* Re: Some thoughts about Emacs performance
2024-02-16 14:08 ` Simon Leinen
@ 2024-02-16 16:15 ` Robert Boyer
2024-02-19 12:29 ` Ihor Radchenko
2024-02-16 16:31 ` Robert Boyer
` (2 subsequent siblings)
3 siblings, 1 reply; 11+ messages in thread
From: Robert Boyer @ 2024-02-16 16:15 UTC (permalink / raw)
To: Simon Leinen; +Cc: Emacs developers
[-- Attachment #1.1: Type: text/plain, Size: 7263 bytes --]
Dear Simon,
That's the kindest message I have ever received. Thank you so very much.
Made my day and my life.
I am hoping, using 'native-compile', about which I have heard great things,
to contribute one function that is the same as 'stable-sort' in Common
Lisp. It runs, in SBCL, about twice as fast as SBCL's stable-sort. See
the attached file n3ms.lisp. The really stunning news to me is that my
friend Grant Passmore, running on an M3 machine only yesterday, and taking
advantage of threads in SBCL, saw a 5x performance improvement over SBCL's
stable-sort. I have not the slightest clue about whether threads are in
Emacs or in its future.
The name of my function is 'msa'. I have attached the file msa.el in its
current state of development. It is far from finished and no one in their
right mind will try to use it! But great Emacs minds might read it and
tell me things I need to know.
The threads code I added to 'msa' for SBCL is I think an amazing comment
upon what a fantastic job the SBCL folks have done for threads. My
addition to 'msa' to take advantage of threads is only about a dozen lines
long, and in my extremely unhumble opinion, it is a dozen lines of the
greatest beauty that I have ever seen. May the Lord bless John von Neumann,
wherever he is, for 'merge-sort'.
But I do fear for the worst. I am 77 and have always worried too much, I
also attach a file that I just sent to rms, because I cannot yet use m-x
report-emacs-bug, due to some mailer problem. I run on a $100 Lenovo
Chromebook, and somehow, by magic Emacs 28 just suddenly appeared a few
days ago and it is great, except for 1. some mailer problem with
report-emacs-bug and 2. crucial to my work on msa, the following bug
report, in the attached file compile-bug.el, on native-compile.
I say I fear for the worst because if that bug is what I think it is, it
would kill 'msa' performance.
Very secondarily, even if the bug is fixed, I have no idea how I could ever
take advantage of it!
Emacs 28 appeared by magic on my Chromebook only a few days ago. I have
always depended upon the kindness of friends and strangers, who have
magically installed Emacs for me. All I did to install Emacs on my
Chromebook was to run this command:
sudo apt-get install emacs
Great minds, I guess you guys with Emacs, at Google, and at Debian should
know that is all I know and all I need to know, so far, about installing
Emacs.
Thanks so very, very much for your extremely kind letter,
Bob
P. S. You mention 'random'. My bet is that unless you fix the bug that I
mention above, no one could ever do a random via native-compiler that would
be competitive. Using *declared *fixnum and vectors is crucial to any
speed-competitive work on 'random' that I can imagine.
P. P. S. My home phone is 512 467 0182. Phone me any time. You or anyone
else doing Emacs development. If you like, since I can call anywhere in the
USA for free, I will hang up and call you right back if you prefer.
P. P. P. S. You kindly mentioned the ancient 'boyer' benchmark. One must
know about it that it has a bug, as far as truth rather than performance
testing, is concerned. Whoever translated that file from Maclisp to Common
Lisp failed to note that 'member' now needs a :test 'equal bit in the call.
Common Lisp defaults the test to 'eql, and that is not what one needs.
Anyway, that code is only for performance testing. Nqthm and ACL2 are real
theorem-proving programs written in Common Lisp and they are both easy to
obtain for 'free', or for 'gratis', as rms is now saying. To some I guess,
'free' may sometimes have a bad connotation; not for me, though. May the
Free Software Foundation forever flourish. What Harvard has done for us
all: Gates, Zuckerberg, Stallman.
On Fri, Feb 16, 2024 at 8:08 AM Simon Leinen <simon.leinen@gmail.com> wrote:
> Bob,
>
> welcome to the emacs-devel list, and thanks a lot for *your* wonderful
> contributions (theorem prover, string search, and Lisp benchmarking -
> I remember boyer.lisp from RPG's reference work[1]).
>
> On Thu, Feb 8, 2024 at 8:15 AM Robert Boyer
> <robertstephenboyer@gmail.com> wrote:
> >
> > Emacs 27.1 has a 'sort' function that takes longer than stable-sort of
> SBCL. Maybe
> > by a factor of 2. See also my attached file 'ms.lisp'.
> >
> > There may be a lot that can be improved in Emacs'
> > handling of cl-loop, setf, elt, or cl-random.
>
> In this case, cl-random seems to be the main culprit for the slow
> initialization—replacing that with plain "random" speeds it up by
> about a factor of ten. There was some discussion on the list recently
> about cl-random vs. random. The main functional difference is that
> cl-random supports a defined state. But the performance difference may
> be due more to the fact that random is written in C, and cl-random in
> Lisp.
>
> As for the sorting itself, both Emacs and SBCL seem to use mergesort
> in their implementations of (stable-)sort. Emacs's implementation is
> written in C, SBCL's in Lisp. Performance is quite similar—on my
> system (Apple Macbook Air M2) Emacs takes about 35% longer to sort a
> million random numbers than SBCL. (On the other hand when sorting it
> again, i.e. when the vector is already fully sorter, Emacs is quite a
> bit faster than SBCL—maybe Emacs chose to optimize for partly-sorted
> vectors at the expense of a bit of performance for random input.)
>
> In general, the Emacs Lisp runtime system and compiler(s) aren't as
> optimized as SBCL for general Lisp use. But it gets quite close!
>
> On the other hand, Emacs has editor-specific code (e.g. redisplay) and
> data structures (e.g. buffers) which are highly optimized and partly
> written in C. But it doesn't try to be a standalone platform for
> performance-oriented Lisp developers. Of course Emacs is very
> suitable as a Software Development Environment for systems such as
> SBCL, and there are many good integration options—personally I use the
> SLIME package these days.
>
> Best regards, and enjoy Lisping in Emacs!
> --
> Simon.
>
> > ;; First some Emacs, with times on my $100 Chromebook.
> >
> > (setq n 6)
> > (defun make-random-array (n)
> > (let ((a (make-vector n 0)))
> > (cl-loop for i below n do
> > (setf (elt a i) (cl-random 1000000)))
> > a))
> > (byte-compile 'make-random-array)
> > (benchmark '(setq foo (make-random-array (expt 10 n))) 1) -- 2.3 seconds
> > (benchmark '(sort foo '<) 1) -- 1 second
> >
> > ;; Second some Common Lisp, with times for SBCL on my $100 Chromebook.
> >
> > (defparameter n 6)
> > (defun make-random-array (n)
> > (declare (fixnum n))
> > (let ((a (make-array n)))
> > (declare (type array a))
> > (loop for i fixnum below n do
> > (setf (aref a i) (random most-positive-fixnum)))
> > a))
> > (time (defparameter foo (make-random-array (expt 10 n)))) -- .041
> seconds
> > (time (progn (stable-sort foo '<) nil)) -- .45 seconds
> >
> > Thanks so much for Emacs, which is so great that I cannot put it
> > into words.
> >
> > Bob
>
>
> [1] https://dreamsongs.com/Files/Timrep.pdf
>
[-- Attachment #1.2: Type: text/html, Size: 8575 bytes --]
[-- Attachment #2: msa.el --]
[-- Type: application/octet-stream, Size: 10001 bytes --]
;; with-temp-file
(defvar msa-scratch (make-vector 1 0))
(defvar msa-input (make-vector 1 0))
(defvar msa-type t)
(defvar msa-predicate '<)
(defvar msa-tmp-file "msa-tmp-file.el")
(defvar msa-tmp-buffer "msa-tmp-buffer")
(defvar msa-1-body
"(defun msa-1 (start end input scratch)
(declare (fixnum start end) (vector input scratch))
(let* ((start+1 (1+ start))
(end-start (- end start))
(mid (+ start (floor end-start 2))))
(declare (fixnum start+1 end-start mid))
(cond
((or (eql start end) (eql start+1 end))
(cl-return-from msa-1 nil))
((eql end-start 2)
(let ((as-start (aref input start))
(as-start+1 (aref input start+1)))
(cond ((msa-predicate (key as-start+1) (key as-start))
(setf (aref input start) as-start+1)
(setf (aref input start+1) as-start)))
(cl-return-from msa-1 nil)))
(t
(let ((k1 (key (aref input start)))
(k2 (key (aref input (1+ start))))
(i start)
(mid-2 (- mid 2)))
(declare (fixnum i mid-2))
(cl-loop
(cond ((eql i mid-2)
(msa-1 mid end input scratch)
(cl-return nil))
((or (msa-equal k1 k2) (msa-predicate k1 k2))
(setq k1 k2)
(setq k2 (key (aref input (+ i 2))))
(cl-incf i))
(t (msa-1 start mid input scratch)
(msa-1 mid end input scratch)
(cl-return nil))))
(let ((i0 start) (i1 mid) (end0 mid) (end1 end))
(declare (fixnum i0 i1 end0 end1))
(let ((si 0))
(declare (fixnum si))
(cl-loop (cond ((= i end-start) (cl-return nil)))
(cond ((eql i0 end0)
(setf (aref scratch si) (aref input i1))
(cl-incf i1))
((eql i1 end1)
(setf (aref scratch si) (aref input i0))
(cl-incf i0))
(t (let* ((x0 (aref input i0))
(x1 (aref input i1))
(k0 (key x0))
(k1 (key x1)))
(cond ((or (msa-equal k0 k1) (msa-predicate k0 k1))
(setf (aref scratch si) x0)
(cl-incf i0))
(t (setf (aref scratch si) x1)
(cl-incf i1))))))))
(let ((i 0) (j start))
(declare (fixnum i j))
(cl-loop (cond ((eql i end-start) (cl-return nil)))
(setf (aref input j) (aref scratch i))
(cl-incf i)))
input))))))")
(setq msa-1-body
"(defun msa-1 (start end input scratch)
(declare (fixnum start end) (vector input scratch))
(let* ((start+1 (1+ start))
(end-start (- end start))
(mid (+ start (floor end-start 2))))
(declare (fixnum start+1 end-start mid))
(cond
((or (eql start end) (eql start+1 end))
(cl-return-from msa-1 nil))
((eql end-start 2)
(let ((as-start (aref input start))
(as-start+1 (aref input start+1)))
(cond ((msa-predicate (key as-start+1) (key as-start))
(setf (aref input start) as-start+1)
(setf (aref input start+1) as-start)))
(cl-return-from msa-1 nil)))
(t
(let ((k1 (key (aref input start)))
(k2 (key (aref input (1+ start))))
(i start)
(mid-2 (- mid 2)))
(declare (fixnum i mid-2))
(cl-loop
(cond ((eql i mid-2)
(msa-1 mid end input scratch)
(cl-return nil))
((or (msa-equal k1 k2) (msa-predicate k1 k2))
(setq k1 k2)
(setq k2 (key (aref input (+ i 2))))
(cl-incf i))
(t (msa-1 start mid input scratch)
(msa-1 mid end input scratch)
(cl-return nil))))
(let ((i0 start) (i1 mid) (end0 mid) (end1 end))
(declare (fixnum i0 i1 end0 end1))
(let ((si 0))
(declare (fixnum si))
(cl-loop (cond ((= i end-start) (cl-return nil)))
(cond ((eql i0 end0)
(setf (aref scratch si) (aref input i1))
(cl-incf i1))
((eql i1 end1)
(setf (aref scratch si) (aref input i0))
(cl-incf i0))
(t (let* ((x0 (aref input i0))
(x1 (aref input i1))
(k0 (key x0))
(k1 (key x1)))
(cond ((or (msa-equal k0 k1) (msa-predicate k0 k1))
(setf (aref scratch si) x0)
(cl-incf i0))
(t (setf (aref scratch si) x1)
(cl-incf i1))))))))
(let ((i 0) (j start))
(declare (fixnum i j))
(cl-loop (cond ((eql i end-start) (cl-return nil)))
(setf (aref input j) (aref scratch i))
(cl-incf i)))
input))))))")
(defun msa (ar msa-predicate key)
(cond ((not (or (null ar) (vectorp ar) (and (consp ar) (null (cdr (last ar))))))
(error "msa: first argument is not a proper sequence.")))
(cond ((not (functionp msa-predicate))
(error "msa: msa-predicate is not a function.")))
(cond ((null key))
((or (not (symbolp key)) (not (functionp key)))
(error "msa: key must be a symbol.")))
(cond ((eq key 'identity) (setq key nil)))
(let ((len (length ar)))
(declare (fixnum len))
(cond ((= len 0) nil)
(t (cond ((<= len (length msa-scratch)))
(t (setq msa-scratch (make-vector len 0))))
(setq msa-input
(cond ((consp ar)
(let ((ans (make-vector len 0)))
(let ((i 0) (l ar))
(declare (fixnum i))
(while (< i len)
(setf (aref ans i) (car l))
(setq l (cdr l))
(cl-incf i)))
ans))
(t ar)))
(let ((msa-all-fixnum t) (msa-all-float t)
(msa-all-keys-real t) (orig-ar ar))
(let ((i 0))
(declare (fixnum i))
(while (< i len)
(let ((x (cond ((null key) (aref msa-input i))
(t (funcall key (aref msa-input i))))))
(cond ((and msa-all-keys-real
(not (member (type-of x) '(float integer))))
(setq msa-all-keys-real nil)))
(cond ((and msa-all-fixnum
(not (fixnump x)))
(setq msa-all-fixnum nil)))
(cond ((and msa-all-float
(not (floatp x)))
(setq msa-all-float nil)))
(cl-incf i)))
(let ((msa-type (cond (msa-all-fixnum 'fixnum)
(msa-all-float 'float)
(t t))))
(switch-to-buffer msa-tmp-buffer)
(kill-region (point-min) (point-max))
(with-output-to-temp-buffer msa-tmp-buffer
(print `(defvar key ',key))
(print `(defmacro key (x)
(cond ((null key) x)
(t `(,key ,x)))))
(print `(defmacro msa-equal (x y)
(cond (msa-all-keys-real `(eql ,x ,y))
(t `(equal ,x ,y)))))
(print `(defmacro msa-predicate (x y)
`(let ((xv ,x) (yv ,y))
(declare (fixnum xv yv))
(,msa-predicate xv yv))))
(princ msa-1-body)
(write-file msa-tmp-file)
;; (write-file "msa-debug.el")
(switch-to-buffer msa-tmp-file)
(emacs-lisp-native-compile-and-load)
;; (kill-buffer msa-tmp-file)
(eval `(msa-1 0 ,len msa-input msa-scratch))
(let ((i 0))
(declare (fixnum i))
(while (< i len)
(setf (aref msa-scratch i) 0)))
(cond ((vectorp orig-ar)
(let ((i 0))
(declare (fixnum i))
(setf (aref orig-ar i) (aref msa-input i))))
(t (let ((i 0) (tail orig-ar))
(declare (fixnum i))
(while tail
(setf (car tail) (aref msa-input i))
(cl-incf i)
(setq tail (cdr tail))))))
(setq msa-input (make-vector 1 0))
ar))))))))
[-- Attachment #3: compile-bug.el --]
[-- Type: application/octet-stream, Size: 1731 bytes --]
;; Let us suppose that this is the file "compile-bug.el"
;; Invoking (native-compile "compile-bug.el") should work, I do believe.
;; However it fails and the error message is printed below.
;; It cannot be emphasized enough how serious this problem seems to me, Bob
;; Boyer, robertstephenboyer@gmail.com.
;; The reason it is so SERIOUS is that ANY Lisp compiler would need to be
;; delighted to see such a typing expression as
;; (declare (fixnum start end) (vector input scratch))
;; Why? BECAUSE it means that the compiler does not have to lay down code to
;; type check the type of start, end, input, and scratch!!!!
;; However, (byte-compile "compile-bug.el") results in the following error message:
;; Compiling file /mnt/chromeos/GoogleDrive/MyDrive/Linux/working/compile-bug.el at Fri Feb 16 08:25:19 2024
;; compile-bug.el:2:45: Warning: Unknown defun property ‘fixnum’ in foo
;; compile-bug.el:2:45: Warning: Unknown defun property ‘vector’ in foo
(defun foo (start end input scratch)
(declare (fixnum start end) (vector input scratch))
(list start end input scratch))
;; foo works fine
;; Here is an example form for the invocation of foo:
;; (foo 1 2 (make-vector 3 4) (make-vector 5 6))
;; That form runs ok if it is running with foo interpreted.
;; (byte-compile 'foo) runs ok.
;; One can run the same form after byte-compiling and it runs ok.
;; However, (native-compile "compile-bug.el") fails with
;; the error report:
;; Compiling file /mnt/chromeos/GoogleDrive/MyDrive/Linux/working/compile-bug.el at Fri Feb 16 08:35:04 2024
;; compile-bug.el:12:45: Warning: Unknown defun property ‘fixnum’ in foo
;; compile-bug.el:12:45: Warning: Unknown defun property ‘vector’ in foo
[-- Attachment #4: n3ms.lisp --]
[-- Type: application/x-lisp, Size: 29191 bytes --]
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Some thoughts about Emacs performance
2024-02-16 16:15 ` Robert Boyer
@ 2024-02-19 12:29 ` Ihor Radchenko
2024-02-19 19:05 ` Robert Boyer
0 siblings, 1 reply; 11+ messages in thread
From: Ihor Radchenko @ 2024-02-19 12:29 UTC (permalink / raw)
To: Robert Boyer; +Cc: Simon Leinen, Emacs developers
Robert Boyer <robertstephenboyer@gmail.com> writes:
> I say I fear for the worst because if that bug is what I think it is, it
> would kill 'msa' performance.
> Very secondarily, even if the bug is fixed, I have no idea how I could ever
> take advantage of it!
As Andrea mentioned in
https://yhetil.org/emacs-devel/yp1cyssanrp.fsf@fencepost.gnu.org/, type
declarations are not yet supported and it is not yet clear how long it
will take to implement type declaration support.
If you wish to contribute a better sorting to Emacs, another option
might be contributing to Emacs C sources. Check out src/sort.c file
(tim_sort) for the sort implementation that is currently in use.
https://git.savannah.gnu.org/cgit/emacs.git/tree/src/sort.c#n915
--
Ihor Radchenko // yantar92,
Org mode contributor,
Learn more about Org mode at <https://orgmode.org/>.
Support Org development at <https://liberapay.com/org-mode>,
or support my work at <https://liberapay.com/yantar92>
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Some thoughts about Emacs performance
2024-02-19 12:29 ` Ihor Radchenko
@ 2024-02-19 19:05 ` Robert Boyer
0 siblings, 0 replies; 11+ messages in thread
From: Robert Boyer @ 2024-02-19 19:05 UTC (permalink / raw)
To: Ihor Radchenko; +Cc: Simon Leinen, Emacs developers
[-- Attachment #1: Type: text/plain, Size: 1693 bytes --]
> Another option might be contributing to Emacs C sources.
Surely you're joking, Mr. Feynman. 1. sort in Emacs is, I do believe,
written in C and seems ok, but could be improved. 2. You probably cannot
imagine, in this day, how totally ignorant I am of C! It is unforgivable.
> it is not yet clear how long it will take to implement type declaration
support.
My bet is that it will be a long time coming, from the vibes I get. But
what do I know?
Bob
On Mon, Feb 19, 2024 at 6:25 AM Ihor Radchenko <yantar92@posteo.net> wrote:
> Robert Boyer <robertstephenboyer@gmail.com> writes:
>
> > I say I fear for the worst because if that bug is what I think it is, it
> > would kill 'msa' performance.
> > Very secondarily, even if the bug is fixed, I have no idea how I could
> ever
> > take advantage of it!
>
> As Andrea mentioned in
> https://yhetil.org/emacs-devel/yp1cyssanrp.fsf@fencepost.gnu.org/, type
> declarations are not yet supported and it is not yet clear how long it
> will take to implement type declaration support.
>
> If you wish to contribute a better sorting to Emacs, another option
> might be contributing to Emacs C sources. Check out src/sort.c file
> (tim_sort) for the sort implementation that is currently in use.
> https://git.savannah.gnu.org/cgit/emacs.git/tree/src/sort.c#n915
>
> --
> Ihor Radchenko // yantar92,
> Org mode contributor,
> Learn more about Org mode at <https://orgmode.org/>.
> Support Org development at <https://liberapay.com/org-mode>,
> or support my work at <https://liberapay.com/yantar92>
>
--
Anything I seem to state should be taken as a question. I am at least 77
and feeble.
[-- Attachment #2: Type: text/html, Size: 2830 bytes --]
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Some thoughts about Emacs performance
2024-02-16 14:08 ` Simon Leinen
2024-02-16 16:15 ` Robert Boyer
@ 2024-02-16 16:31 ` Robert Boyer
2024-02-16 18:06 ` Robert Boyer
2024-02-16 18:34 ` Robert Boyer
3 siblings, 0 replies; 11+ messages in thread
From: Robert Boyer @ 2024-02-16 16:31 UTC (permalink / raw)
To: Simon Leinen; +Cc: Emacs developers
[-- Attachment #1: Type: text/plain, Size: 5936 bytes --]
Dear Simon,
You note:
On the other hand when sorting it
again, i.e. when the vector is already fully sorter, Emacs is quite a
bit faster than SBCL—maybe Emacs chose to optimize for partly-sorted
vectors at the expense of a bit of performance for random input.
You are so right. 'msa' kills 'stable sort' on a few examples, namely an
array
that is all 0's and an array that is 1, 2, 3, ...
Below is some info on sorting arrays that are 21 million long.
But I should mention that I regard what my 'msa' does to catch the easy
cases may be nothing in comparison to what sorting pros can do.
Best,
Bob
Running (test-msa-0 21,000,000).
Timing (MSA TEST-MSA-AR1 '< :KEY 'CAR).
WARNING: redefining COMMON-LISP-USER::KEY in DEFMACRO
WARNING: redefining COMMON-LISP-USER::MSA-EQUAL in DEFMACRO
WARNING: redefining COMMON-LISP-USER::MSA-PREDICATE in DEFMACRO
WARNING: redefining COMMON-LISP-USER::MSA-1 in DEFUN
Evaluation took:
3.334 seconds of real time
3.267129 seconds of total run time (2.323391 user, 0.943738 system)
[ Run times consist of 0.154 seconds GC time, and 3.114 seconds non-GC
time. ]
97.99% CPU
31 forms interpreted
115 lambdas converted
5,990,145,243 processor cycles
174,860,640 bytes consed
Timing (STABLE-SORT TEST-MSA-AR2 '< :KEY 'CAR).
Evaluation took:
12.729 seconds of real time
12.562949 seconds of total run time (11.935063 user, 0.627886 system)
[ Run times consist of 0.232 seconds GC time, and 12.331 seconds non-GC
time. ]
98.70% CPU
22,870,125,432 processor cycles
168,006,352 bytes consed
Running (test-msa-ascending 21,000,000).
Timing (MSA TEST-MSA-AR1 '< :KEY 'CAR).
WARNING: redefining COMMON-LISP-USER::KEY in DEFMACRO
WARNING: redefining COMMON-LISP-USER::MSA-EQUAL in DEFMACRO
WARNING: redefining COMMON-LISP-USER::MSA-PREDICATE in DEFMACRO
WARNING: redefining COMMON-LISP-USER::MSA-1 in DEFUN
Evaluation took:
2.525 seconds of real time
2.514507 seconds of total run time (2.471494 user, 0.043013 system)
99.60% CPU
31 forms interpreted
115 lambdas converted
4,536,551,428 processor cycles
6,823,488 bytes consed
Timing (STABLE-SORT TEST-MSA-AR2 '< :KEY 'CAR).
Evaluation took:
12.914 seconds of real time
12.823063 seconds of total run time (12.407732 user, 0.415331 system)
[ Run times consist of 0.227 seconds GC time, and 12.597 seconds non-GC
time. ]
99.30% CPU
23,203,467,086 processor cycles
168,006,128 bytes consed
On Fri, Feb 16, 2024 at 8:08 AM Simon Leinen <simon.leinen@gmail.com> wrote:
> Bob,
>
> welcome to the emacs-devel list, and thanks a lot for *your* wonderful
> contributions (theorem prover, string search, and Lisp benchmarking -
> I remember boyer.lisp from RPG's reference work[1]).
>
> On Thu, Feb 8, 2024 at 8:15 AM Robert Boyer
> <robertstephenboyer@gmail.com> wrote:
> >
> > Emacs 27.1 has a 'sort' function that takes longer than stable-sort of
> SBCL. Maybe
> > by a factor of 2. See also my attached file 'ms.lisp'.
> >
> > There may be a lot that can be improved in Emacs'
> > handling of cl-loop, setf, elt, or cl-random.
>
> In this case, cl-random seems to be the main culprit for the slow
> initialization—replacing that with plain "random" speeds it up by
> about a factor of ten. There was some discussion on the list recently
> about cl-random vs. random. The main functional difference is that
> cl-random supports a defined state. But the performance difference may
> be due more to the fact that random is written in C, and cl-random in
> Lisp.
>
> As for the sorting itself, both Emacs and SBCL seem to use mergesort
> in their implementations of (stable-)sort. Emacs's implementation is
> written in C, SBCL's in Lisp. Performance is quite similar—on my
> system (Apple Macbook Air M2) Emacs takes about 35% longer to sort a
> million random numbers than SBCL. (On the other hand when sorting it
> again, i.e. when the vector is already fully sorter, Emacs is quite a
> bit faster than SBCL—maybe Emacs chose to optimize for partly-sorted
> vectors at the expense of a bit of performance for random input.)
>
> In general, the Emacs Lisp runtime system and compiler(s) aren't as
> optimized as SBCL for general Lisp use. But it gets quite close!
>
> On the other hand, Emacs has editor-specific code (e.g. redisplay) and
> data structures (e.g. buffers) which are highly optimized and partly
> written in C. But it doesn't try to be a standalone platform for
> performance-oriented Lisp developers. Of course Emacs is very
> suitable as a Software Development Environment for systems such as
> SBCL, and there are many good integration options—personally I use the
> SLIME package these days.
>
> Best regards, and enjoy Lisping in Emacs!
> --
> Simon.
>
> > ;; First some Emacs, with times on my $100 Chromebook.
> >
> > (setq n 6)
> > (defun make-random-array (n)
> > (let ((a (make-vector n 0)))
> > (cl-loop for i below n do
> > (setf (elt a i) (cl-random 1000000)))
> > a))
> > (byte-compile 'make-random-array)
> > (benchmark '(setq foo (make-random-array (expt 10 n))) 1) -- 2.3 seconds
> > (benchmark '(sort foo '<) 1) -- 1 second
> >
> > ;; Second some Common Lisp, with times for SBCL on my $100 Chromebook.
> >
> > (defparameter n 6)
> > (defun make-random-array (n)
> > (declare (fixnum n))
> > (let ((a (make-array n)))
> > (declare (type array a))
> > (loop for i fixnum below n do
> > (setf (aref a i) (random most-positive-fixnum)))
> > a))
> > (time (defparameter foo (make-random-array (expt 10 n)))) -- .041
> seconds
> > (time (progn (stable-sort foo '<) nil)) -- .45 seconds
> >
> > Thanks so much for Emacs, which is so great that I cannot put it
> > into words.
> >
> > Bob
>
>
> [1] https://dreamsongs.com/Files/Timrep.pdf
>
[-- Attachment #2: Type: text/html, Size: 7180 bytes --]
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Some thoughts about Emacs performance
2024-02-16 14:08 ` Simon Leinen
2024-02-16 16:15 ` Robert Boyer
2024-02-16 16:31 ` Robert Boyer
@ 2024-02-16 18:06 ` Robert Boyer
2024-02-16 18:34 ` Robert Boyer
3 siblings, 0 replies; 11+ messages in thread
From: Robert Boyer @ 2024-02-16 18:06 UTC (permalink / raw)
To: Simon Leinen; +Cc: Emacs developers
[-- Attachment #1: Type: text/plain, Size: 23729 bytes --]
Dear Simon,
You mention 'random'. Below is 'random' in SBCL. I am clueless about how
good it is.
It takes people like Knuth to do this sort of thing well. Way over my head,
even though I have a Ph. D. in math.
I once heard the following random assertion: NSA, rms's favorite
organization I am sure, was at the time the largest employer of
mathematicians in the USA.
Did you know that the British spooks invented RSA before RSA did? Of course
those fucking spooks didn't tell the public. Criminal.
I wonder if Churchill had Turing, among others, in mind when he said: Never
in the field of human conflict was so much owed by so many to so few.
'Shit' is all I have to say about security, despite the large part of my
past
salary that might have been federally spook funded! Unlike rms, I like to
get
paid. At NSA's insistence, I got kicked off the board of my company
Computational Logic when I refused to get a security clearance.
Eventually, when funding dried up, the company folded.
Thank God for RSA. I can remember at a party for the MIT AI Lab, when
funding
did not look too hot, Moses said that 'security' might be a way to go to
look
for funding. It may be nice for him that our president gets $2,000,000,000
a year for the White House, in part to make it secure, but if anyone claims
to know
anything about security, I say, how come the news is constantly filled with
horror stories about hackers breaking in.
I once heard that Minsky, bless his soul, refused to let his students spend
time worrying about security on his DEC PDP-10 because it would only lead to
other students wasting time figuring out how to crack the system. But
'random' is one of the coolest ideas in the world. Where would we ever be
without hash tables! I am so glad that 'sxhash' is in Emacs Lisp and Common
Lisp!
I send far too much email. Is there a group Email Addicts Anonymous?
I am 77. I cannot pick up a coin on the floor. I am not to be trusted about
anything I say.
With highest regards,
Bob
I recommend to Emacs folks working on 'native-compiler' that they look at
the
18 uses of "declare" in the following! You want efficiency in Lisp, you use
"declare".
------------------------------------------------------------------------------
;;;; This software is part of the SBCL system. See the README file for
;;;; more information.
;;;;
;;;; This software is derived from the CMU CL system, which was
;;;; written at Carnegie Mellon University and released into the
;;;; public domain. The software is in the public domain and is
;;;; provided with absolutely no warranty. See the COPYING and CREDITS
;;;; files for more information.
(in-package "SB-KERNEL")
;;;; Constants
(defconstant mt19937-n 624)
(defconstant mt19937-m 397)
(defconstant mt19937-upper-mask #x80000000)
(defconstant mt19937-lower-mask #x7FFFFFFF)
(defconstant mt19937-a #x9908B0DF)
(defconstant mt19937-b #x9D2C5680)
(defconstant mt19937-c #xEFC60000)
;;;; RANDOM-STATEs
;;; The state is stored in a (simple-array (unsigned-byte 32) (627))
;;; wrapped in a random-state structure:
;;;
;;; 0-1: Constant matrix A. [0, #x9908b0df]
;;; 2: Index k.
;;; 3-626: State.
(deftype random-state-state () `(simple-array (unsigned-byte 32) (,(+ 3
mt19937-n))))
(defmethod make-load-form ((random-state random-state) &optional
environment)
(make-load-form-saving-slots random-state :environment environment))
(defmethod print-object ((state random-state) stream)
(if (and *print-readably* (not *read-eval*))
(print-not-readable-error state stream)
(format stream "#S(~S ~S #.~S)"
'random-state
':state
`(make-array ,(+ 3 mt19937-n)
:element-type
'(unsigned-byte 32)
:initial-contents
',(coerce (random-state-state state) 'list)))))
;;; Generate and initialize a new random-state array. Index is
;;; initialized to 1 and the states to 32bit integers excluding zero.
;;;
;;; Seed - A 32bit number.
;;;
;;; See Knuth TAOCP Vol2. 3rd Ed. P.106 for multiplier.
;;; In the previous versions, MSBs of the seed affect only MSBs of the
array.
(defun init-random-state (&optional (seed 5489) state)
(declare (type (unsigned-byte 32) seed))
(let ((state (or state (make-array 627 :element-type '(unsigned-byte
32)))))
(check-type state random-state-state)
(setf (aref state 0) 0)
(setf (aref state 1) mt19937-a)
(setf (aref state 2) mt19937-n)
(loop for i below mt19937-n
for p from 3
for s = seed then
(logand #xFFFFFFFF
(+ (* 1812433253
(logxor s (ash s -30)))
i))
do (setf (aref state p) s))
state))
(defvar *random-state*)
(defun !random-cold-init ()
(/show0 "entering !RANDOM-COLD-INIT")
(setf *random-state* (%make-random-state (init-random-state)))
(/show0 "returning from !RANDOM-COLD-INIT"))
;;; Q: Why is there both MAKE-RANDOM-STATE and SEED-RANDOM-STATE?
;;; A: Because the DEFKNOWN for MAKE-RANDOM-STATE is more restricted
;;; and doesn't accept numerical state.
(defun make-random-state (&optional state)
"Make a random state object. The optional STATE argument specifies a seed
for deterministic pseudo-random number generation.
As per the Common Lisp standard,
- If STATE is NIL or not supplied, return a copy of the default
*RANDOM-STATE*.
- If STATE is a random state, return a copy of it.
- If STATE is T, return a randomly initialized state (using operating-system
provided randomness where available, otherwise a poor substitute based on
internal time and PID).
See SB-EXT:SEED-RANDOM-STATE for a SBCL extension to this functionality."
(/show0 "entering MAKE-RANDOM-STATE")
(seed-random-state state))
(defun fallback-random-seed ()
;; When /dev/urandom is not available, we make do with time and pid
;; Thread ID and/or address of a CONS cell would be even better, but...
;; [ADDRESS-BASED-COUNTER-VAL in 'target-sxhash' could be used here]
(/show0 "No /dev/urandom, using randomness from time and pid")
(+ (get-internal-real-time)
(ash (sb-unix:unix-getpid) 32)))
#-win32
(defun os-random-seed ()
(or
;; On unices, we try to read from /dev/urandom and pass the results
;; to our (simple-array (unsigned-byte 32) (*)) processor below.
;; More than 256 bits would provide a false sense of security.
;; If you need more bits than that, you probably also need
;; a better algorithm too.
(ignore-errors
(with-open-file (r "/dev/urandom" :element-type '(unsigned-byte 32)
:direction :input :if-does-not-exist :error)
(let ((a (make-array '(8) :element-type '(unsigned-byte 32))))
(aver (= 8 (read-sequence a r)))
a)))
(fallback-random-seed)))
#+win32
(defun os-random-seed ()
(/show0 "Getting randomness from CryptGenRandom")
(or (sb-win32:crypt-gen-random 32)
(fallback-random-seed)))
(defun seed-random-state (&optional state)
"Make a random state object. The optional STATE argument specifies a seed
for deterministic pseudo-random number generation.
As per the Common Lisp standard for MAKE-RANDOM-STATE,
- If STATE is NIL or not supplied, return a copy of the default
*RANDOM-STATE*.
- If STATE is a random state, return a copy of it.
- If STATE is T, return a randomly initialized state (using operating-system
provided randomness where available, otherwise a poor substitute based on
internal time and pid).
As a supported SBCL extension, we also support receiving as a seed an object
of the following types:
- (SIMPLE-ARRAY (UNSIGNED-BYTE 8) (*))
- UNSIGNED-BYTE
While we support arguments of any size and will mix the provided bits into
the random state, it is probably overkill to provide more than 256 bits
worth
of actual information.
This particular SBCL version also accepts an argument of the following type:
(SIMPLE-ARRAY (UNSIGNED-BYTE 32) (*))
This particular SBCL version uses the popular MT19937 PRNG algorithm, and
its
internal state only effectively contains about 19937 bits of information.
http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html
"
(declare (explicit-check))
(named-let seed-random-state ((state state))
(etypecase state
;; Easy standard cases
(null
(/show0 "copying *RANDOM-STATE*")
(%make-random-state (copy-seq (random-state-state *random-state*))))
(random-state
(/show0 "copying the provided RANDOM-STATE")
(%make-random-state (copy-seq (random-state-state state))))
;; Standard case, less easy: try to randomly initialize a state.
((eql t)
(/show0 "getting randomness from the operating system")
(seed-random-state (os-random-seed)))
;; For convenience to users, we accept (simple-array (unsigned-byte 8)
(*))
;; We just convert it to (simple-array (unsigned-byte 32) (*)) in a
;; completely straightforward way.
;; TODO: probably similarly accept other word sizes.
((simple-array (unsigned-byte 8) (*))
(/show0 "getting random seed from byte vector (converting to 32-bit-word
vector)")
(let* ((l (length state))
(m (ceiling l 4))
(r (if (>= l 2496) 0 (mod l 4)))
(y (make-array (list m) :element-type '(unsigned-byte 32))))
(loop for i from 0 below (- m (if (zerop r) 0 1))
for j = (* i 4) do
(setf (aref y i)
(+ (aref state j)
(ash (aref state (+ j 1)) 8)
(ash (aref state (+ j 2)) 16)
(ash (aref state (+ j 3)) 24))))
(unless (zerop r) ;; The last word may require special treatment.
(let* ((p (1- m)) (q (* 4 p)))
(setf (aref y p)
(+ (aref state q)
(if (< 1 r) (ash (aref state (+ q 1)) 8) 0)
(if (= 3 r) (ash (aref state (+ q 2)) 16) 0)))))
(seed-random-state y)))
;; Also for convenience, we accept non-negative integers as seeds.
;; Small ones get passed to init-random-state, as before.
((unsigned-byte 32)
(/show0 "getting random seed from 32-bit word")
(%make-random-state (init-random-state state)))
;; Larger ones ones get trivially chopped into an array of (unsigned-byte
32)
((unsigned-byte)
(/show0 "getting random seed from bignum (converting to 32-bit-word
vector)")
(loop with l = (ceiling (integer-length state) 32)
with s = (make-array (list l) :element-type '(unsigned-byte 32))
for i below l
for p from 0 by 32
do (setf (aref s i) (ldb (byte 32 p) state))
finally (return (seed-random-state s))))
;; Last but not least, when provided an array of 32-bit words, we truncate
;; it to 19968 bits and mix these into an initial state. We reuse the same
;; method as the authors of the original algorithm. See
;;
http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/MT2002/CODES/mt19937ar.c
;; NB: their mt[i] is our (aref s (+ 3 i))
((simple-array (unsigned-byte 32) (*))
(/show0 "getting random seed from 32-bit-word vector")
(let ((s (init-random-state 19650218))
(i 1) (j 0) (l (length state)))
(loop for k downfrom (max mt19937-n l) above 0 do
(setf (aref s (+ i 3))
(logand #xFFFFFFFF
(+ (logxor (aref s (+ i 3))
(* 1664525
(logxor (aref s (+ i 2))
(ash (aref s (+ i 2)) -30))))
(aref state j) j))) ;; non-linear
(incf i) (when (>= i mt19937-n) (setf (aref s 3) (aref s (+ 2
mt19937-n)) i 1))
(incf j) (when (>= j l) (setf j 0)))
(loop for k downfrom (1- mt19937-n) above 0 do
(setf (aref s (+ i 3))
(logand #xFFFFFFFF
(- (logxor (aref s (+ i 3))
(* 1566083941
(logxor (aref s (+ i 2))
(ash (aref s (+ i 2)) -30))))
i))) ;; non-linear
(incf i) (when (>= i mt19937-n) (setf (aref s 3) (aref s (+ 2
mt19937-n)) i 1)))
(setf (aref s 3) #x80000000) ;; MSB is 1; assuring non-zero initial array
(%make-random-state s))))))
;;;; random entries
;;; This function generates a 32bit integer between 0 and #xffffffff
;;; inclusive.
;;; portable implementation
#-x86
(defun random-mt19937-update (state)
(declare (type random-state-state state)
(optimize (speed 3) (safety 0)))
(let ((y 0))
(declare (type (unsigned-byte 32) y))
(do ((kk 3 (1+ kk)))
((>= kk (+ 3 (- mt19937-n mt19937-m))))
(declare (type (mod 628) kk))
(setf y (logior (logand (aref state kk) mt19937-upper-mask)
(logand (aref state (1+ kk)) mt19937-lower-mask)))
(setf (aref state kk) (logxor (aref state (+ kk mt19937-m))
(ash y -1) (aref state (logand y 1)))))
(do ((kk (+ (- mt19937-n mt19937-m) 3) (1+ kk)))
((>= kk (+ (1- mt19937-n) 3)))
(declare (type (mod 628) kk))
(setf y (logior (logand (aref state kk) mt19937-upper-mask)
(logand (aref state (1+ kk)) mt19937-lower-mask)))
(setf (aref state kk) (logxor (aref state (+ kk (- mt19937-m mt19937-n)))
(ash y -1) (aref state (logand y 1)))))
(setf y (logior (logand (aref state (+ 3 (1- mt19937-n)))
mt19937-upper-mask)
(logand (aref state 3) mt19937-lower-mask)))
(setf (aref state (+ 3 (1- mt19937-n)))
(logxor (aref state (+ 3 (1- mt19937-m)))
(ash y -1) (aref state (logand y 1)))))
(values))
(declaim (start-block random %random-single-float %random-double-float
random-chunk big-random-chunk))
(declaim (inline random-chunk))
#-x86
(defun random-chunk (state)
(declare (type random-state state))
(let* ((state (random-state-state state))
(k (aref state 2)))
(declare (type (mod 628) k))
(when (= k mt19937-n)
(random-mt19937-update state)
(setf k 0))
(setf (aref state 2) (1+ k))
(let ((y (aref state (+ 3 k))))
(declare (type (unsigned-byte 32) y))
(setf y (logxor y (ash y -11)))
(setf y (logxor y (ash (logand y (ash mt19937-b -7)) 7)))
(setf y (logxor y (ash (logand y (ash mt19937-c -15)) 15)))
(setf y (logxor y (ash y -18)))
y)))
;;; Using inline VOP support, only available on the x86 so far.
;;;
;;; FIXME: It would be nice to have some benchmark numbers on this.
;;; My inclination is to get rid of the nonportable implementation
;;; unless the performance difference is just enormous.
#+x86
(defun random-chunk (state)
(declare (type random-state state))
(sb-vm::random-mt19937 (random-state-state state)))
(declaim (inline big-random-chunk))
(defun big-random-chunk (state)
(declare (type random-state state))
(logior (ash (random-chunk state) 32)
(random-chunk state)))
;;; Handle the single or double float case of RANDOM. We generate a
;;; float between 0.0 and 1.0 by clobbering the significand of 1.0
;;; with random bits, then subtracting 1.0. This hides the fact that
;;; we have a hidden bit.
(declaim (inline %random-single-float %random-double-float))
(declaim (ftype (function ((single-float ($0f0)) random-state)
(single-float $0f0))
%random-single-float))
(defun %random-single-float (arg state)
(declare (type (single-float ($0f0)) arg)
(type random-state state))
(loop for candidate of-type single-float
= (* arg
(- (make-single-float
(dpb (ash (random-chunk state)
(- sb-vm:single-float-digits n-random-chunk-bits))
sb-vm:single-float-significand-byte
(single-float-bits $1.0)))
$1.0))
while (#+x86 eql ;; Can't use = due to 80-bit precision
#-x86 =
candidate arg)
finally (return candidate)))
(declaim (ftype (function ((double-float ($0d0)) random-state)
(double-float $0d0))
%random-double-float))
;;; 32-bit version
#+nil
(defun %random-double-float (arg state)
(declare (type (double-float ($0d0)) arg)
(type random-state state))
(* (float (random-chunk state) $1d0) (/ $1d0 (expt 2 32))))
;;; 53-bit version
#-x86
(defun %random-double-float (arg state)
(declare (type (double-float ($0d0)) arg)
(type random-state state))
(loop for candidate of-type double-float
= (* arg
(- (sb-impl::make-double-float
(dpb (ash (random-chunk state)
(- sb-vm:double-float-digits n-random-chunk-bits 32))
sb-vm:double-float-significand-byte
(sb-impl::double-float-high-bits $1d0))
(random-chunk state))
$1d0))
while (= candidate arg)
finally (return candidate)))
;;; using a faster inline VOP
#+x86
(defun %random-double-float (arg state)
(declare (type (double-float ($0d0)) arg)
(type random-state state))
(let ((state-vector (random-state-state state)))
(loop for candidate of-type double-float
= (* arg
(- (sb-impl::make-double-float
(dpb (ash (sb-vm::random-mt19937 state-vector)
(- sb-vm:double-float-digits n-random-chunk-bits
sb-vm:n-word-bits))
sb-vm:double-float-significand-byte
(sb-impl::double-float-high-bits $1d0))
(sb-vm::random-mt19937 state-vector))
$1d0))
;; Can't use = due to 80-bit precision
while (eql candidate arg)
finally (return candidate))))
;;;; random fixnums
;;; Generate and return a pseudo random fixnum less than ARG. To achieve
;;; equidistribution an accept-reject loop is used.
;;; No extra effort is made to detect the case of ARG being a power of
;;; two where rejection is not possible, as the cost of checking for
;;; this case is the same as doing the rejection test. When ARG is
;;; larger than (expt 2 N-RANDOM-CHUNK-BITS), which can only happen if
;;; the random chunk size is half the word size, two random chunks are
;;; used in each loop iteration, otherwise only one. Finally, the
;;; rejection probability could often be reduced by not masking the
;;; chunk but rejecting only values as least as large as the largest
;;; multiple of ARG that fits in a chunk (or two), but this is not done
;;; as the speed gains due to needing fewer loop iterations are by far
;;; outweighted by the cost of the two divisions required (one to find
;;; the multiplier and one to bring the result into the correct range).
(declaim (inline %random-fixnum))
(defun %random-fixnum (arg state)
(declare (type (integer 1 #.most-positive-fixnum) arg)
(type random-state state))
(if (= arg 1)
0
(let* ((n-bits (integer-length (1- arg)))
(mask (1- (ash 1 n-bits))))
(macrolet ((accept-reject-loop (generator)
`(loop
(let ((bits (logand mask (,generator state))))
(when (< bits arg)
(return bits))))))
(aver (<= n-bits (* 2 n-random-chunk-bits)))
(if (<= n-bits n-random-chunk-bits)
(accept-reject-loop random-chunk)
(accept-reject-loop big-random-chunk))))))
(defun random (arg &optional (state *random-state*))
(declare (inline %random-fixnum
%random-single-float %random-double-float
#+long-float %random-long-float))
(declare (explicit-check))
(cond
((and (fixnump arg) (> arg 0))
(%random-fixnum arg state))
((and (typep arg 'single-float) (> arg $0.0f0))
(%random-single-float arg state))
((and (typep arg 'double-float) (> arg $0.0d0))
(%random-double-float arg state))
#+long-float
((and (typep arg 'long-float) (> arg $0.0l0))
(%random-long-float arg state))
((and (bignump arg) (> arg 0))
(%random-bignum arg state))
(t
(error 'simple-type-error
:expected-type '(or (integer 1) (float (0))) :datum arg
:format-control "~@<Argument is neither a positive integer nor a ~
positive float: ~2I~_~S~:>"
:format-arguments (list arg)))))
;;;; This implementation of RANDOM is based on the Mersenne Twister random
;;;; number generator "MT19937" due to Matsumoto and Nishimura. See:
;;;; Makoto Matsumoto and T. Nishimura, "Mersenne twister: A
;;;; 623-dimensionally equidistributed uniform pseudorandom number
;;;; generator.", ACM Transactions on Modeling and Computer Simulation,
;;;; Vol. 8, No. 1, January pp.3-30 (1998) DOI:10.1145/272991.272995
;;;; http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html
On Fri, Feb 16, 2024 at 8:08 AM Simon Leinen <simon.leinen@gmail.com> wrote:
> Bob,
>
> welcome to the emacs-devel list, and thanks a lot for *your* wonderful
> contributions (theorem prover, string search, and Lisp benchmarking -
> I remember boyer.lisp from RPG's reference work[1]).
>
> On Thu, Feb 8, 2024 at 8:15 AM Robert Boyer
> <robertstephenboyer@gmail.com> wrote:
> >
> > Emacs 27.1 has a 'sort' function that takes longer than stable-sort of
> SBCL. Maybe
> > by a factor of 2. See also my attached file 'ms.lisp'.
> >
> > There may be a lot that can be improved in Emacs'
> > handling of cl-loop, setf, elt, or cl-random.
>
> In this case, cl-random seems to be the main culprit for the slow
> initialization—replacing that with plain "random" speeds it up by
> about a factor of ten. There was some discussion on the list recently
> about cl-random vs. random. The main functional difference is that
> cl-random supports a defined state. But the performance difference may
> be due more to the fact that random is written in C, and cl-random in
> Lisp.
>
> As for the sorting itself, both Emacs and SBCL seem to use mergesort
> in their implementations of (stable-)sort. Emacs's implementation is
> written in C, SBCL's in Lisp. Performance is quite similar—on my
> system (Apple Macbook Air M2) Emacs takes about 35% longer to sort a
> million random numbers than SBCL. (On the other hand when sorting it
> again, i.e. when the vector is already fully sorter, Emacs is quite a
> bit faster than SBCL—maybe Emacs chose to optimize for partly-sorted
> vectors at the expense of a bit of performance for random input.)
>
> In general, the Emacs Lisp runtime system and compiler(s) aren't as
> optimized as SBCL for general Lisp use. But it gets quite close!
>
> On the other hand, Emacs has editor-specific code (e.g. redisplay) and
> data structures (e.g. buffers) which are highly optimized and partly
> written in C. But it doesn't try to be a standalone platform for
> performance-oriented Lisp developers. Of course Emacs is very
> suitable as a Software Development Environment for systems such as
> SBCL, and there are many good integration options—personally I use the
> SLIME package these days.
>
> Best regards, and enjoy Lisping in Emacs!
> --
> Simon.
>
> > ;; First some Emacs, with times on my $100 Chromebook.
> >
> > (setq n 6)
> > (defun make-random-array (n)
> > (let ((a (make-vector n 0)))
> > (cl-loop for i below n do
> > (setf (elt a i) (cl-random 1000000)))
> > a))
> > (byte-compile 'make-random-array)
> > (benchmark '(setq foo (make-random-array (expt 10 n))) 1) -- 2.3 seconds
> > (benchmark '(sort foo '<) 1) -- 1 second
> >
> > ;; Second some Common Lisp, with times for SBCL on my $100 Chromebook.
> >
> > (defparameter n 6)
> > (defun make-random-array (n)
> > (declare (fixnum n))
> > (let ((a (make-array n)))
> > (declare (type array a))
> > (loop for i fixnum below n do
> > (setf (aref a i) (random most-positive-fixnum)))
> > a))
> > (time (defparameter foo (make-random-array (expt 10 n)))) -- .041
> seconds
> > (time (progn (stable-sort foo '<) nil)) -- .45 seconds
> >
> > Thanks so much for Emacs, which is so great that I cannot put it
> > into words.
> >
> > Bob
>
>
> [1] https://dreamsongs.com/Files/Timrep.pdf
>
[-- Attachment #2: Type: text/html, Size: 26985 bytes --]
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Some thoughts about Emacs performance
2024-02-16 14:08 ` Simon Leinen
` (2 preceding siblings ...)
2024-02-16 18:06 ` Robert Boyer
@ 2024-02-16 18:34 ` Robert Boyer
3 siblings, 0 replies; 11+ messages in thread
From: Robert Boyer @ 2024-02-16 18:34 UTC (permalink / raw)
To: Simon Leinen; +Cc: Emacs developers, Gerard Huet
[-- Attachment #1: Type: text/plain, Size: 5245 bytes --]
>
> > But it doesn't try to be a standalone platform for
> > performance-oriented Lisp developers.
>
> I hope that Emacs merges with SBCL.
>
> Emacs and SBCL are amongst the greatest inventions in the history of the
> world.
>
> Of course I should mention gcc and Linux and so much that the Free
> Software Foundation has given us, e.g., their 'sort' which can sort files
> of amazing length at amazing speed. It is crucial to my ongoing work on
> automatic theorem proving. If you've got a billion clause records and want
> to flush duplicates, who you gonna call but /usr/bin/sort. That's what I
> say, and what I do. To do that in Lisp might consume more space than I
> have except on my SD card. I run on a $100 chromebook and my 256 gb SD
> card only cost $30, and 1 tb SD cards are coming soon at a price I think I
> will be able to afford.
> But of course, Coq may be the fastest thing except for C. And it has
> 'sanitary' macros.
>
> Where would I be without unsanitary macros! My sorting algorithm 'msa'
> depends upon them!
>
> But what do I know about sanitation.
I think that the following are very unsanitary, but what do I know?
(eval
> '(defmacro key (x)
> (cond (key
> (cond ((symbolp key) `(the ,msa-type (,key ,x)))
> (t `(the ,msa-type (funcall ,key ,x)))))
> (t `(the ,msa-type ,x)))))
> (eval
> '(defmacro msa-equal (x y)
> (cond (msa-all-keys-real `(eql ,x ,y))
> (t `(equal ,x ,y)))))
> (eval
> '(defmacro msa-predicate (x y)
> (cond ((symbolp msa-predicate) `(,msa-predicate ,x ,y))
> (t `(funcall ',msa-predicate ,x ,y)))))
On Fri, Feb 16, 2024 at 8:08 AM Simon Leinen <simon.leinen@gmail.com> wrote:
> Bob,
>
> welcome to the emacs-devel list, and thanks a lot for *your* wonderful
> contributions (theorem prover, string search, and Lisp benchmarking -
> I remember boyer.lisp from RPG's reference work[1]).
>
> On Thu, Feb 8, 2024 at 8:15 AM Robert Boyer
> <robertstephenboyer@gmail.com> wrote:
> >
> > Emacs 27.1 has a 'sort' function that takes longer than stable-sort of
> SBCL. Maybe
> > by a factor of 2. See also my attached file 'ms.lisp'.
> >
> > There may be a lot that can be improved in Emacs'
> > handling of cl-loop, setf, elt, or cl-random.
>
> In this case, cl-random seems to be the main culprit for the slow
> initialization—replacing that with plain "random" speeds it up by
> about a factor of ten. There was some discussion on the list recently
> about cl-random vs. random. The main functional difference is that
> cl-random supports a defined state. But the performance difference may
> be due more to the fact that random is written in C, and cl-random in
> Lisp.
>
> As for the sorting itself, both Emacs and SBCL seem to use mergesort
> in their implementations of (stable-)sort. Emacs's implementation is
> written in C, SBCL's in Lisp. Performance is quite similar—on my
> system (Apple Macbook Air M2) Emacs takes about 35% longer to sort a
> million random numbers than SBCL. (On the other hand when sorting it
> again, i.e. when the vector is already fully sorter, Emacs is quite a
> bit faster than SBCL—maybe Emacs chose to optimize for partly-sorted
> vectors at the expense of a bit of performance for random input.)
>
> In general, the Emacs Lisp runtime system and compiler(s) aren't as
> optimized as SBCL for general Lisp use. But it gets quite close!
>
> On the other hand, Emacs has editor-specific code (e.g. redisplay) and
> data structures (e.g. buffers) which are highly optimized and partly
> written in C. But it doesn't try to be a standalone platform for
> performance-oriented Lisp developers. Of course Emacs is very
> suitable as a Software Development Environment for systems such as
> SBCL, and there are many good integration options—personally I use the
> SLIME package these days.
>
> Best regards, and enjoy Lisping in Emacs!
> --
> Simon.
>
> > ;; First some Emacs, with times on my $100 Chromebook.
> >
> > (setq n 6)
> > (defun make-random-array (n)
> > (let ((a (make-vector n 0)))
> > (cl-loop for i below n do
> > (setf (elt a i) (cl-random 1000000)))
> > a))
> > (byte-compile 'make-random-array)
> > (benchmark '(setq foo (make-random-array (expt 10 n))) 1) -- 2.3 seconds
> > (benchmark '(sort foo '<) 1) -- 1 second
> >
> > ;; Second some Common Lisp, with times for SBCL on my $100 Chromebook.
> >
> > (defparameter n 6)
> > (defun make-random-array (n)
> > (declare (fixnum n))
> > (let ((a (make-array n)))
> > (declare (type array a))
> > (loop for i fixnum below n do
> > (setf (aref a i) (random most-positive-fixnum)))
> > a))
> > (time (defparameter foo (make-random-array (expt 10 n)))) -- .041
> seconds
> > (time (progn (stable-sort foo '<) nil)) -- .45 seconds
> >
> > Thanks so much for Emacs, which is so great that I cannot put it
> > into words.
> >
> > Bob
>
>
> [1] https://dreamsongs.com/Files/Timrep.pdf
>
[-- Attachment #2: Type: text/html, Size: 7835 bytes --]
^ permalink raw reply [flat|nested] 11+ messages in thread