Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

JIT: Assertion failed 'tree->IsCall() || tree->OperIs(GT_COMMA)' during 'Stress gtSplitTree' #83576

Closed
jakobbotsch opened this issue Mar 17, 2023 · 2 comments · Fixed by #83725
Closed

JIT: Assertion failed 'tree->IsCall() || tree->OperIs(GT_COMMA)' during 'Stress gtSplitTree' #83576

jakobbotsch opened this issue Mar 17, 2023 · 2 comments · Fixed by #83725
Assignees
@jakobbotsch
Labels
area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
Milestone
8.0.0

Comments

@jakobbotsch
Copy link
Member

jakobbotsch commented Mar 17, 2023 •
edited
Loading

Seen in https://dev.azure.com/dnceng-public/public/_build/results?buildId=207037&view=ms.vss-test-web.build-test-results-tab&runId=3846126&resultId=116663&paneView=dotnet-dnceng.dnceng-build-release-tasks.helix-test-information-tab.

Assert failure(PID 2810 [0x00000afa], Thread: 2810 [0x0afa]): Assertion failed 'tree->IsCall() || tree->OperIs(GT_COMMA)' in 'JIT.HardwareIntrinsics.General._Vector256.VectorWidenTest__WidenInt16:RunReflectionScenario_UnsafeRead():this' during 'Stress gtSplitTree' (IL size 301; hash 0xf9eeacfa; MinOpts)

    File: /__w/1/s/src/coreclr/jit/fgdiagnostic.cpp Line: 3166
    Image: /datadisks/disk1/work/A86B0945/p/corerun
@dotnet-issue-labeler dotnet-issue-labeler bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Mar 17, 2023
@ghost ghost added the untriaged New issue has not been triaged by the area owner label Mar 17, 2023
@ghost
Copy link

ghost commented Mar 17, 2023

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch, @kunalspathak
See info in area-owners.md if you want to be subscribed.

Issue Details

Seen in https://dev.azure.com/dnceng-public/public/_build/results?buildId=207037&view=ms.vss-test-web.build-test-results-tab&runId=3846126&resultId=116663&paneView=dotnet-dnceng.dnceng-build-release-tasks.helix-test-information-tab.

Author: jakobbotsch
Assignees: -
Labels:

area-CodeGen-coreclr, untriaged
Milestone: -

@jakobbotsch jakobbotsch self-assigned this Mar 17, 2023
@jakobbotsch jakobbotsch removed the untriaged New issue has not been triaged by the area owner label Mar 17, 2023
@jakobbotsch jakobbotsch added this to the 8.0.0 milestone Mar 17, 2023
@jakobbotsch
Copy link
Member Author

jakobbotsch commented Mar 21, 2023 •
edited
Loading

This will only happen in CI once we merge #79283 -- it is adding a stress mode that changes the stress decisions made.
The problem is that we have a tree that looks like:

N018 ( 38, 33) [000327] -ACXGO----- arg1 setup              ├──▌  COMMA     simd32
N011 ( 28, 25) [000328] -ACXGO-----                         │  ├──▌  COMMA     void  
N008 ( 24, 22) [000320] -ACXG---R--                         │  │  ├──▌  ASG       byref 
N007 (  3,  2) [000319] D------N---                         │  │  │  ├──▌  LCL_VAR   byref  V50 tmp45        
N006 ( 20, 19) [000126] --CXG------                         │  │  │  └──▌  CALL help byref  CORINFO_HELP_UNBOX
N004 (  3,  2) [000124] ----------- arg1 in rsi             │  │  │     ├──▌  LCL_VAR   ref    V02 loc1         
N005 (  3, 10) [000125] H---------- arg0 in rdi             │  │  │     └──▌  CNS_INT(h) long   0x7f6e4bb0e500 class
N010 (  4,  3) [000322] ---X-O-----                         │  │  └──▌  NULLCHECK byte  
N009 (  3,  2) [000321] -----------                         │  │     └──▌  LCL_VAR   byref  V50 tmp45        
N017 ( 10,  8) [000331] -A-XGO--R--                         │  └──▌  ASG       simd32 (copy)
N016 (  3,  2) [000330] D------N---                         │     ├──▌  LCL_VAR   simd32 V51 tmp46        
N015 (  6,  5) [000329] ---XGO-N---                         │     └──▌  IND       simd32
N014 (  4,  3) [000326] -----O-N---                         │        └──▌  ADD       byref 
N012 (  3,  2) [000324] -----O-----                         │           ├──▌  LCL_VAR   byref  V50 tmp45        
N013 (  1,  1) [000325] -----------                         │           └──▌  CNS_INT   long   32

I.e. simd32 typed comma but the "value" is a simd32 assignment. gtSplitTree ends up creating

N016 ( 14, 11) [000428] -A-XGO--R--                         └──▌  ASG       simd32 (copy)
N015 (  3,  2) [000427] D------N---                            ├──▌  LCL_VAR   simd32 V76 tmp71        
N014 ( 10,  8) [000331] -A-XGO-NR--                            └──▌  ASG       simd32 (copy)
N013 (  3,  2) [000330] D------N---                               ├──▌  LCL_VAR   simd32 V51 tmp46        
N012 (  6,  5) [000329] ---XGO-N---                               └──▌  IND       simd32
N011 (  4,  3) [000326] -----O-N---                                  └──▌  ADD       byref 
N009 (  3,  2) [000324] -----O-----                                     ├──▌  LCL_VAR   byref  V50 tmp45        
N010 (  1,  1) [000325] -----------                                     └──▌  CNS_INT   long   32

jakobbotsch added a commit to jakobbotsch/runtime that referenced this issue Mar 21, 2023
@jakobbotsch
JIT: Expand handling of non-value commas in gtSplitTree
efb75f4 
We can have a 'value'-typed COMMA whose RHS operand is actually an
assignment. gtSplitTree would create illegal IR for this case.

For example:

N018 ( 38, 33) [000327] -ACXGO----- arg1 setup              ├──▌  COMMA     simd32
N011 ( 28, 25) [000328] -ACXGO-----                         │  ├──▌  COMMA     void
N008 ( 24, 22) [000320] -ACXG---R--                         │  │  ├──▌  ASG       byref
N007 (  3,  2) [000319] D------N---                         │  │  │  ├──▌  LCL_VAR   byref  V50 tmp45
N006 ( 20, 19) [000126] --CXG------                         │  │  │  └──▌  CALL help byref  CORINFO_HELP_UNBOX
N004 (  3,  2) [000124] ----------- arg1 in rsi             │  │  │     ├──▌  LCL_VAR   ref    V02 loc1
N005 (  3, 10) [000125] H---------- arg0 in rdi             │  │  │     └──▌  CNS_INT(h) long   0x7f6e4bb0e500 class
N010 (  4,  3) [000322] ---X-O-----                         │  │  └──▌  NULLCHECK byte
N009 (  3,  2) [000321] -----------                         │  │     └──▌  LCL_VAR   byref  V50 tmp45
N017 ( 10,  8) [000331] -A-XGO--R--                         │  └──▌  ASG       simd32 (copy)
N016 (  3,  2) [000330] D------N---                         │     ├──▌  LCL_VAR   simd32 V51 tmp46
N015 (  6,  5) [000329] ---XGO-N---                         │     └──▌  IND       simd32
N014 (  4,  3) [000326] -----O-N---                         │        └──▌  ADD       byref
N012 (  3,  2) [000324] -----O-----                         │           ├──▌  LCL_VAR   byref  V50 tmp45
N013 (  1,  1) [000325] -----------                         │           └──▌  CNS_INT   long   32

gtSplitTree would believe that [000327] was a value and would create the
following IR shape:

N016 ( 14, 11) [000428] -A-XGO--R--                         ▌  ASG       simd32 (copy)
N015 (  3,  2) [000427] D------N---                         ├──▌  LCL_VAR   simd32 V76 tmp71
N014 ( 10,  8) [000331] -A-XGO-NR--                         └──▌  ASG       simd32 (copy)
N013 (  3,  2) [000330] D------N---                            ├──▌  LCL_VAR   simd32 V51 tmp46
N012 (  6,  5) [000329] ---XGO-N---                            └──▌  IND       simd32
N011 (  4,  3) [000326] -----O-N---                               └──▌  ADD       byref
N009 (  3,  2) [000324] -----O-----                                  ├──▌  LCL_VAR   byref  V50 tmp45
N010 (  1,  1) [000325] -----------                                  └──▌  CNS_INT   long   32

Fix dotnet#83576
@jakobbotsch jakobbotsch mentioned this issue Mar 21, 2023
Merged
@ghost ghost added the in-pr There is an active PR which will close this issue when it is merged label Mar 21, 2023
jakobbotsch added a commit that referenced this issue Mar 21, 2023
@jakobbotsch
JIT: Expand handling of non-value commas in gtSplitTree (#83725)
6439980 
We can have a 'value'-typed COMMA whose RHS operand is actually an
assignment. gtSplitTree would create illegal IR for this case.

For example:

N018 ( 38, 33) [000327] -ACXGO----- arg1 setup              ├──▌  COMMA     simd32
N011 ( 28, 25) [000328] -ACXGO-----                         │  ├──▌  COMMA     void
N008 ( 24, 22) [000320] -ACXG---R--                         │  │  ├──▌  ASG       byref
N007 (  3,  2) [000319] D------N---                         │  │  │  ├──▌  LCL_VAR   byref  V50 tmp45
N006 ( 20, 19) [000126] --CXG------                         │  │  │  └──▌  CALL help byref  CORINFO_HELP_UNBOX
N004 (  3,  2) [000124] ----------- arg1 in rsi             │  │  │     ├──▌  LCL_VAR   ref    V02 loc1
N005 (  3, 10) [000125] H---------- arg0 in rdi             │  │  │     └──▌  CNS_INT(h) long   0x7f6e4bb0e500 class
N010 (  4,  3) [000322] ---X-O-----                         │  │  └──▌  NULLCHECK byte
N009 (  3,  2) [000321] -----------                         │  │     └──▌  LCL_VAR   byref  V50 tmp45
N017 ( 10,  8) [000331] -A-XGO--R--                         │  └──▌  ASG       simd32 (copy)
N016 (  3,  2) [000330] D------N---                         │     ├──▌  LCL_VAR   simd32 V51 tmp46
N015 (  6,  5) [000329] ---XGO-N---                         │     └──▌  IND       simd32
N014 (  4,  3) [000326] -----O-N---                         │        └──▌  ADD       byref
N012 (  3,  2) [000324] -----O-----                         │           ├──▌  LCL_VAR   byref  V50 tmp45
N013 (  1,  1) [000325] -----------                         │           └──▌  CNS_INT   long   32

gtSplitTree would believe that [000327] was a value and would create the
following IR shape:

N016 ( 14, 11) [000428] -A-XGO--R--                         ▌  ASG       simd32 (copy)
N015 (  3,  2) [000427] D------N---                         ├──▌  LCL_VAR   simd32 V76 tmp71
N014 ( 10,  8) [000331] -A-XGO-NR--                         └──▌  ASG       simd32 (copy)
N013 (  3,  2) [000330] D------N---                            ├──▌  LCL_VAR   simd32 V51 tmp46
N012 (  6,  5) [000329] ---XGO-N---                            └──▌  IND       simd32
N011 (  4,  3) [000326] -----O-N---                               └──▌  ADD       byref
N009 (  3,  2) [000324] -----O-----                                  ├──▌  LCL_VAR   byref  V50 tmp45
N010 (  1,  1) [000325] -----------                                  └──▌  CNS_INT   long   32

Fix #83576
@ghost ghost removed the in-pr There is an active PR which will close this issue when it is merged label Mar 21, 2023
@ghost ghost locked as resolved and limited conversation to collaborators Apr 20, 2023
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees

@jakobbotsch jakobbotsch

Labels
area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
Projects
None yet
Milestone
8.0.0
Development

Successfully merging a pull request may close this issue.

JIT: Expand handling of non-value commas in gtSplitTree jakobbotsch/runtime
1 participant
@jakobbotsch