Add ETI macros to simplify instantiations for element types and devic…

…es in communication (#995)
eth-cscs · Oct 6, 2023 · 87ec68a · 87ec68a
1 parent 3bfd3ee
commit 87ec68a
Show file tree

Hide file tree

Showing 9 changed files with 93 additions and 212 deletions.
diff --git a/include/dlaf/common/eti.h b/include/dlaf/common/eti.h
@@ -0,0 +1,49 @@
+//
+// Distributed Linear Algebra with Future (DLAF)
+//
+// Copyright (c) 2018-2023, ETH Zurich
+// All rights reserved.
+//
+// Please, refer to the LICENSE file in the root directory.
+// SPDX-License-Identifier: BSD-3-Clause
+//
+
+/// @file
+
+#pragma once
+
+#include <complex>
+
+#include <dlaf/types.h>
+
+#ifdef DLAF_WITH_GPU
+#define DLAF_EXPAND_ETI_SDCZ_DEVICE(ETI_MACRO, KWORD)  \
+  ETI_MACRO(KWORD, float, Device::CPU);                \
+  ETI_MACRO(KWORD, double, Device::CPU);               \
+  ETI_MACRO(KWORD, std::complex<float>, Device::CPU);  \
+  ETI_MACRO(KWORD, std::complex<double>, Device::CPU); \
+  ETI_MACRO(KWORD, float, Device::GPU);                \
+  ETI_MACRO(KWORD, double, Device::GPU);               \
+  ETI_MACRO(KWORD, std::complex<float>, Device::GPU);  \
+  ETI_MACRO(KWORD, std::complex<double>, Device::GPU);
+#define DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(ETI_MACRO, KWORD, ...)  \
+  ETI_MACRO(KWORD, float, Device::CPU, __VA_ARGS__);                \
+  ETI_MACRO(KWORD, double, Device::CPU, __VA_ARGS__);               \
+  ETI_MACRO(KWORD, std::complex<float>, Device::CPU, __VA_ARGS__);  \
+  ETI_MACRO(KWORD, std::complex<double>, Device::CPU, __VA_ARGS__); \
+  ETI_MACRO(KWORD, float, Device::GPU, __VA_ARGS__);                \
+  ETI_MACRO(KWORD, double, Device::GPU, __VA_ARGS__);               \
+  ETI_MACRO(KWORD, std::complex<float>, Device::GPU, __VA_ARGS__);  \
+  ETI_MACRO(KWORD, std::complex<double>, Device::GPU, __VA_ARGS__);
+#else
+#define DLAF_EXPAND_ETI_SDCZ_DEVICE(ETI_MACRO, KWORD) \
+  ETI_MACRO(KWORD, float, Device::CPU);               \
+  ETI_MACRO(KWORD, double, Device::CPU);              \
+  ETI_MACRO(KWORD, std::complex<float>, Device::CPU); \
+  ETI_MACRO(KWORD, std::complex<double>, Device::CPU);
+#define DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(ETI_MACRO, KWORD, ...) \
+  ETI_MACRO(KWORD, float, Device::CPU, __VA_ARGS__);               \
+  ETI_MACRO(KWORD, double, Device::CPU, __VA_ARGS__);              \
+  ETI_MACRO(KWORD, std::complex<float>, Device::CPU, __VA_ARGS__); \
+  ETI_MACRO(KWORD, std::complex<double>, Device::CPU, __VA_ARGS__);
+#endif
diff --git a/include/dlaf/communication/kernels/all_reduce.h b/include/dlaf/communication/kernels/all_reduce.h
@@ -18,6 +18,7 @@
 
 #include <pika/execution.hpp>
 
+#include <dlaf/common/eti.h>
 #include <dlaf/common/pipeline.h>
 #include <dlaf/communication/communicator.h>
 #include <dlaf/matrix/matrix.h>
@@ -43,6 +44,7 @@ template <class T, Device D>
       MPI_Op reduce_op, dlaf::matrix::ReadOnlyTileSender<Type, Device> tile_in,                        \
       dlaf::matrix::ReadWriteTileSender<Type, Device> tile_out)
 
+DLAF_EXPAND_ETI_SDCZ_DEVICE(DLAF_SCHEDULE_ALL_REDUCE_ETI, extern);
 DLAF_SCHEDULE_ALL_REDUCE_ETI(extern, int, Device::CPU);
 
 /// Schedule an in-place all reduce.
@@ -60,16 +62,6 @@ template <class T, Device D>
       pika::execution::experimental::unique_any_sender<common::Pipeline<Communicator>::Wrapper> pcomm, \
       MPI_Op reduce_op, dlaf::matrix::ReadWriteTileSender<Type, Device> tile)
 
+DLAF_EXPAND_ETI_SDCZ_DEVICE(DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI, extern);
 DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, int, Device::CPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, float, Device::CPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, double, Device::CPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, std::complex<float>, Device::CPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, std::complex<double>, Device::CPU);
-
-#ifdef DLAF_WITH_GPU
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, float, Device::GPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, double, Device::GPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, std::complex<float>, Device::GPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(extern, std::complex<double>, Device::GPU);
-#endif
 }
diff --git a/include/dlaf/communication/kernels/broadcast.h b/include/dlaf/communication/kernels/broadcast.h
@@ -21,6 +21,7 @@
 #include <dlaf/common/assert.h>
 #include <dlaf/common/callable_object.h>
 #include <dlaf/common/data.h>
+#include <dlaf/common/eti.h>
 #include <dlaf/common/pipeline.h>
 #include <dlaf/communication/communicator.h>
 #include <dlaf/communication/message.h>
@@ -71,23 +72,14 @@ template <class T, Device D, class Comm>
       pika::execution::experimental::unique_any_sender<Comm> pcomm,                    \
       dlaf::matrix::ReadOnlyTileSender<Type, Device> tile)
 
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, SizeType, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, float, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, double, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, std::complex<float>, Device::CPU,
-                             common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, std::complex<double>, Device::CPU,
-                             common::Pipeline<Communicator>::Wrapper);
+// clang-format off
+DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(DLAF_SCHEDULE_SEND_BCAST_ETI, extern, common::Pipeline<Communicator>::Wrapper);
 
+DLAF_SCHEDULE_SEND_BCAST_ETI(extern, SizeType, Device::CPU, common::Pipeline<Communicator>::Wrapper);
 #ifdef DLAF_WITH_GPU
 DLAF_SCHEDULE_SEND_BCAST_ETI(extern, SizeType, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, float, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, double, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, std::complex<float>, Device::GPU,
-                             common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(extern, std::complex<double>, Device::GPU,
-                             common::Pipeline<Communicator>::Wrapper);
 #endif
+// clang-format on
 
 /// Schedule a broadcast receive.
 ///
@@ -104,21 +96,12 @@ template <class T, Device D, class Comm>
       pika::execution::experimental::unique_any_sender<Comm> pcomm, comm::IndexT_MPI root_rank, \
       dlaf::matrix::ReadWriteTileSender<Type, Device> tile)
 
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, SizeType, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, float, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, double, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, std::complex<float>, Device::CPU,
-                             common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, std::complex<double>, Device::CPU,
-                             common::Pipeline<Communicator>::Wrapper);
+// clang-format off
+DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(DLAF_SCHEDULE_RECV_BCAST_ETI, extern, common::Pipeline<Communicator>::Wrapper);
 
+DLAF_SCHEDULE_RECV_BCAST_ETI(extern, SizeType, Device::CPU, common::Pipeline<Communicator>::Wrapper);
 #ifdef DLAF_WITH_GPU
 DLAF_SCHEDULE_RECV_BCAST_ETI(extern, SizeType, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, float, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, double, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, std::complex<float>, Device::GPU,
-                             common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(extern, std::complex<double>, Device::GPU,
-                             common::Pipeline<Communicator>::Wrapper);
 #endif
+// clang-format on
 }
diff --git a/include/dlaf/communication/kernels/p2p.h b/include/dlaf/communication/kernels/p2p.h
@@ -18,6 +18,7 @@
 
 #include <pika/execution.hpp>
 
+#include <dlaf/common/eti.h>
 #include <dlaf/common/pipeline.h>
 #include <dlaf/communication/communicator.h>
 #include <dlaf/matrix/tile.h>
@@ -33,31 +34,10 @@ template <class T, Device D, class Comm>
       pika::execution::experimental::unique_any_sender<Comm> pcomm, IndexT_MPI dest, IndexT_MPI tag, \
       dlaf::matrix::ReadOnlyTileSender<Type, Device> tile)
 
-DLAF_SCHEDULE_SEND_ETI(extern, float, Device::CPU, Communicator);
-DLAF_SCHEDULE_SEND_ETI(extern, double, Device::CPU, Communicator);
-DLAF_SCHEDULE_SEND_ETI(extern, std::complex<float>, Device::CPU, Communicator);
-DLAF_SCHEDULE_SEND_ETI(extern, std::complex<double>, Device::CPU, Communicator);
-
-DLAF_SCHEDULE_SEND_ETI(extern, float, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_ETI(extern, double, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_ETI(extern, std::complex<float>, Device::CPU,
-                       common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_ETI(extern, std::complex<double>, Device::CPU,
-                       common::Pipeline<Communicator>::Wrapper);
-
-#ifdef DLAF_WITH_GPU
-DLAF_SCHEDULE_SEND_ETI(extern, float, Device::GPU, Communicator);
-DLAF_SCHEDULE_SEND_ETI(extern, double, Device::GPU, Communicator);
-DLAF_SCHEDULE_SEND_ETI(extern, std::complex<float>, Device::GPU, Communicator);
-DLAF_SCHEDULE_SEND_ETI(extern, std::complex<double>, Device::GPU, Communicator);
-
-DLAF_SCHEDULE_SEND_ETI(extern, float, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_ETI(extern, double, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_ETI(extern, std::complex<float>, Device::GPU,
-                       common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_ETI(extern, std::complex<double>, Device::GPU,
-                       common::Pipeline<Communicator>::Wrapper);
-#endif
+// clang-format off
+DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(DLAF_SCHEDULE_SEND_ETI, extern, Communicator);
+DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(DLAF_SCHEDULE_SEND_ETI, extern, common::Pipeline<Communicator>::Wrapper);
+// clang-format on
 
 template <class T, Device D, class Comm>
 [[nodiscard]] dlaf::matrix::ReadWriteTileSender<T, D> scheduleRecv(
@@ -69,29 +49,8 @@ template <class T, Device D, class Comm>
       pika::execution::experimental::unique_any_sender<Comm> pcomm, IndexT_MPI source, IndexT_MPI tag, \
       dlaf::matrix::ReadWriteTileSender<Type, Device> tile)
 
-DLAF_SCHEDULE_RECV_ETI(extern, float, Device::CPU, Communicator);
-DLAF_SCHEDULE_RECV_ETI(extern, double, Device::CPU, Communicator);
-DLAF_SCHEDULE_RECV_ETI(extern, std::complex<float>, Device::CPU, Communicator);
-DLAF_SCHEDULE_RECV_ETI(extern, std::complex<double>, Device::CPU, Communicator);
-
-DLAF_SCHEDULE_RECV_ETI(extern, float, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_ETI(extern, double, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_ETI(extern, std::complex<float>, Device::CPU,
-                       common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_ETI(extern, std::complex<double>, Device::CPU,
-                       common::Pipeline<Communicator>::Wrapper);
-
-#ifdef DLAF_WITH_GPU
-DLAF_SCHEDULE_RECV_ETI(extern, float, Device::GPU, Communicator);
-DLAF_SCHEDULE_RECV_ETI(extern, double, Device::GPU, Communicator);
-DLAF_SCHEDULE_RECV_ETI(extern, std::complex<float>, Device::GPU, Communicator);
-DLAF_SCHEDULE_RECV_ETI(extern, std::complex<double>, Device::GPU, Communicator);
-
-DLAF_SCHEDULE_RECV_ETI(extern, float, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_ETI(extern, double, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_ETI(extern, std::complex<float>, Device::GPU,
-                       common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_ETI(extern, std::complex<double>, Device::GPU,
-                       common::Pipeline<Communicator>::Wrapper);
-#endif
+// clang-format off
+DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(DLAF_SCHEDULE_RECV_ETI, extern, Communicator);
+DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(DLAF_SCHEDULE_RECV_ETI, extern, common::Pipeline<Communicator>::Wrapper);
+// clang-format on
 }
diff --git a/include/dlaf/communication/kernels/reduce.h b/include/dlaf/communication/kernels/reduce.h
@@ -19,6 +19,7 @@
 #include <pika/execution.hpp>
 
 #include <dlaf/common/data.h>
+#include <dlaf/common/eti.h>
 #include <dlaf/common/pipeline.h>
 #include <dlaf/communication/communicator.h>
 #include <dlaf/communication/message.h>
@@ -40,18 +41,8 @@ template <class T, Device D>
       pika::execution::experimental::unique_any_sender<common::Pipeline<Communicator>::Wrapper> pcomm, \
       MPI_Op reduce_op, dlaf::matrix::ReadWriteTileSender<Type, Device> tile)
 
+DLAF_EXPAND_ETI_SDCZ_DEVICE(DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI, extern);
 DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, int, Device::CPU);
-DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, float, Device::CPU);
-DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, double, Device::CPU);
-DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, std::complex<float>, Device::CPU);
-DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, std::complex<double>, Device::CPU);
-
-#ifdef DLAF_WITH_GPU
-DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, float, Device::GPU);
-DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, double, Device::GPU);
-DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, std::complex<float>, Device::GPU);
-DLAF_SCHEDULE_REDUCE_RECV_IN_PLACE_ETI(extern, std::complex<double>, Device::GPU);
-#endif
 
 /// Schedule a reduction send.
 ///
@@ -69,16 +60,6 @@ template <class T, Device D>
       comm::IndexT_MPI rank_root, MPI_Op reduce_op,                                                    \
       dlaf::matrix::ReadOnlyTileSender<Type, Device> tile)
 
+DLAF_EXPAND_ETI_SDCZ_DEVICE(DLAF_SCHEDULE_REDUCE_SEND_ETI, extern);
 DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, int, Device::CPU);
-DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, float, Device::CPU);
-DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, double, Device::CPU);
-DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, std::complex<float>, Device::CPU);
-DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, std::complex<double>, Device::CPU);
-
-#ifdef DLAF_WITH_GPU
-DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, float, Device::GPU);
-DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, double, Device::GPU);
-DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, std::complex<float>, Device::GPU);
-DLAF_SCHEDULE_REDUCE_SEND_ETI(extern, std::complex<double>, Device::GPU);
-#endif
 }
diff --git a/src/communication/kernels/all_reduce.cpp b/src/communication/kernels/all_reduce.cpp
@@ -16,6 +16,7 @@
 #include <pika/execution.hpp>
 
 #include <dlaf/common/callable_object.h>
+#include <dlaf/common/eti.h>
 #include <dlaf/common/pipeline.h>
 #include <dlaf/communication/communicator.h>
 #include <dlaf/communication/kernels/all_reduce.h>
@@ -115,6 +116,7 @@ template <class T, Device D>
                                      std::move(tile_out));
 }
 
+DLAF_EXPAND_ETI_SDCZ_DEVICE(DLAF_SCHEDULE_ALL_REDUCE_ETI, );
 DLAF_SCHEDULE_ALL_REDUCE_ETI(, int, Device::CPU);
 
 template <class T, Device D>
@@ -150,18 +152,6 @@ template <class T, Device D>
                            RequireContiguous::Yes>(std::move(tile), std::move(all_reduce_in_place));
 }
 
-// TODO: This is only for a test (test_collective_async)
+DLAF_EXPAND_ETI_SDCZ_DEVICE(DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI, );
 DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, int, Device::CPU);
-
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, float, Device::CPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, double, Device::CPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, std::complex<float>, Device::CPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, std::complex<double>, Device::CPU);
-
-#ifdef DLAF_WITH_GPU
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, float, Device::GPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, double, Device::GPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, std::complex<float>, Device::GPU);
-DLAF_SCHEDULE_ALL_REDUCE_IN_PLACE_ETI(, std::complex<double>, Device::GPU);
-#endif
 }
diff --git a/src/communication/kernels/broadcast.cpp b/src/communication/kernels/broadcast.cpp
@@ -18,6 +18,7 @@
 #include <dlaf/common/assert.h>
 #include <dlaf/common/callable_object.h>
 #include <dlaf/common/data.h>
+#include <dlaf/common/eti.h>
 #include <dlaf/communication/communicator.h>
 #include <dlaf/communication/kernels/broadcast.h>
 #include <dlaf/communication/message.h>
@@ -60,23 +61,14 @@ template <class T, Device D, class Comm>
   return internal::scheduleSendBcast(std::move(pcomm), std::move(tile));
 }
 
-DLAF_SCHEDULE_SEND_BCAST_ETI(, SizeType, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(, float, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(, double, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(, std::complex<float>, Device::CPU,
-                             common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(, std::complex<double>, Device::CPU,
-                             common::Pipeline<Communicator>::Wrapper);
+// clang-format off
+DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(DLAF_SCHEDULE_SEND_BCAST_ETI, , common::Pipeline<Communicator>::Wrapper);
 
+DLAF_SCHEDULE_SEND_BCAST_ETI(, SizeType, Device::CPU, common::Pipeline<Communicator>::Wrapper);
 #ifdef DLAF_WITH_GPU
 DLAF_SCHEDULE_SEND_BCAST_ETI(, SizeType, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(, float, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(, double, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(, std::complex<float>, Device::GPU,
-                             common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_SEND_BCAST_ETI(, std::complex<double>, Device::GPU,
-                             common::Pipeline<Communicator>::Wrapper);
 #endif
+// clang-format on
 
 template <class T, Device D, class Comm>
 [[nodiscard]] dlaf::matrix::ReadWriteTileSender<T, D> scheduleRecvBcast(
@@ -112,21 +104,12 @@ template <class T, Device D, class Comm>
                            RequireContiguous::No>(std::move(tile), std::move(recv));
 }
 
-DLAF_SCHEDULE_RECV_BCAST_ETI(, SizeType, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(, float, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(, double, Device::CPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(, std::complex<float>, Device::CPU,
-                             common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(, std::complex<double>, Device::CPU,
-                             common::Pipeline<Communicator>::Wrapper);
+// clang-format off
+DLAF_EXPAND_ETI_SDCZ_DEVICE_VA_ARGS(DLAF_SCHEDULE_RECV_BCAST_ETI, , common::Pipeline<Communicator>::Wrapper);
 
+DLAF_SCHEDULE_RECV_BCAST_ETI(, SizeType, Device::CPU, common::Pipeline<Communicator>::Wrapper);
 #ifdef DLAF_WITH_GPU
 DLAF_SCHEDULE_RECV_BCAST_ETI(, SizeType, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(, float, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(, double, Device::GPU, common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(, std::complex<float>, Device::GPU,
-                             common::Pipeline<Communicator>::Wrapper);
-DLAF_SCHEDULE_RECV_BCAST_ETI(, std::complex<double>, Device::GPU,
-                             common::Pipeline<Communicator>::Wrapper);
 #endif
+// clang-format on
 }