Rewrote the PRNG using an xorshift* algorithm and moved the files from test/ to base/.
Created a simple unit test for the new random number generator. (It mostly tests
that the generated numbers are consistent with the intended distribution, e.g. uniform.
It is not a comprehensive test of the quality of the random numbers.)
Several assertions in OveruseDetectorTest seem to depend on the exact sequence of random numbers. I updated those numbers to work with the new PRNG.
Compute the standard deviation of the expected result in TestReorderFilter instead of passing an uncertainty parameter.
BUG=webrtc:5177
Review URL: https://codereview.webrtc.org/1457023002
Cr-Commit-Position: refs/heads/master@{#10965}
diff --git a/webrtc/base/random.cc b/webrtc/base/random.cc
new file mode 100644
index 0000000..14a9faf
--- /dev/null
+++ b/webrtc/base/random.cc
@@ -0,0 +1,86 @@
+/*
+ * Copyright (c) 2015 The WebRTC project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+#include "webrtc/base/random.h"
+
+#include <math.h>
+
+#include "webrtc/base/checks.h"
+
+namespace webrtc {
+
+Random::Random(uint64_t seed) {
+ RTC_DCHECK(seed != 0x0ull);
+ state_ = seed;
+}
+
+uint32_t Random::Rand(uint32_t t) {
+ // Casting the output to 32 bits will give an almost uniform number.
+ // Pr[x=0] = (2^32-1) / (2^64-1)
+ // Pr[x=k] = 2^32 / (2^64-1) for k!=0
+ // Uniform would be Pr[x=k] = 2^32 / 2^64 for all 32-bit integers k.
+ uint32_t x = NextOutput();
+ // If x / 2^32 is uniform on [0,1), then x / 2^32 * (t+1) is uniform on
+ // the interval [0,t+1), so the integer part is uniform on [0,t].
+ uint64_t result = x * (static_cast<uint64_t>(t) + 1);
+ result >>= 32;
+ return result;
+}
+
+uint32_t Random::Rand(uint32_t low, uint32_t high) {
+ RTC_DCHECK(low <= high);
+ return Rand(high - low) + low;
+}
+
+int32_t Random::Rand(int32_t low, int32_t high) {
+ RTC_DCHECK(low <= high);
+ // We rely on subtraction (and addition) to be the same for signed and
+ // unsigned numbers in two-complement representation. Thus, although
+ // high - low might be negative as an int, it is the correct difference
+ // when interpreted as an unsigned.
+ return Rand(high - low) + low;
+}
+
+template <>
+float Random::Rand<float>() {
+ double result = NextOutput() - 1;
+ result = result / 0xFFFFFFFFFFFFFFFEull;
+ return static_cast<float>(result);
+}
+
+template <>
+double Random::Rand<double>() {
+ double result = NextOutput() - 1;
+ result = result / 0xFFFFFFFFFFFFFFFEull;
+ return result;
+}
+
+template <>
+bool Random::Rand<bool>() {
+ return Rand(0, 1) == 1;
+}
+
+double Random::Gaussian(double mean, double standard_deviation) {
+ // Creating a Normal distribution variable from two independent uniform
+ // variables based on the Box-Muller transform, which is defined on the
+ // interval (0, 1]. Note that we rely on NextOutput to generate integers
+ // in the range [1, 2^64-1]. Normally this behavior is a bit frustrating,
+ // but here it is exactly what we need.
+ const double kPi = 3.14159265358979323846;
+ double u1 = static_cast<double>(NextOutput()) / 0xFFFFFFFFFFFFFFFFull;
+ double u2 = static_cast<double>(NextOutput()) / 0xFFFFFFFFFFFFFFFFull;
+ return mean + standard_deviation * sqrt(-2 * log(u1)) * cos(2 * kPi * u2);
+}
+
+double Random::Exponential(double lambda) {
+ double uniform = Rand<double>();
+ return -log(uniform) / lambda;
+}
+
+} // namespace webrtc