comaniac commented on a change in pull request #4465: [AutoTVM] Tune softmax CUDA schedule URL: https://github.com/apache/incubator-tvm/pull/4465#discussion_r355106974
########## File path: topi/python/topi/cuda/softmax.py ########## @@ -52,13 +60,22 @@ def schedule_softmax(outs): raise ValueError('Tag is expected to be softmax_output or log_softmax_output. \ Got {0}'.format(op_tag)) + # create tuning space + max_num_threads = tvm.target.current_target(allow_none=False).max_num_threads + possible_num_thread = get_powers_of_two_in_range(32, max_num_threads) + cfg.define_knob("num_thread", possible_num_thread) Review comment: I personally think it would be better to use `define_split` directly so that this part could be more concise. `define_split` also has an option to use all power of two numbers in a given range as candidates. In addition, do you think if there will bring any improvements if we create two knobs used at `s[expsum].split(k, factor=num_thread)` and `s[softmax].split(softmax.op.axis[1], nparts=num_thread)`? Maybe we need differernt thread number if `k` and `softmax.op.axis[1]` are different, but that will also increase the tuning space so I'm not 100% for sure if this is a good idea. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services