Rough implementation of scan algorithm.

include/RAJA/policy/sycl.hpp

@@ -29,9 +29,9 @@
 #include "RAJA/policy/sycl/forall.hpp"
 #include "RAJA/policy/sycl/policy.hpp"
 #include "RAJA/policy/sycl/reduce.hpp"
+#include "RAJA/policy/sycl/scan.hpp"
+#include "RAJA/policy/sycl/sort.hpp"
 //#include "RAJA/policy/sycl/multi_reduce.hpp"
-//#include "RAJA/policy/sycl/scan.hpp"
-//#include "RAJA/policy/sycl/sort.hpp"
 #include "RAJA/policy/sycl/kernel.hpp"
 //#include "RAJA/policy/sycl/synchronize.hpp"
 #include "RAJA/policy/sycl/launch.hpp"

include/RAJA/policy/sycl/scan.hpp

@@ -0,0 +1,236 @@
+/*!
+******************************************************************************
+*
+* \file
+*
+* \brief   Header file providing RAJA scan declarations.
+*
+******************************************************************************
+*/
+
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+// Copyright (c) 2016-24, Lawrence Livermore National Security, LLC
+// and RAJA project contributors. See the RAJA/LICENSE file for details.
+//
+// SPDX-License-Identifier: (BSD-3-Clause)
+//~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~//
+
+#ifndef RAJA_scan_sycl_HPP
+#define RAJA_scan_sycl_HPP
+
+#include <cstddef>
+#include "RAJA/config.hpp"
+#include "camp/resource/sycl.hpp"
+
+#if defined(RAJA_ENABLE_SYCL)
+
+#include "RAJA/policy/sycl/policy.hpp"
+
+namespace RAJA
+{
+namespace impl
+{
+namespace scan
+{
+
+template <size_t BLOCK_SIZE, bool Async, typename InputIter, typename Function>
+RAJA_INLINE 
+camp::resources::EventProxy<camp::resources::Sycl> 
+inclusive_inplace(
+    camp::resources::Sycl sycl_res,
+    ::RAJA::policy::sycl::sycl_exec<BLOCK_SIZE, Async>,
+    InputIter begin,
+    InputIter end,
+    Function binary_op)
+{
+    ::sycl::queue* sycl_queue = sycl_res.get_queue();
+
+    using valueT = typename std::remove_reference<decltype(*begin)>::type;
+
+     // Calculate the size of the input range
+    size_t size = std::distance(begin, end);
+
+    ::sycl::buffer<valueT, 1> buffer(begin, end);
+    ::sycl::buffer<valueT, 1> tempAccBuff(begin, ::sycl::range<1>(size));
+
+    int iterations = 0;
+    for (size_t ii = size >> 1; ii > 0; ii >>= 1) {
+        iterations++;
+    }
+    if ((size & (size - 1)) != 0) {
+        iterations++;
+    }
+
+    auto buffPtr = &buffer;
+    auto tempPtr = &tempAccBuff;
+
+    if (iterations % 2 == 0) { 
+        tempPtr = &buffer;
+        buffPtr = &tempAccBuff;
+    } 
+
+    int ii = 1;
+    do {
+        // Submit the kernel to the SYCL queue  
+        sycl_queue->submit([&](::sycl::handler& cgh) {
+            auto buffAccessor = buffPtr->get_access(cgh);
+            auto tempAccessor = tempPtr->get_access(cgh);
+            // outBuffAccessor = outBuff->get_access<::sycl::access::mode::read(cgh);
+
+            cgh.parallel_for(::sycl::range<1>(size), [=](::sycl::item<1> idx) {
+                size_t td = 1 << (ii - 1);
+                size_t thisID = idx.get_id(0);
+                if (thisID < size and thisID >= td) { 
+                    tempAccessor[thisID] = binary_op(buffAccessor[thisID - td], buffAccessor[thisID]);
+                } else {
+                    tempAccessor[thisID] = buffAccessor[thisID];
+                }
+            });
+        });
+
+        std::swap(buffPtr, tempPtr);
+        ii++;
+    } while (  ii <= iterations);
+
+  sycl_res.wait();
+  return camp::resources::EventProxy<camp::resources::Sycl>(sycl_res);
+}
+
+template <size_t BLOCK_SIZE,
+          bool Async,
+          typename InputIter,
+          typename Function,
+          typename TT>
+RAJA_INLINE 
+resources::EventProxy<resources::Sycl> 
+exclusive_inplace(
+    resources::Sycl sycl_res,
+    ::RAJA::policy::sycl::sycl_exec<BLOCK_SIZE, Async> exec,
+    InputIter begin,
+    InputIter end,
+    Function binary_op,
+    TT initVal)
+{
+    ::sycl::queue* sycl_queue = sycl_res.get_queue();
+
+    using valueT = typename std::remove_reference<decltype(*begin)>::type;
+
+     // Calculate the size of the input range
+    size_t size = std::distance(begin, end);
+
+    ::sycl::buffer<valueT, 1> buffer(begin, end);
+    ::sycl::buffer<valueT, 1> tempAccBuff(begin, ::sycl::range<1>(size));
+
+    int iterations = 0;
+    for (size_t ii = size >> 1; ii > 0; ii >>= 1) {
+        iterations++;
+    }
+    if ((size & (size - 1)) != 0) {
+        iterations++;
+    }
+
+    auto buffPtr = &buffer;
+    auto tempPtr = &tempAccBuff;
+
+    if (iterations % 2 != 0) { 
+        tempPtr = &buffer;
+        buffPtr = &tempAccBuff;
+    } 
+
+    // Submit the kernel to the SYCL queue  
+    sycl_queue->submit([&](::sycl::handler& cgh) {
+        auto inAccessor = buffPtr->get_access(cgh);
+        auto outAccessor = tempPtr->get_access(cgh);
+        // outBuffAccessor = outBuff->get_access<::sycl::access::mode::read(cgh);
+
+        cgh.parallel_for(::sycl::range<1>(size), [=](::sycl::item<1> idx) {
+            size_t thisID = idx.get_id(0);
+            // size_t td = 1 << (ii - 1);
+            if (thisID > 0) { 
+                outAccessor[thisID] = inAccessor[thisID - 1];
+            } else {
+                outAccessor[thisID] = initVal;
+            }
+        });
+    });
+
+    std::swap(buffPtr, tempPtr);
+
+    int ii = 1;
+    do {
+        // Submit the kernel to the SYCL queue  
+        sycl_queue->submit([&](::sycl::handler& cgh) {
+            auto buffAccessor = buffPtr->get_access(cgh);
+            auto tempAccessor = tempPtr->get_access(cgh);
+            // outBuffAccessor = outBuff->get_access<::sycl::access::mode::read(cgh);
+
+            cgh.parallel_for(::sycl::range<1>(size), [=](::sycl::item<1> idx) {
+                size_t td = 1 << (ii - 1);
+                size_t thisID = idx.get_id(0);
+                if (thisID < size and thisID >= td) { 
+                    tempAccessor[thisID] = binary_op(buffAccessor[thisID - td], buffAccessor[thisID]);
+                } else {
+                    tempAccessor[thisID] = buffAccessor[thisID];
+                }
+            });
+        });
+
+        std::swap(buffPtr, tempPtr);
+        ii++;
+    } while (  ii <= iterations);
+
+  sycl_res.wait();
+  return camp::resources::EventProxy<camp::resources::Sycl>(sycl_res);
+    return inclusive_inplace(sycl_res, exec, begin, end, binary_op);
+}
+
+template <size_t BLOCK_SIZE,
+          bool Async,
+          typename InputIter,
+          typename OutputIter,
+          typename Function>
+RAJA_INLINE 
+resources::EventProxy<resources::Sycl> 
+inclusive(
+    resources::Sycl sycl_res,
+    ::RAJA::policy::sycl::sycl_exec<BLOCK_SIZE, Async> exec,
+    InputIter begin,
+    InputIter end,
+    OutputIter out,
+    Function binary_op)
+{
+  // ::sycl::joint_inclusive_scan()
+    using std::distance;
+    std::copy(begin, end, out);
+    return inclusive_inplace(sycl_res, exec, out, out + distance(begin, end), binary_op);
+}
+
+template <size_t BLOCK_SIZE,
+          bool Async,
+          typename InputIter,
+          typename OutputIter,
+          typename Function,
+          typename TT>
+RAJA_INLINE 
+resources::EventProxy<resources::Sycl> 
+exclusive(
+    resources::Sycl sycl_res,
+    ::RAJA::policy::sycl::sycl_exec<BLOCK_SIZE, Async> exec,
+    InputIter begin,
+    InputIter end,
+    OutputIter out,
+    Function binary_op,
+    TT initVal)
+{
+    using std::distance;
+    std::copy(begin, end, out);
+    return exclusive_inplace(sycl_res, exec, out, out + distance(begin, end), binary_op, initVal);
+}
+
+}  // namespace scan
+}  // namespace impl
+}  // namespace RAJA
+
+#endif  // closing endif for RAJA enable Sycl guard
+
+#endif  // closing endif for header include guard

test/functional/scan/CMakeLists.txt

@@ -19,6 +19,10 @@ if(RAJA_ENABLE_HIP)
   list(APPEND SCAN_BACKENDS Hip)
 endif()
 
+if(RAJA_ENABLE_SYCL)
+  list(APPEND SCAN_BACKENDS Sycl)
+endif()
+
 
 set(SCAN_TYPES Exclusive ExclusiveInplace Inclusive InclusiveInplace)