diff options
author | Neil Fortner <fortnern@gmail.com> | 2022-10-19 16:13:15 (GMT) |
---|---|---|
committer | GitHub <noreply@github.com> | 2022-10-19 16:13:15 (GMT) |
commit | 93754cae33d4ed45850745664ce5e59f270f38f8 (patch) | |
tree | ac3154f44fe7969dacb452474c32fcaeb84e620d /src/H5Dio.c | |
parent | a898cef6c2e277f74cc99302a3c49c39a7428b5b (diff) | |
download | hdf5-93754cae33d4ed45850745664ce5e59f270f38f8.zip hdf5-93754cae33d4ed45850745664ce5e59f270f38f8.tar.gz hdf5-93754cae33d4ed45850745664ce5e59f270f38f8.tar.bz2 |
Multi Dataset (#2120)
* Fix bug with cross platform compatibility of references within vlens.
No testing yet.
* Merge from multi_rd_wd_coll_io to a more recent branch from develop.
Untested, probably does not work yet.
* Committing clang-format changes
* Committing clang-format changes
* Fix many bugs in multi dataset branch. Mostly works, some issues in
SWMR tests.
* Committing clang-format changes
* Disable test in swmr.c that was failing due to bug in HDF5 unrelated to
multi dataset.
* Committing clang-format changes
* Fixed fortran multi-dataset tests
* Fixed xlf errors
* Added benchmark code for multi-datasets
* loops over datasets
* added missing error arg.
* Added gnuplot formatting
* Jonathan Kim original MD benchmarking code
* updated MD benchmarking code
* code clean-up
* Only make files in feature test mode
* misc clean-up
* removed TEST_MDSET_NO_LAST_DSET_2ND_PROC option
* Committing clang-format changes
* Change multi dataset API to use arrays of individual parameters instead
of the parameter struct.
* Committing clang-format changes
* Update to new multi dataset Fortran API and tests. (#1724)
* Update to new multi dataset Fortran API and tests.
* Sync Fortran with develop.
* skipping h5pget_mpio_actual_io_mode_f for now
* Fixed issue with dxpl_id, changed to variable size dim. (#1770)
* Remove "is_coll_broken" field from H5D_io_info_t struct
* Committing clang-format changes
* Minor cleanup in multi dataset code.
* Committing clang-format changes
* Clean up in multi dataset code.
* Committing clang-format changes
* Committing clang-format changes
* Fix speeling
* Fix bug in parallel compression. Switch base_maddr in io_info to be a
union.
* Committing clang-format changes
* Implement selection I/O support with multi dataset. Will be broken in
parallel until PR 1803 is merged to develop then the MDS branch.
* Committing clang-format changes
* Spelling
* Fix bug in multi dataset that could cause errors when only some of the
datasets in the multi dataset I/O used type conversion.
* Committing clang-format changes
* Integrate multi dataset APIs with VOL layer. Add async versions of
multi dataset APIs.
* Committing clang-format changes
* Spelling fixes
* Fix bug in non-parallel HDF5 compilation.
* Committing clang-format changes
* Fix potential memory/free list error. Minor performance fix. Other minor
changes.
* Committing clang-format changes
* Fix memory leak with memory dataspace for I/O.
* Committing clang-format changes
* Fix stack variables too large. Rename H5D_dset_info_t to
H5D_dset_io_info_t.
* Committing clang-format changes
* Remove mem_space_alloc field from H5D_dset_io_info_t. Each function is
now responsible for freeing any spaces it adds to dset_info.
* Committing clang-format changes
* fixed _multi Fortran declaration
* Refactor various things in (mostly) the serial I/O code path to make
things more maintainable.
* Committing clang-format changes
* updated to array based, doxygen, and examples
* Reinstate H5D_chunk_map_t, stored (via pointer) inside
H5D_dset_io_info_t.
* Change from calloc to malloc for H5D_dset_io_info_t and H5D_chunk_map_t.
Switch temporary dset_infos to be local stack variables.
* Committing clang-format changes
* format cleanup
* format cleanup
* added coll and ind
* Modify all parallel I/O paths to take dset_info instead of assuming
dset_info[0].
* Committing clang-format changes
* fixed output
* Rework parallel I/O code to work properly with multi dataset in more
cases. Fix bug in parallel compression.
* Committing clang-format changes
* Prevent H5D__multi_chunk_collective_io() from messing up collective opt
property for other datasets in I/O. Other minor cleanup. Add new test
case to t_pmulti_dset.c for H5FD_MPIO_INDIVIDUAL_IO, disabled for now
due to failures apparently unrelated to multi dataset code.
* Fix spelling
* Committing clang-format changes
* Replace N log N algorithm for finding chunk in
H5D__multi_chunk_collective_io() with O(N) algorithm, and remove use of
io_info->sel_pieces in that function.
* Committing clang-format changes
* Replace sel_pieces skiplist in io_info with flat array of pointers, use
qsort in I/O routine only when necessary.
* Committing clang-format changes
* Add new test case to mdset.c
* Committing clang-format changes
* Fix spelling
* Very minor fix in H5VL__native_dataset_read()
* Fix bug that could affect filtered parallel multi-dataset I/O.
* Add RM entries for H5Dread_multi(), H5Dread_multi_async(),
H5Dwrite_multi(), and H5Dwrite_multi_async()
* Unskip test in swmr.c
* Committing clang-format changes
* Eliminate H5D__pre_read and H5D__pre_write
* Remove examples/ph5mdsettest.c. Will fix and re-add as a test.
* Enable hyperslab combinations in mdset test
* Committing clang-format changes
* Clarify H5Dread/write_multi documentation.
* Fix bugs in multi-dataset I/O. Expand serial multi dataset test.
Update macro in parallel multi dataset test.
* Committing clang-format changes
* Spelling
* Remove obsolete entry in bin/trace
* Rework type conversion buffer allocation. Only one buffer is shared
between datasets in mdset mode, and it is malloced instead of calloced.
* Committing clang-format changes
* Fix bug in error handling in H5D__read/write
* added multi-dataset fortran check with optional dataset creation id (#2150)
* removed dup. dll entry
* Address comments from code review.
* Remove spurious changes in H5Fmpi.c
* Fix issue with reading unallocated datasets in multi-dataset mode.
Address other comments from code review.
* Committing clang-format changes
* Delay chunk index lookup from io_init to mdio_init so it doesn't add
overhead to single dataset I/O.
* Committing clang-format changes
* Fix inappropriate use of piece_count
* updated copyright on new file, removed benchmark from testing dir.
Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: M. Scot Breitenfeld <brtnfld@hdfgroup.org>
Co-authored-by: Dana Robinson <43805+derobins@users.noreply.github.com>
Diffstat (limited to 'src/H5Dio.c')
-rw-r--r-- | src/H5Dio.c | 1365 |
1 files changed, 870 insertions, 495 deletions
diff --git a/src/H5Dio.c b/src/H5Dio.c index c1427d1..0b52208 100644 --- a/src/H5Dio.c +++ b/src/H5Dio.c @@ -44,15 +44,14 @@ /********************/ /* Setup/teardown routines */ -static herr_t H5D__ioinfo_init(H5D_t *dset, const H5D_type_info_t *type_info, H5D_storage_t *store, - H5D_io_info_t *io_info); -static herr_t H5D__typeinfo_init(const H5D_t *dset, hid_t mem_type_id, hbool_t do_write, - H5D_type_info_t *type_info); +static herr_t H5D__ioinfo_init(size_t count, H5D_dset_io_info_t *dset_info, H5D_io_info_t *io_info); +static herr_t H5D__dset_ioinfo_init(H5D_t *dset, H5D_dset_io_info_t *dset_info, H5D_storage_t *store); +static herr_t H5D__typeinfo_init(H5D_io_info_t *io_info, H5D_dset_io_info_t *dset_info, hid_t mem_type_id); +static herr_t H5D__typeinfo_init_phase2(H5D_io_info_t *io_info); #ifdef H5_HAVE_PARALLEL -static herr_t H5D__ioinfo_adjust(H5D_io_info_t *io_info, const H5D_t *dset, const H5S_t *file_space, - const H5S_t *mem_space, const H5D_type_info_t *type_info); +static herr_t H5D__ioinfo_adjust(H5D_io_info_t *io_info); #endif /* H5_HAVE_PARALLEL */ -static herr_t H5D__typeinfo_term(const H5D_type_info_t *type_info); +static herr_t H5D__typeinfo_term(H5D_io_info_t *io_info, size_t type_info_init); /*********************/ /* Package Variables */ @@ -65,66 +64,63 @@ static herr_t H5D__typeinfo_term(const H5D_type_info_t *type_info); /* Declare a free list to manage blocks of type conversion data */ H5FL_BLK_DEFINE(type_conv); -/* Declare a free list to manage the H5D_chunk_map_t struct */ -H5FL_DEFINE(H5D_chunk_map_t); - /*------------------------------------------------------------------------- * Function: H5D__read * - * Purpose: Reads (part of) a DATASET into application memory BUF. See - * H5Dread() for complete details. + * Purpose: Reads multiple (parts of) DATASETs into application memory BUFs. + * See H5Dread_multi() for complete details. * * Return: Non-negative on success/Negative on failure * - * Programmer: Robb Matzke - * Thursday, December 4, 1997 - * *------------------------------------------------------------------------- */ herr_t -H5D__read(H5D_t *dataset, hid_t mem_type_id, H5S_t *mem_space, H5S_t *file_space, void *buf /*out*/) +H5D__read(size_t count, H5D_dset_io_info_t *dset_info) { - H5D_chunk_map_t *fm = NULL; /* Chunk file<->memory mapping */ - H5D_io_info_t io_info; /* Dataset I/O info */ - H5D_type_info_t type_info; /* Datatype info for operation */ - H5D_layout_t layout_type; /* Dataset's layout type (contig, chunked, compact, etc.) */ - hbool_t type_info_init = FALSE; /* Whether the datatype info has been initialized */ - H5S_t *projected_mem_space = NULL; /* If not NULL, ptr to dataspace containing a */ - /* projection of the supplied mem_space to a new */ - /* dataspace with rank equal to that of */ - /* file_space. */ - /* */ - /* This field is only used if */ - /* H5S_select_shape_same() returns TRUE when */ - /* comparing the mem_space and the data_space, */ - /* and the mem_space have different rank. */ - /* */ - /* Note that if this variable is used, the */ - /* projected mem space must be discarded at the */ - /* end of the function to avoid a memory leak. */ - H5D_storage_t store; /* union of EFL and chunk pointer in file space */ - hsize_t nelmts; /* total number of elmts */ - hbool_t io_op_init = FALSE; /* Whether the I/O op has been initialized */ - char fake_char; /* Temporary variable for NULL buffer pointers */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE_TAG(dataset->oloc.addr) + H5D_io_info_t io_info; /* Dataset I/O info for multi dsets */ + size_t type_info_init = 0; /* Number of datatype info structs that have been initialized */ + H5S_t *orig_mem_space_local; /* Local buffer for orig_mem_space */ + H5S_t **orig_mem_space = NULL; /* If not NULL, ptr to an array of dataspaces */ + /* containing the original memory spaces contained */ + /* in dset_info. This is needed in order to */ + /* restore the original state of dset_info if we */ + /* replaced any mem spaces with equivalents */ + /* projected to a rank equal to that of file_space. */ + /* */ + /* This field is only used if */ + /* H5S_select_shape_same() returns TRUE when */ + /* comparing at least one mem_space and data_space, */ + /* and the mem_space has a different rank. */ + /* */ + /* Note that this is a temporary variable - the */ + /* projected memory space is stored in dset_info, */ + /* and will be freed when that structure is */ + /* freed. */ + H5D_storage_t store_local; /* Local buffer for store */ + H5D_storage_t *store = &store_local; /* Union of EFL and chunk pointer in file space */ + size_t io_op_init = 0; /* Number I/O ops that have been initialized */ + size_t io_skipped = + 0; /* Number I/O ops that have been skipped (due to the dataset not being allocated) */ + size_t i; /* Local index variable */ + char fake_char; /* Temporary variable for NULL buffer pointers */ + herr_t ret_value = SUCCEED; /* Return value */ - /* check args */ - HDassert(dataset && dataset->oloc.file); - HDassert(file_space); - HDassert(mem_space); + FUNC_ENTER_NOAPI(FAIL) - layout_type = dataset->shared->layout.type; + /* Init io_info */ + if (H5D__ioinfo_init(count, dset_info, &io_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't initialize I/O info") + io_info.op_type = H5D_IO_OP_READ; - /* Set up datatype info for operation */ - if (H5D__typeinfo_init(dataset, mem_type_id, FALSE, &type_info) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info") - type_info_init = TRUE; + /* Allocate store buffer if necessary */ + if (count > 1) + if (NULL == (store = (H5D_storage_t *)H5MM_malloc(count * sizeof(H5D_storage_t)))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, "couldn't allocate dset storage info array buffer") #ifdef H5_HAVE_PARALLEL - /* Check for non-MPI-based VFD */ - if (!(H5F_HAS_FEATURE(dataset->oloc.file, H5FD_FEAT_HAS_MPI))) { + /* Check for non-MPI-based VFD. Only need to check first dataset since all + * share the same file. */ + if (!(H5F_HAS_FEATURE(dset_info[0].dset->oloc.file, H5FD_FEAT_HAS_MPI))) { H5FD_mpio_xfer_t io_xfer_mode; /* MPI I/O transfer mode */ /* Get I/O transfer mode */ @@ -137,363 +133,626 @@ H5D__read(H5D_t *dataset, hid_t mem_type_id, H5S_t *mem_space, H5S_t *file_space } /* end if */ #endif /*H5_HAVE_PARALLEL*/ - /* Make certain that the number of elements in each selection is the same */ - nelmts = H5S_GET_SELECT_NPOINTS(mem_space); - if (nelmts != H5S_GET_SELECT_NPOINTS(file_space)) - HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, - "src and dest dataspaces have different number of elements selected") - - /* Check for a NULL buffer */ - if (NULL == buf) { - /* Check for any elements selected (which is invalid) */ - if (nelmts > 0) - HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "no output buffer") - - /* If the buffer is nil, and 0 element is selected, make a fake buffer. - * This is for some MPI package like ChaMPIon on NCSA's tungsten which - * doesn't support this feature. - */ - buf = &fake_char; - } /* end if */ - - /* Make sure that both selections have their extents set */ - if (!(H5S_has_extent(file_space))) - HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "file dataspace does not have extent set") - if (!(H5S_has_extent(mem_space))) - HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "memory dataspace does not have extent set") + /* iterate over all dsets and construct I/O information necessary to do I/O */ + for (i = 0; i < count; i++) { + haddr_t prev_tag = HADDR_UNDEF; + + /* check args */ + if (NULL == dset_info[i].dset) + HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a dataset") + if (NULL == dset_info[i].dset->oloc.file) + HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file") + + /* set metadata tagging with dset oheader addr */ + H5AC_tag(dset_info[i].dset->oloc.addr, &prev_tag); + + /* Set up datatype info for operation */ + if (H5D__typeinfo_init(&io_info, &(dset_info[i]), dset_info[i].mem_type_id) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info") + type_info_init++; + + /* Make certain that the number of elements in each selection is the same, and cache nelmts in + * dset_info */ + dset_info[i].nelmts = H5S_GET_SELECT_NPOINTS(dset_info[i].mem_space); + if (dset_info[i].nelmts != H5S_GET_SELECT_NPOINTS(dset_info[i].file_space)) + HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, + "src and dest dataspaces have different number of elements selected") + + /* Check for a NULL buffer */ + if (NULL == dset_info[i].buf.vp) { + /* Check for any elements selected (which is invalid) */ + if (dset_info[i].nelmts > 0) + HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "no output buffer") + + /* If the buffer is nil, and 0 element is selected, make a fake buffer. + * This is for some MPI package like ChaMPIon on NCSA's tungsten which + * doesn't support this feature. + */ + dset_info[i].buf.vp = &fake_char; + } /* end if */ - /* H5S_select_shape_same() has been modified to accept topologically identical - * selections with different rank as having the same shape (if the most - * rapidly changing coordinates match up), but the I/O code still has - * difficulties with the notion. - * - * To solve this, we check to see if H5S_select_shape_same() returns true, - * and if the ranks of the mem and file spaces are different. If they are, - * construct a new mem space that is equivalent to the old mem space, and - * use that instead. - * - * Note that in general, this requires us to touch up the memory buffer as - * well. - */ - if (nelmts > 0 && TRUE == H5S_SELECT_SHAPE_SAME(mem_space, file_space) && - H5S_GET_EXTENT_NDIMS(mem_space) != H5S_GET_EXTENT_NDIMS(file_space)) { - ptrdiff_t buf_adj = 0; - - /* Attempt to construct projected dataspace for memory dataspace */ - if (H5S_select_construct_projection(mem_space, &projected_mem_space, - (unsigned)H5S_GET_EXTENT_NDIMS(file_space), - type_info.dst_type_size, &buf_adj) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to construct projected memory dataspace") - HDassert(projected_mem_space); - - /* Adjust the buffer by the given amount */ - buf = (void *)(((uint8_t *)buf) + buf_adj); - - /* Switch to using projected memory dataspace & adjusted buffer */ - mem_space = projected_mem_space; - } /* end if */ + /* Make sure that both selections have their extents set */ + if (!(H5S_has_extent(dset_info[i].file_space))) + HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "file dataspace does not have extent set") + if (!(H5S_has_extent(dset_info[i].mem_space))) + HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "memory dataspace does not have extent set") + + /* H5S_select_shape_same() has been modified to accept topologically identical + * selections with different rank as having the same shape (if the most + * rapidly changing coordinates match up), but the I/O code still has + * difficulties with the notion. + * + * To solve this, we check to see if H5S_select_shape_same() returns true, + * and if the ranks of the mem and file spaces are different. If they are, + * construct a new mem space that is equivalent to the old mem space, and + * use that instead. + * + * Note that in general, this requires us to touch up the memory buffer as + * well. + */ + if (dset_info[i].nelmts > 0 && + TRUE == H5S_SELECT_SHAPE_SAME(dset_info[i].mem_space, dset_info[i].file_space) && + H5S_GET_EXTENT_NDIMS(dset_info[i].mem_space) != H5S_GET_EXTENT_NDIMS(dset_info[i].file_space)) { + ptrdiff_t buf_adj = 0; + + /* Allocate original memory space buffer if necessary */ + if (!orig_mem_space) { + if (count > 1) { + /* Allocate buffer */ + if (NULL == (orig_mem_space = (H5S_t **)H5MM_calloc(count * sizeof(H5S_t *)))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, + "couldn't allocate original memory space array buffer") + } + else + /* Use local buffer */ + orig_mem_space = &orig_mem_space_local; + } - /* Retrieve dataset properties */ - /* <none needed in the general case> */ + /* Save original memory space */ + orig_mem_space[i] = dset_info[i].mem_space; + dset_info[i].mem_space = NULL; - /* If space hasn't been allocated and not using external storage, - * return fill value to buffer if fill time is upon allocation, or - * do nothing if fill time is never. If the dataset is compact and - * fill time is NEVER, there is no way to tell whether part of data - * has been overwritten. So just proceed in reading. - */ - if (nelmts > 0 && dataset->shared->dcpl_cache.efl.nused == 0 && - !(*dataset->shared->layout.ops->is_space_alloc)(&dataset->shared->layout.storage) && - !(dataset->shared->layout.ops->is_data_cached && - (*dataset->shared->layout.ops->is_data_cached)(dataset->shared))) { - H5D_fill_value_t fill_status; /* Whether/How the fill value is defined */ - - /* Retrieve dataset's fill-value properties */ - if (H5P_is_fill_value_defined(&dataset->shared->dcpl_cache.fill, &fill_status) < 0) - HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, FAIL, "can't tell if fill value defined") - - /* Should be impossible, but check anyway... */ - if (fill_status == H5D_FILL_VALUE_UNDEFINED && - (dataset->shared->dcpl_cache.fill.fill_time == H5D_FILL_TIME_ALLOC || - dataset->shared->dcpl_cache.fill.fill_time == H5D_FILL_TIME_IFSET)) - HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, - "read failed: dataset doesn't exist, no data can be read") - - /* If we're never going to fill this dataset, just leave the junk in the user's buffer */ - if (dataset->shared->dcpl_cache.fill.fill_time == H5D_FILL_TIME_NEVER) - HGOTO_DONE(SUCCEED) - - /* Go fill the user's selection with the dataset's fill value */ - if (H5D__fill(dataset->shared->dcpl_cache.fill.buf, dataset->shared->type, buf, type_info.mem_type, - mem_space) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "filling buf failed") - else - HGOTO_DONE(SUCCEED) - } /* end if */ + /* Attempt to construct projected dataspace for memory dataspace */ + if (H5S_select_construct_projection(orig_mem_space[i], &dset_info[i].mem_space, + (unsigned)H5S_GET_EXTENT_NDIMS(dset_info[i].file_space), + (hsize_t)dset_info[i].type_info.dst_type_size, &buf_adj) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to construct projected memory dataspace") + HDassert(dset_info[i].mem_space); - /* Set up I/O operation */ - io_info.op_type = H5D_IO_OP_READ; - io_info.u.rbuf = buf; - if (H5D__ioinfo_init(dataset, &type_info, &store, &io_info) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_UNSUPPORTED, FAIL, "unable to set up I/O operation") - - /* Sanity check that space is allocated, if there are elements */ - if (nelmts > 0) - HDassert((*dataset->shared->layout.ops->is_space_alloc)(&dataset->shared->layout.storage) || - (dataset->shared->layout.ops->is_data_cached && - (*dataset->shared->layout.ops->is_data_cached)(dataset->shared)) || - dataset->shared->dcpl_cache.efl.nused > 0 || layout_type == H5D_COMPACT); - - /* Allocate the chunk map */ - if (H5D_CONTIGUOUS != layout_type && H5D_COMPACT != layout_type) { - if (NULL == (fm = H5FL_CALLOC(H5D_chunk_map_t))) - HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, "can't allocate chunk map") - } + /* Adjust the buffer by the given amount */ + dset_info[i].buf.vp = (void *)(((uint8_t *)dset_info[i].buf.vp) + buf_adj); + } /* end if */ - /* Call storage method's I/O initialization routine */ - if (io_info.layout_ops.io_init && - (*io_info.layout_ops.io_init)(&io_info, &type_info, nelmts, file_space, mem_space, fm) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't initialize I/O info") - io_op_init = TRUE; + /* If space hasn't been allocated and not using external storage, + * return fill value to buffer if fill time is upon allocation, or + * do nothing if fill time is never. If the dataset is compact and + * fill time is NEVER, there is no way to tell whether part of data + * has been overwritten. So just proceed in reading. + */ + if (dset_info[i].nelmts > 0 && dset_info[i].dset->shared->dcpl_cache.efl.nused == 0 && + !(*dset_info[i].dset->shared->layout.ops->is_space_alloc)( + &dset_info[i].dset->shared->layout.storage) && + !(dset_info[i].dset->shared->layout.ops->is_data_cached && + (*dset_info[i].dset->shared->layout.ops->is_data_cached)(dset_info[i].dset->shared))) { + H5D_fill_value_t fill_status; /* Whether/How the fill value is defined */ + + /* Retrieve dataset's fill-value properties */ + if (H5P_is_fill_value_defined(&dset_info[i].dset->shared->dcpl_cache.fill, &fill_status) < 0) + HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, FAIL, "can't tell if fill value defined") + + /* Should be impossible, but check anyway... */ + if (fill_status == H5D_FILL_VALUE_UNDEFINED && + (dset_info[i].dset->shared->dcpl_cache.fill.fill_time == H5D_FILL_TIME_ALLOC || + dset_info[i].dset->shared->dcpl_cache.fill.fill_time == H5D_FILL_TIME_IFSET)) + HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, + "read failed: dataset doesn't exist, no data can be read") + + /* If we're never going to fill this dataset, just leave the junk in the user's buffer */ + if (dset_info[i].dset->shared->dcpl_cache.fill.fill_time != H5D_FILL_TIME_NEVER) + /* Go fill the user's selection with the dataset's fill value */ + if (H5D__fill(dset_info[i].dset->shared->dcpl_cache.fill.buf, dset_info[i].dset->shared->type, + dset_info[i].buf.vp, dset_info[i].type_info.mem_type, + dset_info[i].mem_space) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "filling buf failed") + + /* No need to perform any more I/O for this dataset */ + dset_info[i].skip_io = TRUE; + io_skipped++; + } /* end if */ + else { + /* Set up I/O operation */ + if (H5D__dset_ioinfo_init(dset_info[i].dset, &(dset_info[i]), &(store[i])) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_UNSUPPORTED, FAIL, "unable to set up I/O operation") + + /* Sanity check that space is allocated, if there are elements */ + if (dset_info[i].nelmts > 0) + HDassert( + (*dset_info[i].dset->shared->layout.ops->is_space_alloc)( + &dset_info[i].dset->shared->layout.storage) || + (dset_info[i].dset->shared->layout.ops->is_data_cached && + (*dset_info[i].dset->shared->layout.ops->is_data_cached)(dset_info[i].dset->shared)) || + dset_info[i].dset->shared->dcpl_cache.efl.nused > 0 || + dset_info[i].dset->shared->layout.type == H5D_COMPACT); + + /* Call storage method's I/O initialization routine */ + if (dset_info[i].layout_ops.io_init && + (dset_info[i].layout_ops.io_init)(&io_info, &(dset_info[i])) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't initialize I/O info") + dset_info[i].skip_io = FALSE; + io_op_init++; + + /* Reset metadata tagging */ + H5AC_tag(prev_tag, NULL); + } + } /* end of for loop */ + + HDassert(type_info_init == count); + HDassert(io_op_init + io_skipped == count); + + /* If no datasets have I/O, we're done */ + if (io_op_init == 0) + HGOTO_DONE(SUCCEED) + + /* Perform second phase of type info initialization */ + if (H5D__typeinfo_init_phase2(&io_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info (second phase)") #ifdef H5_HAVE_PARALLEL /* Adjust I/O info for any parallel I/O */ - if (H5D__ioinfo_adjust(&io_info, dataset, file_space, mem_space, &type_info) < 0) + if (H5D__ioinfo_adjust(&io_info) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to adjust I/O info for parallel I/O") #endif /*H5_HAVE_PARALLEL*/ - /* Invoke correct "high level" I/O routine */ - if ((*io_info.io_ops.multi_read)(&io_info, &type_info, nelmts, file_space, mem_space, fm) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "can't read data") + /* If multi dataset I/O callback is not provided, perform read IO via + * single-dset path with looping */ + if (io_info.md_io_ops.multi_read_md) { + /* Create sel_pieces array if any pieces are selected */ + if (io_info.piece_count > 0) { + HDassert(!io_info.sel_pieces); + HDassert(io_info.pieces_added == 0); + + /* Allocate sel_pieces array */ + if (NULL == + (io_info.sel_pieces = H5MM_malloc(io_info.piece_count * sizeof(io_info.sel_pieces[0])))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, "unable to allocate array of selected pieces") + } + + /* MDIO-specific second phase initialization */ + for (i = 0; i < count; i++) + if (dset_info[i].layout_ops.mdio_init) { + haddr_t prev_tag = HADDR_UNDEF; + + /* set metadata tagging with dset oheader addr */ + H5AC_tag(dset_info[i].dset->oloc.addr, &prev_tag); + + /* Make second phase IO init call */ + if ((dset_info[i].layout_ops.mdio_init)(&io_info, &(dset_info[i])) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't populate array of selected pieces") + + /* Reset metadata tagging */ + H5AC_tag(prev_tag, NULL); + } + + /* Invoke correct "high level" I/O routine */ + if ((*io_info.md_io_ops.multi_read_md)(&io_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "can't read data") + } /* end if */ + else { + haddr_t prev_tag = HADDR_UNDEF; + + if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info) && io_info.piece_count > 0) { + if (NULL == (io_info.mem_spaces = H5MM_malloc(io_info.piece_count * sizeof(H5S_t *)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, + "memory allocation failed for memory space list") + if (NULL == (io_info.file_spaces = H5MM_malloc(io_info.piece_count * sizeof(H5S_t *)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "memory allocation failed for file space list") + if (NULL == (io_info.addrs = H5MM_malloc(io_info.piece_count * sizeof(haddr_t)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, + "memory allocation failed for piece address list") + if (NULL == (io_info.element_sizes = H5MM_malloc(io_info.piece_count * sizeof(size_t)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, + "memory allocation failed for element size list") + if (NULL == (io_info.rbufs = H5MM_malloc(io_info.piece_count * sizeof(void *)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, + "memory allocation failed for read buffer list") + } + + /* Loop with serial & single-dset read IO path */ + for (i = 0; i < count; i++) { + /* Check for skipped I/O */ + if (dset_info[i].skip_io) + continue; + + /* set metadata tagging with dset oheader addr */ + H5AC_tag(dset_info[i].dset->oloc.addr, &prev_tag); + + /* Invoke correct "high level" I/O routine */ + if ((*dset_info[i].io_ops.multi_read)(&io_info, &dset_info[i]) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "can't read data") + + /* Reset metadata tagging */ + H5AC_tag(prev_tag, NULL); + } + + /* Make final multi dataset selection I/O call if we are using both + * features - in this case the multi_read callbacks did not perform the + * actual I/O */ + H5_CHECK_OVERFLOW(io_info.pieces_added, size_t, uint32_t) + if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info)) + if (H5F_shared_select_read(io_info.f_sh, H5FD_MEM_DRAW, (uint32_t)io_info.pieces_added, + io_info.mem_spaces, io_info.file_spaces, io_info.addrs, + io_info.element_sizes, io_info.rbufs) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "selection read failed") + } done: /* Shut down the I/O op information */ - if (io_op_init && io_info.layout_ops.io_term && (*io_info.layout_ops.io_term)(fm) < 0) - HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, "unable to shut down I/O op info") - if (fm) - fm = H5FL_FREE(H5D_chunk_map_t, fm); + for (i = 0; i < io_op_init; i++) + if (!dset_info[i].skip_io && dset_info[i].layout_ops.io_term && + (*dset_info[i].layout_ops.io_term)(&io_info, &(dset_info[i])) < 0) + HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, "unable to shut down I/O op info") /* Shut down datatype info for operation */ - if (type_info_init && H5D__typeinfo_term(&type_info) < 0) + if (H5D__typeinfo_term(&io_info, type_info_init) < 0) HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, "unable to shut down type info") - /* discard projected mem space if it was created */ - if (NULL != projected_mem_space) - if (H5S_close(projected_mem_space) < 0) - HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, "unable to shut down projected memory dataspace") + /* Discard projected mem spaces and restore originals */ + if (orig_mem_space) { + for (i = 0; i < count; i++) + if (orig_mem_space[i]) { + if (H5S_close(dset_info[i].mem_space) < 0) + HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, + "unable to shut down projected memory dataspace") + dset_info[i].mem_space = orig_mem_space[i]; + } + + /* Free orig_mem_space array if it was allocated */ + if (orig_mem_space != &orig_mem_space_local) + H5MM_free(orig_mem_space); + } + + /* Free global piece array */ + H5MM_xfree(io_info.sel_pieces); + + /* Free selection I/O arrays */ + H5MM_xfree(io_info.mem_spaces); + H5MM_xfree(io_info.file_spaces); + H5MM_xfree(io_info.addrs); + H5MM_xfree(io_info.element_sizes); + H5MM_xfree(io_info.rbufs); - FUNC_LEAVE_NOAPI_TAG(ret_value) + /* Free store array if it was allocated */ + if (store != &store_local) + H5MM_free(store); + + FUNC_LEAVE_NOAPI(ret_value) } /* end H5D__read() */ /*------------------------------------------------------------------------- * Function: H5D__write * - * Purpose: Writes (part of) a DATASET to a file from application memory - * BUF. See H5Dwrite() for complete details. + * Purpose: Writes multiple (part of) DATASETs to a file from application + * memory BUFs. See H5Dwrite_multi() for complete details. * - * Return: Non-negative on success/Negative on failure + * This was referred from H5D__write for multi-dset work. * - * Programmer: Robb Matzke - * Thursday, December 4, 1997 + * Return: Non-negative on success/Negative on failure * *------------------------------------------------------------------------- */ herr_t -H5D__write(H5D_t *dataset, hid_t mem_type_id, H5S_t *mem_space, H5S_t *file_space, const void *buf) +H5D__write(size_t count, H5D_dset_io_info_t *dset_info) { - H5D_chunk_map_t *fm = NULL; /* Chunk file<->memory mapping */ - H5D_io_info_t io_info; /* Dataset I/O info */ - H5D_type_info_t type_info; /* Datatype info for operation */ - H5D_layout_t layout_type; /* Dataset's layout type (contig, chunked, compact, etc.) */ - hbool_t type_info_init = FALSE; /* Whether the datatype info has been initialized */ - hbool_t should_alloc_space = FALSE; /* Whether or not to initialize dataset's storage */ - H5S_t *projected_mem_space = NULL; /* If not NULL, ptr to dataspace containing a */ - /* projection of the supplied mem_space to a new */ - /* dataspace with rank equal to that of */ - /* file_space. */ - /* */ - /* This field is only used if */ - /* H5S_select_shape_same() returns TRUE when */ - /* comparing the mem_space and the data_space, */ - /* and the mem_space have different rank. */ - /* */ - /* Note that if this variable is used, the */ - /* projected mem space must be discarded at the */ - /* end of the function to avoid a memory leak. */ - H5D_storage_t store; /* union of EFL and chunk pointer in file space */ - hsize_t nelmts; /* total number of elmts */ - hbool_t io_op_init = FALSE; /* Whether the I/O op has been initialized */ - char fake_char; /* Temporary variable for NULL buffer pointers */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE_TAG(dataset->oloc.addr) + H5D_io_info_t io_info; /* Dataset I/O info for multi dsets */ + size_t type_info_init = 0; /* Number of datatype info structs that have been initialized */ + H5S_t *orig_mem_space_local; /* Local buffer for orig_mem_space */ + H5S_t **orig_mem_space = NULL; /* If not NULL, ptr to an array of dataspaces */ + /* containing the original memory spaces contained */ + /* in dset_info. This is needed in order to */ + /* restore the original state of dset_info if we */ + /* replaced any mem spaces with equivalents */ + /* projected to a rank equal to that of file_space. */ + /* */ + /* This field is only used if */ + /* H5S_select_shape_same() returns TRUE when */ + /* comparing at least one mem_space and data_space, */ + /* and the mem_space has a different rank. */ + /* */ + /* Note that this is a temporary variable - the */ + /* projected memory space is stored in dset_info, */ + /* and will be freed when that structure is */ + /* freed. */ + H5D_storage_t store_local; /* Local buffer for store */ + H5D_storage_t *store = &store_local; /* Union of EFL and chunk pointer in file space */ + size_t io_op_init = 0; /* Number I/O ops that have been initialized */ + size_t i; /* Local index variable */ + char fake_char; /* Temporary variable for NULL buffer pointers */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Init io_info */ + if (H5D__ioinfo_init(count, dset_info, &io_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't initialize I/O info") + io_info.op_type = H5D_IO_OP_WRITE; - /* check args */ - HDassert(dataset && dataset->oloc.file); - HDassert(file_space); - HDassert(mem_space); + /* Allocate store buffer if necessary */ + if (count > 1) + if (NULL == (store = (H5D_storage_t *)H5MM_malloc(count * sizeof(H5D_storage_t)))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, "couldn't allocate dset storage info array buffer") - layout_type = dataset->shared->layout.type; + /* iterate over all dsets and construct I/O information */ + for (i = 0; i < count; i++) { + hbool_t should_alloc_space = FALSE; /* Whether or not to initialize dataset's storage */ + haddr_t prev_tag = HADDR_UNDEF; - /* All filters in the DCPL must have encoding enabled. */ - if (!dataset->shared->checked_filters) { - if (H5Z_can_apply(dataset->shared->dcpl_id, dataset->shared->type_id) < 0) - HGOTO_ERROR(H5E_PLINE, H5E_CANAPPLY, FAIL, "can't apply filters") + /* check args */ + if (NULL == dset_info[i].dset) + HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a dataset") + if (NULL == dset_info[i].dset->oloc.file) + HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file") - dataset->shared->checked_filters = TRUE; - } /* end if */ + /* set metadata tagging with dset oheader addr */ + H5AC_tag(dset_info[i].dset->oloc.addr, &prev_tag); + + /* All filters in the DCPL must have encoding enabled. */ + if (!dset_info[i].dset->shared->checked_filters) { + if (H5Z_can_apply(dset_info[i].dset->shared->dcpl_id, dset_info[i].dset->shared->type_id) < 0) + HGOTO_ERROR(H5E_PLINE, H5E_CANAPPLY, FAIL, "can't apply filters") - /* Check if we are allowed to write to this file */ - if (0 == (H5F_INTENT(dataset->oloc.file) & H5F_ACC_RDWR)) - HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "no write intent on file") + dset_info[i].dset->shared->checked_filters = TRUE; + } /* end if */ + + /* Check if we are allowed to write to this file */ + if (0 == (H5F_INTENT(dset_info[i].dset->oloc.file) & H5F_ACC_RDWR)) + HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "no write intent on file") - /* Set up datatype info for operation */ - if (H5D__typeinfo_init(dataset, mem_type_id, TRUE, &type_info) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info") - type_info_init = TRUE; + /* Set up datatype info for operation */ + if (H5D__typeinfo_init(&io_info, &(dset_info[i]), dset_info[i].mem_type_id) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info") + type_info_init++; - /* Various MPI based checks */ + /* Various MPI based checks */ #ifdef H5_HAVE_PARALLEL - if (H5F_HAS_FEATURE(dataset->oloc.file, H5FD_FEAT_HAS_MPI)) { - /* If MPI based VFD is used, no VL or region reference datatype support yet. */ - /* This is because they use the global heap in the file and we don't */ - /* support parallel access of that yet */ - if (H5T_is_vl_storage(type_info.mem_type) > 0) - HGOTO_ERROR(H5E_DATASET, H5E_UNSUPPORTED, FAIL, - "Parallel IO does not support writing VL or region reference datatypes yet") - } /* end if */ - else { - H5FD_mpio_xfer_t io_xfer_mode; /* MPI I/O transfer mode */ + if (H5F_HAS_FEATURE(dset_info[i].dset->oloc.file, H5FD_FEAT_HAS_MPI)) { + /* If MPI based VFD is used, no VL or region reference datatype support yet. */ + /* This is because they use the global heap in the file and we don't */ + /* support parallel access of that yet */ + if (H5T_is_vl_storage(dset_info[i].type_info.mem_type) > 0) + HGOTO_ERROR(H5E_DATASET, H5E_UNSUPPORTED, FAIL, + "Parallel IO does not support writing VL or region reference datatypes yet") + } /* end if */ + else { + H5FD_mpio_xfer_t io_xfer_mode; /* MPI I/O transfer mode */ + + /* Get I/O transfer mode */ + if (H5CX_get_io_xfer_mode(&io_xfer_mode) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get MPI-I/O transfer mode") + + /* Collective access is not permissible without a MPI based VFD */ + if (io_xfer_mode == H5FD_MPIO_COLLECTIVE) + HGOTO_ERROR(H5E_DATASET, H5E_UNSUPPORTED, FAIL, "collective access for MPI-based driver only") + } /* end else */ +#endif /*H5_HAVE_PARALLEL*/ + + /* Make certain that the number of elements in each selection is the same, and cache nelmts in + * dset_info */ + dset_info[i].nelmts = H5S_GET_SELECT_NPOINTS(dset_info[i].mem_space); + if (dset_info[i].nelmts != H5S_GET_SELECT_NPOINTS(dset_info[i].file_space)) + HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, + "src and dest dataspaces have different number of elements selected") + + /* Check for a NULL buffer */ + if (NULL == dset_info[i].buf.cvp) { + /* Check for any elements selected (which is invalid) */ + if (dset_info[i].nelmts > 0) + HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "no output buffer") + + /* If the buffer is nil, and 0 element is selected, make a fake buffer. + * This is for some MPI package like ChaMPIon on NCSA's tungsten which + * doesn't support this feature. + */ + dset_info[i].buf.cvp = &fake_char; + } /* end if */ - /* Get I/O transfer mode */ - if (H5CX_get_io_xfer_mode(&io_xfer_mode) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get MPI-I/O transfer mode") + /* Make sure that both selections have their extents set */ + if (!(H5S_has_extent(dset_info[i].file_space))) + HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "file dataspace does not have extent set") + if (!(H5S_has_extent(dset_info[i].mem_space))) + HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "memory dataspace does not have extent set") + + /* H5S_select_shape_same() has been modified to accept topologically + * identical selections with different rank as having the same shape + * (if the most rapidly changing coordinates match up), but the I/O + * code still has difficulties with the notion. + * + * To solve this, we check to see if H5S_select_shape_same() returns + * true, and if the ranks of the mem and file spaces are different. + * If they are, construct a new mem space that is equivalent to the + * old mem space, and use that instead. + * + * Note that in general, this requires us to touch up the memory buffer + * as well. + */ + if (dset_info[i].nelmts > 0 && + TRUE == H5S_SELECT_SHAPE_SAME(dset_info[i].mem_space, dset_info[i].file_space) && + H5S_GET_EXTENT_NDIMS(dset_info[i].mem_space) != H5S_GET_EXTENT_NDIMS(dset_info[i].file_space)) { + ptrdiff_t buf_adj = 0; + + /* Allocate original memory space buffer if necessary */ + if (!orig_mem_space) { + if (count > 1) { + /* Allocate buffer */ + if (NULL == (orig_mem_space = (H5S_t **)H5MM_calloc(count * sizeof(H5S_t *)))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, + "couldn't allocate original memory space array buffer") + } + else + /* Use local buffer */ + orig_mem_space = &orig_mem_space_local; + } - /* Collective access is not permissible without a MPI based VFD */ - if (io_xfer_mode == H5FD_MPIO_COLLECTIVE) - HGOTO_ERROR(H5E_DATASET, H5E_UNSUPPORTED, FAIL, "collective access for MPI-based driver only") - } /* end else */ -#endif /*H5_HAVE_PARALLEL*/ + /* Save original memory space */ + orig_mem_space[i] = dset_info[i].mem_space; + dset_info[i].mem_space = NULL; + + /* Attempt to construct projected dataspace for memory dataspace */ + if (H5S_select_construct_projection(orig_mem_space[i], &dset_info[i].mem_space, + (unsigned)H5S_GET_EXTENT_NDIMS(dset_info[i].file_space), + dset_info[i].type_info.src_type_size, &buf_adj) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to construct projected memory dataspace") + HDassert(dset_info[i].mem_space); + + /* Adjust the buffer by the given amount */ + dset_info[i].buf.cvp = (const void *)(((const uint8_t *)dset_info[i].buf.cvp) + buf_adj); + } /* end if */ - /* Make certain that the number of elements in each selection is the same */ - nelmts = H5S_GET_SELECT_NPOINTS(mem_space); - if (nelmts != H5S_GET_SELECT_NPOINTS(file_space)) - HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, - "src and dest dataspaces have different number of elements selected") - - /* Check for a NULL buffer */ - if (NULL == buf) { - /* Check for any elements selected (which is invalid) */ - if (nelmts > 0) - HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "no output buffer") - - /* If the buffer is nil, and 0 element is selected, make a fake buffer. - * This is for some MPI package like ChaMPIon on NCSA's tungsten which - * doesn't support this feature. + /* Retrieve dataset properties */ + /* <none needed currently> */ + + /* Set up I/O operation */ + if (H5D__dset_ioinfo_init(dset_info[i].dset, &(dset_info[i]), &(store[i])) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up I/O operation") + + /* Allocate dataspace and initialize it if it hasn't been. */ + should_alloc_space = dset_info[i].dset->shared->dcpl_cache.efl.nused == 0 && + !(*dset_info[i].dset->shared->layout.ops->is_space_alloc)( + &dset_info[i].dset->shared->layout.storage); + + /* + * If not using an MPI-based VFD, we only need to allocate + * and initialize storage if there's a selection in the + * dataset's dataspace. Otherwise, we always need to participate + * in the storage allocation since this may use collective + * operations and we will hang if we don't participate. */ - buf = &fake_char; - } /* end if */ + if (!H5F_HAS_FEATURE(dset_info[i].dset->oloc.file, H5FD_FEAT_HAS_MPI)) + should_alloc_space = should_alloc_space && (dset_info[i].nelmts > 0); - /* Make sure that both selections have their extents set */ - if (!(H5S_has_extent(file_space))) - HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "file dataspace does not have extent set") - if (!(H5S_has_extent(mem_space))) - HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "memory dataspace does not have extent set") + if (should_alloc_space) { + hssize_t file_nelmts; /* Number of elements in file dataset's dataspace */ + hbool_t full_overwrite; /* Whether we are over-writing all the elements */ - /* H5S_select_shape_same() has been modified to accept topologically - * identical selections with different rank as having the same shape - * (if the most rapidly changing coordinates match up), but the I/O - * code still has difficulties with the notion. - * - * To solve this, we check to see if H5S_select_shape_same() returns - * true, and if the ranks of the mem and file spaces are different. - * If the are, construct a new mem space that is equivalent to the - * old mem space, and use that instead. - * - * Note that in general, this requires us to touch up the memory buffer - * as well. - */ - if (nelmts > 0 && TRUE == H5S_SELECT_SHAPE_SAME(mem_space, file_space) && - H5S_GET_EXTENT_NDIMS(mem_space) != H5S_GET_EXTENT_NDIMS(file_space)) { - ptrdiff_t buf_adj = 0; - - /* Attempt to construct projected dataspace for memory dataspace */ - if (H5S_select_construct_projection(mem_space, &projected_mem_space, - (unsigned)H5S_GET_EXTENT_NDIMS(file_space), - type_info.src_type_size, &buf_adj) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to construct projected memory dataspace") - HDassert(projected_mem_space); - - /* Adjust the buffer by the given amount */ - buf = (const void *)(((const uint8_t *)buf) + buf_adj); - - /* Switch to using projected memory dataspace & adjusted buffer */ - mem_space = projected_mem_space; - } /* end if */ + /* Get the number of elements in file dataset's dataspace */ + if ((file_nelmts = H5S_GET_EXTENT_NPOINTS(dset_info[i].file_space)) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_BADVALUE, FAIL, + "can't retrieve number of elements in file dataset") - /* Retrieve dataset properties */ - /* <none needed currently> */ + /* Always allow fill values to be written if the dataset has a VL datatype */ + if (H5T_detect_class(dset_info[i].dset->shared->type, H5T_VLEN, FALSE)) + full_overwrite = FALSE; + else + full_overwrite = (hbool_t)((hsize_t)file_nelmts == dset_info[i].nelmts ? TRUE : FALSE); - /* Set up I/O operation */ - io_info.op_type = H5D_IO_OP_WRITE; - io_info.u.wbuf = buf; - if (H5D__ioinfo_init(dataset, &type_info, &store, &io_info) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up I/O operation") + /* Allocate storage */ + if (H5D__alloc_storage(dset_info[i].dset, H5D_ALLOC_WRITE, full_overwrite, NULL) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to initialize storage") + } /* end if */ - /* Allocate dataspace and initialize it if it hasn't been. */ - should_alloc_space = dataset->shared->dcpl_cache.efl.nused == 0 && - !(*dataset->shared->layout.ops->is_space_alloc)(&dataset->shared->layout.storage); + /* Call storage method's I/O initialization routine */ + /* Init io_info.dset_info[] and generate piece_info in skip list */ + if (dset_info[i].layout_ops.io_init && + (*dset_info[i].layout_ops.io_init)(&io_info, &(dset_info[i])) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't initialize I/O info") + dset_info[i].skip_io = FALSE; + io_op_init++; - /* - * If not using an MPI-based VFD, we only need to allocate - * and initialize storage if there's a selection in the - * dataset's dataspace. Otherwise, we always need to participate - * in the storage allocation since this may use collective - * operations and we will hang if we don't participate. - */ - if (!H5F_HAS_FEATURE(dataset->oloc.file, H5FD_FEAT_HAS_MPI)) - should_alloc_space = should_alloc_space && (nelmts > 0); - - if (should_alloc_space) { - hssize_t file_nelmts; /* Number of elements in file dataset's dataspace */ - hbool_t full_overwrite; /* Whether we are over-writing all the elements */ - - /* Get the number of elements in file dataset's dataspace */ - if ((file_nelmts = H5S_GET_EXTENT_NPOINTS(file_space)) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_BADVALUE, FAIL, "can't retrieve number of elements in file dataset") - - /* Always allow fill values to be written if the dataset has a VL datatype */ - if (H5T_detect_class(dataset->shared->type, H5T_VLEN, FALSE)) - full_overwrite = FALSE; - else - full_overwrite = (hbool_t)((hsize_t)file_nelmts == nelmts ? TRUE : FALSE); - - /* Allocate storage */ - if (H5D__alloc_storage(&io_info, H5D_ALLOC_WRITE, full_overwrite, NULL) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to initialize storage") - } /* end if */ + /* Reset metadata tagging */ + H5AC_tag(prev_tag, NULL); + } /* end of for loop */ - /* Allocate the chunk map */ - if (H5D_CONTIGUOUS != layout_type && H5D_COMPACT != layout_type) { - if (NULL == (fm = H5FL_CALLOC(H5D_chunk_map_t))) - HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, "can't allocate chunk map") - } + HDassert(type_info_init == count); + HDassert(io_op_init == count); - /* Call storage method's I/O initialization routine */ - if (io_info.layout_ops.io_init && - (*io_info.layout_ops.io_init)(&io_info, &type_info, nelmts, file_space, mem_space, fm) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't initialize I/O info") - io_op_init = TRUE; + /* Perform second phase of type info initialization */ + if (H5D__typeinfo_init_phase2(&io_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to set up type info (second phase)") #ifdef H5_HAVE_PARALLEL /* Adjust I/O info for any parallel I/O */ - if (H5D__ioinfo_adjust(&io_info, dataset, file_space, mem_space, &type_info) < 0) + if (H5D__ioinfo_adjust(&io_info) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "unable to adjust I/O info for parallel I/O") #endif /*H5_HAVE_PARALLEL*/ - /* Invoke correct "high level" I/O routine */ - if ((*io_info.io_ops.multi_write)(&io_info, &type_info, nelmts, file_space, mem_space, fm) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "can't write data") + /* If multi dataset I/O callback is not provided, perform write IO via + * single-dset path with looping */ + if (io_info.md_io_ops.multi_write_md) { + /* Create sel_pieces array if any pieces are selected */ + if (io_info.piece_count > 0) { + HDassert(!io_info.sel_pieces); + HDassert(io_info.pieces_added == 0); + + /* Allocate sel_pieces array */ + if (NULL == + (io_info.sel_pieces = H5MM_malloc(io_info.piece_count * sizeof(io_info.sel_pieces[0])))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTALLOC, FAIL, "unable to allocate array of selected pieces") + } + + /* MDIO-specific second phase initialization */ + for (i = 0; i < count; i++) + if (dset_info[i].layout_ops.mdio_init) { + haddr_t prev_tag = HADDR_UNDEF; + + /* set metadata tagging with dset oheader addr */ + H5AC_tag(dset_info[i].dset->oloc.addr, &prev_tag); + + /* Make second phase IO init call */ + if ((dset_info[i].layout_ops.mdio_init)(&io_info, &(dset_info[i])) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't populate array of selected pieces") + + /* Reset metadata tagging */ + H5AC_tag(prev_tag, NULL); + } + + /* Invoke correct "high level" I/O routine */ + if ((*io_info.md_io_ops.multi_write_md)(&io_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "can't write data") + } /* end if */ + else { + haddr_t prev_tag = HADDR_UNDEF; + + if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info) && io_info.piece_count > 0) { + if (NULL == (io_info.mem_spaces = H5MM_malloc(io_info.piece_count * sizeof(H5S_t *)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, + "memory allocation failed for memory space list") + if (NULL == (io_info.file_spaces = H5MM_malloc(io_info.piece_count * sizeof(H5S_t *)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "memory allocation failed for file space list") + if (NULL == (io_info.addrs = H5MM_malloc(io_info.piece_count * sizeof(haddr_t)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, + "memory allocation failed for piece address list") + if (NULL == (io_info.element_sizes = H5MM_malloc(io_info.piece_count * sizeof(size_t)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, + "memory allocation failed for element size list") + if (NULL == (io_info.wbufs = H5MM_malloc(io_info.piece_count * sizeof(const void *)))) + HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, + "memory allocation failed for read buffer list") + } + + /* loop with serial & single-dset write IO path */ + for (i = 0; i < count; i++) { + HDassert(!dset_info[i].skip_io); + + /* set metadata tagging with dset oheader addr */ + H5AC_tag(dset_info->dset->oloc.addr, &prev_tag); + + /* Invoke correct "high level" I/O routine */ + if ((*dset_info[i].io_ops.multi_write)(&io_info, &dset_info[i]) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "can't write data") + + /* Reset metadata tagging */ + H5AC_tag(prev_tag, NULL); + } + + /* Make final multi dataset selection I/O call if we are using both + * features - in this case the multi_write callbacks did not perform the + * actual I/O */ + H5_CHECK_OVERFLOW(io_info.pieces_added, size_t, uint32_t) + if (!H5D_LAYOUT_CB_PERFORM_IO(&io_info)) + if (H5F_shared_select_write(io_info.f_sh, H5FD_MEM_DRAW, (uint32_t)io_info.pieces_added, + io_info.mem_spaces, io_info.file_spaces, io_info.addrs, + io_info.element_sizes, io_info.wbufs) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "selection write failed") + } #ifdef OLD_WAY /* @@ -515,88 +774,138 @@ H5D__write(H5D_t *dataset, hid_t mem_type_id, H5S_t *mem_space, H5S_t *file_spac done: /* Shut down the I/O op information */ - if (io_op_init && io_info.layout_ops.io_term && (*io_info.layout_ops.io_term)(fm) < 0) - HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, "unable to shut down I/O op info") - if (fm) - fm = H5FL_FREE(H5D_chunk_map_t, fm); + for (i = 0; i < io_op_init; i++) { + HDassert(!dset_info[i].skip_io); + if (dset_info[i].layout_ops.io_term && + (*dset_info[i].layout_ops.io_term)(&io_info, &(dset_info[i])) < 0) + HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, "unable to shut down I/O op info") + } /* Shut down datatype info for operation */ - if (type_info_init && H5D__typeinfo_term(&type_info) < 0) + if (H5D__typeinfo_term(&io_info, type_info_init) < 0) HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, "unable to shut down type info") - /* discard projected mem space if it was created */ - if (NULL != projected_mem_space) - if (H5S_close(projected_mem_space) < 0) - HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, "unable to shut down projected memory dataspace") + /* Discard projected mem spaces and restore originals */ + if (orig_mem_space) { + for (i = 0; i < count; i++) + if (orig_mem_space[i]) { + if (H5S_close(dset_info[i].mem_space) < 0) + HDONE_ERROR(H5E_DATASET, H5E_CANTCLOSEOBJ, FAIL, + "unable to shut down projected memory dataspace") + dset_info[i].mem_space = orig_mem_space[i]; + } + + /* Free orig_mem_space array if it was allocated */ + if (orig_mem_space != &orig_mem_space_local) + H5MM_free(orig_mem_space); + } + + /* Free global piece array */ + H5MM_xfree(io_info.sel_pieces); - FUNC_LEAVE_NOAPI_TAG(ret_value) -} /* end H5D__write() */ + /* Free store array if it was allocated */ + if (store != &store_local) + H5MM_free(store); + + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5D__write */ /*------------------------------------------------------------------------- * Function: H5D__ioinfo_init * - * Purpose: Routine for determining correct I/O operations for - * each I/O action. + * Purpose: General setup for H5D_io_info_t struct * * Return: Non-negative on success/Negative on failure * - * Programmer: Quincey Koziol - * Thursday, September 30, 2004 + *------------------------------------------------------------------------- + */ +static herr_t +H5D__ioinfo_init(size_t count, H5D_dset_io_info_t *dset_info, H5D_io_info_t *io_info) +{ + FUNC_ENTER_PACKAGE_NOERR + + /* check args */ + HDassert(count > 0); + HDassert(dset_info); + HDassert(dset_info[0].dset->oloc.file); + HDassert(io_info); + + /* Zero out struct */ + HDmemset(io_info, 0, sizeof(*io_info)); + + /* Set up simple fields */ + io_info->f_sh = count > 0 ? H5F_SHARED(dset_info[0].dset->oloc.file) : NULL; + io_info->count = count; + + /* Start without multi-dataset I/O ops. If we're not using the collective + * I/O path then we will call the single dataset callbacks in a loop. */ + + /* Use provided dset_info */ + io_info->dsets_info = dset_info; + + /* Start with selection I/O on if the global is on, layout callback will + * turn it off if appropriate */ + io_info->use_select_io = H5_use_selection_io_g; + +#ifdef H5_HAVE_PARALLEL + /* Determine if the file was opened with an MPI VFD */ + if (count > 0) + io_info->using_mpi_vfd = H5F_HAS_FEATURE(dset_info[0].dset->oloc.file, H5FD_FEAT_HAS_MPI); +#endif /* H5_HAVE_PARALLEL */ + + FUNC_LEAVE_NOAPI(SUCCEED) +} /* end H5D__ioinfo_init() */ + +/*------------------------------------------------------------------------- + * Function: H5D__dset_ioinfo_init + * + * Purpose: Routine for determining correct I/O operations for each I/O action. + * + * Return: Non-negative on success/Negative on failure * *------------------------------------------------------------------------- */ static herr_t -H5D__ioinfo_init(H5D_t *dset, const H5D_type_info_t *type_info, H5D_storage_t *store, H5D_io_info_t *io_info) +H5D__dset_ioinfo_init(H5D_t *dset, H5D_dset_io_info_t *dset_info, H5D_storage_t *store) { FUNC_ENTER_PACKAGE_NOERR /* check args */ HDassert(dset); HDassert(dset->oloc.file); - HDassert(type_info); - HDassert(type_info->tpath); - HDassert(io_info); + HDassert(dset_info->type_info.tpath); /* Set up "normal" I/O fields */ - io_info->dset = dset; - io_info->f_sh = H5F_SHARED(dset->oloc.file); - io_info->store = store; + dset_info->dset = dset; + dset_info->store = store; /* Set I/O operations to initial values */ - io_info->layout_ops = *dset->shared->layout.ops; + dset_info->layout_ops = *dset->shared->layout.ops; /* Set the "high-level" I/O operations for the dataset */ - io_info->io_ops.multi_read = dset->shared->layout.ops->ser_read; - io_info->io_ops.multi_write = dset->shared->layout.ops->ser_write; + dset_info->io_ops.multi_read = dset->shared->layout.ops->ser_read; + dset_info->io_ops.multi_write = dset->shared->layout.ops->ser_write; /* Set the I/O operations for reading/writing single blocks on disk */ - if (type_info->is_xform_noop && type_info->is_conv_noop) { + if (dset_info->type_info.is_xform_noop && dset_info->type_info.is_conv_noop) { /* - * If there is no data transform or type conversion then read directly into - * the application's buffer. This saves at least one mem-to-mem copy. + * If there is no data transform or type conversion then read directly + * into the application's buffer. + * This saves at least one mem-to-mem copy. */ - io_info->io_ops.single_read = H5D__select_read; - io_info->io_ops.single_write = H5D__select_write; + dset_info->io_ops.single_read = H5D__select_read; + dset_info->io_ops.single_write = H5D__select_write; } /* end if */ else { /* * This is the general case (type conversion, usually). */ - io_info->io_ops.single_read = H5D__scatgath_read; - io_info->io_ops.single_write = H5D__scatgath_write; + dset_info->io_ops.single_read = H5D__scatgath_read; + dset_info->io_ops.single_write = H5D__scatgath_write; } /* end else */ - /* Start with selection I/O off, layout callback will turn it on if - * appropriate */ - io_info->use_select_io = FALSE; - -#ifdef H5_HAVE_PARALLEL - /* Determine if the file was opened with an MPI VFD */ - io_info->using_mpi_vfd = H5F_HAS_FEATURE(dset->oloc.file, H5FD_FEAT_HAS_MPI); -#endif /* H5_HAVE_PARALLEL */ - FUNC_LEAVE_NOAPI(SUCCEED) -} /* end H5D__ioinfo_init() */ +} /* end H5D__dset_ioinfo_init() */ /*------------------------------------------------------------------------- * Function: H5D__typeinfo_init @@ -612,8 +921,10 @@ H5D__ioinfo_init(H5D_t *dset, const H5D_type_info_t *type_info, H5D_storage_t *s *------------------------------------------------------------------------- */ static herr_t -H5D__typeinfo_init(const H5D_t *dset, hid_t mem_type_id, hbool_t do_write, H5D_type_info_t *type_info) +H5D__typeinfo_init(H5D_io_info_t *io_info, H5D_dset_io_info_t *dset_info, hid_t mem_type_id) { + H5D_type_info_t *type_info; + const H5D_t *dset; const H5T_t *src_type; /* Source datatype */ const H5T_t *dst_type; /* Destination datatype */ H5Z_data_xform_t *data_transform; /* Data transform info */ @@ -622,7 +933,12 @@ H5D__typeinfo_init(const H5D_t *dset, hid_t mem_type_id, hbool_t do_write, H5D_t FUNC_ENTER_PACKAGE /* check args */ - HDassert(type_info); + HDassert(io_info); + HDassert(dset_info); + + /* Set convenience pointers */ + type_info = &dset_info->type_info; + dset = dset_info->dset; HDassert(dset); /* Patch the top level file pointer for dt->shared->u.vlen.f if needed */ @@ -637,7 +953,7 @@ H5D__typeinfo_init(const H5D_t *dset, hid_t mem_type_id, hbool_t do_write, H5D_t HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a datatype") type_info->dset_type = dset->shared->type; - if (do_write) { + if (io_info->op_type == H5D_IO_OP_WRITE) { src_type = type_info->mem_type; dst_type = dset->shared->type; type_info->src_type_id = mem_type_id; @@ -667,7 +983,6 @@ H5D__typeinfo_init(const H5D_t *dset, hid_t mem_type_id, hbool_t do_write, H5D_t /* Precompute some useful information */ type_info->src_type_size = H5T_get_size(src_type); type_info->dst_type_size = H5T_get_size(dst_type); - type_info->max_type_size = MAX(type_info->src_type_size, type_info->dst_type_size); type_info->is_conv_noop = H5T_path_noop(type_info->tpath); type_info->is_xform_noop = H5Z_xform_noop(data_transform); if (type_info->is_xform_noop && type_info->is_conv_noop) { @@ -675,27 +990,21 @@ H5D__typeinfo_init(const H5D_t *dset, hid_t mem_type_id, hbool_t do_write, H5D_t type_info->need_bkg = H5T_BKG_NO; } /* end if */ else { - void *tconv_buf; /* Temporary conversion buffer pointer */ - void *bkgr_buf; /* Background conversion buffer pointer */ - size_t max_temp_buf; /* Maximum temporary buffer size */ H5T_bkg_t bkgr_buf_type; /* Background buffer type */ - size_t target_size; /* Desired buffer size */ /* Get info from API context */ - if (H5CX_get_max_temp_buf(&max_temp_buf) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve max. temp. buf size") - if (H5CX_get_tconv_buf(&tconv_buf) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve temp. conversion buffer pointer") - if (H5CX_get_bkgr_buf(&bkgr_buf) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve background conversion buffer pointer") if (H5CX_get_bkgr_buf_type(&bkgr_buf_type) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve background buffer type") /* Check if the datatypes are compound subsets of one another */ type_info->cmpd_subset = H5T_path_compound_subset(type_info->tpath); + /* Update io_info->max_type_size */ + io_info->max_type_size = + MAX3(io_info->max_type_size, type_info->src_type_size, type_info->dst_type_size); + /* Check if we need a background buffer */ - if (do_write && H5T_detect_class(dset->shared->type, H5T_VLEN, FALSE)) + if ((io_info->op_type == H5D_IO_OP_WRITE) && H5T_detect_class(dset->shared->type, H5T_VLEN, FALSE)) type_info->need_bkg = H5T_BKG_YES; else { H5T_bkg_t path_bkg; /* Type conversion's background info */ @@ -708,13 +1017,54 @@ H5D__typeinfo_init(const H5D_t *dset, hid_t mem_type_id, hbool_t do_write, H5D_t else type_info->need_bkg = H5T_BKG_NO; /*never needed even if app says yes*/ } /* end else */ + } /* end else */ - /* Set up datatype conversion/background buffers */ +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5D__typeinfo_init() */ + +/*------------------------------------------------------------------------- + * Function: H5D__typeinfo_init_phase2 + * + * Purpose: Finish initializing type info for all datasets after + * calculating the max type size across all datasets. + * + * Return: Non-negative on success/Negative on failure + * + *------------------------------------------------------------------------- + */ +static herr_t +H5D__typeinfo_init_phase2(H5D_io_info_t *io_info) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* check args */ + HDassert(io_info); + + /* Check if we need to allocate a shared type conversion buffer */ + if (io_info->max_type_size) { + void *tconv_buf; /* Temporary conversion buffer pointer */ + void *bkgr_buf; /* Background conversion buffer pointer */ + size_t max_temp_buf; /* Maximum temporary buffer size */ + size_t target_size; /* Desired buffer size */ + size_t i; /* Local index variable */ + + /* Get info from API context */ + if (H5CX_get_max_temp_buf(&max_temp_buf) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve max. temp. buf size") + if (H5CX_get_tconv_buf(&tconv_buf) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve temp. conversion buffer pointer") + if (H5CX_get_bkgr_buf(&bkgr_buf) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't retrieve background conversion buffer pointer") + /* Set up datatype conversion/background buffers */ target_size = max_temp_buf; - /* If the buffer is too small to hold even one element, try to make it bigger */ - if (target_size < type_info->max_type_size) { + /* If the buffer is too small to hold even one element (in the dataset with the largest , try to make + * it bigger */ + if (target_size < io_info->max_type_size) { hbool_t default_buffer_info; /* Whether the buffer information are the defaults */ /* Detect if we have all default settings for buffers */ @@ -724,50 +1074,64 @@ H5D__typeinfo_init(const H5D_t *dset, hid_t mem_type_id, hbool_t do_write, H5D_t /* Check if we are using the default buffer info */ if (default_buffer_info) /* OK to get bigger for library default settings */ - target_size = type_info->max_type_size; + target_size = io_info->max_type_size; else /* Don't get bigger than the application has requested */ HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "temporary buffer max size is too small") } /* end if */ - /* Compute the number of elements that will fit into buffer */ - type_info->request_nelmts = target_size / type_info->max_type_size; - - /* Sanity check elements in temporary buffer */ - if (type_info->request_nelmts == 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "temporary buffer max size is too small") - /* Get a temporary buffer for type conversion unless the app has already * supplied one through the xfer properties. Instead of allocating a - * buffer which is the exact size, we allocate the target size. - */ - if (NULL == (type_info->tconv_buf = (uint8_t *)tconv_buf)) { + * buffer which is the exact size, we allocate the target size. This + * buffer is shared among all datasets in the operation, unlike for the + * background buffer, where each dataset gets its own. */ + if (NULL == (io_info->tconv_buf = (uint8_t *)tconv_buf)) { /* Allocate temporary buffer */ - if (NULL == (type_info->tconv_buf = H5FL_BLK_CALLOC(type_conv, target_size))) + if (NULL == (io_info->tconv_buf = H5FL_BLK_MALLOC(type_conv, target_size))) HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for type conversion") - type_info->tconv_buf_allocated = TRUE; - } /* end if */ - if (type_info->need_bkg && NULL == (type_info->bkg_buf = (uint8_t *)bkgr_buf)) { - size_t bkg_size; /* Desired background buffer size */ - - /* Compute the background buffer size */ - /* (don't try to use buffers smaller than the default size) */ - bkg_size = type_info->request_nelmts * type_info->dst_type_size; - if (bkg_size < max_temp_buf) - bkg_size = max_temp_buf; - - /* Allocate background buffer */ - /* (Need calloc()-like call since memory needs to be initialized) */ - if (NULL == (type_info->bkg_buf = H5FL_BLK_CALLOC(type_conv, bkg_size))) - HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, - "memory allocation failed for background conversion") - type_info->bkg_buf_allocated = TRUE; + io_info->tconv_buf_allocated = TRUE; } /* end if */ - } /* end else */ + + /* Don't use API provided background buffer if there's more than one dataset, since each + * dataset needs its own */ + if (io_info->count > 1) + bkgr_buf = NULL; + + /* Iterate over datasets */ + for (i = 0; i < io_info->count; i++) { + H5D_type_info_t *type_info = &io_info->dsets_info[i].type_info; + + /* Compute the number of elements that will fit into buffer */ + type_info->request_nelmts = target_size / MAX(type_info->src_type_size, type_info->dst_type_size); + ; + + /* Sanity check elements in temporary buffer */ + if (type_info->request_nelmts == 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "temporary buffer max size is too small") + + /* Allocate background buffer if necessary */ + if (type_info->need_bkg && NULL == (type_info->bkg_buf = (uint8_t *)bkgr_buf)) { + size_t bkg_size; /* Desired background buffer size */ + + /* Compute the background buffer size */ + /* (don't try to use buffers smaller than the default size) */ + bkg_size = type_info->request_nelmts * type_info->dst_type_size; + if (bkg_size < max_temp_buf) + bkg_size = max_temp_buf; + + /* Allocate background buffer */ + /* (Need calloc()-like call since memory needs to be initialized) */ + if (NULL == (type_info->bkg_buf = H5FL_BLK_CALLOC(type_conv, bkg_size))) + HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, + "memory allocation failed for background conversion") + type_info->bkg_buf_allocated = TRUE; + } /* end if */ + } + } done: FUNC_LEAVE_NOAPI(ret_value) -} /* end H5D__typeinfo_init() */ +} /* end H5D__typeinfo_init_phase2() */ #ifdef H5_HAVE_PARALLEL @@ -776,30 +1140,28 @@ done: * * Purpose: Adjust operation's I/O info for any parallel I/O * - * Return: Non-negative on success/Negative on failure + * This was derived from H5D__ioinfo_adjust for multi-dset work. * - * Programmer: Quincey Koziol - * Thursday, March 27, 2008 + * Return: Non-negative on success/Negative on failure * *------------------------------------------------------------------------- */ static herr_t -H5D__ioinfo_adjust(H5D_io_info_t *io_info, const H5D_t *dset, const H5S_t *file_space, const H5S_t *mem_space, - const H5D_type_info_t *type_info) +H5D__ioinfo_adjust(H5D_io_info_t *io_info) { + H5D_t *dset0; /* only the first dset , also for single dsets case */ herr_t ret_value = SUCCEED; /* Return value */ FUNC_ENTER_PACKAGE /* check args */ - HDassert(dset); - HDassert(dset->oloc.file); - HDassert(mem_space); - HDassert(file_space); - HDassert(type_info); - HDassert(type_info->tpath); HDassert(io_info); + /* check the first dset, should exist either single or multi dset cases */ + HDassert(io_info->dsets_info[0].dset); + dset0 = io_info->dsets_info[0].dset; + HDassert(dset0->oloc.file); + /* Reset the actual io mode properties to the default values in case * the DXPL (if it's non-default) was previously used in a collective * I/O operation. @@ -819,11 +1181,11 @@ H5D__ioinfo_adjust(H5D_io_info_t *io_info, const H5D_t *dset, const H5S_t *file_ HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get MPI-I/O transfer mode") /* Get MPI communicator */ - if (MPI_COMM_NULL == (io_info->comm = H5F_mpi_get_comm(dset->oloc.file))) + if (MPI_COMM_NULL == (io_info->comm = H5F_mpi_get_comm(dset0->oloc.file))) HGOTO_ERROR(H5E_DATASPACE, H5E_CANTGET, FAIL, "can't retrieve MPI communicator") /* Check if we can set direct MPI-IO read/write functions */ - if ((opt = H5D__mpio_opt_possible(io_info, file_space, mem_space, type_info)) < 0) + if ((opt = H5D__mpio_opt_possible(io_info)) < 0) HGOTO_ERROR(H5E_DATASPACE, H5E_BADRANGE, FAIL, "invalid check for direct IO dataspace ") /* Check if we can use the optimized parallel I/O routines */ @@ -833,10 +1195,10 @@ H5D__ioinfo_adjust(H5D_io_info_t *io_info, const H5D_t *dset, const H5S_t *file_ * handle collective I/O */ /* Check for selection/vector support in file driver? -NAF */ if (!io_info->use_select_io) { - io_info->io_ops.multi_read = dset->shared->layout.ops->par_read; - io_info->io_ops.multi_write = dset->shared->layout.ops->par_write; - io_info->io_ops.single_read = H5D__mpio_select_read; - io_info->io_ops.single_write = H5D__mpio_select_write; + io_info->md_io_ops.multi_read_md = H5D__collective_read; + io_info->md_io_ops.multi_write_md = H5D__collective_write; + io_info->md_io_ops.single_read_md = H5D__mpio_select_read; + io_info->md_io_ops.single_write_md = H5D__mpio_select_write; } /* end if */ } /* end if */ else { @@ -857,28 +1219,38 @@ H5D__ioinfo_adjust(H5D_io_info_t *io_info, const H5D_t *dset, const H5S_t *file_ * with multiple ranks involved; otherwise, there will be metadata * inconsistencies in the file. */ - if (io_info->op_type == H5D_IO_OP_WRITE && io_info->dset->shared->dcpl_cache.pline.nused > 0) { - int comm_size = 0; - - /* Retrieve size of MPI communicator used for file */ - if ((comm_size = H5F_shared_mpi_get_size(io_info->f_sh)) < 0) - HGOTO_ERROR(H5E_FILE, H5E_CANTGET, FAIL, "can't get MPI communicator size") - - if (comm_size > 1) { - char local_no_coll_cause_string[512]; - char global_no_coll_cause_string[512]; - - if (H5D__mpio_get_no_coll_cause_strings(local_no_coll_cause_string, 512, - global_no_coll_cause_string, 512) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, - "can't get reasons for breaking collective I/O") - - HGOTO_ERROR(H5E_IO, H5E_NO_INDEPENDENT, FAIL, - "Can't perform independent write with filters in pipeline.\n" - " The following caused a break from collective I/O:\n" - " Local causes: %s\n" - " Global causes: %s", - local_no_coll_cause_string, global_no_coll_cause_string); + if (io_info->op_type == H5D_IO_OP_WRITE) { + size_t i; + + /* Check all datasets for filters */ + for (i = 0; i < io_info->count; i++) + if (io_info->dsets_info[i].dset->shared->dcpl_cache.pline.nused > 0) + break; + + /* If the above loop didn't complete at least one dataset has a filter */ + if (i < io_info->count) { + int comm_size = 0; + + /* Retrieve size of MPI communicator used for file */ + if ((comm_size = H5F_shared_mpi_get_size(io_info->f_sh)) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTGET, FAIL, "can't get MPI communicator size") + + if (comm_size > 1) { + char local_no_coll_cause_string[512]; + char global_no_coll_cause_string[512]; + + if (H5D__mpio_get_no_coll_cause_strings(local_no_coll_cause_string, 512, + global_no_coll_cause_string, 512) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, + "can't get reasons for breaking collective I/O") + + HGOTO_ERROR(H5E_IO, H5E_NO_INDEPENDENT, FAIL, + "Can't perform independent write with filters in pipeline.\n" + " The following caused a break from collective I/O:\n" + " Local causes: %s\n" + " Global causes: %s", + local_no_coll_cause_string, global_no_coll_cause_string); + } } } @@ -899,31 +1271,34 @@ done: #endif /* H5_HAVE_PARALLEL */ /*------------------------------------------------------------------------- - * Function: H5D__typeinfo_term + * Function: H5D__typeinfo_term * - * Purpose: Common logic for terminating a type info object + * Purpose: Common logic for terminating a type info object * - * Return: Non-negative on success/Negative on failure + * Return: Non-negative on success/Negative on failure * - * Programmer: Quincey Koziol - * Thursday, March 6, 2008 + * Programmer: Quincey Koziol + * Thursday, March 6, 2008 * *------------------------------------------------------------------------- */ static herr_t -H5D__typeinfo_term(const H5D_type_info_t *type_info) +H5D__typeinfo_term(H5D_io_info_t *io_info, size_t type_info_init) { + size_t i; + FUNC_ENTER_PACKAGE_NOERR /* Check for releasing datatype conversion & background buffers */ - if (type_info->tconv_buf_allocated) { - HDassert(type_info->tconv_buf); - (void)H5FL_BLK_FREE(type_conv, type_info->tconv_buf); - } /* end if */ - if (type_info->bkg_buf_allocated) { - HDassert(type_info->bkg_buf); - (void)H5FL_BLK_FREE(type_conv, type_info->bkg_buf); + if (io_info->tconv_buf_allocated) { + HDassert(io_info->tconv_buf); + (void)H5FL_BLK_FREE(type_conv, io_info->tconv_buf); } /* end if */ + for (i = 0; i < type_info_init; i++) + if (io_info->dsets_info[i].type_info.bkg_buf_allocated) { + HDassert(io_info->dsets_info[i].type_info.bkg_buf); + (void)H5FL_BLK_FREE(type_conv, io_info->dsets_info[i].type_info.bkg_buf); + } /* end if */ FUNC_LEAVE_NOAPI(SUCCEED) } /* end H5D__typeinfo_term() */ |