1 @cindex linear algebra, BLAS
2 @cindex matrix, operations
3 @cindex vector, operations
6 @cindex Basic Linear Algebra Subroutines (BLAS)
8 The Basic Linear Algebra Subprograms (@sc{blas}) define a set of fundamental
9 operations on vectors and matrices which can be used to create optimized
10 higher-level linear algebra functionality.
12 The library provides a low-level layer which corresponds directly to the
13 C-language @sc{blas} standard, referred to here as ``@sc{cblas}'', and a
14 higher-level interface for operations on GSL vectors and matrices.
15 Users who are interested in simple operations on GSL vector and matrix
16 objects should use the high-level layer, which is declared in the file
17 @code{gsl_blas.h}. This should satisfy the needs of most users. Note
18 that GSL matrices are implemented using dense-storage so the interface
19 only includes the corresponding dense-storage @sc{blas} functions. The full
20 @sc{blas} functionality for band-format and packed-format matrices is
21 available through the low-level @sc{cblas} interface.
23 The interface for the @code{gsl_cblas} layer is specified in the file
24 @code{gsl_cblas.h}. This interface corresponds to the @sc{blas} Technical
25 Forum's draft standard for the C interface to legacy @sc{blas}
26 implementations. Users who have access to other conforming @sc{cblas}
27 implementations can use these in place of the version provided by the
28 library. Note that users who have only a Fortran @sc{blas} library can
29 use a @sc{cblas} conformant wrapper to convert it into a @sc{cblas}
30 library. A reference @sc{cblas} wrapper for legacy Fortran
31 implementations exists as part of the draft @sc{cblas} standard and can
32 be obtained from Netlib. The complete set of @sc{cblas} functions is
33 listed in an appendix (@pxref{GSL CBLAS Library}).
35 There are three levels of @sc{blas} operations,
39 Vector operations, e.g. @math{y = \alpha x + y}
41 Matrix-vector operations, e.g. @math{y = \alpha A x + \beta y}
43 Matrix-matrix operations, e.g. @math{C = \alpha A B + C}
47 Each routine has a name which specifies the operation, the type of
48 matrices involved and their precisions. Some of the most common
49 operations and their names are given below,
53 scalar product, @math{x^T y}
55 vector sum, @math{\alpha x + y}
57 matrix-vector product, @math{A x}
59 matrix-vector solve, @math{inv(A) x}
61 matrix-matrix product, @math{A B}
63 matrix-matrix solve, @math{inv(A) B}
67 The types of matrices are,
95 Each operation is defined for four precisions,
109 Thus, for example, the name @sc{sgemm} stands for ``single-precision
110 general matrix-matrix multiply'' and @sc{zgemm} stands for
111 ``double-precision complex matrix-matrix multiply''.
114 * GSL BLAS Interface::
116 * BLAS References and Further Reading::
119 @node GSL BLAS Interface
120 @section GSL BLAS Interface
122 GSL provides dense vector and matrix objects, based on the relevant
123 built-in types. The library provides an interface to the @sc{blas}
124 operations which apply to these objects. The interface to this
125 functionality is given in the file @code{gsl_blas.h}.
127 @comment CblasNoTrans, CblasTrans, CblasConjTrans
128 @comment CblasUpper, CblasLower
129 @comment CblasNonUnit, CblasUnit
130 @comment CblasLeft, CblasRight
133 * Level 1 GSL BLAS Interface::
134 * Level 2 GSL BLAS Interface::
135 * Level 3 GSL BLAS Interface::
138 @node Level 1 GSL BLAS Interface
141 @deftypefun int gsl_blas_sdsdot (float @var{alpha}, const gsl_vector_float * @var{x}, const gsl_vector_float * @var{y}, float * @var{result})
142 @cindex DOT, Level-1 BLAS
143 This function computes the sum @math{\alpha + x^T y} for the vectors
144 @var{x} and @var{y}, returning the result in @var{result}.
147 @deftypefun int gsl_blas_sdot (const gsl_vector_float * @var{x}, const gsl_vector_float * @var{y}, float * @var{result})
148 @deftypefunx int gsl_blas_dsdot (const gsl_vector_float * @var{x}, const gsl_vector_float * @var{y}, double * @var{result})
149 @deftypefunx int gsl_blas_ddot (const gsl_vector * @var{x}, const gsl_vector * @var{y}, double * @var{result})
150 These functions compute the scalar product @math{x^T y} for the vectors
151 @var{x} and @var{y}, returning the result in @var{result}.
154 @deftypefun int gsl_blas_cdotu (const gsl_vector_complex_float * @var{x}, const gsl_vector_complex_float * @var{y}, gsl_complex_float * @var{dotu})
155 @deftypefunx int gsl_blas_zdotu (const gsl_vector_complex * @var{x}, const gsl_vector_complex * @var{y}, gsl_complex * @var{dotu})
156 These functions compute the complex scalar product @math{x^T y} for the
157 vectors @var{x} and @var{y}, returning the result in @var{result}
160 @deftypefun int gsl_blas_cdotc (const gsl_vector_complex_float * @var{x}, const gsl_vector_complex_float * @var{y}, gsl_complex_float * @var{dotc})
161 @deftypefunx int gsl_blas_zdotc (const gsl_vector_complex * @var{x}, const gsl_vector_complex * @var{y}, gsl_complex * @var{dotc})
162 These functions compute the complex conjugate scalar product @math{x^H
163 y} for the vectors @var{x} and @var{y}, returning the result in
167 @deftypefun float gsl_blas_snrm2 (const gsl_vector_float * @var{x})
168 @deftypefunx double gsl_blas_dnrm2 (const gsl_vector * @var{x})
169 @cindex NRM2, Level-1 BLAS
170 These functions compute the Euclidean norm
171 @c{$||x||_2 = \sqrt{\sum x_i^2}$}
172 @math{||x||_2 = \sqrt @{\sum x_i^2@}} of the vector @var{x}.
175 @deftypefun float gsl_blas_scnrm2 (const gsl_vector_complex_float * @var{x})
176 @deftypefunx double gsl_blas_dznrm2 (const gsl_vector_complex * @var{x})
177 These functions compute the Euclidean norm of the complex vector @var{x},
181 ||x||_2 = \sqrt{\sum (\Re(x_i)^2 + \Im(x_i)^2)}.
188 ||x||_2 = \sqrt @{\sum (\Re(x_i)^2 + \Im(x_i)^2)@}.
193 @deftypefun float gsl_blas_sasum (const gsl_vector_float * @var{x})
194 @deftypefunx double gsl_blas_dasum (const gsl_vector * @var{x})
195 @cindex ASUM, Level-1 BLAS
196 These functions compute the absolute sum @math{\sum |x_i|} of the
197 elements of the vector @var{x}.
200 @deftypefun float gsl_blas_scasum (const gsl_vector_complex_float * @var{x})
201 @deftypefunx double gsl_blas_dzasum (const gsl_vector_complex * @var{x})
202 These functions compute the sum of the magnitudes of the real and
203 imaginary parts of the complex vector @var{x},
204 @c{$\sum \left( |\Re(x_i)| + |\Im(x_i)| \right)$}
205 @math{\sum |\Re(x_i)| + |\Im(x_i)|}.
208 @deftypefun CBLAS_INDEX_t gsl_blas_isamax (const gsl_vector_float * @var{x})
209 @deftypefunx CBLAS_INDEX_t gsl_blas_idamax (const gsl_vector * @var{x})
210 @deftypefunx CBLAS_INDEX_t gsl_blas_icamax (const gsl_vector_complex_float * @var{x})
211 @deftypefunx CBLAS_INDEX_t gsl_blas_izamax (const gsl_vector_complex * @var{x})
212 @cindex AMAX, Level-1 BLAS
213 These functions return the index of the largest element of the vector
214 @var{x}. The largest element is determined by its absolute magnitude for
215 real vectors and by the sum of the magnitudes of the real and imaginary
216 parts @math{|\Re(x_i)| + |\Im(x_i)|} for complex vectors. If the
217 largest value occurs several times then the index of the first
218 occurrence is returned.
221 @deftypefun int gsl_blas_sswap (gsl_vector_float * @var{x}, gsl_vector_float * @var{y})
222 @deftypefunx int gsl_blas_dswap (gsl_vector * @var{x}, gsl_vector * @var{y})
223 @deftypefunx int gsl_blas_cswap (gsl_vector_complex_float * @var{x}, gsl_vector_complex_float * @var{y})
224 @deftypefunx int gsl_blas_zswap (gsl_vector_complex * @var{x}, gsl_vector_complex * @var{y})
225 @cindex SWAP, Level-1 BLAS
226 These functions exchange the elements of the vectors @var{x} and @var{y}.
229 @deftypefun int gsl_blas_scopy (const gsl_vector_float * @var{x}, gsl_vector_float * @var{y})
230 @deftypefunx int gsl_blas_dcopy (const gsl_vector * @var{x}, gsl_vector * @var{y})
231 @deftypefunx int gsl_blas_ccopy (const gsl_vector_complex_float * @var{x}, gsl_vector_complex_float * @var{y})
232 @deftypefunx int gsl_blas_zcopy (const gsl_vector_complex * @var{x}, gsl_vector_complex * @var{y})
233 @cindex COPY, Level-1 BLAS
234 These functions copy the elements of the vector @var{x} into the vector
239 @deftypefun int gsl_blas_saxpy (float @var{alpha}, const gsl_vector_float * @var{x}, gsl_vector_float * @var{y})
240 @deftypefunx int gsl_blas_daxpy (double @var{alpha}, const gsl_vector * @var{x}, gsl_vector * @var{y})
241 @deftypefunx int gsl_blas_caxpy (const gsl_complex_float @var{alpha}, const gsl_vector_complex_float * @var{x}, gsl_vector_complex_float * @var{y})
242 @deftypefunx int gsl_blas_zaxpy (const gsl_complex @var{alpha}, const gsl_vector_complex * @var{x}, gsl_vector_complex * @var{y})
243 @cindex AXPY, Level-1 BLAS
244 @cindex DAXPY, Level-1 BLAS
245 @cindex SAXPY, Level-1 BLAS
246 These functions compute the sum @math{y = \alpha x + y} for the vectors
250 @deftypefun void gsl_blas_sscal (float @var{alpha}, gsl_vector_float * @var{x})
251 @deftypefunx void gsl_blas_dscal (double @var{alpha}, gsl_vector * @var{x})
252 @deftypefunx void gsl_blas_cscal (const gsl_complex_float @var{alpha}, gsl_vector_complex_float * @var{x})
253 @deftypefunx void gsl_blas_zscal (const gsl_complex @var{alpha}, gsl_vector_complex * @var{x})
254 @deftypefunx void gsl_blas_csscal (float @var{alpha}, gsl_vector_complex_float * @var{x})
255 @deftypefunx void gsl_blas_zdscal (double @var{alpha}, gsl_vector_complex * @var{x})
256 @cindex SCAL, Level-1 BLAS
257 These functions rescale the vector @var{x} by the multiplicative factor
261 @deftypefun int gsl_blas_srotg (float @var{a}[], float @var{b}[], float @var{c}[], float @var{s}[])
262 @deftypefunx int gsl_blas_drotg (double @var{a}[], double @var{b}[], double @var{c}[], double @var{s}[])
263 @cindex ROTG, Level-1 BLAS
264 @cindex Givens Rotation, BLAS
265 These functions compute a Givens rotation @math{(c,s)} which zeroes the
289 [ c s ] [ a ] = [ r ]
295 The variables @var{a} and @var{b} are overwritten by the routine.
298 @deftypefun int gsl_blas_srot (gsl_vector_float * @var{x}, gsl_vector_float * @var{y}, float @var{c}, float @var{s})
299 @deftypefunx int gsl_blas_drot (gsl_vector * @var{x}, gsl_vector * @var{y}, const double @var{c}, const double @var{s})
300 These functions apply a Givens rotation @math{(x', y') = (c x + s y, -s
301 x + c y)} to the vectors @var{x}, @var{y}.
304 @deftypefun int gsl_blas_srotmg (float @var{d1}[], float @var{d2}[], float @var{b1}[], float @var{b2}, float @var{P}[])
305 @deftypefunx int gsl_blas_drotmg (double @var{d1}[], double @var{d2}[], double @var{b1}[], double @var{b2}, double @var{P}[])
306 @cindex Modified Givens Rotation, BLAS
307 @cindex Givens Rotation, Modified, BLAS
308 These functions compute a modified Givens transformation. The modified
309 Givens transformation is defined in the original Level-1 @sc{blas}
310 specification, given in the references.
313 @deftypefun int gsl_blas_srotm (gsl_vector_float * @var{x}, gsl_vector_float * @var{y}, const float @var{P}[])
314 @deftypefunx int gsl_blas_drotm (gsl_vector * @var{x}, gsl_vector * @var{y}, const double @var{P}[])
315 These functions apply a modified Givens transformation.
318 @node Level 2 GSL BLAS Interface
321 @deftypefun int gsl_blas_sgemv (CBLAS_TRANSPOSE_t @var{TransA}, float @var{alpha}, const gsl_matrix_float * @var{A}, const gsl_vector_float * @var{x}, float @var{beta}, gsl_vector_float * @var{y})
322 @deftypefunx int gsl_blas_dgemv (CBLAS_TRANSPOSE_t @var{TransA}, double @var{alpha}, const gsl_matrix * @var{A}, const gsl_vector * @var{x}, double @var{beta}, gsl_vector * @var{y})
323 @deftypefunx int gsl_blas_cgemv (CBLAS_TRANSPOSE_t @var{TransA}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, const gsl_vector_complex_float * @var{x}, const gsl_complex_float @var{beta}, gsl_vector_complex_float * @var{y})
324 @deftypefunx int gsl_blas_zgemv (CBLAS_TRANSPOSE_t @var{TransA}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, const gsl_vector_complex * @var{x}, const gsl_complex @var{beta}, gsl_vector_complex * @var{y})
325 @cindex GEMV, Level-2 BLAS
326 These functions compute the matrix-vector product and sum @math{y =
327 \alpha op(A) x + \beta y}, where @math{op(A) = A},
328 @math{A^T}, @math{A^H} for @var{TransA} = @code{CblasNoTrans},
329 @code{CblasTrans}, @code{CblasConjTrans}.
333 @deftypefun int gsl_blas_strmv (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_matrix_float * @var{A}, gsl_vector_float * @var{x})
334 @deftypefunx int gsl_blas_dtrmv (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_matrix * @var{A}, gsl_vector * @var{x})
335 @deftypefunx int gsl_blas_ctrmv (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_matrix_complex_float * @var{A}, gsl_vector_complex_float * @var{x})
336 @deftypefunx int gsl_blas_ztrmv (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_matrix_complex * @var{A}, gsl_vector_complex * @var{x})
337 @cindex TRMV, Level-2 BLAS
338 These functions compute the matrix-vector product
339 @math{x = op(A) x} for the triangular matrix @var{A}, where
340 @math{op(A) = A}, @math{A^T}, @math{A^H} for @var{TransA} =
341 @code{CblasNoTrans}, @code{CblasTrans}, @code{CblasConjTrans}. When
342 @var{Uplo} is @code{CblasUpper} then the upper triangle of @var{A} is
343 used, and when @var{Uplo} is @code{CblasLower} then the lower triangle
344 of @var{A} is used. If @var{Diag} is @code{CblasNonUnit} then the
345 diagonal of the matrix is used, but if @var{Diag} is @code{CblasUnit}
346 then the diagonal elements of the matrix @var{A} are taken as unity and
351 @deftypefun int gsl_blas_strsv (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_matrix_float * @var{A}, gsl_vector_float * @var{x})
352 @deftypefunx int gsl_blas_dtrsv (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_matrix * @var{A}, gsl_vector * @var{x})
353 @deftypefunx int gsl_blas_ctrsv (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_matrix_complex_float * @var{A}, gsl_vector_complex_float * @var{x})
354 @deftypefunx int gsl_blas_ztrsv (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_matrix_complex * @var{A}, gsl_vector_complex * @var{x})
355 @cindex TRSV, Level-2 BLAS
356 These functions compute @math{inv(op(A)) x} for @var{x}, where
357 @math{op(A) = A}, @math{A^T}, @math{A^H} for @var{TransA} =
358 @code{CblasNoTrans}, @code{CblasTrans}, @code{CblasConjTrans}. When
359 @var{Uplo} is @code{CblasUpper} then the upper triangle of @var{A} is
360 used, and when @var{Uplo} is @code{CblasLower} then the lower triangle
361 of @var{A} is used. If @var{Diag} is @code{CblasNonUnit} then the
362 diagonal of the matrix is used, but if @var{Diag} is @code{CblasUnit}
363 then the diagonal elements of the matrix @var{A} are taken as unity and
368 @deftypefun int gsl_blas_ssymv (CBLAS_UPLO_t @var{Uplo}, float @var{alpha}, const gsl_matrix_float * @var{A}, const gsl_vector_float * @var{x}, float @var{beta}, gsl_vector_float * @var{y})
369 @deftypefunx int gsl_blas_dsymv (CBLAS_UPLO_t @var{Uplo}, double @var{alpha}, const gsl_matrix * @var{A}, const gsl_vector * @var{x}, double @var{beta}, gsl_vector * @var{y})
370 @cindex SYMV, Level-2 BLAS
371 These functions compute the matrix-vector product and sum @math{y =
372 \alpha A x + \beta y} for the symmetric matrix @var{A}. Since the
373 matrix @var{A} is symmetric only its upper half or lower half need to be
374 stored. When @var{Uplo} is @code{CblasUpper} then the upper triangle
375 and diagonal of @var{A} are used, and when @var{Uplo} is
376 @code{CblasLower} then the lower triangle and diagonal of @var{A} are
380 @deftypefun int gsl_blas_chemv (CBLAS_UPLO_t @var{Uplo}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, const gsl_vector_complex_float * @var{x}, const gsl_complex_float @var{beta}, gsl_vector_complex_float * @var{y})
381 @deftypefunx int gsl_blas_zhemv (CBLAS_UPLO_t @var{Uplo}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, const gsl_vector_complex * @var{x}, const gsl_complex @var{beta}, gsl_vector_complex * @var{y})
382 @cindex HEMV, Level-2 BLAS
383 These functions compute the matrix-vector product and sum @math{y =
384 \alpha A x + \beta y} for the hermitian matrix @var{A}. Since the
385 matrix @var{A} is hermitian only its upper half or lower half need to be
386 stored. When @var{Uplo} is @code{CblasUpper} then the upper triangle
387 and diagonal of @var{A} are used, and when @var{Uplo} is
388 @code{CblasLower} then the lower triangle and diagonal of @var{A} are
389 used. The imaginary elements of the diagonal are automatically assumed
390 to be zero and are not referenced.
393 @deftypefun int gsl_blas_sger (float @var{alpha}, const gsl_vector_float * @var{x}, const gsl_vector_float * @var{y}, gsl_matrix_float * @var{A})
394 @deftypefunx int gsl_blas_dger (double @var{alpha}, const gsl_vector * @var{x}, const gsl_vector * @var{y}, gsl_matrix * @var{A})
395 @deftypefunx int gsl_blas_cgeru (const gsl_complex_float @var{alpha}, const gsl_vector_complex_float * @var{x}, const gsl_vector_complex_float * @var{y}, gsl_matrix_complex_float * @var{A})
396 @deftypefunx int gsl_blas_zgeru (const gsl_complex @var{alpha}, const gsl_vector_complex * @var{x}, const gsl_vector_complex * @var{y}, gsl_matrix_complex * @var{A})
397 @cindex GER, Level-2 BLAS
398 @cindex GERU, Level-2 BLAS
399 These functions compute the rank-1 update @math{A = \alpha x y^T + A} of
403 @deftypefun int gsl_blas_cgerc (const gsl_complex_float @var{alpha}, const gsl_vector_complex_float * @var{x}, const gsl_vector_complex_float * @var{y}, gsl_matrix_complex_float * @var{A})
404 @deftypefunx int gsl_blas_zgerc (const gsl_complex @var{alpha}, const gsl_vector_complex * @var{x}, const gsl_vector_complex * @var{y}, gsl_matrix_complex * @var{A})
405 @cindex GERC, Level-2 BLAS
406 These functions compute the conjugate rank-1 update @math{A = \alpha x
407 y^H + A} of the matrix @var{A}.
410 @deftypefun int gsl_blas_ssyr (CBLAS_UPLO_t @var{Uplo}, float @var{alpha}, const gsl_vector_float * @var{x}, gsl_matrix_float * @var{A})
411 @deftypefunx int gsl_blas_dsyr (CBLAS_UPLO_t @var{Uplo}, double @var{alpha}, const gsl_vector * @var{x}, gsl_matrix * @var{A})
412 @cindex SYR, Level-2 BLAS
413 These functions compute the symmetric rank-1 update @math{A = \alpha x
414 x^T + A} of the symmetric matrix @var{A}. Since the matrix @var{A} is
415 symmetric only its upper half or lower half need to be stored. When
416 @var{Uplo} is @code{CblasUpper} then the upper triangle and diagonal of
417 @var{A} are used, and when @var{Uplo} is @code{CblasLower} then the
418 lower triangle and diagonal of @var{A} are used.
421 @deftypefun int gsl_blas_cher (CBLAS_UPLO_t @var{Uplo}, float @var{alpha}, const gsl_vector_complex_float * @var{x}, gsl_matrix_complex_float * @var{A})
422 @deftypefunx int gsl_blas_zher (CBLAS_UPLO_t @var{Uplo}, double @var{alpha}, const gsl_vector_complex * @var{x}, gsl_matrix_complex * @var{A})
423 @cindex HER, Level-2 BLAS
424 These functions compute the hermitian rank-1 update @math{A = \alpha x
425 x^H + A} of the hermitian matrix @var{A}. Since the matrix @var{A} is
426 hermitian only its upper half or lower half need to be stored. When
427 @var{Uplo} is @code{CblasUpper} then the upper triangle and diagonal of
428 @var{A} are used, and when @var{Uplo} is @code{CblasLower} then the
429 lower triangle and diagonal of @var{A} are used. The imaginary elements
430 of the diagonal are automatically set to zero.
433 @deftypefun int gsl_blas_ssyr2 (CBLAS_UPLO_t @var{Uplo}, float @var{alpha}, const gsl_vector_float * @var{x}, const gsl_vector_float * @var{y}, gsl_matrix_float * @var{A})
434 @deftypefunx int gsl_blas_dsyr2 (CBLAS_UPLO_t @var{Uplo}, double @var{alpha}, const gsl_vector * @var{x}, const gsl_vector * @var{y}, gsl_matrix * @var{A})
435 @cindex SYR2, Level-2 BLAS
436 These functions compute the symmetric rank-2 update @math{A = \alpha x
437 y^T + \alpha y x^T + A} of the symmetric matrix @var{A}. Since the
438 matrix @var{A} is symmetric only its upper half or lower half need to be
439 stored. When @var{Uplo} is @code{CblasUpper} then the upper triangle
440 and diagonal of @var{A} are used, and when @var{Uplo} is
441 @code{CblasLower} then the lower triangle and diagonal of @var{A} are
445 @deftypefun int gsl_blas_cher2 (CBLAS_UPLO_t @var{Uplo}, const gsl_complex_float @var{alpha}, const gsl_vector_complex_float * @var{x}, const gsl_vector_complex_float * @var{y}, gsl_matrix_complex_float * @var{A})
446 @deftypefunx int gsl_blas_zher2 (CBLAS_UPLO_t @var{Uplo}, const gsl_complex @var{alpha}, const gsl_vector_complex * @var{x}, const gsl_vector_complex * @var{y}, gsl_matrix_complex * @var{A})
447 @cindex HER2, Level-2 BLAS
448 These functions compute the hermitian rank-2 update @math{A = \alpha x
449 y^H + \alpha^* y x^H A} of the hermitian matrix @var{A}. Since the
450 matrix @var{A} is hermitian only its upper half or lower half need to be
451 stored. When @var{Uplo} is @code{CblasUpper} then the upper triangle
452 and diagonal of @var{A} are used, and when @var{Uplo} is
453 @code{CblasLower} then the lower triangle and diagonal of @var{A} are
454 used. The imaginary elements of the diagonal are automatically set to zero.
457 @node Level 3 GSL BLAS Interface
461 @deftypefun int gsl_blas_sgemm (CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_TRANSPOSE_t @var{TransB}, float @var{alpha}, const gsl_matrix_float * @var{A}, const gsl_matrix_float * @var{B}, float @var{beta}, gsl_matrix_float * @var{C})
462 @deftypefunx int gsl_blas_dgemm (CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_TRANSPOSE_t @var{TransB}, double @var{alpha}, const gsl_matrix * @var{A}, const gsl_matrix * @var{B}, double @var{beta}, gsl_matrix * @var{C})
463 @deftypefunx int gsl_blas_cgemm (CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_TRANSPOSE_t @var{TransB}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, const gsl_matrix_complex_float * @var{B}, const gsl_complex_float @var{beta}, gsl_matrix_complex_float * @var{C})
464 @deftypefunx int gsl_blas_zgemm (CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_TRANSPOSE_t @var{TransB}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, const gsl_matrix_complex * @var{B}, const gsl_complex @var{beta}, gsl_matrix_complex * @var{C})
465 @cindex GEMM, Level-3 BLAS
466 These functions compute the matrix-matrix product and sum @math{C =
467 \alpha op(A) op(B) + \beta C} where @math{op(A) = A}, @math{A^T},
468 @math{A^H} for @var{TransA} = @code{CblasNoTrans}, @code{CblasTrans},
469 @code{CblasConjTrans} and similarly for the parameter @var{TransB}.
473 @deftypefun int gsl_blas_ssymm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, float @var{alpha}, const gsl_matrix_float * @var{A}, const gsl_matrix_float * @var{B}, float @var{beta}, gsl_matrix_float * @var{C})
474 @deftypefunx int gsl_blas_dsymm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, double @var{alpha}, const gsl_matrix * @var{A}, const gsl_matrix * @var{B}, double @var{beta}, gsl_matrix * @var{C})
475 @deftypefunx int gsl_blas_csymm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, const gsl_matrix_complex_float * @var{B}, const gsl_complex_float @var{beta}, gsl_matrix_complex_float * @var{C})
476 @deftypefunx int gsl_blas_zsymm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, const gsl_matrix_complex * @var{B}, const gsl_complex @var{beta}, gsl_matrix_complex * @var{C})
477 @cindex SYMM, Level-3 BLAS
478 These functions compute the matrix-matrix product and sum @math{C =
479 \alpha A B + \beta C} for @var{Side} is @code{CblasLeft} and @math{C =
480 \alpha B A + \beta C} for @var{Side} is @code{CblasRight}, where the
481 matrix @var{A} is symmetric. When @var{Uplo} is @code{CblasUpper} then
482 the upper triangle and diagonal of @var{A} are used, and when @var{Uplo}
483 is @code{CblasLower} then the lower triangle and diagonal of @var{A} are
487 @deftypefun int gsl_blas_chemm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, const gsl_matrix_complex_float * @var{B}, const gsl_complex_float @var{beta}, gsl_matrix_complex_float * @var{C})
488 @deftypefunx int gsl_blas_zhemm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, const gsl_matrix_complex * @var{B}, const gsl_complex @var{beta}, gsl_matrix_complex * @var{C})
489 @cindex HEMM, Level-3 BLAS
490 These functions compute the matrix-matrix product and sum @math{C =
491 \alpha A B + \beta C} for @var{Side} is @code{CblasLeft} and @math{C =
492 \alpha B A + \beta C} for @var{Side} is @code{CblasRight}, where the
493 matrix @var{A} is hermitian. When @var{Uplo} is @code{CblasUpper} then
494 the upper triangle and diagonal of @var{A} are used, and when @var{Uplo}
495 is @code{CblasLower} then the lower triangle and diagonal of @var{A} are
496 used. The imaginary elements of the diagonal are automatically set to
500 @deftypefun int gsl_blas_strmm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, float @var{alpha}, const gsl_matrix_float * @var{A}, gsl_matrix_float * @var{B})
501 @deftypefunx int gsl_blas_dtrmm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, double @var{alpha}, const gsl_matrix * @var{A}, gsl_matrix * @var{B})
502 @deftypefunx int gsl_blas_ctrmm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, gsl_matrix_complex_float * @var{B})
503 @deftypefunx int gsl_blas_ztrmm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, gsl_matrix_complex * @var{B})
504 @cindex TRMM, Level-3 BLAS
505 These functions compute the matrix-matrix product @math{B = \alpha op(A)
506 B} for @var{Side} is @code{CblasLeft} and @math{B = \alpha B op(A)} for
507 @var{Side} is @code{CblasRight}. The matrix @var{A} is triangular and
508 @math{op(A) = A}, @math{A^T}, @math{A^H} for @var{TransA} =
509 @code{CblasNoTrans}, @code{CblasTrans}, @code{CblasConjTrans}. When
510 @var{Uplo} is @code{CblasUpper} then the upper triangle of @var{A} is
511 used, and when @var{Uplo} is @code{CblasLower} then the lower triangle
512 of @var{A} is used. If @var{Diag} is @code{CblasNonUnit} then the
513 diagonal of @var{A} is used, but if @var{Diag} is @code{CblasUnit} then
514 the diagonal elements of the matrix @var{A} are taken as unity and are
519 @deftypefun int gsl_blas_strsm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, float @var{alpha}, const gsl_matrix_float * @var{A}, gsl_matrix_float * @var{B})
520 @deftypefunx int gsl_blas_dtrsm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, double @var{alpha}, const gsl_matrix * @var{A}, gsl_matrix * @var{B})
521 @deftypefunx int gsl_blas_ctrsm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, gsl_matrix_complex_float * @var{B})
522 @deftypefunx int gsl_blas_ztrsm (CBLAS_SIDE_t @var{Side}, CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{TransA}, CBLAS_DIAG_t @var{Diag}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, gsl_matrix_complex * @var{B})
523 @cindex TRSM, Level-3 BLAS
524 These functions compute the inverse-matrix matrix product
525 @math{B = \alpha op(inv(A))B} for @var{Side} is
526 @code{CblasLeft} and @math{B = \alpha B op(inv(A))} for
527 @var{Side} is @code{CblasRight}. The matrix @var{A} is triangular and
528 @math{op(A) = A}, @math{A^T}, @math{A^H} for @var{TransA} =
529 @code{CblasNoTrans}, @code{CblasTrans}, @code{CblasConjTrans}. When
530 @var{Uplo} is @code{CblasUpper} then the upper triangle of @var{A} is
531 used, and when @var{Uplo} is @code{CblasLower} then the lower triangle
532 of @var{A} is used. If @var{Diag} is @code{CblasNonUnit} then the
533 diagonal of @var{A} is used, but if @var{Diag} is @code{CblasUnit} then
534 the diagonal elements of the matrix @var{A} are taken as unity and are
538 @deftypefun int gsl_blas_ssyrk (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, float @var{alpha}, const gsl_matrix_float * @var{A}, float @var{beta}, gsl_matrix_float * @var{C})
539 @deftypefunx int gsl_blas_dsyrk (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, double @var{alpha}, const gsl_matrix * @var{A}, double @var{beta}, gsl_matrix * @var{C})
540 @deftypefunx int gsl_blas_csyrk (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, const gsl_complex_float @var{beta}, gsl_matrix_complex_float * @var{C})
541 @deftypefunx int gsl_blas_zsyrk (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, const gsl_complex @var{beta}, gsl_matrix_complex * @var{C})
542 @cindex SYRK, Level-3 BLAS
543 These functions compute a rank-k update of the symmetric matrix @var{C},
544 @math{C = \alpha A A^T + \beta C} when @var{Trans} is
545 @code{CblasNoTrans} and @math{C = \alpha A^T A + \beta C} when
546 @var{Trans} is @code{CblasTrans}. Since the matrix @var{C} is symmetric
547 only its upper half or lower half need to be stored. When @var{Uplo} is
548 @code{CblasUpper} then the upper triangle and diagonal of @var{C} are
549 used, and when @var{Uplo} is @code{CblasLower} then the lower triangle
550 and diagonal of @var{C} are used.
553 @deftypefun int gsl_blas_cherk (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, float @var{alpha}, const gsl_matrix_complex_float * @var{A}, float @var{beta}, gsl_matrix_complex_float * @var{C})
554 @deftypefunx int gsl_blas_zherk (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, double @var{alpha}, const gsl_matrix_complex * @var{A}, double @var{beta}, gsl_matrix_complex * @var{C})
555 @cindex HERK, Level-3 BLAS
556 These functions compute a rank-k update of the hermitian matrix @var{C},
557 @math{C = \alpha A A^H + \beta C} when @var{Trans} is
558 @code{CblasNoTrans} and @math{C = \alpha A^H A + \beta C} when
559 @var{Trans} is @code{CblasTrans}. Since the matrix @var{C} is hermitian
560 only its upper half or lower half need to be stored. When @var{Uplo} is
561 @code{CblasUpper} then the upper triangle and diagonal of @var{C} are
562 used, and when @var{Uplo} is @code{CblasLower} then the lower triangle
563 and diagonal of @var{C} are used. The imaginary elements of the
564 diagonal are automatically set to zero.
567 @deftypefun int gsl_blas_ssyr2k (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, float @var{alpha}, const gsl_matrix_float * @var{A}, const gsl_matrix_float * @var{B}, float @var{beta}, gsl_matrix_float * @var{C})
568 @deftypefunx int gsl_blas_dsyr2k (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, double @var{alpha}, const gsl_matrix * @var{A}, const gsl_matrix * @var{B}, double @var{beta}, gsl_matrix * @var{C})
569 @deftypefunx int gsl_blas_csyr2k (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, const gsl_matrix_complex_float * @var{B}, const gsl_complex_float @var{beta}, gsl_matrix_complex_float * @var{C})
570 @deftypefunx int gsl_blas_zsyr2k (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, const gsl_matrix_complex * @var{B}, const gsl_complex @var{beta}, gsl_matrix_complex * @var{C})
571 @cindex SYR2K, Level-3 BLAS
572 These functions compute a rank-2k update of the symmetric matrix @var{C},
573 @math{C = \alpha A B^T + \alpha B A^T + \beta C} when @var{Trans} is
574 @code{CblasNoTrans} and @math{C = \alpha A^T B + \alpha B^T A + \beta C} when
575 @var{Trans} is @code{CblasTrans}. Since the matrix @var{C} is symmetric
576 only its upper half or lower half need to be stored. When @var{Uplo} is
577 @code{CblasUpper} then the upper triangle and diagonal of @var{C} are
578 used, and when @var{Uplo} is @code{CblasLower} then the lower triangle
579 and diagonal of @var{C} are used.
582 @deftypefun int gsl_blas_cher2k (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, const gsl_complex_float @var{alpha}, const gsl_matrix_complex_float * @var{A}, const gsl_matrix_complex_float * @var{B}, float @var{beta}, gsl_matrix_complex_float * @var{C})
583 @deftypefunx int gsl_blas_zher2k (CBLAS_UPLO_t @var{Uplo}, CBLAS_TRANSPOSE_t @var{Trans}, const gsl_complex @var{alpha}, const gsl_matrix_complex * @var{A}, const gsl_matrix_complex * @var{B}, double @var{beta}, gsl_matrix_complex * @var{C})
584 @cindex HER2K, Level-3 BLAS
585 These functions compute a rank-2k update of the hermitian matrix @var{C},
586 @math{C = \alpha A B^H + \alpha^* B A^H + \beta C} when @var{Trans} is
587 @code{CblasNoTrans} and @math{C = \alpha A^H B + \alpha^* B^H A + \beta C} when
588 @var{Trans} is @code{CblasConjTrans}. Since the matrix @var{C} is hermitian
589 only its upper half or lower half need to be stored. When @var{Uplo} is
590 @code{CblasUpper} then the upper triangle and diagonal of @var{C} are
591 used, and when @var{Uplo} is @code{CblasLower} then the lower triangle
592 and diagonal of @var{C} are used. The imaginary elements of the
593 diagonal are automatically set to zero.
599 The following program computes the product of two matrices using the
600 Level-3 @sc{blas} function @sc{dgemm},
605 \matrix{0.11&0.12&0.13\cr
615 \matrix{367.76&368.12\cr
624 [ 0.11 0.12 0.13 ] [ 1011 1012 ] [ 367.76 368.12 ]
625 [ 0.21 0.22 0.23 ] [ 1021 1022 ] = [ 674.06 674.72 ]
631 The matrices are stored in row major order, according to the C convention
635 @verbatiminclude examples/blas.c
639 Here is the output from the program,
643 @verbatiminclude examples/blas.out
646 @node BLAS References and Further Reading
647 @section References and Further Reading
649 Information on the @sc{blas} standards, including both the legacy and
650 draft interface standards, is available online from the @sc{blas}
651 Homepage and @sc{blas} Technical Forum web-site.
655 @cite{BLAS Homepage} @*
656 @uref{http://www.netlib.org/blas/}
658 @cite{BLAS Technical Forum} @*
659 @uref{http://www.netlib.org/cgi-bin/checkout/blast/blast.pl}
663 The following papers contain the specifications for Level 1, Level 2 and
668 C. Lawson, R. Hanson, D. Kincaid, F. Krogh, ``Basic Linear Algebra
669 Subprograms for Fortran Usage'', @cite{ACM Transactions on Mathematical
670 Software}, Vol.@: 5 (1979), Pages 308--325.
673 J.J. Dongarra, J. DuCroz, S. Hammarling, R. Hanson, ``An Extended Set of
674 Fortran Basic Linear Algebra Subprograms'', @cite{ACM Transactions on
675 Mathematical Software}, Vol.@: 14, No.@: 1 (1988), Pages 1--32.
678 J.J. Dongarra, I. Duff, J. DuCroz, S. Hammarling, ``A Set of
679 Level 3 Basic Linear Algebra Subprograms'', @cite{ACM Transactions on
680 Mathematical Software}, Vol.@: 16 (1990), Pages 1--28.
684 Postscript versions of the latter two papers are available from
685 @uref{http://www.netlib.org/blas/}. A @sc{cblas} wrapper for Fortran @sc{blas}
686 libraries is available from the same location.