Move preprocessor definitions to configuration header

CHANGELOG.md

@@ -21,6 +21,7 @@ See our [platform support guide](docs/platform-support.md) for a complete list o
 - Add support for SimSYCL as a SYCL implementation (#238)
 - Extend compiler support to GCC (optionally with sanitizers) and C++20 code bases (#238)
 - Add support for profiling with [Tracy](https://github.com/wolfpld/tracy), via `CELERITY_TRACY_SUPPORT` and environment variable `CELERITY_TRACY` (#267)
+- The active SYCL implementation can now be queried via `CELERITY_SYCL_IS_*` macros (#??)
 
 ### Changed
 

CMakeLists.txt

@@ -190,11 +190,14 @@ if(CELERITY_TRACY_SUPPORT)
   fetch_content_from_submodule(Tracy vendor/tracy)
 endif()
 
-configure_file(include/version.h.in include/version.h @ONLY)
+# Deprecated feature flags
+set(CELERITY_FEATURE_SCALAR_REDUCTIONS ON)
+set(CELERITY_FEATURE_SIMPLE_SCALAR_REDUCTIONS ON)
+set(CELERITY_FEATURE_LOCAL_ACCESSOR ON)
+set(CELERITY_FEATURE_UNNAMED_KERNELS ON)
 
 # Add includes to library so they show up in IDEs
 file(GLOB_RECURSE INCLUDES "${CMAKE_CURRENT_SOURCE_DIR}/include/*.h")
-list(APPEND INCLUDES "${CMAKE_CURRENT_BINARY_DIR}/include/version.h")
 
 if(CMAKE_GENERATOR STREQUAL "Ninja")
   # Force colored warnings in Ninja's output, if the compiler has -fdiagnostics-color support.
@@ -238,6 +241,21 @@ elseif(UNIX)
   set(SOURCES ${SOURCES} src/platform_specific/named_threads.unix.cc)
 endif()
 
+# Read by configure_file()
+set(CELERITY_SYCL_IS_ACPP OFF)
+set(CELERITY_SYCL_IS_DPCPP OFF)
+set(CELERITY_SYCL_IS_SIMSYCL OFF)
+if(CELERITY_SYCL_IMPL STREQUAL "AdaptiveCpp")
+  set(CELERITY_SYCL_IS_ACPP ON)
+elseif(CELERITY_SYCL_IMPL STREQUAL "DPC++")
+  set(CELERITY_SYCL_IS_DPCPP ON)
+elseif(CELERITY_SYCL_IMPL STREQUAL "SimSYCL")
+  set(CELERITY_SYCL_IS_SIMSYCL ON)
+endif()
+
+configure_file(include/version.h.in include/version.h @ONLY)
+list(APPEND INCLUDES "${CMAKE_CURRENT_BINARY_DIR}/include/version.h")
+
 add_library(
   celerity_runtime
   STATIC
@@ -288,12 +306,6 @@ target_link_libraries(celerity_runtime PUBLIC
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/src/backend)
 target_link_libraries(celerity_runtime PUBLIC celerity_backends)
 
-# Deprecated feature flags
-set(CELERITY_FEATURE_SCALAR_REDUCTIONS ON)
-set(CELERITY_FEATURE_SIMPLE_SCALAR_REDUCTIONS ON)
-set(CELERITY_FEATURE_LOCAL_ACCESSOR ON)
-set(CELERITY_FEATURE_UNNAMED_KERNELS ON)
-
 # For debug builds, we set the CELERITY_DETAIL_ENABLE_DEBUG preprocessor flag,
 # which allows Celerity to control debug functionality within headers regardless
 # of a user target's build type. (This flag is not intended to be modified by
@@ -308,17 +320,8 @@ target_compile_definitions(celerity_runtime PUBLIC
   # We still mark this as PUBLIC during builds (but not installation),
   # so that the examples and tests receive the correct flag as well.
   $<BUILD_INTERFACE:
-    $<$<CONFIG:Debug>:CELERITY_DETAIL_ENABLE_DEBUG>
+    CELERITY_DETAIL_ENABLE_DEBUG=$<CONFIG:Debug>
   >
-  CELERITY_USE_MIMALLOC=$<BOOL:${CELERITY_USE_MIMALLOC}>
-  CELERITY_FEATURE_SCALAR_REDUCTIONS=$<BOOL:${CELERITY_FEATURE_SCALAR_REDUCTIONS}>
-  CELERITY_FEATURE_SIMPLE_SCALAR_REDUCTIONS=$<BOOL:${CELERITY_FEATURE_SIMPLE_SCALAR_REDUCTIONS}>
-  CELERITY_FEATURE_LOCAL_ACCESSOR=$<BOOL:${CELERITY_FEATURE_LOCAL_ACCESSOR}>
-  CELERITY_FEATURE_UNNAMED_KERNELS=$<BOOL:${CELERITY_FEATURE_UNNAMED_KERNELS}>
-  CELERITY_DETAIL_HAS_NAMED_THREADS=$<BOOL:${CELERITY_DETAIL_HAS_NAMED_THREADS}>
-  CELERITY_ACCESSOR_BOUNDARY_CHECK=$<BOOL:${CELERITY_ACCESSOR_BOUNDARY_CHECK}>
-  CELERITY_ACCESS_PATTERN_DIAGNOSTICS=$<BOOL:${CELERITY_ACCESS_PATTERN_DIAGNOSTICS}>
-  CELERITY_TRACY_SUPPORT=$<BOOL:${CELERITY_TRACY_SUPPORT}>
 )
 
 # Collect version information from git in src/version.cc. This target is always out of date, but the timestamp
@@ -346,7 +349,7 @@ add_sycl_to_target(
 if(MSVC)
   target_compile_options(celerity_runtime PRIVATE /MP /W3)
 elseif(CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang|AppleClang")
-  target_compile_options(celerity_runtime PRIVATE -Wall -Wextra -Wno-unused-parameter -Werror=return-type -Werror=init-self)
+  target_compile_options(celerity_runtime PRIVATE -Wall -Wextra -Wno-unused-parameter -Werror=return-type -Werror=init-self -Werror=undef)
 endif()
 
 if(CMAKE_COMPILER_ID STREQUAL "GNU")

cmake/AddToTarget.cmake

@@ -13,7 +13,6 @@ if(CELERITY_SYCL_IMPL STREQUAL "DPC++")
       -fsycl
       -sycl-std=2020
       "-fsycl-targets=${CELERITY_DPCPP_TARGETS}"
-      -DCELERITY_DPCPP=1
       -Wno-sycl-strict  # -Wsycl-strict produces false-positive warnings in DPC++'s own SYCL headers as of 2022-10-06
     )
     target_compile_options(${ADD_SYCL_TARGET} PUBLIC ${DPCPP_FLAGS})
@@ -30,8 +29,6 @@ elseif(CELERITY_SYCL_IMPL STREQUAL "SimSYCL")
         "${multi_value_args}"
         ${ARGN}
     )
-    target_compile_options(${ADD_SYCL_TARGET} PUBLIC -DCELERITY_SIMSYCL=1)
-    target_link_options(${ADD_SYCL_TARGET} PUBLIC -DCELERITY_SIMSYCL=1)
   endfunction()
 endif()
 

examples/matmul/matmul.cc

@@ -2,7 +2,7 @@
 
 #include <celerity.h>
 
-#if !defined(NDEBUG) || CELERITY_SIMSYCL
+#if !defined(NDEBUG) || CELERITY_SYCL_IS_SIMSYCL
 const size_t MAT_SIZE = 128;
 #else
 const size_t MAT_SIZE = 1024;

include/accessor.h

@@ -10,6 +10,7 @@
 #include "closure_hydrator.h"
 #include "handler.h"
 #include "sycl_wrappers.h"
+#include "version.h"
 
 namespace celerity {
 

include/closure_hydrator.h

@@ -7,6 +7,7 @@
 #include "ranges.h"
 #include "sycl_wrappers.h"
 #include "types.h"
+#include "version.h"
 
 namespace celerity::detail {
 

include/handler.h

@@ -16,6 +16,7 @@
 #include "ranges.h"
 #include "task.h"
 #include "types.h"
+#include "version.h"
 #include "workaround.h"
 
 namespace celerity {

include/instruction_graph.h

@@ -4,6 +4,7 @@
 #include "launcher.h"
 #include "ranges.h"
 #include "types.h"
+#include "version.h"
 
 #include <algorithm>
 #include <cstdlib>

include/version.h.in

@@ -1,5 +1,21 @@
 #pragma once
 
+#cmakedefine01 CELERITY_SYCL_IS_ACPP
+#cmakedefine01 CELERITY_SYCL_IS_DPCPP
+#cmakedefine01 CELERITY_SYCL_IS_SIMSYCL
+
+// CELERITY_DETAIL_ENABLE_DEBUG is specified on the command line
+#cmakedefine01 CELERITY_USE_MIMALLOC
+#cmakedefine01 CELERITY_DETAIL_HAS_NAMED_THREADS
+#cmakedefine01 CELERITY_ACCESSOR_BOUNDARY_CHECK
+#cmakedefine01 CELERITY_ACCESS_PATTERN_DIAGNOSTICS
+#cmakedefine01 CELERITY_TRACY_SUPPORT
+
+#cmakedefine01 CELERITY_FEATURE_SCALAR_REDUCTIONS
+#cmakedefine01 CELERITY_FEATURE_SIMPLE_SCALAR_REDUCTIONS
+#cmakedefine01 CELERITY_FEATURE_LOCAL_ACCESSOR
+#cmakedefine01 CELERITY_FEATURE_UNNAMED_KERNELS
+
 #define CELERITY_VERSION_MAJOR @CELERITY_VERSION_MAJOR@
 #define CELERITY_VERSION_MINOR @CELERITY_VERSION_MINOR@
 #define CELERITY_VERSION_PATCH @CELERITY_VERSION_PATCH@

include/workaround.h

@@ -1,16 +1,18 @@
 #pragma once
 
+#include "version.h"
+
 #include <cassert>
 
 #include <sycl/sycl.hpp>
 
-#if defined(CELERITY_DPCPP)
+#if CELERITY_SYCL_IS_DPCPP
 #define CELERITY_WORKAROUND_DPCPP 1
 #else
 #define CELERITY_WORKAROUND_DPCPP 0
 #endif
 
-#if defined(__HIPSYCL__)
+#if CELERITY_SYCL_IS_ACPP
 #define CELERITY_WORKAROUND_ACPP 1
 #define CELERITY_WORKAROUND_VERSION_MAJOR HIPSYCL_VERSION_MAJOR
 #define CELERITY_WORKAROUND_VERSION_MINOR HIPSYCL_VERSION_MINOR
@@ -19,7 +21,7 @@
 #define CELERITY_WORKAROUND_ACPP 0
 #endif
 
-#if defined(CELERITY_SIMSYCL)
+#if CELERITY_SYCL_IS_SIMSYCL
 #define CELERITY_WORKAROUND_SIMSYCL 1
 #else
 #define CELERITY_WORKAROUND_SIMSYCL 0

src/backend/sycl_cuda_backend.cc

@@ -8,8 +8,7 @@
 #include "system_info.h"
 #include "tracy.h"
 #include "utils.h"
-#include "workaround.h"
-
+#include "version.h"
 
 #define CELERITY_STRINGIFY2(f) #f
 #define CELERITY_STRINGIFY(f) CELERITY_STRINGIFY2(f)
@@ -75,7 +74,7 @@ void nd_copy_device_async(cudaStream_t stream, const void* const source_base, vo
 //   - There are no real thread-safety guarantees. DPC++ currently does not submit kernels from background threads, but if it ever starts doing so, this will
 //     break more-or-less silently.
 // There is an open GitHub issue on the matter: https://github.com/intel/llvm/issues/13706
-#if defined(CELERITY_DPCPP)
+#if CELERITY_SYCL_IS_DPCPP
 
 struct cuda_native_event_deleter {
 	void operator()(const cudaEvent_t evt) const { CELERITY_CUDA_CHECK(cudaEventDestroy, evt); }
@@ -117,7 +116,7 @@ class cuda_event final : public async_event_impl {
 	unique_cuda_native_event m_after;
 };
 
-#endif // defined(CELERITY_DPCPP)
+#endif // CELERITY_SYCL_IS_DPCPP
 
 bool can_enable_peer_access(const int id_device, const int id_peer) {
 	// RTX 30xx and 40xx GPUs do not support peer access, but Nvidia Driver < 550 incorrectly reports that it does, causing kernel panics when enabling it
@@ -151,15 +150,15 @@ namespace celerity::detail::sycl_backend_detail {
 async_event nd_copy_device_cuda(sycl::queue& queue, const void* const source_base, void* const dest_base, const box<3>& source_box, const box<3>& dest_box,
     const region<3>& copy_region, const size_t elem_size, bool enable_profiling) //
 {
-#if defined(__HIPSYCL__)
+#if CELERITY_SYCL_IS_ACPP
 	// AdaptiveCpp provides first-class custom backend op submission without a host round-trip like sycl::queue::host_task would require.
 	auto event = queue.AdaptiveCpp_enqueue_custom_operation([=](sycl::interop_handle handle) {
 		const auto stream = handle.get_native_queue<sycl::backend::cuda>();
 		cuda_backend_detail::nd_copy_device_async(stream, source_base, dest_base, source_box, dest_box, copy_region, elem_size);
 	});
 	sycl_backend_detail::flush(queue);
 	return make_async_event<sycl_event>(std::move(event), enable_profiling);
-#elif defined(CELERITY_DPCPP)
+#elif CELERITY_SYCL_IS_DPCPP
 	// With DPC++, we must submit from the executor thread - see the comment on cuda_native_event above.
 	const auto stream = sycl::get_native<sycl::backend::ext_oneapi_cuda>(queue);
 	auto before = enable_profiling ? cuda_backend_detail::record_native_event(stream, enable_profiling) : nullptr;
@@ -171,7 +170,7 @@ async_event nd_copy_device_cuda(sycl::queue& queue, const void* const source_bas
 #endif
 }
 
-#if defined(CELERITY_DPCPP)
+#if CELERITY_SYCL_IS_DPCPP
 constexpr sycl::backend sycl_cuda_backend = sycl::backend::ext_oneapi_cuda;
 #else
 constexpr sycl::backend sycl_cuda_backend = sycl::backend::cuda;

src/config.cc

@@ -106,7 +106,7 @@ namespace detail {
 		if(parsed_and_validated_envs.ok()) {
 			// ------------------------------- CELERITY_LOG_LEVEL ---------------------------------
 
-#if defined(CELERITY_DETAIL_ENABLE_DEBUG)
+#if CELERITY_DETAIL_ENABLE_DEBUG
 			m_log_lvl = parsed_and_validated_envs.get_or(env_log_level, log_level::debug);
 #else
 			m_log_lvl = parsed_and_validated_envs.get_or(env_log_level, log_level::info);

src/live_executor.cc

@@ -11,6 +11,7 @@
 #include "tracy.h"
 #include "types.h"
 #include "utils.h"
+#include "version.h"
 
 #include <deque>
 #include <memory>

src/platform_specific/named_threads.unix.cc

@@ -1,4 +1,5 @@
 #include "named_threads.h"
+#include "version.h"
 
 #include <cassert>
 #include <type_traits>

src/runtime.cc

@@ -80,7 +80,7 @@ namespace detail {
 	}
 
 	static const char* get_build_type() {
-#if defined(CELERITY_DETAIL_ENABLE_DEBUG)
+#if CELERITY_DETAIL_ENABLE_DEBUG
 		return "debug";
 #else
 		return "release";
@@ -96,11 +96,11 @@ namespace detail {
 	}
 
 	static std::string get_sycl_version() {
-#if defined(__HIPSYCL__) || defined(__HIPSYCL_TRANSFORM__)
+#if CELERITY_SYCL_IS_ACPP
 		return fmt::format("AdaptiveCpp {}.{}.{}", HIPSYCL_VERSION_MAJOR, HIPSYCL_VERSION_MINOR, HIPSYCL_VERSION_PATCH);
-#elif CELERITY_DPCPP
+#elif CELERITY_SYCL_IS_DPCPP
 		return "DPC++ / Clang " __clang_version__;
-#elif CELERITY_SIMSYCL
+#elif CELERITY_SYCL_IS_SIMSYCL
 		return "SimSYCL " SIMSYCL_VERSION;
 #else
 #error "unknown SYCL implementation"

test/backend_tests.cc

@@ -114,7 +114,7 @@ TEST_CASE("backend allocations are properly aligned", "[backend]") {
 TEST_CASE("backend allocations are pattern-filled in debug builds", "[backend]") {
 	test_utils::allow_backend_fallback_warnings();
 
-#if defined(CELERITY_DETAIL_ENABLE_DEBUG)
+#if CELERITY_DETAIL_ENABLE_DEBUG
 	const auto [backend_type, backend, sycl_devices] = generate_backends_with_devices();
 	CAPTURE(backend_type, sycl_devices);