Petr Kraus

> Shouldn't it most likely be VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT to match For the triangle demo, yea. Though it is used for all the demos, so hard to be sure. > should have...

What layers say is bit irrelevant. It is the low bar and necessary but not the only condition. That layers are silent does not imply it is good code. This...

Right, I faced the same problem. Sync is fundamentally untestable. And I don't really want to inspect each of the 50 examples manually neither. Let's perhaps start with just the...

> The spec also doesn't have any VUID which says anything about format extensions or format features to be enabled. https://www.khronos.org/registry/vulkan/specs/1.3-extensions/html/vkspec.html#fundamentals-validusage-extensions > So, based on "or if the feature is...

That's weird. Almost all built-in constants are "Implementation-dependent". `gl_SubgroupSize` is practically the same. It is a physical device limit too.

It is bit unfortunate. That would disqualify its use in e.g. `subgroupBroadcast()`. Shouldn't something like `subgroupBroadcast(x, gl_SubgroupSize - 1)` be allowed?

> because we don't want to make the ISV change the `.conf` file per target platform. This is also identical problem as for the other physical device limits, no? The...

`VK_PIPELINE_SHADER_STAGE_CREATE_ALLOW_VARYING_SUBGROUP_SIZE_BIT_EXT` is indeed problematic here, which makes it "only" a subgroup-scope uniform. It quickly gets messy to express all the levels of conditional constness...

PS: Maybe the original issue should be closed. The window in which it could have been fixed trivially is gone. Now it is settled and extensions are made on top...

Sometimes I feel people troll tabs on purpose. No wonder Rust rather goes with spaces like it is the 16th century...