| [eigen] some more benchmark results for mapped matrices |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen <eigen@xxxxxxxxxxxxxxxxxxx>
- Subject: [eigen] some more benchmark results for mapped matrices
- From: Sameer Agarwal <sameeragarwal@xxxxxxxxxx>
- Date: Wed, 11 Jan 2012 01:20:10 -0800
- Cc: Paul Tucker <tucker@xxxxxxxxxx>, Keir Mierle <keir@xxxxxxxxxx>
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=gamma; h=mime-version:date:message-id:subject:from:to:cc:x-system-of-record :content-type; bh=LPiV4n83fs0+ArqGMEkoOHDEqA86gyJHiOBewIFf+W8=; b=hkAwvTf1sLDSOD0J405mYbv70zFryzgCoGM3Ip/spyXkWJ/uThBdShpx6h4Mv0mUJ8 SizZ9QMXPH4ZJMOEL/nnU6uSaSRcI4nGpIQnjDxZNuxRis0MJrxUKWhNN08TrSXyvqjS xslo+OB4Po+FaKsNdzRYLHVtRpAKMfv1Z/vrc=
I redid some of the experiments (fixing some bugs along the way) and here are two experiments. In both cases we report a table of ratios of eigen3 to eigen2 times.
1. matrix-vector multiply
y += A x
y is n x 1
A is n x m
x is m x 1
1 2 3 4 5 6 7 8 9 10
------------------------------------------------------------
1| 4.40 4.80 4.14 4.00 3.88 4.25 3.90 3.40 3.36 3.64
2| 3.67 3.57 3.22 2.73 2.69 2.64 2.62 2.06 2.26 2.30
3| 3.29 3.25 3.30 2.50 2.71 2.44 2.71 2.11 2.14 2.18
4| 2.88 2.55 2.54 2.06 2.11 2.10 2.00 1.63 1.82 1.77
5| 3.00 2.64 2.57 1.94 2.00 2.00 2.00 1.67 1.83 1.81
6| 2.67 2.14 2.12 1.75 1.72 1.68 1.62 1.38 1.43 1.51
7| 2.78 2.14 2.17 1.67 1.80 1.71 1.78 1.33 1.36 1.41
8| 2.36 1.88 1.81 1.48 1.43 1.43 1.53 1.16 1.08 1.21
9| 2.45 2.06 2.00 1.52 1.52 1.46 1.47 1.08 1.17 1.31
10| 2.17 1.79 1.67 1.31 1.34 1.32 1.15 1.06 1.07 1.13
2. Matrix-Matrix product.
Y += A * X
Y = n x n
A = n x m
X = m x n
1 2 3 4 5 6 7 8 9 10
------------------------------------------------------------
1| 7.10 6.55 6.73 6.42 6.50 6.15 5.86 5.93 5.80 5.56
2| 3.12 3.04 3.04 2.81 2.68 2.67 2.54 2.44 2.52 2.45
3| 2.65 2.54 2.55 2.49 2.42 2.37 2.35 2.35 2.25 2.24
4| 1.65 1.59 1.53 1.44 1.42 1.38 1.40 1.36 1.38 1.33
5| 1.79 1.70 1.60 1.51 1.45 1.42 1.41 1.43 1.28 1.36
6| 1.39 1.28 1.24 1.17 1.13 1.12 1.16 1.17 1.13 1.08
7| 1.51 1.37 1.32 1.24 1.20 1.19 1.21 1.09 1.07 1.05
8| 1.11 1.01 1.01 0.91 0.93 0.93 0.92 0.92 0.89 0.87
9| 1.29 1.11 1.07 1.00 1.02 0.85 0.92 0.83 0.88 0.84
10| 1.11 0.94 0.95 0.88 0.88 0.83 0.88 0.83 0.80 0.80
As can be seen, for small matrices, eigen3 is significantly worse than eigen2 on unaligned mapped matrices/vectors.
Sameer