Landing CG on EARTH A case study of fine-grained multithreading on an evolutionary path

A1

A2

A3A4

q1

q2

q3q4

............11110000000011110000111100001111000011110000111100001111000011110000111100001111000011110000111100001111000011110000111100001111000011110000111100001111000011111100001100110011001100110011001100110011001100110011001100110011001111000011001100110011110000110011001100111100001100110011001111000011001100110011BROADSCATDATASYNCEDATADATASYNCDUCEDREUCREreadyeventqueuequeue

SU

PE

PE

..

PE

Interconnection NetworklocalmemoryEU

cache2520SpnDualExtsuIntsuRoundtrip Times15105012481632Number of Nodes64961207060SpnDualExtsuIntsu50Roundtrip Times40302010012481632Number of Nodes6496120Speedup65.0060.0055.0050.0045.0040.0035.0030.0025.0020.0015.0010.005.000.00No. of Processors0.0020.0040.0060.0080.00100.00120.00LinearSingleDualExtSuIntSuSpeedup100.0095.0090.0085.0080.0075.0070.0065.0060.0055.0050.0045.0040.0035.0030.0025.0020.0015.0010.005.000.00No. of Processors0.0020.0040.0060.0080.00100.00120.00LinearSingleDualExtSuIntSuSpeedup100.0095.0090.0085.0080.0075.0070.0065.0060.0055.0050.0045.0040.0035.0030.0025.0020.0015.0010.005.000.00No. of Processors0.0020.0040.0060.0080.00100.00120.00LinearDualExtSuIntSu