Closed bkloppenborg closed 10 years ago
The normalize_float.cl kernel could be accelerated by computing the 1/x and storing it into local/global memory. Right now this kernel achieves ~60% occupancy and accounts for < 0.1% of GPU time. Therefore this is quite low in terms of priority.
Closed in 89c82c0.
The normalize_float.cl kernel could be accelerated by computing the 1/x and storing it into local/global memory. Right now this kernel achieves ~60% occupancy and accounts for < 0.1% of GPU time. Therefore this is quite low in terms of priority.