Cyphall

[Cyphall @Cyphall]

@RaymanGame 2K is 1080p, why does it say "2560x1440"?

[Cyphall @Cyphall]

@SheriefFYI @hossamkhero Most video/image codecs have specific endianness defined in their spec. If you don't take that into account you're going to have a bad time decoding them.

[Cyphall @Cyphall]

@Gorgonner6 @gamesandgraphs @OfficialPCMR This only works at high framerate where the distance between the discrete teleportations is small enough that the game movement tends towards a contiguous movement that the eyes need, hence why game motion blur is only important at low framerate.

[Cyphall @Cyphall]

@unusual_whales Alright I've seen enough, time to build the Blackwall

[Cyphall @Cyphall]

@Dreammur1 @gamesandgraphs @OfficialPCMR I'm obviously talking about the motion blur for in-game movements. Your brain filters-out most of received visual data during eye movement so there is hardly any motion blur there anyway.

[Cyphall @Cyphall]

@OfficialPCMR Also, comparing motion blur with side-by-side screenshots is about as honest as saying 60 FPS is not better than 10 FPS by showing identical side-by-side screenshots.

[Cyphall @Cyphall]

@OfficialPCMR Motion blur trades image clarity for perceived smoothness, so it entirely depends on the framerate.
At 30 fps it is literally required, at 120+ it is useless.
24 FPS movies would be unwatchable without camera's natural motion blur (pause during movement in any movie to see it).

[Cyphall @Cyphall]

@gamesandgraphs @OfficialPCMR Eye motion blur works when a surface actually moves from point A to point B (relative to the eye). It does nothing when said surface teleports from point A to point B (two frames on a monitor).

[Cyphall @Cyphall]

@SebAaltonen Update: the issue is somewhat fixed (band-aid fix for now, and a more permanent fix by not using STORAGE_BUFFER at all is planned)

[Cyphall @Cyphall]

@SebAaltonen I've heard there is a KHR version of shader_object in the works that aims at being efficiently implementable by all vendors

[Cyphall @Cyphall]

@SebAaltonen Intel Windows driver is borderline unusable for anything remotely modern.
Extension support is lagging hard, and last time I tried to make Slang BDA work on it, the driver compiler would just crash or return a NULL pipeline handle.

[Cyphall @Cyphall]

@TheGingerBill @SebAaltonen Agree but it's the least bad of the options unfortunately

[Cyphall @Cyphall]

@lukas_lipp @SebAaltonen An image view with SAMPLED or STORAGE usage is already a descriptor.
I'm pretty sure this extension just let you get the heap index/address of the descriptor, so this is just descriptor_heap with extra steps.

[Cyphall @Cyphall]

@SebAaltonen What triggers me with attachments is that they still need proper VkImageView instead of taking VkImageViewCreateInfo like descriptor_heap (or a more specialized struct since half the members are irrelevant for attachments)

[Cyphall @Cyphall]

@SebAaltonen Worth knowing then that Slang's C layout also changes bools to be 8-bit instead of 32-bit

[Cyphall @Cyphall]

@leidegre @codeshaunted Getting the support added to the driver is the problem.
On Linux, his API will likely support down to GCN 1.
On Windows however, even RDNA2 will be left out, thanks to AMD.

[Cyphall @Cyphall]

@GaySpaceAngel @IrisShaders I agree that buffer descriptors should really not exist anymore but my point is that unlike storage buffers which generally hold generic data, uniform buffers are basically shader parameters, so they are heavily tied to a single shader and don't benefit from being bindless at all

[Cyphall @Cyphall]

@IrisShaders On pre-Turing, uniform buffers are bound to hardware slots whose indices are baked into the shader code.
This actively goes against bindless uniform buffers (hence the limit), but tbh I don't see any case where using bindless uniform buffers would be of any use anyway.

[Cyphall @Cyphall]

@IrisShaders Have you considered using push descriptor for uniform buffers? It's usually one of the fastest paths for them and you don't have such a low limit anymore (other descriptors should stay in the bindless heap ofc)