submit cactuBSSN for SPEC v8 CPU benchmark

[Roland Haas]

• deadline: Feb 2021
• competition page: https://www.spec.org/cpuv8

Steps required:

• Step 1: Complete the legal paperwork. Submit a Search Program Proposal to SPEC through the entry form found at: https://www.spec.org/cpu/cpuv8/entry_form.html including:

  • Submitter's name and contact information (email address and shipping address),
  • Name of the application,
  • Programming(s) language used and approximate lines of code,
  • A description of what the application does and how it is used, and
  • A statement declaring that the Submitter has the authority to give SPEC the right to distribute the application as part of a SPEC benchmark suite.

• Step 2 Initial porting and workload ($1500 upon successful completion)

• Step 3 Provide workloads and demonstrate profile ($1500 or $2500 upon successful completion)
• Step 4 Complete Code Testing ($1500 or $2000 upon successful completion)
• Step 5 Recommendation to the Open System Group (no monetary award)
• Step 6 Acceptance into the SPEC CPU Suite ($3000 if accepted)

Malcolm Tobias submitted an ADM code for SPEC CPU 2006: https://www.spec.org/cpu2006/Docs/436.cactusADM.html

Erik, Gab, Peter, and Jian Tao submitted a BSSN code for SPEC CPU 2017: https://www.spec.org/cpu2017/Docs/benchmarks/507.cactuBSSN_r.html https://www.spec.org/cpu2017/Docs/benchmarks/607.cactuBSSN_s.html

SPEC CPU benchmarks are widely used in both academia and industry and have a broad impact in many fields. I hope we could keep Cactus in the SPEC benchmark suite.

Getting ADM into SPEC took several years (http://cactuscode.org/pipermail/users/2006-August/001962.html)

[Roland Haas]

Freely distributable part of the code from last time

[Roland Haas]

@Steven R. Brandt submitted paperwork. Next step is to be accepted by SPEC.

[Roland Haas]

@Steven R. Brandt was contacted by the SPEC team, who seem willing to accept another Cactus / Einstein Toolkit test but would like to not use the very same one as before again.

[Roland Haas]

For the SPECv8 contribution we had at one point mentioned 3 options
that we could propose:

• McLachlan's ccZ4 formulation (in master)
• McLachlan's Z4c formulation (in rhaas/Z4c)
• a version of McLachlan that pre-computes all derivatives as grid
  functions the way that the CUDA code does (in
  rhaas/separatedderivatives)

Please use this form:

https://docs.google.com/forms/d/e/1FAIpQLSdtnvxu-PA5gxY7o-fFWARPFOby8joz4-nC51ru_t3oKlJT7w/viewform?usp=sf_link

to voice preference for one or the other.

At this point none have been extracted and incorporated into the
min-driver used for SPEC.

[Roland Haas]

The attached thornlist and diff for McLachlan (removes LoopControl and some other things we do not want to support) may serve as a starting point. I left in the Scalar output but one could remove IOScalar and LocalReduce and PUGHReduce most likely

[Roland Haas]

Assigning to Steve so that both I (reporter) and Steve (assignee) may receive notifications for comments (or so I hope).

[Roland Haas]

upload corrected thornlist with LoopControl and CycleClock removed.

[Roland Haas]

A tarball using the updated thornlist and a standalone makefile.

[Roland Haas]

A version of the tarball containing CCZ4 and ASCII clean source files.

[Roland Haas]

An initial version using CCZ4, 8th order FD, and ShiftedGaugeWave as initial data.

[Roland Haas]

More parameters hard-coded and some IfThen removed from runtime. Better support for dependency tracking in makefile.