1. Dataflow Model of Computation (From Dataflow to Multithreading)
Guang R. Gao
ACM Fellow and IEEE Fellow
Endowed Distinguished Professor
Electrical & Computer Engineering
University of Delaware
652-11S-Topic-Gao-Dataflow

2. Fine-Grain non-preemptive thread-
CPU
Memory
Executor
Locus
A Single
Thread
Coarse-Grain thread-
The family home model
Unit
Executor
Locus
CPU
Memory
Fine-Grain non-preemptive thread-
The “hotel” model
Thread
Unit
A Pool
Thread
Coarse-Grain vs. Fine-Grain Multithreading
[Gao: invited talk at Fran Allen’s Retirement Workshop, 07/2002]
652-11S-Topic-Gao-Dataflow

3. Evolution of Multithreaded Execution and Architecture Models
CHoPP’77
CHoPP’87
Non-dataflow
based
MASA
Halstead
1986
Alwife
Agarwal
Eldorado
HEP
B. Smith
1978
Tera
B. Smith
1990-
CDC 6600
1964
CASCADE
Flynn’s
Processor
1969
Cosmic Cube
Seiltz
1985
J-Machine
Dally
M-Machine
Dally
Others: Multiscalar (1994), SMT (1995), etc.
Monsoon
Papadopoulos
& Culler
1988
Dataflow
model inspired
P-RISC
Nikhil &
Arvind
1989
*T/Start-NG
MIT/Motorola
1991-
MIT TTDA
Arvind
1980
Cilk
Leiserson
LAU
Syre
1976
Iannuci’s
TAM
Culler
1990
Manchester
Gurd & Watson
1982
SIGMA-I
Shimada
1988
EM-5/4/X
RWC-1
Static
Dataflow
Dennis 1972
MIT
Arg-Fetching
Dataflow
DennisGao
MDFA
Gao
MTA
HumTheobald
Gao 94
EARTH CARE
PACT95’,
ISCA96,
Theobald99
Marquez04
652-11S-Topic-Gao-Dataflow

4. The Von Neumann-type Processing
begin
for i = 1 … …
endfor
end
Compiler
Sequential Machine Representation
Source Code
CPU
Load
Processor
652-11S-Topic-Gao-Dataflow

5. A Multithreaded Architecture
To Other PE’s
One PE
652-11S-Topic-Gao-Dataflow

6. McGill Data Flow Architecture Model (MDFA)
652-11S-Topic-Gao-Dataflow

7. 652-11S-Topic-Gao-Dataflown1 n1
store
store
store
fetch
fetch
fetch n2
fetch n3 n2 n3
Argument –flow Principle
Argument –fetching Principle
652-11S-Topic-Gao-Dataflow

8. A Dataflow Program Tuple
Program Tuple = { P-Code . S-Code }
S-Code
P-Code 2 3 n1 a b n2 c d
N1: x = a + b;
N2: y = c – d;
N3: z = x * y;
IPU
ISU
652-11S-Topic-Gao-Dataflow

9. The McGill Dataflow Architecture Model
Pipelined Instruction Processing Unit (PIPU)
Fire
Done
Dataflow Instruction Scheduling Unit (DISU)
Enable Memory & Controller
Signal Processing
652-11S-Topic-Gao-Dataflow

10. The McGill Dataflow Architecture Model
Pipelined Instruction Processing Unit (PIPU)
Important Features
Fire
Done
Dataflow Instruction Scheduling Unit (DISU)
Pipeline can be kept fully utilized provided that the program has sufficient parallelism
= PC
Enabled Instructions
Waiting Instructions
652-11S-Topic-Gao-Dataflow

11. The Scheduling Memory (Enable)
Dataflow Instruction Scheduling Unit (DISU) C O N T R L E 1
Signal Processing
Fire
Done
Count Signal(s) 1
Enabled Instructions
Waiting Instructions
652-11S-Topic-Gao-Dataflow

12. Advantages of the McGill Dataflow Architecture Model
Eliminate unnecessary token copying and transmission overhead
Instruction scheduling is separated from the main datapath of the processor (e.g. asynchronous, decoupled)
652-11S-Topic-Gao-Dataflow

13. Von Neumann Threads as Macro Dataflow Nodes1 2 3 k
A sequence of instructions is “packed” into a macro-dataflow node
Synchronization is done at the macro-node level
652-11S-Topic-Gao-Dataflow

14. 652-11S-Topic-Gao-Dataflow
Hybrid Evaluation Von Neumann Style Instruction Execution” on the McGill Dataflow Architecture
Group a “sequence” of dataflow instruction into a “thread” or a macro dataflow node.
Data-driven synchronization among threads.
“Von Neumann style sequencing” within a thread.
Advantage:
Preserves the parallelism among threads but avoids unnecessary fine-grain synchronization between instructions within a sequential thread.
652-11S-Topic-Gao-Dataflow

15. 652-11S-Topic-Gao-Dataflow
What Do We Get?
A hybrid architecture model without sacrificing the advantage of fine-grain parallelism!
(latency-hiding, pipelining support)
652-11S-Topic-Gao-Dataflow

16. A Realization of the Hybrid Evaluation
Shortcut
Pipelined Instruction Processing Unit (PIPU)
Von Neumann bit 1 2 k
Fire
Done
Dataflow Instruction Scheduling Unit (DISU)
652-11S-Topic-Gao-Dataflow

17. Case Studies – Dataflow Model Insired Multithreading
McGill Dataflow Model ( )
EARTH Model (1993 – mid 2000s )
The UHPC/Runnemede Model ( )
5/14/2019
F/Topic-3-II-FineGrain-Cases