llvm dev - Feb 2019 - Intrinsics for RISCV CSR instructions

I notice that no intrinsics have been defined for the CSRRW/CSRRS/CSRRC
instructions.

It would be convenient to have intrinsics for these to allow CSR
manipulation directly from IR code.

Interestingly, this seems to be true for PowerPC (no intrinsics for
mfdcr/mtdcr) and X86 (no in/out) as well.

Are there plans to define standard RISCV intrinsics for this?
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190213/d8f473a9/attachment.html>

On Wed, 13 Feb 2019, 21:17 David Jones via llvm-dev <llvm-dev at
lists.llvm.org
wrote:
> I notice that no intrinsics have been defined for the CSRRW/CSRRS/CSRRC
> instructions.
>
> It would be convenient to have intrinsics for these to allow CSR
> manipulation directly from IR code.
>
> Interestingly, this seems to be true for PowerPC (no intrinsics for
> mfdcr/mtdcr) and X86 (no in/out) as well.
>
> Are there plans to define standard RISCV intrinsics for this?
>
Its not something I'd considered but I'd be happy to review a proposal
or
patches if there's precedent in LLVM with other backends l a good use case
justification. Why would intrinsics be preferred to just using inline asm?

Best,

Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190213/558e60aa/attachment.html>

I would like to generate LLVM IR from my custom compiler that will lower to
a CSR access instruction.

Can I emit arbitrary inline asm from LLVM IR without using clang?


On Wed, Feb 13, 2019 at 4:45 PM Alex Bradbury <asb at lowrisc.org> wrote:
>
>
> On Wed, 13 Feb 2019, 21:17 David Jones via llvm-dev <
> llvm-dev at lists.llvm.org wrote:
>
>> I notice that no intrinsics have been defined for the CSRRW/CSRRS/CSRRC
>> instructions.
>>
>> It would be convenient to have intrinsics for these to allow CSR
>> manipulation directly from IR code.
>>
>> Interestingly, this seems to be true for PowerPC (no intrinsics for
>> mfdcr/mtdcr) and X86 (no in/out) as well.
>>
>> Are there plans to define standard RISCV intrinsics for this?
>>
>
> Its not something I'd considered but I'd be happy to review a
proposal or
> patches if there's precedent in LLVM with other backends l a good use
case
> justification. Why would intrinsics be preferred to just using inline asm?
>
> Best,
>
> Alex
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190213/d72e5c0f/attachment.html>