unittests/Support/BranchProbabilityTest.cpp

   1 //===- unittest/Support/BranchProbabilityTest.cpp - BranchProbability tests -=//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9
  10 #include "llvm/Support/BranchProbability.h"
  11 #include "llvm/Support/raw_ostream.h"
  12 #include "gtest/gtest.h"
  13
  14 using namespace llvm;
  15
  16 namespace llvm {
  17 void PrintTo(const BranchProbability &P, ::std::ostream *os) {
  18   *os << P.getNumerator() << "/" << P.getDenominator();
  19 }
  20 }
  21 namespace {
  22
  23 typedef BranchProbability BP;
  24 TEST(BranchProbabilityTest, Accessors) {
  25   EXPECT_EQ(1u, BP(1, 7).getNumerator());
  26   EXPECT_EQ(7u, BP(1, 7).getDenominator());
  27   EXPECT_EQ(0u, BP::getZero().getNumerator());
  28   EXPECT_EQ(1u, BP::getZero().getDenominator());
  29   EXPECT_EQ(1u, BP::getOne().getNumerator());
  30   EXPECT_EQ(1u, BP::getOne().getDenominator());
  31 }
  32
  33 TEST(BranchProbabilityTest, Operators) {
  34   EXPECT_TRUE(BP(1, 7) < BP(2, 7));
  35   EXPECT_TRUE(BP(1, 7) < BP(1, 4));
  36   EXPECT_TRUE(BP(5, 7) < BP(3, 4));
  37   EXPECT_FALSE(BP(1, 7) < BP(1, 7));
  38   EXPECT_FALSE(BP(1, 7) < BP(2, 14));
  39   EXPECT_FALSE(BP(4, 7) < BP(1, 2));
  40   EXPECT_FALSE(BP(4, 7) < BP(3, 7));
  41
  42   EXPECT_FALSE(BP(1, 7) > BP(2, 7));
  43   EXPECT_FALSE(BP(1, 7) > BP(1, 4));
  44   EXPECT_FALSE(BP(5, 7) > BP(3, 4));
  45   EXPECT_FALSE(BP(1, 7) > BP(1, 7));
  46   EXPECT_FALSE(BP(1, 7) > BP(2, 14));
  47   EXPECT_TRUE(BP(4, 7) > BP(1, 2));
  48   EXPECT_TRUE(BP(4, 7) > BP(3, 7));
  49
  50   EXPECT_TRUE(BP(1, 7) <= BP(2, 7));
  51   EXPECT_TRUE(BP(1, 7) <= BP(1, 4));
  52   EXPECT_TRUE(BP(5, 7) <= BP(3, 4));
  53   EXPECT_TRUE(BP(1, 7) <= BP(1, 7));
  54   EXPECT_TRUE(BP(1, 7) <= BP(2, 14));
  55   EXPECT_FALSE(BP(4, 7) <= BP(1, 2));
  56   EXPECT_FALSE(BP(4, 7) <= BP(3, 7));
  57
  58   EXPECT_FALSE(BP(1, 7) >= BP(2, 7));
  59   EXPECT_FALSE(BP(1, 7) >= BP(1, 4));
  60   EXPECT_FALSE(BP(5, 7) >= BP(3, 4));
  61   EXPECT_TRUE(BP(1, 7) >= BP(1, 7));
  62   EXPECT_TRUE(BP(1, 7) >= BP(2, 14));
  63   EXPECT_TRUE(BP(4, 7) >= BP(1, 2));
  64   EXPECT_TRUE(BP(4, 7) >= BP(3, 7));
  65
  66   EXPECT_FALSE(BP(1, 7) == BP(2, 7));
  67   EXPECT_FALSE(BP(1, 7) == BP(1, 4));
  68   EXPECT_FALSE(BP(5, 7) == BP(3, 4));
  69   EXPECT_TRUE(BP(1, 7) == BP(1, 7));
  70   EXPECT_TRUE(BP(1, 7) == BP(2, 14));
  71   EXPECT_FALSE(BP(4, 7) == BP(1, 2));
  72   EXPECT_FALSE(BP(4, 7) == BP(3, 7));
  73
  74   EXPECT_TRUE(BP(1, 7) != BP(2, 7));
  75   EXPECT_TRUE(BP(1, 7) != BP(1, 4));
  76   EXPECT_TRUE(BP(5, 7) != BP(3, 4));
  77   EXPECT_FALSE(BP(1, 7) != BP(1, 7));
  78   EXPECT_FALSE(BP(1, 7) != BP(2, 14));
  79   EXPECT_TRUE(BP(4, 7) != BP(1, 2));
  80   EXPECT_TRUE(BP(4, 7) != BP(3, 7));
  81 }
  82
  83 TEST(BlockProbabilityTest, MoreOperators) {
  84   BP A(4, 5);
  85   BP B(4U << 29, 5U << 29);
  86   BP C(3, 4);
  87
  88   EXPECT_TRUE(A == B);
  89   EXPECT_FALSE(A != B);
  90   EXPECT_FALSE(A < B);
  91   EXPECT_FALSE(A > B);
  92   EXPECT_TRUE(A <= B);
  93   EXPECT_TRUE(A >= B);
  94
  95   EXPECT_FALSE(B == C);
  96   EXPECT_TRUE(B != C);
  97   EXPECT_FALSE(B < C);
  98   EXPECT_TRUE(B > C);
  99   EXPECT_FALSE(B <= C);
 100   EXPECT_TRUE(B >= C);
 101
 102   BP BigZero(0, UINT32_MAX);
 103   BP BigOne(UINT32_MAX, UINT32_MAX);
 104   EXPECT_FALSE(BigZero == BigOne);
 105   EXPECT_TRUE(BigZero != BigOne);
 106   EXPECT_TRUE(BigZero < BigOne);
 107   EXPECT_FALSE(BigZero > BigOne);
 108   EXPECT_TRUE(BigZero <= BigOne);
 109   EXPECT_FALSE(BigZero >= BigOne);
 110 }
 111
 112 TEST(BranchProbabilityTest, getCompl) {
 113   EXPECT_EQ(BP(5, 7), BP(2, 7).getCompl());
 114   EXPECT_EQ(BP(2, 7), BP(5, 7).getCompl());
 115   EXPECT_EQ(BP::getZero(), BP(7, 7).getCompl());
 116   EXPECT_EQ(BP::getOne(), BP(0, 7).getCompl());
 117 }
 118
 119 TEST(BranchProbabilityTest, scale) {
 120   // Multiply by 1.0.
 121   EXPECT_EQ(UINT64_MAX, BP(1, 1).scale(UINT64_MAX));
 122   EXPECT_EQ(UINT64_MAX, BP(7, 7).scale(UINT64_MAX));
 123   EXPECT_EQ(UINT32_MAX, BP(1, 1).scale(UINT32_MAX));
 124   EXPECT_EQ(UINT32_MAX, BP(7, 7).scale(UINT32_MAX));
 125   EXPECT_EQ(0u, BP(1, 1).scale(0));
 126   EXPECT_EQ(0u, BP(7, 7).scale(0));
 127
 128   // Multiply by 0.0.
 129   EXPECT_EQ(0u, BP(0, 1).scale(UINT64_MAX));
 130   EXPECT_EQ(0u, BP(0, 1).scale(UINT64_MAX));
 131   EXPECT_EQ(0u, BP(0, 1).scale(0));
 132
 133   auto Two63 = UINT64_C(1) << 63;
 134   auto Two31 = UINT64_C(1) << 31;
 135
 136   // Multiply by 0.5.
 137   EXPECT_EQ(Two63 - 1, BP(1, 2).scale(UINT64_MAX));
 138
 139   // Big fractions.
 140   EXPECT_EQ(1u, BP(Two31, UINT32_MAX).scale(2));
 141   EXPECT_EQ(Two31, BP(Two31, UINT32_MAX).scale(Two31 * 2));
 142   EXPECT_EQ(Two63 + Two31, BP(Two31, UINT32_MAX).scale(UINT64_MAX));
 143
 144   // High precision.
 145   EXPECT_EQ(UINT64_C(9223372047592194055),
 146             BP(Two31 + 1, UINT32_MAX - 2).scale(UINT64_MAX));
 147 }
 148
 149 TEST(BranchProbabilityTest, scaleByInverse) {
 150   // Divide by 1.0.
 151   EXPECT_EQ(UINT64_MAX, BP(1, 1).scaleByInverse(UINT64_MAX));
 152   EXPECT_EQ(UINT64_MAX, BP(7, 7).scaleByInverse(UINT64_MAX));
 153   EXPECT_EQ(UINT32_MAX, BP(1, 1).scaleByInverse(UINT32_MAX));
 154   EXPECT_EQ(UINT32_MAX, BP(7, 7).scaleByInverse(UINT32_MAX));
 155   EXPECT_EQ(0u, BP(1, 1).scaleByInverse(0));
 156   EXPECT_EQ(0u, BP(7, 7).scaleByInverse(0));
 157
 158   // Divide by something very small.
 159   EXPECT_EQ(UINT64_MAX, BP(1, UINT32_MAX).scaleByInverse(UINT64_MAX));
 160   EXPECT_EQ(uint64_t(UINT32_MAX) * UINT32_MAX,
 161             BP(1, UINT32_MAX).scaleByInverse(UINT32_MAX));
 162   EXPECT_EQ(UINT32_MAX, BP(1, UINT32_MAX).scaleByInverse(1));
 163
 164   auto Two63 = UINT64_C(1) << 63;
 165   auto Two31 = UINT64_C(1) << 31;
 166
 167   // Divide by 0.5.
 168   EXPECT_EQ(UINT64_MAX - 1, BP(1, 2).scaleByInverse(Two63 - 1));
 169   EXPECT_EQ(UINT64_MAX, BP(1, 2).scaleByInverse(Two63));
 170
 171   // Big fractions.
 172   EXPECT_EQ(1u, BP(Two31, UINT32_MAX).scaleByInverse(1));
 173   EXPECT_EQ(2u, BP(Two31 - 1, UINT32_MAX).scaleByInverse(1));
 174   EXPECT_EQ(Two31 * 2 - 1, BP(Two31, UINT32_MAX).scaleByInverse(Two31));
 175   EXPECT_EQ(Two31 * 2 + 1, BP(Two31 - 1, UINT32_MAX).scaleByInverse(Two31));
 176   EXPECT_EQ(UINT64_MAX, BP(Two31, UINT32_MAX).scaleByInverse(Two63 + Two31));
 177
 178   // High precision.  The exact answers to these are close to the successors of
 179   // the floor.  If we were rounding, these would round up.
 180   EXPECT_EQ(UINT64_C(18446744065119617030),
 181             BP(Two31 + 2, UINT32_MAX - 2)
 182                 .scaleByInverse(UINT64_C(9223372047592194055)));
 183   EXPECT_EQ(UINT64_C(18446744065119617026),
 184             BP(Two31 + 1, UINT32_MAX).scaleByInverse(Two63 + Two31));
 185 }
 186
 187 TEST(BlockProbabilityTest, scaleBruteForce) {
 188   struct {
 189     uint64_t Num;
 190     uint32_t Prob[2];
 191     uint64_t Result;
 192   } Tests[] = {
 193     // Data for scaling that results in <= 64 bit division.
 194     { 0x1423e2a50ULL, { 0x64819521, 0x7765dd13 }, 0x10f418889ULL },
 195     { 0x35ef14ceULL, { 0x28ade3c7, 0x304532ae }, 0x2d73c33aULL },
 196     { 0xd03dbfbe24ULL, { 0x790079, 0xe419f3 }, 0x6e776fc1fdULL },
 197     { 0x21d67410bULL, { 0x302a9dc2, 0x3ddb4442 }, 0x1a5948fd6ULL },
 198     { 0x8664aeadULL, { 0x3d523513, 0x403523b1 }, 0x805a04cfULL },
 199     { 0x201db0cf4ULL, { 0x35112a7b, 0x79fc0c74 }, 0xdf8b07f6ULL },
 200     { 0x13f1e4430aULL, { 0x21c92bf, 0x21e63aae }, 0x13e0cba15ULL },
 201     { 0x16c83229ULL, { 0x3793f66f, 0x53180dea }, 0xf3ce7b6ULL },
 202     { 0xc62415be8ULL, { 0x9cc4a63, 0x4327ae9b }, 0x1ce8b71caULL },
 203     { 0x6fac5e434ULL, { 0xe5f9170, 0x1115e10b }, 0x5df23dd4cULL },
 204     { 0x1929375f2ULL, { 0x3a851375, 0x76c08456 }, 0xc662b082ULL },
 205     { 0x243c89db6ULL, { 0x354ebfc0, 0x450ef197 }, 0x1bf8c1661ULL },
 206     { 0x310e9b31aULL, { 0x1b1b8acf, 0x2d3629f0 }, 0x1d69c93f9ULL },
 207     { 0xa1fae921dULL, { 0xa7a098c, 0x10469f44 }, 0x684413d6cULL },
 208     { 0xc1582d957ULL, { 0x498e061, 0x59856bc }, 0x9edc5f4e7ULL },
 209     { 0x57cfee75ULL, { 0x1d061dc3, 0x7c8bfc17 }, 0x1476a220ULL },
 210     { 0x139220080ULL, { 0x294a6c71, 0x2a2b07c9 }, 0x1329e1c76ULL },
 211     { 0x1665d353cULL, { 0x7080db5, 0xde0d75c }, 0xb590d9fbULL },
 212     { 0xe8f14541ULL, { 0x5188e8b2, 0x736527ef }, 0xa4971be5ULL },
 213     { 0x2f4775f29ULL, { 0x254ef0fe, 0x435fcf50 }, 0x1a2e449c1ULL },
 214     { 0x27b85d8d7ULL, { 0x304c8220, 0x5de678f2 }, 0x146e3bef9ULL },
 215     { 0x1d362e36bULL, { 0x36c85b12, 0x37a66f55 }, 0x1cc19b8e6ULL },
 216     { 0x155fd48c7ULL, { 0xf5894d, 0x1256108 }, 0x11e383602ULL },
 217     { 0xb5db2d15ULL, { 0x39bb26c5, 0x5bdcda3e }, 0x72499259ULL },
 218     { 0x153990298ULL, { 0x48921c09, 0x706eb817 }, 0xdb3268e8ULL },
 219     { 0x28a7c3ed7ULL, { 0x1f776fd7, 0x349f7a70 }, 0x184f73ae1ULL },
 220     { 0x724dbeabULL, { 0x1bd149f5, 0x253a085e }, 0x5569c0b3ULL },
 221     { 0xd8f0c513ULL, { 0x18c8cc4c, 0x1b72bad0 }, 0xc3e30643ULL },
 222     { 0x17ce3dcbULL, { 0x1e4c6260, 0x233b359e }, 0x1478f4afULL },
 223     { 0x1ce036ce0ULL, { 0x29e3c8af, 0x5318dd4a }, 0xe8e76196ULL },
 224     { 0x1473ae2aULL, { 0x29b897ba, 0x2be29378 }, 0x13718185ULL },
 225     { 0x1dd41aa68ULL, { 0x3d0a4441, 0x5a0e8f12 }, 0x1437b6bbfULL },
 226     { 0x1b49e4a53ULL, { 0x3430c1fe, 0x5a204aed }, 0xfcd6852fULL },
 227     { 0x217941b19ULL, { 0x12ced2bd, 0x21b68310 }, 0x12aca65b1ULL },
 228     { 0xac6a4dc8ULL, { 0x3ed68da8, 0x6fdca34c }, 0x60da926dULL },
 229     { 0x1c503a4e7ULL, { 0xfcbbd32, 0x11e48d17 }, 0x18fec7d38ULL },
 230     { 0x1c885855ULL, { 0x213e919d, 0x25941897 }, 0x193de743ULL },
 231     { 0x29b9c168eULL, { 0x2b644aea, 0x45725ee7 }, 0x1a122e5d5ULL },
 232     { 0x806a33f2ULL, { 0x30a80a23, 0x5063733a }, 0x4db9a264ULL },
 233     { 0x282afc96bULL, { 0x143ae554, 0x1a9863ff }, 0x1e8de5204ULL },
 234     // Data for scaling that results in > 64 bit division.
 235     { 0x23ca5f2f672ca41cULL, { 0xecbc641, 0x111373f7 }, 0x1f0301e5e8295ab5ULL },
 236     { 0x5e4f2468142265e3ULL, { 0x1ddf5837, 0x32189233 }, 0x383ca7ba9fdd2c8cULL },
 237     { 0x277a1a6f6b266bf6ULL, { 0x415d81a8, 0x61eb5e1e }, 0x1a5a3e1d41b30c0fULL },
 238     { 0x1bdbb49a237035cbULL, { 0xea5bf17, 0x1d25ffb3 }, 0xdffc51c53d44b93ULL },
 239     { 0x2bce6d29b64fb8ULL, { 0x3bfd5631, 0x7525c9bb }, 0x166ebedda7ac57ULL },
 240     { 0x3a02116103df5013ULL, { 0x2ee18a83, 0x3299aea8 }, 0x35be8922ab1e2a84ULL },
 241     { 0x7b5762390799b18cULL, { 0x12f8e5b9, 0x2563bcd4 }, 0x3e960077aca01209ULL },
 242     { 0x69cfd72537021579ULL, { 0x4c35f468, 0x6a40feee }, 0x4be4cb3848be98a3ULL },
 243     { 0x49dfdf835120f1c1ULL, { 0x8cb3759, 0x559eb891 }, 0x79663f7120edadeULL },
 244     { 0x74b5be5c27676381ULL, { 0x47e4c5e0, 0x7c7b19ff }, 0x4367d2dff36a1028ULL },
 245     { 0x4f50f97075e7f431ULL, { 0x9a50a17, 0x11cd1185 }, 0x2af952b34c032df4ULL },
 246     { 0x2f8b0d712e393be4ULL, { 0x1487e386, 0x15aa356e }, 0x2d0df36478a776aaULL },
 247     { 0x224c1c75999d3deULL, { 0x3b2df0ea, 0x4523b100 }, 0x1d5b481d145f08aULL },
 248     { 0x2bcbcea22a399a76ULL, { 0x28b58212, 0x48dd013e }, 0x187814d084c47cabULL },
 249     { 0x1dbfca91257cb2d1ULL, { 0x1a8c04d9, 0x5e92502c }, 0x859cf7d00f77545ULL },
 250     { 0x7f20039b57cda935ULL, { 0xeccf651, 0x323f476e }, 0x25720cd976461a77ULL },
 251     { 0x40512c6a586aa087ULL, { 0x113b0423, 0x398c9eab }, 0x1341c03de8696a7eULL },
 252     { 0x63d802693f050a11ULL, { 0xf50cdd6, 0xfce2a44 }, 0x60c0177bb5e46846ULL },
 253     { 0x2d956b422838de77ULL, { 0xb2d345b, 0x1321e557 }, 0x1aa0ed16b6aa5319ULL },
 254     { 0x5a1cdf0c1657bc91ULL, { 0x1d77bb0c, 0x1f991ff1 }, 0x54097ee94ff87560ULL },
 255     { 0x3801b26d7e00176bULL, { 0xeed25da, 0x1a819d8b }, 0x1f89e96a3a639526ULL },
 256     { 0x37655e74338e1e45ULL, { 0x300e170a, 0x5a1595fe }, 0x1d8cfb55fddc0441ULL },
 257     { 0x7b38703f2a84e6ULL, { 0x66d9053, 0xc79b6b9 }, 0x3f7d4c91774094ULL },
 258     { 0x2245063c0acb3215ULL, { 0x30ce2f5b, 0x610e7271 }, 0x113b916468389235ULL },
 259     { 0x6bc195877b7b8a7eULL, { 0x392004aa, 0x4a24e60c }, 0x530594fb17db6ba5ULL },
 260     { 0x40a3fde23c7b43dbULL, { 0x4e712195, 0x6553e56e }, 0x320a799bc76a466aULL },
 261     { 0x1d3dfc2866fbccbaULL, { 0x5075b517, 0x5fc42245 }, 0x18917f0061595bc3ULL },
 262     { 0x19aeb14045a61121ULL, { 0x1bf6edec, 0x707e2f4b }, 0x6626672a070bcc7ULL },
 263     { 0x44ff90486c531e9fULL, { 0x66598a, 0x8a90dc }, 0x32f6f2b0525199b0ULL },
 264     { 0x3f3e7121092c5bcbULL, { 0x1c754df7, 0x5951a1b9 }, 0x14267f50b7ef375dULL },
 265     { 0x60e2dafb7e50a67eULL, { 0x4d96c66e, 0x65bd878d }, 0x49e31715ac393f8bULL },
 266     { 0x656286667e0e6e29ULL, { 0x9d971a2, 0xacda23b }, 0x5c6ee315ead6cb4fULL },
 267     { 0x1114e0974255d507ULL, { 0x1c693, 0x2d6ff }, 0xaae42e4b35f6e60ULL },
 268     { 0x508c8baf3a70ff5aULL, { 0x3b26b779, 0x6ad78745 }, 0x2c98387636c4b365ULL },
 269     { 0x5b47bc666bf1f9cfULL, { 0x10a87ed6, 0x187d358a }, 0x3e1767155848368bULL },
 270     { 0x50954e3744460395ULL, { 0x7a42263, 0xcdaa048 }, 0x2fe739f0aee1fee1ULL },
 271     { 0x20020b406550dd8fULL, { 0x3318539, 0x42eead0 }, 0x186f326325fa346bULL },
 272     { 0x5bcb0b872439ffd5ULL, { 0x6f61fb2, 0x9af7344 }, 0x41fa1e3bec3c1b30ULL },
 273     { 0x7a670f365db87a53ULL, { 0x417e102, 0x3bb54c67 }, 0x8642a558304fd9eULL },
 274     { 0x1ef0db1e7bab1cd0ULL, { 0x2b60cf38, 0x4188f78f }, 0x147ae0d6226b2ee6ULL }
 275   };
 276
 277   for (const auto &T : Tests) {
 278     EXPECT_EQ(T.Result, BP(T.Prob[0], T.Prob[1]).scale(T.Num));
 279   }
 280 }
 281
 282 }