ADI: add MPID_Allocate_vci and MPID_Deallocate_vci

This ADI allows MPIR layer to request for explicit vcis.

src/mpi/errhan/errnames.txt

@@ -950,6 +950,11 @@ is too big (> MPIU_SHMW_GHND_SZ)
 **set_thread_affinity:Failed to set the async thread affinity
 **set_thread_affinity %d:Failed to set the async thread affinity to the logical processor [%d]
 
+## MPIX_Stream
+**ch3nostream:Stream is not supported in ch3.
+**ch4nostream:No streams available. Configure --enable-thread-cs=per-vci and --with-ch4-max-vcis=# to enable streams.
+**outofstream:No streams available. Use MPIR_CVAR_CH4_RESERVE_VCIS to reserve the number of streams can be allocated.
+
 # -----------------------------------------------------------------------------
 # The following names are defined but not used (see the -careful option 
 # for extracterrmsgs) (still to do: move the undefined names here)

src/mpid/ch3/include/mpidpre.h

@@ -532,6 +532,9 @@ int MPID_Init(int required, int *provided);
 
 int MPID_InitCompleted( void );
 
+int MPID_Allocate_vci(int *vci);
+int MPID_Deallocate_vci(int vci);
+
 int MPID_Finalize(void);
 
 int MPID_Abort( MPIR_Comm *comm, int mpi_errno, int exit_code, const char *error_msg );

src/mpid/ch3/src/mpid_init.c

@@ -265,6 +265,19 @@ int MPID_InitCompleted( void )
     /* --END ERROR HANDLING-- */
 }
 
+int MPID_Allocate_vci(int *vci)
+{
+    int mpi_errno = MPI_SUCCESS;
+    *vci = 0;
+    MPIR_ERR_SET(mpi_errno, MPI_ERR_OTHER, "**ch3nostream");
+    return mpi_errno;
+}
+
+int MPID_Deallocate_vci(int vci)
+{
+    MPIR_Assert(0);
+    return MPI_SUCCESS;
+}
 /*
  * Initialize the process group structure by using PMI calls.
  * This routine initializes PMI and uses PMI calls to setup the 

src/mpid/ch4/include/mpidch4.h

@@ -14,6 +14,8 @@
 
 int MPID_Init(int, int *);
 int MPID_InitCompleted(void);
+int MPID_Allocate_vci(int *vci);
+int MPID_Deallocate_vci(int vci);
 MPL_STATIC_INLINE_PREFIX int MPID_Cancel_recv(MPIR_Request *) MPL_STATIC_INLINE_SUFFIX;
 MPL_STATIC_INLINE_PREFIX int MPID_Cancel_send(MPIR_Request *) MPL_STATIC_INLINE_SUFFIX;
 int MPID_Comm_disconnect(MPIR_Comm *);

src/mpid/ch4/src/ch4_init.c

@@ -518,6 +518,40 @@ int MPID_InitCompleted(void)
     goto fn_exit;
 }
 
+int MPID_Allocate_vci(int *vci)
+{
+    int mpi_errno = MPI_SUCCESS;
+
+    *vci = 0;
+#if MPIDI_CH4_MAX_VCIS == 1
+    MPIR_ERR_SET(mpi_errno, MPI_ERR_OTHER, "**ch4nostream");
+#else
+
+    if (MPIDI_global.n_vcis + MPIDI_global.n_reserved_vcis >= MPIDI_global.n_total_vcis) {
+        MPIR_ERR_SET(mpi_errno, MPI_ERR_OTHER, "**outofstream");
+    } else {
+        MPIDI_global.n_reserved_vcis++;
+        for (int i = MPIDI_global.n_vcis; i < MPIDI_global.n_total_vcis; i++) {
+            if (!MPIDI_VCI(i).allocated) {
+                MPIDI_VCI(i).allocated = true;
+                *vci = i;
+                break;
+            }
+        }
+    }
+#endif
+    return mpi_errno;
+}
+
+int MPID_Deallocate_vci(int vci)
+{
+    MPIR_Assert(vci < MPIDI_global.n_total_vcis && vci >= MPIDI_global.n_vcis);
+    MPIR_Assert(MPIDI_VCI(vci).allocated);
+    MPIDI_VCI(vci).allocated = false;
+    MPIDI_global.n_reserved_vcis--;
+    return MPI_SUCCESS;
+}
+
 int MPID_Finalize(void)
 {
     int mpi_errno;

src/mpid/ch4/src/ch4_types.h

@@ -254,6 +254,7 @@ typedef struct MPIDI_per_vci {
     MPL_atomic_uint64_t exp_seq_no;
     MPL_atomic_uint64_t nxt_seq_no;
 
+    bool allocated;
     char pad MPL_ATTR_ALIGNED(MPL_CACHELINE_SIZE);
 } MPIDI_per_vci_t;