D3D: remove load texture on creation optimization

[degasus]

No description provided.

[degasus]

on irc: I would expect that very similar work is required (copy resource to system visible memory, then copy to video private).

[kayru]

On the second thought, it might be the case that upload to video private is done synchronously. This would make it faster than UpdateSubresource.

[degasus]

@kayru So do you think this optimization matters? I want to implement a texture pool like your rendertarget one. So this optimization is in my way :/

[kayru]

I think the best thing is always to measure real impact.

[JMC47]

NSMBWii EFB2RAM 3x IR Master: 89 FPS
NSMBWii EFB2RAM 3x IR PR: 71 FPS

1x IR is 119 fps on both builds.

[pauldacheez]

Texture pooling would likely be more beneficial than this optimization anyway. If you can reimplement this later for a win-win, though, that'd be cool.

[kayru]

@degasus what is the problem with passing in initial data on texture creation? You should still be able to use UpdateSubresource to change content later.

[degasus]

@kayru With pooling, we'll hit this path less often. It's also in the way of decoding directly into a mapped staging buffer.

[PatrickFerry]

So why remove the optimisation if it badly affects higher IR values?

[JMC47]

To make things faster in the long term.

[PatrickFerry]

So maybe this PR can wait until a replacement gets made?

Edit: Please :( puppy eyes

[degasus]

@ZephyrSurfer That's the reason why this PR wasn't touched in the last three weeks ;'-)

[degasus]

degasus> so is this PR as good as master?
JMC47> for me, at leats
JMC47> it seems fine in NSMBWii
JMC47> I don't see a big slowdown
mimimi> in D3D with 3xIR, pr1501 and master 5010 are at the same speed

So ready to merge?

The old slowdown wasn't because of this optimization. I did also remove the hint to mark 1-level textures as default instead of dynamic. But it seems to be faster to create map-able textures.

[CrossVR]

LGTM, the changes in the PR look to be purely cosmetic.

[waddlesplash]

This PR is the cause of the D3D backend crashing with Intel cards. With previous revision, tried multiple startups of both games & the emulator and it never crashes, with this PR it crashes 100% of the time in SMG2 just past the Wiimote screen.