Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
uvahotspot
GitHub Repository: uvahotspot/HotSpot
Path: blob/master/examples/example6/extra_information.txt
612 views
1
i7. HotFloorplan:
2
3
a) As mentioned above, one of the inputs to HotFloorplan is a
4
floorplan description file that has information about the
5
blocks to be floorplanned. The maximum aspect ratio constraint is set
6
to 1:3 except when a block has a bigger aspect ratio in the
7
base floorplan. In that case, the block's aspect
8
ratio itself forms the upper limit. Also,
9
the connectivity section of 'ev6.desc' currently lists 13
10
major interconnects that we thought were important at the
11
architecture level. They are assumed to be of equal wire
12
density. Also, the wire_density field of the floorplan
13
description can be used to assign weights for the different
14
wires according to their contribution to performance. In
15
'ev6.desc', the assignment of these weights is uniform
16
across all wires.
17
b) HotFloorplan includes a first-order wire-delay model for
18
wires in the global and intermediate metal layers. It is
19
a simple stand-alone model that converts wire length to
20
wire delay and can be used in any tool, whether the thermal
21
model, floorplan model, or any other simulator. It is
22
adapted from Otten and Brayton, DAC'98, and related work.
23
The interfaces for the model are provided 'wire.[ch]'. An
24
appropriate process technology node can be chosen by using
25
the TECHNODE #define in 'wire.h'. After that, the model is
26
a simple call to the function 'wirelength2delay'.
27
c) HotFloorplan can be configured through many command line options.
28
The 'hotspot.config' file clearly lists them and what they do.
29
Among the options, following merit further discussion:
30
i) Annealing parameters:
31
The meaning of each of these can be easily obtained from any
32
standard text on VLSI CAD that deals with simulated
33
annealing (e.g.: Sarrafzadeh and Wong, "An Introduction to
34
VLSI Physical Design", McGraw-Hill 1996). The original
35
DAC '86 paper by Wong and Liu on floorplanning is a very good
36
starting point.
37
ii) Compaction ratio:
38
During floorplanning, empty spaces can arise in a floorplan.
39
Sometimes, these spaces are just an artifact of the discrete
40
nature of the aspect ratio function of the individual
41
blocks. Typically, about 50% of the maximum possible number
42
of dead spaces is due to this artifact. Such dead spaces
43
are very tiny and hence can be safely ignored without any
44
impact on the thermal model. In fact, ignoring such blocks
45
reduces the size of the problem input to the thermal solver.
46
Since the thermal model's solver algorithm is a higher order
47
polynomial function in the number of functional blocks,
48
ignoring the tiny blocks improves the performance of the
49
floorplanner significantly. The 'compact_ratio' option
50
sets a threshold on the area of such tiny blocks. The
51
default is 0.5% of the area of the encompassing 'parent
52
rectangle'. So, dead spaces that are smaller than 0.5% of
53
the parent rectangle are ignored during floorplanning. This
54
results in a maximum error in core area of about 0.46% for
55
our experiments. This setting, combined with the 'n_orients'
56
setting that determines the magnitude of discreteness of
57
the blocks' aspect ratio function, can be tuned to manage
58
the trade-off between the speed and accuracy of the
59
floorplanner. It is to be noted that in HotFloorplan, the
60
empty spaces that are not compacted are named in the form
61
_0, _1 and so on. It can be seen that the 'output.flp'
62
generated from the steps discussed above has blocks named
63
in that form.
64
iii) Weights for the metric (objective function):
65
HotFloorplan uses an objective function that is of the form
66
lambdaA * A + lambdaT * T + lambdaW * W where A, T and W are
67
the area, temperature and wire length respectively. The
68
weights lambdaA lambdaT and lambdaW can be set through
69
configuration parameters. Please note that for HotFloorplan,
70
A is in the order or hundreds of mm, T is in the order of
71
hundreds of Kelvin and W is in the order of tens of mm. In
72
combining them to a single metric, the weights have to be
73
assigned in a manner that not only takes into account the
74
desired importance of the variables A, T and W but also
75
compensates for the mismatch in their units of measurement.
76
The default weights in HotFloorplan are chosen based on this
77
principle. If one is interested in objective functions that
78
are not in the above-mentioned form, it is quite easy to
79
make the change in HotFloorplan. The design of the simulated
80
annealing algorithm is such that any arbitrary metric (objective
81
function) can be incorporated. In HotFloorplan, this is done
82
just by changing the return statement of the
83
'flp_evaluate_metric' function at the beginning of 'flp.c' into
84
an appropriate objective function of A, T and W.
85
86
87