diff options
| author |  Abseil Team <absl-team@google.com> | 2020-07-09 08:13:12 -0700 |
|---|---|---|
| committer |  Gennadiy Rozental <rogeeff@google.com> | 2020-07-09 11:59:40 -0400 |
| commit | d5269a8b6dbe7836049417d0ff2c88b8363cc1fc (patch) | |
| tree | 007496a6f35159aabff38bf1a258e324801faf90 /absl/synchronization/mutex.h | |
| parent | bf655de09b67fd8b924814cbb369cb65ddd0bd24 (diff) | |
Export of internal Abseil changes
--
4833151c207fac9f57a735efe6d5db4c83368415 by Gennadiy Rozental <rogeeff@google.com>:
Import of CCTZ from GitHub.
PiperOrigin-RevId: 320398694
--
a1becb36b223230f0a45f204a5fb33b83d2deffe by Gennadiy Rozental <rogeeff@google.com>:
Update CMakeLists.txt
Import of https://github.com/abseil/abseil-cpp/pull/737
PiperOrigin-RevId: 320391906
--
b529c45856fe7a3447f1f3259286d57e13b1f292 by Abseil Team <absl-team@google.com>:
Improves a comment about use of absl::Condition.
PiperOrigin-RevId: 320384329
--
c7b1dacda2739c10dc1ccbfb56b07ed7fe2464a4 by Laramie Leavitt <lar@google.com>:
Improve FastUniformBits performance for std::minstd_rand.
The rejection algorithm was too pessimistic before, and not in line with the [rand.adapt.ibits]. Specifically, when sampling from an URBG with a non power of 2 range, FastUniformBits constructed a rejection threshold with a power-of-2 range that was too restrictive.
For example, minstd_rand has a range of
[1, 2147483646], which has a range of 2145386495, or about 30.999 bits.
Before FastUniformBits rejected values between 1<<30 and 2145386495, which includes approximately 50% of the generated values. However, since a minimum of 3 calls are required to generate a full 64-bit value from an entropy pool of 30.9 bits, the correct value for rejection sampling is the range value which masks 21 (0x7fe00000) or 22 bits and rejects values greater than that. This reduces the probability of rejecting a sample to about 0.1%
NOTE: Abseil random does not guarantee sequence stability over time, and this is expected to change sequences in some cases.
PiperOrigin-RevId: 320285836
--
15800a39557a07dd52e0add66a0ab67aed00590b by Gennadiy Rozental <rogeeff@google.com>:
Internal change.
PiperOrigin-RevId: 320220913
--
ef39348360873f6d19669755fe0b5d09a945a501 by Gennadiy Rozental <rogeeff@google.com>:
Internal change
PiperOrigin-RevId: 320181729
--
4f9f6ef8034a24da1832e4c838c72f80fc2ea062 by Gennadiy Rozental <rogeeff@google.com>:
Internal change
PiperOrigin-RevId: 320176084
--
6bfc8008462801657d231585bd5c37fc18bb25b6 by Gennadiy Rozental <rogeeff@google.com>:
Internal change
PiperOrigin-RevId: 320176070
--
b35b055ab1f41e6056031ff0641cabab23530027 by Abseil Team <absl-team@google.com>:
Disabling using header module as well as building one for randen_hwaes_impl
PiperOrigin-RevId: 320024299
GitOrigin-RevId: 4833151c207fac9f57a735efe6d5db4c83368415
Change-Id: I9cf102dbf46ed07752a508b7cda3ab3858857d0d
Diffstat (limited to 'absl/synchronization/mutex.h')
| -rw-r--r-- | absl/synchronization/mutex.h | 5 |
1 files changed, 5 insertions, 0 deletions
diff --git a/absl/synchronization/mutex.h b/absl/synchronization/mutex.h index 876698ca..52401fe3 100644 --- a/absl/synchronization/mutex.h +++ b/absl/synchronization/mutex.h @@ -685,6 +685,11 @@ class Condition { // return processed_ >= current; // }; // mu_.Await(Condition(&reached)); + // + // NOTE: never use "mu_.AssertHeld()" instead of "mu_.AssertReadHeld()" in the + // lambda as it may be called when the mutex is being unlocked from a scope + // holding only a reader lock, which will make the assertion not fulfilled and + // crash the binary. // See class comment for performance advice. In particular, if there // might be more than one waiter for the same condition, make sure |