Fix gpuudpate and gpucomm matrix docs

Improves the outdated documentation/description of the matrices
clarifying the use/role of these.

Also removed gpubufferops matrix which is not needed in the release
branch (the feature is automatically triggered when required by some
other features).

Change-Id: I097e1d6b150767a95b9d638b3c6031a24ece995f

diff --git a/admin/builds/gpubufferops-matrix.txt b/admin/builds/gpubufferops-matrix.txt
deleted file mode 100644 (file)
index bab7e2a..0000000
--- a/admin/builds/gpubufferops-matrix.txt
+++ /dev/null
@@ -1,38 +0,0 @@
-# This matrix is intended to permit Jenkins on-demand testing
-# of code hidden behind the GMX_USE_BUFFER_OPS feature flag
-# during development. When the feature flag is removed, the
-# normal test matrices will be adapted to cover this code path.
-#
-# Comment line(s) preceding each configuration document the main
-# intent behind that configuration, so that we can correctly judge
-# whether to preserve that during maintenance decisions.
-#
-# Both configurations currently target bs_nix1204, for better load
-# balance with pre-submit matrix, which makes heavier use of
-# bs_nix1310 agent.
-
-# Test an older version of CUDA
-# Test MPI with CUDA
-# Test MPMD PME with library MPI
-# Test gpubufferops features in the above combination
-gcc-5 gpuhw=nvidia cuda-9.0 mpi npme=1 nranks=2 openmp gpubufferops
-
-# Test newest gcc supported by newest CUDA at time of release
-# Test thread-MPI with CUDA
-# Test gpubufferops features in the above combination
-gcc-8 gpuhw=nvidia nranks=1 gpu_id=1 cuda-10.1 thread-mpi openmp cmake-3.10.0 release-with-assert simd=avx2_256 hwloc libhwloc-2.0.4 gpubufferops
-
-# Test non-default use of mdrun -gpu_id
-# Test GPU-sharing among 4 PP ranks
-# Test no hwloc build, tests internal CPU topology detection (mainly for x86)
-# Test gpubufferops in the above combination
-gcc-7 gpuhw=nvidia nranks=4 gpu_id=1 cuda-10.0 no-hwloc release-with-assert gpubufferops
-
-# Test CUDA build on a agent with no CUDA devices
-# Test without TNG support
-# Test gpubufferops in the above combination
-gcc-7 gpuhw=none cuda-10.0 openmp no-tng release-with-assert gpubufferops
-
-# Test OpenCL build with gpudev features
-# Test gpubufferops on the OpenCL path where it is unsupported
-clang-8 openmp gpuhw=amd opencl-1.2 clFFT-2.14 simd=None gpubufferops

diff --git a/admin/builds/gpucomm-matrix.txt b/admin/builds/gpucomm-matrix.txt
index ef7895fe30e6fedff9ebcff4367ec8d53943148f..644334dba5ba20d9cb8d8721482efa4a0e0342ba 100644 (file)
--- a/admin/builds/gpucomm-matrix.txt
+++ b/admin/builds/gpucomm-matrix.txt
@@ -1,7 +1,10 @@
-# This matrix is intended to permit Jenkins on-demand testing
-# of code hidden behind the GMX_USE_BUFFER_OPS feature flag
-# during development. When the feature flag is removed, the
-# normal test matrices will be adapted to cover this code path.
+# This matrix is intended to permit Jenkins:
+# - nigtly testing of the GPU direct communations experimental features in the 2020 release
+# - on-demand testing of the GPU direct communication code path
+# This is neceassary due to:
+# i) the limitations of the integration tests which only test the deafult code-path
+# ii) limitations of the regresssion tests which are not able to rerun update with a
+#     different offload mode unless the default is observed to be GPU.
 #
 # Comment line(s) preceding each configuration document the main
 # intent behind that configuration, so that we can correctly judge

diff --git a/admin/builds/gpuupdate-matrix.txt b/admin/builds/gpuupdate-matrix.txt
index cd0c31b09e1a3c09c7ead449fbb3981a200ffc80..f31acc2e34565c31af36dbe76e6eb6423302cbbb 100644 (file)
--- a/admin/builds/gpuupdate-matrix.txt
+++ b/admin/builds/gpuupdate-matrix.txt
@@ -1,5 +1,10 @@
-# This matrix is intended to permit Jenkins on-demand testing
-# of GPU update code path during development.
+# This matrix is intended to permit Jenkins:
+# - nigtly testing of the GPU update features in the 2020 release
+# - on-demand testing of the GPU update code path
+# This is neceassary due to:
+# i) the limitations of the integration tests which only test the deafult code-path
+# ii) limitations of the regresssion tests which are not able to rerun update with a
+#     different offload mode unless the default is observed to be GPU.
 #
 # Comment line(s) preceding each configuration document the main
 # intent behind that configuration, so that we can correctly judge