[llvm-dev] Updating RegMask inline
Mehdi Amini via llvm-dev
llvm-dev at lists.llvm.org
Fri May 27 18:58:38 PDT 2016
Sent from my iPhone
> On May 27, 2016, at 6:55 PM, vivek pandya <vivekvpandya at gmail.com> wrote:
>
>
>
>> On Sat, May 28, 2016 at 12:23 AM, vivek pandya <vivekvpandya at gmail.com> wrote:
>>
>>
>>> On Sat, May 28, 2016 at 12:21 AM, Mehdi Amini <mehdi.amini at apple.com> wrote:
>>>
>>> > On May 27, 2016, at 11:49 AM, vivek pandya <vivekvpandya at gmail.com> wrote:
>>> >
>>> > Hello Mentors,
>>> >
>>> > I have completed writing simple register mask calculator pass, an immutable pass that stores RegMasks and provides API to query them, and a target specific pass for X86 target which currently iterates through all MI in given MF and if a call instruction is found then it quires for RegMask to Immutable pass and prints which registers are clobbered and preserved as per actual register allocation for callee function.
>>> >
>>> > So now I need to update RegMask for call instruction, so as per our previous discussion on IRC if we add a method to update RegMask, Do we have to do it in target specific way? Is it possible to have RegMask at fixed position in Operands array in MI? Also provide some more details on updating cached RegMask in various other class.
>>>
>>> Add a method "setOperandMask()" in class MachineOperand, and then when you find a call you just walk all the operands till you find one which "isRegMask()" at which point you can update it.
>>
>> Ok.
>
> Is it ok to add "setRegMaskOperand()" in class MI and looping through operands to find RegMask is also included in method ?
With this behavior I'd expect the method to be named updateRegMask...
Also, what is the semantic if there is no regmask?
--
Mehdi
> -Vivek
>
>
>> -Vivek
>>> --
>>> Mehdi
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20160527/b9470fa2/attachment.html>
More information about the llvm-dev
mailing list