Remove usage of -DDISABLE_OPENACC (GPU build simplification)

[pramodk]

• always use Random123 via unified memory support
• some of the cpp files were avoiding it via -DDISABLE_OPENACC but with the addition of RANDOM construct, the Random123 streams are allocated during model setup and hence should be allocated via unified memory. So it's better to do this in one way everywhere (unless there is performance degradation).

See Olli's comment:

OL210813: this shouldn't be needed anymore, but it may have a small performance benefit

For the setup functions (under io/*) and two artificial cells, I don't think there can be a performance impact.

CI_BRANCHES:NMODL_BRANCH=master,

[azure-pipelines]

✔️ 2fd11a1 -> Azure artifacts URL

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (a0f1848) 66.19% compared to head (2fd11a1) 66.18%.

❗ Current head 2fd11a1 differs from pull request most recent head 47e79ba. Consider uploading reports for the commit 47e79ba to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2653      +/-   ##
==========================================
- Coverage   66.19%   66.18%   -0.02%     
==========================================
  Files         559      559              
  Lines      108937   108939       +2     
==========================================
- Hits        72109    72098      -11     
- Misses      36828    36841      +13     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[iomaganaris]

I must say I'm not familiar why we disabled OpenACC with those files but I don't see any issue enabling OpenACC/Unified memory.
Have you tested this branch with corenrn-perf CI in case there is any performance degradation?

[pramodk]

Have you tested this branch with corenrn-perf CI in case there is any performance degradation?

@iomaganaris. Yes - https://bluebrainproject.slack.com/archives/C032L6U24MB/p1704732829835149

I must say I'm not familiar why we disabled OpenACC with those files but I don't see any issue enabling OpenACC/Unified memory.

Here is the background:

• Before Olli introduced cuda unified memory support for Random123, we had Random123 streams allocated either on GPU or CPU.
• Artificial cells are executed always on the CPU. So you don't want to allocate Random123 streams on GPU. This is why we generate code with -DISABLE_OPENACC:
  https://github.com/BlueBrain/nmodl/blob/58bb7fd938aee2d0066eb3ba47159ad7fe9419fe/src/codegen/codegen_acc_visitor.cpp#L70
• When we set Random123 global seed, we use set_seed related API from nrn_setup() related code and hence we wanted those files to also not use the "CUDA/GPU version" of Random123 APIs.

Hence the initial introduction of the -DISABLE_OPENACC macro.

Today, we don't have this constraint - thanks to unified memory-related changes from Olli. But we thought if artificial cells are going to execute on CPU, why allocate those streams using CUDA unified memory? So we kept it as is.

With Michael changes from Random123, we are going to allocate Random123 streams from io/phase2.cpp. And then there was a bit of confusion between where we allocate Random123 streams with CUDA unified memory vs where it's plain malloc/new.

And just to avoid this confusion, I thought why not remove it and make it simple? (NMODL should also remove DISABLE_OPENACC but that could be separate update)

[azure-pipelines]

✔️ 47e79ba -> Azure artifacts URL

[azure-pipelines]

✔️ c5a31fd -> Azure artifacts URL

[sonarqubecloud]

Quality Gate passed

Kudos, no new issues were introduced!

0 New issues
0 Security Hotspots
No data about Coverage
0.0% Duplication on New Code

See analysis details on SonarCloud

[azure-pipelines]

✔️ e41dd09 -> Azure artifacts URL

[pramodk]

The result from the coreneuron perf CI pipeline with this neuron + nmodl branches:

[bbpbuildbot]

Logfiles from GitLab pipeline #188727 (:white_check_mark:) have been uploaded here!

Status and direct links:

• ✅ mac_m1_cmake_build: [cmake, ON, OFF, OFF, address]
• ✅ spack_setup
• ✅ build:nmodl
• ✅ build:neuron:nmodl:intel:legacy
• ✅ build:neuron:nmodl:intel:shared
• ✅ build:neuron:nmodl:nvhpc:acc:legacy
• ✅ build:neuron:nmodl:nvhpc:acc:shared
• ✅ build:neuron:nmodl:nvhpc:omp:legacy
• ✅ build:neuron:nmodl:nvhpc:omp
• ✅ test:neuron:nmodl:intel:legacy
• ✅ test:neuron:nmodl:intel:shared
• ✅ test:neuron:nmodl:nvhpc:acc:legacy
• ✅ test:neuron:nmodl:nvhpc:acc:shared
• ✅ test:neuron:nmodl:nvhpc:omp:legacy
• ✅ test:neuron:nmodl:nvhpc:omp