github.com/gonum/lapack@v0.0.0-20181123203213-e4cdc5a0bff9/native/dormhr.go (about) 1 // Copyright ©2016 The gonum Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 package native 6 7 import "github.com/gonum/blas" 8 9 // Dormhr multiplies an m×n general matrix C with an nq×nq orthogonal matrix Q 10 // Q * C, if side == blas.Left and trans == blas.NoTrans, 11 // Q^T * C, if side == blas.Left and trans == blas.Trans, 12 // C * Q, if side == blas.Right and trans == blas.NoTrans, 13 // C * Q^T, if side == blas.Right and trans == blas.Trans, 14 // where nq == m if side == blas.Left and nq == n if side == blas.Right. 15 // 16 // Q is defined implicitly as the product of ihi-ilo elementary reflectors, as 17 // returned by Dgehrd: 18 // Q = H_{ilo} H_{ilo+1} ... H_{ihi-1}. 19 // Q is equal to the identity matrix except in the submatrix 20 // Q[ilo+1:ihi+1,ilo+1:ihi+1]. 21 // 22 // ilo and ihi must have the same values as in the previous call of Dgehrd. It 23 // must hold that 24 // 0 <= ilo <= ihi < m, if m > 0 and side == blas.Left, 25 // ilo = 0 and ihi = -1, if m = 0 and side == blas.Left, 26 // 0 <= ilo <= ihi < n, if n > 0 and side == blas.Right, 27 // ilo = 0 and ihi = -1, if n = 0 and side == blas.Right. 28 // 29 // a and lda represent an m×m matrix if side == blas.Left and an n×n matrix if 30 // side == blas.Right. The matrix contains vectors which define the elementary 31 // reflectors, as returned by Dgehrd. 32 // 33 // tau contains the scalar factors of the elementary reflectors, as returned by 34 // Dgehrd. tau must have length m-1 if side == blas.Left and n-1 if side == 35 // blas.Right. 36 // 37 // c and ldc represent the m×n matrix C. On return, c is overwritten by the 38 // product with Q. 39 // 40 // work must have length at least max(1,lwork), and lwork must be at least 41 // max(1,n), if side == blas.Left, and max(1,m), if side == blas.Right. For 42 // optimum performance lwork should be at least n*nb if side == blas.Left and 43 // m*nb if side == blas.Right, where nb is the optimal block size. On return, 44 // work[0] will contain the optimal value of lwork. 45 // 46 // If lwork == -1, instead of performing Dormhr, only the optimal value of lwork 47 // will be stored in work[0]. 48 // 49 // If any requirement on input sizes is not met, Dormhr will panic. 50 // 51 // Dormhr is an internal routine. It is exported for testing purposes. 52 func (impl Implementation) Dormhr(side blas.Side, trans blas.Transpose, m, n, ilo, ihi int, a []float64, lda int, tau, c []float64, ldc int, work []float64, lwork int) { 53 var ( 54 nq int // The order of Q. 55 nw int // The minimum length of work. 56 ) 57 switch side { 58 case blas.Left: 59 nq = m 60 nw = n 61 case blas.Right: 62 nq = n 63 nw = m 64 default: 65 panic(badSide) 66 } 67 switch { 68 case trans != blas.NoTrans && trans != blas.Trans: 69 panic(badTrans) 70 case ilo < 0 || max(1, nq) <= ilo: 71 panic(badIlo) 72 case ihi < min(ilo, nq-1) || nq <= ihi: 73 panic(badIhi) 74 case lwork < max(1, nw) && lwork != -1: 75 panic(badWork) 76 case len(work) < max(1, lwork): 77 panic(shortWork) 78 } 79 if lwork != -1 { 80 checkMatrix(m, n, c, ldc) 81 checkMatrix(nq, nq, a, lda) 82 if len(tau) != nq-1 && nq > 0 { 83 panic(badTau) 84 } 85 86 } 87 88 nh := ihi - ilo 89 var nb int 90 if side == blas.Left { 91 opts := "LN" 92 if trans == blas.Trans { 93 opts = "LT" 94 } 95 nb = impl.Ilaenv(1, "DORMQR", opts, nh, n, nh, -1) 96 } else { 97 opts := "RN" 98 if trans == blas.Trans { 99 opts = "RT" 100 } 101 nb = impl.Ilaenv(1, "DORMQR", opts, m, nh, nh, -1) 102 } 103 lwkopt := max(1, nw) * nb 104 if lwork == -1 { 105 work[0] = float64(lwkopt) 106 return 107 } 108 109 if m == 0 || n == 0 || nh == 0 { 110 work[0] = 1 111 return 112 } 113 if side == blas.Left { 114 impl.Dormqr(side, trans, nh, n, nh, a[(ilo+1)*lda+ilo:], lda, 115 tau[ilo:ihi], c[(ilo+1)*ldc:], ldc, work, lwork) 116 } else { 117 impl.Dormqr(side, trans, m, nh, nh, a[(ilo+1)*lda+ilo:], lda, 118 tau[ilo:ihi], c[ilo+1:], ldc, work, lwork) 119 } 120 work[0] = float64(lwkopt) 121 }