Hi,
I’ve noticed something odd when using the icamax and icamin functions from the CUBLAS library. It appears that icamax finds the index of the maximum of the sum of the absolute value of the real part and the absolute value of the imaginary part, but icamin finds the index of the minimum of the Euclidean norm. In other words, icamax performs a calculation similar to scacum, but icamin performs a calculation similar to scnrm2. Is this discrepancy intentional, or is it a bug in CUBLAS?
I encountered this discrepancy in CUDA toolkit version 2.0 on 32-bit Windows XP, compiled with Microsoft Visual C++ 2008 Express Edition.
Thanks in advance for any information,
–Chris