* Stop trying to be smart and generate repeats on the fly; this breaks
time consistency. Also if repeats are not handled in time this causes
infinite loops.
* Move rarely-used functions to external files, simplify stuff, get rid
of internal driver events; saves ~1 kB per add-in overall.
(*) O_APPEND is *not* functional yet, this is just a hack for write-only
streams. Currently O_APPEND does not reposition the cursor at the end of
the file before every write.
There is no evidence that BFile_Create() keeps the address and writes
to it later on, but I'd rather be overly careful than have to debug a
stack corruption problem in half a year :)
I'm pretty sure it makes no difference because the OS does not rely on
interrupts for most (if not all) of its DMA operations, but it's better
to keep it clean anyway.
This helped locate some bugs:
* read() could read past EOF due to BFile_Read() allowing you to read up
until the end of the last sector, beyond the file size
* pread() did not restore the file offset because the negative seek at
the end is not relative (that was the CASIOWIN fs API), so pread()
could not actually be written without knowing the current position
* lseek() would clamp you to EOF but still return its out-of-bounds
arguments, as a direct result of BFile_Seek() doing that
Benefits:
* Made pread() a generic function
I saw a crash with the 12 kB stack. Added an error message to diagnose
further similar issues, and bumbed the stack to 14 kB. That's a lot of
space just for BFile but stability is queen... :x
* Mark SPU memory as sleep-blocking.
* Perform 4-byte accesses only in dma_memset() and dma_memcpy() (32-byte
accesses freeze as one would expect).
This change does *NOT* implement support for SPU's integrated DMAC.
The checks for VRAM access account for image columns intersecting the
longword before the start of a VRAM line, but not the longword after the
start of a VRAM line. This is now fixed.
