Phylanx Report: October 2019

The twenty-sixth month of work focused on work in the following areas:

• Improvement to Traveler-Integrated
• Additional support for 4D operations
• Work on Lanczas solver

Organizational Activities and Results

• Weekly group-meeting, minutes:
  • October 9th: https://drive.google.com/open?id=1jssSX0tZGY4caTjjjIrzo_a3faF_Mds9
  • October 16th: https://drive.google.com/open?id=1idfNESBZkhBT1WSaolVJPiXw65w_8lWZ
  • October 23rd: https://drive.google.com/open?id=1QIfnqgjqWRtEZgdAWixAXN4cvZWkI9ZO

Development Activities and Results

• Traveler
  • Traveler-Integrated now shows Parent-Child relationships among tasks
    • Clicking on a task will show a traceback of relationships
  • Traveler-Integrated backend now reading PAPI metrics from OTF2
  • UI/Bug Fixes:
    • Addressed bugs when clicking on Gantt chart selections in Traveler-Integrated
    • Auxiliary lines showing in Execution Trees (Atria) for Traveler-Integrated
• Tiling
  • Theory:
    • Worked on the k-server problem
      • Looking at trees and cycle metrics
  • Implementation
    • Wrote a version of the Cannon’s matrix multiplication algorithm
• Primitives & Algorithms
  • The following features were added to Phylanx:
    • Added 4D support to squeeze operation (PR #1043)
    • Added conv2d operation (PR #1042)
    • Added 4D support for softmax operation (PR #1048)
    • Added Conv2d_transpose operation (PR #1051)
    • Lanczos solver was also added (PR #1050)
  • The following improvements were made to hpxMP
    • Add omp API for omp set get nested which is need by mkl (PR #58)
    • Fix building when build with ‘-DHPX_WITH_HPXMP=ON’ in HPX (PR #57)
    • fix intrusive ptr without template argument complain generated by clang8 (PR #56)
• Python
  • Enabling assignment to dtype
  • Add variable _keras_shape attribute
  • Improve the translation between primitives and Python objects
  • Simplify defining lazy functions

Repository Activity October 1st – October 31st:

Code Statistics:

-------------------------------------------------------------------------------

Language                     files          blank        comment           code

-------------------------------------------------------------------------------
C++                            479          18978           6241          97175
C/C++ Header                   313           6993           4407          34196
Python                         142           2671           1829           6942
CMake                          109            958            794           4057
YAML                             2             24             77            669
Markdown                         2             17              0             46
Dockerfile                       2              6             28             45
CSS                              1             14             10             38
HTML                             2              5              6             22
DOS Batch                        2              0              0             12
INI                              1              0              0              8
-------------------------------------------------------------------------------
SUM:                          1055          29666          13392         143210
-------------------------------------------------------------------------------

Impact on Other Projects

• Blaze Pull Requests
  • Excluded CUDA assignable structures from existing smpAssign strategies (PR #37)
• HPX Pull Requests:
  • Adding basic support for stackless threads (PR #4145)
  • Stackless coroutines now can refer to themselves (through get_self() and friends) (PR #4156)
  • Using target_link_libraries for cmake versions >= 3.12 (PR #4160)
• Blaze_CUDA:
  • Fixed cudaAssign() backend interface for matrices (Commit 752b63b)
  • CUDA backend interface should be complete now (Commit 16380e1)
  • Added output iterator adapter for blaze/thrust interoperability (Commit 8de4bc0)