thr3ads.net - llvm dev - [llvm-dev] [SelectionDAG] Assertion due to MachineMemOperand flags difference. [Oct 2017]

If this information is useful, please help other people find it:
Share via:

Sean Fertile via llvm-dev

2017-Oct-13 23:05 UTC

[llvm-dev] [SelectionDAG] Assertion due to MachineMemOperand flags difference.

Hello,

I've hit an assertion in SelectionDAG where we try to merge 2 loads
that have the same operands but their MMO flags differ. One is
dereferenceable and one is not. I'm not sure what the underlying issue
here is:

1) MDSDNode with the same operands should have the same flags set on
their respective MMO. The fact the flags differ when the
opcode,types,operands and address-space are the same is the problem.

2) Having MDSDnodes with the same operands but different MMO flags is
possible, so the Flags should be added to the FoldingSetNodeID.

3) Something else I haven't considered.

I have a patch posted implementing 2, but don't know if I should look
at fixing 1 as well (or perhaps instead). The loads that trigger the
assertion are:

t47: v4i32,ch = load<LD16[%0+80](align=8)(dereferenceable)> t20, t46,
undef:i64
t69: v4i32,ch = load<LD16[FixedStack1+80](align=8)> t50, t46, undef:i64

I would expect the the second load should also be marked
dereferenceable since its loading from one of the TargetFrames. Am I
on the right track here?

Thanks
Sean
-------------- next part --------------
Initial selection DAG: BB#0 '_Z3fn2v:entry'
SelectionDAG has 122 nodes:
  t4: i64 = GlobalAddress<void (%class.F*)* @_Z10EmitLValuev> 0
  t10: i64 = add Register:i64 %X1, Constant:i64<32>
        t0: ch = EntryToken
      t3: ch = lifetime.start t0, TargetFrameIndex:i64<1>
    t7: ch,glue = callseq_start t3, TargetConstant:i64<32>,
TargetConstant:i64<0>
  t12: ch,glue = CopyToReg t7, Register:i64 %X3, FrameIndex:i64<1>
  t16: ch,glue = PPCISD::CALL_NOP t12, TargetGlobalAddress:i64<void
(%class.F*)* @_Z10EmitLValuev> 0, Register:i64 %X3, Register:i64 %X2,
RegisterMask:Untyped, t12:1
    t17: ch,glue = callseq_end t16, TargetConstant:i64<32>,
TargetConstant:i64<0>, t16:1
  t20: ch = lifetime.start t17, TargetFrameIndex:i64<0>
  t21: i64 = Constant<96>
  t22: i64 = Constant<0>
  t24: v4i32,ch = load<LD16[%0](align=8)(dereferenceable)> t20,
FrameIndex:i64<1>, undef:i64
  t27: i64 = add FrameIndex:i64<1>, Constant:i64<16>
  t28: v4i32,ch = load<LD16[%0+16](align=8)(dereferenceable)> t20, t27,
undef:i64
  t31: i64 = add FrameIndex:i64<1>, Constant:i64<32>
  t32: v4i32,ch = load<LD16[%0+32](align=8)(dereferenceable)> t20, t31,
undef:i64
  t36: i64 = add FrameIndex:i64<1>, Constant:i64<48>
  t37: v4i32,ch = load<LD16[%0+48](align=8)(dereferenceable)> t20, t36,
undef:i64
  t41: i64 = add FrameIndex:i64<1>, Constant:i64<64>
  t42: v4i32,ch = load<LD16[%0+64](align=8)(dereferenceable)> t20, t41,
undef:i64
  t46: i64 = add FrameIndex:i64<1>, Constant:i64<80>
  t47: v4i32,ch = load<LD16[%0+80](align=8)(dereferenceable)> t20, t46,
undef:i64
    t25: ch = store<ST16[%1]> t20, t24, FrameIndex:i64<0>, undef:i64
      t29: i64 = add FrameIndex:i64<0>, Constant:i64<16>
    t30: ch = store<ST16[%1+16]> t20, t28, t29, undef:i64
      t33: i64 = add FrameIndex:i64<0>, Constant:i64<32>
    t34: ch = store<ST16[%1+32]> t20, t32, t33, undef:i64
      t38: i64 = add FrameIndex:i64<0>, Constant:i64<48>
    t39: ch = store<ST16[%1+48]> t20, t37, t38, undef:i64
      t43: i64 = add FrameIndex:i64<0>, Constant:i64<64>
    t44: ch = store<ST16[%1+64]> t20, t42, t43, undef:i64
      t48: i64 = add FrameIndex:i64<0>, Constant:i64<80>
    t49: ch = store<ST16[%1+80]> t20, t47, t48, undef:i64
  t50: ch = TokenFactor t24:1, t25, t28:1, t30, t32:1, t34, t37:1, t39, t42:1,
t44, t47:1, t49
  t51: i64 = GlobalAddress<void (%class.F*)* @_Z3fn11F> 0
  t53: ch,glue = callseq_start t50, TargetConstant:i64<128>,
TargetConstant:i64<0>
  t54: i32 = Constant<96>
  t55: v4i32,ch = load<LD16[FixedStack1](align=8)> t50,
FrameIndex:i64<1>, undef:i64
  t57: v4i32,ch = load<LD16[FixedStack1+16](align=8)> t50, t27, undef:i64
  t60: v4i32,ch = load<LD16[FixedStack1+32](align=8)> t50, t31, undef:i64
  t63: v4i32,ch = load<LD16[FixedStack1+48](align=8)> t50, t36, undef:i64
  t66: v4i32,ch = load<LD16[FixedStack1+64](align=8)> t50, t41, undef:i64
  t69: v4i32,ch = load<LD16[FixedStack1+80](align=8)> t50, t46, undef:i64
      t56: ch = store<ST16[<unknown>](align=8)> t50, t55, t10,
undef:i64
        t58: i64 = add t10, Constant:i64<16>
      t59: ch = store<ST16[<unknown>](align=8)> t50, t57, t58,
undef:i64
        t61: i64 = add t10, Constant:i64<32>
      t62: ch = store<ST16[<unknown>](align=8)> t50, t60, t61,
undef:i64
        t64: i64 = add t10, Constant:i64<48>
      t65: ch = store<ST16[<unknown>](align=8)> t50, t63, t64,
undef:i64
        t67: i64 = add t10, Constant:i64<64>
      t68: ch = store<ST16[<unknown>](align=8)> t50, t66, t67,
undef:i64
        t70: i64 = add t10, Constant:i64<80>
      t71: ch = store<ST16[<unknown>](align=8)> t50, t69, t70,
undef:i64
    t72: ch = TokenFactor t55:1, t56, t57:1, t59, t60:1, t62, t63:1, t65, t66:1,
t68, t69:1, t71
  t73: ch,glue = callseq_start t72, TargetConstant:i64<128>,
TargetConstant:i64<0>
  t74: i64,ch = load<LD8[FixedStack1]> t73, FrameIndex:i64<1>,
undef:i64
    t76: i64 = add FrameIndex:i64<1>, Constant:i64<8>
  t77: i64,ch = load<LD8[FixedStack1+8]> t73, t76, undef:i64
  t78: i64,ch = load<LD8[FixedStack1+16]> t73, t27, undef:i64
    t80: i64 = add FrameIndex:i64<1>, Constant:i64<24>
  t81: i64,ch = load<LD8[FixedStack1+24]> t73, t80, undef:i64
  t82: i64,ch = load<LD8[FixedStack1+32]> t73, t31, undef:i64
    t84: i64 = add FrameIndex:i64<1>, Constant:i64<40>
  t85: i64,ch = load<LD8[FixedStack1+40]> t73, t84, undef:i64
  t86: i64,ch = load<LD8[FixedStack1+48]> t73, t36, undef:i64
    t88: i64 = add FrameIndex:i64<1>, Constant:i64<56>
  t89: i64,ch = load<LD8[FixedStack1+56]> t73, t88, undef:i64
    t90: ch = TokenFactor t74:1, t77:1, t78:1, t81:1, t82:1, t85:1, t86:1, t89:1
  t91: ch,glue = CopyToReg t90, Register:i64 %X3, t74
  t93: ch,glue = CopyToReg t91, Register:i64 %X4, t77, t91:1
  t95: ch,glue = CopyToReg t93, Register:i64 %X5, t78, t93:1
  t97: ch,glue = CopyToReg t95, Register:i64 %X6, t81, t95:1
  t99: ch,glue = CopyToReg t97, Register:i64 %X7, t82, t97:1
  t101: ch,glue = CopyToReg t99, Register:i64 %X8, t85, t99:1
  t103: ch,glue = CopyToReg t101, Register:i64 %X9, t86, t101:1
  t105: ch,glue = CopyToReg t103, Register:i64 %X10, t89, t103:1
  t107: ch,glue = PPCISD::CALL_NOP t105, TargetGlobalAddress:i64<void
(%class.F*)* @_Z3fn11F> 0, Register:i64 %X3, Register:i64 %X4, Register:i64
%X5, Register:i64 %X6, Register:i64 %X7, Register:i64 %X8, Register:i64 %X9,
Register:i64 %X10, Register:i64 %X2, RegisterMask:Untyped, t105:1
  t109: i64 = GlobalAddress<i32 (%class.F*)* @_ZN1F11isGlobalRegEv> 0
      t108: ch,glue = callseq_end t107, TargetConstant:i64<128>,
TargetConstant:i64<0>, t107:1
    t110: ch,glue = callseq_start t108, TargetConstant:i64<32>,
TargetConstant:i64<0>
  t111: ch,glue = CopyToReg t110, Register:i64 %X3, FrameIndex:i64<0>
  t113: ch,glue = PPCISD::CALL_NOP t111, TargetGlobalAddress:i64<i32
(%class.F*)* @_ZN1F11isGlobalRegEv> 0, Register:i64 %X3, Register:i64 %X2,
RegisterMask:Untyped, t111:1
  t114: ch,glue = callseq_end t113, TargetConstant:i64<32>,
TargetConstant:i64<0>, t113:1
  t115: i64,ch,glue = CopyFromReg t114, Register:i64 %X3, t114:1
    t117: i64 = AssertSext t115, ValueType:ch:i32
  t118: i32 = truncate t117
      t119: ch = lifetime.end t115:1, TargetFrameIndex:i64<0>
    t120: ch = lifetime.end t119, TargetFrameIndex:i64<1>
  t121: ch = PPCISD::RET_FLAG t120
-------------- next part --------------
A non-text attachment was scrubbed...
Name: MMO_Flag_Mismatch.ll
Type: application/octet-stream
Size: 1726 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20171013/a804467a/attachment.obj>

Hal Finkel via llvm-dev

2017-Oct-26 00:28 UTC

head link

[llvm-dev] [SelectionDAG] Assertion due to MachineMemOperand flags difference.

On 10/13/2017 06:05 PM, Sean Fertile via llvm-dev wrote:> Hello,
>
> I've hit an assertion in SelectionDAG where we try to merge 2 loads
> that have the same operands but their MMO flags differ. One is
> dereferenceable and one is not. I'm not sure what the underlying issue
> here is:
>
> 1) MDSDNode with the same operands should have the same flags set on
> their respective MMO. The fact the flags differ when the
> opcode,types,operands and address-space are the same is the problem.
>
> 2) Having MDSDnodes with the same operands but different MMO flags is
> possible, so the Flags should be added to the FoldingSetNodeID.
>
> 3) Something else I haven't considered.
>
> I have a patch posted implementing 2, but don't know if I should look
> at fixing 1 as well (or perhaps instead). The loads that trigger the
> assertion are:
>
> t47: v4i32,ch = load<LD16[%0+80](align=8)(dereferenceable)> t20, t46,
undef:i64
> t69: v4i32,ch = load<LD16[FixedStack1+80](align=8)> t50, t46,
undef:i64
>
> I would expect the the second load should also be marked
> dereferenceable since its loading from one of the TargetFrames. Am I
> on the right track here?
Hi, Sean,

To follow-up on this, first I'll note that, IIRC, we've fixed a couple 
of related problems in the recent past. Specifically, the following come 
to mind: r309930 and r306404. In this case, it seems like something is 
dropping this information somewhere. Or, perhaps, I don't see why we 
have a "FixedStack" location that's not marked as dereferenceable.
That
having been said...

As for whether "having MDSDnodes with the same operands but different 
MMO flags is possible", the answer is yes. Of the bits that are 
currently defined, MOLoad and MOStore are attached to fundamental 
properties of the corresponding SDNode. Of the remainder: MOVolatile, 
MONonTemporal, MODereferenceable, and MOInvariant, some of these can be 
set independently on a per-operation basis. For example, one can 
certainly have a two volatile loads of the same address, one which is 
non-temporal and one which is not. However, those bits are already added 
to the FoldingSetNodeID, because they're mirrored by the MemSDNode's 
MemSDNodeBits (which has bits for IsVolatile, IsNonTemporal, 
IsDereferenceable, and IsInvariant), and these should be returned by 
getRawSubclassData, which is called in all of the places that D38898 
modifies). Thus, in short, I don't understand why the IsDereferenceable 
bit is not already included in the FoldingSetNodeID.

  -Hal
>
> Thanks
> Sean
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-- 
Hal Finkel
Lead, Compiler Technology and Programming Languages
Leadership Computing Facility
Argonne National Laboratory

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20171025/a7b43730/attachment-0001.html>

Sean Fertile via llvm-dev

2017-Oct-26 21:19 UTC

head link

[llvm-dev] [SelectionDAG] Assertion due to MachineMemOperand flags difference.

Hi Hal,

Thanks for the reply, that clears up a fair bit for me. I think the
reason we have the 'FixedStack'  location without it being marked as
dereferenceable is because its created in CreateCopyOfByValArgument
where we use a default MachinePointerInfo in the call to getMemcpy. As
for fixing the assertion I think I understand now why we look up a
node with different MMO flags, its a 'MemIntrinsicSDNode' which
doesn't add its subclass data to the FoldingSetNode. I'll update the
patch on D38898 accordingly.

Thanks
Sean


On Wed, Oct 25, 2017 at 8:28 PM, Hal Finkel <hfinkel at anl.gov>
wrote:> On 10/13/2017 06:05 PM, Sean Fertile via llvm-dev wrote:
>
> Hello,
>
> I've hit an assertion in SelectionDAG where we try to merge 2 loads
> that have the same operands but their MMO flags differ. One is
> dereferenceable and one is not. I'm not sure what the underlying issue
> here is:
>
> 1) MDSDNode with the same operands should have the same flags set on
> their respective MMO. The fact the flags differ when the
> opcode,types,operands and address-space are the same is the problem.
>
> 2) Having MDSDnodes with the same operands but different MMO flags is
> possible, so the Flags should be added to the FoldingSetNodeID.
>
> 3) Something else I haven't considered.
>
> I have a patch posted implementing 2, but don't know if I should look
> at fixing 1 as well (or perhaps instead). The loads that trigger the
> assertion are:
>
> t47: v4i32,ch = load<LD16[%0+80](align=8)(dereferenceable)> t20, t46,
> undef:i64
> t69: v4i32,ch = load<LD16[FixedStack1+80](align=8)> t50, t46,
undef:i64
>
> I would expect the the second load should also be marked
> dereferenceable since its loading from one of the TargetFrames. Am I
> on the right track here?
>
>
> Hi, Sean,
>
> To follow-up on this, first I'll note that, IIRC, we've fixed a
couple of
> related problems in the recent past. Specifically, the following come to
> mind: r309930 and r306404. In this case, it seems like something is
dropping
> this information somewhere. Or, perhaps, I don't see why we have a
> "FixedStack" location that's not marked as dereferenceable.
That having been
> said...
>
> As for whether "having MDSDnodes with the same operands but different
MMO
> flags is possible", the answer is yes. Of the bits that are currently
> defined, MOLoad and MOStore are attached to fundamental properties of the
> corresponding SDNode. Of the remainder: MOVolatile, MONonTemporal,
> MODereferenceable, and MOInvariant, some of these can be set independently
> on a per-operation basis. For example, one can certainly have a two
volatile
> loads of the same address, one which is non-temporal and one which is not.
> However, those bits are already added to the FoldingSetNodeID, because
> they're mirrored by the MemSDNode's MemSDNodeBits (which has bits
for
> IsVolatile, IsNonTemporal, IsDereferenceable, and IsInvariant), and these
> should be returned by getRawSubclassData, which is called in all of the
> places that D38898 modifies). Thus, in short, I don't understand why
the
> IsDereferenceable bit is not already included in the FoldingSetNodeID.
>
>  -Hal
>
>
> Thanks
> Sean
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
> --
> Hal Finkel
> Lead, Compiler Technology and Programming Languages
> Leadership Computing Facility
> Argonne National Laboratory

Seemingly Similar Threads

Search for more apparently analagous threads

llvm dev - Oct 2017 - [SelectionDAG] Assertion due to MachineMemOperand flags difference.

[llvm-dev] [SelectionDAG] Assertion due to MachineMemOperand flags difference.

[llvm-dev] [SelectionDAG] Assertion due to MachineMemOperand flags difference.

[llvm-dev] [SelectionDAG] Assertion due to MachineMemOperand flags difference.

Seemingly Similar Threads