Automatically determine location of nvcc when compiling programs using charmc in accel
When compiling programs that take advantage of the accel framework charmc implicitly invokes nvcc to compile generated cuda code. However, the location of nvcc is hardcoded to /usr/local/cuda/bin instead of whatever is detected when charm is built. Therefore this is related to those issues as well.
There is also nastiness here about checking CMK_CUDA, creating new flags is a separate issue. Ultimately this fix should go in the final accel commit.
Potentially related fix for gpu manager: #511
#4 Updated by Michael Robson almost 2 years ago
- Target version changed from 6.8.1 to 6.9.0
- Category changed from Build & Test Automation to GPU Support
Sam White wrote:
Is this critical for 6.8.1? Retarget to 6.9.0 if not
After looking at the title of this bug I realized it was for accel which isn't even integrated yet... It should be (in some form) in 6.9.0 so I'll target it there.