Merge pull request #5 from NU-CUCIS/remove_mem_profile

Remove use of H5get_alloc_stats

Author: wkliao

RELEASE_NOTES.md

@@ -51,6 +51,8 @@ Bug fix:
   + Fix compression threshold calculation. See a9ea841
 
 Other changes
+  + Remove the HDF5 metadata memory footprint profiling, as H5get_alloc_stats
+    has been removed from HDF5 since 1.14.0. See PR #5.
   + Remove the requirement of group `/spill`. See d4e0852
   + Change chunk setting for small datasets. Set the chunk size to 256K
     element, no matter if the dataset dim[0] is smaller than 256K element or

configure.ac

@@ -354,9 +354,6 @@ AC_CHECK_DECL(H5Ovisit3,
 AC_CHECK_DECL(H5Treclaim,
               [AC_DEFINE(HAS_H5TRECLAIM, 1, [Define if API H5Treclaim is defined])],
               [], [[#include <hdf5.h>]])
-AC_CHECK_DECL(H5get_alloc_stats,
-              [AC_DEFINE(HAS_H5GET_ALLOC_STATS, 1, [Define if API H5get_alloc_stats is defined])],
-              [], [[#include <hdf5.h>]])
 
 AC_ARG_VAR(TESTMPIRUN, [MPI run command for "make check", @<:@default: mpiexec@:>@])
 if test "x${TESTMPIRUN}" = x ; then

main.cpp

@@ -361,10 +361,6 @@ int main(int argc, char **argv)
 
     GET_MEM(step_vmrss[0], step_vmsize[0])
 
-#if defined HAS_H5GET_ALLOC_STATS && HAS_H5GET_ALLOC_STATS
-    size_t cur_bytes, md_malloc;
-    H5get_alloc_stats(NULL, &cur_bytes, NULL, NULL, NULL, NULL, NULL);
-#endif
     SET_TIMER(ts)
     /* Each process reads assigned input files to collect all group and dataset
      * metadata and stores them in the object concat. The input datasets are
@@ -379,10 +375,6 @@ int main(int argc, char **argv)
     GET_TIMER(ts, step_time[0])
     PRN_TIMER(step_time[0], "Read metadata from input files")
     GET_MEM(step_vmrss[1], step_vmsize[1])
-#if defined HAS_H5GET_ALLOC_STATS && HAS_H5GET_ALLOC_STATS
-    H5get_alloc_stats(NULL, &md_malloc, NULL, NULL, NULL, NULL, NULL);
-    md_malloc -= cur_bytes;
-#endif
 
     SET_TIMER(ts)
     if (opt.append_mode)
@@ -641,60 +633,6 @@ int main(int argc, char **argv)
         printf("                                 VmSize: %9s %9s %9ld\n", "","", total_vmsize[8]);
         printf("\n");
     }
-
-#if defined HAS_H5GET_ALLOC_STATS && HAS_H5GET_ALLOC_STATS
-    unsigned long long total_alloc_bytes, max_ull, min_ull, avg_ull;
-    size_t curr_alloc_bytes, peak_alloc_bytes, max_block_size;
-    size_t total_alloc_blocks_count, curr_alloc_blocks_count;
-    size_t peak_alloc_blocks_count, zd[7], max_zd[7], min_zd[7], avg_zd[7];
-
-    err = H5get_alloc_stats(&total_alloc_bytes, &curr_alloc_bytes,
-                            &peak_alloc_bytes, &max_block_size,
-                            &total_alloc_blocks_count,
-                            &curr_alloc_blocks_count,
-                            &peak_alloc_blocks_count);
-
-    MPI_Reduce(&total_alloc_bytes, &min_ull, 1, MPI_UNSIGNED_LONG_LONG, MPI_MIN, 0, MPI_COMM_WORLD);
-    MPI_Reduce(&total_alloc_bytes, &max_ull, 1, MPI_UNSIGNED_LONG_LONG, MPI_MAX, 0, MPI_COMM_WORLD);
-    MPI_Reduce(&total_alloc_bytes, &avg_ull, 1, MPI_UNSIGNED_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
-    min_ull /= 1048576;
-    max_ull /= 1048576;
-    avg_ull /= nprocs * 1048576;
-    zd[0] = md_malloc;
-    zd[1] = curr_alloc_bytes;
-    zd[2] = peak_alloc_bytes;
-    zd[3] = max_block_size;
-    zd[4] = total_alloc_blocks_count;
-    zd[5] = curr_alloc_blocks_count;
-    zd[6] = peak_alloc_blocks_count;
-    MPI_Reduce(zd, &min_zd, 7, MPI_UNSIGNED_LONG, MPI_MIN, 0, MPI_COMM_WORLD);
-    MPI_Reduce(zd, &max_zd, 7, MPI_UNSIGNED_LONG, MPI_MAX, 0, MPI_COMM_WORLD);
-    MPI_Reduce(zd, &avg_zd, 7, MPI_UNSIGNED_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
-    for (int i=0; i<4; i++) {
-        min_zd[i] /= 1048576;
-        max_zd[i] /= 1048576;
-        avg_zd[i] /= nprocs * 1048576;
-    }
-    for (int i=4; i<7; i++) {
-        min_zd[i] /= 1024;
-        max_zd[i] /= 1024;
-        avg_zd[i] /= nprocs * 1024;
-    }
-
-    if (!opt.quiet && rank == 0) { /* only rank 0 reports timings */
-        printf("-------------------------------------------------------------\n");
-        printf("Memory footprints (min, max, avg among all processes):\n");
-        printf("construct metadata     (MiB) min=%8zd max=%8zd avg=%8zd\n",min_zd[0],max_zd[0],avg_zd[0]);
-        printf("total_alloc_bytes      (MiB) min=%8llu max=%8llu avg=%8llu\n",min_ull,max_ull,avg_ull);
-        // printf("curr_alloc_bytes       (MiB) min=%8zd max=%8zd avg=%8zd\n",min_zd[1],max_zd[1],avg_zd[1]);
-        printf("peak_alloc_bytes       (MiB) min=%8zd max=%8zd avg=%8zd\n",min_zd[2],max_zd[2],avg_zd[2]);
-        printf("max_block_size         (MiB) min=%8zd max=%8zd avg=%8zd\n",min_zd[3],max_zd[3],avg_zd[3]);
-        printf("total_alloc_blocks_count (K) min=%8zd max=%8zd avg=%8zd\n",min_zd[4],max_zd[4],avg_zd[4]);
-        // printf("curr_alloc_blocks_count  (K) min=%8zd max=%8zd avg=%8zd\n",min_zd[5],max_zd[5],avg_zd[5]);
-        printf("peak_alloc_blocks_count  (K) min=%8zd max=%8zd avg=%8zd\n",min_zd[6],max_zd[6],avg_zd[6]);
-        printf("-------------------------------------------------------------\n");
-    }
-#endif
 #endif
 
 prog_exit: