[llvm-dev] how to prevent LLVM back-end from reordering instructions at instruction scheduling?
Ryan Taylor via llvm-dev
llvm-dev at lists.llvm.org
Wed Nov 16 05:46:39 PST 2016
You'd think isBarrier might work but this is a sync mechanism and doesn't
prevent instructions from crossing the barrier that aren't use dependent.
On Tue, Nov 15, 2016 at 11:12 PM, Madhur Amilkanthwar <madhur13490 at gmail.com
> wrote:
> Just a thought. What about putting a dummy barrier between the
> instructions during selectionDAG creation?
>
> On Nov 16, 2016 4:00 AM, "Ryan Taylor via llvm-dev" <
> llvm-dev at lists.llvm.org> wrote:
>
>> Setting the MI as isTerminator should have the same impact, yes? I'm not
>> sure of the other consequences of this though, if any, have to look into it.
>>
>> Thanks.
>>
>> -Ryan
>>
>> On Tue, Nov 15, 2016 at 5:18 PM, Krzysztof Parzyszek <
>> kparzysz at codeaurora.org> wrote:
>>
>>> You can override TargetInstrInfo::isSchedulingBoundary for that.
>>>
>>> -Krzysztof
>>>
>>> On 11/15/2016 4:13 PM, Ryan Taylor wrote:
>>>
>>>> I have the same issue, would it be easier and more useful to attach a
>>>> flag to the instruction to tell the scheduler not to move instructions
>>>> across this boundary?
>>>>
>>>> -Ryan
>>>>
>>>> On Tue, Nov 15, 2016 at 5:11 PM, Krzysztof Parzyszek via llvm-dev
>>>> <llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>> wrote:
>>>>
>>>> You can create a DAG mutation that adds artificial dependencies
>>>> between A, B and Z.
>>>>
>>>> -Krzysztof
>>>>
>>>>
>>>> On 11/15/2016 3:26 PM, Wei Ding via llvm-dev wrote:
>>>>
>>>> Hello,
>>>>
>>>> I have a LLVM backend question regarding how to prevent compiler
>>>> from
>>>> reordering instructions. For example, I have the following
>>>> instructions.
>>>> Z_instruction is the one which I want to insert.
>>>>
>>>>
>>>> // instruction order which I am looking for
>>>> /////////////////////////////////////////
>>>> A_instruction
>>>> B_instruction
>>>>
>>>> *Z_instruction*
>>>>
>>>> C_instruction
>>>> D_instruction
>>>> E_instruction
>>>> F_instruction
>>>>
>>>> *Z_instruction*
>>>>
>>>> G_instruction
>>>> F_instruction
>>>>
>>>>
>>>> ==================================================
>>>>
>>>> But I found compiler reordered those instructions like the
>>>> following.
>>>> How could I prevent compiler from moving *A_instruction* and
>>>> *B_instruction* after the first *Z_instruction*? Any
>>>> recommendations and
>>>> suggestion are greatly appreciated here!!!
>>>>
>>>> *Z_instruction*
>>>> *
>>>> *
>>>> A_instruction
>>>> B_instruction
>>>>
>>>> C_instruction
>>>> D_instruction
>>>> E_instruction
>>>> F_instruction
>>>>
>>>> *Z_instruction*
>>>>
>>>> G_instruction
>>>> F_instruction
>>>>
>>>> BTW, I have define input/output chain for : *C_instruction,
>>>> D_instruction, E_instruction, F_instruction and Z_instruction*
>>>> *
>>>> *
>>>> Thank you so much!
>>>>
>>>> --
>>>> Wei Ding
>>>>
>>>>
>>>> _______________________________________________
>>>> LLVM Developers mailing list
>>>> llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>
>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>> <http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev>
>>>>
>>>>
>>>> --
>>>> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
>>>> hosted by The Linux Foundation
>>>> _______________________________________________
>>>> LLVM Developers mailing list
>>>> llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>
>>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>> <http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev>
>>>>
>>>>
>>>>
>>> --
>>> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
>>> hosted by The Linux Foundation
>>>
>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> llvm-dev at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20161116/ec8b5f3b/attachment.html>
More information about the llvm-dev
mailing list