FPChecker
FPChecker copied to clipboard
LLNL
A dynamic analysis tool to detect floating-point errors in HPC applications.
After generating trace, and running fpc-create-report command, gives the following error. Tools/FPChecker/build/bin/fpc-create-report Generating FPChecker report... Trace files found: 9 Traceback (most recent call last): File "Tools/FPChecker/build/bin/fpc-create-report", line 471, in loadEvents(fileList)...
When I run the example in the static directory, clang reported a segment error The configuration of my ubuntu is ubuntu16.04 、 the CUDA's version is CUDA9.0、the version of clang...
This was incorrectly instrumented: This line: `fptr[idx_dst] = (*fine_volumes)(SAMRAI::pdat::CellIndex(*bi));` Became: `fptr[idx_dst] = _FPC_CHECK_((*fine_volumes)(SAMRAI::pdat::CellIndex(* bi), ), 125, “..ineOperator.cxx”);` Here a function pointer is assigned to an array element.
Add support for checking on basic operations(+,-,*,/): from this: `_FPC_CHECK(x + y - z)` change to this: `_FPC_CHECK(x + _FPC_CHECK(y - _FPC_CHECK(z)))`
Currently tests assume a fixed compute capability of sm_65. We should provide a mechanism to determine the compute capability dynamically (when tests are run).
We need to provide a platform-independent RAJA test for the llvm version
Test from master run on python 2.x and tests for clang_plugin runs on python 3.x. We should be consistent.
We should add functionality to check for almost subnormal numbers in the host checks.
The plugin tests are not functional when we make the tool.