force disable Kokkos_ENABLE_IMPL_CUDA_MALLOC_ASYNC in tests

as it is now enabled by default in Kokkos 4.2
kokkos/kokkos@ac23f36

doc/python_requirements.txt

@@ -3,7 +3,7 @@
 # Doxygen https://www.doxygen.nl/download.html
 #
 # Install Sphinx and required plugins with the following command
-# python -m pip install -r python-requirements.txt
+# python -m pip install -r python_requirements.txt
 
 wheel>=0.38.4 # help forward compatibility for pip with old sphinx plugins
 sphinx==4.5.0

doc/source/faq.rst

@@ -50,6 +50,10 @@ How can I stop the code without loosing the current calculation?
 I'm doing performance measures. How do I disable all outputs in *Idefix*?
   Add ``-nowrite`` when you call *Idefix* executable.
 
+Using Cuda+MPI, I get an error ``The call to cuIpcGetMemHandle failed. This means the GPU RDMA protocol cannot be used. cuIpcGetMemHandle return value:   1``
+  This is because you are using an old version of MPI with UCX. You should disable ``Kokkos_ENABLE_IMPL_CUDA_MALLOC_ASYNC`` in cmake.
+
+
 
 Developement
 ------------

pytools/idfx_test.py

@@ -116,6 +116,8 @@ def configure(self,definitionFile=""):
       comm.append("-DKokkos_ENABLE_CUDA=ON")
       # disable fmad operations on Cuda to make it compatible with CPU arithmetics
       comm.append("-DIdefix_CXX_FLAGS=--fmad=false")
+      # disable Async cuda malloc for tests performed on old UCX implementations
+      comm.append("-DKokkos_ENABLE_IMPL_CUDA_MALLOC_ASYNC=OFF")
 
     if self.hip:
       comm.append("-DKokkos_ENABLE_HIP=ON")