<div dir="ltr">Recall that last. Always measure at the command line <sigh>. I'm seeing FlVector at 30% faster. <div><br></div><div style><div>#lang typed/racket</div><div><br></div><div>(provide main)</div><div> <br></div><div>(require racket/unsafe/ops</div><div><span class="" style="white-space:pre"> </span> racket/fixnum</div><div> racket/flonum)</div><div><br></div><div>(define-type NF Nonnegative-Fixnum)</div><div><br> </div><div>(: matrix* :</div><div> FlVector FlVector Index Index Index</div><div> -> FlVector)</div><div>(define (matrix* A B m p n)</div><div> (define size (* m n))</div><div> (define: v : FlVector (make-flvector size 0.0))</div> <div> (define index 0)</div><div> (: loop-i : NF -> FlVector)</div><div> (define (loop-i i)</div><div> (cond [(< i m)</div><div> (loop-j i 0)</div><div> (loop-i (+ i 1))]</div><div> [else v]))</div> <div> (: loop-j : NF NF -> Void)</div><div> (define (loop-j i j)</div><div> (when (< j n)</div><div> (loop-k i j 0 0.0)</div><div> (set! index (+ index 1))</div><div> (loop-j i (+ j 1))))</div><div> (: loop-k : NF NF NF Flonum -> Void)</div><div> (define (loop-k i j k sum)</div><div> (define i*p (* i p))</div><div> (define k*n (* k n))</div><div> (cond</div><div> [(< k p)</div><div> (loop-k i j (+ k 1)</div> <div><span class="" style="white-space:pre"> </span> (+ sum</div><div><span class="" style="white-space:pre"> </span> (* (unsafe-flvector-ref A (+ i*p k))</div><div><span class="" style="white-space:pre"> </span>(unsafe-flvector-ref B (+ k*n j)))))]</div> <div> [else</div><div> (unsafe-flvector-set! v index sum)]))</div><div> (loop-i 0))</div><div><br></div><div>(: mkFlVector (Fixnum (-> Flonum) -> FlVector))</div><div>(define (mkFlVector sz init)</div><div> (let ((v (make-flvector sz)))</div><div> (do ((i #{0 : Fixnum} (+ i 1)))</div><div><span class="" style="white-space:pre"> </span>((>= i sz) v)</div><div> (unsafe-flvector-set! v i (init)))))</div><div><br></div> <div>(define: dim : Index 300)</div><div>(define: A : FlVector (mkFlVector (fx* dim dim) random))</div><div>(define: B : FlVector (mkFlVector (fx* dim dim) random))</div><div><br></div><div>(define (main)</div><div> (collect-garbage)</div> <div> (collect-garbage)</div><div> (let ((m (time (matrix* A B dim dim dim))))</div><div> (void)))</div><div><br></div></div></div><div class="gmail_extra"><br><br><div class="gmail_quote">On Sun, Jan 20, 2013 at 4:25 PM, Ray Racine <span dir="ltr"><<a href="mailto:ray.racine@gmail.com" target="_blank">ray.racine@gmail.com</a>></span> wrote:<br> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">Modification of the above to use FlVector did not show a difference.</div><div class="HOEnZb"><div class="h5"> <div class="gmail_extra"><br><br><div class="gmail_quote">On Sun, Jan 20, 2013 at 2:35 PM, Jens Axel Søgaard <span dir="ltr"><<a href="mailto:jensaxel@soegaard.net" target="_blank">jensaxel@soegaard.net</a>></span> wrote:<br> <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">2013/1/20 Berthold Bäuml <<a href="mailto:berthold.baeuml@dlr.de" target="_blank">berthold.baeuml@dlr.de</a>>:<br> <div>>> Ah! The Numpy example uses floats and not doubles.<br> ><br> > On my machine Numpy says it is using float64, hence, double.<br> > <a href="http://big1.dtype.name" target="_blank">big1.dtype.name</a> -> float64<br> <br> </div>Sorry for the confusion. I am used to the Racket documentation<br> were float means single precision, so I misinterpreted the Numpy<br> documentation.<br> <br> Also<br> <br> > numpy.random.random((5,5))<br> array([[ 0.04748764, 0.15242073, 0.2366255 , 0.86926654, 0.16586114],<br> [ 0.33117124, 0.7504899 , 0.43179013, 0.1992083 , 0.83266119],<br> [ 0.25741902, 0.36912007, 0.38273193, 0.29622522, 0.75817261],<br> [ 0.49313726, 0.20514653, 0.45329344, 0.964105 , 0.77459153],<br> [ 0.5267172 , 0.87269101, 0.34530438, 0.38218051, 0.05107181]])<br> <br> suggested, that the numbers were single precision. It turns out<br> that random only produces 53 bit random floats.<br> <br> Anyways, here is a standard matrix multiplication algorithm in<br> Typed Racket.<br> <br> #lang typed/racket<br> (require racket/unsafe/ops<br> racket/flonum)<br> <br> (define-type NF Nonnegative-Fixnum)<br> <br> (: matrix* :<br> (Vectorof Flonum) (Vectorof Flonum) Index Index Index<br> -> (Vectorof Flonum))<br> (define (matrix* A B m p n)<br> (define size (* m n))<br> (define: v : (Vectorof Flonum) (make-vector size 0.0))<br> (define index 0)<br> (: loop-i : NF -> (Vectorof Flonum))<br> (define (loop-i i)<br> (cond [(< i m)<br> (loop-j i 0)<br> (loop-i (+ i 1))]<br> [else v]))<br> (: loop-j : NF NF -> Void)<br> (define (loop-j i j)<br> (when (< j n)<br> (loop-k i j 0 0.0)<br> (set! index (+ index 1))<br> (loop-j i (+ j 1))))<br> (: loop-k : NF NF NF Flonum -> Void)<br> (define (loop-k i j k sum)<br> (define i*p (* i p))<br> (define k*n (* k n))<br> (cond<br> [(< k p)<br> (loop-k i j (+ k 1)<br> (+ sum<br> (* (unsafe-vector-ref A (+ i*p k))<br> (unsafe-vector-ref B (+ k*n j)))))]<br> [else<br> (vector-set! v index sum)]))<br> (loop-i 0))<br> <br> (define dim 200)<br> (define A (build-vector (* dim dim) (λ (_) (random))))<br> (define B (build-vector (* dim dim) (λ (_) (random))))<br> (collect-garbage)<br> (collect-garbage)<br> (define A*B (time (matrix* A B dim dim dim)))<br> <span><font color="#888888"><br> /Jens Axel<br> </font></span><div><div><br> ____________________<br> Racket Users list:<br> <a href="http://lists.racket-lang.org/users" target="_blank">http://lists.racket-lang.org/users</a><br> </div></div></blockquote></div><br></div> </div></div></blockquote></div><br></div>