GETRF NOPIV: Use explicit workspaces to control the communication flow

See merge request !307

Matthieu KUHN authored 2 years ago and Mathieu Faverge committed 5 months ago

getrf_nopiv: Add a getrf nopiv algorithm using lookhaed buffers to enforce a ring of communications and limit the number of ongoing communication though the dependencies.

Matthieu KUHN authored 2 years ago and Mathieu Faverge committed 5 months ago

getrf_nopiv: Call the option structure alloc/free functions in the getrf nopiv and gesv nopiv functions to switch from initial algorithm to the one with explicit communication management

Matthieu KUHN authored 2 years ago and Mathieu Faverge committed 5 months ago

getrf_nopiv: Add Alloc/Free functions to manage temporary buffers of the pzgetrf_nopiv algorithm to better control the MPI transfers overhead

Matthieu KUHN authored 2 years ago and Mathieu Faverge committed 5 months ago

compute_z.h: Add an option structure for the MPI getrf nopiv algorithm to manage explicitely the communications and enforce a lookahead turnover in the communications

plafrim benchmarks: avoid to get warned if regressions on small matrices (high variability)

See merge request !491

Enable insert_task_zlacpy_on_remote_node

See merge request solverstack/chameleon!483

Enable the official Gitlab - Sonarqube integration

See merge request !490

Mixed precision

See merge request !488

testing/parameter: Make sure we don't use mixed precision with global and out-of-core matrices allocations

lapacke_dlamch: Add a chameleon wrapper to encapsulate and refactor the test on CHAMELEON_SIMULATION

A small set of modifications to allow external applications to use the RUNTIME API

See merge request !475

starpu: Try to split what can be public if the user wants to add its own insert_task functions from what needs to remain private

Fix warning

See merge request !486