Henrik Rydgård
2a1f6bca97
Fix NEON 4444 CheckAlpha, workaround for failed bNEON detection on ARM64
2015-05-31 12:29:08 +02:00
KentuckyCompass
880697f40a
Add NEON versions of the CheckAlpha family
2015-05-25 18:04:52 -07:00
Henrik Rydgard
b309c83973
Initial work on ARM64, based on the ARM jit.
2015-04-06 18:13:01 +02:00
Unknown W. Brackets
1b12b57e21
Fix reversed consts in hand-neon texhash.
2014-10-29 16:38:40 -07:00
Unknown W. Brackets
faea860c27
Attempt to match the SSE2 version of the hash.
...
Not tested, don't currently have an ARM device. May or may not have a
performance hit.
2014-10-29 16:37:45 -07:00
Unknown W. Brackets
3977f5a8ec
Add a non-SSE implementation of the QuickTexHash.
...
This way we can use it as a stable value for texture replacements.
2014-10-29 16:37:44 -07:00
Unknown W. Brackets
2b16b5b79b
Use XXH64 on 64-bit systems.
...
I'm not able to reproduce any real performance difference, but I know
there are some games which do more hashing and it may help there.
2014-10-26 17:49:24 -07:00
Unknown W. Brackets
b800762ceb
Add a NEON-optimized version of XXH32.
...
This takes at least 40% less time to hash on NEON/ARM devices.
2014-03-25 00:34:55 -07:00
Unknown W. Brackets
a26e6ce4b6
Minor improvements to QuickTexHashNEON().
...
Cuts a VFP switch, oops.
2014-03-23 11:56:25 -07:00
Unknown W. Brackets
970dd4f2f8
Use vldmia not vld1 for QuickTexHashNEON().
...
It turns out this is ~20% slower usually, so aligned is not worth the
pipeline hit.
2014-03-23 11:56:24 -07:00
Unknown W. Brackets
2482b2a1e0
Use aligned NEON loads in texhash.
2014-03-22 21:58:50 -07:00
Unknown W. Brackets
b44d10a91e
Move texture unswizzling to decoder, use NEON.
2014-03-22 21:35:16 -07:00
Unknown W. Brackets
2623f6bfe2
Disable the optimized hash on iOS, seems to crash.
...
Except not for me, not sure why.
2013-11-23 10:41:58 -08:00
Henrik Rydgård
662ad32d4f
Fix and re-enable @unknownbracket's handcoded NEON tex hash
2013-11-11 17:12:38 +01:00
Henrik Rydgård
53800bd413
Disable the hand-coded NEON texture hash again. It's slow, for unknown reasons.
2013-11-11 14:39:29 +01:00
Henrik Rydgård
1713465e9f
Revert "Temporarily disable the handcoded NEON texture hash"
...
and
Revert "Temporarily disable the handcoded non-NEON tex hash"
This reverts commit 068af6fda8df8069bc0491fe98e75ae007f5b650.
This reverts commit 91543e7a8d27b4a8c906c1e1c349f489a449c432.
2013-11-11 14:13:47 +01:00
Henrik Rydgård
c80403c377
Temporarily disable the handcoded NEON texture hash
2013-11-11 14:13:47 +01:00
Unknown W. Brackets
fa79d7f13c
Handcode the non-NEON texcache hash as well on ARM.
...
Well, it could be a better algorithm, but at least this uses LDMIA.
2013-11-09 14:51:23 -08:00
Unknown W. Brackets
2dfa2379f4
Rewrite QuickTexHashNEON directly in asm.
...
Seems like gcc couldn't optimize it to this.
2013-11-09 12:54:55 -08:00
Unknown W. Brackets
64e977db08
Improve the non-NEON tex hash path on ARM.
...
This generates better looking disassembly, though a small change.
2013-11-03 07:43:10 -08:00
Unknown W. Brackets
4d47ccd5df
Add a NEON version of the tex hash.
...
Should be used only for NEON devices. Currently only compiled on Android.
2013-11-02 02:09:54 -07:00
