Add WGSL support for slang-test

[aleino-nv]

• Add Tint as a downstream compiler
• Add some plumbing for slang-test to target WebGPU and WGSL using the newly added RHI support
• Disable several asserting slang-test tests in tests/compute

[aleino-nv]

@csyonghe Should I use the breaking label instead? I'm adding some stuff to the middle of an enum in slang.h.
Maybe I should just move that value to the end?

[aleino-nv]

There are some things that seem a bit strange about this: I have added "WGSL SPIR-V" and "WGSL SPIR-V assembly" targets in a bunch of places just to be able to express that I want Tint to produce the SPIR-V.

@csyonghe
Maybe there are other mechanisms to achieve this that don't involve adding these strange "redundant" targets?

[csyonghe]

To answer your questions: all enums need to be added at the end to avoid binary breakage. Defining wasm-spirv and wasm-spirv-asm as separate targets is fine.

[csyonghe]

Instead of adding DISABLED_TEST lines to individual .slang files, please add the test names to tests/expected-failure-github.txt so we can use the list as a centralized place to track which tests are being disabled.

[aleino-nv]

Instead of adding DISABLED_TEST lines to individual .slang files, please add the test names to tests/expected-failure-github.txt so we can use the list as a centralized place to track which tests are being disabled.

I can do that, but these tests are asserting which seems to interrupt the test run a lot. Disabling gets rid of that problem.

Do you still think I should just mark these as test failures?

[aleino-nv]

Instead of adding DISABLED_TEST lines to individual .slang files, please add the test names to tests/expected-failure-github.txt so we can use the list as a centralized place to track which tests are being disabled.

I can do that, but these tests are asserting which seems to interrupt the test run a lot. Disabling gets rid of that problem.

Do you still think I should just mark these as test failures?

As discussed on Slack I will keep these disabled and then also add to the expected-failure list.

[csyonghe]

Use String instead of std::string.

[aleino-nv]

Done

[csyonghe]

Leaking memory on tint_compileResult.

[csyonghe]

Use SLANG_DEFER to ensure things are freed:

auto rs = m_compile(&result);
SLANG_DEFER(m_free(&result));
if (rs != 0) ...

[aleino-nv]

Do you mean m_compile fails? A common convention (which is followed in this case) is to make sure that no resources are actually returned if the call fails.

extern "C"
SLANG_SHARED_LIB_EXPORT
int tint_compile(tint_CompileRequest* request, tint_CompileResult* result)
{
    ///...
    uint8_t *const resultBuffer {static_cast<uint8_t*>(::malloc(resultBufferSize))};
    if(resultBuffer == nullptr)
        return 1;
    ::memcpy(resultBuffer, spirvCode.data(), resultBufferSize);

    result->buffer = resultBuffer;
    result->bufferSize = resultBufferSize;
    return 0;
}

[aleino-nv]

(But of course I may as well add a DEFER anyway.)

[csyonghe]

How do we report back diagnostics returned by tint?

[aleino-nv]

Done. I added an error field to the slang-tint result struct.

https://github.com/shader-slang/dawn/blob/slang-tint/src/slang-tint/slang-tint.cpp#L62

[csyonghe]

On coding style:
No ); on its own line.
No ( on its own line.
No const unless necessary (required by c++ standard).
No T x{} initialization. Should be T c= {}
No return on its own line.

Long function signature should either be

Blah longFunc(
    All parameters);

Or

Blah longFunc(
   Param1
   ParamN);

[csyonghe]

Reading the ci log it doesn't seem wgpu is reported as supported and no tests are run for wgpu.

I merged Simon's and see if ci can pick up the tests this time.