diff options
Diffstat (limited to 'mfbt/FastBernoulliTrial.h')
| -rw-r--r-- | mfbt/FastBernoulliTrial.h | 7 |
1 files changed, 3 insertions, 4 deletions
diff --git a/mfbt/FastBernoulliTrial.h b/mfbt/FastBernoulliTrial.h index 7e38b70ab..7618a552a 100644 --- a/mfbt/FastBernoulliTrial.h +++ b/mfbt/FastBernoulliTrial.h @@ -1,5 +1,4 @@ /* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ -/* vim: set ts=8 sts=2 et sw=2 tw=80: */ /* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ @@ -251,7 +250,7 @@ class FastBernoulliTrial { * - The gaps below 1 are 2**-53, so that interval is (0, 1-2**-53]. * * - Because the floating-point gaps near 1 are wider than those near - * zero, there are many small positive doubles ε such that 1-ε rounds to + * zero, there are many small positive doubles ?? such that 1-?? rounds to * exactly 1. However, 2**-53 can be represented exactly. So * 1-mProbability is in [2**-53, 1]. * @@ -264,9 +263,9 @@ class FastBernoulliTrial { * * - How much of the range of mProbability does this cause us to ignore? * The only value for which log returns 0 is exactly 1; the slope of log - * at 1 is 1, so for small ε such that 1 - ε != 1, log(1 - ε) is -ε, + * at 1 is 1, so for small ?? such that 1 - ?? != 1, log(1 - ??) is -??, * never 0. The gaps near one are larger than the gaps near zero, so if - * 1 - ε wasn't 1, then -ε is representable. So if log(1 - mProbability) + * 1 - ?? wasn't 1, then -?? is representable. So if log(1 - mProbability) * isn't 0, then 1 - mProbability isn't 1, which means that mProbability * is at least 2**-53, as discussed earlier. This is a sampling * likelihood of roughly one in ten trillion, which is unlikely to be |