enforce-baseline-decomposition
enforce-baseline-decomposition
Use the same local size
and SIMD width as measured on a baseline device.
Syntax
--enforce-baseline-decomposition
--no-enforce-baseline-decomposition
Default
Off (no-enforce-baseline-decomposition)
Actions Modified
collect
=projection
--profile-gpu
Usage
This option is applicable only to the GPU-to-GPU
performance modeling workflow.
When used with the Performance Modeling (
collect=projection
) as part of
the GPU Roofline Insights perspective for an application executed on a GPU,
your application performance is modeled for a baseline GPU device as a target.
The estimated performance is compared with the actual application performance
to add more recommendations for performance optimization.
This option is recommended to be used with the
model-baseline-gpu
option.
Example
Run the GPU-to-GPU Performance Modeling for the
baseline GPU using the same local size and SIMD width as on the baseline
device.
advisor --collect=projection --profile-gpu --model-baseline-gpu --enforce-baseline-decomposition --project-dir=./advi_results -- ./myApplication