After profiling with nvprof in a machine with the following configuration (V100, cc 7.0, Ubuntu 20.04.2 ) I noticed that the value returned to the parameter “inst_executed” is much smaller than “inst_integer” or even “inst_fp_64”. Are inst_executed the instructions executed in one warp and the other instructions parameters the total amount of instructions? The values don’t seem to sum up.
Below are the values I mentioned.
inst_executed - Avg 5717846
inst_fp_64 - 5242880
inst_integer - 15729152
I attached a doc with the values I obtained with the command line " nvprof --metrics all ./sumconstant"
Thank you in advance
metrics_sumconstant_nvprof.txt (48.6 KB)