vectorized-eigensolver/python_interface.pyx at master · kipfstuhl/vectorized-eigensolver · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220


cimport numpy as np
import numpy as np
cimport cython
import cython
from libcpp cimport bool
ctypedef np.float64_t dtype_t

cimport openmp

cdef extern from "utils.cpp":
    void repMatrixSoA "repMatrixSoA"(int dim, int block_cnt, int thread_cnt, double* matrix, double* out) nogil
    void repVectorSoA "repVectorSoA"(int dim, int block_cnt, int thread_cnt, double *vec, double *out) nogil
    void rearrangeMatrices "rearrangeMatrices"(int dim, int block_cnt, int thread_cnt,
                                                   double *inputMat, double *outputMat) nogil
    void call_zlahqr "call_zlahqr_new"(int dim, int block_cnt, int thread_cnt, double *rE, double *iE,
                                           double *rM, double *iM, double *rS, double *iS) nogil
    void call_zgemv "call_zgemv"(bool hermitian, int dim, int block_cnt, int thread_cnt, double *rM, double *iM,
                                     double *rx, double *ix, double *ry, double *iy,
                                     double rbeta, double ibeta) nogil
    void call_ztrevc "call_ztrevc"(int dim, int block_cnt, int thread_cnt,
                                       double *rT, double *iT, double *rR, double *iR) nogil
    void call_backsubs "call_backsubstitution"(int dim, int block_cnt, int thread_cnt,
                                                   double *rT, double *iT, double *rb, double *ib) nogil
    void call_zgehrd "call_zgehrd"(int dim, int block_cnt, int thread_cnt, double *rA, double *iA, double *rtau, double *itau) nogil
    void call_zunghr "call_zunghr"(int dim, int block_cnt, int thread_cnt, double *rA, double *iA, double *rtau, double *itau) nogil
    void hessenberg "hessenberg"(int dim, int block_cnt, int thread_cnt, double *rA, double *iA, double *rQ, double *iQ) nogil


def my_repMatrix(int dim, int block_cnt, int thread_cnt,
                     np.ndarray[dtype_t, ndim=1] matrix, np.ndarray[dtype_t, ndim=1] out):
    with nogil:
        repMatrixSoA(dim, block_cnt, thread_cnt, &matrix[0], &out[0])

def arrangeMatrices(int dim, int block_cnt, int thread_cnt,
                        np.ndarray[dtype_t, ndim=1] inputMat, np.ndarray[dtype_t, ndim=1] outputMat):
    with nogil:
        rearrangeMatrices(dim, block_cnt, thread_cnt, &inputMat[0], &outputMat[0])

def my_zlahqr(int dim, int block_cnt, int thread_cnt,
                  np.ndarray[dtype_t, ndim=1] rE, np.ndarray[dtype_t, ndim=1] iE,
                  np.ndarray[dtype_t, ndim=1] rM, np.ndarray[dtype_t, ndim=1] iM,
                  np.ndarray[dtype_t, ndim=1] rS, np.ndarray[dtype_t, ndim=1] iS):
    with nogil:
        call_zlahqr(dim, block_cnt, thread_cnt, &rE[0], &iE[0], &rM[0], &iM[0], &rS[0], &iS[0])

def my_backsubs(int dim, int block_cnt, int thread_cnt,
                    np.ndarray[dtype_t, ndim=1] rT, np.ndarray[dtype_t, ndim=1] iT,
                    np.ndarray[dtype_t, ndim=1] rb, np.ndarray[dtype_t, ndim=1] ib):
    with nogil:
        call_backsubs(dim, block_cnt, thread_cnt, &rT[0], &iT[0], &rb[0], &ib[0])

def my_repVector(int dim, int block_cnt, int thread_cnt,
                     np.ndarray[dtype_t, ndim=1] vector, np.ndarray[dtype_t, ndim=1] out):
    with nogil:
        repVectorSoA(dim, block_cnt, thread_cnt, &vector[0], &out[0])

def my_ztrevec(int dim, int block_cnt, int thread_cnt,
                   np.ndarray[dtype_t, ndim=1] rT, np.ndarray[dtype_t, ndim=1] iT,
                   np.ndarray[dtype_t, ndim=1] rR, np.ndarray[dtype_t, ndim=1] iR):
    with nogil:
        call_ztrevc(dim, block_cnt, thread_cnt, &rT[0], &iT[0], &rR[0], &iR[0])

def my_zgemv(bool hermitian, int dim, int block_cnt, int thread_cnt,
                 np.ndarray[dtype_t, ndim=1] rA, np.ndarray[dtype_t, ndim=1] iA,
                 np.ndarray[dtype_t, ndim=1] rx, np.ndarray[dtype_t, ndim=1] ix,
                 double rbeta, double ibeta,
                 np.ndarray[dtype_t, ndim=1] ry, np.ndarray[dtype_t, ndim=1] iy):
    with nogil:
        call_zgemv(hermitian, dim, block_cnt, thread_cnt,
                       &rA[0], &iA[0], &rx[0], &ix[0], &ry[0], &iy[0], rbeta, ibeta)

def my_zgehrd(int dim, int block_cnt, int thread_cnt,
                  np.ndarray[dtype_t, ndim=1] rA, np.ndarray[dtype_t, ndim=1] iA,
                  np.ndarray[dtype_t, ndim=1] rtau, np.ndarray[dtype_t, ndim=1] itau):
    with nogil:
        call_zgehrd(dim, block_cnt, thread_cnt, &rA[0], &iA[0], &rtau[0], &itau[0])

def my_zunghr(int dim, int block_cnt, int thread_cnt,
                  np.ndarray[dtype_t, ndim=1] rA, np.ndarray[dtype_t, ndim=1] iA,
                  np.ndarray[dtype_t, ndim=1] rtau, np.ndarray[dtype_t, ndim=1] itau):
    with nogil:
        call_zunghr(dim, block_cnt, thread_cnt, &rA[0], &iA[0], &rtau[0], &itau[0])

def my_hessenberg(int dim, int block_cnt, int thread_cnt,
                      np.ndarray[dtype_t, ndim=1] rA, np.ndarray[dtype_t, ndim=1] iA,
                      np.ndarray[dtype_t, ndim=1] rQ, np.ndarray[dtype_t, ndim=1] iQ):
    with nogil:
        hessenberg(dim, block_cnt, thread_cnt, &rA[0], &iA[0], &rQ[0], &iQ[0])

def extractMatrix(int dim, int thread_cnt, np.ndarray[dtype_t, ndim=1] matrix):
    return matrix[::thread_cnt][:dim*dim].reshape(dim,dim).T

def extractVector(int dim, int thread_cnt, np.ndarray[dtype_t, ndim=1] vector):
    return vector[::thread_cnt][:dim]

def test_backsubs(int dim, int block_cnt, int thread_cnt):
    a = np.random.random((dim, dim))
    a = np.triu(a)
    ra = np.zeros(dim*dim*thread_cnt*block_cnt, dtype='double')
    ia = np.zeros_like(ra)
    my_repMatrix(dim, block_cnt, thread_cnt, a.T.reshape(dim*dim), ra)

    x = np.random.random(dim)
    rx = np.zeros(dim*thread_cnt*block_cnt, dtype='double')
    ix = np.zeros_like(rx)
    my_repVector(dim, block_cnt, thread_cnt, x, rx)

    my_backsubs(dim, block_cnt, thread_cnt, ra, ia, rx, ix)

    sol = extractVector(dim, thread_cnt, rx)
    return a.dot(sol) - x

def test_zgehrd(int dim, int block_cnt, int thread_cnt):
    a = np.random.random((dim, dim))
    ra = np.zeros(dim*dim * thread_cnt * block_cnt, dtype='double')
    ia = np.zeros_like(ra)
    my_repMatrix(dim, block_cnt, thread_cnt, a.T.reshape(dim*dim), ra)

    rtau = np.zeros(dim * thread_cnt * block_cnt, dtype='double')
    itau = np.zeros_like(rtau)
    my_zgehrd(dim, block_cnt, thread_cnt, ra, ia, rtau, itau)
    erg = extractMatrix(dim, thread_cnt, ra) + 1.0j * extractMatrix(dim, thread_cnt, ia)
    return erg, a

def test_hessenberg(int dim, int block_cnt, int thread_cnt):
    a = np.random.random((dim, dim))
    ra = np.zeros(dim*dim * thread_cnt * block_cnt, dtype='double')
    ia = np.zeros_like(ra)
    my_repMatrix(dim, block_cnt, thread_cnt, a.T.reshape(dim*dim), ra)

    rq = np.zeros_like(ra)
    iq = np.zeros_like(rq)

    my_hessenberg(dim, block_cnt, thread_cnt, ra, ia, rq, iq)

    h = extractMatrix(dim, thread_cnt, ra) + 1.0j * extractMatrix(dim, thread_cnt, ia)
    q = extractMatrix(dim, thread_cnt, rq) + 1.0j * extractMatrix(dim, thread_cnt, iq)

    return h, q, a

def test_zgemv(int dim, int block_cnt, int thread_cnt, bool hermitian = False):
    a = np.random.random((dim,dim))
    ra = np.zeros(dim*dim*thread_cnt*block_cnt, dtype='double')
    ia = np.zeros_like(ra)
    my_repMatrix(dim, block_cnt, thread_cnt, a.T.reshape(dim*dim), ra)

    y = np.random.random(dim)
    ry = np.zeros(dim*thread_cnt*block_cnt, dtype='double')
    iy = np.zeros_like(ry)
    my_repVector(dim, block_cnt, thread_cnt, y, ry)

    x = np.random.random(dim)
    rx = np.zeros_like(iy)
    ix = np.zeros_like(rx)
    my_repVector(dim, block_cnt, thread_cnt, x, rx)

    my_zgemv(hermitian, dim, block_cnt, thread_cnt, ra, ia, rx, ix, 1.0, 0.0, ry, iy)
    axpy = extractVector(dim, thread_cnt, ry)

    if hermitian:
        sol = np.conj(a.T).dot(x) + y
    else:
        sol = a.dot(x) + y
    return np.abs(axpy - sol)

def test_zlahqr(int dim, int block_cnt, int thread_cnt):
    dim2 = dim * dim
    size = dim2 * block_cnt * thread_cnt
    sizev = dim * block_cnt * thread_cnt

    a = np.random.random((dim,dim))
    a = np.triu(a,-1)
    ra = np.zeros(size, dtype='double')
    ia = np.zeros_like(ra)
    my_repMatrix(dim, block_cnt, thread_cnt, a.T.reshape(dim2), ra)

    s = np.eye(dim)
    rschur = np.zeros(size, dtype='double')
    ischur = np.zeros(size, dtype='double')
    my_repMatrix(dim, block_cnt, thread_cnt, s.T.reshape(dim2), rschur)

    re = np.zeros(sizev, dtype='double')
    ie = np.zeros(sizev, dtype='double')

    my_zlahqr(dim, block_cnt, thread_cnt, re, ie, ra, ia, rschur, ischur)

    eigs = extractVector(dim, thread_cnt, re) + 1.0j * extractVector(dim, thread_cnt, ie)
    schurv = extractMatrix(dim, thread_cnt, rschur) + 1.0j * extractMatrix(dim, thread_cnt, ischur)
    t = extractMatrix(dim, thread_cnt, ra) + 1.0j * extractMatrix(dim, thread_cnt, ia)

    return (eigs, schurv, t, a)

def test_eigenvectors(int dim, int block_cnt, int thread_cnt):
    dim2 = dim * dim
    size = dim2 * block_cnt * thread_cnt
    sizev = dim * block_cnt * thread_cnt

    a = np.random.random((dim, dim))
    a = np.triu(a, -1)
    ra = np.zeros(size, dtype='double')
    ia = np.zeros_like(ra)
    my_repMatrix(dim, block_cnt, thread_cnt, a.T.reshape(dim2), ra)

    s = np.eye(dim)
    rsch = np.zeros(size, dtype='double')
    isch = np.zeros(size, dtype='double')
    my_repMatrix(dim, block_cnt, thread_cnt, s.T.reshape(dim2), rsch)

    re = np.zeros(sizev, dtype='double')
    ie = np.zeros(sizev, dtype='double')

    my_zlahqr(dim, block_cnt, thread_cnt, re, ie, ra, ia, rsch, isch)
    my_ztrevec(dim, block_cnt, thread_cnt, ra, ia, rsch, isch)

    eigs = extractVector(dim, thread_cnt, re) + 1.0j * extractVector(dim, thread_cnt, ie)
    vecs = extractMatrix(dim, thread_cnt, rsch) + 1.0j * extractMatrix(dim, thread_cnt, isch)

    return(eigs, vecs, a)