Hey Ali,
On Fri, May 17, 2013 at 3:46 PM, Ali Javadi <aj14889 at yahoo.com> wrote:
> Hi,
>
> I have some newly defined intrinsic functions in my llvm IR code, which I
> want to translate to X86 instruction set. As a first step, I want to be
> able to generate "nop" for these instructions, so the program at
least
> compiles successfully.
>
> The call to my intrinsic function looks like this in the IR:
> call void @llvm.X(i16 %43)
>
> From what I understand it may be possible to pattern match this to the nop
> instructions in the backend.
I would first attempt this in LowerINTRINSIC_WO_CHAIN(...)
from lib/Target/X86/X86ISelLowering.cpp. That may or may not be the right
way to go... just a thought.
> I did not find an example of intrinsic function translation in the X86
> backend. Could you explain how this can be done, or point me to some places
> to look?
>
As for matching intrinsics, check out lib/Target/X86/X86InstrSSE.td. For
example, here is a pattern that matches a 256 bit unaligned store intrinsic
to the AVX vmovups instruction.
def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
(VMOVUPSYmr addr:$dst, VR256:$src)>;
-Cameron
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130517/35275dfa/attachment.html>