Merge pull request #891 from rolfv/pr/minor-cuda-verbosity-fixes

Fix cuda verbosity messages

opal/mca/common/cuda/common_cuda.c

@@ -1103,9 +1103,9 @@ int cuda_getmemhandle(void *base, size_t size, mca_mpool_base_registration_t *ne
  */
 int cuda_ungetmemhandle(void *reg_data, mca_mpool_base_registration_t *reg)
 {
-    CUDA_DUMP_EVTHANDLE((100, ((mca_mpool_common_cuda_reg_t *)reg)->data.evtHandle, "cuda_ungetmemhandle"));
     opal_output_verbose(10, mca_common_cuda_output,
                         "CUDA: cuda_ungetmemhandle (no-op): base=%p", reg->base);
+    CUDA_DUMP_MEMHANDLE((100, ((mca_mpool_common_cuda_reg_t *)reg)->data.memHandle, "cuda_ungetmemhandle"));
 
     return OPAL_SUCCESS;
 }
@@ -1691,9 +1691,9 @@ static void cuda_dump_memhandle(int verbose, void *memHandle, char *str) {
     }
     memcpy(&memH, memHandle, sizeof(memH));
     opal_output_verbose(verbose, mca_common_cuda_output,
-                        "%s:ctxId=%d, pid=%d, size=%d, blocksize=%d, offset=%d, gpuId=%d, "
+                        "%s:ctxId=%p, pid=%d, size=%d, blocksize=%d, offset=%d, gpuId=%d, "
                         "subDeviceIndex=%d, serial=%d",
-                        str, (int)memH.ctxId, memH.pid, (int)memH.size, (int)memH.blocksize, (int)memH.offset,
+                        str, (void *)memH.ctxId, memH.pid, (int)memH.size, (int)memH.blocksize, (int)memH.offset,
                         memH.gpuId, memH.subDeviceIndex, (int)memH.serial);
 }
 
@@ -1705,12 +1705,9 @@ static void cuda_dump_evthandle(int verbose, void *evtHandle, char *str) {
 
     struct InterprocessEventHandleInternal
     {
-        /* The first two entries are the CUinterprocessCtxHandle */
+        unsigned long pid;
-        int64_t ctxId; /* unique (within a process) id of the sharing context */
+        unsigned long serial;
-        int     pid;   /* pid of sharing context */
+        int index;
-
-        int     pad;   /* pad to match the structure */
-        int     index;
     } evtH;
 
     if (NULL == str) {
@@ -1718,8 +1715,8 @@ static void cuda_dump_evthandle(int verbose, void *evtHandle, char *str) {
     }
     memcpy(&evtH, evtHandle, sizeof(evtH));
     opal_output_verbose(verbose, mca_common_cuda_output,
-                        "CUDA: %s:ctxId=%d, pid=%d, index=%d",
+                        "CUDA: %s:pid=%d, serial=%d, index=%d",
-                        str, (int)evtH.ctxId, evtH.pid, (int)evtH.index);
+                        str, (int)evtH.pid, (int)evtH.serial, (int)evtH.index);
 }