andrei-pokrovsky / p100_exp Goto Github PK

Microbenchmarking experiments for the P100

# compile
nvcc -o <executable> <sources...>

# executing
./<executable>

# profiling
nvprof <executable>
options:
* --print-gpu-trace       | breaks down run time, register / memory usage, allocations
* --events warps_launched | shows num of warps scheduled
* --output-profile <file> | dumps profile data to use / visualize elsewhere

# dumping object
cuobjdump <executable>
options:
* -reg-usage | show allocated resources (registers, memory, etc)
* -sass -ptx | disassembles into PTX