQueue>
-  std::enable_if_t<not std::is_array_v<T>, host_buffer<T>> make_host_buffer(TQueue const& queue [[maybe_unused]]) {
+  std::enable_if_t<not std::is_array_v<T>, host_buffer<T>> make_host_buffer(TQueue const& queue) {
     if constexpr (allocator_policy<alpaka::Dev<TQueue>> == AllocatorPolicy::Caching) {
       return allocCachedBuf<T, Idx>(host, queue, Scalar{});
     } else {
-      return make_host_buffer<T>();
+      return alpaka::allocMappedBuf<T, Idx>(host, alpaka::getDev(queue), Scalar{});
     }
   }
 
   template <typename T, typename TQueue>
   std::enable_if_t<cms::is_unbounded_array_v<T> and not std::is_array_v<std::remove_extent_t<T>>, host_buffer<T>>
-  make_host_buffer(TQueue const& queue [[maybe_unused]], Extent extent) {
+  make_host_buffer(TQueue const& queue, Extent extent) {
     if constexpr (allocator_policy<alpaka::Dev<TQueue>> == AllocatorPolicy::Caching) {
       return allocCachedBuf<std::remove_extent_t<T>, Idx>(host, queue, Vec1D{extent});
     } else {
-      return make_host_buffer<T>(extent);
+      return alpaka::allocMappedBuf<std::remove_extent_t<T>, Idx>(host, alpaka::getDev(queue), Vec1D{extent});
     }
   }
 
   template <typename T, typename TQueue>
   std::enable_if_t<cms::is_bounded_array_v<T> and not std::is_array_v<std::remove_extent_t<T>>, host_buffer<T>>
-  make_host_buffer(TQueue const& queue [[maybe_unused]]) {
+  make_host_buffer(TQueue const& queue) {
     if constexpr (allocator_policy<alpaka::Dev<TQueue>> == AllocatorPolicy::Caching) {
       return allocCachedBuf<std::remove_extent_t<T>, Idx>(host, queue, Vec1D{std::extent_v<T>});
     } else {
-      return make_host_buffer<T>();
+      return alpaka::allocMappedBuf<std::remove_extent_t<T>, Idx>(host, alpaka::getDev(queue), Vec1D{std::extent_v<T>});
     }
   }
 
diff --git a/src/alpaka/AlpakaCore/backend.h b/src/alpaka/AlpakaCore/backend.h
index 387154013..8d58953cc 100644
--- a/src/alpaka/AlpakaCore/backend.h
+++ b/src/alpaka/AlpakaCore/backend.h
@@ -3,4 +3,15 @@
 
 enum class Backend { SERIAL, TBB, CUDA, HIP };
 
+inline std::string const& name(Backend backend) {
+  static const std::string names[] = {"serial_sync", "tbb_async", "cuda_async", "rocm_async"};
+  return names[static_cast<int>(backend)];
+}
+
+template <typename T>
+inline T& operator<<(T& out, Backend backend) {
+  out << name(backend);
+  return out;
+}
+
 #endif  // AlpakaCore_backend_h
diff --git a/src/alpaka/AlpakaCore/initialise.h b/src/alpaka/AlpakaCore/initialise.h
index 98ba966a1..1114ff5f9 100644
--- a/src/alpaka/AlpakaCore/initialise.h
+++ b/src/alpaka/AlpakaCore/initialise.h
@@ -1,31 +1,26 @@
 #ifndef AlpakaCore_initialise_h
 #define AlpakaCore_initialise_h
 
-#include <iostream>
-
-#include <alpaka/alpaka.hpp>
-
-#include "AlpakaCore/alpakaDevices.h"
-#include "Framework/demangle.h"
+#include "AlpakaCore/alpakaConfig.h"
 
 namespace cms::alpakatools {
 
   template <typename TPlatform>
-  void initialise() {
-    constexpr const char* suffix[] = {"devices.", "device:", "devices:"};
-
-    if (devices<TPlatform>.empty()) {
-      devices<TPlatform> = enumerate<TPlatform>();
-      auto size = devices<TPlatform>.size();
-      //std::cout << edm::demangle<TPlatform> << " platform succesfully initialised." << std::endl;
-      std::cout << "Found " << size << " " << suffix[size < 2 ? size : 2] << std::endl;
-      for (auto const& device : devices<TPlatform>) {
-        std::cout << "  - " << alpaka::getName(device) << std::endl;
-      }
-    } else {
-      //std::cout << edm::demangle<TPlatform> << " platform already initialised." << std::endl;
-    }
-  }
+  void initialise();
+
+  // explicit template instantiation declaration
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
+  extern template void initialise<alpaka_serial_sync::Platform>();
+#endif
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
+  extern template void initialise<alpaka_tbb_async::Platform>();
+#endif
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
+  extern template void initialise<alpaka_cuda_async::Platform>();
+#endif
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
+  extern template void initialise<alpaka_rocm_async::Platform>();
+#endif
 
 }  // namespace cms::alpakatools
 
diff --git a/src/alpaka/Makefile b/src/alpaka/Makefile
index bc56cab44..2cc62e515 100644
--- a/src/alpaka/Makefile
+++ b/src/alpaka/Makefile
@@ -31,12 +31,12 @@ EXE_DEP := $(EXE_OBJ:$.o=$.d)
 
 LIBNAMES := $(filter-out plugin-% bin test Makefile% plugins.txt%,$(wildcard *))
 PLUGINNAMES := $(patsubst plugin-%,%,$(filter plugin-%,$(wildcard *)))
-MY_CXXFLAGS := -I$(TARGET_DIR) -DSRC_DIR=$(TARGET_DIR) -DLIB_DIR=$(LIB_DIR)/$(TARGET_NAME) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED -DALPAKA_HOST_ONLY
+MY_CXXFLAGS := -I$(TARGET_DIR) -DLIB_DIR=$(LIB_DIR)/$(TARGET_NAME) -DALPAKA_HOST_ONLY -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT -DALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
 ifdef CUDA_BASE
-MY_CXXFLAGS += -DALPAKA_ACC_GPU_CUDA_ENABLED -DALPAKA_ACC_GPU_CUDA_ONLY_MODE
+MY_CXXFLAGS += -DALPAKA_ACC_GPU_CUDA_PRESENT -DALPAKA_ACC_GPU_CUDA_ONLY_MODE
 endif
 ifdef ROCM_BASE
-MY_CXXFLAGS += -DALPAKA_ACC_GPU_HIP_ENABLED -DALPAKA_ACC_GPU_HIP_ONLY_MODE
+MY_CXXFLAGS += -DALPAKA_ACC_GPU_HIP_PRESENT -DALPAKA_ACC_GPU_HIP_ONLY_MODE
 endif
 MY_LDFLAGS := -ldl -Wl,-rpath,$(LIB_DIR)/$(TARGET_NAME)
 LIB_LDFLAGS := -L$(LIB_DIR)/$(TARGET_NAME)
@@ -182,8 +182,12 @@ $(foreach test,$(TESTS_ROCM_EXE),$(eval $(call RUNTEST_template,$(test),amdgpu))
 -include $(ALL_DEPENDS)
 
 # Build targets
-$(TARGET): $(EXE_OBJ) $(LIBS) $(PLUGINS) | $(TESTS_EXE)
-	$(CXX) $(EXE_OBJ) $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+$(LIB_DIR)/$(TARGET_NAME)/plugins.txt: $(PLUGINS)
+	nm -A -C -D -P --defined-only $(PLUGINS) | sed -n -e"s#$(LIB_DIR)/$(TARGET_NAME)/\(plugin\w\+\.so\): typeinfo for edm::\(PluginFactory\|ESPluginFactory\)::impl::Maker<\([A-Za-z0-9_:]\+\)> V .* .*#\3 \1#p" | sort > $@
+
+$(TARGET): $(EXE_OBJ) $(LIBS) $(PLUGINS) $(LIB_DIR)/$(TARGET_NAME)/plugins.txt | $(TESTS_EXE)
+	# Link all libraries, also the "portable" ones
+	$(CXX) $(EXE_OBJ) $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_SERIAL_LDFLAGS) $($(lib)_TBB_LDFLAGS) $($(lib)_CUDA_LDFLAGS) $($(lib)_ROCM_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 
 define BUILD_template
 $(OBJ_DIR)/$(2)/%.cc.o: $(SRC_DIR)/$(2)/%.cc
@@ -219,7 +223,7 @@ $$($(1)_ROCM_LIB): $$($(1)_ROCM_OBJ) $$(foreach dep,$(EXTERNAL_DEPENDS_H),$$($$(
 # Portable code, for serial backend
 $(OBJ_DIR)/$(2)/alpaka/%.cc.serial.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 	@[ -d $$(@D) ] || mkdir -p $$(@D)
-	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
+	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
 	@cp $(OBJ_DIR)/$(2)/alpaka/$$*.cc.serial.d $(OBJ_DIR)/$(2)/alpaka/$$*.cc.serial.d.tmp; \
 	  sed 's#\($(2)/alpaka/$$*\)\.o[ :]*#\1.o \1.d : #g' < $(OBJ_DIR)/$(2)/alpaka/$$*.cc.serial.d.tmp > $(OBJ_DIR)/$(2)/alpaka/$$*.cc.serial.d; \
 	  sed -e 's/#.*//' -e 's/^[^:]*: *//' -e 's/ *\\$$$$//' \
@@ -229,7 +233,7 @@ $(OBJ_DIR)/$(2)/alpaka/%.cc.serial.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 # Portable code, for TBB backend
 $(OBJ_DIR)/$(2)/alpaka/%.cc.tbb.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 	@[ -d $$(@D) ] || mkdir -p $$(@D)
-	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
+	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
 	@cp $(OBJ_DIR)/$(2)/alpaka/$$*.cc.tbb.d $(OBJ_DIR)/$(2)/alpaka/$$*.cc.tbb.d.tmp; \
 	  sed 's#\($(2)/alpaka/$$*\)\.o[ :]*#\1.o \1.d : #g' < $(OBJ_DIR)/$(2)/alpaka/$$*.cc.tbb.d.tmp > $(OBJ_DIR)/$(2)/alpaka/$$*.cc.tbb.d; \
 	  sed -e 's/#.*//' -e 's/^[^:]*: *//' -e 's/ *\\$$$$//' \
@@ -240,7 +244,7 @@ $(OBJ_DIR)/$(2)/alpaka/%.cc.tbb.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 ifdef CUDA_BASE
 $(OBJ_DIR)/$(2)/alpaka/%.cc.cuda.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 	@[ -d $$(@D) ] || mkdir -p $$(@D)
-	$(CUDA_NVCC) -x cu $(CUDA_CUFLAGS) $(CUDA_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND -UALPAKA_HOST_ONLY $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_NVCC_CXXFLAGS)) -c $$< -o $$@ -MMD
+	$(CUDA_NVCC) -x cu $(CUDA_CUFLAGS) $(CUDA_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_CUDA_ENABLED -DALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND -UALPAKA_HOST_ONLY $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_NVCC_CXXFLAGS)) -c $$< -o $$@ -MMD
 
 $$($(1)_CUDADLINK): $$($(1)_CUOBJ)
 	$(CUDA_NVCC) $(CUDA_DLINKFLAGS) $(CUDA_LDFLAGS) $$($(1)_CUOBJ) -o $$@
@@ -250,7 +254,7 @@ endif
 ifdef ROCM_BASE
 $(OBJ_DIR)/$(2)/alpaka/%.cc.rocm.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 	@[ -d $$(@D) ] || mkdir -p $$(@D)
-	$(ROCM_HIPCC) $(HIPCC_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_HIP_ASYNC_BACKEND -UALPAKA_HOST_ONLY $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
+	$(ROCM_HIPCC) $(HIPCC_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_HIP_ENABLED -DALPAKA_ACC_GPU_HIP_ASYNC_BACKEND -UALPAKA_HOST_ONLY $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
 endif
 
 endef
@@ -271,7 +275,7 @@ $(OBJ_DIR)/$(TARGET_NAME)/bin/%.cc.o: $(SRC_DIR)/$(TARGET_NAME)/bin/%.cc
 # Serial backend
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.serial.o: $(SRC_DIR)/$(TARGET_NAME)/test/alpaka/%.cc
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
+	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
 	@cp $(@D)/$*.cc.serial.d $(@D)/$*.cc.serial.d.tmp; \
 	  sed 's#\($(TARGET_NAME)/$*\)\.o[ :]*#\1.o \1.d : #g' < $(@D)/$*.cc.serial.d.tmp > $(@D)/$*.cc.serial.d; \
 	  sed -e 's/#.*//' -e 's/^[^:]*: *//' -e 's/ *\\$$//' \
@@ -280,12 +284,12 @@ $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.serial.o: $(SRC_DIR)/$(TARGET_NAME)/t
 
 $(TEST_DIR)/$(TARGET_NAME)/%.serial: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.serial.o | $(LIBS)
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_SERIAL_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 
 # TBB backend
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.tbb.o: $(SRC_DIR)/$(TARGET_NAME)/test/alpaka/%.cc
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
+	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
 	@cp $(@D)/$*.cc.tbb.d $(@D)/$*.cc.tbb.d.tmp; \
 	  sed 's#\($(TARGET_NAME)/$*\)\.o[ :]*#\1.o \1.d : #g' < $(@D)/$*.cc.tbb.d.tmp > $(@D)/$*.cc.tbb.d; \
 	  sed -e 's/#.*//' -e 's/^[^:]*: *//' -e 's/ *\\$$//' \
@@ -294,29 +298,29 @@ $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.tbb.o: $(SRC_DIR)/$(TARGET_NAME)/test
 
 $(TEST_DIR)/$(TARGET_NAME)/%.tbb: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.tbb.o | $(LIBS)
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_TBB_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 
 # CUDA backend
 ifdef CUDA_BASE
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.cuda.o: $(SRC_DIR)/$(TARGET_NAME)/test/alpaka/%.cc
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CUDA_NVCC) -x cu $(CUDA_CUFLAGS) $(CUDA_CXXFLAGS) $(CUDA_TEST_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND -UALPAKA_HOST_ONLY $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_NVCC_CXXFLAGS)) -c $< -o $@ -MMD
+	$(CUDA_NVCC) -x cu $(CUDA_CUFLAGS) $(CUDA_CXXFLAGS) $(CUDA_TEST_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_CUDA_ENABLED -DALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND -UALPAKA_HOST_ONLY $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_NVCC_CXXFLAGS)) -c $< -o $@ -MMD
 
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cudadlink.o: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.o
 	$(CUDA_NVCC) $(CUDA_DLINKFLAGS) $(CUDA_LDFLAGS) $< -o $@
 
 $(TEST_DIR)/$(TARGET_NAME)/%.cuda: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.cuda.o $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.cuda.cudadlink.o | $(LIBS)
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_CUDA_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 endif
 
 # ROCm backend
 ifdef ROCM_BASE
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.rocm.o: $(SRC_DIR)/$(TARGET_NAME)/test/alpaka/%.cc
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(ROCM_HIPCC) $(HIPCC_CXXFLAGS) $(ROCM_TEST_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_HIP_ASYNC_BACKEND -UALPAKA_HOST_ONLY $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
+	$(ROCM_HIPCC) $(HIPCC_CXXFLAGS) $(ROCM_TEST_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_HIP_ENABLED -DALPAKA_ACC_GPU_HIP_ASYNC_BACKEND -UALPAKA_HOST_ONLY $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
 
 $(TEST_DIR)/$(TARGET_NAME)/%.rocm: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.rocm.o | $(LIBS)
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_ROCM_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 endif
diff --git a/src/alpaka/bin/EventProcessor.cc b/src/alpaka/bin/EventProcessor.cc
index 3d7a3546f..7e84f5114 100644
--- a/src/alpaka/bin/EventProcessor.cc
+++ b/src/alpaka/bin/EventProcessor.cc
@@ -1,3 +1,4 @@
+#include <cmath>
 #include <exception>
 #include <filesystem>
 #include <string>
@@ -13,7 +14,7 @@ namespace edm {
   EventProcessor::EventProcessor(int maxEvents,
                                  int runForMinutes,
                                  int numberOfStreams,
-                                 std::vector<std::string> const& path,
+                                 Alternatives alternatives,
                                  std::vector<std::string> const& esproducers,
                                  std::filesystem::path const& datadir,
                                  bool validation)
@@ -24,9 +25,23 @@ namespace edm {
       esp->produce(eventSetup_);
     }
 
+    // normalise the total weight to the number of streams
+    float total = 0.;
+    for (auto const& alternative : alternatives) {
+      total += alternative.weight;
+    }
     //schedules_.reserve(numberOfStreams);
-    for (int i = 0; i < numberOfStreams; ++i) {
-      schedules_.emplace_back(registry_, pluginManager_, &source_, &eventSetup_, i, path);
+    float cumulative = 0.;
+    int lower_range = 0;
+    int upper_range = 0;
+    for (auto& alternative : alternatives) {
+      cumulative += alternative.weight;
+      lower_range = upper_range;
+      upper_range = static_cast<int>(std::round(cumulative * numberOfStreams / total));
+      for (int i = lower_range; i < upper_range; ++i) {
+        schedules_.emplace_back(registry_, pluginManager_, &source_, &eventSetup_, i, alternative.path);
+      }
+      streamsPerBackend_.emplace_back(alternative.backend, upper_range - lower_range);
     }
   }
 
diff --git a/src/alpaka/bin/EventProcessor.h b/src/alpaka/bin/EventProcessor.h
index 5fb20f0b9..e1ca9fe01 100644
--- a/src/alpaka/bin/EventProcessor.h
+++ b/src/alpaka/bin/EventProcessor.h
@@ -5,6 +5,7 @@
 #include <string>
 #include <vector>
 
+#include "AlpakaCore/backend.h"
 #include "Framework/EventSetup.h"
 
 #include "PluginManager.h"
@@ -12,18 +13,31 @@
 #include "Source.h"
 
 namespace edm {
+  struct Alternative {
+    Alternative() = default;
+    Alternative(Backend backend, float weight, std::vector<std::string> path)
+        : backend{backend}, weight{weight}, path{std::move(path)} {}
+
+    Backend backend;
+    float weight;
+    std::vector<std::string> path;
+  };
+
+  using Alternatives = std::vector<Alternative>;
+
   class EventProcessor {
   public:
     explicit EventProcessor(int maxEvents,
                             int runForMinutes,
                             int numberOfStreams,
-                            std::vector<std::string> const& path,
+                            Alternatives alternatives,
                             std::vector<std::string> const& esproducers,
                             std::filesystem::path const& datadir,
                             bool validation);
 
     int maxEvents() const { return source_.maxEvents(); }
     int processedEvents() const { return source_.processedEvents(); }
+    std::vector<std::pair<Backend, int>> const& backends() const { return streamsPerBackend_; }
 
     void runToCompletion();
 
@@ -35,6 +49,7 @@ namespace edm {
     Source source_;
     EventSetup eventSetup_;
     std::vector<StreamSchedule> schedules_;
+    std::vector<std::pair<Backend, int>> streamsPerBackend_;
   };
 }  // namespace edm
 
diff --git a/src/alpaka/bin/PluginManager.cc b/src/alpaka/bin/PluginManager.cc
index 0fd46bbff..d1fa1d0f7 100644
--- a/src/alpaka/bin/PluginManager.cc
+++ b/src/alpaka/bin/PluginManager.cc
@@ -6,9 +6,6 @@
 
 #include "PluginManager.h"
 
-#ifndef SRC_DIR
-#error "SRC_DIR undefined"
-#endif
 #ifndef LIB_DIR
 #error "LIB_DIR undefined"
 #endif
@@ -18,7 +15,7 @@
 
 namespace edmplugin {
   PluginManager::PluginManager() {
-    std::ifstream pluginMap(STR(SRC_DIR) "/plugins.txt");
+    std::ifstream pluginMap(STR(LIB_DIR) "/plugins.txt");
     std::string plugin, library;
     while (pluginMap >> plugin >> library) {
       //std::cout << "plugin " << plugin << " in " << library << std::endl;
diff --git a/src/alpaka/bin/main.cc b/src/alpaka/bin/main.cc
index 6e863d8f3..9c997f4af 100644
--- a/src/alpaka/bin/main.cc
+++ b/src/alpaka/bin/main.cc
@@ -6,6 +6,7 @@
 #include <iomanip>
 #include <ios>
 #include <iostream>
+#include <unordered_map>
 #include <stdexcept>
 #include <string>
 #include <utility>
@@ -25,31 +26,31 @@ namespace {
   void print_help(std::string const& name) {
     std::cout
         << name << ": "
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
         << "[--serial] "
 #endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
         << "[--tbb] "
 #endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
         << "[--cuda] "
 #endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
         << "[--hip] "
 #endif
         << "[--numberOfThreads NT] [--numberOfStreams NS] [--maxEvents ME] [--data PATH] "
-           "[--transfer] [--validation]\n\n"
+           "[--transfer] [--validation] [--histogram]\n\n"
         << "Options\n"
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
         << " --serial            Use CPU Serial backend\n"
 #endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
         << " --tbb               Use CPU TBB backend\n"
 #endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
         << " --cuda              Use CUDA backend\n"
 #endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
         << " --hip               Use ROCm/HIP backend\n"
 #endif
         << " --numberOfThreads   Number of threads to use (default 1, use 0 to use all CPU cores)\n"
@@ -64,13 +65,62 @@ namespace {
         << " --empty             Ignore all producers (for testing only)\n"
         << std::endl;
   }
-
 }  // namespace
 
+bool getOptionalArgument(std::vector<std::string> const& args, std::vector<std::string>::iterator& i, int& value) {
+  auto it = i;
+  ++it;
+  if (it == args.end()) {
+    return false;
+  }
+  try {
+    value = std::stoi(*it);
+    ++i;
+    return true;
+  } catch (...) {
+    return false;
+  }
+}
+
+bool getOptionalArgument(std::vector<std::string> const& args, std::vector<std::string>::iterator& i, float& value) {
+  auto it = i;
+  ++it;
+  if (it == args.end()) {
+    return false;
+  }
+  try {
+    value = std::stof(*it);
+    ++i;
+    return true;
+  } catch (...) {
+    return false;
+  }
+}
+
+bool getOptionalArgument(std::vector<std::string> const& args,
+                         std::vector<std::string>::iterator& i,
+                         std::filesystem::path& value) {
+  auto it = i;
+  ++it;
+  if (it == args.end()) {
+    return false;
+  }
+  value = *it;
+  return true;
+}
+
+template <typename T>
+void getArgument(std::vector<std::string> const& args, std::vector<std::string>::iterator& i, T& value) {
+  if (not getOptionalArgument(args, i, value)) {
+    std::cerr << "error: " << *i << " expects an argument" << std::endl;
+    exit(EXIT_FAILURE);
+  }
+}
+
 int main(int argc, char** argv) {
   // Parse command line arguments
   std::vector<std::string> args(argv, argv + argc);
-  std::vector<Backend> backends;
+  std::unordered_map<Backend, float> backends;
   int numberOfThreads = 1;
   int numberOfStreams = 0;
   int maxEvents = -1;
@@ -84,37 +134,40 @@ int main(int argc, char** argv) {
     if (*i == "-h" or *i == "--help") {
       print_help(args.front());
       return EXIT_SUCCESS;
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
     } else if (*i == "--serial") {
-      backends.emplace_back(Backend::SERIAL);
+      float weight = 1.;
+      getOptionalArgument(args, i, weight);
+      backends.insert_or_assign(Backend::SERIAL, weight);
 #endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
     } else if (*i == "--tbb") {
-      backends.emplace_back(Backend::TBB);
+      float weight = 1.;
+      getOptionalArgument(args, i, weight);
+      backends.insert_or_assign(Backend::TBB, weight);
 #endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
     } else if (*i == "--cuda") {
-      backends.emplace_back(Backend::CUDA);
+      float weight = 1.;
+      getOptionalArgument(args, i, weight);
+      backends.insert_or_assign(Backend::CUDA, weight);
 #endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
     } else if (*i == "--hip") {
-      backends.emplace_back(Backend::HIP);
+      float weight = 1.;
+      getOptionalArgument(args, i, weight);
+      backends.insert_or_assign(Backend::HIP, weight);
 #endif
     } else if (*i == "--numberOfThreads") {
-      ++i;
-      numberOfThreads = std::stoi(*i);
+      getArgument(args, i, numberOfThreads);
     } else if (*i == "--numberOfStreams") {
-      ++i;
-      numberOfStreams = std::stoi(*i);
+      getArgument(args, i, numberOfStreams);
     } else if (*i == "--maxEvents") {
-      ++i;
-      maxEvents = std::stoi(*i);
+      getArgument(args, i, maxEvents);
     } else if (*i == "--runForMinutes") {
-      ++i;
-      runForMinutes = std::stoi(*i);
+      getArgument(args, i, runForMinutes);
     } else if (*i == "--data") {
-      ++i;
-      datadir = *i;
+      getArgument(args, i, datadir);
     } else if (*i == "--transfer") {
       transfer = true;
     } else if (*i == "--validation") {
@@ -150,76 +203,78 @@ int main(int argc, char** argv) {
   }
 
   // Initialiase the selected backends
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
-  if (std::find(backends.begin(), backends.end(), Backend::SERIAL) != backends.end()) {
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
+  if (backends.find(Backend::SERIAL) != backends.end()) {
     cms::alpakatools::initialise<alpaka_serial_sync::Platform>();
   }
 #endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
-  if (std::find(backends.begin(), backends.end(), Backend::TBB) != backends.end()) {
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
+  if (backends.find(Backend::TBB) != backends.end()) {
     cms::alpakatools::initialise<alpaka_tbb_async::Platform>();
   }
 #endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
-  if (std::find(backends.begin(), backends.end(), Backend::CUDA) != backends.end()) {
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
+  if (backends.find(Backend::CUDA) != backends.end()) {
     cms::alpakatools::initialise<alpaka_cuda_async::Platform>();
   }
 #endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-  if (std::find(backends.begin(), backends.end(), Backend::HIP) != backends.end()) {
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
+  if (backends.find(Backend::HIP) != backends.end()) {
     cms::alpakatools::initialise<alpaka_rocm_async::Platform>();
   }
 #endif
 
   // Initialize EventProcessor
-  std::vector<std::string> edmodules;
   std::vector<std::string> esmodules;
+  edm::Alternatives alternatives;
   if (not empty) {
+    // host-only ESModules
     esmodules = {"BeamSpotESProducer", "SiPixelFedIdsESProducer"};
-    auto addModules = [&](std::string const& accelerator_namespace, Backend backend) {
-      if (std::find(backends.begin(), backends.end(), backend) != backends.end()) {
-        edmodules.emplace_back(accelerator_namespace + "::" + "BeamSpotToAlpaka");
-        edmodules.emplace_back(accelerator_namespace + "::" + "SiPixelRawToCluster");
-        edmodules.emplace_back(accelerator_namespace + "::" + "SiPixelRecHitAlpaka");
-        edmodules.emplace_back(accelerator_namespace + "::" + "CAHitNtupletAlpaka");
-        edmodules.emplace_back(accelerator_namespace + "::" + "PixelVertexProducerAlpaka");
-        if (transfer) {
-          edmodules.emplace_back(accelerator_namespace + "::" + "PixelTrackSoAFromAlpaka");
-          edmodules.emplace_back(accelerator_namespace + "::" + "PixelVertexSoAFromAlpaka");
-        }
-        if (validation) {
-          edmodules.emplace_back(accelerator_namespace + "::" + "CountValidator");
-        }
-        if (histogram) {
-          edmodules.emplace_back(accelerator_namespace + "::" + "HistoValidator");
-        }
-        esmodules.emplace_back(accelerator_namespace + "::" + "SiPixelFedCablingMapESProducer");
-        esmodules.emplace_back(accelerator_namespace + "::" + "SiPixelGainCalibrationForHLTESProducer");
-        esmodules.emplace_back(accelerator_namespace + "::" + "PixelCPEFastESProducer");
+    for (auto const& [backend, weight] : backends) {
+      std::string prefix = "alpaka_" + name(backend) + "::";
+      // "portable" ESModules
+      esmodules.emplace_back(prefix + "SiPixelFedCablingMapESProducer");
+      esmodules.emplace_back(prefix + "SiPixelGainCalibrationForHLTESProducer");
+      esmodules.emplace_back(prefix + "PixelCPEFastESProducer");
+      // "portable" EDModules
+      std::vector<std::string> edmodules;
+      edmodules.emplace_back(prefix + "BeamSpotToAlpaka");
+      edmodules.emplace_back(prefix + "SiPixelRawToCluster");
+      edmodules.emplace_back(prefix + "SiPixelRecHitAlpaka");
+      edmodules.emplace_back(prefix + "CAHitNtupletAlpaka");
+      edmodules.emplace_back(prefix + "PixelVertexProducerAlpaka");
+      if (transfer) {
+        edmodules.emplace_back(prefix + "PixelTrackSoAFromAlpaka");
+        edmodules.emplace_back(prefix + "PixelVertexSoAFromAlpaka");
       }
-    };
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
-    addModules("alpaka_serial_sync", Backend::SERIAL);
-#endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
-    addModules("alpaka_tbb_async", Backend::TBB);
-#endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
-    addModules("alpaka_cuda_async", Backend::CUDA);
-#endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-    addModules("alpaka_rocm_async", Backend::HIP);
-#endif
+      if (validation) {
+        edmodules.emplace_back(prefix + "CountValidator");
+      }
+      if (histogram) {
+        edmodules.emplace_back(prefix + "HistoValidator");
+      }
+      alternatives.emplace_back(backend, weight, std::move(edmodules));
+    }
   }
   edm::EventProcessor processor(
-      maxEvents, runForMinutes, numberOfStreams, std::move(edmodules), std::move(esmodules), datadir, validation);
+      maxEvents, runForMinutes, numberOfStreams, std::move(alternatives), std::move(esmodules), datadir, validation);
 
   if (runForMinutes < 0) {
-    std::cout << "Processing " << processor.maxEvents() << " events, of which " << numberOfStreams
-              << " concurrently, with " << numberOfThreads << " threads." << std::endl;
+    std::cout << "Processing " << processor.maxEvents() << " events,";
   } else {
-    std::cout << "Processing for about " << runForMinutes << " minutes with " << numberOfStreams
-              << " concurrent events and " << numberOfThreads << " threads." << std::endl;
+    std::cout << "Processing for about " << runForMinutes << " minutes,";
+  }
+  {
+    std::cout << " with " << numberOfStreams << " concurrent events (";
+    bool need_comma = false;
+    for (auto const& [backend, streams] : processor.backends()) {
+      if (need_comma) {
+        std::cout << ", ";
+      }
+      std::cout << streams << " on " << backend;
+      need_comma = true;
+    }
+    std::cout << ") and " << numberOfThreads << " threads." << std::endl;
   }
 
   // Initialize the TBB thread pool
diff --git a/src/alpaka/plugins.txt b/src/alpaka/plugins.txt
deleted file mode 100644
index 35ed33a40..000000000
--- a/src/alpaka/plugins.txt
+++ /dev/null
@@ -1,50 +0,0 @@
-BeamSpotESProducer pluginBeamSpotProducer.so
-alpaka_cuda_async::BeamSpotToAlpaka pluginBeamSpotProducer_cuda.so
-alpaka_rocm_async::BeamSpotToAlpaka pluginBeamSpotProducer_rocm.so
-alpaka_tbb_async::BeamSpotToAlpaka pluginBeamSpotProducer_tbb.so
-alpaka_serial_sync::BeamSpotToAlpaka pluginBeamSpotProducer_serial.so
-alpaka_cuda_async::CAHitNtupletAlpaka pluginPixelTriplets_cuda.so
-alpaka_rocm_async::CAHitNtupletAlpaka pluginPixelTriplets_rocm.so
-alpaka_tbb_async::CAHitNtupletAlpaka pluginPixelTriplets_tbb.so
-alpaka_serial_sync::CAHitNtupletAlpaka pluginPixelTriplets_serial.so
-alpaka_cuda_async::PixelTrackSoAFromAlpaka pluginPixelTrackFitting_cuda.so
-alpaka_rocm_async::PixelTrackSoAFromAlpaka pluginPixelTrackFitting_rocm.so
-alpaka_tbb_async::PixelTrackSoAFromAlpaka pluginPixelTrackFitting_tbb.so
-alpaka_serial_sync::PixelTrackSoAFromAlpaka pluginPixelTrackFitting_serial.so
-alpaka_cuda_async::PixelVertexProducerAlpaka pluginPixelVertexFinding_cuda.so
-alpaka_rocm_async::PixelVertexProducerAlpaka pluginPixelVertexFinding_rocm.so
-alpaka_tbb_async::PixelVertexProducerAlpaka pluginPixelVertexFinding_tbb.so
-alpaka_serial_sync::PixelVertexProducerAlpaka pluginPixelVertexFinding_serial.so
-alpaka_cuda_async::PixelVertexSoAFromAlpaka pluginPixelVertexFinding_cuda.so
-alpaka_rocm_async::PixelVertexSoAFromAlpaka pluginPixelVertexFinding_rocm.so
-alpaka_tbb_async::PixelVertexSoAFromAlpaka pluginPixelVertexFinding_tbb.so
-alpaka_serial_sync::PixelVertexSoAFromAlpaka pluginPixelVertexFinding_serial.so
-alpaka_cuda_async::SiPixelRawToCluster pluginSiPixelClusterizer_cuda.so
-alpaka_rocm_async::SiPixelRawToCluster pluginSiPixelClusterizer_rocm.so
-alpaka_tbb_async::SiPixelRawToCluster pluginSiPixelClusterizer_tbb.so
-alpaka_serial_sync::SiPixelRawToCluster pluginSiPixelClusterizer_serial.so
-SiPixelFedIdsESProducer pluginSiPixelClusterizer.so
-alpaka_cuda_async::SiPixelFedCablingMapESProducer pluginSiPixelClusterizer_cuda.so
-alpaka_rocm_async::SiPixelFedCablingMapESProducer pluginSiPixelClusterizer_rocm.so
-alpaka_tbb_async::SiPixelFedCablingMapESProducer pluginSiPixelClusterizer_tbb.so
-alpaka_serial_sync::SiPixelFedCablingMapESProducer pluginSiPixelClusterizer_serial.so
-alpaka_cuda_async::SiPixelGainCalibrationForHLTESProducer pluginSiPixelClusterizer_cuda.so
-alpaka_rocm_async::SiPixelGainCalibrationForHLTESProducer pluginSiPixelClusterizer_rocm.so
-alpaka_tbb_async::SiPixelGainCalibrationForHLTESProducer pluginSiPixelClusterizer_tbb.so
-alpaka_serial_sync::SiPixelGainCalibrationForHLTESProducer pluginSiPixelClusterizer_serial.so
-alpaka_cuda_async::PixelCPEFastESProducer pluginSiPixelRecHits_cuda.so
-alpaka_rocm_async::PixelCPEFastESProducer pluginSiPixelRecHits_rocm.so
-alpaka_tbb_async::PixelCPEFastESProducer pluginSiPixelRecHits_tbb.so
-alpaka_serial_sync::PixelCPEFastESProducer pluginSiPixelRecHits_serial.so
-alpaka_cuda_async::SiPixelRecHitAlpaka pluginSiPixelRecHits_cuda.so
-alpaka_rocm_async::SiPixelRecHitAlpaka pluginSiPixelRecHits_rocm.so
-alpaka_tbb_async::SiPixelRecHitAlpaka pluginSiPixelRecHits_tbb.so
-alpaka_serial_sync::SiPixelRecHitAlpaka pluginSiPixelRecHits_serial.so
-alpaka_cuda_async::CountValidator pluginValidation_cuda.so
-alpaka_rocm_async::CountValidator pluginValidation_rocm.so
-alpaka_tbb_async::CountValidator pluginValidation_tbb.so
-alpaka_serial_sync::CountValidator pluginValidation_serial.so
-alpaka_cuda_async::HistoValidator pluginValidation_cuda.so
-alpaka_rocm_async::HistoValidator pluginValidation_rocm.so
-alpaka_tbb_async::HistoValidator pluginValidation_tbb.so
-alpaka_serial_sync::HistoValidator pluginValidation_serial.so
diff --git a/src/alpakatest/AlpakaCore/CachingAllocator.h b/src/alpakatest/AlpakaCore/CachingAllocator.h
index 5411ecdc5..20b26ebe0 100644
--- a/src/alpakatest/AlpakaCore/CachingAllocator.h
+++ b/src/alpakatest/AlpakaCore/CachingAllocator.h
@@ -90,6 +90,11 @@ namespace cms::alpakatools {
     using Event = alpaka::Event<Queue>;  // the events used to synchronise the operations
     using Buffer = alpaka::Buf<Device, std::byte, alpaka::DimInt<1u>, size_t>;
 
+    // The "memory device" type can either be the same as the "synchronisation device" type, or be the host CPU.
+    static_assert(std::is_same_v<Device, alpaka::Dev<Queue>> or std::is_same_v<Device, alpaka::DevCpu>,
+                  "The \"memory device\" type can either be the same as the \"synchronisation device\" type, or be the "
+                  "host CPU.");
+
     struct CachedBytes {
       size_t free = 0;       // total bytes freed and cached on this device
       size_t live = 0;       // total bytes currently in use oin this device
@@ -311,11 +316,24 @@ namespace cms::alpakatools {
       return false;
     }
 
+    Buffer allocateBuffer(size_t bytes, Queue const& queue) {
+      if constexpr (std::is_same_v<Device, alpaka::Dev<Queue>>) {
+        // allocate device memory
+        return alpaka::allocBuf<std::byte, size_t>(device_, bytes);
+      } else if constexpr (std::is_same_v<Device, alpaka::DevCpu>) {
+        // allocate pinned host memory
+        return alpaka::allocMappedBuf<std::byte, size_t>(device_, alpaka::getDev(queue), bytes);
+      } else {
+        // unsupported combination
+        static_assert(std::is_same_v<Device, alpaka::Dev<Queue>> or std::is_same_v<Device, alpaka::DevCpu>,
+                      "The \"memory device\" type can either be the same as the \"synchronisation device\" type, or be "
+                      "the host CPU.");
+      }
+    }
+
     void allocateNewBlock(BlockDescriptor& block) {
       try {
-        // FIXME simplify alpaka::Vec<alpaka::DimInt<1u>, size_t>{block.bytes} to block.bytes ?
-        block.buffer =
-            alpaka::allocBuf<std::byte, size_t>(device_, alpaka::Vec<alpaka::DimInt<1u>, size_t>{block.bytes});
+        block.buffer = allocateBuffer(block.bytes, *block.queue);
       } catch (std::runtime_error const& e) {
         // the allocation attempt failed: free all cached blocks on the device and retry
         if (debug_) {
@@ -329,25 +347,8 @@ namespace cms::alpakatools {
         freeAllCached();
 
         // throw an exception if it fails again
-        block.buffer =
-            alpaka::allocBuf<std::byte, size_t>(device_, alpaka::Vec<alpaka::DimInt<1u>, size_t>{block.bytes});
-      }
-
-      // for host memory, pin the newly allocated block
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
-      if (not cms::alpakatools::devices<alpaka::PltfCudaRt>.empty()) {
-        // it is possible to initialise the CUDA runtime and call cudaHostRegister
-        // only if the system has at least one supported GPU
-        alpaka::prepareForAsyncCopy(*block.buffer);
-      }
-#endif  // ALPAKA_ACC_GPU_CUDA_ENABLED
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-      if (not cms::alpakatools::devices<alpaka::PltfHipRt>.empty()) {
-        // it is possible to initialise the ROCm runtime and call hipHostRegister
-        // only if the system has at least one supported GPU
-        alpaka::prepareForAsyncCopy(*block.buffer);
+        block.buffer = allocateBuffer(block.bytes, *block.queue);
       }
-#endif  // ALPAKA_ACC_GPU_HIP_ENABLED
 
       // create a new event associated to the "synchronisation device"
       block.event = Event{block.device()};
diff --git a/src/alpakatest/AlpakaCore/HostOnlyTask.h b/src/alpakatest/AlpakaCore/HostOnlyTask.h
index dc9c5b7af..d010dc3eb 100644
--- a/src/alpakatest/AlpakaCore/HostOnlyTask.h
+++ b/src/alpakatest/AlpakaCore/HostOnlyTask.h
@@ -24,6 +24,8 @@ namespace alpaka {
     //! The CUDA async queue enqueue trait specialization for "safe tasks"
     template <>
     struct Enqueue<QueueCudaRtNonBlocking, HostOnlyTask> {
+      using TApi = ApiCudaRt;
+
       static void CUDART_CB callback(cudaStream_t /*queue*/, cudaError_t /*status*/, void* arg) {
         //ALPAKA_UNIFORM_CUDA_HIP_RT_CHECK(status);
         std::unique_ptr<HostOnlyTask> pTask(static_cast<HostOnlyTask*>(arg));
@@ -42,7 +44,9 @@ namespace alpaka {
     //! The HIP async queue enqueue trait specialization for "safe tasks"
     template <>
     struct Enqueue<QueueHipRtNonBlocking, HostOnlyTask> {
-      static void HIPRT_CB callback(hipStream_t /*queue*/, hipError_t /*status*/, void* arg) {
+      using TApi = ApiHipRt;
+
+      static void callback(hipStream_t /*queue*/, hipError_t /*status*/, void* arg) {
         //ALPAKA_UNIFORM_CUDA_HIP_RT_CHECK(status);
         std::unique_ptr<HostOnlyTask> pTask(static_cast<HostOnlyTask*>(arg));
         (*pTask)();
diff --git a/src/alpakatest/AlpakaCore/alpaka/initialise.cc b/src/alpakatest/AlpakaCore/alpaka/initialise.cc
new file mode 100644
index 000000000..690975af8
--- /dev/null
+++ b/src/alpakatest/AlpakaCore/alpaka/initialise.cc
@@ -0,0 +1,32 @@
+#include <iostream>
+
+#include <alpaka/alpaka.hpp>
+
+#include "AlpakaCore/alpakaConfig.h"
+#include "AlpakaCore/alpakaDevices.h"
+#include "AlpakaCore/initialise.h"
+#include "Framework/demangle.h"
+
+namespace cms::alpakatools {
+
+  template <typename TPlatform>
+  void initialise() {
+    constexpr const char* suffix[] = {"devices.", "device:", "devices:"};
+
+    if (devices<TPlatform>.empty()) {
+      devices<TPlatform> = enumerate<TPlatform>();
+      auto size = devices<TPlatform>.size();
+      //std::cout << edm::demangle<TPlatform> << " platform succesfully initialised." << std::endl;
+      std::cout << "Found " << size << " " << suffix[size < 2 ? size : 2] << std::endl;
+      for (auto const& device : devices<TPlatform>) {
+        std::cout << "  - " << alpaka::getName(device) << std::endl;
+      }
+    } else {
+      //std::cout << edm::demangle<TPlatform> << " platform already initialised." << std::endl;
+    }
+  }
+
+  // explicit template instantiation definition
+  template void initialise<ALPAKA_ACCELERATOR_NAMESPACE::Platform>();
+
+}  // namespace cms::alpakatools
diff --git a/src/alpakatest/AlpakaCore/alpakaConfig.h b/src/alpakatest/AlpakaCore/alpakaConfig.h
index 447ced874..ae4d6c058 100644
--- a/src/alpakatest/AlpakaCore/alpakaConfig.h
+++ b/src/alpakatest/AlpakaCore/alpakaConfig.h
@@ -1,9 +1,7 @@
 #ifndef AlpakaCore_alpakaConfig_h
 #define AlpakaCore_alpakaConfig_h
 
-#include <type_traits>
-
-#include <alpaka/alpaka.hpp>
+#include "AlpakaCore/alpakaFwd.h"
 
 namespace alpaka_common {
 
@@ -32,7 +30,7 @@ namespace alpaka_common {
 
   // host types
   using DevHost = alpaka::DevCpu;
-  using PltfHost = alpaka::Pltf<DevHost>;
+  using PltfHost = alpaka::PltfCpu;
 
 }  // namespace alpaka_common
 
@@ -44,7 +42,7 @@ namespace alpaka_common {
 #define DEFINE_FWK_ALPAKA_EVENTSETUP_MODULE(name) \
   DEFINE_FWK_ALPAKA_EVENTSETUP_MODULE2(ALPAKA_ACCELERATOR_NAMESPACE::name)
 
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
 namespace alpaka_cuda_async {
   using namespace alpaka_common;
 
@@ -61,13 +59,13 @@ namespace alpaka_cuda_async {
 
 }  // namespace alpaka_cuda_async
 
-#endif  // ALPAKA_ACC_GPU_CUDA_ENABLED
+#endif  // ALPAKA_ACC_GPU_CUDA_PRESENT
 
 #ifdef ALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_cuda_async
 #endif  // ALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND
 
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
 namespace alpaka_rocm_async {
   using namespace alpaka_common;
 
@@ -84,13 +82,13 @@ namespace alpaka_rocm_async {
 
 }  // namespace alpaka_rocm_async
 
-#endif  // ALPAKA_ACC_GPU_HIP_ENABLED
+#endif  // ALPAKA_ACC_GPU_HIP_PRESENT
 
 #ifdef ALPAKA_ACC_GPU_HIP_ASYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_rocm_async
 #endif  // ALPAKA_ACC_GPU_HIP_ASYNC_BACKEND
 
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
 namespace alpaka_serial_sync {
   using namespace alpaka_common;
 
@@ -107,13 +105,13 @@ namespace alpaka_serial_sync {
 
 }  // namespace alpaka_serial_sync
 
-#endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
 
 #ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_serial_sync
 #endif  // ALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND
 
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
 namespace alpaka_tbb_async {
   using namespace alpaka_common;
 
@@ -130,13 +128,13 @@ namespace alpaka_tbb_async {
 
 }  // namespace alpaka_tbb_async
 
-#endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
 
 #ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_tbb_async
 #endif  // ALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND
 
-#ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_PRESENT
 namespace alpaka_omp2_async {
   using namespace alpaka_common;
 
@@ -153,7 +151,7 @@ namespace alpaka_omp2_async {
 
 }  // namespace alpaka_omp2_async
 
-#endif  // ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ENABLED
+#endif  // ALPAKA_ACC_CPU_B_OMP2_T_SEQ_PRESENT
 
 #ifdef ALPAKA_ACC_CPU_B_OMP2_T_SEQ_ASYNC_BACKEND
 #define ALPAKA_ACCELERATOR_NAMESPACE alpaka_omp2_async
diff --git a/src/alpakatest/AlpakaCore/alpakaFwd.h b/src/alpakatest/AlpakaCore/alpakaFwd.h
new file mode 100644
index 000000000..2234e440b
--- /dev/null
+++ b/src/alpakatest/AlpakaCore/alpakaFwd.h
@@ -0,0 +1,96 @@
+#ifndef AlpakaCore_alpakaFwd_h
+#define AlpakaCore_alpakaFwd_h
+
+#include <cstddef>
+#include <cstdint>
+#include <type_traits>
+
+/**
+ * This file forward declares specific types defined in Alpaka
+ * (depending on the backend-enabling macros) so that these types
+ * would be available throughout CMSSW without a direct dependence on
+ * Alpaka in order to avoid the constraints that would impose
+ * (primarily the device compiler)
+ *
+ * This is a little bit brittle, but let's see how it goes.
+ */
+namespace alpaka {
+
+  // miscellanea
+  template <std::size_t N>
+  using DimInt = std::integral_constant<std::size_t, N>;
+
+  template <typename TDim, typename TVal>
+  class Vec;
+
+  template <typename TDim, typename TIdx>
+  class WorkDivMembers;
+
+  // API
+  struct ApiCudaRt;
+  struct ApiHipRt;
+
+  // Platforms
+  class PltfCpu;
+  template <typename TApi>
+  class PltfUniformCudaHipRt;
+  using PltfCudaRt = PltfUniformCudaHipRt<ApiCudaRt>;
+  using PltfHipRt = PltfUniformCudaHipRt<ApiHipRt>;
+
+  // Devices
+  class DevCpu;
+  template <typename TApi>
+  class DevUniformCudaHipRt;
+  using DevCudaRt = DevUniformCudaHipRt<ApiCudaRt>;
+  using DevHipRt = DevUniformCudaHipRt<ApiHipRt>;
+
+  // Queues
+  template <typename TDev>
+  class QueueGenericThreadsBlocking;
+  using QueueCpuBlocking = QueueGenericThreadsBlocking<DevCpu>;
+
+  template <typename TDev>
+  class QueueGenericThreadsNonBlocking;
+  using QueueCpuNonBlocking = QueueGenericThreadsNonBlocking<DevCpu>;
+
+  namespace uniform_cuda_hip::detail {
+    template <typename TApi, bool TBlocking>
+    class QueueUniformCudaHipRt;
+  }
+  using QueueCudaRtBlocking = uniform_cuda_hip::detail::QueueUniformCudaHipRt<ApiCudaRt, true>;
+  using QueueCudaRtNonBlocking = uniform_cuda_hip::detail::QueueUniformCudaHipRt<ApiCudaRt, false>;
+  using QueueHipRtBlocking = uniform_cuda_hip::detail::QueueUniformCudaHipRt<ApiHipRt, true>;
+  using QueueHipRtNonBlocking = uniform_cuda_hip::detail::QueueUniformCudaHipRt<ApiHipRt, false>;
+
+  // Events
+  template <typename TDev>
+  class EventGenericThreads;
+  using EventCpu = EventGenericThreads<DevCpu>;
+
+  template <typename TApi>
+  class EventUniformCudaHipRt;
+  using EventCudaRt = EventUniformCudaHipRt<ApiCudaRt>;
+  using EventHipRt = EventUniformCudaHipRt<ApiHipRt>;
+
+  // Accelerators
+  template <typename TApi, typename TDim, typename TIdx>
+  class AccGpuUniformCudaHipRt;
+
+  template <typename TDim, typename TIdx>
+  using AccGpuCudaRt = AccGpuUniformCudaHipRt<ApiCudaRt, TDim, TIdx>;
+
+  template <typename TDim, typename TIdx>
+  using AccGpuHipRt = AccGpuUniformCudaHipRt<ApiHipRt, TDim, TIdx>;
+
+  template <typename TDim, typename TIdx>
+  class AccCpuSerial;
+
+  template <typename TDim, typename TIdx>
+  class AccCpuTbbBlocks;
+
+  template <typename TDim, typename TIdx>
+  class AccCpuOmp2Blocks;
+
+}  // namespace alpaka
+
+#endif  // AlpakaCore_alpakaFwd_h
diff --git a/src/alpakatest/AlpakaCore/alpakaMemory.h b/src/alpakatest/AlpakaCore/alpakaMemory.h
index 0ec3167c3..5ab53bd59 100644
--- a/src/alpakatest/AlpakaCore/alpakaMemory.h
+++ b/src/alpakatest/AlpakaCore/alpakaMemory.h
@@ -72,86 +72,59 @@ namespace cms::alpakatools {
 
   }  // namespace detail
 
-  // helper function for pinning memory buffers
-
-  template <typename TBuf>
-  void pin_buffer(TBuf& buffer) {
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
-    if (not cms::alpakatools::devices<alpaka::PltfCudaRt>.empty()) {
-      // it is possible to initialise the CUDA runtime and call cudaHostRegister
-      // only if the system has at least one supported GPU
-      alpaka::prepareForAsyncCopy(buffer);
-    }
-#endif  // ALPAKA_ACC_GPU_CUDA_ENABLED
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-    if (not cms::alpakatools::devices<alpaka::PltfHipRt>.empty()) {
-      // it is possible to initialise the ROCm runtime and call hipHostRegister
-      // only if the system has at least one supported GPU
-      alpaka::prepareForAsyncCopy(buffer);
-    }
-#endif  // ALPAKA_ACC_GPU_HIP_ENABLED
-  }
-
   // scalar and 1-dimensional host buffers
 
   template <typename T>
   using host_buffer = typename detail::buffer_type<DevHost, T>::type;
 
-  // non-cached, scalar and 1-dimensional host buffers
-  // the memory is pinned explicitly
+  // non-cached, non-pinned, scalar and 1-dimensional host buffers
 
   template <typename T>
   std::enable_if_t<not std::is_array_v<T>, host_buffer<T>> make_host_buffer() {
-    auto buffer = alpaka::allocBuf<T, Idx>(host, Scalar{});
-    pin_buffer(buffer);
-    return buffer;
+    return alpaka::allocBuf<T, Idx>(host, Scalar{});
   }
 
   template <typename T>
   std::enable_if_t<cms::is_unbounded_array_v<T> and not std::is_array_v<std::remove_extent_t<T>>, host_buffer<T>>
   make_host_buffer(Extent extent) {
-    auto buffer = alpaka::allocBuf<std::remove_extent_t<T>, Idx>(host, Vec1D{extent});
-    pin_buffer(buffer);
-    return buffer;
+    return alpaka::allocBuf<std::remove_extent_t<T>, Idx>(host, Vec1D{extent});
   }
 
   template <typename T>
   std::enable_if_t<cms::is_bounded_array_v<T> and not std::is_array_v<std::remove_extent_t<T>>, host_buffer<T>>
   make_host_buffer() {
-    auto buffer = alpaka::allocBuf<std::remove_extent_t<T>, Idx>(host, Vec1D{std::extent_v<T>});
-    pin_buffer(buffer);
-    return buffer;
+    return alpaka::allocBuf<std::remove_extent_t<T>, Idx>(host, Vec1D{std::extent_v<T>});
   }
 
-  // potentially cached, scalar and 1-dimensional host buffers, associated to a work queue
-  // the memory is pinned by the caching allocator, or explicitly if it is not used
+  // potentially cached, pinned, scalar and 1-dimensional host buffers, associated to a work queue
+  // the memory is pinned according to the device associated to the queue
 
   template <typename T, typename TQueue>
-  std::enable_if_t<not std::is_array_v<T>, host_buffer<T>> make_host_buffer(TQueue const& queue [[maybe_unused]]) {
+  std::enable_if_t<not std::is_array_v<T>, host_buffer<T>> make_host_buffer(TQueue const& queue) {
     if constexpr (allocator_policy<alpaka::Dev<TQueue>> == AllocatorPolicy::Caching) {
       return allocCachedBuf<T, Idx>(host, queue, Scalar{});
     } else {
-      return make_host_buffer<T>();
+      return alpaka::allocMappedBuf<T, Idx>(host, alpaka::getDev(queue), Scalar{});
     }
   }
 
   template <typename T, typename TQueue>
   std::enable_if_t<cms::is_unbounded_array_v<T> and not std::is_array_v<std::remove_extent_t<T>>, host_buffer<T>>
-  make_host_buffer(TQueue const& queue [[maybe_unused]], Extent extent) {
+  make_host_buffer(TQueue const& queue, Extent extent) {
     if constexpr (allocator_policy<alpaka::Dev<TQueue>> == AllocatorPolicy::Caching) {
       return allocCachedBuf<std::remove_extent_t<T>, Idx>(host, queue, Vec1D{extent});
     } else {
-      return make_host_buffer<T>(extent);
+      return alpaka::allocMappedBuf<std::remove_extent_t<T>, Idx>(host, alpaka::getDev(queue), Vec1D{extent});
     }
   }
 
   template <typename T, typename TQueue>
   std::enable_if_t<cms::is_bounded_array_v<T> and not std::is_array_v<std::remove_extent_t<T>>, host_buffer<T>>
-  make_host_buffer(TQueue const& queue [[maybe_unused]]) {
+  make_host_buffer(TQueue const& queue) {
     if constexpr (allocator_policy<alpaka::Dev<TQueue>> == AllocatorPolicy::Caching) {
       return allocCachedBuf<std::remove_extent_t<T>, Idx>(host, queue, Vec1D{std::extent_v<T>});
     } else {
-      return make_host_buffer<T>();
+      return alpaka::allocMappedBuf<std::remove_extent_t<T>, Idx>(host, alpaka::getDev(queue), Vec1D{std::extent_v<T>});
     }
   }
 
diff --git a/src/alpakatest/AlpakaCore/backend.h b/src/alpakatest/AlpakaCore/backend.h
index 387154013..8d58953cc 100644
--- a/src/alpakatest/AlpakaCore/backend.h
+++ b/src/alpakatest/AlpakaCore/backend.h
@@ -3,4 +3,15 @@
 
 enum class Backend { SERIAL, TBB, CUDA, HIP };
 
+inline std::string const& name(Backend backend) {
+  static const std::string names[] = {"serial_sync", "tbb_async", "cuda_async", "rocm_async"};
+  return names[static_cast<int>(backend)];
+}
+
+template <typename T>
+inline T& operator<<(T& out, Backend backend) {
+  out << name(backend);
+  return out;
+}
+
 #endif  // AlpakaCore_backend_h
diff --git a/src/alpakatest/AlpakaCore/initialise.h b/src/alpakatest/AlpakaCore/initialise.h
index 98ba966a1..1114ff5f9 100644
--- a/src/alpakatest/AlpakaCore/initialise.h
+++ b/src/alpakatest/AlpakaCore/initialise.h
@@ -1,31 +1,26 @@
 #ifndef AlpakaCore_initialise_h
 #define AlpakaCore_initialise_h
 
-#include <iostream>
-
-#include <alpaka/alpaka.hpp>
-
-#include "AlpakaCore/alpakaDevices.h"
-#include "Framework/demangle.h"
+#include "AlpakaCore/alpakaConfig.h"
 
 namespace cms::alpakatools {
 
   template <typename TPlatform>
-  void initialise() {
-    constexpr const char* suffix[] = {"devices.", "device:", "devices:"};
-
-    if (devices<TPlatform>.empty()) {
-      devices<TPlatform> = enumerate<TPlatform>();
-      auto size = devices<TPlatform>.size();
-      //std::cout << edm::demangle<TPlatform> << " platform succesfully initialised." << std::endl;
-      std::cout << "Found " << size << " " << suffix[size < 2 ? size : 2] << std::endl;
-      for (auto const& device : devices<TPlatform>) {
-        std::cout << "  - " << alpaka::getName(device) << std::endl;
-      }
-    } else {
-      //std::cout << edm::demangle<TPlatform> << " platform already initialised." << std::endl;
-    }
-  }
+  void initialise();
+
+  // explicit template instantiation declaration
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
+  extern template void initialise<alpaka_serial_sync::Platform>();
+#endif
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
+  extern template void initialise<alpaka_tbb_async::Platform>();
+#endif
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
+  extern template void initialise<alpaka_cuda_async::Platform>();
+#endif
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
+  extern template void initialise<alpaka_rocm_async::Platform>();
+#endif
 
 }  // namespace cms::alpakatools
 
diff --git a/src/alpakatest/Makefile b/src/alpakatest/Makefile
index 281871c13..c39793a16 100644
--- a/src/alpakatest/Makefile
+++ b/src/alpakatest/Makefile
@@ -31,12 +31,12 @@ EXE_DEP := $(EXE_OBJ:$.o=$.d)
 
 LIBNAMES := $(filter-out plugin-% bin test Makefile% plugins.txt%,$(wildcard *))
 PLUGINNAMES := $(patsubst plugin-%,%,$(filter plugin-%,$(wildcard *)))
-MY_CXXFLAGS := -I$(TARGET_DIR) -DSRC_DIR=$(TARGET_DIR) -DLIB_DIR=$(LIB_DIR)/$(TARGET_NAME) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED -DALPAKA_HOST_ONLY
+MY_CXXFLAGS := -I$(TARGET_DIR) -DLIB_DIR=$(LIB_DIR)/$(TARGET_NAME) -DALPAKA_HOST_ONLY -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT -DALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
 ifdef CUDA_BASE
-MY_CXXFLAGS += -DALPAKA_ACC_GPU_CUDA_ENABLED -DALPAKA_ACC_GPU_CUDA_ONLY_MODE
+MY_CXXFLAGS += -DALPAKA_ACC_GPU_CUDA_PRESENT -DALPAKA_ACC_GPU_CUDA_ONLY_MODE
 endif
 ifdef ROCM_BASE
-MY_CXXFLAGS += -DALPAKA_ACC_GPU_HIP_ENABLED -DALPAKA_ACC_GPU_HIP_ONLY_MODE
+MY_CXXFLAGS += -DALPAKA_ACC_GPU_HIP_PRESENT -DALPAKA_ACC_GPU_HIP_ONLY_MODE
 endif
 MY_LDFLAGS := -ldl -Wl,-rpath,$(LIB_DIR)/$(TARGET_NAME)
 LIB_LDFLAGS := -L$(LIB_DIR)/$(TARGET_NAME)
@@ -182,8 +182,12 @@ $(foreach test,$(TESTS_ROCM_EXE),$(eval $(call RUNTEST_template,$(test),amdgpu))
 -include $(ALL_DEPENDS)
 
 # Build targets
-$(TARGET): $(EXE_OBJ) $(LIBS) $(PLUGINS) | $(TESTS_EXE)
-	$(CXX) $(EXE_OBJ) $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+$(LIB_DIR)/$(TARGET_NAME)/plugins.txt: $(PLUGINS)
+	nm -A -C -D -P --defined-only $(PLUGINS) | sed -n -e"s#$(LIB_DIR)/$(TARGET_NAME)/\(plugin\w\+\.so\): typeinfo for edm::\(PluginFactory\|ESPluginFactory\)::impl::Maker<\([A-Za-z0-9_:]\+\)> V .* .*#\3 \1#p" | sort > $@
+
+$(TARGET): $(EXE_OBJ) $(LIBS) $(PLUGINS) $(LIB_DIR)/$(TARGET_NAME)/plugins.txt | $(TESTS_EXE)
+	# Link all libraries, also the "portable" ones
+	$(CXX) $(EXE_OBJ) $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_SERIAL_LDFLAGS) $($(lib)_TBB_LDFLAGS) $($(lib)_CUDA_LDFLAGS) $($(lib)_ROCM_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 
 define BUILD_template
 $(OBJ_DIR)/$(2)/%.cc.o: $(SRC_DIR)/$(2)/%.cc
@@ -219,7 +223,7 @@ $$($(1)_ROCM_LIB): $$($(1)_ROCM_OBJ) $$(foreach dep,$(EXTERNAL_DEPENDS_H),$$($$(
 # Portable code, for serial backend
 $(OBJ_DIR)/$(2)/alpaka/%.cc.serial.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 	@[ -d $$(@D) ] || mkdir -p $$(@D)
-	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
+	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
 	@cp $(OBJ_DIR)/$(2)/alpaka/$$*.cc.serial.d $(OBJ_DIR)/$(2)/alpaka/$$*.cc.serial.d.tmp; \
 	  sed 's#\($(2)/alpaka/$$*\)\.o[ :]*#\1.o \1.d : #g' < $(OBJ_DIR)/$(2)/alpaka/$$*.cc.serial.d.tmp > $(OBJ_DIR)/$(2)/alpaka/$$*.cc.serial.d; \
 	  sed -e 's/#.*//' -e 's/^[^:]*: *//' -e 's/ *\\$$$$//' \
@@ -229,7 +233,7 @@ $(OBJ_DIR)/$(2)/alpaka/%.cc.serial.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 # Portable code, for TBB backend
 $(OBJ_DIR)/$(2)/alpaka/%.cc.tbb.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 	@[ -d $$(@D) ] || mkdir -p $$(@D)
-	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
+	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
 	@cp $(OBJ_DIR)/$(2)/alpaka/$$*.cc.tbb.d $(OBJ_DIR)/$(2)/alpaka/$$*.cc.tbb.d.tmp; \
 	  sed 's#\($(2)/alpaka/$$*\)\.o[ :]*#\1.o \1.d : #g' < $(OBJ_DIR)/$(2)/alpaka/$$*.cc.tbb.d.tmp > $(OBJ_DIR)/$(2)/alpaka/$$*.cc.tbb.d; \
 	  sed -e 's/#.*//' -e 's/^[^:]*: *//' -e 's/ *\\$$$$//' \
@@ -240,7 +244,7 @@ $(OBJ_DIR)/$(2)/alpaka/%.cc.tbb.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 ifdef CUDA_BASE
 $(OBJ_DIR)/$(2)/alpaka/%.cc.cuda.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 	@[ -d $$(@D) ] || mkdir -p $$(@D)
-	$(CUDA_NVCC) -x cu $(CUDA_CUFLAGS) $(CUDA_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND -UALPAKA_HOST_ONLY $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_NVCC_CXXFLAGS)) -c $$< -o $$@ -MMD
+	$(CUDA_NVCC) -x cu $(CUDA_CUFLAGS) $(CUDA_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_CUDA_ENABLED -DALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND -UALPAKA_HOST_ONLY $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_NVCC_CXXFLAGS)) -c $$< -o $$@ -MMD
 
 $$($(1)_CUDADLINK): $$($(1)_CUOBJ)
 	$(CUDA_NVCC) $(CUDA_DLINKFLAGS) $(CUDA_LDFLAGS) $$($(1)_CUOBJ) -o $$@
@@ -250,7 +254,7 @@ endif
 ifdef ROCM_BASE
 $(OBJ_DIR)/$(2)/alpaka/%.cc.rocm.o: $(SRC_DIR)/$(2)/alpaka/%.cc
 	@[ -d $$(@D) ] || mkdir -p $$(@D)
-	$(ROCM_HIPCC) $(HIPCC_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_HIP_ASYNC_BACKEND -UALPAKA_HOST_ONLY $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
+	$(ROCM_HIPCC) $(HIPCC_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_HIP_ENABLED -DALPAKA_ACC_GPU_HIP_ASYNC_BACKEND -UALPAKA_HOST_ONLY $$(foreach dep,$(EXTERNAL_DEPENDS),$$($$(dep)_CXXFLAGS)) -c $$< -o $$@ -MMD
 endif
 
 endef
@@ -271,7 +275,7 @@ $(OBJ_DIR)/$(TARGET_NAME)/bin/%.cc.o: $(SRC_DIR)/$(TARGET_NAME)/bin/%.cc
 # Serial backend
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.serial.o: $(SRC_DIR)/$(TARGET_NAME)/test/alpaka/%.cc
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
+	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
 	@cp $(@D)/$*.cc.serial.d $(@D)/$*.cc.serial.d.tmp; \
 	  sed 's#\($(TARGET_NAME)/$*\)\.o[ :]*#\1.o \1.d : #g' < $(@D)/$*.cc.serial.d.tmp > $(@D)/$*.cc.serial.d; \
 	  sed -e 's/#.*//' -e 's/^[^:]*: *//' -e 's/ *\\$$//' \
@@ -280,12 +284,12 @@ $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.serial.o: $(SRC_DIR)/$(TARGET_NAME)/t
 
 $(TEST_DIR)/$(TARGET_NAME)/%.serial: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.serial.o | $(LIBS)
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_SERIAL_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 
 # TBB backend
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.tbb.o: $(SRC_DIR)/$(TARGET_NAME)/test/alpaka/%.cc
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
+	$(CXX) $(CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED -DALPAKA_ACC_CPU_B_TBB_T_SEQ_ASYNC_BACKEND $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
 	@cp $(@D)/$*.cc.tbb.d $(@D)/$*.cc.tbb.d.tmp; \
 	  sed 's#\($(TARGET_NAME)/$*\)\.o[ :]*#\1.o \1.d : #g' < $(@D)/$*.cc.tbb.d.tmp > $(@D)/$*.cc.tbb.d; \
 	  sed -e 's/#.*//' -e 's/^[^:]*: *//' -e 's/ *\\$$//' \
@@ -294,29 +298,29 @@ $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.tbb.o: $(SRC_DIR)/$(TARGET_NAME)/test
 
 $(TEST_DIR)/$(TARGET_NAME)/%.tbb: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.tbb.o | $(LIBS)
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_TBB_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 
 # CUDA backend
 ifdef CUDA_BASE
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.cuda.o: $(SRC_DIR)/$(TARGET_NAME)/test/alpaka/%.cc
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CUDA_NVCC) -x cu $(CUDA_CUFLAGS) $(CUDA_CXXFLAGS) $(CUDA_TEST_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND -UALPAKA_HOST_ONLY $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_NVCC_CXXFLAGS)) -c $< -o $@ -MMD
+	$(CUDA_NVCC) -x cu $(CUDA_CUFLAGS) $(CUDA_CXXFLAGS) $(CUDA_TEST_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_CUDA_ENABLED -DALPAKA_ACC_GPU_CUDA_ASYNC_BACKEND -UALPAKA_HOST_ONLY $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_NVCC_CXXFLAGS)) -c $< -o $@ -MMD
 
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cudadlink.o: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.o
 	$(CUDA_NVCC) $(CUDA_DLINKFLAGS) $(CUDA_LDFLAGS) $< -o $@
 
 $(TEST_DIR)/$(TARGET_NAME)/%.cuda: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.cuda.o $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.cuda.cudadlink.o | $(LIBS)
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_CUDA_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 endif
 
 # ROCm backend
 ifdef ROCM_BASE
 $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.rocm.o: $(SRC_DIR)/$(TARGET_NAME)/test/alpaka/%.cc
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(ROCM_HIPCC) $(HIPCC_CXXFLAGS) $(ROCM_TEST_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_HIP_ASYNC_BACKEND -UALPAKA_HOST_ONLY $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
+	$(ROCM_HIPCC) $(HIPCC_CXXFLAGS) $(ROCM_TEST_CXXFLAGS) $(MY_CXXFLAGS) -DALPAKA_ACC_GPU_HIP_ENABLED -DALPAKA_ACC_GPU_HIP_ASYNC_BACKEND -UALPAKA_HOST_ONLY $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_CXXFLAGS)) -c $< -o $@ -MMD
 
 $(TEST_DIR)/$(TARGET_NAME)/%.rocm: $(OBJ_DIR)/$(TARGET_NAME)/test/alpaka/%.cc.rocm.o | $(LIBS)
 	@[ -d $(@D) ] || mkdir -p $(@D)
-	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(patsubst %,-l%,$(LIBNAMES)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
+	$(CXX) $^ $(LDFLAGS) $(MY_LDFLAGS) -o $@ -L$(LIB_DIR)/$(TARGET_NAME) $(foreach lib,$(LIBNAMES),$($(lib)_LDFLAGS) $($(lib)_ROCM_LDFLAGS)) $(foreach dep,$(EXTERNAL_DEPENDS),$($(dep)_LDFLAGS))
 endif
diff --git a/src/alpakatest/bin/EventProcessor.cc b/src/alpakatest/bin/EventProcessor.cc
index 3d7a3546f..7e84f5114 100644
--- a/src/alpakatest/bin/EventProcessor.cc
+++ b/src/alpakatest/bin/EventProcessor.cc
@@ -1,3 +1,4 @@
+#include <cmath>
 #include <exception>
 #include <filesystem>
 #include <string>
@@ -13,7 +14,7 @@ namespace edm {
   EventProcessor::EventProcessor(int maxEvents,
                                  int runForMinutes,
                                  int numberOfStreams,
-                                 std::vector<std::string> const& path,
+                                 Alternatives alternatives,
                                  std::vector<std::string> const& esproducers,
                                  std::filesystem::path const& datadir,
                                  bool validation)
@@ -24,9 +25,23 @@ namespace edm {
       esp->produce(eventSetup_);
     }
 
+    // normalise the total weight to the number of streams
+    float total = 0.;
+    for (auto const& alternative : alternatives) {
+      total += alternative.weight;
+    }
     //schedules_.reserve(numberOfStreams);
-    for (int i = 0; i < numberOfStreams; ++i) {
-      schedules_.emplace_back(registry_, pluginManager_, &source_, &eventSetup_, i, path);
+    float cumulative = 0.;
+    int lower_range = 0;
+    int upper_range = 0;
+    for (auto& alternative : alternatives) {
+      cumulative += alternative.weight;
+      lower_range = upper_range;
+      upper_range = static_cast<int>(std::round(cumulative * numberOfStreams / total));
+      for (int i = lower_range; i < upper_range; ++i) {
+        schedules_.emplace_back(registry_, pluginManager_, &source_, &eventSetup_, i, alternative.path);
+      }
+      streamsPerBackend_.emplace_back(alternative.backend, upper_range - lower_range);
     }
   }
 
diff --git a/src/alpakatest/bin/EventProcessor.h b/src/alpakatest/bin/EventProcessor.h
index 5fb20f0b9..e1ca9fe01 100644
--- a/src/alpakatest/bin/EventProcessor.h
+++ b/src/alpakatest/bin/EventProcessor.h
@@ -5,6 +5,7 @@
 #include <string>
 #include <vector>
 
+#include "AlpakaCore/backend.h"
 #include "Framework/EventSetup.h"
 
 #include "PluginManager.h"
@@ -12,18 +13,31 @@
 #include "Source.h"
 
 namespace edm {
+  struct Alternative {
+    Alternative() = default;
+    Alternative(Backend backend, float weight, std::vector<std::string> path)
+        : backend{backend}, weight{weight}, path{std::move(path)} {}
+
+    Backend backend;
+    float weight;
+    std::vector<std::string> path;
+  };
+
+  using Alternatives = std::vector<Alternative>;
+
   class EventProcessor {
   public:
     explicit EventProcessor(int maxEvents,
                             int runForMinutes,
                             int numberOfStreams,
-                            std::vector<std::string> const& path,
+                            Alternatives alternatives,
                             std::vector<std::string> const& esproducers,
                             std::filesystem::path const& datadir,
                             bool validation);
 
     int maxEvents() const { return source_.maxEvents(); }
     int processedEvents() const { return source_.processedEvents(); }
+    std::vector<std::pair<Backend, int>> const& backends() const { return streamsPerBackend_; }
 
     void runToCompletion();
 
@@ -35,6 +49,7 @@ namespace edm {
     Source source_;
     EventSetup eventSetup_;
     std::vector<StreamSchedule> schedules_;
+    std::vector<std::pair<Backend, int>> streamsPerBackend_;
   };
 }  // namespace edm
 
diff --git a/src/alpakatest/bin/PluginManager.cc b/src/alpakatest/bin/PluginManager.cc
index 0fd46bbff..d1fa1d0f7 100644
--- a/src/alpakatest/bin/PluginManager.cc
+++ b/src/alpakatest/bin/PluginManager.cc
@@ -6,9 +6,6 @@
 
 #include "PluginManager.h"
 
-#ifndef SRC_DIR
-#error "SRC_DIR undefined"
-#endif
 #ifndef LIB_DIR
 #error "LIB_DIR undefined"
 #endif
@@ -18,7 +15,7 @@
 
 namespace edmplugin {
   PluginManager::PluginManager() {
-    std::ifstream pluginMap(STR(SRC_DIR) "/plugins.txt");
+    std::ifstream pluginMap(STR(LIB_DIR) "/plugins.txt");
     std::string plugin, library;
     while (pluginMap >> plugin >> library) {
       //std::cout << "plugin " << plugin << " in " << library << std::endl;
diff --git a/src/alpakatest/bin/main.cc b/src/alpakatest/bin/main.cc
index ba18f372e..8e7a36c6c 100644
--- a/src/alpakatest/bin/main.cc
+++ b/src/alpakatest/bin/main.cc
@@ -6,6 +6,7 @@
 #include <iomanip>
 #include <ios>
 #include <iostream>
+#include <unordered_map>
 #include <stdexcept>
 #include <string>
 #include <utility>
@@ -25,37 +26,37 @@ namespace {
   void print_help(std::string const& name) {
     std::cout
         << name << ": "
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
         << "[--serial] "
 #endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
         << "[--tbb] "
 #endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
         << "[--cuda] "
 #endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
         << "[--hip] "
 #endif
         << "[--numberOfThreads NT] [--numberOfStreams NS] [--maxEvents ME] [--data PATH] "
            "[--transfer]\n\n"
         << "Options\n"
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
         << " --serial            Use CPU Serial backend\n"
 #endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
         << " --tbb               Use CPU TBB backend\n"
 #endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
         << " --cuda              Use CUDA backend\n"
 #endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
         << " --hip               Use ROCm/HIP backend\n"
 #endif
         << " --numberOfThreads   Number of threads to use (default 1, use 0 to use all CPU cores)\n"
         << " --numberOfStreams   Number of concurrent events (default 0 = numberOfThreads)\n"
         << " --maxEvents         Number of events to process (default -1 for all events in the input file)\n"
-        << " --runForMinutes     Continue processing the set of 1000 events until this many minutes have passed"
+        << " --runForMinutes     Continue processing the set of 1000 events until this many minutes have passed "
            "(default -1 for disabled; conflicts with --maxEvents)\n"
         << " --data              Path to the 'data' directory (default 'data' in the directory of the executable)\n"
         << " --transfer          Transfer results from GPU to CPU (default is to leave them on GPU)\n"
@@ -64,10 +65,60 @@ namespace {
   }
 }  // namespace
 
+bool getOptionalArgument(std::vector<std::string> const& args, std::vector<std::string>::iterator& i, int& value) {
+  auto it = i;
+  ++it;
+  if (it == args.end()) {
+    return false;
+  }
+  try {
+    value = std::stoi(*it);
+    ++i;
+    return true;
+  } catch (...) {
+    return false;
+  }
+}
+
+bool getOptionalArgument(std::vector<std::string> const& args, std::vector<std::string>::iterator& i, float& value) {
+  auto it = i;
+  ++it;
+  if (it == args.end()) {
+    return false;
+  }
+  try {
+    value = std::stof(*it);
+    ++i;
+    return true;
+  } catch (...) {
+    return false;
+  }
+}
+
+bool getOptionalArgument(std::vector<std::string> const& args,
+                         std::vector<std::string>::iterator& i,
+                         std::filesystem::path& value) {
+  auto it = i;
+  ++it;
+  if (it == args.end()) {
+    return false;
+  }
+  value = *it;
+  return true;
+}
+
+template <typename T>
+void getArgument(std::vector<std::string> const& args, std::vector<std::string>::iterator& i, T& value) {
+  if (not getOptionalArgument(args, i, value)) {
+    std::cerr << "error: " << *i << " expects an argument" << std::endl;
+    exit(EXIT_FAILURE);
+  }
+}
+
 int main(int argc, char** argv) {
   // Parse command line arguments
   std::vector<std::string> args(argv, argv + argc);
-  std::vector<Backend> backends;
+  std::unordered_map<Backend, float> backends;
   int numberOfThreads = 1;
   int numberOfStreams = 0;
   int maxEvents = -1;
@@ -79,37 +130,40 @@ int main(int argc, char** argv) {
     if (*i == "-h" or *i == "--help") {
       print_help(args.front());
       return EXIT_SUCCESS;
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
     } else if (*i == "--serial") {
-      backends.emplace_back(Backend::SERIAL);
+      float weight = 1.;
+      getOptionalArgument(args, i, weight);
+      backends.insert_or_assign(Backend::SERIAL, weight);
 #endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
     } else if (*i == "--tbb") {
-      backends.emplace_back(Backend::TBB);
+      float weight = 1.;
+      getOptionalArgument(args, i, weight);
+      backends.insert_or_assign(Backend::TBB, weight);
 #endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
     } else if (*i == "--cuda") {
-      backends.emplace_back(Backend::CUDA);
+      float weight = 1.;
+      getOptionalArgument(args, i, weight);
+      backends.insert_or_assign(Backend::CUDA, weight);
 #endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
     } else if (*i == "--hip") {
-      backends.emplace_back(Backend::HIP);
+      float weight = 1.;
+      getOptionalArgument(args, i, weight);
+      backends.insert_or_assign(Backend::HIP, weight);
 #endif
     } else if (*i == "--numberOfThreads") {
-      ++i;
-      numberOfThreads = std::stoi(*i);
+      getArgument(args, i, numberOfThreads);
     } else if (*i == "--numberOfStreams") {
-      ++i;
-      numberOfStreams = std::stoi(*i);
+      getArgument(args, i, numberOfStreams);
     } else if (*i == "--maxEvents") {
-      ++i;
-      maxEvents = std::stoi(*i);
+      getArgument(args, i, maxEvents);
     } else if (*i == "--runForMinutes") {
-      ++i;
-      runForMinutes = std::stoi(*i);
+      getArgument(args, i, runForMinutes);
     } else if (*i == "--data") {
-      ++i;
-      datadir = *i;
+      getArgument(args, i, datadir);
     } else if (*i == "--transfer") {
       transfer = true;
     } else if (*i == "--empty") {
@@ -139,65 +193,65 @@ int main(int argc, char** argv) {
   }
 
   // Initialiase the selected backends
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
-  if (std::find(backends.begin(), backends.end(), Backend::SERIAL) != backends.end()) {
+#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_PRESENT
+  if (backends.find(Backend::SERIAL) != backends.end()) {
     cms::alpakatools::initialise<alpaka_serial_sync::Platform>();
   }
 #endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
-  if (std::find(backends.begin(), backends.end(), Backend::TBB) != backends.end()) {
+#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_PRESENT
+  if (backends.find(Backend::TBB) != backends.end()) {
     cms::alpakatools::initialise<alpaka_tbb_async::Platform>();
   }
 #endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
-  if (std::find(backends.begin(), backends.end(), Backend::CUDA) != backends.end()) {
+#ifdef ALPAKA_ACC_GPU_CUDA_PRESENT
+  if (backends.find(Backend::CUDA) != backends.end()) {
     cms::alpakatools::initialise<alpaka_cuda_async::Platform>();
   }
 #endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-  if (std::find(backends.begin(), backends.end(), Backend::HIP) != backends.end()) {
+#ifdef ALPAKA_ACC_GPU_HIP_PRESENT
+  if (backends.find(Backend::HIP) != backends.end()) {
     cms::alpakatools::initialise<alpaka_rocm_async::Platform>();
   }
 #endif
 
   // Initialize EventProcessor
-  std::vector<std::string> edmodules;
   std::vector<std::string> esmodules;
+  edm::Alternatives alternatives;
   if (not empty) {
-    auto addModules = [&](std::string const& accelerator_namespace, Backend backend) {
-      if (std::find(backends.begin(), backends.end(), backend) != backends.end()) {
-        edmodules.emplace_back(accelerator_namespace + "::" + "TestProducer");
-        edmodules.emplace_back(accelerator_namespace + "::" + "TestProducer3");
-        edmodules.emplace_back(accelerator_namespace + "::" + "TestProducer2");
-      }
-    };
-
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
-    addModules("alpaka_serial_sync", Backend::SERIAL);
-#endif
-#ifdef ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
-    addModules("alpaka_tbb_async", Backend::TBB);
-#endif
-#ifdef ALPAKA_ACC_GPU_CUDA_ENABLED
-    addModules("alpaka_cuda_async", Backend::CUDA);
-#endif
-#ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-    addModules("alpaka_rocm_async", Backend::HIP);
-#endif
+    // host-only ESModules
     esmodules = {"IntESProducer"};
-    if (transfer) {
-      // add modules for transfer
+    for (auto const& [backend, weight] : backends) {
+      std::string prefix = "alpaka_" + name(backend) + "::";
+      // "portable" EDModules
+      std::vector<std::string> edmodules;
+      edmodules.emplace_back(prefix + "TestProducer");
+      edmodules.emplace_back(prefix + "TestProducer3");
+      edmodules.emplace_back(prefix + "TestProducer2");
+      if (transfer) {
+        // add modules for transfer
+      }
+      alternatives.emplace_back(backend, weight, std::move(edmodules));
     }
   }
   edm::EventProcessor processor(
-      maxEvents, runForMinutes, numberOfStreams, std::move(edmodules), std::move(esmodules), datadir, false);
+      maxEvents, runForMinutes, numberOfStreams, std::move(alternatives), std::move(esmodules), datadir, false);
 
   if (runForMinutes < 0) {
-    std::cout << "Processing " << processor.maxEvents() << " events, of which " << numberOfStreams
-              << " concurrently, with " << numberOfThreads << " threads." << std::endl;
+    std::cout << "Processing " << processor.maxEvents() << " events,";
   } else {
-    std::cout << "Processing for about " << runForMinutes << " minutes with " << numberOfStreams
-              << " concurrent events and " << numberOfThreads << " threads." << std::endl;
+    std::cout << "Processing for about " << runForMinutes << " minutes,";
+  }
+  {
+    std::cout << " with " << numberOfStreams << " concurrent events (";
+    bool need_comma = false;
+    for (auto const& [backend, streams] : processor.backends()) {
+      if (need_comma) {
+        std::cout << ", ";
+      }
+      std::cout << streams << " on " << backend;
+      need_comma = true;
+    }
+    std::cout << ") and " << numberOfThreads << " threads." << std::endl;
   }
 
   // Initialize the TBB thread pool
diff --git a/src/alpakatest/plugins.txt b/src/alpakatest/plugins.txt
deleted file mode 100644
index 6c4d0c386..000000000
--- a/src/alpakatest/plugins.txt
+++ /dev/null
@@ -1,13 +0,0 @@
-IntESProducer pluginTest1.so
-alpaka_serial_sync::TestProducer pluginTest1_serial.so
-alpaka_tbb_async::TestProducer pluginTest1_tbb.so
-alpaka_cuda_async::TestProducer pluginTest1_cuda.so
-alpaka_rocm_async::TestProducer pluginTest1_rocm.so
-alpaka_serial_sync::TestProducer2 pluginTest2_serial.so
-alpaka_tbb_async::TestProducer2 pluginTest2_tbb.so
-alpaka_cuda_async::TestProducer2 pluginTest2_cuda.so
-alpaka_rocm_async::TestProducer2 pluginTest2_rocm.so
-alpaka_serial_sync::TestProducer3 pluginTest2_serial.so
-alpaka_tbb_async::TestProducer3 pluginTest2_tbb.so
-alpaka_cuda_async::TestProducer3 pluginTest2_cuda.so
-alpaka_rocm_async::TestProducer3 pluginTest2_rocm.so
diff --git a/src/alpakatest/test/alpaka/hello.cc b/src/alpakatest/test/alpaka/hello.cc
index 7d8d134ae..fe5fd4201 100644
--- a/src/alpakatest/test/alpaka/hello.cc
+++ b/src/alpakatest/test/alpaka/hello.cc
@@ -1,16 +1,19 @@
 #include <iostream>
 
-#include "AlpakaCore/alpakaConfig.h"
-
 int main() {
-  std::cout << "Hello from "
-#ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_SYNC_BACKEND
-            << "CPU serial"
-#elif defined ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
-            << "CPU TBB"
-#elif defined ALPAKA_ACC_GPU_CUDA_ENABLED
-            << "CUDA"
+  std::cout << "Hello from the "
+#if defined ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
+            << "CPU serial "
+#endif
+#if defined ALPAKA_ACC_CPU_B_TBB_T_SEQ_ENABLED
+            << "CPU TBB "
+#endif
+#if defined ALPAKA_ACC_GPU_CUDA_ENABLED
+            << "CUDA "
+#endif
+#if defined ALPAKA_ACC_GPU_HIP_ENABLED
+            << "HIP/ROCm "
 #endif
-            << " backend" << std::endl;
+            << "backend" << std::endl;
   return 0;
 }