Ahmad Nawab

Install python virtual environment with contributed ruamel.yaml

3

This PR contributes the installation of a python virtual environment that contains the contributed fypp and ruamel.yaml modules. A minimal `fckit_yaml_reader` module is also installed that creates an API consistent...

Block-index injection transformations

3

In order to mitigate the cost of creating thread-local copies of data structures, per-block views of fields can be replaced with full field views and the block-index can be inserted...

Currently, to protect from misaligned addresses on device, every allocation is padded to 8 bytes in the pool allocator. This would mitigate a lot of the device memory bandwidth benefits...

If module imports are used to define temporary array sizes deep in a call-tree, these variables might be undefined in intermediate layers when the temporaries are hoisted up to the...

Add unique ID to `Node`

2

Currently, the only mechanism for differentiating two string-identical but distinct nodes is the `.source` property. However this property is only instantiated for parsed nodes; if two identical but distinct nodes...

The driver loop pragma manipulations in the pool allocator transformation feel a little clumsy. For an ostensibly simple operation, the current implementation has a nested conditional and extensive use of...

Local scalar variables defined in vector parallel regions

1

Consider the following example: ``` subroutine some_kernel(flag1, flag2, ...) implicit none logical, intent(in) :: flag1, flag2 logical :: local_flag ... !$acc routine vector !$acc loop vector do ij=1,NPROMA local_flag =...

Skip privatization of arrays with existing data declarations

2

Arrays with existing data declarations should not be privatized in the driver gang loop. This PR adds the functionality to gather information from encompassing `!$acc data` regions and filter out...

Populate

1

This PR adds a populate script that can be used to cache python dependencies. The python venv installation instructions are updated accordingly as well as the README. Python version support...

Various fixes to SCC separator nodes

2

This PR includes the following fixes/updates to the identification of SCC separator nodes: - Vector inline calls, typically an array reduction intrinsic like `SUM`, are now marked as separator nodes...