[LLVMdev] Using the New Attributes Classes

David Blaikie dblaikie at gmail.com
Sat Feb 9 12:56:19 PST 2013


On Feb 9, 2013 8:45 AM, "Bill Wendling" <wendling at apple.com> wrote:
>
>                       Using the New Attributes Classes

Thanks for writing this up, Bill. Makes a fair bit of sense & should be
useful.

I have a few questions/comments but I'm not suggesting anything should be
changed (even if my ideas were solid, which I've no reason to believe they
are, I'm not really a stakeholder in this part of LLVM)

> Attributes in LLVM have changed in some fundamental ways. It was
necessary to do
> this to support expanding the attributes to encompass more than a handful
of
> attributes --- e.g. command line options. The old way of handling
attributes
> consisted of representing them as a bit mask of values. This bit mask was
stored
> in a "list" structure that was reference counted. The advantage of this
was that
> attributes could be manipulated with 'or's and 'and's. The disadvantage
of this
> was that there was limited room for expansion, and virtually no support
for
> attribute-value pairs other than alignment.
>
> In the new scheme, an Attribute object represents a single attribute
that's
> uniqued. You use the "Attribute::get" methods to create a new Attribute
> object. An attribute can be a single "enum" value (the enum being the
> Attribute::AttrKind enum), a string representing a target-dependent
attribute,
> or an attribute-value pair. Some examples:
>
>    Target-independent:   noinline, zext
>    Target-dependent:     "no-sse", "thumb2"
>    Attribute-value pair: "cpu" = "cortex-a8", align = 4
>
> (Note: for an attribute value pair, we expect a target-dependent
attribute to
> have a string for the value.)
>
> An Attribute object is designed to be passed around by value.
>
> Because attributes are no longer represented as a bit mask, you will need
to
> convert any code which does treat them as a bit mask to use the new query
> methods on the Attribute class. This should be straight forward. If there
is
> missing functionality on the Attribute class, which you feel should be
there,
> please let me know about it.
>
> The next class is the AttributeSet class. This replaces the old
AttributeList
> class. The AttributeSet stores a collection of Attribute objects for each
kind
> of object that may have an attribute associated with it: the function as a
> whole, the return type, or the function's parameters. A function's
attributes
> are at index "AttributeSet::FunctionIndex"; the return type's attributes
are at
> index "AttributeSet::ReturnIndex"; and the function's parameters'
attributes are
> at indices 1, ..., n (where 'n' is the number of parameters). Most
methods on
> the AttributeSet class take an index parameter.

This seems like a slightly curious factoring: AttributeSet is actually a
list of sets of attributes, yes? Why does the list end up encapsulated
within this abstraction rather than something more like
vector<AttributeSet>? (Or even attaching the respective AttributeSets to
the various pieces (parameters/return type etc) - though I expect that
might not work as well/easily)

> An AttributeSet is also a uniqued and immutable object. You create an
> AttributeSet through the "AttributeSet::get" methods. You can add and
remove
> attributes, which result in the creation of a new AttributeSet.
>
> An AttributeSet object is designed to be passed around by value.
>
> Note: It is advised that you do *not* use the AttributeSet "Introspection"
> methods (e.g. 'Raw', 'getRawPointer', etc.). These methods break
encapsulation,
> and may be removed in a future release (i.e. 4.0).
>
> Lastly, we have a 'builder' class to help create the AttributeSet object
without
> having to create several different intermediate uniqued AttributeSet
> objects. The AttrBuilder class allows you to add and remove attributes at
> will. The attributes won't be uniqued until you call the appropriate
> "AttributeSet::get" method.
>
> An AttrBuilder object is *not* designed to be passed around by value. It
should
> be passed by reference.
>
> Note: It is advised that you do *not* use the "AttrBuilder::addRawValue()"
> method or the "AttrBuilder(uint64_t Val)" c'tor. These are for backwards
> compatibility and may be removed in a future release (i.e. 4.0).
>
> And that's basically it! A lot of functionality is hidden behind these
classes,
> but the interfaces are pretty straight forward. Please let me know if you
have
> any questions about how to use these classes, or if there is any
functionality
> you feel is missing.
>
> -bw
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130209/ce6fdd7c/attachment.html>


More information about the llvm-dev mailing list