Commit 6bcd830490 ported an optimization to decode_subframe_fixed()
from FFmpeg (upstream commit 08965b22e2). This contains an out of
bounds read, which doesn't affect the decoder output, but makes ASAN
complain.
FFmpeg fixed the out of bounds read (upstream commit 0ec7b71de8) but
that appears to increase code size a lot.
Inlining the initialization of a, b, c, d into the switch produces
similar code as the non-bounds-checked version with only a handful
of instructions of overhead (checked on MIPS & ARM).
Change-Id: I053fac4efc4676b133eb7545c80e23f37fb00d86
Not really sure how best to deal with the warnings (redefinition of
ARRAYLEN, etc.) short of splitting things out of system.h but this
is good enough for now.
Change-Id: Ibea9f693d128c7995f564f0f5c81d572462a5553
These are more efficient than separate pin/unpin calls because
pin count increment and decrement can be done cheaply when the
data pointer is known.
Secondly, pinned access can be made safe against preemption by
hardware interrupts or other CPU cores; buflib_get_data() can't.
This makes it more useful under different threading models and
for SMP targets; both of which are not particularly relevant to
Rockbox now, but might be in the future.
Change-Id: I09284251b83bbbc59ef88a494c8fda26a7f7ef26
This is intended for improving the effectiveness of tools like
ASAN when debugging memory errors in the sim. It's not meant to
be a serious allocator for hosted targets.
Enable it by changing the buflib backend in config.h.
Change-Id: I0cf23cefa47ee35dede7b49e0e5b72dac60e8d3e
To minimize code duplication between buflib backends move the
public part of the API to buflib.h. Also rewrote documentation
for the whole API.
Change-Id: I4d7ed6d02084d7130cb41511e63c25ec45b51703
You could only add single files to playlists
from the database browser before. This
enables adding any database selection to
a new or existing playlist.
Change-Id: I811c7167641c589944bb2afc18dcc1d299a7b979
move seconds and sizes to arrays indiexed by enums
use a loop for the display and talk of each
stop exiting on SYS_EVENTs
Change-Id: I49d5b9827df4e711b38326e5fef3c54292000370
bad tracks now get skipped with a message at the end of scanning
rather than an error killing the count
Change-Id: I6d8c14ce00e78416b772bc5e9093a889351bc3de
Sansa Clip and Clip+ have a split monochrome screen
some versions have a dead line of pixels
having text split at this line makes it hard to read
move text picker below this split on these devices
Change-Id: I1ebcb4c4c7b1ea950f38e35fed06ed85437a657f
This reverts commit ebebef5566.
Reason for revert: I don't want to get too deep into this till I come back to the keyboard rewrite (hopefully)
Change-Id: Ia273f1a19a042be2dd0f1ee46690c03f2865cd95
Gate buflib_get_data() checking, debug printing, and buflib
integrity checks behind individual defines in buflib.h, and
turn them all off by default. If needed, they can be turned
on manually when compiling.
The buflib debug menu is only available if debug printing is
enabled, so after this commit it will no longer be included
in normal builds -- it isn't very useful to end users.
Change-Id: Iab25b7852bc7c5592ce04c9c45762046a87d5bc3
Removing the variable-length name in the middle of buflib metadata
makes it a lot easier to deal with.
Change-Id: I6eaf236c2285cae40fb6ff0fa5acf827972cdd8b
The CRC is a fairly useless safety check because we already
have specific checks to validate the metadata, and CRCs are
only verified before calling the move callback. Removing the
check should not significantly reduce buflib's robustness.
Change-Id: Ica99bd92fc514819b4fd9f359b4272e581020f75
Remove the name handling code, but leave the allocation structure
otherwise untouched; it's as if all callers provided a NULL name.
The public API still accepts names but names are no longer stored
or returned.
Change-Id: I6c4defcdfd255774f02030949a0fd731477e6a54
Naming your allocations is more of a debugging feature; it's a
bad idea to abuse this feature for storing some random string.
The font code does exactly that to remember the path used to
load a font, and it's the only code that uses buflib names in
this way.
Store the path inside the font allocation instead as a regular
char array. For simplicity it's MAX_PATH sized, so this'll use
a bit more memory than the buflib name method, maybe 1-2 KiB if
a few fonts are loaded.
Change-Id: Ie286a1a273d6477af9e5d3f76e0534b62f72e8f7
dc_thread_playlist was asking for invalid indices
since previously the name buffer would have been valid
it just got whatever junk data was left over
add dc_discard_playlist_pointers for HAVE_DIRCACHE targets
this allows the dc_playlist_thread to stop its current lookup loop
Change-Id: I6f25b97b8c4e314d27c5e1e6ff0925b5a3e93f26
dc_playlist_thread may continue loading pointers even while
underlying indices are changing instead stop the loop by marking
pointers as clean and rebuild later
Change-Id: If154f673b4af8d6e9c364499d58f1321834a76a4
It's usually only called from init() in main.c, so this is safe.
There is one more call in system-dm320.c from system_init(), but
that's also "safe". I don't know if it's okay to call i2c_init()
twice, but presumably it works...
Change-Id: I9c1cd918d162d9955f7cf03209e836cbd5e30c57
On single-core targets, system_init() is only called from init()
in main.c, which is also INIT_ATTR, so this is safe.
Multi-core targets additionally call system_init() from cop_main(),
which isn't marked INIT_ATTR. I'm fairly certain that it would be
safe to add INIT_ATTR to cop_main() with a bit of refactoring, but
I don't have a way of testing it. So for now, leave INIT_ATTR off
for multi-core targets.
Change-Id: I8fe4358f975880ca317d46bda01d7d9db529d3e3
It seems removing this causes a crash on the Clip+ when playing
any file. Appears to be a timing-related issue as replacing the
loop with an mdelay() also fixes it. Needs further investigation
to identify the real cause of the problem, see FS#13386.
Change-Id: Ia93a2887a79b376de50563d6bb3bbc79cee11a1c
This reverts commit 89c021fbfa.
Reason for revert: crash when playing on disk playlist then playing a directory as an in-ram playlist
Change-Id: Ia5cf5da9f46f8c10c5c0f3707e7978c05664b8a4
chunk_alloc allows arrays (or any data) to be allocated in smaller chunks
you have to save the indices..
variable data will have variable indices you need to
store these as [chunk_alloc] doesn't keep track
if you have a fixed size for each
alloc you can do indice / sizeof(data)
or index * sizeof(data) to convert
Lots of debug stuff still in and it needs optimization
User provides chunk_size and max_chunks
max_chunks * struct chunk will be allocated at start
with (1) chunk_size allocation initially
alloc_chunk() with size = 0 shrinks the last allocation to the size of the data used
add OOM checks on buflib_alloc -- oops
move bytes available to the header -- less memory per chunk & better alignment
keep track of the current in use chunk index -- should speed things up a bit
Now allows:
realloc chunk header
larger allocations than chunk size
reallocs smaller than existing will shrink the current array
rather than alloc a new and copy data
Comments welcome :)
Change-Id: I8ed170eef73da95da19430a80b32e5debf0c8276
Context menu gains new option to view
metadata for individual tracks or albums.
Will display an album's length and total file size.
Other fields are displayed only if they are identical
across all tracks (except for the album year, which
is determined by the highest value encountered).
Change-Id: Ibc14cfaf2cb3d91b8d1cfbee05c6261db4975355
voicefont.c expected a max of 999 voice IDs we are at 1013 or so
bad stuff happened TM new limit is 2048 and added an error message
(no file will be created if limit exceeded)
Change-Id: Ifda6dc5c45883551f8ae8f0d4efc9f7acdb7c90f
give user option to load playlist from disk when
namebuffer is too small -- max_playlist size still
limits the total entries but filename length
won't stop them from playing the directory
Change-Id: I1787689417661ea670a211f575f2c52e84465869
Fixes a bug introduced by commit 5d0c382a59 that caused graphical
corruption on anti-aliased fonts.
Change-Id: I6052ca758382bd9a1154d2e2208dee633dd17715
