MAGMA  1.2.0
MatrixAlgebraonGPUandMulticoreArchitectures
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups
zgessm_gpu.cpp
Go to the documentation of this file.
1 /*
2  -- MAGMA (version 1.2.0) --
3  Univ. of Tennessee, Knoxville
4  Univ. of California, Berkeley
5  Univ. of Colorado, Denver
6  May 2012
7 
8  @author Hatem Ltaief
9  @author Mathieu Faverge
10 
11  @precisions normal z -> c d s
12 
13 */
14 #include "common_magma.h"
15 
16 #define magma_zgemm magmablas_zgemm
17 //#define magma_ztrsm magmablas_ztrsm
18 //#define magma_ztrmm magmablas_ztrmm
19 
20 extern "C" magma_int_t
21 magma_zgessm_gpu( char storev, magma_int_t m, magma_int_t n, magma_int_t k, magma_int_t ib,
22  magma_int_t *ipiv,
23  cuDoubleComplex *dL1, magma_int_t lddl1,
24  cuDoubleComplex *dL, magma_int_t lddl,
25  cuDoubleComplex *dA, magma_int_t ldda,
26  magma_int_t *info)
27 {
28 /* -- MAGMA (version 1.2.0) --
29  Univ. of Tennessee, Knoxville
30  Univ. of California, Berkeley
31  Univ. of Colorado, Denver
32  May 2012
33 
34  Purpose
35  =======
36 
37  SGETRF computes an LU factorization of a general M-by-N matrix A
38  using partial pivoting with row interchanges.
39 
40  The factorization has the form
41  A = P * L * U
42  where P is a permutation matrix, L is lower triangular with unit
43  diagonal elements (lower trapezoidal if m > n), and U is upper
44  triangular (upper trapezoidal if m < n).
45 
46  This is the right-looking Level 3 BLAS version of the algorithm.
47 
48  Arguments
49  =========
50 
51  M (input) INTEGER
52  The number of rows of the matrix A. M >= 0.
53 
54  N (input) INTEGER
55  The number of columns of the matrix A. N >= 0.
56 
57  A (input/output) REAL array on the GPU, dimension (LDA,N).
58  On entry, the M-by-N matrix to be factored.
59  On exit, the factors L and U from the factorization
60  A = P*L*U; the unit diagonal elements of L are not stored.
61 
62  LDA (input) INTEGER
63  The leading dimension of the array A. LDA >= max(1,M).
64 
65  IPIV (output) INTEGER array, dimension (min(M,N))
66  The pivot indices; for 1 <= i <= min(M,N), row i of the
67  matrix was interchanged with row IPIV(i).
68 
69  INFO (output) INTEGER
70  = 0: successful exit
71  < 0: if INFO = -i, the i-th argument had an illegal value
72  or another error occured, such as memory allocation failed.
73  > 0: if INFO = i, U(i,i) is exactly zero. The factorization
74  has been completed, but the factor U is exactly
75  singular, and division by zero will occur if it is used
76  to solve a system of equations.
77 
78  ===================================================================== */
79 
80 #define AT(i,j) (dAT + (i)*ldda + (j) )
81 #define L(i,j) (dL + (i) + (j)*lddl )
82 #define dL1(j) (dL1 + (j)*lddl1)
83 
84  cuDoubleComplex c_one = MAGMA_Z_ONE;
85  cuDoubleComplex c_neg_one = MAGMA_Z_NEG_ONE;
86 
87  int i, s, sb;
88  cuDoubleComplex *dAT;
89 
90  /* Check arguments */
91  *info = 0;
92  if (m < 0)
93  *info = -1;
94  else if (n < 0)
95  *info = -2;
96  else if (ldda < max(1,m))
97  *info = -4;
98 
99  if (*info != 0) {
100  magma_xerbla( __func__, -(*info) );
101  return *info;
102  }
103 
104  /* Quick return if possible */
105  if (m == 0 || n == 0)
106  return *info;
107 
108  if ( (storev == 'C') || (storev == 'c') ) {
109  magmablas_zgetmo_in( dA, dAT, ldda, m, n );
110  } else {
111  dAT = dA;
112  }
113 
114  s = k / ib;
115  for(i = 0; i < k; i += ib) {
116  sb = min(ib, k-i);
117 
118  magmablas_zlaswp( n, dAT, ldda, i+1, i+sb, ipiv, 1 );
119 
120 #ifndef WITHOUTTRTRI
121  magma_ztrmm( MagmaRight, MagmaLower, MagmaTrans, MagmaUnit,
122  n, sb,
123  c_one, dL1(i), lddl1,
124  AT(i, 0), ldda);
125 #else
126  magma_ztrsm( MagmaRight, MagmaLower, MagmaTrans, MagmaUnit,
127  n, sb,
128  c_one, L( i, i), lddl,
129  AT(i, 0), ldda);
130 #endif
131 
132  if ( (i+sb) < m) {
133  magma_zgemm( MagmaNoTrans, MagmaTrans,
134  n, m-(i+sb), sb,
135  c_neg_one, AT(i, 0), ldda,
136  L( i+sb, i), lddl,
137  c_one, AT(i+sb, 0), ldda );
138  }
139  }
140 
141  if ( (storev == 'C') || (storev == 'c') ) {
142  magmablas_zgetmo_in( dA, dAT, ldda, m, n );
143  }
144 
145  return *info;
146  /* End of MAGMA_ZGETRF_GPU */
147 }