From ac2e0d548eaab1cb3c3875ed4cd842de8aed6b41 Mon Sep 17 00:00:00 2001 From: Leon Arber Date: Tue, 28 Jun 2005 14:20:56 -0500 Subject: [svn-r10996] Purpose: Bug fix for parallel case for new "Some objects were not comparable" patch to h5diff. Description: The parallel h5diff wouldn't print out "Some objects were not comparable" because the worker tasks were not communicating the not_cmp flag of the diff_opt_t struct back to the manager, who ultimately prints everything. Also, some miscellaneous fixes for error printing. Some errors were printed out with printf instead of fprintf(stderr,...). In parallel environments, this can result in output getting lost. Solution: Had the worker tasks pass along the not_cmp flag to the manager when they sent along the number of differences they found. Platforms tested: heping (pp), sol (pp) Misc. update: --- tools/h5diff/h5diff_common.c | 62 ++++++++++++-------------- tools/h5diff/h5diff_common.h | 3 ++ tools/h5diff/ph5diff_main.c | 13 +++--- tools/lib/h5diff.c | 102 ++++++++++++++++++++++++------------------- tools/lib/h5diff_util.c | 2 +- tools/lib/ph5diff.h | 6 +++ 6 files changed, 101 insertions(+), 87 deletions(-) diff --git a/tools/h5diff/h5diff_common.c b/tools/h5diff/h5diff_common.c index 48bcedb..5387f2a 100644 --- a/tools/h5diff/h5diff_common.c +++ b/tools/h5diff/h5diff_common.c @@ -17,9 +17,6 @@ #include "h5diff.h" #include "h5diff_common.h" -extern int g_Parallel; -extern int g_nTasks; - /*------------------------------------------------------------------------- * Function: parse_input * @@ -185,36 +182,35 @@ void parse_input(int argc, const char* argv[], const char** fname1, const char** void print_results(hsize_t nfound, diff_opt_t* options) { -/*------------------------------------------------------------------------- - * print how many differences were found - *------------------------------------------------------------------------- - */ - if (!options->m_quiet) - { - printf("----------------------------------------------------\n"); - printf("Summary\n"); - printf("----------------------------------------------------\n"); - - if (options->cmn_objs==0 && !options->err_stat) - { - printf("No common objects found. Files are not comparable.\n"); - if (!options->m_verbose) - printf("Use -v for a list of objects.\n"); - } - else - { - /* no errors found */ - if (!options->err_stat) - { - /* objects were not compared */ - if (options->not_cmp==1) - printf("Some objects are not comparable\n"); - else - /* objects were compared, print the number of differences */ - print_found(nfound); - } - } - } + /*------------------------------------------------------------------------- + * print how many differences were found + *------------------------------------------------------------------------- + */ + if (!options->m_quiet) + { + printf("----------------------------------------------------\n"); + printf("Summary\n"); + printf("----------------------------------------------------\n"); + if (options->cmn_objs==0 && !options->err_stat) + { + printf("No common objects found. Files are not comparable.\n"); + if (!options->m_verbose) + printf("Use -v for a list of objects.\n"); + } + else + { + /* no errors found */ + if (!options->err_stat) + { + /* objects were not compared */ + if (options->not_cmp==1) + printf("Some objects are not comparable\n"); + else + /* objects were compared, print the number of differences */ + print_found(nfound); + } + } + } } /*------------------------------------------------------------------------- diff --git a/tools/h5diff/h5diff_common.h b/tools/h5diff/h5diff_common.h index 3a6f01d..728b022 100644 --- a/tools/h5diff/h5diff_common.h +++ b/tools/h5diff/h5diff_common.h @@ -1,3 +1,6 @@ +extern unsigned char g_Parallel; +extern int g_nTasks; + void usage(void); int check_n_input( const char* ); int check_f_input( const char* ); diff --git a/tools/h5diff/ph5diff_main.c b/tools/h5diff/ph5diff_main.c index 082a6bd..3b6267d 100644 --- a/tools/h5diff/ph5diff_main.c +++ b/tools/h5diff/ph5diff_main.c @@ -131,6 +131,7 @@ ph5diff_worker(int nID) char filenames[2][1024]; char out_data[PRINT_DATA_MAX_SIZE] = {0}; hsize_t nfound=0; + struct diffs_found diffs; int i; MPI_Status Status; @@ -169,6 +170,8 @@ ph5diff_worker(int nID) MPI_Recv(&args, sizeof(struct diff_args), MPI_BYTE, 0, MPI_TAG_ARGS, MPI_COMM_WORLD, &Status); /*Do the diff */ nfound = diff(file1_id, args.name, file2_id, args.name, &(args.options), args.type); + diffs.nfound = nfound; + diffs.not_cmp = args.options.not_cmp; /*If print buffer has something in it, request print token.*/ if(outBuffOffset>0) @@ -213,16 +216,10 @@ ph5diff_worker(int nID) memset(outBuff, 0, OUTBUFF_SIZE); outBuffOffset = 0; - /* Since the data type of diff value is hsize_t which can - * be arbitary large such that there is no MPI type that - * matches it, the value is passed between processes as - * an array of bytes in order to be portable. But this - * may not work in non-homogeneous MPI environments. - */ - MPI_Send(&nfound, sizeof(nfound), MPI_BYTE, 0, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD); + MPI_Send(&diffs, sizeof(diffs), MPI_BYTE, 0, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD); } else - MPI_Send(&nfound, sizeof(nfound), MPI_BYTE, 0, MPI_TAG_DONE, MPI_COMM_WORLD); + MPI_Send(&diffs, sizeof(diffs), MPI_BYTE, 0, MPI_TAG_DONE, MPI_COMM_WORLD); } else if(Status.MPI_TAG == MPI_TAG_END) { diff --git a/tools/lib/h5diff.c b/tools/lib/h5diff.c index c7828e8..16e5af7 100644 --- a/tools/lib/h5diff.c +++ b/tools/lib/h5diff.c @@ -52,7 +52,7 @@ void phdiff_dismiss_workers(void) { int i; for(i=1; i0) && g_Parallel) - { - printf("%s", outBuff); + /* If there was something we buffered, let's print it now */ + if( (outBuffOffset>0) && g_Parallel) + { + printf("%s", outBuff); - if(overflow_file) - { - int tmp; + if(overflow_file) + { + int tmp; - rewind(overflow_file); - while((tmp = getc(overflow_file)) >= 0) - putchar(tmp); + rewind(overflow_file); + while((tmp = getc(overflow_file)) >= 0) + putchar(tmp); - fclose(overflow_file); - overflow_file = NULL; - } + fclose(overflow_file); + overflow_file = NULL; + } - fflush(stdout); - memset(outBuff, 0, OUTBUFF_SIZE); - outBuffOffset = 0; - } - else if( (outBuffOffset>0) && !g_Parallel) - { - printf("h5diff error: outBuffOffset>0, but we're not in parallel!\n"); - } + fflush(stdout); + memset(outBuff, 0, OUTBUFF_SIZE); + outBuffOffset = 0; + } + else if( (outBuffOffset>0) && !g_Parallel) + { + fprintf(stderr, "h5diff error: outBuffOffset>0, but we're not in parallel!\n"); + } } /*------------------------------------------------------------------------- @@ -119,17 +119,17 @@ static void print_incoming_data(void) char data[PRINT_DATA_MAX_SIZE+1]; int incomingMessage; MPI_Status Status; - + do { - MPI_Iprobe(MPI_ANY_SOURCE, MPI_TAG_PRINT_DATA, MPI_COMM_WORLD, &incomingMessage, &Status); - if(incomingMessage) - { - memset(data, 0, PRINT_DATA_MAX_SIZE+1); - MPI_Recv(data, PRINT_DATA_MAX_SIZE, MPI_CHAR, Status.MPI_SOURCE, MPI_TAG_PRINT_DATA, MPI_COMM_WORLD, &Status); - - printf("%s", data); - } + MPI_Iprobe(MPI_ANY_SOURCE, MPI_TAG_PRINT_DATA, MPI_COMM_WORLD, &incomingMessage, &Status); + if(incomingMessage) + { + memset(data, 0, PRINT_DATA_MAX_SIZE+1); + MPI_Recv(data, PRINT_DATA_MAX_SIZE, MPI_CHAR, Status.MPI_SOURCE, MPI_TAG_PRINT_DATA, MPI_COMM_WORLD, &Status); + + printf("%s", data); + } } while(incomingMessage); } #endif @@ -163,6 +163,7 @@ h5diff (const char *fname1, hid_t file1_id=(-1), file2_id=(-1); char filenames[2][1024]; hsize_t nfound = 0; + int not_cmp = 0; memset(filenames, 0, 1024*2); @@ -308,7 +309,7 @@ h5diff (const char *fname1, if( (strlen(fname1) > 1024) || (strlen(fname2) > 1024)) { - printf("The parallel diff only supports path names up to 1024 characters\n"); + fprintf(stderr, "The parallel diff only supports path names up to 1024 characters\n"); MPI_Abort(MPI_COMM_WORLD, 0); } @@ -323,6 +324,7 @@ h5diff (const char *fname1, #endif nfound = diff_match (file1_id, nobjects1, info1, file2_id, nobjects2, info2, options); + } @@ -475,7 +477,7 @@ diff_match (hid_t file1_id, char* workerTasks = malloc((g_nTasks-1) * sizeof(char)); int n; int busyTasks=0; - hsize_t nFoundbyWorker; + struct diffs_found nFoundbyWorker; struct diff_args args; int havePrintToken = 1; MPI_Status Status; @@ -532,7 +534,8 @@ diff_match (hid_t file1_id, { workerTasks[Status.MPI_SOURCE-1] = 1; MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_DONE, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; busyTasks--; } @@ -547,7 +550,8 @@ diff_match (hid_t file1_id, { workerTasks[Status.MPI_SOURCE-1] = 1; MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; busyTasks--; havePrintToken = 1; } @@ -598,9 +602,10 @@ diff_match (hid_t file1_id, * if we don't have the token, some task is currently printing so we'll wait for that task to return it. */ if(!havePrintToken) { - MPI_Recv(&nFoundbyWorker, sizeof(hsize_t), MPI_BYTE, MPI_ANY_SOURCE, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD, &Status); + MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, MPI_ANY_SOURCE, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD, &Status); havePrintToken = 1; - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; /* send this task the work unit. */ MPI_Send(&args, sizeof(struct diff_args), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_ARGS, MPI_COMM_WORLD); } @@ -614,7 +619,8 @@ diff_match (hid_t file1_id, if(Status.MPI_TAG == MPI_TAG_DONE) { MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_DONE, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; MPI_Send(&args, sizeof(struct diff_args), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_ARGS, MPI_COMM_WORLD); } else if(Status.MPI_TAG == MPI_TAG_TOK_REQUEST) @@ -623,7 +629,8 @@ diff_match (hid_t file1_id, MPI_Send(NULL, 0, MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_PRINT_TOK, MPI_COMM_WORLD); MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; MPI_Send(&args, sizeof(struct diff_args), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_ARGS, MPI_COMM_WORLD); } else @@ -647,14 +654,16 @@ diff_match (hid_t file1_id, MPI_Probe(MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_COMM_WORLD, &Status); if(Status.MPI_TAG == MPI_TAG_DONE) { - MPI_Recv(&nFoundbyWorker, sizeof(hsize_t), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_DONE, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_DONE, MPI_COMM_WORLD, &Status); + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; busyTasks--; } else if(Status.MPI_TAG == MPI_TAG_TOK_RETURN) { MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_DONE, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; busyTasks--; havePrintToken = 1; } @@ -665,14 +674,16 @@ diff_match (hid_t file1_id, { MPI_Send(NULL, 0, MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_PRINT_TOK, MPI_COMM_WORLD); MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; busyTasks--; } else /* someone else must have it...wait for them to return it, then give it to the task that just asked for it. */ { int source = Status.MPI_SOURCE; MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, MPI_ANY_SOURCE, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; busyTasks--; MPI_Send(NULL, 0, MPI_BYTE, source, MPI_TAG_PRINT_TOK, MPI_COMM_WORLD); } @@ -680,7 +691,8 @@ diff_match (hid_t file1_id, else if(Status.MPI_TAG == MPI_TAG_TOK_RETURN) { MPI_Recv(&nFoundbyWorker, sizeof(nFoundbyWorker), MPI_BYTE, Status.MPI_SOURCE, MPI_TAG_TOK_RETURN, MPI_COMM_WORLD, &Status); - nfound += nFoundbyWorker; + nfound += nFoundbyWorker.nfound; + options->not_cmp = options->not_cmp | nFoundbyWorker.not_cmp; busyTasks--; havePrintToken = 1; } diff --git a/tools/lib/h5diff_util.c b/tools/lib/h5diff_util.c index d29434c..3a9e7e4 100644 --- a/tools/lib/h5diff_util.c +++ b/tools/lib/h5diff_util.c @@ -64,7 +64,7 @@ void parallel_print(const char* format, ...) overflow_file = HDtmpfile(); if(overflow_file == NULL) - printf("Warning: Could not create overflow file. Output may be truncated.\n"); + fprintf(stderr, "Warning: Could not create overflow file. Output may be truncated.\n"); else bytes_written = HDvfprintf(overflow_file, format, ap); } diff --git a/tools/lib/ph5diff.h b/tools/lib/ph5diff.h index b8b38b2..864ae49 100644 --- a/tools/lib/ph5diff.h +++ b/tools/lib/ph5diff.h @@ -44,6 +44,12 @@ struct diff_args diff_opt_t options; }; +struct diffs_found +{ + hsize_t nfound; + int not_cmp; +}; + #ifdef H5_HAVE_PARALLEL #include #endif -- cgit v0.12