llvm dev - Jun 2017 - Clarification: machine operand subreg, <def, read-undef>, and IMPLICIT_DEF

Hey,

Suppose I have this MachineInstr:
    %vreg190:sub_32_1<def,read-undef> = IMPLICIT_DEF; PairRegs:%vreg190

(on my backend, the 'PairRegs' register class hold 2-vectors of i32
values)

I want to make sure I understand the combination of subreg and read-undef
correctly.  According to the comment in MachineOperand.h, quote:
  /// ...  On a sub-register def operand, it refers to the part of the
  /// register that isn't written. ...

So, am I correct to say that the above instruction is equivalent to the
following?
    %vreg190<def> = IMPLICIT_DEF; PairRegs:%vreg190

Or if there is a subtle distinction, please help me understand it. 

Thank you,
Nick Johnson
D. E. Shaw Research

I believe the short answer is yes, they are pretty much equivalent, but 
only because the value being written is IMPLICIT_DEF.

The first instruction results in:
  %vreg190:sub_32_1 -> i32 undef (from IMPLICIT_DEF)
  %vreg190:sub_32_0? -> i32 undef (from read-undef)

The second instruction results in:
   %vreg190 -> i32 2-vector undef (from IMPLICIT_DEF)

Assuming the i32 2-vector can hold any two i32 values, they are 
equivalent, other than the fact that they will likely generate 
instructions that write different widths.

On 6/7/2017 10:12 AM, Johnson, Nicholas Paul via llvm-dev
wrote:
> Hey,
>
> Suppose I have this MachineInstr:
>      %vreg190:sub_32_1<def,read-undef> = IMPLICIT_DEF;
PairRegs:%vreg190
>
> (on my backend, the 'PairRegs' register class hold 2-vectors of i32
values)
>
> I want to make sure I understand the combination of subreg and read-undef
correctly.  According to the comment in MachineOperand.h, quote:
>    /// ...  On a sub-register def operand, it refers to the part of the
>    /// register that isn't written. ...
>
> So, am I correct to say that the above instruction is equivalent to the
following?
>      %vreg190<def> = IMPLICIT_DEF; PairRegs:%vreg190
>
> Or if there is a subtle distinction, please help me understand it.
>
> Thank you,
> Nick Johnson
> D. E. Shaw Research
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-- 
Geoff Berry
Employee of Qualcomm Datacenter Technologies, Inc.
  Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm
Technologies, Inc.  Qualcomm Technologies, Inc. is a member of the Code Aurora
Forum, a Linux Foundation Collaborative Project.