Enable tensor ukernels by default
This enables the MLIR tensor ukernels by default as they provide the best performance. All benchmark/production workloads I am aware of seem to set this flag to true already, so this should reduce the number of flags needed and avoid sudden regressions when people forget to set/carry the flag.
I will follow up with PRs in other repos that remove the usage of this flag.
@MaheshRavishankar Do we want to run an additional (nightly) test suite on this to make sure it's stable?
ci-extra: test_torch
@jtuyls I added a ci-extra trailer so the torch ci will run on this again and retriggered the ci. I'd recommend disabling the ukernel flags (if enabled) in the ci compiler flags to check if everything works as intended.
@jtuyls I added a ci-extra trailer so the torch ci will run on this again and retriggered the ci. I'd recommend disabling the ukernel flags (if enabled) in the ci compiler flags to check if everything works as intended.
Cool, thanks!
This enables the MLIR tensor ukernels by default as they provide the best performance. All benchmark/production workloads I am aware of seem to set this flag to true already, so this should reduce the number of flags needed and avoid sudden regressions when people forget to set/carry the flag.
I will follow up with PRs in other repos that remove the usage of this flag.
@MaheshRavishankar Do we want to run an additional (nightly) test suite on this to make sure it's stable?
ci-extra: test_torch
I think as you said most places already use this, but let me ping you internally how to run some additional tests on this.
Reminder: please add the message to the release tracker: https://github.com/iree-org/iree/issues/22307