MapleShrine/mkxp
1
0
Fork 0
Code Issues 32 Pull requests 11 Projects Releases Wiki Activity

Optimize archive reading

Browse source 
This should be almost as fast as reading unencrypted
files from disk now. I also don't see any possible further
optimizations, so this is probably as fast as it gets.
This commit is contained in:
Jonas Kulla 2013-10-04 07:52:31 +02:00
parent a9fc44f79d
commit 958af38442
1 changed files with 78 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

90
src/filesystem.cpp
View file

@ -148,26 +148,92 @@ RGSS_ioRead(PHYSFS_Io *self, void *buffer, PHYSFS_uint64 len)
{ {
RGSS_entryHandle *entry = static_cast<RGSS_entryHandle*>(self->opaque); RGSS_entryHandle *entry = static_cast<RGSS_entryHandle*>(self->opaque);
PHYSFS_Io *io = entry->io;
uint64_t toRead = min<uint64_t>(entry->data.size - entry->currentOffset, len); uint64_t toRead = min<uint64_t>(entry->data.size - entry->currentOffset, len);
uint64_t offs = entry->currentOffset; uint64_t offs = entry->currentOffset;
entry->io->seek(entry->io, entry->data.offset + offs); io->seek(io, entry->data.offset + offs);
uint64_t buffI = 0; /* We divide up the bytes to be read in 3 categories:
for (uint64_t o = offs; o < offs + toRead;) *
* preAlign: If the current read address is not dword
* aligned, this is the number of bytes to read til
* we reach alignment again (therefore can only be
* 3 or less).
*
* align: The number of aligned dwords we can read
* times 4 (= number of bytes).
*
* postAlign: The number of bytes to read after the
* last aligned dword. Always 3 or less.
*
* Treating the pre- and post aligned reads specially,
* we can read all aligned dwords in one syscall directly
* into the write buffer and then run the xor chain on
* it afterwards. */
uint8_t preAlign = 4 - (offs % 4);
if (preAlign == 4)
preAlign = 0;
else
preAlign = min<uint64_t>(preAlign, len);
uint8_t postAlign = (len > preAlign) ? (offs + len) % 4 : 0;
uint64_t align = len - (preAlign + postAlign);
/* Byte buffer pointer */
uint8_t *bBufferP = static_cast<uint8_t*>(buffer);
if (preAlign > 0)
{ {
uint8_t bitOffset = (0x8 * (o % 4)); uint32_t dword;
uint8_t magicByte = (entry->currentMagic >> bitOffset) & 0xFF; io->read(io, &dword, preAlign);
uint8_t byte; /* Need to align the bytes with the
entry->io->read(entry->io, &byte, 1); * magic before xoring */
dword <<= 8 * (offs % 4);
dword ^= entry->currentMagic;
((uint8_t*) buffer)[buffI++] = byte ^ magicByte; /* Shift them back to normal */
dword >>= 8 * (offs % 4);
memcpy(bBufferP, &dword, preAlign);
if (++o % 4 == 0) bBufferP += preAlign;
/* Only advance the magic if we actually
* reached the next alignment */
if ((offs+preAlign) % 4 == 0)
advanceMagic(entry->currentMagic); advanceMagic(entry->currentMagic);
} }
if (align > 0)
{
/* Double word buffer pointer */
uint32_t *dwBufferP = reinterpret_cast<uint32_t*>(bBufferP);
/* Read aligned dwords in one go */
io->read(io, bBufferP, align);
/* Then xor them */
for (uint64_t i = 0; i < (align / 4); ++i)
dwBufferP[i] ^= advanceMagic(entry->currentMagic);
bBufferP += align;
}
if (postAlign > 0)
{
uint32_t dword;
io->read(io, &dword, postAlign);
/* Bytes are already aligned with magic */
dword ^= entry->currentMagic;
memcpy(bBufferP, &dword, postAlign);
}
entry->currentOffset += toRead; entry->currentOffset += toRead;
return toRead; return toRead;
@ -191,10 +257,10 @@ RGSS_ioSeek(PHYSFS_Io *self, PHYSFS_uint64 offset)
entry->currentMagic = entry->data.startMagic; entry->currentMagic = entry->data.startMagic;
} }
/* For each 4 bytes sought, advance magic */ /* For each overstepped alignment, advance magic */
uint64_t currentDword = entry->currentOffset / 4; uint64_t currentDword = entry->currentOffset / 4;
uint64_t soughtDword = offset / 4; uint64_t targetDword = offset / 4;
uint64_t dwordsSought = soughtDword - currentDword; uint64_t dwordsSought = targetDword - currentDword;
for (uint64_t i = 0; i < dwordsSought; ++i) for (uint64_t i = 0; i < dwordsSought; ++i)
advanceMagic(entry->currentMagic); advanceMagic(entry->currentMagic);