Implementing Kronecker Products Efficiently
In the algorithm described above we encountered expressions of the form A1⊗A2⊗⋯⊗An which we denote by ⊗i=1nAi. To calculate the product (⊗i,Ai)x it is computationally advantageous to factor ⊗iAi into terms of the form I⊗Ai⊗I[1]. Then each term represents a set of copies of Ai. First, recall the following property of Kronecker products
AB⊗CD=(A⊗C)(B⊗D).(1)
This property can be used to factor ⊗iAi in the following way. Let the number of rows and columns of Ai be denoted by ri and ci respectively. Then
A1⊗A2 | = | A1Ic1⊗Ir2A2 |
= | (A1⊗Ir2)(Ic1⊗A2). |
But we can also write
A1⊗A2 | = | Ir1A1⊗A2Ic2 |
= | (Ir1⊗A2)(A1⊗Ic2). |
Note that in factorization Equation 2, copies of A2 are applied to the data vector x first, followed by copies of A1. On the other hand, in factorization Equation 3, copies of A1 are applied to the data vector x first, followed by copies of A2. These two factorizations can be distinguished by the sequence in which A1 and A2 are ordered.
Lets compare the computational complexity of factorizations Equation 2 and Equation 3. Notice that Equation 2 consists of r2 copies of A1 and c1 copies of A2, therefore Equation 2 has a computational cost of r2Q1+c1Q2 where Qi is the computational cost of Ai. On the other hand, the computational cost of Equation 3 is c2Q1+r1Q2. That is, the factorizations Equation 2 andEquation 3 have in general different computational costs when Ai are not square. Further, observe that Equation 2 is the more efficient factorization exactly when
r2Q1+c1Q2<c2Q1+r1Q2(4)
or equivalently, when
r1−c1 |
Q1 |
r2−c2 |
Q2 |
Consequently, in the more efficient factorization, the operation Ai applied to the data vector x first is the one for which the ratio (ri−ci)/Qi is the more negative. If r1>c1 and r2<c2 thenEquation 4 is always true (Qi is always positive). Therefore, in the most computationally efficient factorization of A1⊗A2, matrices with fewer rows than columns are always applied to the data vector x before matrices with more rows than columns. If both matrices are square, then their ordering does not affect the computational efficiency, because in that case each ordering has the same computation cost.
We now consider the Kronecker product of more than two matrices. For the Kronecker product ⊗i=1nAi there are n! possible different ways in which to order the operations Ai. For example
A1⊗A2⊗A3 | = | (A1⊗Ir2r3)(Ic1⊗A2⊗Ir3)(Ic1c2⊗A3) |
= | (A1⊗Ir2r3)(Ic1r2⊗A3)(Ic1⊗A2⊗Ic3) | |
= | (Ir1⊗A2⊗Ir3)(A1⊗Ic2r3)(Ic1c2⊗A3) | |
= | (Ir1⊗A2⊗Ir3)(Ir1c2⊗A3)(A1⊗Ic2c3) | |
= | (Ir1r2⊗A3)(A1⊗Ir2c3)(Ic1⊗A2⊗Ic3) | |
= | (Ir1r2⊗A3)(Ir1⊗A2⊗Ic3)(A1⊗Ic2c3) |
Each factorization of ⊗iAi can be described by a permutation g(·) of {1,...,n{ which gives the order in which Ai is applied to the data vector x. Ag(1) is the first operation applied to the data vector x, Ag(2) is the second, and so on. For example, the factorization Equation 6 is described by the permutation g(1)=3, g(2)=1, g(3)=2. For n=3, the computational cost of each factorization can be written as
C(g)=Qg(1)cg(2)cg(3)+rg(1)Qg(2)cg(3)+rg(1)rg(2)Qg(3)(7)
In general
C(g)=
(
,rg(j))Qg(i)(
,cg(j)).(8)
n |
∑ |
i=1 |
i−1 |
∏ |
j=1 |
n |
∏ |
j=i+1 |
Therefore, the most efficient factorization of ⊗iAi is described by the permutation g(·) that minimizes C.
It turns out that for the Kronecker product of more than two matrices, the ordering of operations that describes the most efficient factorization of ⊗iAi also depends only on the ratios(ri−ci)/Qi. To show that this is the case, suppose u(·) is the permutation that minimizes C, then u(·) has the property that
ru(k)−cu(k) |
Qu(k) |
ru(k+1)−cu(k+1) |
Qu(k+1) |
for k=1,⋯,n−1. To support this, note that since u(·) is the permutation that minimizes C, we have in particular
C(u)≤C(v)(10)
where v(·) is the permutation defined by the following:
v(i)={
.(11)
u(i) | i<k,i>k+1 |
u(k+1) | i=k |
u(k) | i=k+1 |
Because only two terms in Equation 8 are different, we have from Equation 10
k+1 |
∑ |
i=k |
i−1 |
∏ |
j=1 |
n |
∏ |
j=i+1 |
k+1 |
∑ |
i=k |
i−1 |
∏ |
j=1 |
n |
∏ |
j=i+1 |
which, after canceling common terms from each side, gives
Qu(k)cu(k+1)+ru(k)Qu(k+1)≤Qv(k)cv(k+1)+rv(k)Qv(k+1).(13)
Since v(k)=u(k+1) and v(k+1)=u(k) this becomes
Qu(k)cu(k+1)+ru(k)Qu(k+1)≤Qu(k+1)cu(k)+ru(k+1)Qu(k)(14)
which is equivalent to Equation 9. Therefore, to find the best factorization of ⊗iAi it is necessary only to compute the ratios (ri−ci)/Qi and to order them in an non-decreasing order. The operation Ai whose index appears first in this list is applied to the data vector x first, and so on
As above, if ru(k+1)>cu(k+1) and ru(k)<cu(k) then Equation 14 is always true. Therefore, in the most computationally efficient factorization of ⊗iAi, all matrices with fewer rows than columns are always applied to the data vector x before any matrices with more rows than columns. If some matrices are square, then their ordering does not affect the computational efficiency as long as they are applied after all matrices with fewer rows than columns and before all matrices with more rows than columns.
Once the permutation g(·) that minimizes C is determined by ordering the ratios (ri−ci)/Qi, ⊗iAi can be written as
n |
⊗ |
i=1 |
1 |
∏ |
i=n |
where
a(i)=
γ(i,k)(16)
g(i)−1 |
∏ |
k=1 |
b(i)=
γ(i,k)(17)
n |
∏ |
k=g(i)+1 |
and where γ(·) is defined by
γ(i,k)={
.(18)
rk | ifg(i)>g(k) |
ck | ifg(i)<g(k) |
Some Matlab Code
A Matlab program that computes the permutation that describes the computationally most efficient factorization of ⊗i=1nAi is
cgc()
. It also gives the resulting computational cost. It requires the computational cost of each of the matrices Ai and the number of rows and columns of each.function [g,C] = cgc(Q,r,c,n) % [g,C] = cgc(Q,r,c,n); % Compute g and C % g : permutation that minimizes C % C : computational cost of Kronecker product of A(1),...,A(n) % Q : computation cost of A(i) % r : rows of A(i) % c : columns of A(i) % n : number of terms f = find(Q==0); Q(f) = eps * ones(size(Q(f))); Q = Q(:); r = r(:); c = c(:); [s,g] = sort((r-c)./Q); C = 0; for i = 1:n C = C + prod(r(g(1:i-1)))*Q(g(i))*prod(c(g(i+1:n))); end C = round(C);
The Matlab program
kpi()
implements the Kronecker product ⊗i=1nAi.function y = kpi(d,g,r,c,n,x) % y = kpi(d,g,r,c,n,x); % Kronecker Product : A(d(1)) kron ... kron A(d(n)) % g : permutation of 1,...,n % r : [r(1),...,r(n)] % c : [c(1),..,c(n)] % r(i) : rows of A(d(i)) % c(i) : columns of A(d(i)) % n : number of terms for i = 1:n a = 1; for k = 1:(g(i)-1) if i > find(g==k) a = a * r(k); else a = a * c(k); end end b = 1; for k = (g(i)+1):n if i > find(g==k) b = b * r(k); else b = b * c(k); end end % y = (I(a) kron A(d(g(i))) kron I(b)) * x; y = IAI(d(g(i)),a,b,x); end
where the last line of code calls a function that implements (Ia⊗Ad(g(i))⊗Ib)x. That is, the program
IAI(i,a,b,x)
implements (Ia⊗A(i)⊗Ib)x.The Matlab program
IAI
implements y=(Im⊗A⊗In)xfunction y = IAI(A,r,c,m,n,x) % y = (I(m) kron A kron I(n))x % r : number of rows of A % c : number of columns of A v = 0:n:n*(r-1); u = 0:n:n*(c-1); for i = 0:m-1 for j = 0:n-1 y(v+i*r*n+j+1) = A * x(u+i*c*n+j+1); end end
It simply uses two loops to implement the mn copies of A. Each copy of A is applied to a different subset of the elements of x.
Vector/Parallel Interpretation
The command I⊗A⊗I where ⊗ is the Kronecker (or Tensor) product can be interpreted as a vector/parallel command [2], [3]. In these references, the implementation of these commands is discussed in detail and they have found that the Tensor product is “an extremely useful tool for matching algorithms to computer architectures [2].”
The expression I⊗A can easily be seen to represent a parallel command:
I⊗A=[
].(19)
A | |||
A | |||
⋱ | |||
A |
Each block along the diagonal acts on non-overlapping sections of the data vector - so that each section can be performed in parallel. Since each section represents exactly the same operation, this form is amenable to implementation on a computer with a parallel architectural configuration. The expression A⊗I can be similarly seen to represent a vector command, see [2].
It should also be noted that by employing `stride' permutations, the command (I⊗A⊗I)x can be replaced by either (I⊗A)x or (A⊗I)x[2], [3]. It is only necessary to permute the input and output. It is also the case that these stride permutations are natural loading and storing commands for some architectures.
In the programs we have written in conjunction with this paper we implement the commands y=(I⊗A⊗I)x with loops in a set of subroutines. The circular convolution and prime length FFT programs we present, however, explicitly use the form I⊗A⊗I to make clear the structure of the algorithm, to make them more modular and simpler, and to make them amenable to implementation on special architectures. In fact, in [2] it is suggested that it might be practical to develop tensor product compilers. The FFT programs we have generated will be well suited for such compilers.
REFERENCES
- Agarwal, R. C. and Cooley, J. W. (1977, October). New Algorithms for Digital Convolution. IEEE Trans. Acoust., Speech, Signal Proc., 25(5), 392-410.
- Granata, J. and Conner, M. and Tolimieri, R. (1992, January). The Tensor Product: A Mathematical Programming Language for FFTs and other Fast DSP Operations. IEEE Signal Processing Magazine, 9(1), 40-48.
- Tolimieri, R. and An, M. and Lu, C. (1989). Algorithms for Discrete Fourier Transform and Convolution. Springer-Verlag.
0 comments:
Post a Comment