FILE_CACHE_BLOCK flag used more safely (copied in afile_read)
add file_sector_size (set from sysdep code)
documented and organized FileFlags
moved publically visible file_buf* decls to file.h; add FileBufFlags
file_buf_get: moved to file_io and renamed because it is not a part of
the file_buf manager itself
FILE_BLOCK_SIZE no longer exposed to users - it's an implementation
detail. for alignment purposes, you should use file_sector_size (less
wasteful)
add page_aligned allocator (mmap)
write buffers now allocated via file_buf_alloc(FB_FROM_HEAP) (avoids
wasting cache space)
append trace runs into the trace file (-> better input data for archive
ordering)
document lib_errors.h
tex_load: add provision for file_flags, used by ogl_tex_load to prevent
caching file data
simulation, entitymanager: add perf measurements
This was SVN commit r3725.
adts: slight improvements
allocators: change POOL_VARIABLE_ALLOCS to -1 for safety; fix invalid
assumptions of it being 0
lib: more expressive param names
file: add file_get_random_name for randomized trace generation; allow
file_cache.h to declare externally visible file_buf_alloc API
file_cache: zero-length file bugfix; also add several flags to the API
routines allowing statistics gathering to be disabled (avoids distorting
stats due to e.g. trace_entry_causes_io cache simulation)
file_stats: add archive builder info; separate IO timing and seek
accounting (-> correct results)
trace: add random generation capability
vfs: made responsible for cache hit/miss accounting (fixes cache miss
rate display)
This was SVN commit r3681.
- compression: now use "fastest" compression mode and more memory unless
in FINAL build. this greatly speeds up archive generation at expense of
1.5% size increase (fine during development). done because profiling
shows ZLib accounts for 78% of CPU; practically nothing else shows up,
which shows effectiveness of compressing in parallel with reading from
disk.
- file_cache: documented; fixed several potential issues with exact_buf
when invalidating files.
This was SVN commit r3586.
everything else: bugfix: buffers added to extant list via
file_buf_retrieve (i.e. moved from cache to extant list) didn't have
their long_lived flag carried over; this caused a "not freed
immediately" warning.
This was SVN commit r3580.
- allocators.cpp: fix: make sure all pool fields are reset in
pool_free_all; correct matrix self-test
- lib: fix self-test
- lib_error: give "informational" return codes values > 100 (allows
multiplexing with percent indicator)
- archive: add clarification to afile_enum; use vfs_load when building
archive (fixes caching problem); add support for progressive archive
creation
- file: allow pp_set_dir's path to end in slash (simplifies other code);
add file_exists
- file_cache: add stats+sanity checks to allocator; improve ExtantBufMgr
find(); add ExactBufOracle to translate padded address returned by Cache
to what was actually allocated; relax infinite-loop check; fix
file_cache_reset
- file_io: refactor callback logic so it can be used by vfs (needed for
vfs_load)
- vfs: vfs_load now supports calling back (required for archive builder)
- vfs_mount: optimize path caching in afile_cb: due to atom_fn system,
string compare can be done via pointer compare
- vfs_optimizer: support auto-build of archive (progressive); improve
should-build-archive check (WIP)
- vfs_path: add util functions (also used in screenshot code)
- vfs_tree: simplify pathname storage
- wposix: mmap bugfix
- main: call progressive archive builder; skip reload when doing so
(documented)
- JSInterface_VFS+scriptglue: add archiveBuilderCancel
- ps/Util: use new vfs_path helper function for cleverer screenshot
naming (correctly handles some being deleted)
This was SVN commit r3578.
lib: add u32_hi/lo
archive: move archive builder logic here from vf_optimizer
file_cache: add flush() call - used when simulating cache
vfs_optimizer: WIP, not yet functional: file gatherer, 50% of TSP tour
generator
This was SVN commit r3486.
archive: afile_read bugfix (wasn't returning byte count)
byte_order: add to_leXX, implement read_leXX in terms of that, add
write_leXX (and also big-endian versions)
compression: support 0 length buffers + minor fixes
file: no longer support passing fn_len (that didn't work anyway due to
DynHashTbl find(const char*) interface); add FILE_DONT_SET_FN
optimization that prevents cluttering name cache.
file_cache: add logic to verify load/free/load/free sequence; add
FILE_LONG_LIVED flag to specify exceptions (e.g. XMB files that don't
free their buffer immediately)
vfs_mount/vfs_tree: bugfix (was mixing up portable and VFS paths)
vfs_optimizer: small fixes; creating Zip files now works
vfs_tree: add PathName that stores pointer to path and name component
for efficiency + convenience
zip: bugfixes (wasn't endian safe; incorrect handling of cmethod; forgot
to initialize ZipArchive members)
config: add -buildarchive flag
Xeromyces.cpp, FileUnpacker.cpp: add FILE_LONG_LIVED
mapreader: bump time estimate for ReadXML
This was SVN commit r3446.
allocators: add freelist capability to Bucket; add provision for
variable XOR fixed size allocs
archive: re-tag file buffers if reading uncompressed from archive;
improve LFH fixup logic
file_cache: add cache line invalidation; lock down pages (readonly) when
IO finished
file_io: cleanup+docs; properly cut off at EOF without breaking
alignment.
file_stats: add seek accounting (WIP)
vfs_optimizer: also record file_buf_free in the trace. initial
implementation of archive builder (WIP)
zip: lfh_fixup now more efficient (does not involve buffer manager -
instead it grabs LFH from temp blocks)
tex: plug FileIOBuf leak. avoid writing to tex.hm because that is a
read-only file_buf.
This was SVN commit r3428.
previous zip.* split up into archive, zip and compression. add O(1)
lookup via memento. add support for writing Zip archive. simplify IO by
having compression module allow queuing input buffers.
file: split up into file, file_Io, file_cache. add atom_fn that enables
string comparison via pointer compare. (stores all filenames/paths in
one place). add statistics gathering.
file_cache: landlord algorithm for caching files as well as LRU block
cache.
file_io: aio code that splits transfers into blocks revamped and made
into class.
vfs: filelist/trace code moved into new vfs_optimizer. add caching logic
to vfs_load.
vfs_optimizer: will generate zip archive (to replace current archive
builder script). not yet done
vfs_tree: major cleanup - now has nice base class rather than union
hackery.
This was SVN commit r3420.