Release 0.99.3

[pelesh]

Description

This is final PR before 0.99.3 release (first beta).

Proposed changes

This PR only updates version number. Please check that all the documentation is (reasonably) accurate, all tests pass and that examples work correctly on all supported architectures.

Checklist

• All tests pass (make test and make test_install per testing instructions). Code tested on
  • CPU backend
  • CUDA backend
  • HIP backend
• I have manually run the non-experimental examples and verified that residuals are close to machine precision. (In your build directory run: ./examples/<your_example>.exe -h to get instructions how to run examples). Code tested on:
  • CPU backend
  • CUDA backend
  • HIP backend
• Code compiles cleanly with flags -Wall -Wpedantic -Wconversion -Wextra.
• The new code follows Re::Solve style guidelines.
• There are unit tests for the new code.
• The new code is documented.
• The feature branch is rebased with respect to the target branch.
• I have updated CHANGELOG.md to reflect the changes in this PR. If this is a minor PR that is part of a larger fix already included in the file, state so.

Further comments

[shakedregev]

Looks good

[andrewxu319]

Tested the following on both Frontier (CPU & HIP) and a local CPU-only build:

• All tests pass
• All examples (including experimental) work fine with the ACTIVSg2000 series systems
• make test_install passes

[tamar-dewilde]

I also tested CPU, CUDA, and HIP at a8f8995. The automated test suites and test_install passed on all three (CPU 60/60, CUDA 67/67, HIP 64/64).

With the ACTIVSg10k systems, I saw a couple of additional findings:

• Across CPU, CUDA, and HIP, some manual examples did not reach near-machine precision. randGmres/sysGmres reached 2500 iterations with residuals around 7e-3, and kluFactor system 2 gave residuals of 6.17e-5 on CPU and CUDA and 1.04e-4 on HIP.
• On HIP, the refactor examples with iterative refinement hit the Gram-Schmidt zero-norm assertion on system 2; the corresponding runs without IR completed.

The strict warning builds also produced warnings; details are included in the attached logs.

Full validation logs:

• cpu-resolve-0993-validation.log
• cuda-resolve-0993-validation.log
• hip-release-0993-validation.log

[pelesh]

I also tested CPU, CUDA, and HIP at a8f8995. The automated test suites and test_install passed on all three (CPU 60/60, CUDA 67/67, HIP 64/64).

With the ACTIVSg10k systems, I saw a couple of additional findings:

• Across CPU, CUDA, and HIP, some manual examples did not reach near-machine precision. randGmres/sysGmres reached 2500 iterations with residuals around 7e-3, and kluFactor system 2 gave residuals of 6.17e-5 on CPU and CUDA and 1.04e-4 on HIP.
• On HIP, the refactor examples with iterative refinement hit the Gram-Schmidt zero-norm assertion on system 2; the corresponding runs without IR completed.

The strict warning builds also produced warnings; details are included in the attached logs.

The example runs results are what we expected. We need nicer matrices for GMRES solver examples.

I guess we need to fix some compiler warnings before we can release. I think we can ignore CUDA deprecation warnings nudging us to use cuDSS instead of cuSOLVER.

[tamar-dewilde]

Small thing, but I noticed a typo in my last name. It should be DeWilde with a capital W. I noticed it was changed in the README too.