added memory barrier intrinsics to spirv-std

[hannes-vernooij]

Added memory barrier intrinsics to get feature parity with GLSL and HLSL. Currently adding barriers is quite verbose and most used usecases are covered with these intrinsics. I used the DXC to compile the hlsl equivalents to SPIR-V.

[repi]

Nice. Could you add some function-level docs for each of the barrier functions? Could describe it on a high-level and then maybe reference the GLSL/HLSL functions (as you do internally in the function body, but in the docs) and link to more details about them?

[khyperia]

Additionally, please add some compiletests!

[hrydgard]

Linking to related issue #696

[hannes-vernooij]

Additionally, please add some compiletests!

I added compile tests for all intrinsics but I was not able to validate the constants, they are probably defined outside of the scope.

%1 = OpFunction  %2  None %3
%4 = OpLabel
OpLine %5 40 4
OpControlBarrier %6 %7 %8
OpLine %9 8 1
OpReturn
OpFunctionEnd

As you can see in the example above %6, %7 and %8 are not output. Do you maybe have an idea on how I can get this incuded in scope for the tests?

[khyperia]

Ah, sorry, should have been more clear - I mostly wanted compiletests to make sure it just compiles/validates, rather than disassembling - just a compiletest (without --disassemble-xyz) is fine. How it is now is perfectly OK too, though!

Review-request @hrydgard to make sure the names here are OK, and no more are needed to solve your use cases.

[hrydgard]

Looks great.

[hannes-vernooij]

There is one problem regarding the tests; the workgroup and all_memory barrier intrinsics currently rely on the VulkanMemoryModelDeviceScopeKHR and SPV_KHR_vulkan_memory_model features. I added these to the target_features defined in tests/src/main.rs but this causes some tests to fail due to the added OpCapabilities, Is it okay to re-generate and push these changes?

[khyperia]

Hmm. You should not need to add the capabilities to tests/src/main.rs, but rather adding them to compile-flags: -C target-feature like other tests do. The outputs of other tests should not have to be updated.

[hannes-vernooij]

Hmm. You should not need to add the capabilities to tests/src/main.rs, but rather adding them to compile-flags: -C target-feature like other tests do. The outputs of other tests should not have to be updated.

Oh sorry about that, I could have figured that out myself. I was thrown off a bit by the features above that were only used by read_cloack_khr.rs.

[khyperia]

Yeah, in that case there were complications, which, should probably also happen here, but I'm way too tired :c and how it is now is totally fine