APEX MAIN: 21.48199 - 100.0000% [1] {min=21.4820, max=21.4820, mean=21.4820, var=0.0000, std dev=0.0000} | hipMemcpy: 21.35977 - 99.4311% [20] {min=0.0013, max=2.1064, mean=1.0680, var=1.0671, std dev=1.0330} | | GPU: CopyDeviceToHost: 0.00714 - 0.0334% [10] {min=0.0007, max=0.0008, mean=0.0007, var=0.0000, std dev=0.0000} | | GPU: CopyHostToDevice: 0.00657 - 0.0307% [10] {min=0.0006, max=0.0007, mean=0.0007, var=0.0000, std dev=0.0000} | | Remainder: 21.3461 - 99.3673% | hipLaunchKernel: 0.00133 - 0.0062% [10] {min=0.0000, max=0.0012, mean=0.0001, var=0.0000, std dev=0.0004} | | GPU: __device_stub__bodyForce: 20.96895 - 1581492.5955% [10] {min=2.0921, max=2.1054, mean=2.0969, var=0.0000, std dev=0.0043} | hipFree: 0.00017 - 0.0008% [1] {min=0.0002, max=0.0002, mean=0.0002, var=0.0000, std dev=0.0000} | hipMalloc: 0.00010 - 0.0005% [1] {min=0.0001, max=0.0001, mean=0.0001, var=0.0000, std dev=0.0000} | __hipPushCallConfiguration: 0.00002 - 0.0001% [10] {min=0.0000, max=0.0000, mean=0.0000, var=0.0000, std dev=0.0000} | __hipPopCallConfiguration: 0.00001 - 0.0000% [10] {min=0.0000, max=0.0000, mean=0.0000, var=0.0000, std dev=0.0000} | Remainder: 0.1206 - 0.5613%