Skip to content

Fix statistics management with multiple GPU.#454

Merged
bosilca merged 1 commit intoICLDisco:masterfrom
bosilca:fix/multi_gpu_446
Nov 10, 2022
Merged

Fix statistics management with multiple GPU.#454
bosilca merged 1 commit intoICLDisco:masterfrom
bosilca:fix/multi_gpu_446

Conversation

@bosilca
Copy link
Copy Markdown
Contributor

@bosilca bosilca commented Nov 10, 2022

The data_in_array_size was initialized using parsec_nb_devices, but that variable is not constant until all the devices have been added to a context. Thus, each module was having statistics arrays with different sizes. This patch delays the statistics arrays allocation until the end of the device registration, a moment where we know how many devices are overall.

Fixes #446

@bosilca bosilca added the blocker Blocking release or critical use case label Nov 10, 2022
@bosilca bosilca added this to the v4.0 milestone Nov 10, 2022
@bosilca bosilca requested a review from a team as a code owner November 10, 2022 20:45
The data_in_array_size was initialized using parsec_nb_devices, but that
variable is not constant until all the devices have been added to a
context. Thus, each module was having statistics arrays with different
sizes. This patch delays the statistics arrays allocation until the end
of the device registration, a moment where we know how many devices are
overall.

Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
@bosilca bosilca merged commit 3d78d8c into ICLDisco:master Nov 10, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

blocker Blocking release or critical use case

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Multi GPU dplasma GEMM runs fail on master branch

3 participants