To optimize my code, first I’m trying to optimize register use.
How can I measure the number of registers required per thread?
NVVP tells me that I’m using 63 registers per thread which is maximum, but I don’t have that much scalar variables.
To optimize my code, first I’m trying to optimize register use.
How can I measure the number of registers required per thread?
NVVP tells me that I’m using 63 registers per thread which is maximum, but I don’t have that much scalar variables.
How can I measure the number of registers required per thread?
You can see the number the number of registers used via the flag “-ta=tesla:ptxinfo”. You can set the maximum number of registers to use per thread via the flag “-ta=tesla:maxregcount:n” where “n” is the number of registers.