Merge pull request #7913 from hjelmn/btl_base_atomics_are_awesome

btl: change argument type of BTL receive callbacks

Author: hjelmn

ompi/mca/pml/ob1/pml_ob1.h

@@ -311,8 +311,8 @@ void mca_pml_ob1_process_pending_rdma(void);
  * Compute the total number of bytes on supplied descriptor
  */
 static inline size_t
-mca_pml_ob1_compute_segment_length_base(mca_btl_base_segment_t *segments,
-                                        size_t count, size_t hdrlen)
+mca_pml_ob1_compute_segment_length_base (const mca_btl_base_segment_t *segments,
+                                         size_t count, size_t hdrlen)
 {
     size_t i, length = 0;
 
@@ -323,7 +323,7 @@ mca_pml_ob1_compute_segment_length_base(mca_btl_base_segment_t *segments,
 }
 
 static inline size_t
-mca_pml_ob1_compute_segment_length_remote (size_t seg_size, void *segments,
+mca_pml_ob1_compute_segment_length_remote (size_t seg_size, const void *segments,
                                            size_t count, ompi_proc_t *rem_proc)
 {
     mca_btl_base_segment_t *segment = (mca_btl_base_segment_t *) segments;

ompi/mca/pml/ob1/pml_ob1_recvfrag.c

@@ -14,6 +14,7 @@
  * Copyright (c) 2011      Sandia National Laboratories. All rights reserved.
  * Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2020      Google, LLC. All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -113,60 +114,49 @@ do {                                                                    \
  *  Callback from BTL on receipt of a recv_frag (match).
  */
 
-extern void mca_pml_ob1_recv_frag_callback_match( mca_btl_base_module_t *btl,
-                                                  mca_btl_base_tag_t tag,
-                                                  mca_btl_base_descriptor_t* descriptor,
-                                                  void* cbdata );
+extern void mca_pml_ob1_recv_frag_callback_match (mca_btl_base_module_t *btl,
+                                                  const mca_btl_base_receive_descriptor_t *descriptor);
 
 /**
  *  Callback from BTL on receipt of a recv_frag (rndv).
  */
 
-extern void mca_pml_ob1_recv_frag_callback_rndv( mca_btl_base_module_t *btl,
-                                                 mca_btl_base_tag_t tag,
-                                                 mca_btl_base_descriptor_t* descriptor,
-                                                 void* cbdata );
+extern void mca_pml_ob1_recv_frag_callback_rndv (mca_btl_base_module_t *btl,
+                                                 const mca_btl_base_receive_descriptor_t *descriptor);
 /**
  *  Callback from BTL on receipt of a recv_frag (rget).
  */
 
-extern void mca_pml_ob1_recv_frag_callback_rget( mca_btl_base_module_t *btl,
-                                                 mca_btl_base_tag_t tag,
-                                                 mca_btl_base_descriptor_t* descriptor,
-                                                 void* cbdata );
+extern void mca_pml_ob1_recv_frag_callback_rget (mca_btl_base_module_t *btl,
+                                                 const mca_btl_base_receive_descriptor_t *descriptor);
 
 /**
  *  Callback from BTL on receipt of a recv_frag (ack).
  */
 
-extern void mca_pml_ob1_recv_frag_callback_ack( mca_btl_base_module_t *btl,
-                                                mca_btl_base_tag_t tag,
-                                                mca_btl_base_descriptor_t* descriptor,
-                                                void* cbdata );
+extern void mca_pml_ob1_recv_frag_callback_ack (mca_btl_base_module_t *btl,
+                                                const mca_btl_base_receive_descriptor_t *descriptor);
+
 /**
  *  Callback from BTL on receipt of a recv_frag (frag).
  */
 
-extern void mca_pml_ob1_recv_frag_callback_frag( mca_btl_base_module_t *btl,
-                                                 mca_btl_base_tag_t tag,
-                                                 mca_btl_base_descriptor_t* descriptor,
-                                                 void* cbdata );
+extern void mca_pml_ob1_recv_frag_callback_frag (mca_btl_base_module_t *btl,
+                                                 const mca_btl_base_receive_descriptor_t *descriptor);
+
 /**
  *  Callback from BTL on receipt of a recv_frag (put).
  */
 
-extern void mca_pml_ob1_recv_frag_callback_put( mca_btl_base_module_t *btl,
-                                                mca_btl_base_tag_t tag,
-                                                mca_btl_base_descriptor_t* descriptor,
-                                                void* cbdata );
+extern void mca_pml_ob1_recv_frag_callback_put (mca_btl_base_module_t *btl,
+                                                const mca_btl_base_receive_descriptor_t *descriptor);
+
 /**
  *  Callback from BTL on receipt of a recv_frag (fin).
  */
 
-extern void mca_pml_ob1_recv_frag_callback_fin( mca_btl_base_module_t *btl,
-                                                mca_btl_base_tag_t tag,
-                                                mca_btl_base_descriptor_t* descriptor,
-                                                void* cbdata );
+extern void mca_pml_ob1_recv_frag_callback_fin (mca_btl_base_module_t *btl,
+                                                const mca_btl_base_receive_descriptor_t *descriptor);
 
 /**
  * Extract the next fragment from the cant_match ordered list. This fragment

ompi/mca/pml/ob1/pml_ob1_recvfrag.h

@@ -20,6 +20,7 @@
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2018      Sandia National Laboratories
  *                         All rights reserved.
+ * Copyright (c) 2020      Google, LLC. All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -517,7 +518,7 @@ int mca_pml_ob1_recv_request_get_frag (mca_pml_ob1_rdma_frag_t *frag)
 
 void mca_pml_ob1_recv_request_progress_frag( mca_pml_ob1_recv_request_t* recvreq,
                                              mca_btl_base_module_t* btl,
-                                             mca_btl_base_segment_t* segments,
+                                             const mca_btl_base_segment_t* segments,
                                              size_t num_segments )
 {
     size_t bytes_received, data_offset = 0;
@@ -652,7 +653,7 @@ void mca_pml_ob1_recv_request_frag_copy_finished( mca_btl_base_module_t* btl,
 
 void mca_pml_ob1_recv_request_progress_rget( mca_pml_ob1_recv_request_t* recvreq,
                                              mca_btl_base_module_t* btl,
-                                             mca_btl_base_segment_t* segments,
+                                             const mca_btl_base_segment_t* segments,
                                              size_t num_segments )
 {
     mca_pml_ob1_rget_hdr_t* hdr = (mca_pml_ob1_rget_hdr_t*)segments->seg_addr.pval;
@@ -806,7 +807,7 @@ void mca_pml_ob1_recv_request_progress_rget( mca_pml_ob1_recv_request_t* recvreq
 
 void mca_pml_ob1_recv_request_progress_rndv( mca_pml_ob1_recv_request_t* recvreq,
                                              mca_btl_base_module_t* btl,
-                                             mca_btl_base_segment_t* segments,
+                                             const mca_btl_base_segment_t* segments,
                                              size_t num_segments )
 {
     size_t bytes_received = 0;
@@ -876,7 +877,7 @@ void mca_pml_ob1_recv_request_progress_rndv( mca_pml_ob1_recv_request_t* recvreq
  */
 void mca_pml_ob1_recv_request_progress_match( mca_pml_ob1_recv_request_t* recvreq,
                                               mca_btl_base_module_t* btl,
-                                              mca_btl_base_segment_t* segments,
+                                              const mca_btl_base_segment_t* segments,
                                               size_t num_segments )
 {
     size_t bytes_received, data_offset = 0;
@@ -932,7 +933,7 @@ void mca_pml_ob1_recv_request_progress_match( mca_pml_ob1_recv_request_t* recvre
 
 void mca_pml_ob1_recv_request_matched_probe( mca_pml_ob1_recv_request_t* recvreq,
                                              mca_btl_base_module_t* btl,
-                                             mca_btl_base_segment_t* segments,
+                                             const mca_btl_base_segment_t* segments,
                                              size_t num_segments )
 {
     size_t bytes_packed = 0;

ompi/mca/pml/ob1/pml_ob1_recvreq.c

@@ -15,6 +15,7 @@
  *                         reserved.
  * Copyright (c) 2014      Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2020      Google, LLC. All rights reserved.
  *
  * $COPYRIGHT$
  *
@@ -239,7 +240,7 @@ static inline void prepare_recv_req_converter(mca_pml_ob1_recv_request_t *req)
     recv_req_matched(request, hdr)
 
 static inline void recv_req_matched(mca_pml_ob1_recv_request_t *req,
-                                    mca_pml_ob1_match_hdr_t *hdr)
+                                    const mca_pml_ob1_match_hdr_t *hdr)
 {
     req->req_recv.req_base.req_ompi.req_status.MPI_SOURCE = hdr->hdr_src;
     req->req_recv.req_base.req_ompi.req_status.MPI_TAG = hdr->hdr_tag;
@@ -278,7 +279,7 @@ do {
         uint32_t iov_count = 0;                                                   \
         size_t max_data = bytes_received;                                         \
         size_t n, offset = seg_offset;                                            \
-        mca_btl_base_segment_t* segment = segments;                               \
+        const mca_btl_base_segment_t *segment = segments;                         \
                                                                                   \
         for( n = 0; n < num_segments; n++, segment++ ) {                          \
             if(offset >= segment->seg_len) {                                      \
@@ -314,7 +315,7 @@ do {
 void mca_pml_ob1_recv_request_progress_match(
     mca_pml_ob1_recv_request_t* req,
     struct mca_btl_base_module_t* btl,
-    mca_btl_base_segment_t* segments,
+    const mca_btl_base_segment_t* segments,
     size_t num_segments);
 
 /**
@@ -324,7 +325,7 @@ void mca_pml_ob1_recv_request_progress_match(
 void mca_pml_ob1_recv_request_progress_frag(
     mca_pml_ob1_recv_request_t* req,
     struct mca_btl_base_module_t* btl,
-    mca_btl_base_segment_t* segments,
+    const mca_btl_base_segment_t* segments,
     size_t num_segments);
 
 #if OPAL_CUDA_SUPPORT
@@ -347,7 +348,7 @@ void mca_pml_ob1_recv_request_frag_copy_finished(struct mca_btl_base_module_t* b
 void mca_pml_ob1_recv_request_progress_rndv(
     mca_pml_ob1_recv_request_t* req,
     struct mca_btl_base_module_t* btl,
-    mca_btl_base_segment_t* segments,
+    const mca_btl_base_segment_t* segments,
     size_t num_segments);
 
 /**
@@ -357,7 +358,7 @@ void mca_pml_ob1_recv_request_progress_rndv(
 void mca_pml_ob1_recv_request_progress_rget(
     mca_pml_ob1_recv_request_t* req,
     struct mca_btl_base_module_t* btl,
-    mca_btl_base_segment_t* segments,
+    const mca_btl_base_segment_t* segments,
     size_t num_segments);
 
 /**
@@ -367,7 +368,7 @@ void mca_pml_ob1_recv_request_progress_rget(
 void mca_pml_ob1_recv_request_matched_probe(
     mca_pml_ob1_recv_request_t* req,
     struct mca_btl_base_module_t* btl,
-    mca_btl_base_segment_t* segments,
+    const mca_btl_base_segment_t* segments,
     size_t num_segments);
 
 /**

ompi/mca/pml/ob1/pml_ob1_recvreq.h

@@ -1216,9 +1216,9 @@ int mca_pml_ob1_send_request_put_frag( mca_pml_ob1_rdma_frag_t *frag )
  *  (3) Queue the RDMA put
  */
 
-void mca_pml_ob1_send_request_put( mca_pml_ob1_send_request_t* sendreq,
-                                   mca_btl_base_module_t* btl,
-                                   mca_pml_ob1_rdma_hdr_t* hdr )
+void mca_pml_ob1_send_request_put (mca_pml_ob1_send_request_t *sendreq,
+                                   mca_btl_base_module_t *btl,
+                                   const mca_pml_ob1_rdma_hdr_t *hdr)
 {
     mca_bml_base_endpoint_t *bml_endpoint = sendreq->req_endpoint;
     mca_pml_ob1_rdma_frag_t* frag;

ompi/mca/pml/ob1/pml_ob1_sendreq.c

@@ -512,9 +512,9 @@ mca_pml_ob1_send_request_start( mca_pml_ob1_send_request_t* sendreq )
  *  Initiate a put scheduled by the receiver.
  */
 
-void mca_pml_ob1_send_request_put( mca_pml_ob1_send_request_t* sendreq,
-                                   mca_btl_base_module_t* btl,
-                                   mca_pml_ob1_rdma_hdr_t* hdr );
+void mca_pml_ob1_send_request_put (mca_pml_ob1_send_request_t *sendreq,
+                                   mca_btl_base_module_t *btl,
+                                   const mca_pml_ob1_rdma_hdr_t *hdr);
 
 int mca_pml_ob1_send_request_put_frag(mca_pml_ob1_rdma_frag_t* frag);
 

ompi/mca/pml/ob1/pml_ob1_sendreq.h

@@ -450,13 +450,12 @@ typedef struct mca_btl_base_segment_t mca_btl_base_segment_t;
 #define MCA_BTL_BASE_SEGMENT_NTOH(s)
 #endif
 /**
- * A descriptor that holds the parameters to a send/put/get
+ * @brief Descriptor to hold outgoing send messages
+ *
+ * This descriptor holds the parameters to perform a send
  * operation along w/ a callback routine that is called on
  * completion of the request.
- * Note: receive callbacks will store the incomming data segments in
- *       des_segments
  */
-
 struct mca_btl_base_descriptor_t {
     opal_free_list_item_t super;
     mca_btl_base_segment_t *des_segments;     /**< local segments */
@@ -519,6 +518,28 @@ OPAL_DECLSPEC OBJ_CLASS_DECLARATION(mca_btl_base_descriptor_t);
  */
 #define MCA_BTL_REG_HANDLE_MAX_SIZE 256
 
+/**
+ * @brief Descriptor for incoming messages
+ *
+ * This descriptor holds the information on incoming messages. The
+ * data this descriptor describes is only valid during a callback.
+ */
+struct mca_btl_base_receive_descriptor_t {
+    /** Incoming endpoint. May be NULL if the endpoint is not readily
+     * available for this BTL. */
+    struct mca_btl_base_endpoint_t *endpoint;
+    /** Local segment data. Copy this data if it is needed after the
+     * receive callback returns. **/
+    const mca_btl_base_segment_t *des_segments;
+    /** Number of segment is des_segments */
+    size_t des_segment_count;
+    mca_btl_base_tag_t tag;
+    /** Callback data supplied at callback registration time. */
+    void *cbdata;
+};
+typedef struct mca_btl_base_receive_descriptor_t mca_btl_base_receive_descriptor_t;
+
+
 /*
  *  BTL base header, stores the tag at a minimum
  */
@@ -581,25 +602,20 @@ typedef int (*mca_btl_base_component_progress_fn_t)(void);
 /**
  * Callback function that is called asynchronously on receipt
  * of data by the transport layer.
- * Note that the the mca_btl_base_descriptor_t is only valid within the
+ * Note that the the mca_btl_base_receive_descriptor_t is only valid within the
  * completion function, this implies that all data payload in the
  * mca_btl_base_descriptor_t must be copied out within this callback or
  * forfeited back to the BTL.
  * Note also that descriptor segments (des_segments) must be base
  * segments for all callbacks.
  *
  * @param[IN] btl        BTL module
- * @param[IN] tag        The active message receive callback tag value
  * @param[IN] descriptor The BTL descriptor (contains the receive payload)
- * @param[IN] cbdata     Opaque callback data
  */
 
 typedef void (*mca_btl_base_module_recv_cb_fn_t)(
-    struct mca_btl_base_module_t* btl,
-    mca_btl_base_tag_t tag,
-    mca_btl_base_descriptor_t* descriptor,
-    void* cbdata
-);
+    struct mca_btl_base_module_t *btl,
+    const mca_btl_base_receive_descriptor_t *descriptor);
 
 typedef struct mca_btl_active_message_callback_t {
     mca_btl_base_module_recv_cb_fn_t cbfunc;
@@ -1262,22 +1278,29 @@ struct mca_btl_base_module_t {
 };
 typedef struct mca_btl_base_module_t mca_btl_base_module_t;
 
+#define MCA_BTL_BASE_MAJOR_VERSION 3
+#define MCA_BTL_BASE_MINOR_VERSION 2
+#define MCA_BTL_BASE_PATCH_VERSION 0
+
 /*
- * Macro for use in modules that are of type btl v3.1.0
+ * Macro for use in modules that are of type btl v3.2.0
  */
-#define MCA_BTL_BASE_VERSION_3_1_0              \
-    OPAL_MCA_BASE_VERSION_2_1_0("btl", 3, 1, 0)
+#define MCA_BTL_BASE_VERSION_3_2_0                          \
+    OPAL_MCA_BASE_VERSION_2_1_0("btl",                      \
+                                MCA_BTL_BASE_MAJOR_VERSION, \
+                                MCA_BTL_BASE_MINOR_VERSION, \
+                                MCA_BTL_BASE_PATCH_VERSION)
 
 #define MCA_BTL_DEFAULT_VERSION(name)                       \
-    MCA_BTL_BASE_VERSION_3_1_0,                             \
+    MCA_BTL_BASE_VERSION_3_2_0,                             \
     .mca_component_name = name,                             \
     MCA_BASE_MAKE_VERSION(component, OPAL_MAJOR_VERSION, OPAL_MINOR_VERSION, \
                           OPAL_RELEASE_VERSION)
 
 /**
  * Convinience macro for detecting the BTL interface version.
  */
-#define BTL_VERSION 310
+#define BTL_VERSION 320
 
 END_C_DECLS