[Workaround] gdb error REGMAP_MAX_ENTRIES failed

Accelerated Computing CUDA CUDA Programming and Performance
1

i try to debug large piece of code inside cuda using nsight

any ideas how i can overcome the following error?

cuda-gdb/7.6/gdb/cuda-regmap.c:163: internal-error: regmap_get_location_index: Assertion `idx < REGMAP_MAX_ENTRIES' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Create a core file of GDB? (y or n) [answered Y; input not from terminal]
2

I don’t have any suggestions for how to overcome that error, but if you can develop a sequence of steps that will reliably reproduce that error, I would suggest filing a bug at developer.nvidia.com

Are you using the latest CUDA version? Bugs get fixed all the time.

3

The error comes and goes when i modify the code,
every time in different place,

i tried to run the code on the latest cuda (7.5)
it just moved the error to a new location.

4

we have found a workaround.

split the function into two.

void foo{
line 1
line 2
line 3
line 4
}

to

void foo{
line 1
line 2
foo2();
}
void foo2{
line 3
line 4
}

i would file a bug report, but it is not possible to share the code, and i failed to replicate it in standalone example. any ideas how to reach them?

5

Not sure what you mean by “any ideas how to reach them?”.

Bug reports filed with NVIDIA are visible to the filer and NVIDIA engineers who work on their resolution, not the public at large. Depending on the gravity of your concerns about the proprietary code, you could also additionally obfuscate the code (e.g. by the use of nonsense variable names). Depending on which organization you work for, they may also have a dedicated contacted at NVIDIA.