Topic/cuda aware communications#671

Open

bosilca wants to merge 18 commits intoICLDisco:masterfrom

bosilca:topic/cuda_aware_communications

Commits on Feb 13, 2026

Allow JDF with no dependencies, no datatype and no arenas.
bosilca
committed
Add a CUDA-based RTT test.
bosilca
committed
Transfer data to and from GPU.
bosilca
committed
Mostly improvement to the debuging output.
bosilca
committed
Correctly initialize an unlock atomic lock.
bosilca
committed
gpu-datain: when we are executing a CPU body we need to retrieve the CPU

abouteiller
authored and
bosilca
committed
gpu-datain: pure distribution collections have no data_of/data_of_key

abouteiller
authored and
bosilca
committed
Use lock initializers instead of static temps

abouteiller
authored and
bosilca
committed
gpu-datain: when the RO data-in is from D2D we need to increase its

abouteiller
authored and
bosilca
committed
gpu-datain: add a configure option and change some namings

abouteiller
authored and
bosilca
committed
When managing CPU-only tasks that received a GPU data copy, skip output-only flows, for which checking if they are control flows segfaults

therault
authored and
bosilca
committed
mpigpu: fix a number of crashes in release_self_contained_data

abouteiller
authored and
bosilca
committed
testing for dev_index==0 is not always true, use the mca_device_is_gpu

abouteiller
authored and
bosilca
committed
Use PRIx64 when printing a parsec_data_key_t

devreal
authored and
bosilca
committed
Add support for CUDA 13.
bosilca
committed
Allow command line arguments for the stress test.
bosilca
committed
Always propagate the CPU copy to successors.
bosilca
committed
Set the data_in to NULL before using the temporary task.
bosilca
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Topic/cuda aware communications#671

Topic/cuda aware communications#671
bosilca wants to merge 18 commits intoICLDisco:masterfrom
bosilca:topic/cuda_aware_communications

Commits on Feb 13, 2026

Allow JDF with no dependencies, no datatype and no arenas.

Add a CUDA-based RTT test.

Transfer data to and from GPU.

Mostly improvement to the debuging output.

Correctly initialize an unlock atomic lock.

gpu-datain: when we are executing a CPU body we need to retrieve the CPU

gpu-datain: pure distribution collections have no data_of/data_of_key

Use lock initializers instead of static temps

gpu-datain: when the RO data-in is from D2D we need to increase its

gpu-datain: add a configure option and change some namings

When managing CPU-only tasks that received a GPU data copy, skip output-only flows, for which checking if they are control flows segfaults

mpigpu: fix a number of crashes in release_self_contained_data

testing for dev_index==0 is not always true, use the mca_device_is_gpu

Use PRIx64 when printing a parsec_data_key_t

Add support for CUDA 13.

Allow command line arguments for the stress test.

Always propagate the CPU copy to successors.

Set the data_in to NULL before using the temporary task.