#510 allow internal VT calls to access MPI

- Scoped grants have to be enabled at all usage sites
  and cannot extend into the scheduler itself.

examples/collection/transpose.cc

@@ -43,6 +43,7 @@
 */
 
 #include <vt/transport.h>
+#include "vt/runtime/mpi_access.h"
 
 #include <vector>
 
@@ -206,9 +207,13 @@ vt::NodeType my_map(IndexT* idx, IndexT* max_idx, vt::NodeType num_nodes) {
 
 // group-targeted handler for the sub-solve
 /*static*/ void SubSolveInfo::subSolveHandler(SubSolveMsg* msg) {
+  // Normally MPI calls are NOT ALLOWED inside user code.
+  VT_ALLOW_MPI_CALLS;
+
   vt::NodeType this_node = vt::theContext()->getNode();
   vt::NodeType num_nodes = vt::theContext()->getNumNodes();
 
+
   auto const group_id = vt::envelopeGetGroup(msg->env);
   MPI_Comm sub_comm = vt::theGroup()->getGroupComm(group_id);
   int sub_size = 0, sub_rank = 0;

src/vt/event/event_record.cc

@@ -44,6 +44,7 @@
 
 #include "vt/event/event.h"
 #include "vt/event/event_record.h"
+#include "vt/runtime/mpi_access.h"
 
 #include <memory>
 
@@ -53,6 +54,7 @@ namespace vt { namespace event {
 
 EventRecord::EventRecord(EventRecordType const& type, EventType const& id)
   : event_id_(id), type_(type) {
+  VT_ALLOW_MPI_CALLS;
 
   switch (type) {
   case EventRecordType::MPI_EventRecord:
@@ -79,6 +81,8 @@ EventRecord::EventRecord(EventRecordType const& type, EventType const& id)
 }
 
 bool EventRecord::testMPIEventReady() {
+  VT_ALLOW_MPI_CALLS;
+
   int flag = 0;
   MPI_Request* req = getRequest();
   MPI_Status stat;

src/vt/group/collective/group_info_collective.cc

@@ -56,6 +56,7 @@
 #include "vt/collective/tree/tree.h"
 #include "vt/collective/collective_alg.h"
 #include "vt/collective/collective_ops.h"
+#include "vt/runtime/mpi_access.h"
 
 #include <memory>
 #include <set>
@@ -101,13 +102,17 @@ MPI_Comm InfoColl::getComm() const {
 }
 
 void InfoColl::freeComm() {
+  VT_ALLOW_MPI_CALLS;
+
   if (mpi_group_comm != MPI_COMM_WORLD) {
     MPI_Comm_free(&mpi_group_comm);
     mpi_group_comm = MPI_COMM_WORLD;
   }
 }
 
 void InfoColl::setupCollective() {
+  VT_ALLOW_MPI_CALLS;
+
   auto const& this_node = theContext()->getNode();
   auto const& num_nodes = theContext()->getNumNodes();
   auto const& group_ = getGroupID();

src/vt/messaging/active.cc

@@ -52,6 +52,7 @@
 #include "vt/runnable/general.h"
 #include "vt/timing/timing.h"
 #include "vt/scheduler/priority.h"
+#include "vt/runtime/mpi_access.h"
 
 namespace vt { namespace messaging {
 
@@ -106,6 +107,8 @@ EventType ActiveMessenger::sendMsgBytesWithPut(
   NodeType const& dest, MsgSharedPtr<BaseMsgType> const& base,
   MsgSizeType const& msg_size, TagType const& send_tag
 ) {
+  VT_ALLOW_MPI_CALLS;
+
   auto msg = base.get();
   auto const& is_term = envelopeIsTerm(msg->env);
   auto const& is_put = envelopeIsPut(msg->env);
@@ -179,6 +182,8 @@ EventType ActiveMessenger::sendMsgBytes(
   NodeType const& dest, MsgSharedPtr<BaseMsgType> const& base,
   MsgSizeType const& msg_size, TagType const& send_tag
 ) {
+  VT_ALLOW_MPI_CALLS;
+
   auto const& msg = base.get();
 
   auto const epoch = envelopeIsEpochType(msg->env) ?
@@ -320,6 +325,8 @@ EventType ActiveMessenger::doMessageSend(
 ActiveMessenger::SendDataRetType ActiveMessenger::sendData(
   RDMA_GetType const& ptr, NodeType const& dest, TagType const& tag
 ) {
+  VT_ALLOW_MPI_CALLS;
+
   auto const& data_ptr = std::get<0>(ptr);
   auto const& num_bytes = std::get<1>(ptr);
   auto const send_tag = tag == no_tag ? cur_direct_buffer_tag_++ : tag;
@@ -429,6 +436,8 @@ bool ActiveMessenger::recvDataMsgBuffer(
   RDMA_ContinuationDeleteType next
 ) {
   if (not enqueue) {
+    VT_ALLOW_MPI_CALLS;
+
     CountType num_probe_bytes;
     MPI_Status stat;
     int flag;

src/vt/messaging/irecv_holder.h

@@ -48,6 +48,9 @@
 #include "vt/config.h"
 #include "vt/configs/arguments/args.h"
 
+// Unfortunate header leak for VT_ALLOW_MPI_CALLS
+#include "vt/runtime/mpi_access.h"
+
 #if backend_check_enabled(trace_enabled)
   #include "vt/trace/trace_headers.h"
 #endif
@@ -94,6 +97,7 @@ struct IRecvHolder {
    */
   template <typename Callable>
   bool testAll(Callable c) {
+    VT_ALLOW_MPI_CALLS;
 
 #   if backend_check_enabled(trace_enabled)
     std::size_t const holder_size_start = holder_.size();

src/vt/rdma/channel/rdma_channel.cc

@@ -43,6 +43,7 @@
 */
 
 #include "vt/rdma/channel/rdma_channel.h"
+#include "vt/runtime/mpi_access.h"
 
 #define PRINT_RDMA_OP_TYPE(OP) ((OP) == RDMA_TypeType::Get ? "GET" : "PUT")
 
@@ -87,6 +88,8 @@ Channel::Channel(
 
 void
 Channel::initChannelGroup() {
+  VT_ALLOW_MPI_CALLS;
+
   debug_print(
     rdma_channel, node,
     "channel: initChannelGroup: target={}, non_target={}, han={}\n",
@@ -236,6 +239,8 @@ void
 Channel::writeDataToChannel(
   RDMA_PtrType const& ptr, ByteType const& ptr_num_bytes, ByteType const& offset
 ) {
+  VT_ALLOW_MPI_CALLS;
+
   vtAssert(initialized_, "Channel must be initialized");
   vtAssert(not is_target_, "The target can not write to this channel");
 
@@ -278,6 +283,8 @@ Channel::writeDataToChannel(
 
 void
 Channel::freeChannel() {
+  VT_ALLOW_MPI_CALLS;
+
   if (locked_) {
     unlockChannelForOp();
   }
@@ -294,6 +301,8 @@ Channel::freeChannel() {
 
 void
 Channel::initChannelWindow() {
+  VT_ALLOW_MPI_CALLS;
+
   debug_print(
     rdma_channel, node,
     "channel: create window: num_bytes={}\n", num_bytes_

src/vt/runtime/runtime.cc

@@ -69,6 +69,7 @@
 #include "vt/configs/error/stack_out.h"
 #include "vt/configs/error/pretty_print_stack.h"
 #include "vt/utils/memory/memory_usage.h"
+#include "vt/runtime/mpi_access.h"
 
 #include <memory>
 #include <iostream>
@@ -1162,6 +1163,7 @@ void Runtime::setup() {
 
   // wait for all nodes to start up to initialize the runtime
   theCollective->barrierThen([this]{
+    VT_ALLOW_MPI_CALLS;
     MPI_Barrier(theContext->getComm());
   });