Hello Author!
I have attempted to run the smoke-test script within the container you provided. When extracting traces, the tracer.cu file executed successfully until the last line, but encountered an error upon exiting the CUDA context.
------------- NVBit (NVidia Binary Instrumentation Tool v1.5.1) Loaded --------------
NVBit core environment variables (mostly for nvbit-devs):
NVDISASM = nvdisasm - override default nvdisasm found in PATH
NOBANNER = 0 - if set, does not print this banner
---------------------------------------------------------------------------------
INSTR_BEGIN = 0 - Beginning of the instruction interval where to apply instrumentation
INSTR_END = 4294967295 - End of the instruction interval where to apply instrumentation
----------------------------------------------------------------------------------------------------
WG size of kernel = 16 X 16
pyramidHeight: 2
gridSize: [512, 512]
border:[2, 2]
blockGrid:[43, 43]
targetBlock:[12, 12]
Start computing the transient temperature
Starting kernel #1...
calculate_temp(int, float*, float*, float*, int, int, int, int, float, float, float, float, float, float)
Exiting kernel #1...
Ending simulation
# Collected COMPUTE + MEMORY traces for 1 kernels
# issue.config for 1 kernels
# instn.config for 1 kernels
trace extraction finished!
ASSERT FAIL: nvbit_imp.cpp:508:void Nvbit::destroy_ctx(CUcontext, bool): FAIL !(context_map.find(cuctx) != context_map.end())
The error seems to be occurring within the NVBit library, making it difficult for me to pinpoint the exact location and cause of the issue. I ran the provided container on RTX 2080 Ti. Any possible solution will be appreciated!
By the way, could you please inform me about the standard format for the traces? As previously mentioned, tracer.cu was able to execute, allowing me to obtain some traces(though I am uncertain of its accuracy).But upon reviewing the sections of the source code that handle traces, I seem unable to match them with my traces.
Firstly, the code requires the trace directory to be xxx_traces/kernel-1, but I do not have the kernel-1 subdirectory. My directory path is the same as the commented-out part in the code. Secondly, the SASS traces processed by the code appear to have three data pieces per line, yet mine only contain pc and mask_str,without gwarp_id. I am unsure if these discrepancies are caused by the aforementioned error.
Looking forward to your reply! Thanks in advance!
Hello Author!
I have attempted to run the
smoke-testscript within the container you provided. When extracting traces, thetracer.cufile executed successfully until the last line, but encountered an error upon exiting the CUDA context.The error seems to be occurring within the NVBit library, making it difficult for me to pinpoint the exact location and cause of the issue. I ran the provided container on RTX 2080 Ti. Any possible solution will be appreciated!
By the way, could you please inform me about the standard format for the traces? As previously mentioned,
tracer.cuwas able to execute, allowing me to obtain some traces(though I am uncertain of its accuracy).But upon reviewing the sections of the source code that handle traces, I seem unable to match them with my traces.Firstly, the code requires the trace directory to be
xxx_traces/kernel-1, but I do not have thekernel-1subdirectory. My directory path is the same as the commented-out part in the code. Secondly, the SASS traces processed by the code appear to have three data pieces per line, yet mine only containpcandmask_str,withoutgwarp_id. I am unsure if these discrepancies are caused by the aforementioned error.Looking forward to your reply! Thanks in advance!