| // Copyright ©2020 The Gonum Authors. All rights reserved. |
| // Use of this source code is governed by a BSD-style |
| // license that can be found in the LICENSE file. |
| |
| package testlapack |
| |
| import ( |
| "fmt" |
| "math" |
| "testing" |
| |
| "golang.org/x/exp/rand" |
| |
| "gonum.org/v1/gonum/blas" |
| "gonum.org/v1/gonum/floats" |
| "gonum.org/v1/gonum/lapack" |
| ) |
| |
| type Dlantber interface { |
| Dlantb(norm lapack.MatrixNorm, uplo blas.Uplo, diag blas.Diag, n, k int, a []float64, lda int, work []float64) float64 |
| } |
| |
| func DlantbTest(t *testing.T, impl Dlantber) { |
| rnd := rand.New(rand.NewSource(1)) |
| for _, norm := range []lapack.MatrixNorm{lapack.MaxAbs, lapack.MaxRowSum, lapack.MaxColumnSum, lapack.Frobenius} { |
| for _, uplo := range []blas.Uplo{blas.Lower, blas.Upper} { |
| for _, diag := range []blas.Diag{blas.NonUnit, blas.Unit} { |
| name := normToString(norm) + uploToString(uplo) + diagToString(diag) |
| t.Run(name, func(t *testing.T) { |
| for _, n := range []int{0, 1, 2, 3, 4, 5, 10} { |
| for _, k := range []int{0, 1, 2, 3, n, n + 2} { |
| for _, lda := range []int{k + 1, k + 3} { |
| for iter := 0; iter < 10; iter++ { |
| dlantbTest(t, impl, rnd, norm, uplo, diag, n, k, lda) |
| } |
| } |
| } |
| } |
| }) |
| } |
| } |
| } |
| } |
| |
| func dlantbTest(t *testing.T, impl Dlantber, rnd *rand.Rand, norm lapack.MatrixNorm, uplo blas.Uplo, diag blas.Diag, n, k, lda int) { |
| const tol = 1e-14 |
| |
| name := fmt.Sprintf("n=%v,k=%v,lda=%v", n, k, lda) |
| |
| // Deal with zero-sized matrices early. |
| if n == 0 { |
| got := impl.Dlantb(norm, uplo, diag, n, k, nil, lda, nil) |
| if got != 0 { |
| t.Errorf("%v: unexpected result for zero-sized matrix", name) |
| } |
| return |
| } |
| |
| a := make([]float64, max(0, (n-1)*lda+k+1)) |
| if rnd.Float64() < 0.5 { |
| // Sometimes fill A with elements between -0.5 and 0.5 so that for |
| // blas.Unit matrices the largest element is the 1 on the main diagonal. |
| for i := range a { |
| // Between -0.5 and 0.5. |
| a[i] = rnd.Float64() - 0.5 |
| } |
| } else { |
| for i := range a { |
| // Between -2 and 2. |
| a[i] = 4*rnd.Float64() - 2 |
| } |
| } |
| // Sometimes put a NaN into A. |
| if rnd.Float64() < 0.5 { |
| a[rnd.Intn(len(a))] = math.NaN() |
| } |
| // Make a copy of A for later comparison. |
| aCopy := make([]float64, len(a)) |
| copy(aCopy, a) |
| |
| var work []float64 |
| if norm == lapack.MaxColumnSum { |
| work = make([]float64, n) |
| } |
| // Fill work with random garbage. |
| for i := range work { |
| work[i] = rnd.NormFloat64() |
| } |
| |
| got := impl.Dlantb(norm, uplo, diag, n, k, a, lda, work) |
| |
| if !floats.Same(a, aCopy) { |
| t.Fatalf("%v: unexpected modification of a", name) |
| } |
| |
| // Generate a dense representation of A and compute the wanted result. |
| ldaGen := n |
| aGen := make([]float64, n*ldaGen) |
| if uplo == blas.Upper { |
| for i := 0; i < n; i++ { |
| for j := 0; j < min(n-i, k+1); j++ { |
| aGen[i*ldaGen+i+j] = a[i*lda+j] |
| } |
| } |
| } else { |
| for i := 0; i < n; i++ { |
| for j := max(0, k-i); j < k+1; j++ { |
| aGen[i*ldaGen+i-(k-j)] = a[i*lda+j] |
| } |
| } |
| } |
| if diag == blas.Unit { |
| for i := 0; i < n; i++ { |
| aGen[i*ldaGen+i] = 1 |
| } |
| } |
| want := dlange(norm, n, n, aGen, ldaGen) |
| |
| if math.IsNaN(want) { |
| if !math.IsNaN(got) { |
| t.Errorf("%v: unexpected result with NaN element; got %v, want %v", name, got, want) |
| } |
| return |
| } |
| |
| if norm == lapack.MaxAbs { |
| if got != want { |
| t.Errorf("%v: unexpected result; got %v, want %v", name, got, want) |
| } |
| return |
| } |
| diff := math.Abs(got - want) |
| if diff > tol { |
| t.Errorf("%v: unexpected result; got %v, want %v, diff=%v", name, got, want, diff) |
| } |
| } |