llvm dev - Dec 2008 - [LLVMdev] Intrinsics question

Is it necessary to modify the front-end (in this case, for C) to generate
new intrinsic calls for a particular target?  For instance, say I wanted to
allow programmers to call a function "foo" that does some
target-specific
magic.  I want "foo" to be handled as an intrinsic by the back-end,
but all
intrinsics needs the "llvm." prefix, and I can't declare a
function called
"llvm.foo" in C.  Can it be done some other way?  Or do I need to hack
it
with inline assembly or some sort of intermediary find/replace
transformation.

Marc
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20081205/3a025154/attachment.html>

On Dec 5, 2008, at 8:29 PM, Marc de Kruijf wrote:
> Is it necessary to modify the front-end (in this case, for C) to  
> generate new intrinsic calls for a particular target?  For instance,  
> say I wanted to allow programmers to call a function "foo" that
does
> some target-specific magic.  I want "foo" to be handled as an  
> intrinsic by the back-end, but all intrinsics needs the "llvm."  
> prefix, and I can't declare a function called "llvm.foo" in
C.  Can
> it be done some other way?  Or do I need to hack it with inline  
> assembly or some sort of intermediary find/replace transformation.
You can use the gnu asm renaming feature to do this:


void llvmmemset(char*, char val, char len, int align)  
__asm__("llvm.memset.i8");

void foo(short X) {
   llvmmemset(&X, 1, 2, 1);
}

Except that I see that llvm-gcc is being more strict about checking  
parameter attrs:

$ llvm-gcc t.c -S -o - -emit-llvm -O3
t.c: In function ‘foo’:
t.c:4: warning: passing argument 1 of ‘llvmmemset’ from incompatible  
pointer type
Intrinsic has wrong parameter attributes!
void (i8*, i8, i8, i32)* @llvm.memset.i8
Broken module found, compilation aborted!
t.c:3: internal compiler error: Abort trap

It would probably be easy enough to make this work: after applying the  
asm rename, llvm-gcc could check to see if it is being renamed to an  
intrinsic, if so, add attrs and check the types as appropriate.

-Chris
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20081205/7aa878a1/attachment.html>