Order of OpSwitch Target Operands

[vili-1]

This example is from here https://github.com/afd/spirv-control-flow/blob/main/scraped-validator-tests/spv_1_0_FULLY_VALID_ValidateCFG_SwitchMultipleDefaultWithFallThroughGood.spv that we discussed in our meeting. I've annotated in the graph all the .als and .asm encodings. I think the problem is that first does not immediately precede third. Attached the .als and .asm files.

@afd @johnwickerson you mentioned in the meeting that we may not have considered occurences of default in our model - I can see that our model is in line with the suggested new rules, but please correct me if I'm missing something.

Archive.zip

[afd]

Thanks! Will look. What I meant regarding not having considered default properly is that we didn't consider it fully in the new rules (I'm sure the model is in line with the rules, so perhaps the rules need a small tweak.)

[afd]

I looked at this. I agree that it is invalid according to the suggested new rules, and I think that's fine - i.e., I think this example should indeed be regarded as invalid.

Let's open an issue so that we can discuss it with Alan.

[afd]

KhronosGroup/SPIRV-Tools#4799

[afd]

It turned out that we did want this example to be valid, and we worked some more on the wording of the rules around switch statements to make it so.

Could discuss in paper under minor issues: an example of a conformance test suite informing our model.