[btl/smcuda] OMPI build fails when cuda is enabled

[AboorvaDevarajan]

Background information

OMPI version

master (git clone)

If you are building/installing from a git clone, please copy-n-paste the output from git submodule status.

 4a43c39c89037f52b4e25927e58caf08f3707c33 opal/mca/hwloc/hwloc2/hwloc (hwloc-2.1.0rc2-53-g4a43c39)
 b18f8ff3f62bd93f17e01c4230e7eca4a30aa204 opal/mca/pmix/pmix4x/openpmix (v1.1.3-2460-gb18f8ff)
 17ace07e7d41f50d035dc42dfd6233f8802e4405 prrte (dev-30660-g17ace07)

---

Details of the problem

OMPI CUDA enabled build fails,

./autogen.pl 
./configure --prefix=/nfs_smpi_ci/abd/pr/ompi-install --with-cuda=/usr/local/cuda
make -j 20

Error:

make[2]: Entering directory `/nfs_smpi_ci/abd/pr/ompi/opal/mca/btl/smcuda'
  CC       mca_btl_smcuda_la-btl_smcuda_component.lo
btl_smcuda_component.c: In function ‘mca_btl_smcuda_send_cuda_ipc_ack’:
btl_smcuda_component.c:632:9: warning: variable ‘rc’ set but not used [-Wunused-but-set-variable]
     int rc;
         ^
btl_smcuda_component.c: In function ‘mca_btl_smcuda_component_init’:
btl_smcuda_component.c:938:68: warning: assignment from incompatible pointer type [enabled by default]
     mca_btl_base_active_message_trigger[MCA_BTL_TAG_SMCUDA].cbfunc = btl_smcuda_control;
                                                                    ^
btl_smcuda_component.c: In function ‘mca_btl_smcuda_component_progress’:
btl_smcuda_component.c:1071:29: warning: passing argument 2 of ‘reg->cbfunc’ makes pointer from integer without a cast [enabled by default]
                             reg->cbdata);
                             ^
btl_smcuda_component.c:1071:29: note: expected ‘const struct mca_btl_base_receive_descriptor_t *’ but argument is of type ‘mca_btl_base_tag_t’
btl_smcuda_component.c:1071:29: error: too many arguments to function ‘reg->cbfunc’
make[2]: *** [mca_btl_smcuda_la-btl_smcuda_component.lo] Error 1
make[2]: Leaving directory `/nfs_smpi_ci/abd/pr/ompi/opal/mca/btl/smcuda'

Looks like the corresponding changes are missing in btl_smcuda here
#7913

[jsquyres]

@hjelmn Could you have a look?

[hjelmn]

Easy enough to fix. Missed this location. Should have a PR up this afternoon.

[bosilca]

This is more complicated than a simple sed, as the BTL and the PML are exchanging data in the descriptor.

[hjelmn]

I think the extra line setting Frag.hdr doesn't actually do anything. Double-checking now. If that is the case this is a simple sed.

[hjelmn]

I see what you mean. Will have to think about that.

[hjelmn]

Actually, the new callback makes the smcuda code cleaner. The endpoint is provided in the callback structure so the hdr is no longer needed. Will open a PR shortly.

[awlauria]

@AboorvaDevarajan can you verify that #7931 fixes this and close if it does? Thanks.

[AboorvaDevarajan]

This is fixed now. Closing it.