Here's what the documentation says:
"Share of Preference (Logit) Rule: Respondents are allowed to split their votes across the items included in the simulation set. The probability that an item is selected is equal to the antilog of the item's raw score divided by the summation of the antilogs of the raw scores for all items in the set."
Let me give a numeric example to illustrate. Imagine you are simulating using just three items: A, B, and C. Imagine the utilities (just for respondent #1) are:
A = 1
B = 2
C = 3
You first take the antilog of each of the scores. You can do this in Excel using the =EXP() function.
=EXP(1) = 2.72
=EXP(2) = 7.39
=EXP(3) = 20.09
You sum those (30.19). Then, divide each of the exponentiated scores by their sum, leading to:
A = 0.09
B = 0.24
C = 0.67
Note that this is done for each respondent separately. Then, the scores are averaged across respondents to obtain the population estimate.
Note also that the market simulation math using the logit rule is different from the "Probability of Choice" transformation that is done on the scores within the MaxDiff Analyzer. That transformation is described in more detail within Appendix K of the CBC/HB manual: http://www.sawtoothsoftware.com/support/manuals/cbc-hb-help