hisbug
September 7, 2020, 9:21am
1
/tests/CPUOperatorTest
./tests/CPUMLTest
LLVMIRCodeGen/libjit/libjit.cpp:902: void (anonymous namespace)::find_min_max_f(float *, dim_t, float &, float &): Assertion `!std::isnan(tensor[i]) && “NaN value found!”’ failed.
Partition failed: the number of given(Interpreter) devices(1) is fewer than the required minimal partitions(2).
Please help me, I am new in glow compiler, I do not know how to start learning this compiler, so I run some test and get 3 errors, please give me some advice for learning this compiler
jfix
September 14, 2020, 6:44am
2
Hi @hisbug , for the error on CPUMLTest, I think that is just a buggy test. Do you see the error every time you run it, or is it inconsistent?
For the error on the Operator test, I’m not sure why you are seeing this error. What cmd are you running exactly?
hisbug
September 14, 2020, 8:37am
3
@jfix Thanks for replying, CPUMLTest has two outputs, it is inconsistent,
/home/hye/tmp/build_Debug/tests/CPUMLTest --gtest_output=xml:CPUMLTest.xml
most of the time, the output like this:
[----------] Global test environment tear-down
[==========] 20 tests from 1 test suite ran. (17844 ms total)
[ PASSED ] 19 tests.
[ SKIPPED ] 1 test, listed below:
[ SKIPPED ] CPU/MLTest.learnSparseLengthsSumEmbeddings/0
sometimes the output like this:
CPUMLTest: /home/hye/tmp/glow/lib/Backends/CPU/../../LLVMIRCodeGen/libjit/libjit.cpp:902: void (anonymous namespace)::find_min_max_f(float *, dim_t, float &, float &): Assertion `!std::isnan(tensor[i]) && "NaN value found!"' failed.
#0 0x000000000300d21e llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/home/hye/tmp/build_Debug/tests/CPUMLTest+0x300d21e)
#1 0x000000000300b114 llvm::sys::RunSignalHandlers() (/home/hye/tmp/build_Debug/tests/CPUMLTest+0x300b114)
#2 0x000000000300b295 SignalHandler(int) (/home/hye/tmp/build_Debug/tests/CPUMLTest+0x300b295)
#3 0x00007f5fd261eb20 __restore_rt (/lib64/libpthread.so.0+0x14b20)
#4 0x00007f5fd2124625 raise (/lib64/libc.so.6+0x3c625)
#5 0x00007f5fd210d8d9 abort (/lib64/libc.so.6+0x258d9)
#6 0x00007f5fd210d7a9 _nl_load_domain.cold (/lib64/libc.so.6+0x257a9)
#7 0x00007f5fd211ca66 (/lib64/libc.so.6+0x34a66)
#8 0x00007f5fd33726b8
For the Operator test, my cmd :
/home/hye/tmp/build_Debug/tests/CPUOperatorTest --gtest_output=xml:CPUOperatorTest.xml
The third failure is unit test 40, my cmd:
/home/hye/tmp/build_Debug/tests/InterpreterMLTest --gtest_output=xml:InterpreterMLTest.xml
output is:
[----------] Global test environment tear-down
[==========] 20 tests from 1 test suite ran. (20686 ms total)
[ PASSED ] 19 tests.
[ SKIPPED ] 1 test, listed below:
[ SKIPPED ] Interpreter/MLTest.learnSparseLengthsSumEmbeddings/0
jfix
September 14, 2020, 7:37pm
4
I’m a bit confused on what the issue is for the CPUOperatorTest. But the MLTest issues should be fixed as of this commit which just landed.