2017-04-01 09:19:00 +00:00
|
|
|
#include <Functions/FunctionsRandom.h>
|
2020-05-28 13:01:00 +00:00
|
|
|
#include <Functions/FunctionFactory.h>
|
2020-05-20 15:43:01 +00:00
|
|
|
#include <Functions/VectorExtension.h>
|
2018-12-02 01:57:01 +00:00
|
|
|
#include <Common/HashTable/Hash.h>
|
|
|
|
#include <Common/randomSeed.h>
|
2021-10-02 07:13:14 +00:00
|
|
|
#include <base/unaligned.h>
|
2020-05-28 15:38:07 +00:00
|
|
|
#if USE_MULTITARGET_CODE
|
|
|
|
# include <x86intrin.h>
|
|
|
|
#endif
|
2018-12-02 01:57:01 +00:00
|
|
|
|
2020-05-29 09:18:39 +00:00
|
|
|
|
2014-08-22 00:57:20 +00:00
|
|
|
namespace DB
|
|
|
|
{
|
|
|
|
|
2018-12-02 01:57:01 +00:00
|
|
|
namespace
|
2017-07-25 17:00:38 +00:00
|
|
|
{
|
2020-05-15 12:00:20 +00:00
|
|
|
/// NOTE Probably
|
|
|
|
/// http://www.pcg-random.org/
|
|
|
|
/// or http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/SFMT/
|
|
|
|
/// or http://docs.yeppp.info/c/group__yep_random___w_e_l_l1024a.html
|
|
|
|
/// could go better.
|
|
|
|
|
2018-12-02 01:57:01 +00:00
|
|
|
struct LinearCongruentialGenerator
|
|
|
|
{
|
2020-05-15 12:00:20 +00:00
|
|
|
/// Constants from `man lrand48_r`.
|
2018-12-02 01:57:01 +00:00
|
|
|
static constexpr UInt64 a = 0x5DEECE66D;
|
|
|
|
static constexpr UInt64 c = 0xB;
|
|
|
|
|
2020-05-15 12:00:20 +00:00
|
|
|
/// And this is from `head -c8 /dev/urandom | xxd -p`
|
2018-12-02 01:57:01 +00:00
|
|
|
UInt64 current = 0x09826f4a081cee35ULL;
|
|
|
|
|
|
|
|
void seed(UInt64 value)
|
|
|
|
{
|
|
|
|
current = value;
|
|
|
|
}
|
|
|
|
|
|
|
|
UInt32 next()
|
|
|
|
{
|
|
|
|
current = current * a + c;
|
|
|
|
return current >> 16;
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
2020-05-20 12:42:21 +00:00
|
|
|
UInt64 calcSeed(UInt64 rand_seed, UInt64 additional_seed)
|
|
|
|
{
|
|
|
|
return intHash64(rand_seed ^ intHash64(additional_seed));
|
|
|
|
}
|
|
|
|
|
|
|
|
void seed(LinearCongruentialGenerator & generator, UInt64 rand_seed, intptr_t additional_seed)
|
2017-07-25 17:00:38 +00:00
|
|
|
{
|
2020-05-20 12:42:21 +00:00
|
|
|
generator.seed(calcSeed(rand_seed, additional_seed));
|
2017-07-25 17:00:38 +00:00
|
|
|
}
|
2020-05-20 12:42:21 +00:00
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
/// The array of random numbers from 'head -c8 /dev/urandom | xxd -p'.
|
|
|
|
/// Can be used for creating seeds for random generators.
|
|
|
|
constexpr std::array<UInt64, 32> random_numbers = {
|
|
|
|
0x0c8ff307dabc0c4cULL, 0xf4bce78bf3821c1bULL, 0x4eb628a1e189c21aULL, 0x85ae000d253e0dbcULL,
|
|
|
|
0xc98073e6480f8a10ULL, 0xb17e9b70a084d570ULL, 0x1361c752b768da8cULL, 0x3d915f60c06d144dULL,
|
|
|
|
0xd5bc9b7aced79587ULL, 0x66c28000ba8a66cfULL, 0x0fb58da7a48820f5ULL, 0x540ee1b57aa861a1ULL,
|
|
|
|
0x212f11936ef2db04ULL, 0xa3939cd900edcc58ULL, 0xc676c84420170102ULL, 0xcbdc824e8b4bf3edULL,
|
|
|
|
|
|
|
|
0x8296f9d93cc94e3bULL, 0x78a7e826d62085b2ULL, 0xaa30620211fc6c69ULL, 0xbd38de52f0a93677ULL,
|
|
|
|
0x19983de8d79dcc4eULL, 0x8afe883ef2199e6fULL, 0xb7160f7ed022b60aULL, 0x2ce173d373ddafd4ULL,
|
|
|
|
0x15762761bb55b9acULL, 0x3e448fc94fdd28e7ULL, 0xa5121232adfbe70aULL, 0xb1e0f6d286112804ULL,
|
|
|
|
0x6062e96de9554806ULL, 0xcc679b329c28882aULL, 0x5c6d29f45cbc060eULL, 0x1af1325a86ffb162ULL,
|
|
|
|
};
|
2017-07-25 17:00:38 +00:00
|
|
|
}
|
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
DECLARE_DEFAULT_CODE(
|
|
|
|
|
2018-12-02 01:57:01 +00:00
|
|
|
void RandImpl::execute(char * output, size_t size)
|
|
|
|
{
|
|
|
|
LinearCongruentialGenerator generator0;
|
|
|
|
LinearCongruentialGenerator generator1;
|
|
|
|
LinearCongruentialGenerator generator2;
|
|
|
|
LinearCongruentialGenerator generator3;
|
2017-07-25 17:00:38 +00:00
|
|
|
|
2020-05-20 12:42:21 +00:00
|
|
|
UInt64 rand_seed = randomSeed();
|
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
seed(generator0, rand_seed, random_numbers[0] + reinterpret_cast<intptr_t>(output));
|
|
|
|
seed(generator1, rand_seed, random_numbers[1] + reinterpret_cast<intptr_t>(output));
|
|
|
|
seed(generator2, rand_seed, random_numbers[2] + reinterpret_cast<intptr_t>(output));
|
|
|
|
seed(generator3, rand_seed, random_numbers[3] + reinterpret_cast<intptr_t>(output));
|
2017-07-25 17:00:38 +00:00
|
|
|
|
2018-12-02 01:57:01 +00:00
|
|
|
for (const char * end = output + size; output < end; output += 16)
|
|
|
|
{
|
2019-06-28 16:21:05 +00:00
|
|
|
unalignedStore<UInt32>(output, generator0.next());
|
|
|
|
unalignedStore<UInt32>(output + 4, generator1.next());
|
|
|
|
unalignedStore<UInt32>(output + 8, generator2.next());
|
|
|
|
unalignedStore<UInt32>(output + 12, generator3.next());
|
2018-12-02 01:57:01 +00:00
|
|
|
}
|
2020-05-15 12:00:20 +00:00
|
|
|
/// It is guaranteed (by PaddedPODArray) that we can overwrite up to 15 bytes after end.
|
2014-08-22 00:57:20 +00:00
|
|
|
}
|
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
) // DECLARE_DEFAULT_CODE
|
2020-05-20 15:43:01 +00:00
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
DECLARE_AVX2_SPECIFIC_CODE(
|
2020-05-20 12:42:21 +00:00
|
|
|
|
2020-05-20 15:43:01 +00:00
|
|
|
using namespace VectorExtension;
|
2020-05-20 12:42:21 +00:00
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
/* Takes 2 vectors with LinearCongruentialGenerator states and combines them into vector with random values.
|
|
|
|
* From every rand-state we use only bits 15...47 to generate random vector.
|
|
|
|
*/
|
2020-06-21 18:47:16 +00:00
|
|
|
inline UInt64x4 combineValues(UInt64x4 a, UInt64x4 b)
|
2020-05-19 10:54:41 +00:00
|
|
|
{
|
2020-05-24 12:25:07 +00:00
|
|
|
auto xa = reinterpret_cast<__m256i>(a);
|
|
|
|
auto xb = reinterpret_cast<__m256i>(b);
|
|
|
|
/// Every state is 8-byte value and we need to use only 4 from the middle.
|
|
|
|
/// Swap the low half and the high half of every state to move these bytes from the middle to sides.
|
|
|
|
/// xa = xa[1, 0, 3, 2, 5, 4, 7, 6]
|
|
|
|
xa = _mm256_shuffle_epi32(xa, 0xb1);
|
|
|
|
/// Now every 8-byte value in xa is xx....xx and every value in xb is ..xxxx.. where x is random byte we want to use.
|
|
|
|
/// Just blend them to get the result vector.
|
2020-05-25 15:16:19 +00:00
|
|
|
/// result = xa[0],xb[1,2],xa[3,4],xb[5,6],xa[7,8],xb[9,10],xa[11,12],xb[13,14],xa[15]
|
2020-05-26 11:15:44 +00:00
|
|
|
__m256i result = _mm256_blend_epi16(xa, xb, 0x66);
|
2020-05-25 15:16:19 +00:00
|
|
|
return reinterpret_cast<UInt64x4>(result);
|
2020-05-24 12:25:07 +00:00
|
|
|
}
|
2020-05-20 12:42:21 +00:00
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
void RandImpl::execute(char * output, size_t size)
|
|
|
|
{
|
2020-05-19 10:54:41 +00:00
|
|
|
if (size == 0)
|
|
|
|
return;
|
2020-05-20 16:44:01 +00:00
|
|
|
|
2020-05-19 10:54:41 +00:00
|
|
|
char * end = output + size;
|
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
constexpr int vec_size = 4;
|
2020-05-19 10:54:41 +00:00
|
|
|
constexpr int safe_overwrite = 15;
|
2020-05-24 12:25:07 +00:00
|
|
|
constexpr int bytes_per_write = 4 * sizeof(UInt64x4);
|
2020-05-20 12:42:21 +00:00
|
|
|
|
|
|
|
UInt64 rand_seed = randomSeed();
|
2020-05-20 15:43:01 +00:00
|
|
|
|
|
|
|
UInt64 a = LinearCongruentialGenerator::a;
|
|
|
|
// TODO(dakovalkov): try to remove this.
|
|
|
|
/// Note: GCC likes to expand multiplication by a constant into shifts + additions.
|
|
|
|
/// In this case a few multiplications become tens of shifts and additions. That leads to a huge slow down.
|
|
|
|
/// To avoid it we pretend that 'a' is not a constant. Actually we hope that rand_seed is never 0.
|
|
|
|
if (rand_seed == 0)
|
|
|
|
a = LinearCongruentialGenerator::a + 2;
|
|
|
|
|
|
|
|
constexpr UInt64 c = LinearCongruentialGenerator::c;
|
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
UInt64x4 gens1{};
|
|
|
|
UInt64x4 gens2{};
|
|
|
|
UInt64x4 gens3{};
|
|
|
|
UInt64x4 gens4{};
|
2020-05-25 15:16:19 +00:00
|
|
|
|
2020-05-24 12:25:07 +00:00
|
|
|
for (int i = 0; i < vec_size; ++i)
|
2020-05-20 12:42:21 +00:00
|
|
|
{
|
2020-05-20 15:43:01 +00:00
|
|
|
gens1[i] = calcSeed(rand_seed, random_numbers[i] + reinterpret_cast<intptr_t>(output));
|
2020-05-24 12:25:07 +00:00
|
|
|
gens2[i] = calcSeed(rand_seed, random_numbers[i + vec_size] + reinterpret_cast<intptr_t>(output));
|
|
|
|
gens3[i] = calcSeed(rand_seed, random_numbers[i + 2 * vec_size] + reinterpret_cast<intptr_t>(output));
|
|
|
|
gens4[i] = calcSeed(rand_seed, random_numbers[i + 3 * vec_size] + reinterpret_cast<intptr_t>(output));
|
2020-05-20 12:42:21 +00:00
|
|
|
}
|
2020-05-19 10:54:41 +00:00
|
|
|
|
|
|
|
while ((end - output) + safe_overwrite >= bytes_per_write)
|
|
|
|
{
|
2020-05-20 15:43:01 +00:00
|
|
|
gens1 = gens1 * a + c;
|
|
|
|
gens2 = gens2 * a + c;
|
2020-06-21 18:47:16 +00:00
|
|
|
unalignedStore<UInt64x4>(output, combineValues(gens1, gens2));
|
2020-05-24 12:25:07 +00:00
|
|
|
gens3 = gens3 * a + c;
|
|
|
|
gens4 = gens4 * a + c;
|
2020-06-21 18:47:16 +00:00
|
|
|
unalignedStore<UInt64x4>(output + sizeof(UInt64x4), combineValues(gens3, gens4));
|
2020-05-24 12:25:07 +00:00
|
|
|
gens1 = gens1 * a + c;
|
|
|
|
gens2 = gens2 * a + c;
|
2020-06-21 18:47:16 +00:00
|
|
|
unalignedStore<UInt64x4>(output + 2 * sizeof(UInt64x4), combineValues(gens1, gens2));
|
2020-05-20 15:43:01 +00:00
|
|
|
gens3 = gens3 * a + c;
|
|
|
|
gens4 = gens4 * a + c;
|
2020-06-21 18:47:16 +00:00
|
|
|
unalignedStore<UInt64x4>(output + 3 * sizeof(UInt64x4), combineValues(gens3, gens4));
|
2020-05-19 10:54:41 +00:00
|
|
|
output += bytes_per_write;
|
|
|
|
}
|
2020-05-20 15:43:01 +00:00
|
|
|
|
|
|
|
// Process tail
|
2020-05-20 12:42:21 +00:00
|
|
|
while ((end - output) > 0)
|
2020-05-19 10:54:41 +00:00
|
|
|
{
|
2020-05-24 12:25:07 +00:00
|
|
|
gens1 = gens1 * a + c;
|
|
|
|
gens2 = gens2 * a + c;
|
2020-06-21 18:47:16 +00:00
|
|
|
UInt64x4 values = combineValues(gens1, gens2);
|
2020-05-24 12:25:07 +00:00
|
|
|
for (int i = 0; i < vec_size && (end - output) > 0; ++i)
|
2020-05-19 10:54:41 +00:00
|
|
|
{
|
2020-05-24 12:25:07 +00:00
|
|
|
unalignedStore<UInt64>(output, values[i]);
|
|
|
|
output += sizeof(UInt64);
|
2020-05-19 10:54:41 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-05-20 15:43:01 +00:00
|
|
|
) // DECLARE_AVX2_SPECIFIC_CODE
|
2020-05-20 12:43:33 +00:00
|
|
|
|
2014-08-22 00:57:20 +00:00
|
|
|
}
|