Niketan Pansare created SYSTEMML-933:
----------------------------------------
Summary: Chose the block dimensions of custom kernels using CUDA's
occupancy API
Key: SYSTEMML-933
URL: https://issues.apache.org/jira/browse/SYSTEMML-933
Project: SystemML
Issue Type: Improvement
Reporter: Niketan Pansare
See
https://devblogs.nvidia.com/parallelforall/cuda-pro-tip-occupancy-api-simplifies-launch-configuration/.
Modify org.apache.sysml.runtime.instructions.gpu.context.ExecutionConfig's
getConfigForSimpleMatrixOperations and getConfigForSimpleVectorOperations
methods
Test case before merging (try on different CUDA versions):
org.apache.sysml.test.integration.functions.binary.matrix_full_other
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)