Download presentation

Presentation is loading. Please wait.

Published bySharon Brumitt Modified over 3 years ago

1
Daniel OrozcoDaniel Orozco Guang GaoGuang Gao

2
Mapping FDTD to Many-Cores ------- Daniel Orozco2

3
Mapping FDTD to Many-Cores Daniel Orozco3 FDTD = Finite Difference Time Domain FDTD simulates the propagation of electromagnetic waves through materials.

4
Mapping FDTD to Many-Cores ------- Daniel Orozco4

5
5

6
6

7
7

8
8 Read Create New Overwrite

9
Mapping FDTD to Many-Cores ------- Daniel Orozco9 Memory Loads Per Element Computed: 9 Memory Loads Per Element Computed: 1.44

10
Mapping FDTD to Many-Cores ------- Daniel Orozco10

11
Mapping FDTD to Many-Cores ------- Daniel Orozco11

12
Mapping FDTD to Many-Cores ------- Daniel Orozco12

13
Mapping FDTD to Many-Cores ------- Daniel Orozco13

14
Mapping FDTD to Many-Cores ------- Daniel Orozco14

15
Mapping FDTD to Many-Cores ------- Daniel Orozco15

16
Logical View Tile shape Mapping FDTD to Many-Cores ------- Daniel Orozco16 No Lost Computations Tiles are fully parallel. No lost computations. Very simple programming.

17
Logical View Mapping FDTD to Many-Cores ------- Daniel Orozco17 Tiles are fully parallel. No lost computations. Useful to understand the problem.

18
Mapping FDTD to Many-Cores ------- Daniel Orozco18 Reuse = 40 Reuse = 5

19
Mapping FDTD to Many-Cores ------- Daniel Orozco19 No Tiling Simple Tiling Skewed Tiling Overlapped Tiling Split Tiling Triangle Tiling Parametric Tiling p = 0.5 Diamond Tiling Not Embarrassingly Parallel Developed at CAPSL The Fine Print: Values are for a tile size of 100. Reuse values change with the size of the tile. Results apply to 1 Dimensional Stencil Computation with dependencies similar to those of the examples.

20
Mapping FDTD to Many-Cores ------- Daniel Orozco20 No Tiling Triangle Size = 16 Triangle Size = 64 Diamond Size = 64 Diamond Size = 16 The Fine Print: Simulated Speedup Results for FDTD 1D running on Cyclops-64 using FAST simulator. Problem size varies for each test, and was selected as big as possible. Only the computation time was measured. Problem data located in DRAM. Tiling done manually. GCC 3.4, -O3 used.

21
Mapping FDTD to Many-Cores ------- Daniel Orozco21 Diamond Size = N Parametric Size = N Low Reuse High Reuse Low Reuse High Reuse

22
Mapping FDTD to Many-Cores ------- Daniel Orozco22

23
Mapping FDTD to Many-Cores ------- Daniel Orozco23

24
Mapping FDTD to Many-Cores ------- Daniel Orozco24

25
Mapping FDTD to Many-Cores ------- Daniel Orozco25

26
Mapping FDTD to Many-Cores ------- Daniel Orozco26

Similar presentations

OK

Hybrid Parallel Implementation of The DG Method Advanced Computing Department/ CAAM 03/03/2016 N. Chaabane, B. Riviere, H. Calandra, M. Sekachev, S. Hamlaoui.

Hybrid Parallel Implementation of The DG Method Advanced Computing Department/ CAAM 03/03/2016 N. Chaabane, B. Riviere, H. Calandra, M. Sekachev, S. Hamlaoui.

© 2017 SlidePlayer.com Inc.

All rights reserved.

Ads by Google