From: Damien Mattei <damien.mattei@gmail.com>
To: kawa mailing list <kawa@sourceware.org>
Subject: infix precedence optimizer for Scheme+ for Kawa
Date: Wed, 29 Nov 2023 10:02:54 +0100 [thread overview]
Message-ID: <CADEOadcwRtEPyxPjcLepg4XnF4piMuv0qs15h+0Pv0GEhzNrDQ@mail.gmail.com> (raw)
[-- Attachment #1: Type: text/plain, Size: 4116 bytes --]
hello,
i have added optimisation features for infix precedence (
https://github.com/damien-mattei/Scheme-PLUS-for-Kawa/blob/main/optimize-infix.scm)
for the kawa parser version of Scheme+ (
https://github.com/damien-mattei/Scheme-PLUS-for-Kawa/blob/main/curly-infix2prefix4kawa.scm
)
for example for a given code in Scheme+ like this:
; modify coefficients layer(define (modification_des_poids M_i_o η
z_input z_output z̃_output ᐁ_i_o მzⳆმz̃) ; derivative of activation
function of the layer ; the length of output and input layer
with coeff. used for bias update {(len_layer_output
len_layer_input_plus1forBias) <+ (dim-matrix M_i_o)} ; use values and
define-values to create bindings {len_layer_input <+
{len_layer_input_plus1forBias - 1}} (for-each-in (j (in-range
len_layer_output)) ; line (for-each-in (i (in-range
len_layer_input)) ; column , parcours les colonnes de la ligne sauf le
bias {M_i_o[j {i + 1}] <- M_i_o[j {i + 1}] - (- η) *
z_input[i] * მzⳆმz̃(z_output[j] z̃_output[j]) * ᐁ_i_o[j]}) ;
and update the bias {M_i_o[j 0] <- M_i_o[j 0] - (- η)
* 1.0 * მzⳆმz̃(z_output[j] z̃_output[j]) * ᐁ_i_o[j]}))
without optimization it is transformed after parsing in:
(define (modification_des_poids M_i_o η z_input z_output z̃_output
ᐁ_i_o მzⳆმz̃) (<+ (len_layer_output len_layer_input_plus1forBias)
(dim-matrix M_i_o)) (<+ len_layer_input (-
len_layer_input_plus1forBias 1)) (for-each-in (j (in-range
len_layer_output)) (for-each-in (i (in-range len_layer_input))
($nfx$ (bracket-apply M_i_o j (+ i 1)) <- (bracket-apply M_i_o j (+ i
1)) - (- η) * (bracket-apply z_input i) * (მzⳆმz̃ (bracket-apply
z_output j) (bracket-apply z̃_output j)) * (bracket-apply ᐁ_i_o
j))) ($nfx$ (bracket-apply M_i_o j 0) <- (bracket-apply M_i_o j 0) -
(- η) * 1.0 * (მzⳆმz̃ (bracket-apply z_output j) (bracket-apply
z̃_output j)) * (bracket-apply ᐁ_i_o j))))
this cause the $nfx$ algorithm to be executed at runtime.
for example in the above code
{M_i_o[j {i + 1}] <- M_i_o[j {i + 1}] - (- η) * z_input[i] *
მzⳆმz̃(z_output[j] z̃_output[j]) * ᐁ_i_o[j]}
at each iteration of loop $nfx$ is called with its arguments:
($nfx$ (bracket-apply M_i_o j (+ i 1)) <- (bracket-apply M_i_o j (+
i 1)) - (- η) * (bracket-apply z_input i) * (მzⳆმz̃
(bracket-apply z_output j) (bracket-apply z̃_output j)) *
(bracket-apply ᐁ_i_o j))
Now the infix parser, before compilation produce a code like this,
where $nfx$ calls are replaced by the symbolic expression
according to the operator precedence rules:
(define (modification_des_poids M_i_o η z_input z_output z̃_output
ᐁ_i_o მzⳆმz̃) (<+ (len_layer_output len_layer_input_plus1forBias)
(dim-matrix M_i_o)) (<+ len_layer_input (-
len_layer_input_plus1forBias 1)) (for-each-in (j (in-range
len_layer_output)) (for-each-in (i (in-range len_layer_input)) (<-
(bracket-apply M_i_o j (+ i 1)) (- (bracket-apply M_i_o j (+ i 1))
(* (- η) (* (bracket-apply z_input i) (* (მzⳆმz̃
(bracket-apply z_output j) (bracket-apply z̃_output j))
(bracket-apply ᐁ_i_o j))))))) (<- (bracket-apply M_i_o j 0) (-
(bracket-apply M_i_o j 0) (* (- η) (* 1.0 (* (მzⳆმz̃
(bracket-apply z_output j) (bracket-apply z̃_output j))
(bracket-apply ᐁ_i_o j))))))))
so for the above expressions the simplest scheme code is evaluated at runtime:
(<- (bracket-apply M_i_o j (+ i 1)) (- (bracket-apply M_i_o j (+
i 1)) (* (- η) (* (bracket-apply z_input i) (* (მzⳆმz̃
(bracket-apply z_output j) (bracket-apply z̃_output j))
(bracket-apply ᐁ_i_o j))))))
the performance at runtime in mathematic computation algorithms are
now _twice_ faster just with this optimization.
Next optimization will concern bracket-apply but this is a big work,
inference type need to be performed on many expressions of all files
of code..
Damien
reply other threads:[~2023-11-29 9:03 UTC|newest]
Thread overview: [no followups] expand[flat|nested] mbox.gz Atom feed
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CADEOadcwRtEPyxPjcLepg4XnF4piMuv0qs15h+0Pv0GEhzNrDQ@mail.gmail.com \
--to=damien.mattei@gmail.com \
--cc=kawa@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).