Alpaka-related updates [13.0.x]

FWCore/ParameterSet/python/Config.py

@@ -1538,10 +1538,10 @@ def handleProcessAccelerators(self, parameterSet):
                    resolved.update(acc)
             # Sanity check
             if len(invalid) != 0:
-                raise ValueError("Invalid pattern{} of {} in process.options.accelerators, valid values are {} or a pattern matching to some of them.".format(
+                raise ValueError("Invalid pattern{} of '{}' in process.options.accelerators, valid values are '{}' or a pattern matching some of them.".format(
                     "s" if len(invalid) > 2 else "",
-                    ",".join(invalid),
-                    ",".join(sorted(list(allAccelerators)))))
+                    "', '".join(invalid),
+                    "', '".join(sorted(list(allAccelerators)))))
             selectedAccelerators = sorted(list(resolved))
         parameterSet.addVString(False, "@selected_accelerators", selectedAccelerators)
 

HeterogeneousCore/AlpakaCore/interface/EventCache.h

@@ -21,7 +21,7 @@ namespace cms::alpakatools {
     friend class alpaka_cuda_async::AlpakaService;
 #endif
 #ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-    friend class alpaka_hip_async::AlpakaService;
+    friend class alpaka_rocm_async::AlpakaService;
 #endif
 #ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
     friend class alpaka_serial_sync::AlpakaService;

HeterogeneousCore/AlpakaCore/interface/QueueCache.h

@@ -19,7 +19,7 @@ namespace cms::alpakatools {
     friend class alpaka_cuda_async::AlpakaService;
 #endif
 #ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-    friend class alpaka_hip_async::AlpakaService;
+    friend class alpaka_rocm_async::AlpakaService;
 #endif
 #ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
     friend class alpaka_serial_sync::AlpakaService;

HeterogeneousCore/AlpakaCore/interface/MakerMacros.h → HeterogeneousCore/AlpakaCore/interface/alpaka/MakerMacros.h

@@ -1,5 +1,5 @@
-#ifndef HeterogeneousCore_AlpakaCore_interface_MakerMacros_h
-#define HeterogeneousCore_AlpakaCore_interface_MakerMacros_h
+#ifndef HeterogeneousCore_AlpakaCore_interface_alpaka_MakerMacros_h
+#define HeterogeneousCore_AlpakaCore_interface_alpaka_MakerMacros_h
 
 #include "FWCore/Framework/interface/MakerMacros.h"
 #include "HeterogeneousCore/AlpakaInterface/interface/config.h"
@@ -16,4 +16,4 @@
 #define DEFINE_FWK_ALPAKA_MODULE(name) DEFINE_FWK_ALPAKA_MODULE2(ALPAKA_ACCELERATOR_NAMESPACE::name)
 #endif
 
-#endif  // HeterogeneousCore_AlpakaCore_interface_MakerMacros_h
+#endif  // HeterogeneousCore_AlpakaCore_interface_alpaka_MakerMacros_h

HeterogeneousCore/AlpakaInterface/interface/AlpakaServiceFwd.h

@@ -13,9 +13,9 @@ namespace alpaka_cuda_async {
 #endif  // ALPAKA_ACC_GPU_CUDA_ENABLED
 
 #ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-namespace alpaka_hip_async {
+namespace alpaka_rocm_async {
   class AlpakaService;
-}  // namespace alpaka_hip_async
+}  // namespace alpaka_rocm_async
 #endif  // ALPAKA_ACC_GPU_HIP_ENABLED
 
 #ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED

HeterogeneousCore/AlpakaInterface/interface/CachingAllocator.h

@@ -91,7 +91,7 @@ namespace cms::alpakatools {
     friend class alpaka_cuda_async::AlpakaService;
 #endif
 #ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-    friend class alpaka_hip_async::AlpakaService;
+    friend class alpaka_rocm_async::AlpakaService;
 #endif
 #ifdef ALPAKA_ACC_CPU_B_SEQ_T_SEQ_ENABLED
     friend class alpaka_serial_sync::AlpakaService;

HeterogeneousCore/AlpakaInterface/interface/config.h

@@ -65,7 +65,7 @@ namespace alpaka_cuda_async {
 #endif  // ALPAKA_ACC_GPU_CUDA_ENABLED
 
 #ifdef ALPAKA_ACC_GPU_HIP_ENABLED
-namespace alpaka_hip_async {
+namespace alpaka_rocm_async {
   using namespace alpaka_common;
 
   using Platform = alpaka::PltfHipRt;
@@ -79,13 +79,13 @@ namespace alpaka_hip_async {
   using Acc2D = Acc<Dim2D>;
   using Acc3D = Acc<Dim3D>;
 
-}  // namespace alpaka_hip_async
+}  // namespace alpaka_rocm_async
 
 #ifdef ALPAKA_ACCELERATOR_NAMESPACE
 #define ALPAKA_DUPLICATE_NAMESPACE
 #else
-#define ALPAKA_ACCELERATOR_NAMESPACE alpaka_hip_async
-#define ALPAKA_TYPE_SUFFIX HipAsync
+#define ALPAKA_ACCELERATOR_NAMESPACE alpaka_rocm_async
+#define ALPAKA_TYPE_SUFFIX ROCmAsync
 #endif
 
 #endif  // ALPAKA_ACC_GPU_HIP_ENABLED

HeterogeneousCore/AlpakaTest/plugins/alpaka/TestAlpakaGlobalProducer.cc

@@ -59,5 +59,5 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
 
 }  // namespace ALPAKA_ACCELERATOR_NAMESPACE
 
-#include "HeterogeneousCore/AlpakaCore/interface/MakerMacros.h"
+#include "HeterogeneousCore/AlpakaCore/interface/alpaka/MakerMacros.h"
 DEFINE_FWK_ALPAKA_MODULE(TestAlpakaGlobalProducer);

HeterogeneousCore/AlpakaTest/plugins/alpaka/TestAlpakaGlobalProducerOffset.cc

@@ -63,5 +63,5 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
 
 }  // namespace ALPAKA_ACCELERATOR_NAMESPACE
 
-#include "HeterogeneousCore/AlpakaCore/interface/MakerMacros.h"
+#include "HeterogeneousCore/AlpakaCore/interface/alpaka/MakerMacros.h"
 DEFINE_FWK_ALPAKA_MODULE(TestAlpakaGlobalProducerOffset);

HeterogeneousCore/AlpakaTest/plugins/alpaka/TestAlpakaProducer.cc

@@ -46,5 +46,5 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
 
 }  // namespace ALPAKA_ACCELERATOR_NAMESPACE
 
-#include "HeterogeneousCore/AlpakaCore/interface/MakerMacros.h"
+#include "HeterogeneousCore/AlpakaCore/interface/alpaka/MakerMacros.h"
 DEFINE_FWK_ALPAKA_MODULE(TestAlpakaProducer);

HeterogeneousCore/AlpakaTest/plugins/alpaka/TestAlpakaStreamProducer.cc

@@ -66,5 +66,5 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
 
 }  // namespace ALPAKA_ACCELERATOR_NAMESPACE
 
-#include "HeterogeneousCore/AlpakaCore/interface/MakerMacros.h"
+#include "HeterogeneousCore/AlpakaCore/interface/alpaka/MakerMacros.h"
 DEFINE_FWK_ALPAKA_MODULE(TestAlpakaStreamProducer);

HeterogeneousCore/AlpakaTest/plugins/alpaka/TestAlpakaStreamSynchronizingProducer.cc

@@ -48,5 +48,5 @@ namespace ALPAKA_ACCELERATOR_NAMESPACE {
 
 }  // namespace ALPAKA_ACCELERATOR_NAMESPACE
 
-#include "HeterogeneousCore/AlpakaCore/interface/MakerMacros.h"
+#include "HeterogeneousCore/AlpakaCore/interface/alpaka/MakerMacros.h"
 DEFINE_FWK_ALPAKA_MODULE(TestAlpakaStreamSynchronizingProducer);

HeterogeneousCore/AlpakaTest/test/writer.py

@@ -6,39 +6,19 @@
 process.source = cms.Source('EmptySource')
 
 process.load('Configuration.StandardSequences.Accelerators_cff')
+process.load('HeterogeneousCore.AlpakaCore.ProcessAcceleratorAlpaka_cfi')
 
 # enable logging for the AlpakaService and TestAlpakaAnalyzer
 process.MessageLogger.TestAlpakaAnalyzer = cms.untracked.PSet()
-process.MessageLogger.AlpakaService = cms.untracked.PSet()
 
-# enable alpaka-based heterogeneous modules
-process.AlpakaServiceCudaAsync = cms.Service('AlpakaServiceCudaAsync')
-process.AlpakaServiceSerialSync = cms.Service('AlpakaServiceSerialSync')
-
-# run the producer on a CUDA gpu (if available)
-process.testProducerCuda = cms.EDProducer('alpaka_cuda_async::TestAlpakaProducer',
-    size = cms.int32(42)
-)
-
-# run the producer on the cpu
-process.testProducerCpu = cms.EDProducer('alpaka_serial_sync::TestAlpakaProducer',
-    size = cms.int32(42)
-)
-
-# either run the producer on a CUDA gpu (if available) and copy the product to the cpu, or run the producer directly on the cpu
-#
-# TODO: the cuda case needs currently explicitly to restrict the
-# host-only data products. In the near future the constraints of
-# SwichProducer will be relaxed to not consider transient data
-# products (as defined in classes_def.xml) after which the explicit
-# set of data products can be removed.
-process.testProducer = SwitchProducerCUDA(
-    cpu = cms.EDAlias(
-        testProducerCpu = cms.VPSet(cms.PSet(type = cms.string('*')))
-    ),
-    cuda = cms.EDAlias(
-        testProducerCuda = cms.VPSet(cms.PSet(type = cms.string('128falseportabletestTestSoALayoutPortableHostCollection')))
-    )
+# either run the producer on a gpu (if available) and copy the product to the cpu, or run the producer directly on the cpu
+process.testProducer = cms.EDProducer('TestAlpakaProducer@alpaka',
+    size = cms.int32(42),
+    # alpaka.backend can be set to a specific backend to force using it, or be omitted or left empty to use the defult backend;
+    # depending on the architecture and available hardware, the supported backends are "serial_sync", "cuda_async", "rocm_async"
+    #alpaka = cms.untracked.PSet(
+    #    backend = cms.untracked.string("")
+    #)
 )
 
 # analyse the product
@@ -50,6 +30,13 @@
 process.testProducerSerial = cms.EDProducer('alpaka_serial_sync::TestAlpakaProducer',
     size = cms.int32(99)
 )
+# an alternative approach would be to use
+#process.testProducerSerial = cms.EDProducer('TestAlpakaProducer@alpaka',
+#    size = cms.int32(99),
+#    alpaka = cms.untracked.PSet(
+#        backend = cms.untracked.string("serial_sync")
+#    )
+#)
 
 # analyse the second product
 process.testAnalyzerSerial = cms.EDAnalyzer('TestAlpakaAnalyzer',
@@ -66,12 +53,9 @@
   )
 )
 
-process.producer_task = cms.Task(process.testProducerCuda, process.testProducerCpu)
-
 process.process_path = cms.Path(
     process.testProducer +
-    process.testAnalyzer,
-    process.producer_task)
+    process.testAnalyzer)
 
 process.serial_path = cms.Path(
     process.testProducerSerial +