20.11.2014 Views

PPKE ITK PhD and MPhil Thesis Classes

PPKE ITK PhD and MPhil Thesis Classes

PPKE ITK PhD and MPhil Thesis Classes

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

2.3 Computational Fluid Flow Simulation on Body Fitted Mesh Geometry<br />

with IBM Cell Broadb<strong>and</strong> Engine <strong>and</strong> FPGA Architecture 63<br />

cells of different layers are connected to the cell of layer ρu at position (i, j).<br />

⎡<br />

A ρu<br />

1 = 1 ⎣<br />

2∆x<br />

A ρu<br />

2 = 1<br />

2∆x<br />

⎡<br />

A ρu<br />

3 = 1 ⎣<br />

2∆x<br />

0 0 0<br />

ρu 2 + p 0 −(ρu 2 + p)<br />

0 0 0<br />

⎡<br />

⎣<br />

0 −ρuv 0<br />

0 0 0<br />

0 ρuv 0<br />

⎤<br />

0 ρv 0<br />

ρu −2ρu − 2ρv ρu<br />

0 ρv 0<br />

⎤<br />

⎦ (2.18)<br />

⎦ (2.19)<br />

⎤<br />

⎦ (2.20)<br />

The template values for ρ, ρv <strong>and</strong> E layers can be defined analogously.<br />

The ρuu + p, ρuv, ρu <strong>and</strong> ρv terms can be reused during the computation of<br />

the neighboring cells <strong>and</strong> they should be computed only once in each iteration<br />

step. This solution requires additional memory elements but greatly reduces the<br />

area requirement of the arithmetic unit.<br />

Other trick can be applied if we choose ∆t to be integer power of two because<br />

the multiplication in this case can be done by shifts so we can eliminate several<br />

multipliers from the hardware <strong>and</strong> additionally the area requirements will be<br />

greatly reduced.<br />

In the second-order case limiter function should be used on the primitive<br />

variables <strong>and</strong> the conservative variables are computed from these results. The<br />

limited values will be different for the four interfaces <strong>and</strong> cannot be reused in<br />

the computation of the neighboring cells.<br />

Therefore, this approach does not<br />

make it possible to derive CNN templates for the solution. However a specialized<br />

arithmetic unit still can be designed to solve it directly.<br />

2.3.2.7 Results <strong>and</strong> performance<br />

The previous results solving the Euler equations on a rectangular grid shows that<br />

only few specialized arithmetic unit can be implemented even on the largest FP-<br />

GAs. In the body fitted case additional area is required to take into account<br />

the geometry of the mesh during the computation. Symmetrical nature of the<br />

problem, which can be seen on the templates (2.18)-(2.20), enable further optimization<br />

of the arithmetic unit which compensates the area increase due to

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!