On Wed, May 14, 2008 at 5:02 PM, Liam Healy lhealy@common-lisp.net wrote:
On Wed, May 14, 2008 at 2:23 PM, Mirko Vukovic mirko.vukovic@gmail.com wrote:
Hi,
I was playing with gsll and blas. I modified the cblas-transpose definition in blas2.lisp (see top of file) to capture a few more values:
Here is the new version:
(cffi:defcenum cblas-transpose "CBLAS_TRANSPOSE from /usr/include/gsl/gsl_cblas.h." (:notrans 111) (:trans 112) (:conjtrans 113))
I fail to see the difference with what is in there now: (cffi:defcenum cblas-transpose "CBLAS_TRANSPOSE from /usr/include/gsl/gsl_cblas.h." (:notrans 111) :trans :conjtrans) See http://common-lisp.net/project/cffi/manual/html_node/defcenum.html#defcenum "If value is omitted its value will either be 0, if it's the first entry, or it it will continue the progression from the last specified value." so :trans gets 112, :conjtrans 113 by default.
I was wondering why the other values were omitted.
I am posting it here because I was not able to download with svn so I don't have versioning for gsll.
Also, here is a code snippet that I used to test for tridiagonal matrix solving that also uses gsll-s blas2 routine gemv to double check that things work.
(in-package :gsll) ;; solve a tridiagonal system and then verify result by matrix multiplication ;; uses solve-tridiagonal and gemv (let* ((dim 6) (dim- (- dim 1))) (let ((d (make-array dim :element-type 'double-float :initial-element -2.d0)) (e (make-array dim- :element-type 'double-float :initial-element 0.2d0)) (f (make-array dim- :element-type 'double-float :initial-element 1.d0)) (x (make-array dim :element-type 'double-float :initial-element 0.d0)) (b (make-array dim :element-type 'double-float :initial-element 0.5d0)) (mat (make-array (list dim dim) :element-type 'double-float :initial-element 0.d0)))
;; load matrix (loop for i from 0 below dim- do (progn (setf (aref mat i i) (aref d i)) (setf (aref mat (+ i 1) i) (aref f i)) (setf (aref mat i (+ i 1)) (aref e i)))) (setf (aref mat dim- dim-) (aref d dim-)) ;; gsl operations (letm ((d* (vector-double-float d)) (e* (vector-double-float e)) (f* (vector-double-float f)) (x* (vector-double-float x)) (b* (vector-double-float b)) (v2* (vector-double-float (make-array dim :element-type 'double-float :initial-element 0.d0))) (mat* (matrix-double-float mat))) (solve-tridiagonal d* e* f* b* x*) ;; gemv returns the result in the last argument (gemv :notrans 1.d0 mat* x* 0.d0 v2*) (data v2*))))I am sure this code betrays my lack of lisp experience. So feel free to clean it up if you want it included in the verification part.
Mirko
I might make a few simplifications; #'1- to do decrementing and I don't think there's a need to make arrays in CL and then remake them on the foreign side. It is possibly a candidate for example/ regression tests.
Liam
I agree for regression that there is no need to duplicate the arrays. But,
1) I will be using the arrays outside of (letm ...) to plot them etc 2) I am still feeling my way around. Even the above point may be un-necessary, with, for example the (data ... function)
Mirko