LAPACK 3.3.0
|
00001 SUBROUTINE SSTECT( N, A, B, SHIFT, NUM ) 00002 * 00003 * -- LAPACK test routine (version 3.1) -- 00004 * Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. 00005 * November 2006 00006 * 00007 * .. Scalar Arguments .. 00008 INTEGER N, NUM 00009 REAL SHIFT 00010 * .. 00011 * .. Array Arguments .. 00012 REAL A( * ), B( * ) 00013 * .. 00014 * 00015 * Purpose 00016 * ======= 00017 * 00018 * SSTECT counts the number NUM of eigenvalues of a tridiagonal 00019 * matrix T which are less than or equal to SHIFT. T has 00020 * diagonal entries A(1), ... , A(N), and offdiagonal entries 00021 * B(1), ..., B(N-1). 00022 * See W. Kahan "Accurate Eigenvalues of a Symmetric Tridiagonal 00023 * Matrix", Report CS41, Computer Science Dept., Stanford 00024 * University, July 21, 1966 00025 * 00026 * Arguments 00027 * ========= 00028 * 00029 * N (input) INTEGER 00030 * The dimension of the tridiagonal matrix T. 00031 * 00032 * A (input) REAL array, dimension (N) 00033 * The diagonal entries of the tridiagonal matrix T. 00034 * 00035 * B (input) REAL array, dimension (N-1) 00036 * The offdiagonal entries of the tridiagonal matrix T. 00037 * 00038 * SHIFT (input) REAL 00039 * The shift, used as described under Purpose. 00040 * 00041 * NUM (output) INTEGER 00042 * The number of eigenvalues of T less than or equal 00043 * to SHIFT. 00044 * 00045 * ===================================================================== 00046 * 00047 * .. Parameters .. 00048 REAL ZERO, ONE, THREE 00049 PARAMETER ( ZERO = 0.0E0, ONE = 1.0E0, THREE = 3.0E0 ) 00050 * .. 00051 * .. Local Scalars .. 00052 INTEGER I 00053 REAL M1, M2, MX, OVFL, SOV, SSHIFT, SSUN, SUN, TMP, 00054 $ TOM, U, UNFL 00055 * .. 00056 * .. External Functions .. 00057 REAL SLAMCH 00058 EXTERNAL SLAMCH 00059 * .. 00060 * .. Intrinsic Functions .. 00061 INTRINSIC ABS, MAX, SQRT 00062 * .. 00063 * .. Executable Statements .. 00064 * 00065 * Get machine constants 00066 * 00067 UNFL = SLAMCH( 'Safe minimum' ) 00068 OVFL = SLAMCH( 'Overflow' ) 00069 * 00070 * Find largest entry 00071 * 00072 MX = ABS( A( 1 ) ) 00073 DO 10 I = 1, N - 1 00074 MX = MAX( MX, ABS( A( I+1 ) ), ABS( B( I ) ) ) 00075 10 CONTINUE 00076 * 00077 * Handle easy cases, including zero matrix 00078 * 00079 IF( SHIFT.GE.THREE*MX ) THEN 00080 NUM = N 00081 RETURN 00082 END IF 00083 IF( SHIFT.LT.-THREE*MX ) THEN 00084 NUM = 0 00085 RETURN 00086 END IF 00087 * 00088 * Compute scale factors as in Kahan's report 00089 * At this point, MX .NE. 0 so we can divide by it 00090 * 00091 SUN = SQRT( UNFL ) 00092 SSUN = SQRT( SUN ) 00093 SOV = SQRT( OVFL ) 00094 TOM = SSUN*SOV 00095 IF( MX.LE.ONE ) THEN 00096 M1 = ONE / MX 00097 M2 = TOM 00098 ELSE 00099 M1 = ONE 00100 M2 = TOM / MX 00101 END IF 00102 * 00103 * Begin counting 00104 * 00105 NUM = 0 00106 SSHIFT = ( SHIFT*M1 )*M2 00107 U = ( A( 1 )*M1 )*M2 - SSHIFT 00108 IF( U.LE.SUN ) THEN 00109 IF( U.LE.ZERO ) THEN 00110 NUM = NUM + 1 00111 IF( U.GT.-SUN ) 00112 $ U = -SUN 00113 ELSE 00114 U = SUN 00115 END IF 00116 END IF 00117 DO 20 I = 2, N 00118 TMP = ( B( I-1 )*M1 )*M2 00119 U = ( ( A( I )*M1 )*M2-TMP*( TMP / U ) ) - SSHIFT 00120 IF( U.LE.SUN ) THEN 00121 IF( U.LE.ZERO ) THEN 00122 NUM = NUM + 1 00123 IF( U.GT.-SUN ) 00124 $ U = -SUN 00125 ELSE 00126 U = SUN 00127 END IF 00128 END IF 00129 20 CONTINUE 00130 RETURN 00131 * 00132 * End of SSTECT 00133 * 00134 END