Ryan Harrison

Should we remove the downloads provided by the builders?

6

Points I can immediately think about this: Reason to remove: 1. These builds are not pegged to any specific release, instead they are just most recent build that has succeeded....

It would be nice to have a dashboard or server with the current state of test coverage data for the library, instead of having to manually generate it by hand.

Have some sort of dashboard for current build/tree status

1

There are ways for people to get the status of specific builders, if they know the correct arcane incantations. We should have some sort of dashboard that is easy to...

Fix test breakage in shaderc_util_compiler

7

CompilerTest.HlslFunctionality1Enabled in this test suite is failing, because the third_party/spirv-tools has not been rolled to pick up the SPIR-V 1.4 changes yet, but the test expectations were updated in #608...

Precision requirements for `atan2` appear to be too restrictive for some backends

6

While working CTS tests for `atan2` I have encountered backends, specifically Nvidia & Intel running DirectX, that appear operating with different accuracy requirements than what is currently in the spec....

Accuracy of `acosh` is does not include Vulkan CTS alternative

8

Currently `acosh` is listed as having an accuracy of `log(x + sqrt(x * x - 1.0))`. By this definition `acosh(1)` explodes, because eventually has a 1/0 for the value of...

Accuracy of `mix` does not cover all common implementations of linear blend

2

Currently `mix`'s accuracy is listed as: ``` Inherited from x * (1.0 - z) + y * z ``` The definition of mix uses this as an example implementation, but...

Currently `ldexp(e1, e2)` is defined as `Returns e1 * 2^e2`. This definition does not explicitly state any over/underflow behaviours, thus we fall back to the general over/underflow rules for WGSL....