lapack/testlapack: simplify randSymBand and use it in Dpb* tests

This commit is contained in:
Vladimir Chalupecky
2019-06-16 18:51:40 +02:00
committed by Vladimír Chalupecký
parent e307a7a43c
commit ce6986a678
4 changed files with 50 additions and 145 deletions

View File

@@ -6,7 +6,6 @@ package testlapack
import ( import (
"fmt" "fmt"
"math"
"testing" "testing"
"golang.org/x/exp/rand" "golang.org/x/exp/rand"
@@ -39,24 +38,8 @@ func dpbtf2Test(t *testing.T, impl Dpbtf2er, rnd *rand.Rand, uplo blas.Uplo, n,
name := fmt.Sprintf("uplo=%v,n=%v,kd=%v,ldab=%v", string(uplo), n, kd, ldab) name := fmt.Sprintf("uplo=%v,n=%v,kd=%v,ldab=%v", string(uplo), n, kd, ldab)
// Allocate a band matrix and fill it with random numbers. // Generate a random symmetric positive definite band matrix.
ab := make([]float64, n*ldab) ab := randSymBand(uplo, n, kd, ldab, rnd)
for i := range ab {
ab[i] = rnd.NormFloat64()
}
// Make sure that the matrix U or L has a sufficiently positive diagonal.
switch uplo {
case blas.Upper:
for i := 0; i < n; i++ {
ab[i*ldab] = 2 + rnd.Float64()
}
case blas.Lower:
for i := 0; i < n; i++ {
ab[i*ldab+kd] = 2 + rnd.Float64()
}
}
// Compute U^T*U or L*L^T. The resulting (symmetric) matrix A will be positive definite.
dsbmm(uplo, n, kd, ab, ldab)
// Compute the Cholesky decomposition of A. // Compute the Cholesky decomposition of A.
abFac := make([]float64, len(ab)) abFac := make([]float64, len(ab))
@@ -66,30 +49,12 @@ func dpbtf2Test(t *testing.T, impl Dpbtf2er, rnd *rand.Rand, uplo blas.Uplo, n,
t.Fatalf("%v: bad test matrix, Dpbtf2 failed", name) t.Fatalf("%v: bad test matrix, Dpbtf2 failed", name)
} }
if n == 0 {
return
}
// Reconstruct an symmetric band matrix from the U^T*U or L*L^T factorization, overwriting abFac. // Reconstruct an symmetric band matrix from the U^T*U or L*L^T factorization, overwriting abFac.
dsbmm(uplo, n, kd, abFac, ldab) dsbmm(uplo, n, kd, abFac, ldab)
// Compute and check the max-norm distance between the reconstructed and original matrix A. // Compute and check the max-norm distance between the reconstructed and original matrix A.
var diff float64 dist := distSymBand(uplo, n, kd, abFac, ldab, ab, ldab)
switch uplo { if dist > tol {
case blas.Upper: t.Errorf("%v: unexpected result, diff=%v", name, dist)
for i := 0; i < n; i++ {
for j := 0; j < min(kd+1, n-i); j++ {
diff = math.Max(diff, math.Abs(abFac[i*ldab+j]-ab[i*ldab+j]))
}
}
case blas.Lower:
for i := 0; i < n; i++ {
for j := max(0, kd-i); j < kd+1; j++ {
diff = math.Max(diff, math.Abs(abFac[i*ldab+j]-ab[i*ldab+j]))
}
}
}
if diff > tol {
t.Errorf("%v: unexpected result, diff=%v", name, diff)
} }
} }

View File

@@ -6,7 +6,6 @@ package testlapack
import ( import (
"fmt" "fmt"
"math"
"testing" "testing"
"golang.org/x/exp/rand" "golang.org/x/exp/rand"
@@ -44,24 +43,8 @@ func dpbtrfTest(t *testing.T, impl Dpbtrfer, uplo blas.Uplo, n, kd int, ldab int
name := fmt.Sprintf("uplo=%v,n=%v,kd=%v,ldab=%v", string(uplo), n, kd, ldab) name := fmt.Sprintf("uplo=%v,n=%v,kd=%v,ldab=%v", string(uplo), n, kd, ldab)
// Allocate a band matrix and fill it with random numbers. // Generate a random symmetric positive definite band matrix.
ab := make([]float64, n*ldab) ab := randSymBand(uplo, n, kd, ldab, rnd)
for i := range ab {
ab[i] = rnd.NormFloat64()
}
// Make sure that the matrix U or L has a sufficiently positive diagonal.
switch uplo {
case blas.Upper:
for i := 0; i < n; i++ {
ab[i*ldab] = 2 + rnd.Float64()
}
case blas.Lower:
for i := 0; i < n; i++ {
ab[i*ldab+kd] = 2 + rnd.Float64()
}
}
// Compute U^T*U or L*L^T. The resulting (symmetric) matrix A will be positive definite.
dsbmm(uplo, n, kd, ab, ldab)
// Compute the Cholesky decomposition of A. // Compute the Cholesky decomposition of A.
abFac := make([]float64, len(ab)) abFac := make([]float64, len(ab))
@@ -71,31 +54,13 @@ func dpbtrfTest(t *testing.T, impl Dpbtrfer, uplo blas.Uplo, n, kd int, ldab int
t.Fatalf("%v: bad test matrix, Dpbtrf failed", name) t.Fatalf("%v: bad test matrix, Dpbtrf failed", name)
} }
if n == 0 {
return
}
// Reconstruct an symmetric band matrix from the U^T*U or L*L^T factorization, overwriting abFac. // Reconstruct an symmetric band matrix from the U^T*U or L*L^T factorization, overwriting abFac.
dsbmm(uplo, n, kd, abFac, ldab) dsbmm(uplo, n, kd, abFac, ldab)
// Compute and check the max-norm distance between the reconstructed and original matrix A. // Compute and check the max-norm distance between the reconstructed and original matrix A.
var diff float64 dist := distSymBand(uplo, n, kd, abFac, ldab, ab, ldab)
switch uplo { if dist > tol*float64(n) {
case blas.Upper: t.Errorf("%v: unexpected result, diff=%v", name, dist)
for i := 0; i < n; i++ {
for j := 0; j < min(kd+1, n-i); j++ {
diff = math.Max(diff, math.Abs(abFac[i*ldab+j]-ab[i*ldab+j]))
}
}
case blas.Lower:
for i := 0; i < n; i++ {
for j := max(0, kd-i); j < kd+1; j++ {
diff = math.Max(diff, math.Abs(abFac[i*ldab+j]-ab[i*ldab+j]))
}
}
}
if diff > tol {
t.Errorf("%v: unexpected result, diff=%v", name, diff)
} }
} }

View File

@@ -46,25 +46,8 @@ func dpbtrsTest(t *testing.T, impl Dpbtrser, rnd *rand.Rand, uplo blas.Uplo, n,
name := fmt.Sprintf("uplo=%v,n=%v,kd=%v,nrhs=%v,ldab=%v,ldb=%v", string(uplo), n, kd, nrhs, ldab, ldb) name := fmt.Sprintf("uplo=%v,n=%v,kd=%v,nrhs=%v,ldab=%v,ldb=%v", string(uplo), n, kd, nrhs, ldab, ldb)
// Allocate a band matrix and fill it with random numbers. // Generate a random symmetric positive definite band matrix.
ab := make([]float64, n*ldab) ab := randSymBand(uplo, n, kd, ldab, rnd)
for i := range ab {
ab[i] = rnd.NormFloat64()
}
// Make sure that the matrix U or L has a sufficiently positive diagonal.
switch uplo {
case blas.Upper:
for i := 0; i < n; i++ {
ab[i*ldab] = float64(n) + rnd.Float64()
}
case blas.Lower:
for i := 0; i < n; i++ {
ab[i*ldab+kd] = float64(n) + rnd.Float64()
}
}
// Compute U^T*U or L*L^T. The resulting (symmetric) matrix A will be
// positive definite and well-conditioned.
dsbmm(uplo, n, kd, ab, ldab)
// Compute the Cholesky decomposition of A. // Compute the Cholesky decomposition of A.
abFac := make([]float64, len(ab)) abFac := make([]float64, len(ab))

View File

@@ -1031,57 +1031,49 @@ func equalApproxSymmetric(a, b blas64.Symmetric, tol float64) bool {
} }
// randSymBand returns an n×n random symmetric positive definite band matrix // randSymBand returns an n×n random symmetric positive definite band matrix
// with kd diagonals, and the equivalent symmetric dense matrix. // with kd diagonals.
func randSymBand(ul blas.Uplo, n, kd, ldab int, rnd *rand.Rand) (blas64.SymmetricBand, blas64.Symmetric) { func randSymBand(uplo blas.Uplo, n, kd, ldab int, rnd *rand.Rand) []float64 {
if n == 0 { // Allocate a triangular band matrix U or L and fill it with random numbers.
return blas64.SymmetricBand{Uplo: ul, Stride: 1}, blas64.Symmetric{Uplo: ul, Stride: 1} ab := make([]float64, n*ldab)
for i := range ab {
ab[i] = rnd.NormFloat64()
} }
// A matrix is positive definite if and only if it has a Cholesky // Make sure that the matrix U or L has a sufficiently positive diagonal.
// decomposition. Generate a random lower triangular band matrix L with switch uplo {
// strictly positive diagonal, and construct the random symmetric band case blas.Upper:
// matrix as L*L^T.
a := make([]float64, n*n)
for i := 0; i < n; i++ { for i := 0; i < n; i++ {
for j := max(0, i-kd); j <= i; j++ { ab[i*ldab] = float64(n) + rnd.Float64()
a[i*n+j] = rnd.NormFloat64()
} }
a[i*n+i] = math.Abs(a[i*n+i]) case blas.Lower:
// Add an extra amount to the diagonal in order to improve the condition number. for i := 0; i < n; i++ {
a[i*n+i] += 1.5 * rnd.Float64() ab[i*ldab+kd] = float64(n) + rnd.Float64()
} }
agen := blas64.General{
Rows: n,
Cols: n,
Stride: n,
Data: a,
} }
// Compute U^T*U or L*L^T. The resulting (symmetric) matrix A will be
// positive definite and well-conditioned.
dsbmm(uplo, n, kd, ab, ldab)
return ab
}
// Construct the SymDense from a*a^T // distSymBand returns the max-norm distance between the symmetric band matrices
c := make([]float64, n*n) // A and B.
cgen := blas64.General{ func distSymBand(uplo blas.Uplo, n, kd int, a []float64, lda int, b []float64, ldb int) float64 {
Rows: n, var dist float64
Cols: n, switch uplo {
Stride: n, case blas.Upper:
Data: c, for i := 0; i < n; i++ {
for j := 0; j < min(kd+1, n-i); j++ {
dist = math.Max(dist, math.Abs(a[i*lda+j]-b[i*ldb+j]))
} }
blas64.Gemm(blas.NoTrans, blas.Trans, 1, agen, agen, 0, cgen)
sym := blas64.Symmetric{
N: n,
Stride: n,
Data: c,
Uplo: ul,
} }
case blas.Lower:
b := symToSymBand(ul, c, n, n, kd, ldab) for i := 0; i < n; i++ {
band := blas64.SymmetricBand{ for j := max(0, kd-i); j < kd+1; j++ {
N: n, dist = math.Max(dist, math.Abs(a[i*lda+j]-b[i*ldb+j]))
K: kd,
Stride: ldab,
Data: b,
Uplo: ul,
} }
}
return band, sym }
return dist
} }
// symToSymBand takes the data in a Symmetric matrix and returns a // symToSymBand takes the data in a Symmetric matrix and returns a