fix getFunctionAttributes for the SYCL backend

alpaka-group · Aug 7, 2024 · 106a497 · 106a497
1 parent f36e115
commit 106a497
Show file tree

Hide file tree

Showing 4 changed files with 93 additions and 27 deletions.
diff --git a/include/alpaka/kernel/TaskKernelCpuSycl.hpp b/include/alpaka/kernel/TaskKernelCpuSycl.hpp
@@ -15,6 +15,37 @@ namespace alpaka
 
     template<typename TDim, typename TIdx, typename TKernelFnObj, typename... TArgs>
     using TaskKernelCpuSycl = TaskKernelGenericSycl<AccCpuSycl<TDim, TIdx>, TDim, TIdx, TKernelFnObj, TArgs...>;
+
+    namespace trait
+    {
+        //! \brief Specialisation of the class template FunctionAttributes
+        //! \tparam TDev The device type.
+        //! \tparam TDim The dimensionality of the accelerator device properties.
+        //! \tparam TIdx The idx type of the accelerator device properties.
+        //! \tparam TKernelFn Kernel function object type.
+        //! \tparam TArgs Kernel function object argument types as a parameter pack.
+        template<typename TDev, typename TDim, typename TIdx, typename TKernelFn, typename... TArgs>
+        struct FunctionAttributes<AccCpuSycl<TDim, TIdx>, TDev, KernelBundle<TKernelFn, TArgs...>>
+        {
+            //! \param dev The device instance
+            //! \param kernelBundle Kernel bundeled with it's arguments. The function attributes of this kernel will be
+            //! determined. Max threads per block is one of the attributes.
+            //! \return KernelFunctionAttributes instance. The default version always returns an instance with zero
+            //! fields. For CPU, the field of max threads allowed by kernel function for the block is 1.
+            ALPAKA_FN_HOST static auto getFunctionAttributes(
+                TDev const& dev,
+                [[maybe_unused]] KernelBundle<TKernelFn, TArgs...> const& kernelBundle)
+                -> alpaka::KernelFunctionAttributes
+            {
+                alpaka::KernelFunctionAttributes kernelFunctionAttributes;
+
+                // set function properties for maxThreadsPerBlock to device properties
+                auto const& props = alpaka::getAccDevProps<AccCpuSycl<TDim, TIdx>>(dev);
+                kernelFunctionAttributes.maxThreadsPerBlock = static_cast<int>(props.m_blockThreadCountMax);
+                return kernelFunctionAttributes;
+            }
+        };
+    } // namespace trait
 } // namespace alpaka
 
 #endif
diff --git a/include/alpaka/kernel/TaskKernelFpgaSyclIntel.hpp b/include/alpaka/kernel/TaskKernelFpgaSyclIntel.hpp
@@ -16,6 +16,37 @@ namespace alpaka
     template<typename TDim, typename TIdx, typename TKernelFnObj, typename... TArgs>
     using TaskKernelFpgaSyclIntel
         = TaskKernelGenericSycl<AccFpgaSyclIntel<TDim, TIdx>, TDim, TIdx, TKernelFnObj, TArgs...>;
+
+    namespace trait
+    {
+        //! \brief Specialisation of the class template FunctionAttributes
+        //! \tparam TDev The device type.
+        //! \tparam TDim The dimensionality of the accelerator device properties.
+        //! \tparam TIdx The idx type of the accelerator device properties.
+        //! \tparam TKernelFn Kernel function object type.
+        //! \tparam TArgs Kernel function object argument types as a parameter pack.
+        template<typename TDev, typename TDim, typename TIdx, typename TKernelFn, typename... TArgs>
+        struct FunctionAttributes<AccFpgaSyclIntel<TDim, TIdx>, TDev, KernelBundle<TKernelFn, TArgs...>>
+        {
+            //! \param dev The device instance
+            //! \param kernelBundle Kernel bundeled with it's arguments. The function attributes of this kernel will be
+            //! determined. Max threads per block is one of the attributes.
+            //! \return KernelFunctionAttributes instance. The default version always returns an instance with zero
+            //! fields. For CPU, the field of max threads allowed by kernel function for the block is 1.
+            ALPAKA_FN_HOST static auto getFunctionAttributes(
+                TDev const& dev,
+                [[maybe_unused]] KernelBundle<TKernelFn, TArgs...> const& kernelBundle)
+                -> alpaka::KernelFunctionAttributes
+            {
+                alpaka::KernelFunctionAttributes kernelFunctionAttributes;
+
+                // set function properties for maxThreadsPerBlock to device properties
+                auto const& props = alpaka::getAccDevProps<AccFpgaSyclIntel<TDim, TIdx>>(dev);
+                kernelFunctionAttributes.maxThreadsPerBlock = static_cast<int>(props.m_blockThreadCountMax);
+                return kernelFunctionAttributes;
+            }
+        };
+    } // namespace trait
 } // namespace alpaka
 
 #endif
diff --git a/include/alpaka/kernel/TaskKernelGenericSycl.hpp b/include/alpaka/kernel/TaskKernelGenericSycl.hpp
@@ -279,33 +279,6 @@ namespace alpaka::trait
         using type = TIdx;
     };
 
-    //! \brief Specialisation of the class template FunctionAttributes
-    //! \tparam TDev The device type.
-    //! \tparam TDim The dimensionality of the accelerator device properties.
-    //! \tparam TIdx The idx type of the accelerator device properties.
-    //! \tparam TKernelFn Kernel function object type.
-    //! \tparam TArgs Kernel function object argument types as a parameter pack.
-    template<typename TDev, typename TDim, typename TIdx, typename TKernelFn, typename... TArgs>
-    struct FunctionAttributes<AccGenericSycl<TDim, TIdx>, TDev, KernelBundle<TKernelFn, TArgs...>>
-    {
-        //! \param dev The device instance
-        //! \param kernelBundle Kernel bundeled with it's arguments. The function attributes of this kernel will be
-        //! determined. Max threads per block is one of the attributes.
-        //! \return KernelFunctionAttributes instance. The default version always returns an instance with zero
-        //! fields. For CPU, the field of max threads allowed by kernel function for the block is 1.
-        ALPAKA_FN_HOST static auto getFunctionAttributes(
-            TDev const& dev,
-            [[maybe_unused]] KernelBundle<TKernelFn, TArgs...> const& kernelBundle) -> alpaka::KernelFunctionAttributes
-        {
-            alpaka::KernelFunctionAttributes kernelFunctionAttributes;
-
-            // set function properties for maxThreadsPerBlock to device properties
-            auto const& props = alpaka::getAccDevProps<AccGenericSycl<TDim, TIdx>>(dev);
-            kernelFunctionAttributes.maxThreadsPerBlock = static_cast<int>(props.m_blockThreadCountMax);
-            return kernelFunctionAttributes;
-        }
-    };
-
 } // namespace alpaka::trait
 
 #    undef LAUNCH_SYCL_KERNEL_IF_SUBGROUP_SIZE_IS

diff --git a/include/alpaka/kernel/TaskKernelGpuSyclIntel.hpp b/include/alpaka/kernel/TaskKernelGpuSyclIntel.hpp
@@ -16,6 +16,37 @@ namespace alpaka
     template<typename TDim, typename TIdx, typename TKernelFnObj, typename... TArgs>
     using TaskKernelGpuSyclIntel
         = TaskKernelGenericSycl<AccGpuSyclIntel<TDim, TIdx>, TDim, TIdx, TKernelFnObj, TArgs...>;
+
+    namespace trait
+    {
+        //! \brief Specialisation of the class template FunctionAttributes
+        //! \tparam TDev The device type.
+        //! \tparam TDim The dimensionality of the accelerator device properties.
+        //! \tparam TIdx The idx type of the accelerator device properties.
+        //! \tparam TKernelFn Kernel function object type.
+        //! \tparam TArgs Kernel function object argument types as a parameter pack.
+        template<typename TDev, typename TDim, typename TIdx, typename TKernelFn, typename... TArgs>
+        struct FunctionAttributes<AccGpuSyclIntel<TDim, TIdx>, TDev, KernelBundle<TKernelFn, TArgs...>>
+        {
+            //! \param dev The device instance
+            //! \param kernelBundle Kernel bundeled with it's arguments. The function attributes of this kernel will be
+            //! determined. Max threads per block is one of the attributes.
+            //! \return KernelFunctionAttributes instance. The default version always returns an instance with zero
+            //! fields. For CPU, the field of max threads allowed by kernel function for the block is 1.
+            ALPAKA_FN_HOST static auto getFunctionAttributes(
+                TDev const& dev,
+                [[maybe_unused]] KernelBundle<TKernelFn, TArgs...> const& kernelBundle)
+                -> alpaka::KernelFunctionAttributes
+            {
+                alpaka::KernelFunctionAttributes kernelFunctionAttributes;
+
+                // set function properties for maxThreadsPerBlock to device properties
+                auto const& props = alpaka::getAccDevProps<AccGpuSyclIntel<TDim, TIdx>>(dev);
+                kernelFunctionAttributes.maxThreadsPerBlock = static_cast<int>(props.m_blockThreadCountMax);
+                return kernelFunctionAttributes;
+            }
+        };
+    } // namespace trait
 } // namespace alpaka
 
 #endif