Merge pull request #132 from pmodels/issue-122

shmem_calloc needs to allocate count*size not size

src/include/oshmpi_impl.h

@@ -357,6 +357,7 @@ void *OSHMPI_malloc(size_t size);
 void OSHMPI_free(void *ptr);
 void *OSHMPI_realloc(void *ptr, size_t size);
 void *OSHMPI_align(size_t alignment, size_t size);
+void *OSHMPI_calloc(size_t count, size_t size);
 
 void OSHMPI_strided_initialize(void);
 void OSHMPI_strided_finalize(void);

src/internal/mem_impl.c

@@ -70,3 +70,17 @@ void *OSHMPI_align(size_t alignment, size_t size)
     OSHMPI_barrier_all();
     return ptr;
 }
+
+void *OSHMPI_calloc(size_t count, size_t size)
+{
+    void *ptr = NULL;
+
+    OSHMPI_THREAD_ENTER_CS(&OSHMPI_global.symm_heap_mspace_cs);
+    ptr = mspace_calloc(OSHMPI_global.symm_heap_mspace, count, size);
+    OSHMPI_THREAD_EXIT_CS(&OSHMPI_global.symm_heap_mspace_cs);
+
+    OSHMPI_DBGMSG("count %ld, size %ld, ptr %p, disp 0x%lx\n", count, size, ptr,
+                  (MPI_Aint) ptr - (MPI_Aint) OSHMPI_global.symm_heap_attr.base);
+    OSHMPI_barrier_all();
+    return ptr;
+}

src/shmem/mem.c

@@ -47,9 +47,15 @@ void *shmem_calloc(size_t count, size_t size)
 {
     void *ptr = NULL;
 
+    /* OpenSHMEM 1.5 Section 9.3.3:
+     * When count or size is 0, the shmem_calloc routine returns without performing a barrier.
+     * Otherwise, this routine calls a procedure that is semantically equivalent to shmem_barrier_all on exit. */
+    if ((count == 0) || (size == 0)) {
+        return ptr;
+    }
+
     OSHMPI_NOINLINE_RECURSIVE()
-        ptr = OSHMPI_malloc(size);
-    memset(ptr, 0, count * size);
+        ptr = OSHMPI_calloc(count, size);
 
     return ptr;
 }