bug-gnu-emacs
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

bug#69249: bug in native-compile?


From: Robert Boyer
Subject: bug#69249: bug in native-compile?
Date: Sun, 18 Feb 2024 17:46:09 -0600

Robert Boyer robertstephenboyer@gmail.com

4:39 PM (1 hour ago)
Reply to all
to AndreasEli69249AndreaStefanrms
Look out, here it comes, as Tracy Schwarz says.

The following DISASSEMBLE might tell you what you need to do for native-compile. 

Speaking with the utmost sincerity, I know less about the following than I do about what Proust called the greatest painting in the world, the View of Delft by Vermmer.  Check it out.

https://www.mauritshuis.nl/en/our-collection/artworks/92-view-of-delft/

Again, this is in SBCL on my $100 Lenovo Chromebook.

* (disassemble 'foo-with-declare)
disassembly for FOO-WITH-DECLARE
; Size: 60 bytes. Origin: #x52A4AF19                          ; FOO-WITH-DECLARE
; 19:       4883EC10         SUB RSP, 16
; 1D:       BA14000000       MOV EDX, 20
; 22:       B904000000       MOV ECX, 4
; 27:       48892C24         MOV [RSP], RBP
; 2B:       488BEC           MOV RBP, RSP
; 2E:       B842BD3550       MOV EAX, #x5035BD42              ; #<FDEFN EXPT>
; 33:       FFD0             CALL RAX
; 35:       31C0             XOR EAX, EAX
; 37:       31C9             XOR ECX, ECX
; 39:       EB0C             JMP L1
; 3B:       0F1F440000       NOP
; 40: L0:   4801C1           ADD RCX, RAX
; 43:       4883C002         ADD RAX, 2
; 47: L1:   4839D0           CMP RAX, RDX
; 4A:       7EF4             JLE L0
; 4C:       488BD1           MOV RDX, RCX
; 4F:       488BE5           MOV RSP, RBP
; 52:       F8               CLC
; 53:       5D               POP RBP
; 54:       C3               RET
NIL



On Sun, Feb 18, 2024 at 5:42 PM Robert Boyer <robertstephenboyer@gmail.com> wrote:
 If so, we'd be interested to see your code to try and improve our compiler.

I hope this helps.  If not, please let me know.



Robert Boyer robertstephenboyer@gmail.com

4:19 PM (1 hour ago)
Reply to all
to AndreasEli69249AndreaStefanrms
This message is about how important DECLARE and THE are.  If you do not
handle DECLARE and THE right, people will laugh at native-compile.

Here is an SBCL transcript.  I am running on a $100 Lenovo Chromebook.

I got SBCL with the command sudo apt-get install sbcl.

* (declaim (optimize (safety 0) (speed 3) (debug 0)))
NIL
* (defun foo-with-declare (n)
  (let ((i 0)
        (sum 0)
        (max (expt 10 n)))
    (declare (fixnum i sum max))
    (loop (cond ((<= i max)
                 (setq sum (the fixnum (+ sum i))))
                (t (return sum)))
          (incf i))))
FOO-WITH-DECLARE
* (time (foo-with-declare 8))
Evaluation took:
  0.125 seconds of real time
  0.123203 seconds of total run time (0.123193 user, 0.000010 system)
  98.40% CPU
  135,905,516 processor cycles
  0 bytes consed
5000000050000000
* (defun foo-without-declare (n)
  (let ((i 0)
        (sum 0)
        (max (expt 10 n)))
    (loop (cond ((<= i max)
                 (setq sum (+ sum i)))
                (t (return sum)))
          (incf i))))
WARNING: redefining COMMON-LISP-USER::FOO-WITHOUT-DECLARE in DEFUN
FOO-WITHOUT-DECLARE
* (time (foo-without-declare 8))
Evaluation took:
  1.080 seconds of real time
  1.072932 seconds of total run time (1.072932 user, 0.000000 system)
  99.35% CPU
  1,181,369,303 processor cycles
  0 bytes consed
5000000050000000

People go wow about 8X.  How does native-compile do?

There is a great song by Alabama 'If you gonna play in Texas, you gotta have a fiddle in the band'. Check it out!


Bob


Bob


On Sun, Feb 18, 2024 at 3:28 PM Andreas Schwab <schwab@linux-m68k.org> wrote:
On Feb 18 2024, Robert Boyer wrote:

> Furthermore, such declare forms are necessary for life as we know it
> in the free world, so that fixnum arithmetic can be used where
> appropriate!

If you are programming in Emacs Lisp, you need to follow the rules of
Emacs Lisp.

--
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."


--
Anything I seem to state should be taken as a question.  I am at least 77 and feeble.
Reply allReplyForward

On Sun, Feb 18, 2024 at 5:32 PM Stefan Monnier <monnier@iro.umontreal.ca> wrote:
> ;; However, (native-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

First thing: these are not error messages!  They're *warnings*!

`batch-byte-compile` also gives those warnings, because ELisp does not
know those declarations.  The native compiler does support some type
annotations, but they do not take this shape.

I presume you took those from some other Lisp dialect, but that won't do
you much good.  Scheme is not Clojure is not Common Lisp is not ELisp ...

BTW, A more recent Emacs would give you an additional warning:

    Warning: file has no ‘lexical-binding’ directive on its first line

So, I see no bug.  Did the resulting native-compile misbehave in
any way?  I presume you're worried about its performance.
If so, do you have concrete code where you measured the performance and
were disappointed and where you have good reasons to believe that type
annotations would make a difference?  If so, we'd be interested to see
your code to try and improve our compiler.


        Stefan



--
Anything I seem to state should be taken as a question.  I am at least 77 and feeble.


--
Anything I seem to state should be taken as a question.  I am at least 77 and feeble.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]