Fix clear detection mistake

[unknownbrackets]

Fixes #13320, was just a mistake in 1432cc9.

It's possible this may re-break #12858 because it may have been incorrectly setting the safe size when it was not, in fact, safe. That said, I think it'll still work.

-[Unknown]

[hrydgard]

Nice! Those masks are tricky heh.

Let's try it.

[i30817]

I think this might have fixed the 'menu' problem in 3rd birthday in retroarch. I just tried it on a GLES retroarch build (but on the desktop, it's the only way i can run ppsspp libretro) with the frameskip gone and no weird judder apparent.

[unknownbrackets]

If it fixed a retroarch-only bug, it probably means something like function replacements are being forced off in retroarch.

-[Unknown]

[i30817]

I'm not sure, i didn't make the bisect.

#13250 and #13201

I'm also unsure what this commit would have to do with function replacements but since i didn't bisect i can't comment. Maybe i will to make sure.

[unknownbrackets]

Well, some of our replacements would do something this commit also can do.

-[Unknown]

[i30817]

Ah I got you. I'll make a bisect to make sure then open a bug report. I'm pretty sure that i'm placing the files in the right dir but maybe it's a question of having renamed the retroarch 'system' dir from its default 'system' name to 'bios' (i did it ages ago in retroarch.cfg and never changed), and the ppsspp core part trying to read from hardcoded 'system'.

[i30817]

Whats the name of the function replacements file in the assets, 'compat.ini'?

[unknownbrackets]

You're probably thinking of knownfuncs.ini, but that file is not used by default and not required for function replacements.

The list of function hooks are compiled into PPSSPP and enabled normally by an ini setting. No assets are needed.

-[Unknown]

[i30817]

I was mistaken it was 5313fc5 that fixed it. Possibly only on opengl, but maybe the last vulkan fix takes care of that too (i don't have a vulkan capable gpu).

[hrydgard]

This change somehow caused #13344 as a side effect.

[unknownbrackets]

Hm, likely from #12952 then ultimately - this fixes a typo in that.

Hm. Maybe the game does a block transfer into the current target or something? But we'd detect that. If it really is a full screen sprite, how could that overwrite the wrong memory? I suppose we don't check scissor X1/Y1...

-[Unknown]

[hrydgard]

Let's move any discussion to #13344 . Like I said there I don't think this commit is wrong, this fix just stopped covering the original bug, which #12952 kept doing due to its own bug hehe.