Function attributes for memory side-effects

Hi All

I am writing a pass which requires checking dependences, and am having trouble dealing with function calls. Specifically, I want to be able to know when a called function does not have any side effects (e.g. math library functions like sqrt), and was wondering if there are attributes that specify this behavior (I know there is the ‘noread’ attribute but wasn’t sure if there’s something similar for writes)? Also, how can I tell clang to generate those attributes at the function declaration? Any information would be helpful.

Thanks

-Adel Ejjeh

Specifically, I want to be able to know when a called function does not have any side effects (e.g. math library functions like sqrt)

Apologies for the pedantry, but I believe sqrt may set errno, so it actually can have side effects. :frowning: See -fno-math-errno and the documentation around it.

, and was wondering if there are attributes that specify this behavior (I know there is the ‘noread’ attribute but wasn’t sure if there’s something similar for writes)? Also, how can I tell clang to generate those attributes at the function declaration? Any information would be helpful.

Yep, I believe the IR attributes are readonly and readnone. Readonly implies no side effects (no writes), and readnone implies no memory dependencies at all. The return value is a pure function of the arguments. Two calls with the same arguments can be folded together.

There are a couple of passes (Attributor, FunctionAttrs, I’m not up-to-date) that will infer these attributes if they can see a precise definition of the function body. Learning anything interesting usually requires expanding the scope of analysis with LTO, so these passes can look across translation unit boundaries.

HTH

>> Specifically, I want to be able to know when a called function does not
>> have any side effects (e.g. math library functions like sqrt)
>>
>
> Apologies for the pedantry, but I believe sqrt may set errno, so it
> actually can have side effects. :frowning: See -fno-math-errno and the
> documentation around it.
>
>> , and was wondering if there are attributes that specify this behavior (I
>> know there is the ‘noread’ attribute but wasn’t sure if there’s something
>> similar for writes)? Also, how can I tell clang to generate those
>> attributes at the function declaration? Any information would be helpful.
>>
>
> Yep, I believe the IR attributes are `readonly` and `readnone`. Readonly
> implies no side effects (no writes), and readnone implies no memory
> dependencies at all. The return value is a pure function of the arguments.
> Two calls with the same arguments can be folded together.
>
> There are a couple of passes (Attributor, FunctionAttrs, I'm not
> up-to-date) that will infer these attributes if they can see a precise
> definition of the function body. Learning anything interesting usually
> requires expanding the scope of analysis with LTO, so these passes can look
> across translation unit boundaries.

Often true. In case of library functions we actually "know" the side
effects and will add the appropriate attributes. As you said, fast math
flags are needed for math library functions that may otherwise write
errno.

The full list of attributes we have so far is:

access locations: \`readnone\`, \`inaccessiblememonly\`, \`argmemonly\`, and \`inaccessiblemem\_or\_argmemonly\`

and access "kinds": `readonly` and `writeonly`

Except for `readnone` you can combine a location attribute with a "kind"
or have one of either alone. The Attributor does internally derive more
"locations", basically any combination of:
local memory
constant memory
internal global memory
external global memory
argument memory
inaccessible memory
malloced memory (returned by a function with the `noalias` return attribute)
unknown memory
I want to add some/all of these as attributes but didn't find the time
yet.

Cheers,

Johannes

Hi Johannes,

This got me puzzled:

In case of library functions we actually “know” the side effects and will add the appropriate attributes.

How can we make sure that a function in the LLVM IR is a specific library function and not something that the user wrote themself? For example, we see a call to “printf” in the code but it could be a function that just happened to be named “printf” that does all sorts of things that we wouldn’t expect from a genuine libc printf.

Alexey

Frontends are supposed to add a "nobuiltin" attribute to functions
that do not correspond to the semantics of the C library function with
the same name.

Michael

What Michael said. You can also use

-fno-builtin-<value> Disable implicit builtin knowledge of a specific function

-fno-builtin Disable implicit builtin knowledge of functions

Hello Neil

Thanks for your reply. After experimenting in the Compiler Explorer, it seems that the “readnone” attribute is assigned for functions that neither read from nor write to memory (otherwise “readonly” or “writeonly” are used accordingly when the function does one but not the other). However, one of my problems still exists which is using library function calls (like math library calls – e.g. sqrt()). When the declaration for such functions is being generated in the ll file, it does not include any attributes. Is there a way to handle such scenario?

Thanks

-Adel

Johannes, Reid

Thanks for this input. I somehow missed this entire chain earlier and only just saw your replies! I tried with the fast-math flag and it indeed did produce the readnone attribute for the sqrt function (the attribute was not generated without fast-math).

Regards
-Adel

     >
     >> Specifically, I want to be able to know when a called function does not
     >> have any side effects (e.g. math library functions like sqrt)
     >>
     >
     > Apologies for the pedantry, but I believe sqrt may set errno, so it
     > actually can have side effects. :frowning: See -fno-math-errno and the
     > documentation around it.
     >
     >
     >> , and was wondering if there are attributes that specify this
    behavior (I
     >> know there is the ‘noread’ attribute but wasn’t sure if there’s
    something
     >> similar for writes)? Also, how can I tell clang to generate those
     >> attributes at the function declaration? Any information would be
    helpful.
     >>
     >
     > Yep, I believe the IR attributes are `readonly` and `readnone`. Readonly
     > implies no side effects (no writes), and readnone implies no memory
     > dependencies at all. The return value is a pure function of the
    arguments.
     > Two calls with the same arguments can be folded together.
     >
     > There are a couple of passes (Attributor, FunctionAttrs, I'm not
     > up-to-date) that will infer these attributes if they can see a precise
     > definition of the function body. Learning anything interesting usually
     > requires expanding the scope of analysis with LTO, so these passes
    can look
     > across translation unit boundaries.

    Often true. In case of library functions we actually "know" the side
    effects and will add the appropriate attributes. As you said, fast math
    flags are needed for math library functions that may otherwise write
    errno.

    The full list of attributes we have so far is:

        access locations: `readnone`, `inaccessiblememonly`, `argmemonly`,
    and `inaccessiblemem_or_argmemonly`
      and access "kinds": `readonly` and `writeonly`

    Except for `readnone` you can combine a location attribute with a "kind"
    or have one of either alone. The Attributor does internally derive more
    "locations", basically any combination of:
       local memory
       constant memory
       internal global memory
       external global memory
       argument memory
       inaccessible memory
       malloced memory (returned by a function with the `noalias` return
    attribute)
       unknown memory
    I want to add some/all of these as attributes but didn't find the time
    yet.

    Cheers,

       Johannes