Bug: NVCC 11.3 diagnoses False Positive "unused parameter" with parameter pack

Accelerated Computing CUDA CUDA NVCC Compiler
1

Full example here:

It should not warn on line 13, since the parameter pack is passed to the foo function.

This doesn’t show up in 11.2 and earlier. But 11.2 also doesn’t catch any true positives.

How should we proceed?

2

Could you please file us a ticket following the instruction here Getting Help with CUDA NVCC Compiler also with this topic link in the bug description , so that we can keep you informed about the investigation . Thanks.

3

@Yuki_Ni Bug submitted: * 3432790

What is the turnaround time for bugs? I opened 3373815 more than 2 months ago and haven’t got any feedback yet.

4

Your original issue is talked internally and here is the info
The problem is the duplicate common.o in both libraries. If you copy common.cu to common2.cu, and then link common2.o in libb, common.o in liba, then it works.

Also here is my other workaround on trying your case

yni@node1:~/yni/CustomerBug/nvbug3373815$ nvcc -arch=sm_75 -dlink common.cu.o a.cu.o b.cu.o -o ab.dlink.o
yni@node1:~/yni/CustomerBug/nvbug3373815$

yni@node1:~/yni/CustomerBug/nvbug3373815$ g++  -shared common.cu.o a.cu.o b.cu.o ab.dlink.o -L /usr/local/cuda-11.5/lib64 -lcuda -lcudart -o libmy.so
yni@node1:~/yni/CustomerBug/nvbug3373815$ 
yni@node1:~/yni/CustomerBug/nvbug3373815$ g++ main.cpp -L . -lmy -o myMain
yni@node1:~/yni/CustomerBug/nvbug3373815$
yni@node1:~/yni/CustomerBug/nvbug3373815$
yni@node1:~/yni/CustomerBug/nvbug3373815$ LD_LIBRARY_PATH=. ./myMain
no error
Running A: 579
no error
Running B: 444
5

Thanks, I think this info should be posted in the related thread/bug. Let’s not clutter this post with unrelated info