llvm dev - Aug 2015 - [LLVMdev] [LLVMDev] How to use memcpy with getOrInsertFunction()

A)
In LLVM, why are memcpy (and other memory related functions) treated
specially? i.e why do they have this intrinsic notion.
The IR generated is like the following, while the normal functions differ.

call void @llvm.memcpy.p0i8.p0i8.i64(....)
call void @my_function(...)

Can anyone explain me what the p0* mean in the above IR of memcpy?

B)
How to insert a call to "memcpy(A, B, 252);" before a particular basic
block. Assume A and B are integer64 array of 252 size each.
Can anyone help me figure out the parameter that needs to be placed in
'??' below

    my_memcpy = M.getOrInsertFunction("memcpy",
                                    FunctionType::getInt8PtrTy(Context),
                                    Type::getInt8PtrTy(Context),
                                    Type::getInt8PtrTy(Context),
                                    Type::getInt64Ty(Context),
                                    NULL);


        Value *my_params[] = {
            builder.CreateIntToPtr(builder.getInt64( ?? ),
Type::getInt8PtrTy(Context)),
            builder.CreateIntToPtr(builder.getInt64( ??),
Type::getInt8PtrTy(Context)),
            ConstantInt::get(Context, APInt(64, 252)),
        };
        Function *func_ptr= cast<Function>(my_memcpy);
        builder.CreateCall(func_ptr, my_params);


Hoping for a reply,
Prem Anand

Hi,

I have a partial answer for you, see below. Hopefully someone else
will answer the other part of your question.

2015-08-04 4:34 GMT+02:00 Prem Anand <prem.iamme at
gmail.com>:
>
> A)
> In LLVM, why are memcpy (and other memory related functions) treated
> specially? i.e why do they have this intrinsic notion.
The intrinsics capture information about alignment which can be used
to produce better code. See [1] and [2] for some more information.
> The IR generated is like the following, while the normal functions differ.
>
> call void @llvm.memcpy.p0i8.p0i8.i64(....)
> call void @my_function(...)
>
> Can anyone explain me what the p0* mean in the above IR of memcpy?
The 'p' means that the argument is a pointer type and '0' is the
address space of the pointer type. If you are interested, the name is
created by "Intrinsic::getName" in "lib/IR/Function.cpp".

/ David
>
> B)
> How to insert a call to "memcpy(A, B, 252);" before a particular
basic
> block. Assume A and B are integer64 array of 252 size each.
> Can anyone help me figure out the parameter that needs to be placed in
> '??' below
>
>     my_memcpy = M.getOrInsertFunction("memcpy",
>                                     FunctionType::getInt8PtrTy(Context),
>                                     Type::getInt8PtrTy(Context),
>                                     Type::getInt8PtrTy(Context),
>                                     Type::getInt64Ty(Context),
>                                     NULL);
>
>
>         Value *my_params[] = {
>             builder.CreateIntToPtr(builder.getInt64( ?? ),
> Type::getInt8PtrTy(Context)),
>             builder.CreateIntToPtr(builder.getInt64( ??),
> Type::getInt8PtrTy(Context)),
>             ConstantInt::get(Context, APInt(64, 252)),
>         };
>         Function *func_ptr= cast<Function>(my_memcpy);
>         builder.CreateCall(func_ptr, my_params);
>
>
> Hoping for a reply,
> Prem Anand
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
[1]
http://article.gmane.org/gmane.comp.compilers.llvm.devel/12102/match=memcpy+intrinsic
[2]
http://stackoverflow.com/questions/25302796/why-are-there-some-intrinsics-in-llvm-language