Some tests failed on MSVC 2019 (x64 build) #478
Comments
|
|
|
MSVC 19.21 - |
|
Hello! Do you see these errors every time you start tests, or do they run sporadically? |
|
Windows 7 SP1 - every time, Pull request #480 does not solve this issue. |
|
Can you provide a more detailed description of your hardware? |
|
Windows 7 SP1: Windows 10 Bare metal, |
|
Commit b3fb839 |
|
We tried to get the same errors on the test built with same parameters, but did not get the result described by you. Could you please give a more detailed description of the hardware and software on which the tests are falling. As well as a complete set of steps for building and running tests, for the most accurate reproduction of the error. |
|
Before testing, I completely reinstalled the system.
On Windows 7 tests failed every time. Test hardware: VM: |
|
Windows 7 SP1 x64 without virtualization on CPU Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz (4 core, 8 thread) 2021.3.0: commit 68e075c: |
|
Hello, we tried to reproduce the errors on the available for us MSVC 19.21, but we could not reproduce them. Can you please update your MSVC compiler to the latest version and build the library in Debug mode. If errors are still reproducible, then please send the most complete call stack |
|
Hello! |
|
Windows 10, MSVC 19.21 and 19.29. We can`t use Windows 7 because it is no longer supported and has outdated security protocols. |
|
Have you tried the latest version of the compiler? |
|
Windows 7 screen: Remove |
|
It seems weird: if the alignment the only issue, almost all the tests should fail because each parallel construction uses tasks. What is the allocated
Does it mean that the issue reproduced with only |
|
I don't know why this helped. I have no idea. And I have not found similar errors on the Internet.
I know, but this change gave me the idea to check the alignments. Before remove Other tests without |
|
So, it is specific to -struct task_proxy : public d1::task {
+class alignas(d1::task_alignment) task_proxy : public d1::task {
+public: |
|
Not working :( |
|
On CPU Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz (8 core) Tests (msvc 19.20): In virtualized environment with 2 virtual cores this tests is passed. |
|
Commit 68e075c: On CPU Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz (8 core):
Tests (msvc 19.20): Test I think this is a different problem. Not related to alignas. |
It seems we are stuck. Is it acceptable to create a dump file and share it with us? If yes, can you share the dump of failed conformance_parallel_for with unmodified version of oneTBB in RelWithDebInfo mode? To create a dump file in Visual Studio: "Debug"->"Save Dump As...". |
|
Done. |
|
Thank you for the dump. At failed address Can you please share tbb12.pdb that I can connect assembler with the code that generates this instruction? |
|
The fail is occurred in construction of proxy task I looked at the code generated with msvc 19.28 - the compiler does not suppose any alignment: It seems as an issue of msvc 19.20 that it uses aligned accesses where it cannot guarantee the required alignment. Looking at the source code, I could not find any UB that can cause abnormal behavior of the compiler. @phprus, Do you agree with the analysis that msvc 19.20 generates broken code and it does not make sense to fix anything in that regard? (I am speaking only about the tests related to this issue, as for hangs, it seems another story) |
|
Very strange compiler behavior... At the beginning of the week, I'll check the hanging tests again and try to find more information. |
|
Commit a080baf Threads: Main Thread call stack:
Other threads call stack: All MSVC up to 19.29 has same issues. |
|
Test If the test is successful, the warning " Maybe there is a race condition in the initialization of the library? |
|
More examples of errors: |
|
Thank you for the logs. I will try to figure out what is going on. |
|
@phprus can you please try reproduce hand on test_task_group with current master? |
|
Commit: 4df48f9 Hangs on 2 core cpu:
Tomorrow I will run all tests on an 8 core cpu. |
|
Hangs on 8 core cpu:
|
|
Some of the launches are 2 or more warnings and the other part of the launches is one warning. This is expected behavior? |
|
Yes, it's a lot of random in this test. |
|
New infinity loop on Linux now! Commit: 1ecde27 Backtrace from SLES 11sp3 (Release build): |
|
Commit: 8584c45 cmake command ("-O3 -g"): Output before hangs: Backtrace: |
|
I have checked
I ran this test ~ 30-40 times for each configuration, but it always passes |
|
@phprus , it seems we cannot reproduce the issue. Can you share a dump (and pdb files for tbb library and test application) of some of the hanged tests? E.g. |
|
Unfortunately, I cannot give access to the servers where the error is reproduced :( Dump and pdb for |
|
Hello @alexey-katranov, Commit: d2405e3 Dump and PDB files in attach: |
|
I have a suspicion about the possible issue, can you please try the fix in -#if USE_WINTHREAD
+#if __TBB_USE_WINAPI-#if __TBB_WIN8UI_SUPPORT && (_WIN32_WINNT < 0x0A00)
+#if __TBB_USE_WINAPI |
To fix this error, I replaced On 8 core computer, the error in tests:
is no longer reproducible. But why did this change solve the problem? |
See #553. The root cause is that stack size is incorrectly calculated. In your environment it causes stack anchor overflow that leads to hangs. |
|
I have a similar issue, but on a very different system: I've build Now, I see that #553 is very specific to Windows, but seeing as the symptoms I'm facing are very similar: do you think the root cause could somehow be similar? And do you have any suggestion for how this could be fixed? |
|
@casparvl #553 is really specific to Windows, can you please try 2021.3 or master to check if the issue is still present (there was a big set of fixes)? If present, feel free to open a separate issue (because this one is really overloaded and I am not sure if we investigated all the issues in this thread :) ) |
|
Thanks @alexey-katranov for the quick response. I'll check 2021.3 and if still present, I'll submit a new (seperate) issue :) |
|
Fixed by #553. |
Release build:
Tests:
RelWithDebInfo build:
Tests:
LOG:
The text was updated successfully, but these errors were encountered: