Merge pull request #1137 from davidozog/pr/cxi_quiet_rate

OFI/CXI: add intermittent quiet for CXI workaround

src/shmem_comm.h

@@ -336,7 +336,14 @@ shmem_internal_atomicv(shmem_ctx_t ctx, void *target, const void *source,
         shmem_internal_fetch_atomic(ctx, ((uint8_t *) target) + (i * type_size),
                                     ((uint8_t *) source) + (i * type_size), &tmp_fetch, type_size,
                                     pe, op, datatype);
+#ifdef SOS_OFI_CXI_QUIET_RATE
+        /* FIXME: This intermittent quiet should not be required: */
+        if (i % SOS_OFI_CXI_QUIET_RATE == 0) {
+            shmem_transport_put_quiet((shmem_transport_ctx_t *)ctx);
+        }
+#endif
     }
+    *completion += 1;
 #else
     if (shmem_shr_transport_use_atomic(ctx, target, len, pe, datatype)) {
         shmem_shr_transport_atomicv(ctx, target, source, len, pe, op, datatype);

src/transport_ofi.h

@@ -40,6 +40,13 @@
 #define ENABLE_TARGET_CNTR 0
 #endif
 
+#if DISABLE_NONFETCH_AMO
+/* FIXME: The progress of back-to-back atomic ops over CXI is improved with an
+ * occasional quiet operation. This constant sets the number of atomic ops that
+ * occur between calls to quiet within shmem_internal_atomicv. */
+#define SOS_OFI_CXI_QUIET_RATE 200
+#endif
+
 #if ENABLE_TARGET_CNTR
 extern struct fid_cntr*                 shmem_transport_ofi_target_cntrfd;
 #endif