[SYCL] do addifitional fixes per reviewer comments

This fix also disables the LIT test reduction_usm.cpp for Linux
until the GPU drivers used by CI is updated to 20.06.15619 or newer.

Signed-off-by: Vyacheslav N Klochkov <[email protected]>

Author: v-klochkov

sycl/include/CL/sycl/handler.hpp

@@ -856,7 +856,7 @@ class __SYCL_EXPORT handler {
   parallel_for(nd_range<Dims> Range, Reduction &Redu, KernelType KernelFunc) {
     if (Reduction::is_usm)
       Redu.associateWithHandler(*this);
-    auto QueueCopy = MQueue;
+    shared_ptr_class<detail::queue_impl> QueueCopy = MQueue;
     auto Acc = Redu.getUserAccessor();
     intel::detail::reduCGFunc<KernelName>(*this, KernelFunc, Range, Redu, Acc);
 
@@ -887,7 +887,7 @@ class __SYCL_EXPORT handler {
   detail::enable_if_t<Reduction::accessor_mode == access::mode::discard_write &&
                       Reduction::has_fast_atomics>
   parallel_for(nd_range<Dims> Range, Reduction &Redu, KernelType KernelFunc) {
-    auto QueueCopy = MQueue;
+    shared_ptr_class<detail::queue_impl> QueueCopy = MQueue;
     auto RWAcc = Redu.getReadWriteScalarAcc(*this);
     intel::detail::reduCGFunc<KernelName>(*this, KernelFunc, Range, Redu,
                                           RWAcc);
@@ -940,7 +940,7 @@ class __SYCL_EXPORT handler {
     if (Reduction::is_usm && NWorkGroups == 1)
       Redu.associateWithHandler(*this);
     intel::detail::reduCGFunc<KernelName>(*this, KernelFunc, Range, Redu);
-    auto QueueCopy = MQueue;
+    shared_ptr_class<detail::queue_impl> QueueCopy = MQueue;
     this->finalize();
 
     // 2. Run the additional aux kernel as many times as needed to reduce

sycl/test/reduction/reduction_placeholder.cpp

@@ -2,11 +2,12 @@
 // Reductions use work-group builtins not yet supported by CUDA.
 
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUNx: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: Enable the test for HOST when it supports intel::reduce() and barrier()
+
 // This test performs basic checks of parallel_for(nd_range, reduction, func)
 // with reductions initialized with a placeholder accessor.
 

sycl/test/reduction/reduction_usm.cpp

@@ -1,12 +1,16 @@
 // UNSUPPORTED: cuda
 // Reductions use work-group builtins not yet supported by CUDA.
 
+// UNSUPPORTED: linux
+// TODO: Enable the test for Linux when CI uses GPU driver 20.06.15619 or newer.
+
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUNx: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: Enable the test for HOST when it supports intel::reduce() and barrier()
+
 // This test performs basic checks of parallel_for(nd_range, reduction, func)
 // with reductions initialized with USM var.