[P4-dev] Where should the operations on arbitrary-precision integers be declared?

Javier richardqa at gmail.com
Mon Jun 20 16:26:28 EDT 2016


It is "9a2d5aa38" march 24:(

Please let me try to do it again with the new commit. I write then

Thank you so much
Javier

2016-06-20 17:17 GMT-03:00 Antonin Bas <antonin at barefootnetworks.com>:

> What is the latest commit in your behavioral-model clone? Because I added
> modulo support this morning...
>
> On Mon, Jun 20, 2016 at 1:12 PM, Javier <richardqa at gmail.com> wrote:
>
>> Hi Antonin,
>>
>> Thank you so much for the reply,
>>
>> I tried to replicate the code simple_router.p4 with p4v1.1 using the
>> ideas that you comment: "http://pastebin.com/hukUDq4e"
>>
>> As you see, I added one new action "myaction" and one new table "c".
>>
>> From command.txt, I just add one entry:
>>
>> table_add c myaction 0 => 5 3 0
>>
>> Now, in the action ""set_nhop", I changed this line:
>>
>> +    ipv4.ttl = ipv4.ttl - 1;
>> -     ipv4.ttl = ipv4.ttl - routing_metadata.f1;
>>
>> Now, I execute run_demo.sh and all ok, but when I do ping between h1 and
>> h2  I should get this value 62? (64 - 5mod3), However i didn't get to ping
>> between them (no reply).
>> If I change the operator '%' to '*' operator It works  64 - 3*5=49 when
>> ping h1,h2.
>>
>> Could you tell me whether am I doing something bad please?
>>
>>
>> Thank you in advance,
>> Javier
>>
>>
>>
>>
>> 2016-06-20 15:39 GMT-03:00 Antonin Bas <antonin at barefootnetworks.com>:
>>
>>> Hi,
>>>
>>>
>>> I just noticed that modulo and divide where not supported by the P4
>>> frontend, so I made the change. In P4v1.1, using modulo would look like
>>> this:
>>> action my_action(in int<32> a, in int<32> b) {
>>>     modify_field(*h1.f1*, a % b);  // a >= 0, b > 0
>>>     // or *h1.f1* = a % b;
>>> }
>>> The second syntax is supported by the p4-hlir compiler, but I don't
>>> think it is described in the official spec.
>>> This action can then be referenced by a table.
>>> Really now that I have made the change in the frontend, using "modulo"
>>> is no different from using "add" or "substract". See
>>> https://github.com/p4lang/tutorials/blob/master/p4v1_1/simple_router/p4src/simple_router.p4#L129
>>> for a P4v1.1 program.
>>>
>>> Thanks,
>>>
>>> Antonin
>>>
>>> On Mon, Jun 20, 2016 at 4:53 AM, Christian Esteve Rothenberg <
>>> chesteve at dca.fee.unicamp.br> wrote:
>>>
>>>> Thanks Antonin for your comprehensive reply!
>>>>
>>>> Are you aware of any publicly available P4 code example using the
>>>> modulo operation?
>>>>
>>>> Just a code snippet would suffice...
>>>>
>>>> thanks,
>>>> Christian
>>>>
>>>> On Fri, Jun 17, 2016 at 5:05 PM, Antonin Bas
>>>> <antonin at barefootnetworks.com> wrote:
>>>> > Hi Javier,
>>>> >
>>>> > In P4v1.1, the only variables you have are fields (packet fields and
>>>> > metadata fields), and they have global scope. These variables all
>>>> have a
>>>> > type which includes their bitwidth (with the exception of the very
>>>> specific
>>>> > variable-length field case). For example, the ethernet header can be
>>>> > declared as follows:
>>>> >
>>>> > header_type ethernet_t {
>>>> >     fields {
>>>> >         bit<48> dstAddr;
>>>> >         bit<48> srcAddr;
>>>> >         bit<16> etherType;
>>>> >     }
>>>> > }
>>>> >
>>>> >
>>>> > Notice that all three fields are bitstrings. One can also declare
>>>> fields has
>>>> > signed integers (int<w>), usually this is more useful for some
>>>> metadata
>>>> > fields.
>>>> > As mentioned in the spec, you cannot declare fields a "int" (without a
>>>> > width). This is reserved for evaluation of compile-time constants.
>>>> The "int"
>>>> > type is used by the compiler but is hidden from the P4 programmer. For
>>>> > example, let's look at the following code:
>>>> > ethernet.etherType = (bit<16>) (11111111 - 1);
>>>> > The compiler will use the internal "int" type to represent
>>>> compile-time
>>>> > constants 11111111 and 1. It will also compute the difference,
>>>> 11111110,
>>>> > which is also of type "int". This is what we call "arbitrary-precision
>>>> > arithmetic" since the compiler can handle arbitrarily large compile
>>>> time
>>>> > constants. Finally the compiler should give you a warning because
>>>> 11111110
>>>> > cannot be represented using 16 bits and the most significant bits
>>>> will be
>>>> > lost.
>>>> >
>>>> > Best,
>>>> >
>>>> > Antonin
>>>> >
>>>> >
>>>> > On Thu, Jun 16, 2016 at 4:13 PM, Javier <richardqa at gmail.com> wrote:
>>>> >>
>>>> >> Hi everyone,
>>>> >>
>>>> >> I am trying to use the operations on "arbitrary-precison constant
>>>> >> integers", however I dont get to know how to declare these constant
>>>> >> integers. In the P4 spec for v1.1 tell that "the only values that
>>>> can have
>>>> >> the type int are compile-time constants" but it doesn't show how to
>>>> use it.
>>>> >> For example, i want to use the operator modulo to get the modulo of
>>>> two
>>>> >> variables, where and how should I declare these variables?.
>>>> >>
>>>> >> Thank in advance,
>>>> >> Javier
>>>> >>
>>>> >>
>>>> >> --
>>>> >> Saludos Cordiales
>>>> >> Javier
>>>> >> Fingerprint: 7AF8 4E03 66C2 2663 3A77  7067 6B5A D8F9 EE91 47BE
>>>> >>
>>>> >> _______________________________________________
>>>> >> P4-dev mailing list
>>>> >> P4-dev at lists.p4.org
>>>> >> http://lists.p4.org/mailman/listinfo/p4-dev_lists.p4.org
>>>> >
>>>> >
>>>> >
>>>> >
>>>> > --
>>>> > Antonin
>>>> >
>>>> > _______________________________________________
>>>> > P4-dev mailing list
>>>> > P4-dev at lists.p4.org
>>>> > http://lists.p4.org/mailman/listinfo/p4-dev_lists.p4.org
>>>>
>>>
>>>
>>>
>>> --
>>> Antonin
>>>
>>
>>
>>
>> --
>> Saludos Cordiales
>> Javier
>> PhD Student (UNICAMP)
>> http://www.linkedin.com/in/richardqa
>> Fingerprint: 7AF8 4E03 66C2 2663 3A77  7067 6B5A D8F9 EE91 47BE
>>
>
>
>
> --
> Antonin
>



-- 
Saludos Cordiales
Javier
PhD Student (UNICAMP)
http://www.linkedin.com/in/richardqa
Fingerprint: 7AF8 4E03 66C2 2663 3A77  7067 6B5A D8F9 EE91 47BE
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.p4.org/pipermail/p4-dev_lists.p4.org/attachments/20160620/60e27bff/attachment-0002.html>


More information about the P4-dev mailing list