summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorBill Wendling <wendling@ncsa.uiuc.edu>2003-03-19 23:41:28 (GMT)
committerBill Wendling <wendling@ncsa.uiuc.edu>2003-03-19 23:41:28 (GMT)
commit9697077305b892b69262b263b775eeeed7b8717d (patch)
treeceb67b839d8b809ba8523ef79124deed50b84ad7 /src
parent43eb81fb3c6997ded3a94593d543c413dfa2ce5a (diff)
downloadhdf5-9697077305b892b69262b263b775eeeed7b8717d.zip
hdf5-9697077305b892b69262b263b775eeeed7b8717d.tar.gz
hdf5-9697077305b892b69262b263b775eeeed7b8717d.tar.bz2
[svn-r6505] Purpose:
Update Description: Have the FPHDF5 module initialized on startup. Have only the captain process perform certain functions (like flushing during an F_close call or allocating the superblock). H5Fistore needed a few checks to see if it was working with an FPHDF5 driver. Done similarly to how MPI and MPIPOSIX drivers are checked.. Platforms tested: Linux, Modi4, Sol Misc. update:
Diffstat (limited to 'src')
-rw-r--r--src/H5Distore.c20
-rw-r--r--src/H5F.c206
-rw-r--r--src/H5Fistore.c20
3 files changed, 179 insertions, 67 deletions
diff --git a/src/H5Distore.c b/src/H5Distore.c
index 6ac8ee3..c90d71b 100644
--- a/src/H5Distore.c
+++ b/src/H5Distore.c
@@ -1802,12 +1802,13 @@ H5F_istore_read(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
chunk_addr!=HADDR_UNDEF)
/*
- * If MPIO or MPIPOSIX is used and file can be written to, we must bypass the
- * chunk-cache scheme because other MPI processes could be writing to
- * other elements in the same chunk.
- * Do a direct write-through of only the elements requested.
+ * If MPIO, MPIPOSIX, or FPHDF5 is used and file can be written
+ * to, we must bypass the chunk-cache scheme because other MPI
+ * processes could be writing to other elements in the same
+ * chunk. Do a direct write-through of only the elements
+ * requested.
*/
- || ((IS_H5FD_MPIO(f) ||IS_H5FD_MPIPOSIX(f)) && (H5F_ACC_RDWR & f->shared->flags))
+ || ((IS_H5FD_MPIO(f) ||IS_H5FD_MPIPOSIX(f) || IS_H5FD_FPHDF5(f)) && (H5F_ACC_RDWR & f->shared->flags))
) {
H5O_layout_t l; /* temporary layout */
@@ -1983,11 +1984,12 @@ H5F_istore_write(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
chunk_addr!=HADDR_UNDEF)
/*
- * If MPIO or MPIPOSIX is used, must bypass the chunk-cache scheme because other
- * MPI processes could be writing to other elements in the same chunk.
- * Do a direct write-through of only the elements requested.
+ * If MPIO, MPIPOSIX, or FPHDF5 is used, must bypass the
+ * chunk-cache scheme because other MPI processes could be
+ * writing to other elements in the same chunk. Do a direct
+ * write-through of only the elements requested.
*/
- || ((IS_H5FD_MPIO(f) ||IS_H5FD_MPIPOSIX(f)) && (H5F_ACC_RDWR & f->shared->flags))
+ || ((IS_H5FD_MPIO(f) ||IS_H5FD_MPIPOSIX(f) || IS_H5FD_FPHDF5(f)) && (H5F_ACC_RDWR & f->shared->flags))
) {
H5O_layout_t l; /* temporary layout */
diff --git a/src/H5F.c b/src/H5F.c
index fdca262..c096c21 100644
--- a/src/H5F.c
+++ b/src/H5F.c
@@ -12,13 +12,13 @@
* access to either file, you may request a copy from hdfhelp@ncsa.uiuc.edu. *
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
-/****************************************************************************
-* *
-* MODIFICATIONS *
-* Robb Matzke, 30 Aug 1997 *
-* Added `ERRORS' fields to function prologues. *
-* *
-****************************************************************************/
+/*****************************************************************************
+ * *
+ * MODIFICATIONS *
+ * Robb Matzke, 30 Aug 1997 *
+ * Added `ERRORS' fields to function prologues. *
+ * *
+ ****************************************************************************/
/* $Id$ */
@@ -27,6 +27,7 @@
/* Predefined file drivers */
#include "H5FDcore.h" /*temporary in-memory files */
#include "H5FDfamily.h" /*family of files */
+#include "H5FDfphdf5.h" /*FPHDF5 */
#include "H5FDgass.h" /*GASS I/O */
#include "H5FDlog.h" /* sec2 driver with logging, for debugging */
#include "H5FDmpio.h" /*MPI-2 I/O */
@@ -46,6 +47,7 @@
#include "H5Fpkg.h" /*file access */
#include "H5FDprivate.h" /*file driver */
#include "H5FLprivate.h" /*Free Lists */
+#include "H5FPprivate.h" /*Flexible Parallel HDF5 */
#include "H5Iprivate.h" /*object IDs */
#include "H5Gprivate.h" /*symbol tables */
#include "H5MMprivate.h" /*core memory management */
@@ -326,6 +328,9 @@ H5F_init_interface(void)
#ifdef H5_HAVE_PARALLEL
if ((status=H5FD_MPIO)<0) goto end_registration;
if ((status=H5FD_MPIPOSIX)<0) goto end_registration;
+#ifdef H5_HAVE_FPHDF5
+ if ((status=H5FD_FPHDF5)<0) goto end_registration;
+#endif /* H5_HAVE_FPHDF5 */
#endif /* H5_HAVE_PARALLEL */
#ifdef H5_HAVE_STREAM
if ((status=H5FD_STREAM)<0) goto end_registration;
@@ -1642,7 +1647,7 @@ H5F_dest(H5F_t *f, hid_t dxpl_id)
--f->nrefs;
}
-done:
+done:
FUNC_LEAVE_NOAPI(ret_value);
}
@@ -1873,6 +1878,33 @@ H5F_open(const char *name, unsigned flags, hid_t fcpl_id, hid_t fapl_id, hid_t d
* empty or not.
*/
if (0==H5FD_get_eof(lf) && (flags & H5F_ACC_RDWR)) {
+ /*
+ * We've just opened a fresh new file (or truncated one). We need
+ * to write the superblock.
+ */
+#ifdef H5_HAVE_FPHDF5
+ /*
+ * If this is an FPHDF5 driver, then we want to set a property
+ * which says so and that only the captain process will be able
+ * to allocate things for the duration. That is, only the captain
+ * process should allocate the superblock of a file.
+ */
+ if (H5FD_is_fphdf5_driver(lf)) {
+ unsigned value = 1;
+ H5P_genplist_t *d_plist;
+
+ /* Get the data xfer property list */
+ if ((d_plist = H5I_object(dxpl_id)) == NULL)
+ HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, NULL, "not a dataset transfer list");
+
+ /* Set that only captain should allocate */
+ if (H5P_insert(d_plist, H5FD_FPHDF5_CAPTN_ALLOC_ONLY,
+ H5FD_FPHDF5_CAPTN_ALLOC_SIZE, &value,
+ NULL, NULL, NULL, NULL, NULL) < 0)
+ HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, NULL, "can't insert FPHDF5 property");
+ }
+#endif /* H5_HAVE_FPHDF5 */
+
/* Get values to cache from the FCPL */
if(H5P_get(c_plist, H5F_CRT_ADDR_BYTE_NUM_NAME,&shared->sizeof_addr)<0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, NULL, "unable to set byte number in an address");
@@ -1900,7 +1932,24 @@ H5F_open(const char *name, unsigned flags, hid_t fcpl_id, hid_t fapl_id, hid_t d
/* Create and open the root group */
if (H5G_mkroot(file, dxpl_id, NULL)<0)
HGOTO_ERROR(H5E_FILE, H5E_CANTINIT, NULL, "unable to create/open root group");
-
+
+#ifdef H5_HAVE_FPHDF5
+ /*
+ * If this is an FPHDF5 driver, then remove the property which
+ * says so since we're allocating the superblock.
+ */
+ if (H5FD_is_fphdf5_driver(lf)) {
+ H5P_genplist_t *d_plist;
+
+ /* Get the data xfer property list */
+ if ((d_plist = H5I_object(dxpl_id)) == NULL)
+ HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, NULL, "not a dataset transfer list");
+
+ if (H5P_remove(dxpl_id, d_plist, H5FD_FPHDF5_CAPTN_ALLOC_ONLY) < 0)
+ HGOTO_ERROR(H5E_PLIST, H5E_CANTDELETE, NULL,
+ "can't remove FPHDF5 property");
+ }
+#endif /* H5_HAVE_FPHDF5 */
} else if (1==shared->nrefs) {
/* Read the superblock if it hasn't been read before. */
if (HADDR_UNDEF==(shared->boot_addr=H5F_locate_signature(lf,dxpl_id)))
@@ -2463,10 +2512,11 @@ H5F_flush(H5F_t *f, hid_t dxpl_id, H5F_scope_t scope, unsigned flags)
FUNC_ENTER_NOINIT(H5F_flush);
/*
- * Nothing to do if the file is read only. This determination is made at
- * the shared open(2) flags level, implying that opening a file twice,
- * once for read-only and once for read-write, and then calling
- * H5F_flush() with the read-only handle, still causes data to be flushed.
+ * Nothing to do if the file is read only. This determination is
+ * made at the shared open(2) flags level, implying that opening a
+ * file twice, once for read-only and once for read-write, and then
+ * calling H5F_flush() with the read-only handle, still causes data
+ * to be flushed.
*/
if (0 == (H5F_ACC_RDWR & f->shared->flags))
HGOTO_DONE(SUCCEED);
@@ -2622,31 +2672,56 @@ H5F_flush(H5F_t *f, hid_t dxpl_id, H5F_scope_t scope, unsigned flags)
if (flags & H5F_FLUSH_ALLOC_ONLY) {
haddr_t addr;
- /*
- * Allocate space for the userblock, superblock, and driver info
- * block. We do it with one allocation request because the
- * userblock and superblock need to be at the beginning of the
- * file and only the first allocation request is required to
- * return memory at format address zero.
- */
- H5_CHECK_OVERFLOW(f->shared->base_addr,haddr_t,hsize_t);
- addr = H5FD_alloc(f->shared->lf, H5FD_MEM_SUPER, dxpl_id,
- ((hsize_t)f->shared->base_addr + superblock_size + driver_size));
+#ifdef H5_HAVE_FPHDF5
+ /*
+ * If this is an FPHDF5 file driver, then we only want the
+ * captain process allocating the space. The rest of the
+ * processes should just get the broadcast message sent from the
+ * captain.
+ */
+ if (!H5FD_is_fphdf5_driver(f->shared->lf) ||
+ H5FD_fphdf5_is_captain(f->shared->lf)) {
+#endif /* H5_HAVE_FPHDF5 */
- if (HADDR_UNDEF == addr)
- HGOTO_ERROR(H5E_FILE, H5E_CANTINIT, FAIL,
- "unable to allocate file space for userblock and/or superblock");
+ /*
+ * Allocate space for the userblock, superblock, and driver
+ * info block. We do it with one allocation request because
+ * the userblock and superblock need to be at the beginning
+ * of the file and only the first allocation request is
+ * required to return memory at format address zero.
+ */
+ H5_CHECK_OVERFLOW(f->shared->base_addr,haddr_t,hsize_t);
+
+ addr = H5FD_alloc(f->shared->lf, H5FD_MEM_SUPER, dxpl_id,
+ ((hsize_t)f->shared->base_addr +
+ superblock_size + driver_size));
+
+#ifdef H5_HAVE_FPHDF5
+ }
+
+ if (H5FD_is_fphdf5_driver(f->shared->lf)) {
+ int mrc;
- if (0 != addr)
- HGOTO_ERROR(H5E_FILE, H5E_CANTINIT, FAIL,
+ if ((mrc = MPI_Bcast(&addr, 1, HADDR_AS_MPI_TYPE,
+ (int)H5FP_capt_barrier_rank,
+ H5FP_SAP_BARRIER_COMM)) != MPI_SUCCESS)
+ HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", mrc);
+ }
+#endif /* H5_HAVE_FPHDF5 */
+
+ if (HADDR_UNDEF == addr)
+ HGOTO_ERROR(H5E_FILE, H5E_CANTINIT, FAIL,
+ "unable to allocate file space for userblock and/or superblock");
+ if (0 != addr)
+ HGOTO_ERROR(H5E_FILE, H5E_CANTINIT, FAIL,
"file driver failed to allocate userblock and/or superblock at address zero");
- /*
- * The file driver information block begins immediately after the
- * superblock.
- */
- if (driver_size > 0)
- f->shared->driver_addr = superblock_size;
+ /*
+ * The file driver information block begins immediately after
+ * the superblock.
+ */
+ if (driver_size > 0)
+ f->shared->driver_addr = superblock_size;
} else {
/* Compute boot block checksum */
assert(sizeof(chksum)==sizeof(f->shared->boot_chksum));
@@ -2688,11 +2763,25 @@ H5F_flush(H5F_t *f, hid_t dxpl_id, H5F_scope_t scope, unsigned flags)
} /* end if */
} /* end else */
- /* If we're not just allocating... */
- if ((flags & H5F_FLUSH_ALLOC_ONLY) == 0)
- /* ...flush file buffers to disk. */
- if (H5FD_flush(f->shared->lf, dxpl_id, (flags&H5F_FLUSH_CLOSING)>0) < 0)
- HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "low level flush failed");
+#ifdef H5_HAVE_FPHDF5
+ /*
+ * We only want the captain to perform the flush of the metadata to
+ * the file.
+ */
+ if (!H5FD_is_fphdf5_driver(f->shared->lf) ||
+ H5FD_fphdf5_is_captain(f->shared->lf)) {
+#endif /* H5_HAVE_FPHDF5 */
+
+ /* If we're not just allocating... */
+ if ((flags & H5F_FLUSH_ALLOC_ONLY) == 0)
+ /* ...flush file buffers to disk. */
+ if (H5FD_flush(f->shared->lf, dxpl_id,
+ (unsigned)((flags & H5F_FLUSH_CLOSING) > 0)) < 0)
+ HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "low level flush failed");
+
+#ifdef H5_HAVE_FPHDF5
+ }
+#endif /* H5_HAVE_FPHDF5 */
/* Check flush errors for children - errors are already on the stack */
ret_value = (nerrors ? FAIL : SUCCEED);
@@ -2785,7 +2874,9 @@ H5F_close(H5F_t *f)
} /* end if */
f->mtab.nmounts = 0;
- /* Close file according to close degree:
+ /*
+ * Close file according to close degree:
+ *
* H5F_CLOSE_WEAK: if there are still objects open, wait until
* they are all closed.
* H5F_CLOSE_SEMI: if there are still objects open, return fail;
@@ -2887,15 +2978,32 @@ H5F_close(H5F_t *f)
} /* end switch */
/* Only flush at this point if the file will be closed */
- if(closing) {
- /* Dump debugging info */
- H5AC_debug(f);
- H5F_istore_stats(f, FALSE);
-
- /* Flush and destroy all caches */
- if (H5F_flush(f, H5AC_dxpl_id, H5F_SCOPE_LOCAL,
- H5F_FLUSH_INVALIDATE | H5F_FLUSH_CLOSING) < 0)
- HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "unable to flush cache");
+ if (closing) {
+#ifdef H5_HAVE_FPHDF5
+ /*
+ * We only want the captain to perform the flush of the metadata
+ * to the file.
+ */
+ if (!H5FD_is_fphdf5_driver(f->shared->lf) ||
+ H5FD_fphdf5_is_captain(f->shared->lf)) {
+#endif /* H5_HAVE_FPHDF5 */
+
+ /* Dump debugging info */
+ H5AC_debug(f);
+ H5F_istore_stats(f, FALSE);
+
+ /* Flush and destroy all caches */
+ if (H5F_flush(f, H5AC_dxpl_id, H5F_SCOPE_LOCAL,
+ H5F_FLUSH_INVALIDATE | H5F_FLUSH_CLOSING) < 0)
+ HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "unable to flush cache");
+
+#ifdef H5_HAVE_FPHDF5
+ }
+
+ /* Wait for the captain to finish up... */
+ if (H5FD_is_fphdf5_driver(f->shared->lf))
+ MPI_Barrier(H5FP_SAP_BARRIER_COMM);
+#endif /* H5_HAVE_FPHDF5 */
} /* end if */
/*
diff --git a/src/H5Fistore.c b/src/H5Fistore.c
index 6ac8ee3..c90d71b 100644
--- a/src/H5Fistore.c
+++ b/src/H5Fistore.c
@@ -1802,12 +1802,13 @@ H5F_istore_read(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
chunk_addr!=HADDR_UNDEF)
/*
- * If MPIO or MPIPOSIX is used and file can be written to, we must bypass the
- * chunk-cache scheme because other MPI processes could be writing to
- * other elements in the same chunk.
- * Do a direct write-through of only the elements requested.
+ * If MPIO, MPIPOSIX, or FPHDF5 is used and file can be written
+ * to, we must bypass the chunk-cache scheme because other MPI
+ * processes could be writing to other elements in the same
+ * chunk. Do a direct write-through of only the elements
+ * requested.
*/
- || ((IS_H5FD_MPIO(f) ||IS_H5FD_MPIPOSIX(f)) && (H5F_ACC_RDWR & f->shared->flags))
+ || ((IS_H5FD_MPIO(f) ||IS_H5FD_MPIPOSIX(f) || IS_H5FD_FPHDF5(f)) && (H5F_ACC_RDWR & f->shared->flags))
) {
H5O_layout_t l; /* temporary layout */
@@ -1983,11 +1984,12 @@ H5F_istore_write(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
chunk_addr!=HADDR_UNDEF)
/*
- * If MPIO or MPIPOSIX is used, must bypass the chunk-cache scheme because other
- * MPI processes could be writing to other elements in the same chunk.
- * Do a direct write-through of only the elements requested.
+ * If MPIO, MPIPOSIX, or FPHDF5 is used, must bypass the
+ * chunk-cache scheme because other MPI processes could be
+ * writing to other elements in the same chunk. Do a direct
+ * write-through of only the elements requested.
*/
- || ((IS_H5FD_MPIO(f) ||IS_H5FD_MPIPOSIX(f)) && (H5F_ACC_RDWR & f->shared->flags))
+ || ((IS_H5FD_MPIO(f) ||IS_H5FD_MPIPOSIX(f) || IS_H5FD_FPHDF5(f)) && (H5F_ACC_RDWR & f->shared->flags))
) {
H5O_layout_t l; /* temporary layout */