summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorQuincey Koziol <koziol@hdfgroup.org>2002-06-24 13:47:14 (GMT)
committerQuincey Koziol <koziol@hdfgroup.org>2002-06-24 13:47:14 (GMT)
commitace1f474b7135b1c479c5232643cabc477038350 (patch)
tree169a3af1de0b5b5dc937c30c831446832cd8149e
parent4e184b81a232187ae6f21daf783c35f0bc6dacc1 (diff)
downloadhdf5-ace1f474b7135b1c479c5232643cabc477038350.zip
hdf5-ace1f474b7135b1c479c5232643cabc477038350.tar.gz
hdf5-ace1f474b7135b1c479c5232643cabc477038350.tar.bz2
[svn-r5690] Purpose:
Bug Fix Description: When parallel I/O is used, the MPI-I/O VFL driver uses a "lazy" model to call MPI_File_set_view() in order to reduce the number of calls to this function. However, this is unsafe, because if a collective I/O which uses MPI derived types (and thus uses MPI_File_set_view()) is immediately followed by an independent I/O, the code will attempt to call MPI_File_set_view() in order to switch back to the default view of the file. MPI_File_set_view() is a collective call however, and this causes the application to hang. Solution: Removed "lazy" MPI_File_set_view() code, instead set the file view when it is needed (with MPI derived types) and immediately set the file view back to the default view before leaving the I/O routine. Platforms tested: IRIX64 6.5 (modi4) w/parallel. Also, tested with the latest development and release code for the SAF library, which now works correctly with this change. (Although the release branch of the SAF library seems to have a bug, this 1.4.4 release candidate code gets as far as the version the SAF library is released on top of (1.4.2-patch1, I believe)).
-rw-r--r--release_docs/RELEASE.txt3
-rw-r--r--src/H5FDmpio.c71
2 files changed, 45 insertions, 29 deletions
diff --git a/release_docs/RELEASE.txt b/release_docs/RELEASE.txt
index c07a2a3..58fc2b5 100644
--- a/release_docs/RELEASE.txt
+++ b/release_docs/RELEASE.txt
@@ -35,6 +35,9 @@ Bug Fixes since HDF5-1.4.0
Library
-------
+ * Fixed bug in parallel I/O routines where a collective I/O which used
+ MPI derived types, followed by an independent I/O would cause the library
+ to hang. QAK 2002/06/24
* Fixed bug in chunking routines where they were using internal allocation
free routines, instead of malloc/free, preventing user filters from
working correctly. Chunks are now allocated/freed with malloc/free and
diff --git a/src/H5FDmpio.c b/src/H5FDmpio.c
index 3fee0d9..7f36862 100644
--- a/src/H5FDmpio.c
+++ b/src/H5FDmpio.c
@@ -56,7 +56,6 @@ typedef struct H5FD_mpio_t {
haddr_t eof; /*end-of-file marker */
haddr_t eoa; /*end-of-address marker */
haddr_t last_eoa; /* Last known end-of-address marker */
- unsigned old_use_view; /*remember value of use_view */
} H5FD_mpio_t;
/* Prototypes */
@@ -1178,6 +1177,11 @@ H5FD_mpio_get_eof(H5FD_t *_file)
* the address of the dataset in MPI_File_set_view() calls, as
* necessary.
*
+ * Quincey Koziol - 2002/06/24
+ * Removed "lazy" MPI_File_set_view() calls, since they would fail
+ * if the first I/O was a collective I/O using MPI derived types
+ * and the next I/O was an independent I/O.
+ *
*-------------------------------------------------------------------------
*/
static herr_t
@@ -1191,7 +1195,7 @@ H5FD_mpio_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t dxpl_id, haddr_t add
MPI_Status mpi_stat;
MPI_Datatype buf_type, file_type;
int size_i, bytes_read, n;
- unsigned use_view_this_time=0, used_view_last_time;
+ unsigned use_view_this_time=0;
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value=SUCCEED;
@@ -1256,7 +1260,8 @@ H5FD_mpio_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t dxpl_id, haddr_t add
*/
mpi_disp=mpi_off;
mpi_off=0;
- } else {
+ } /* end if */
+ else {
/*
* Prepare for a simple xfer of a contiguous block of bytes. The
* btype, ftype, and disp fields are not used.
@@ -1264,23 +1269,17 @@ H5FD_mpio_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t dxpl_id, haddr_t add
buf_type = MPI_BYTE;
file_type = MPI_BYTE;
mpi_disp = 0; /* mpi_off is alread set */
- }
+ } /* end else */
/*
- * Don't bother to reset the view if we're not using the types this time,
- * and did we didn't use them last time either.
+ * Set the file view when we are using MPI derived types
*/
- used_view_last_time = file->old_use_view;
- if (used_view_last_time || /* change to new ftype or MPI_BYTE */
- use_view_this_time) { /* almost certainly a different ftype */
+ if (use_view_this_time) {
/*OKAY: CAST DISCARDS CONST QUALIFIER*/
if (MPI_SUCCESS != MPI_File_set_view(file->f, mpi_disp, MPI_BYTE, file_type, (char*)"native", file->info))
HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_set_view failed");
- }
+ } /* end if */
- /* Keep the 'use view' flag around for the next I/O */
- file->old_use_view = use_view_this_time;
-
/* Read the data. */
assert(H5FD_MPIO_INDEPENDENT==dx->xfer_mode || H5FD_MPIO_COLLECTIVE==dx->xfer_mode);
if (H5FD_MPIO_INDEPENDENT==dx->xfer_mode) {
@@ -1342,6 +1341,15 @@ H5FD_mpio_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t dxpl_id, haddr_t add
HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed");
/*
+ * Reset the file view when we used MPI derived types
+ */
+ if (use_view_this_time) {
+ /*OKAY: CAST DISCARDS CONST QUALIFIER*/
+ if (MPI_SUCCESS != MPI_File_set_view(file->f, 0, MPI_BYTE, MPI_BYTE, (char*)"native", file->info))
+ HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_set_view failed");
+ } /* end if */
+
+ /*
* This gives us zeroes beyond end of physical MPI file. What about
* reading past logical end of HDF5 file???
*/
@@ -1471,6 +1479,11 @@ done:
* the address of the dataset in MPI_File_set_view() calls, as
* necessary.
*
+ * Quincey Koziol - 2002/06/24
+ * Removed "lazy" MPI_File_set_view() calls, since they would fail
+ * if the first I/O was a collective I/O using MPI derived types
+ * and the next I/O was an independent I/O.
+ *
*-------------------------------------------------------------------------
*/
static herr_t
@@ -1484,7 +1497,7 @@ H5FD_mpio_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr,
MPI_Status mpi_stat;
MPI_Datatype buf_type, file_type;
int size_i, bytes_written;
- unsigned use_view_this_time=0, used_view_last_time;
+ unsigned use_view_this_time=0;
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value=SUCCEED;
@@ -1549,7 +1562,8 @@ H5FD_mpio_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr,
*/
mpi_disp=mpi_off;
mpi_off=0;
- } else {
+ } /* end if */
+ else {
/*
* Prepare for a simple xfer of a contiguous block of bytes.
* The btype, ftype, and disp fields are not used.
@@ -1557,27 +1571,17 @@ H5FD_mpio_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr,
buf_type = MPI_BYTE;
file_type = MPI_BYTE;
mpi_disp = 0; /* mpi_off is already set */
- }
+ } /* end else */
/*
- * Don't bother to reset the view if we're not using the types this time,
- * and did we didn't use them last time either.
+ * Set the file view when we are using MPI derived types
*/
- used_view_last_time = file->old_use_view;
- if (used_view_last_time || /* change to new ftype or MPI_BYTE */
- use_view_this_time) { /* almost certainly a different ftype */
+ if (use_view_this_time) {
/*OKAY: CAST DISCARDS CONST QUALIFIER*/
if (MPI_SUCCESS != MPI_File_set_view(file->f, mpi_disp, MPI_BYTE, file_type, (char*)"native", file->info))
HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_set_view failed");
- }
+ } /* end if */
- /*
- * We always set the use_view flag to 0 because the default is not to
- * use types next time, unless someone explicitly requests it by setting
- * this flag to !=0.
- */
- file->old_use_view = use_view_this_time;
-
/* Only p<round> will do the actual write if all procs in comm write same data */
if ((type!=H5FD_MEM_DRAW) && H5_mpi_1_metawrite_g) {
if (file->mpi_rank != file->mpi_round) {
@@ -1654,6 +1658,15 @@ H5FD_mpio_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr,
if (bytes_written<0 || bytes_written>size_i)
HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file write failed");
+ /*
+ * Reset the file view when we used MPI derived types
+ */
+ if (use_view_this_time) {
+ /*OKAY: CAST DISCARDS CONST QUALIFIER*/
+ if (MPI_SUCCESS != MPI_File_set_view(file->f, 0, MPI_BYTE, MPI_BYTE, (char*)"native", file->info))
+ HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_File_set_view failed");
+ } /* end if */
+
/* Forget the EOF value (see H5FD_mpio_get_eof()) --rpm 1999-08-06 */
file->eof = HADDR_UNDEF;