Merge pull request #2253 from DARMA-tasking/2063-convert-buffer-point…

…ers-to-std-byte

#2063: Convert buffer pointers to `std::byte*`

examples/rdma/rdma_simple_get.cc

@@ -86,7 +86,7 @@ static vt::RDMA_GetType test_get_fn(
     this_node, num_bytes, tag
   );
   return vt::RDMA_GetType{
-    my_data.get() + tag, num_bytes == vt::no_byte ? sizeof(double)*10 : num_bytes
+    reinterpret_cast<std::byte*>(my_data.get() + tag), num_bytes == vt::no_byte ? sizeof(double)*10 : num_bytes
   };
 }
 

examples/rdma/rdma_simple_put.cc

@@ -81,7 +81,7 @@ static void put_data_fn(HandleMsg* msg) {
     }
 
     vt::theRDMA()->putData(
-      handle, local_data, sizeof(double)*local_data_len,
+      handle, reinterpret_cast<std::byte*>(local_data), sizeof(double)*local_data_len,
       (this_node-1)*local_data_len, vt::no_tag, vt::rdma::rdma_default_byte_size,
       [=]{
         delete [] local_data;
@@ -111,7 +111,7 @@ static void put_handler_fn(
   for (decltype(count) i = 0; i < count; i++) {
     ::fmt::print(
       "{}: put_handler_fn: data[{}] = {}\n",
-      this_node, i, static_cast<double*>(in_ptr)[i]
+      this_node, i, reinterpret_cast<double*>(in_ptr)[i]
     );
   }
 

src/vt/collective/scatter/scatter.cc

@@ -52,19 +52,19 @@ Scatter::Scatter()
   : tree::Tree(tree::tree_cons_tag_t)
 { }
 
-char* Scatter::applyScatterRecur(
-  NodeType node, char* ptr, std::size_t elm_size, FuncSizeType size_fn,
+std::byte* Scatter::applyScatterRecur(
+  NodeType node, std::byte* ptr, std::size_t elm_size, FuncSizeType size_fn,
   FuncDataType data_fn
 ) {
   // pre-order k-ary tree traversal for data layout
   auto children = Tree::getChildren(node);
-  char* cur_ptr = ptr;
+  auto cur_ptr = ptr;
   vt_debug_print(
     normal, scatter,
     "Scatter::applyScatterRecur: elm_size={}, ptr={}, node={}\n",
     elm_size, print_ptr(ptr), node
   );
-  data_fn(node, reinterpret_cast<void*>(cur_ptr));
+  data_fn(node, cur_ptr);
   cur_ptr += elm_size;
   for (auto&& child : children) {
     vt_debug_print(
@@ -100,7 +100,7 @@ void Scatter::scatterIn(ScatterMsg* msg) {
       child, num_children, child_bytes_size
     );
     auto const child_remaining_size =
-      thePool()->remainingSize(reinterpret_cast<void*>(child_msg.get()));
+      thePool()->remainingSize(reinterpret_cast<std::byte*>(child_msg.get()));
     child_msg->user_han = user_handler;
     auto ptr = reinterpret_cast<char*>(child_msg.get()) + sizeof(ScatterMsg);
     vt_debug_print(
@@ -118,7 +118,7 @@ void Scatter::scatterIn(ScatterMsg* msg) {
   });
 
   auto const& active_fn = auto_registry::getScatterAutoHandler(user_handler);
-  active_fn->dispatch(in_base_ptr, nullptr);
+  active_fn->dispatch(reinterpret_cast<std::byte*>(in_base_ptr), nullptr);
 }
 
 /*static*/ void Scatter::scatterHandler(ScatterMsg* msg) {

src/vt/collective/scatter/scatter.h

@@ -66,7 +66,7 @@ namespace vt { namespace collective { namespace scatter {
  */
 struct Scatter : virtual collective::tree::Tree {
   using FuncSizeType = std::function<std::size_t(NodeType)>;
-  using FuncDataType = std::function<void(NodeType, void*)>;
+  using FuncDataType = std::function<void(NodeType, std::byte*)>;
 
   /**
    * \internal \brief Construct a scatter manager
@@ -130,8 +130,8 @@ struct Scatter : virtual collective::tree::Tree {
    *
    * \return incremented point after scatter is complete
    */
-  char* applyScatterRecur(
-    NodeType node, char* ptr, std::size_t elm_size, FuncSizeType size_fn,
+  std::byte* applyScatterRecur(
+    NodeType node, std::byte* ptr, std::size_t elm_size, FuncSizeType size_fn,
     FuncDataType data_fn
   );
 

src/vt/collective/scatter/scatter.impl.h

@@ -65,10 +65,10 @@ void Scatter::scatter(
   auto scatter_msg =
     makeMessageSz<ScatterMsg>(combined_size, combined_size, elm_size);
   vtAssert(total_size == combined_size, "Sizes must be consistent");
-  auto ptr = reinterpret_cast<char*>(scatter_msg.get()) + sizeof(ScatterMsg);
+  auto ptr = reinterpret_cast<std::byte*>(scatter_msg.get()) + sizeof(ScatterMsg);
 #if vt_check_enabled(memory_pool)
   auto remaining_size =
-    thePool()->remainingSize(reinterpret_cast<void*>(scatter_msg.get()));
+    thePool()->remainingSize(reinterpret_cast<std::byte*>(scatter_msg.get()));
   vtAssertInfo(
     remaining_size >= combined_size, "Remaining size must be sufficient",
     total_size, combined_size, remaining_size, elm_size

src/vt/configs/types/types_rdma.h

@@ -52,7 +52,7 @@
 
 namespace vt {
 
-using RDMA_PtrType                = void *;
+using RDMA_PtrType                = std::byte *;
 using RDMA_ElmType                = uint64_t;
 using RDMA_BlockType              = int64_t;
 using RDMA_HandleType             = int64_t;

src/vt/context/runnable_context/collection.h

@@ -87,9 +87,9 @@ struct Collection {
   void resume();
 
 private:
-  std::function<void(void* in_elm)> set_;   /**< Set context function */
+  std::function<void(std::byte* in_elm)> set_;   /**< Set context function */
   std::function<void()> clear_;             /**< Clear context function */
-  void* elm_ = nullptr;                     /**< The element (untyped) */
+  std::byte* elm_ = nullptr;                     /**< The element (untyped) */
 };
 
 }} /* end namespace vt::ctx */

src/vt/context/runnable_context/collection.impl.h

@@ -54,9 +54,9 @@ template <typename IndexT>
 /*explicit*/ Collection::Collection(
   vrt::collection::Indexable<IndexT>* elm
 ) {
-  elm_ = elm;
-  set_ = [](void* in_elm){
-    auto e = static_cast<vrt::collection::Indexable<IndexT>*>(in_elm);
+  elm_ = reinterpret_cast<std::byte*>(elm);
+  set_ = [](std::byte* in_elm){
+    auto e = reinterpret_cast<vrt::collection::Indexable<IndexT>*>(in_elm);
     auto& idx_ = e->getIndex();
     auto proxy_ = e->getProxy();
     vrt::collection::CollectionContextHolder<IndexT>::set(&idx_, proxy_);

src/vt/group/msg/group_msg.h

@@ -129,13 +129,13 @@ struct GroupListMsg : GroupInfoMsg<GroupMsg<::vt::PayloadMessage>> {
       )
   {
     setPut(
-      in_range->getBound(),
+      reinterpret_cast<const std::byte*>(in_range->getBound()),
       in_range->getSize() * sizeof(RangeType::BoundType)
     );
   }
 
   RangeType getRange() {
-    auto const& ptr = static_cast<RangeType::BoundType*>(getPut());
+    auto const& ptr = reinterpret_cast<RangeType::BoundType*>(getPut());
     return region::ShallowList(ptr, getCount());
   }
 };

src/vt/messaging/active.cc

@@ -184,15 +184,15 @@ trace::TraceEventIDType ActiveMessenger::makeTraceCreationSend(
 }
 
 MsgSizeType ActiveMessenger::packMsg(
-  MessageType* msg, MsgSizeType size, void* ptr, MsgSizeType ptr_bytes
+  MessageType* msg, MsgSizeType size, std::byte* ptr, MsgSizeType ptr_bytes
 ) {
   vt_debug_print(
     verbose, active,
     "packMsg: msg_size={}, put_size={}, ptr={}\n",
     size, ptr_bytes, print_ptr(ptr)
   );
 
-  auto const can_grow = thePool()->tryGrowAllocation(msg, ptr_bytes);
+  auto const can_grow = thePool()->tryGrowAllocation(reinterpret_cast<std::byte*>(msg), ptr_bytes);
   // Typically this should be checked by the caller in advance
   vtAssert(can_grow, "not enough space to pack message" );
 
@@ -236,7 +236,7 @@ EventType ActiveMessenger::sendMsgBytesWithPut(
     auto const& put_ptr = envelopeGetPutPtr(msg->env);
     auto const& put_size = envelopeGetPutSize(msg->env);
     bool const& memory_pool_active = thePool()->active_env();
-    auto const& rem_size = thePool()->remainingSize(msg);
+    auto const& rem_size = thePool()->remainingSize(reinterpret_cast<std::byte*>(msg));
     /*
      * Directly pack if the pool is active (which means it may have
      * overallocated and the remaining size of the (envelope) buffer is
@@ -255,7 +255,7 @@ EventType ActiveMessenger::sendMsgBytesWithPut(
         verbose, active,
         "sendMsgBytesWithPut: (put) put_ptr={}, size:[msg={},put={},rem={}],"
         "dest={}, max_pack_size={}, direct_buf_pack={}\n",
-        put_ptr, base.size(), put_size, rem_size, dest, max_pack_direct_size,
+        print_ptr(put_ptr), base.size(), put_size, rem_size, dest, max_pack_direct_size,
         print_bool(direct_buf_pack)
       );
     }
@@ -300,7 +300,7 @@ struct MultiMsg : vt::Message {
 }
 
 void ActiveMessenger::handleChunkedMultiMsg(MultiMsg* msg) {
-  auto buf = static_cast<char*>(thePool()->alloc(msg->getSize()));
+  auto buf = thePool()->alloc(msg->getSize());
 
   auto const size = msg->getSize();
   auto const info = msg->getInfo();
@@ -376,7 +376,7 @@ EventType ActiveMessenger::sendMsgMPI(
     auto tag = allocateNewTag();
 
     // Send the actual data in multiple chunks
-    PtrLenPairType tup = std::make_tuple(untyped_msg, msg_size);
+    PtrLenPairType tup = std::make_tuple(reinterpret_cast<std::byte*>(untyped_msg), msg_size);
     SendInfo info = sendData(tup, dest, tag);
 
     auto event_id = info.getEvent();
@@ -539,7 +539,7 @@ SendInfo ActiveMessenger::sendData(
   vt_debug_print(
     terse, active,
     "sendData: ptr={}, num_bytes={} dest={}, tag={}, send_tag={}\n",
-    data_ptr, num_bytes, dest, tag, send_tag
+    print_ptr(data_ptr), num_bytes, dest, tag, send_tag
   );
 
   vtAbortIf(
@@ -564,7 +564,7 @@ SendInfo ActiveMessenger::sendData(
 std::tuple<EventType, int> ActiveMessenger::sendDataMPI(
   PtrLenPairType const& payload, NodeType const& dest, TagType const& tag
 ) {
-  auto ptr = static_cast<char*>(std::get<0>(payload));
+  auto ptr = reinterpret_cast<char*>(std::get<0>(payload));
   auto remainder = std::get<1>(payload);
   int num_sends = 0;
   std::vector<EventType> events;
@@ -670,7 +670,7 @@ bool ActiveMessenger::tryProcessDataMsgRecv() {
 }
 
 bool ActiveMessenger::recvDataMsgBuffer(
-  int nchunks, void* const user_buf, TagType const& tag,
+  int nchunks, std::byte* const user_buf, TagType const& tag,
   NodeType const& node, bool const& enqueue, ActionType dealloc,
   ContinuationDeleterType next, bool is_user_buf
 ) {
@@ -681,7 +681,7 @@ bool ActiveMessenger::recvDataMsgBuffer(
 }
 
 bool ActiveMessenger::recvDataMsgBuffer(
-  int nchunks, void* const user_buf, PriorityType priority, TagType const& tag,
+  int nchunks, std::byte* const user_buf, PriorityType priority, TagType const& tag,
   NodeType const& node, bool const& enqueue, ActionType dealloc_user_buf,
   ContinuationDeleterType next, bool is_user_buf
 ) {
@@ -702,9 +702,9 @@ bool ActiveMessenger::recvDataMsgBuffer(
     if (flag == 1) {
       MPI_Get_count(&stat, MPI_BYTE, &num_probe_bytes);
 
-      char* buf = user_buf == nullptr ?
-        static_cast<char*>(thePool()->alloc(num_probe_bytes)) :
-        static_cast<char*>(user_buf);
+      std::byte* buf = user_buf == nullptr ?
+        thePool()->alloc(num_probe_bytes) :
+        user_buf;
 
       NodeType const sender = stat.MPI_SOURCE;
 
@@ -743,15 +743,15 @@ void ActiveMessenger::recvDataDirect(
   int nchunks, TagType const tag, NodeType const from, MsgSizeType len,
   ContinuationDeleterType next
 ) {
-  char* buf = static_cast<char*>(thePool()->alloc(len));
+  std::byte* buf = thePool()->alloc(len);
 
   recvDataDirect(
     nchunks, buf, tag, from, len, default_priority, nullptr, next, false
   );
 }
 
 void ActiveMessenger::recvDataDirect(
-  int nchunks, void* const buf, TagType const tag, NodeType const from,
+  int nchunks, std::byte* const buf, TagType const tag, NodeType const from,
   MsgSizeType len, PriorityType prio, ActionType dealloc,
   ContinuationDeleterType next, bool is_user_buf
 ) {
@@ -760,7 +760,7 @@ void ActiveMessenger::recvDataDirect(
   std::vector<MPI_Request> reqs;
   reqs.resize(nchunks);
 
-  char* cbuf = static_cast<char*>(buf);
+  std::byte* cbuf = buf;
   MsgSizeType remainder = len;
   auto const max_per_send = theConfig()->vt_max_mpi_send_size;
   for (int i = 0; i < nchunks; i++) {
@@ -838,7 +838,7 @@ void ActiveMessenger::finishPendingDataMsgAsyncRecv(InProgressDataIRecv* irecv)
     vt_debug_print(
       normal, active,
       "finishPendingDataMsgAsyncRecv: continuation user_buf={}, buf={}\n",
-      user_buf, buf
+      print_ptr(user_buf), print_ptr(buf)
     );
 
     if (user_buf == nullptr) {
@@ -998,7 +998,7 @@ bool ActiveMessenger::tryProcessIncomingActiveMsg() {
   if (flag == 1) {
     MPI_Get_count(&stat, MPI_BYTE, &num_probe_bytes);
 
-    char* buf = static_cast<char*>(thePool()->alloc(num_probe_bytes));
+    std::byte* buf = thePool()->alloc(num_probe_bytes);
 
     NodeType const sender = stat.MPI_SOURCE;
 
@@ -1051,7 +1051,7 @@ bool ActiveMessenger::tryProcessIncomingActiveMsg() {
 }
 
 void ActiveMessenger::finishPendingActiveMsgAsyncRecv(InProgressIRecv* irecv) {
-  char* buf = irecv->buf;
+  std::byte* buf = irecv->buf;
   auto num_probe_bytes = irecv->probe_bytes;
   auto sender = irecv->sender;
 
@@ -1089,14 +1089,14 @@ void ActiveMessenger::finishPendingActiveMsgAsyncRecv(InProgressIRecv* irecv) {
     if (put_tag == PutPackedTag) {
       auto const put_size = envelopeGetPutSize(msg->env);
       auto const msg_size = num_probe_bytes - put_size;
-      char* put_ptr = buf + msg_size;
+      std::byte* put_ptr = buf + msg_size;
 
       if (!is_term || vt_check_enabled(print_term_msgs)) {
         vt_debug_print(
           verbose, active,
           "finishPendingActiveMsgAsyncRecv: packed put: ptr={}, msg_size={}, "
           "put_size={}\n",
-          put_ptr, msg_size, put_size
+          print_ptr(put_ptr), msg_size, put_size
         );
       }