Hi Eliot,


> Interesting! No, there’s no swap as yet. If SqueakV3PlusClosures has an unused bytecode I’d say go for it. (SistaV1 definitely has unused bytecodes).


Great to hear this! I would be glad to join the implementation of a new bytecode :-)


> pop and store temp: N (where N gets to sp - 1)

Which bytecode are you referring to? #genStoreTemp:? This sounds indeed hacky ...

Just another question about bytecodes: Do we currently have any tools for exploring and playing around with the bytecodes? I would love to do say something like

Encoder default generateMethod: [:stack :encoder |
    encoder genPushReceiver.
    encoder genReturnTop].

But afaik, at the moment I have to manipulate the implementation of #emitCodeForValue:encoder: or so if I would like to test something new. Am I missing any existing tools to do so?

(Imagine if you could write your own bytecodes directly into a CodeHolder ... :D)

Best,
Christoph

________________________________
Von: Squeak-dev <squeak-dev-bounces at lists.squeakfoundation.org> im Auftrag von Eliot Miranda <eliot.miranda at gmail.com>
Gesendet: Sonntag, 23. Februar 2020 18:19 Uhr
An: The general-purpose Squeak developers list
Betreff: Re: [squeak-dev] Swap operation for stack machinery?



On Feb 23, 2020, at 9:01 AM, Thiede, Christoph <Christoph.Thiede at student.hpi.uni-potsdam.de> wrote:



Hi all,


I am curious to know whether the Squeak Context model provides any operation to swap the latest two elements from the stack. Something like:


doSwap

    | first second |

    first := self pop.

    second := self pop.

    self push: first.

    self push: second.

If it exists, I could not find it yet. If it does not exist, would you see any need to implement it (for both VM & simulation, of course)?


To me, this appears to be the most performant approach to implement a parametrized #caseError:<http://forum.world.st/Problems-with-caseError-tp5111930p5112251.html>. Something like:


81 <75> pushConstant: 0

82 <88> dup

83 <76> pushConstant: 1

84 <B6> send: =

85 <9A> jumpFalse: 89

86 <87> pop

87 <22> pushConstant: #one

88 <96> jumpTo: 97

89 <77> pushConstant: 2

90 <B6> send: =

91 <99> jumpFalse: 94

92 <21> pushConstant: #two

93 <91> jumpTo: 97

94 <70> self

95 <xx> swap

96 <D0> send: caseError:

97 <87> pop

Otherwise, we would need an extra temporary variable here.

BTW there is a horrible hack which avoids the extra temp.  Since the stack pointer and stack height are known during code generation then instead of this:

94 <70> self

95 <xx> swap

96 <D0> send: caseError:

one could emit this sequence:

    dup
    push self
    pop and store temp: N (where N gets to sp - 1)
    send caseError:

Looking forward to your answers.


Best,

Christoph

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20200223/be4ec5ea/attachment.html>