Commit Graph

328 Commits

Author SHA1 Message Date
Unknown W. Brackets
c53386244d Don't hash the clut when it's not even used.
Many games load the clut even when it wasn't changed, and we don't flush
between when this happens.  There shouldn't be any need to rehash the clut
either.

Technically, the clut could have actually changed, but this would
generally require them to clear the dcache or something in a way the GE
probably doesn't support.
2013-05-26 12:41:36 -07:00
Unknown W. Brackets
373561b90e Don't store clutformat in the texcache entry. 2013-05-20 01:00:14 -07:00
Unknown W. Brackets
3efe7ae24d Cache based on the full clutformat.
Not just 2 bits of it.  Oops.  Simpler this way.
2013-05-20 00:53:39 -07:00
Unknown W. Brackets
5619c84432 Defer palette conversion after clut load.
Because the format can easily be specified afterward.
2013-05-12 10:57:41 -07:00
Unknown W. Brackets
5223ee3d1b Move the font clut opt check to clut load.
And remove the report for mipmap sharing, seems to work...
2013-05-12 09:26:17 -07:00
Unknown W. Brackets
901a7b804e Implement mipmap clut sharing/not sharing.
It was kinda already there, probably from JPCSP.  Not well tested,
but this is what JPCSP does and it makes sense.
2013-05-12 09:04:50 -07:00
Unknown W. Brackets
7e23299c36 Immediately load the clut on LOADCLUT. 2013-05-12 01:56:26 -07:00
Unknown W. Brackets
4c43da9c23 Only apply the clut offset when using the clut.
This may fix games that use the clut offset having wrong colors.
2013-05-12 01:56:25 -07:00
Unknown W. Brackets
5a81bfd1fb Don't invalidate textures based on the clutaddr.
The cluthash should be sufficient, and this is less to pass around.
2013-05-12 01:56:25 -07:00
Unknown W. Brackets
7dd0d242e2 Load the clut only using LOADCLUT, not every tex.
This is a small optimization, and should match PSP behavior better.
It also allows us to hash the clut and convert colors less often.
2013-05-12 01:56:24 -07:00
Unknown W. Brackets
342ae6c322 Reuse the fullhash when it triggers a reload.
Instead of hashing it again a few microseconds later...
2013-05-12 01:56:24 -07:00
Unknown W. Brackets
724495cbb3 Mix add and xor in the texcache for better results. 2013-05-12 01:56:23 -07:00
Unknown W. Brackets
4b91c7e707 Implement DXT3 alpha. 2013-05-10 00:49:13 -07:00
Unknown W. Brackets
47c404b925 Clean up some comments. 2013-05-09 01:19:55 -07:00
Unknown W. Brackets
4a30221c95 Ignore texfunc RGBA when texture is full alpha.
This allows us to skip some work in the fragment shader.
2013-05-08 23:14:20 -07:00
Unknown W. Brackets
cb217189c5 Keep track of whether a texture has alpha.
This way we know what sorts of blending we even have to do.  Commonly,
textures don't have any real alpha.

Not yet used for anything.
2013-05-08 23:14:19 -07:00
Unknown W. Brackets
34da6f983a Fix reporting typo logging wrong things. 2013-05-08 00:35:17 -07:00
Unknown W. Brackets
76c812a428 Allow larger texture buffer widths.
Fixes #1631.
2013-05-06 23:54:41 -07:00
Henrik Rydgård
d8f9790abc Some Mac build fixes from vit9696 and some more. Update submodules. 2013-05-07 00:08:57 +02:00
Unknown W. Brackets
3c0ed69fd2 Improve GPU invalidation interface, check fonts.
When a font is rendered, tell the GPU about it, but don't rehash.

This not only improves perf (less hashing) but makes font changes more
immediate (no deferred hashing.)  But only if they use sceFont.
2013-05-05 22:52:01 -07:00
Unknown W. Brackets
995ab510e9 Optimize alpha-only CLUTs, common for fonts.
This optimizes the case where fonts are drawn using animation (common in
RPGs, for example), and textures have to be decoded over and over.

Could probably be faster with SIMD.
2013-05-05 22:34:35 -07:00
Unknown W. Brackets
d86253cef2 Buildfix for Linux. 2013-05-05 19:58:59 -07:00
Unknown W. Brackets
bd692ddf66 Check for low GPU memory and try to disable cache.
This may help with phones and even desktops without much VRAM/RAM.
Normally OpenGL will copy between them, though, so it should not happen
often, especially not on desktops.
2013-05-05 00:00:22 -07:00
Unknown W. Brackets
5b970d76e3 Use SSE2 in the texture hashing function.
Speeds it up significantly, but only when it's called a lot.
2013-05-04 21:39:12 -07:00
Unknown W. Brackets
02c88b9517 Add a secondary texture cache again.
This time, only used when invalidations happen several times, but not
too often, and still hashes the entire texture.
2013-05-04 20:33:31 -07:00