[llvm-dev] How to interpret Selection DAG error output

[Jonathan Roelofs via llvm-dev]

On 2/18/16 12:32 PM, Rail Shafigulin wrote:
>
>
> On Thu, Feb 18, 2016 at 11:22 AM, Jonathan Roelofs
> <jonathan at codesourcery.com <mailto:jonathan at codesourcery.com>> wrote:
>
>
>
>     On 2/18/16 12:19 PM, Rail Shafigulin via llvm-dev wrote:
>
>
>
>         On Thu, Feb 18, 2016 at 10:50 AM, Krzysztof Parzyszek via llvm-dev
>         <llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>
>         <mailto:llvm-dev at lists.llvm.org
>         <mailto:llvm-dev at lists.llvm.org>>> wrote:
>
>              On 2/18/2016 12:43 PM, Rail Shafigulin via llvm-dev wrote:
>
>
>                  LLVM ERROR: Cannot select: 0x3284268: glue =
>         EsenciaISD::SET_FLAG
>                  0x3283608, 0x3283710, 0x3283e48 [ORD=3] [ID=11]
>                      0x3283608: i32,ch = CopyFromReg 0x3257980,
>         0x3283500 [ORD=1]
>                  [ID=9]
>                        0x3283500: i32 = Register %vreg5 [ID=1]
>                      0x3283710: i32 = Constant<3> [ID=2]
>                      0x3283e48: i32 = Constant<20> [ID=8]
>                  In function: fib
>
>                  As far as I can understand LLVM cannot select it
>         because the
>                  pattern is
>                  not specified. Unfortunately for me, I can't understand
>         how to
>                  interpret
>                  this output. Would anybody be able to help me out?
>
>
>
>              The DAG node that has failed selection is "SET_FLAG %vreg5,
>         3, 20",
>              the type of the node is "glue" and all 3 arguments are i32.
>
>         I understand where SET_FLAG, 3 and 20 come from, but how did you
>         arrive
>         the conclusion that %vreg5 should be part of the pattern?
>
>
>     The hex numbers are pointers. You can see the Use of 0x3283500 in
>     the CopyFromReg inst, and the def is here:
>
>     0x3283500: i32 = Register %vreg5 [ID=1]
>
>
> I think this is where I'm loosing the "thread". Based on what I'm seeing
> SET_FLAG has three operands, the first of which is a CopyFromReg.

Yes.

> So how come the pattern is SET_FLAG %vreg5, 3, 20 and not SET_FLAG
> CopyFromReg, 3, 20?

Because it's not... the latter is a better way of understanding it. The 
former is incorrect.

> In other words how do we go from CopyFromReg to %vreg5?

Written imperatively, the DAG from the dump you pasted is:

  something = dunno
  reg = i32 Register %vreg5 [ID=1]
  cpy = i32,ch CopyFromReg something, reg [ORD=1]
  c1 = i32 Constant<3> [ID=2]
  c2 = i32 Constant<20> [ID=8]
  foo = EsenciaISD::SET_FLAG cpy, c1, c2 [ORD=3] [ID=11]

where:

  something is 0x3257980
  reg is 0x3283500
  c1 is 0x3283710
  c2 is 0x3283e48
  foo is 0x3283608

These are just names I made up... the names don't actually exist as I've 
written them, but maybe that'll help your understanding?

This imperative order doesn't really exist either. Because it's a DAG, 
and the DAG isn't necessarily flattened like that... could be flattened 
in many different ways.


Jon

>
>
>     It's /just/ a dag.
>
>
>     Jon
>
>
>
>              The top line shows the node that caused the failure.  The hex
>              numbers on that line are the addresses of the arguments.  These
>              arguments are then listed below.
>
>              Newer versions of LLVM show node ids instead of addresses,
>         but AFAIK
>              it was dependent on some configuration option.
>
>              -Krzysztof
>
>
>              --
>              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>
>         <mailto:llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>>
>         http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
>
>
>         --
>         Rail Shafigulin
>         Software Engineer
>         Esencia Technologies
>
>
>         _______________________________________________
>         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
>
>
>     --
>     Jon Roelofs
>     jonathan at codesourcery.com <mailto:jonathan at codesourcery.com>
>     CodeSourcery / Mentor Embedded
>
>
>
>
> --
> Rail Shafigulin
> Software Engineer
> Esencia Technologies

-- 
Jon Roelofs
jonathan at codesourcery.com
CodeSourcery / Mentor Embedded