Handler-less kernel submit API

sycl/cmake/modules/AddSYCLUnitTest.cmake

@@ -1,6 +1,6 @@
 # Internal function to create SYCL unit tests with code reuse
-# add_sycl_unittest_internal(test_dirname SHARED|OBJECT is_preview file1.cpp, file2.cpp ...)
-function(add_sycl_unittest_internal test_dirname link_variant is_preview)
+# add_sycl_unittest_internal(test_dirname SHARED|OBJECT is_preview is_no_cgh file1.cpp, file2.cpp ...)
+function(add_sycl_unittest_internal test_dirname link_variant is_preview is_no_cgh)
   # Enable exception handling for these unit tests
   set(LLVM_REQUIRES_EH ON)
   set(LLVM_REQUIRES_RTTI ON)
@@ -34,7 +34,11 @@ function(add_sycl_unittest_internal test_dirname link_variant is_preview)
   # Chaning CMAKE_CURRENT_BINARY_DIR should not affect this variable in its
   # parent scope.
   if (${is_preview})
-    set(CMAKE_CURRENT_BINARY_DIR "${CMAKE_CURRENT_BINARY_DIR}/Preview")
+      set(CMAKE_CURRENT_BINARY_DIR "${CMAKE_CURRENT_BINARY_DIR}/Preview")
+  endif()
+
+  if (${is_no_cgh})
+    set(CMAKE_CURRENT_BINARY_DIR "${CMAKE_CURRENT_BINARY_DIR}/NoCGH")
   endif()
 
   if ("${link_variant}" MATCHES "SHARED")
@@ -65,6 +69,18 @@ function(add_sycl_unittest_internal test_dirname link_variant is_preview)
     set(sycl_cache_suffix "_preview")
   endif()
 
+  if (${is_no_cgh})
+    set(sycl_cache_suffix "_no_cgh")
+  endif()
+
+  if (${is_no_cgh})
+    target_compile_definitions(
+      ${test_dirname}
+      PRIVATE
+        __DPCPP_ENABLE_UNFINISHED_NO_CGH_SUBMIT
+    )
+  endif()
+
   if (SYCL_ENABLE_XPTI_TRACING)
     target_compile_definitions(${test_dirname}
       PRIVATE XPTI_ENABLE_INSTRUMENTATION XPTI_STATIC_LIBRARY)
@@ -150,7 +166,6 @@ function(add_sycl_unittest_internal test_dirname link_variant is_preview)
         -Wno-inconsistent-missing-override
     )
   endif()
-
   target_compile_definitions(${test_dirname} PRIVATE SYCL_DISABLE_FSYCL_SYCLHPP_WARNING)
 endfunction()
 
@@ -160,6 +175,7 @@ endfunction()
 # the SYCL preview features enabled.
 # Produces two binaries, named `basename(test_name_prefix_non_preview)` and `basename(test_name_prefix_preview)`
 macro(add_sycl_unittest test_name_prefix link_variant)
-  add_sycl_unittest_internal(${test_name_prefix}_non_preview ${link_variant} FALSE ${ARGN})
-  add_sycl_unittest_internal(${test_name_prefix}_preview ${link_variant} TRUE ${ARGN})
+  add_sycl_unittest_internal(${test_name_prefix}_non_preview ${link_variant} FALSE FALSE ${ARGN})
+  add_sycl_unittest_internal(${test_name_prefix}_no_cgh ${link_variant} FALSE TRUE ${ARGN})
+  add_sycl_unittest_internal(${test_name_prefix}_preview ${link_variant} TRUE FALSE ${ARGN})
 endmacro()

sycl/include/sycl/ext/oneapi/experimental/enqueue_functions.hpp

@@ -109,6 +109,26 @@ event submit_with_event_impl(const queue &Q, PropertiesT Props,
                              const sycl::detail::code_location &CodeLoc) {
   return Q.submit_with_event(Props, detail::type_erased_cgfo_ty{CGF}, CodeLoc);
 }
+
+template <typename KernelName, typename PropertiesT, typename KernelType,
+          int Dims>
+void submit_kernel_direct_impl(const queue &Q, PropertiesT Props,
+                               nd_range<Dims> Range,
+                               const KernelType &KernelFunc,
+                               const sycl::detail::code_location &CodeLoc) {
+  Q.submit_kernel_direct_without_event<KernelName, PropertiesT, KernelType,
+                                       Dims>(Props, Range, KernelFunc, CodeLoc);
+}
+
+template <typename KernelName, typename PropertiesT, typename KernelType,
+          int Dims>
+event submit_kernel_direct_with_event_impl(
+    const queue &Q, PropertiesT Props, nd_range<Dims> Range,
+    const KernelType &KernelFunc, const sycl::detail::code_location &CodeLoc) {
+  return Q.submit_kernel_direct_with_event<KernelName, PropertiesT, KernelType,
+                                           Dims>(Props, Range, KernelFunc,
+                                                 CodeLoc);
+}
 } // namespace detail
 
 template <typename CommandGroupFunc, typename PropertiesT>
@@ -126,6 +146,17 @@ void submit(const queue &Q, CommandGroupFunc &&CGF,
   submit(Q, empty_properties_t{}, std::forward<CommandGroupFunc>(CGF), CodeLoc);
 }
 
+template <typename KernelName = sycl::detail::auto_name, typename PropertiesT,
+          typename KernelType, int Dims>
+void submit(const queue &Q, PropertiesT Props, nd_range<Dims> Range,
+            const KernelType &KernelFunc,
+            const sycl::detail::code_location &CodeLoc =
+                sycl::detail::code_location::current()) {
+  sycl::ext::oneapi::experimental::detail::submit_kernel_direct_impl<
+      KernelName, PropertiesT, KernelType, Dims>(Q, Props, Range, KernelFunc,
+                                                 CodeLoc);
+}
+
 template <typename CommandGroupFunc, typename PropertiesT>
 event submit_with_event(const queue &Q, PropertiesT Props,
                         CommandGroupFunc &&CGF,
@@ -143,6 +174,18 @@ event submit_with_event(const queue &Q, CommandGroupFunc &&CGF,
                            std::forward<CommandGroupFunc>(CGF), CodeLoc);
 }
 
+template <typename KernelName = sycl::detail::auto_name, typename PropertiesT,
+          typename KernelType, int Dims>
+event submit_with_event(const queue &Q, PropertiesT Props, nd_range<Dims> Range,
+                        const KernelType &KernelFunc,
+                        const sycl::detail::code_location &CodeLoc =
+                            sycl::detail::code_location::current()) {
+  return sycl::ext::oneapi::experimental::detail::
+      submit_kernel_direct_with_event_impl<KernelName, PropertiesT, KernelType,
+                                           Dims>(Q, Props, Range, KernelFunc,
+                                                 CodeLoc);
+}
+
 template <typename KernelName = sycl::detail::auto_name, typename KernelType>
 void single_task(handler &CGH, const KernelType &KernelObj) {
   CGH.single_task<KernelName>(KernelObj);
@@ -259,10 +302,21 @@ template <typename KernelName = sycl::detail::auto_name, int Dimensions,
           typename KernelType, typename... ReductionsT>
 void nd_launch(queue Q, nd_range<Dimensions> Range, const KernelType &KernelObj,
                ReductionsT &&...Reductions) {
+#ifdef __DPCPP_ENABLE_UNFINISHED_NO_CGH_SUBMIT
+  if constexpr (sizeof...(ReductionsT) == 0) {
+    submit<KernelName>(std::move(Q), empty_properties_t{}, Range, KernelObj);
+  } else {
+    submit(std::move(Q), [&](handler &CGH) {
+      nd_launch<KernelName>(CGH, Range, KernelObj,
+                            std::forward<ReductionsT>(Reductions)...);
+    });
+  }
+#else
   submit(std::move(Q), [&](handler &CGH) {
     nd_launch<KernelName>(CGH, Range, KernelObj,
                           std::forward<ReductionsT>(Reductions)...);
   });
+#endif
 }
 
 template <typename KernelName = sycl::detail::auto_name, int Dimensions,
@@ -283,10 +337,25 @@ template <typename KernelName = sycl::detail::auto_name, int Dimensions,
           typename Properties, typename KernelType, typename... ReductionsT>
 void nd_launch(queue Q, launch_config<nd_range<Dimensions>, Properties> Config,
                const KernelType &KernelObj, ReductionsT &&...Reductions) {
+#ifdef __DPCPP_ENABLE_UNFINISHED_NO_CGH_SUBMIT
+  if constexpr (sizeof...(ReductionsT) == 0) {
+    ext::oneapi::experimental::detail::LaunchConfigAccess<nd_range<Dimensions>,
+                                                          Properties>
+        ConfigAccess(Config);
+    submit<KernelName>(std::move(Q), ConfigAccess.getProperties(),
+                       ConfigAccess.getRange(), KernelObj);
+  } else {
+    submit(std::move(Q), [&](handler &CGH) {
+      nd_launch<KernelName>(CGH, Config, KernelObj,
+                            std::forward<ReductionsT>(Reductions)...);
+    });
+  }
+#else
   submit(std::move(Q), [&](handler &CGH) {
     nd_launch<KernelName>(CGH, Config, KernelObj,
                           std::forward<ReductionsT>(Reductions)...);
   });
+#endif
 }
 
 template <int Dimensions, typename... ArgsT>

sycl/include/sycl/khr/free_function_commands.hpp

@@ -153,27 +153,42 @@ void launch_grouped(const queue &q, range<1> r, range<1> size,
                     const KernelType &k,
                     const sycl::detail::code_location &codeLoc =
                         sycl::detail::code_location::current()) {
+#ifdef __DPCPP_ENABLE_UNFINISHED_NO_CGH_SUBMIT
+  submit(q, ext::oneapi::experimental::empty_properties_t{},
+         nd_range<1>(r, size), k);
+#else
   submit(
       q, [&](handler &h) { launch_grouped<KernelType>(h, r, size, k); },
       codeLoc);
+#endif
 }
 template <typename KernelType>
 void launch_grouped(const queue &q, range<2> r, range<2> size,
                     const KernelType &k,
                     const sycl::detail::code_location &codeLoc =
                         sycl::detail::code_location::current()) {
+#ifdef __DPCPP_ENABLE_UNFINISHED_NO_CGH_SUBMIT
+  submit(q, ext::oneapi::experimental::empty_properties_t{},
+         nd_range<2>(r, size), k);
+#else
   submit(
       q, [&](handler &h) { launch_grouped<KernelType>(h, r, size, k); },
       codeLoc);
+#endif
 }
 template <typename KernelType>
 void launch_grouped(const queue &q, range<3> r, range<3> size,
                     const KernelType &k,
                     const sycl::detail::code_location &codeLoc =
                         sycl::detail::code_location::current()) {
+#ifdef __DPCPP_ENABLE_UNFINISHED_NO_CGH_SUBMIT
+  submit(q, ext::oneapi::experimental::empty_properties_t{},
+         nd_range<3>(r, size), k);
+#else
   submit(
       q, [&](handler &h) { launch_grouped<KernelType>(h, r, size, k); },
       codeLoc);
+#endif
 }
 
 template <typename... Args>