d1c9a2b06f
Rationale: Mostly an AOT 64-bit improvement of DIV and MOV by constant on X64_64. Also a few minor supporting optimizations that were missed. With unit tests. https://github.com/dart-lang/sdk/issues/37789 Change-Id: I2f12d6a24bbe983324521d7d4d778f20b34d18a1 Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/114901 Commit-Queue: Aart Bik <ajcbik@google.com> Reviewed-by: Ryan Macnak <rmacnak@google.com>
201 lines
5.7 KiB
C++
201 lines
5.7 KiB
C++
// Copyright (c) 2012, the Dart project authors. Please see the AUTHORS file
|
|
// for details. All rights reserved. Use of this source code is governed by a
|
|
// BSD-style license that can be found in the LICENSE file.
|
|
|
|
#include "platform/utils.h"
|
|
|
|
namespace dart {
|
|
|
|
// Implementation is from "Hacker's Delight" by Henry S. Warren, Jr.,
|
|
// figure 3-3, page 48, where the function is called clp2.
|
|
uintptr_t Utils::RoundUpToPowerOfTwo(uintptr_t x) {
|
|
x = x - 1;
|
|
x = x | (x >> 1);
|
|
x = x | (x >> 2);
|
|
x = x | (x >> 4);
|
|
x = x | (x >> 8);
|
|
x = x | (x >> 16);
|
|
#if defined(ARCH_IS_64_BIT)
|
|
x = x | (x >> 32);
|
|
#endif // defined(ARCH_IS_64_BIT)
|
|
return x + 1;
|
|
}
|
|
|
|
// TODO(koda): Compare to flsll call/intrinsic.
|
|
int Utils::HighestBit(int64_t v) {
|
|
uint64_t x = static_cast<uint64_t>((v > 0) ? v : -v);
|
|
uint64_t t;
|
|
int r = 0;
|
|
if ((t = x >> 32) != 0) {
|
|
x = t;
|
|
r += 32;
|
|
}
|
|
if ((t = x >> 16) != 0) {
|
|
x = t;
|
|
r += 16;
|
|
}
|
|
if ((t = x >> 8) != 0) {
|
|
x = t;
|
|
r += 8;
|
|
}
|
|
if ((t = x >> 4) != 0) {
|
|
x = t;
|
|
r += 4;
|
|
}
|
|
if ((t = x >> 2) != 0) {
|
|
x = t;
|
|
r += 2;
|
|
}
|
|
if (x > 1) r += 1;
|
|
return r;
|
|
}
|
|
|
|
// Implementation according to H.S.Warren's "Hacker's Delight"
|
|
// (Addison Wesley, 2002) Chapter 10 and T.Grablund, P.L.Montogomery's
|
|
// "Division by Invariant Integers Using Multiplication" (PLDI 1994).
|
|
void Utils::CalculateMagicAndShiftForDivRem(int64_t divisor,
|
|
int64_t* magic,
|
|
int64_t* shift) {
|
|
ASSERT(divisor <= -2 || divisor >= 2);
|
|
/* The magic number M and shift S can be calculated in the following way:
|
|
* Let nc be the most positive value of numerator(n) such that nc = kd - 1,
|
|
* where divisor(d) >= 2.
|
|
* Let nc be the most negative value of numerator(n) such that nc = kd + 1,
|
|
* where divisor(d) <= -2.
|
|
* Thus nc can be calculated like:
|
|
* nc = exp + exp % d - 1, where d >= 2 and exp = 2^63.
|
|
* nc = -exp + (exp + 1) % d, where d >= 2 and exp = 2^63.
|
|
*
|
|
* So the shift p is the smallest p satisfying
|
|
* 2^p > nc * (d - 2^p % d), where d >= 2
|
|
* 2^p > nc * (d + 2^p % d), where d <= -2.
|
|
*
|
|
* The magic number M is calculated by
|
|
* M = (2^p + d - 2^p % d) / d, where d >= 2
|
|
* M = (2^p - d - 2^p % d) / d, where d <= -2.
|
|
*/
|
|
int64_t p = 63;
|
|
const uint64_t exp = 1LL << 63;
|
|
|
|
// Initialize the computations.
|
|
uint64_t abs_d = (divisor >= 0) ? divisor : -divisor;
|
|
uint64_t sign_bit = static_cast<uint64_t>(divisor) >> 63;
|
|
uint64_t tmp = exp + sign_bit;
|
|
uint64_t abs_nc = tmp - 1 - (tmp % abs_d);
|
|
uint64_t quotient1 = exp / abs_nc;
|
|
uint64_t remainder1 = exp % abs_nc;
|
|
uint64_t quotient2 = exp / abs_d;
|
|
uint64_t remainder2 = exp % abs_d;
|
|
|
|
// To avoid handling both positive and negative divisor,
|
|
// "Hacker's Delight" introduces a method to handle these
|
|
// two cases together to avoid duplication.
|
|
uint64_t delta;
|
|
do {
|
|
p++;
|
|
quotient1 = 2 * quotient1;
|
|
remainder1 = 2 * remainder1;
|
|
if (remainder1 >= abs_nc) {
|
|
quotient1++;
|
|
remainder1 = remainder1 - abs_nc;
|
|
}
|
|
quotient2 = 2 * quotient2;
|
|
remainder2 = 2 * remainder2;
|
|
if (remainder2 >= abs_d) {
|
|
quotient2++;
|
|
remainder2 = remainder2 - abs_d;
|
|
}
|
|
delta = abs_d - remainder2;
|
|
} while (quotient1 < delta || (quotient1 == delta && remainder1 == 0));
|
|
|
|
*magic = (divisor > 0) ? (quotient2 + 1) : (-quotient2 - 1);
|
|
*shift = p - 64;
|
|
}
|
|
|
|
uint32_t Utils::StringHash(const char* data, int length) {
|
|
// This implementation is based on the public domain MurmurHash
|
|
// version 2.0. It assumes that the underlying CPU can read from
|
|
// unaligned addresses. The constants M and R have been determined
|
|
// to work well experimentally.
|
|
// TODO(3158902): need to account for unaligned address access on ARM.
|
|
const uint32_t M = 0x5bd1e995;
|
|
const int R = 24;
|
|
int size = length;
|
|
uint32_t hash = size;
|
|
|
|
// Mix four bytes at a time into the hash.
|
|
const uint8_t* cursor = reinterpret_cast<const uint8_t*>(data);
|
|
while (size >= 4) {
|
|
uint32_t part = *reinterpret_cast<const uint32_t*>(cursor);
|
|
part *= M;
|
|
part ^= part >> R;
|
|
part *= M;
|
|
hash *= M;
|
|
hash ^= part;
|
|
cursor += 4;
|
|
size -= 4;
|
|
}
|
|
|
|
// Handle the last few bytes of the string.
|
|
switch (size) {
|
|
case 3:
|
|
hash ^= cursor[2] << 16;
|
|
FALL_THROUGH;
|
|
case 2:
|
|
hash ^= cursor[1] << 8;
|
|
FALL_THROUGH;
|
|
case 1:
|
|
hash ^= cursor[0];
|
|
hash *= M;
|
|
}
|
|
|
|
// Do a few final mixes of the hash to ensure the last few bytes are
|
|
// well-incorporated.
|
|
hash ^= hash >> 13;
|
|
hash *= M;
|
|
hash ^= hash >> 15;
|
|
return hash;
|
|
}
|
|
|
|
uint32_t Utils::WordHash(intptr_t key) {
|
|
// TODO(iposva): Need to check hash spreading.
|
|
// This example is from http://www.concentric.net/~Ttwang/tech/inthash.htm
|
|
// via. http://web.archive.org/web/20071223173210/http://www.concentric.net/~Ttwang/tech/inthash.htm
|
|
uword a = static_cast<uword>(key);
|
|
a = (a + 0x7ed55d16) + (a << 12);
|
|
a = (a ^ 0xc761c23c) ^ (a >> 19);
|
|
a = (a + 0x165667b1) + (a << 5);
|
|
a = (a + 0xd3a2646c) ^ (a << 9);
|
|
a = (a + 0xfd7046c5) + (a << 3);
|
|
a = (a ^ 0xb55a4f09) ^ (a >> 16);
|
|
return static_cast<uint32_t>(a);
|
|
}
|
|
|
|
char* Utils::SCreate(const char* format, ...) {
|
|
va_list args;
|
|
va_start(args, format);
|
|
char* buffer = VSCreate(format, args);
|
|
va_end(args);
|
|
return buffer;
|
|
}
|
|
|
|
char* Utils::VSCreate(const char* format, va_list args) {
|
|
// Measure.
|
|
va_list measure_args;
|
|
va_copy(measure_args, args);
|
|
intptr_t len = VSNPrint(NULL, 0, format, measure_args);
|
|
va_end(measure_args);
|
|
|
|
char* buffer = reinterpret_cast<char*>(malloc(len + 1));
|
|
ASSERT(buffer != NULL);
|
|
|
|
// Print.
|
|
va_list print_args;
|
|
va_copy(print_args, args);
|
|
VSNPrint(buffer, len + 1, format, print_args);
|
|
va_end(print_args);
|
|
return buffer;
|
|
}
|
|
|
|
} // namespace dart
|