[NFC][SYCL] Prefer (void) over std::ignore

One of my recent PRs to reduce header dependencies of
`id/item/range/etc.` has caused some compile time errors downstream with
`std::ignore` not being declared because we use different version of STL
than the one in the upstream CI.

We do want to have our device headers as lightweight as
possible (especially in the context of SYCL RTC), so it makes sense to
just use `(void)` uniformly instead of `std::ignore`. UR experience
also found that the compile-time price of `(void)` is smaller as well.

Author: aelovikov-intel

sycl/include/sycl/accessor_image.hpp

@@ -907,7 +907,7 @@ class __SYCL_EBO unsampled_image_accessor :
   DataT read(const CoordT &Coords) const noexcept {
 #ifdef __SYCL_DEVICE_ONLY__
     // Currently not reachable on device.
-    std::ignore = Coords;
+    (void)Coords;
     return {0, 0, 0, 0};
 #else
     return host_base_class::read<DataT>(Coords);
@@ -925,8 +925,8 @@ class __SYCL_EBO unsampled_image_accessor :
   void write(const CoordT &Coords, const DataT &Color) const {
 #ifdef __SYCL_DEVICE_ONLY__
     // Currently not reachable on device.
-    std::ignore = Coords;
-    std::ignore = Color;
+    (void)Coords;
+    (void)Color;
 #else
     host_base_class::write<DataT>(Coords, Color);
 #endif // __SYCL_DEVICE_ONLY__
@@ -938,7 +938,7 @@ class __SYCL_EBO unsampled_image_accessor :
       : host_base_class{Impl}
 #endif // __SYCL_DEVICE_ONLY__
   {
-    std::ignore = Impl;
+    (void)Impl;
   }
 
   template <class Obj>
@@ -1216,7 +1216,7 @@ class __SYCL_EBO sampled_image_accessor :
   DataT read(const CoordT &Coords) const noexcept {
 #ifdef __SYCL_DEVICE_ONLY__
     // Currently not reachable on device.
-    std::ignore = Coords;
+    (void)Coords;
     return {0, 0, 0, 0};
 #else
     return host_base_class::read<DataT>(Coords);
@@ -1229,7 +1229,7 @@ class __SYCL_EBO sampled_image_accessor :
       : host_base_class{Impl}
 #endif // __SYCL_DEVICE_ONLY__
   {
-    std::ignore = Impl;
+    (void)Impl;
   }
 
   template <class Obj>

sycl/include/sycl/detail/spirv.hpp

@@ -924,7 +924,7 @@ template <typename GroupT, typename T>
 EnableIfNativeShuffle<T> Shuffle(GroupT g, T x, id<1> local_id) {
   uint32_t LocalId = MapShuffleID(g, local_id);
 #ifndef __NVPTX__
-  std::ignore = g;
+  (void)g;
   if constexpr (ext::oneapi::experimental::is_user_constructed_group_v<
                     GroupT> &&
                 detail::is_vec<T>::value) {
@@ -957,7 +957,7 @@ EnableIfNativeShuffle<T> Shuffle(GroupT g, T x, id<1> local_id) {
 template <typename GroupT, typename T>
 EnableIfNativeShuffle<T> ShuffleXor(GroupT g, T x, id<1> mask) {
 #ifndef __NVPTX__
-  std::ignore = g;
+  (void)g;
   if constexpr (ext::oneapi::experimental::is_user_constructed_group_v<
                     GroupT> &&
                 detail::is_vec<T>::value) {

sycl/include/sycl/ext/intel/esimd/memory.hpp

@@ -2155,7 +2155,7 @@ __ESIMD_API simd<Tx, N> block_load(AccessorTy acc,
   return block_load<Tx, N>(__ESIMD_DNS::accessorToPointer<Tx>(acc, byte_offset),
                            flags);
 #else
-  std::ignore = flags;
+  (void)flags;
   constexpr unsigned Sz = sizeof(T) * N;
   static_assert(Sz >= detail::OperandSize::OWORD,
                 "block size must be at least 1 oword");

sycl/include/sycl/ext/intel/experimental/pipes.hpp

@@ -110,7 +110,7 @@ class pipe : public pipe_base {
   static _dataT read(queue &Q, bool &Success,
                      memory_order Order = memory_order::seq_cst) {
     // Order is currently unused.
-    std::ignore = Order;
+    (void)Order;
 
     const device Dev = Q.get_device();
     bool IsPipeSupported =
@@ -139,7 +139,7 @@ class pipe : public pipe_base {
   static void write(queue &Q, const _dataT &Data, bool &Success,
                     memory_order Order = memory_order::seq_cst) {
     // Order is currently unused.
-    std::ignore = Order;
+    (void)Order;
 
     const device Dev = Q.get_device();
     bool IsPipeSupported =
@@ -267,7 +267,7 @@ class pipe : public pipe_base {
   // Host API
   static _dataT read(queue &Q, memory_order Order = memory_order::seq_cst) {
     // Order is currently unused.
-    std::ignore = Order;
+    (void)Order;
 
     const device Dev = Q.get_device();
     bool IsPipeSupported =
@@ -290,7 +290,7 @@ class pipe : public pipe_base {
   static void write(queue &Q, const _dataT &Data,
                     memory_order Order = memory_order::seq_cst) {
     // Order is currently unused.
-    std::ignore = Order;
+    (void)Order;
 
     const device Dev = Q.get_device();
     bool IsPipeSupported =

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

@@ -90,9 +90,9 @@ void prefetch_impl(T *ptr, size_t bytes, Properties properties) {
   }
   __spirv_ocl_prefetch(ptrAnnotated, bytes);
 #else
-  std::ignore = ptr;
-  std::ignore = bytes;
-  std::ignore = properties;
+  (void)ptr;
+  (void)bytes;
+  (void)properties;
 #endif
 }
 
@@ -101,7 +101,7 @@ void joint_prefetch_impl(Group g, T *ptr, size_t bytes, Properties properties) {
   // Although calling joint_prefetch is functionally equivalent to calling
   // prefetch from every work-item in a group, native suppurt may be added to to
   // issue cooperative prefetches more efficiently on some hardware.
-  std::ignore = g;
+  (void)g;
   prefetch_impl(ptr, bytes, properties);
 }
 } // namespace detail

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

@@ -33,10 +33,10 @@ T reduce_over_group_impl(GroupHelper group_helper, T x, size_t num_elements,
   group_barrier(g);
   return group_broadcast(g, result);
 #else
-  std::ignore = group_helper;
-  std::ignore = x;
-  std::ignore = num_elements;
-  std::ignore = binary_op;
+  (void)group_helper;
+  (void)x;
+  (void)num_elements;
+  (void)binary_op;
   throw exception(make_error_code(errc::runtime),
                   "Group algorithms are not supported on host.");
 #endif
@@ -73,7 +73,7 @@ reduce_over_group(GroupHelper group_helper, V x, T init,
 #ifdef __SYCL_DEVICE_ONLY__
   return binary_op(init, reduce_over_group(group_helper, x, binary_op));
 #else
-  std::ignore = group_helper;
+  (void)group_helper;
   throw exception(make_error_code(errc::runtime),
                   "Group algorithms are not supported on host.");
 #endif
@@ -107,10 +107,10 @@ joint_reduce(GroupHelper group_helper, Ptr first, Ptr last,
   return detail::reduce_over_group_impl(group_helper, partial, num_elements,
                                         binary_op);
 #else
-  std::ignore = group_helper;
-  std::ignore = first;
-  std::ignore = last;
-  std::ignore = binary_op;
+  (void)group_helper;
+  (void)first;
+  (void)last;
+  (void)binary_op;
   throw exception(make_error_code(errc::runtime),
                   "Group algorithms are not supported on host.");
 #endif
@@ -129,8 +129,8 @@ joint_reduce(GroupHelper group_helper, Ptr first, Ptr last, T init,
 #ifdef __SYCL_DEVICE_ONLY__
   return binary_op(init, joint_reduce(group_helper, first, last, binary_op));
 #else
-  std::ignore = group_helper;
-  std::ignore = last;
+  (void)group_helper;
+  (void)last;
   throw exception(make_error_code(errc::runtime),
                   "Group algorithms are not supported on host.");
 #endif