Make basic CDI functions constexpr

[alexrlongne]

Background

• It would be nice to have a clear delineation of which CDI functions (static or otherwise) use class data. This PR just moves the lowest level functions that don't use the static member variable frequencyGroupBoundaries. I've also marked the low-level functions constexpr do that the GPU can call them more easily.

Purpose of Pull Request

• Make it easier to call simple Planck integration functions.

Description of changes

• Move functions that don't require any class member data outside CDI class
• Mark basic integration functions as constexpr
• Remove files that formerly held low-level planck integration functions

Status

• Reference:
  • Pre-Merge Code Review
  • CDash Status
• Checks
  • [x] Travis/Appveyor CI checks pass
  • [x] Code coverage does not decrease
  • [x] Valgrind test passes
  • [x] Toss3 checks pass
  • [ ] Trinitite checks pass
  • [x] Code reviewed/approved, sufficient DbC checks, testing, documentation

[KineticTheory]

@alexrlongne It is interesting that LLVM accepts your code, but gcc and msvc fail with the same error, expression not constexpr.

[alexrlongne]

@KineticTheory Ahhh ok, I was curious if this would happen. Looks like the "data" function of std::array is not constexpr to those other compilers. I'll see if I can rework this.

[attom]

Will this change the function signatures in such a way that it will break downstream codes? (Isn't a problem; I'd just like a heads-up and @jhchang-lanl would probably too :) )

[alexrlongne]

@attom Yeah, it might, sorry, I should have marked this WIP right at submission. I'll do a quick grep through your repos in a minute.

[alexrlongne]

OK! I think I fixed the constexpr issue. It's something to look again when we go to C++17.

@attom I have a branch with the fixes for Capsaicin. A few functions live in the rtt_cdi namespace now, the rest of the API is the same. I'll push that branch to trt in a moment.

[KineticTheory]

@alexrlongne Don't you wish there were some language standards that compiler's would adhere to? Oh, wait, we have those. 😃

[kgbudge]

[Z]

From: "Kelly (KT) Thompson" @.> Reply-To: lanl/Draco @.> Date: Wednesday, August 18, 2021 at 8:22 AM To: lanl/Draco @.> Cc: Subscribed @.> Subject: [EXTERNAL] Re: [lanl/Draco] WIP: Make basic CDI functions constexpr (#1109)

@alexrlongnehttps://github.com/alexrlongne Don't you wish there were some language standards that compiler's would adhere to? Oh, wait, we have those. 😃

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/lanl/Draco/pull/1109#issuecomment-901157668, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AE4AAATPD7AQO3JPD55GNZTT5O6ZZANCNFSM5CIAVTHA. Triage notifications on the go with GitHub Mobile for iOShttps://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Androidhttps://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email.

[alexrlongne]

@KineticTheory hahah, right? Stack Overflow that it's actually MSVC that's correct here and that GCC provides a "built-in" constexpr version (indeed cppreference says there's not a constexpr std::pow).

[alexrlongne]

Whoops, looks like MSVC also doesn't do constexpr exp or whatever it uses for the FMA. I'll back off on the constexpr and just mark them with the the macro we have for __host__ __device__.

[codecov[bot]]

Codecov Report

Merging #1109 (60200f6) into develop (3c93de8) will increase coverage by 0.0%. The diff coverage is 98.6%.

@@           Coverage Diff           @@
##           develop   #1109   +/-   ##
=======================================
  Coverage     88.9%   89.0%           
=======================================
  Files          376     373    -3     
  Lines        19569   19573    +4     
=======================================
+ Hits         17416   17420    +4     
  Misses        2153    2153           

[alexrlongne]

This should be ready for review again.