transformer_engine/common/fused_attn/utils.h · a150d286a76f6cd4e34b0e16e8b9cc11c4dba34c · OpenDAS / TransformerEngine

"git@developer.sourcefind.cn:yangql/googletest.git" did not exist on "b9347b31c338851879d0105f0fe32d09007f0433"

[C/PyTorch] Add workspace optimization for fused attention arbitrary_seqlen backend (#396) · a150d286

cyanguwa authored Sep 12, 2023



* add workspace optimization for arbitrary_seqlen fused attn
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix whitespace for lint
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* add use_workspace_opt to cudnn plan cache and fix workspace estimate
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* modify workspace opt logic; move zero fill to FP8 API only; other minor fixes
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix lint
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix try/catch
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* fix std string error when input is nullptr
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* remove comments
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>

* Add = for required vs allowed workspace comparison
Signed-off-by: cyanguwa <8636796+cyanguwa@users.noreply.github.com>

---------
Signed-off-by: Charlene Yang <8636796+cyanguwa@users.noreply.github.com>
Signed-off-by: cyanguwa <8636796+cyanguwa@users.noreply.github.com>

a150d286

utils.h 4.66 KB

Replace utils.h