[libcxx-commits] [libcxx] [libc++] Implement std::gcd using the binary version (PR #77747)
Amir Ayupov via libcxx-commits
libcxx-commits at lists.llvm.org
Thu Jan 11 12:15:12 PST 2024
https://github.com/aaupov updated https://github.com/llvm/llvm-project/pull/77747
>From 6749aa2270f2f063def04263bd30b24eb2a1ec66 Mon Sep 17 00:00:00 2001
From: serge-sans-paille <sguelton at mozilla.com>
Date: Mon, 13 Mar 2023 21:06:01 +0100
Subject: [PATCH] [libc++] Implement std::gcd using the binary version
The binary version is four times faster than current implementation
in my setup, and generally considered a better implementation.
Code inspired by https://en.algorithmica.org/hpc/algorithms/gcd/
which itself is inspired by https://lemire.me/blog/2013/12/26/fastest-way-to-compute-the-greatest-common-divisor/
Fix #77648
---
libcxx/include/__bit/countr.h | 17 +++-
libcxx/include/__numeric/gcd_lcm.h | 22 ++++-
.../test/libcxx/transitive_includes/cxx03.csv | 1 +
.../test/libcxx/transitive_includes/cxx11.csv | 1 +
.../test/libcxx/transitive_includes/cxx14.csv | 1 +
.../test/libcxx/transitive_includes/cxx17.csv | 1 +
.../test/libcxx/transitive_includes/cxx20.csv | 1 +
.../test/libcxx/transitive_includes/cxx26.csv | 23 ++++++
.../numeric.ops/numeric.ops.gcd/gcd.pass.cpp | 80 ++++++++++++++++++-
9 files changed, 141 insertions(+), 6 deletions(-)
diff --git a/libcxx/include/__bit/countr.h b/libcxx/include/__bit/countr.h
index 0cc679f87a99d9..93f1a06dc88970 100644
--- a/libcxx/include/__bit/countr.h
+++ b/libcxx/include/__bit/countr.h
@@ -12,6 +12,7 @@
#include <__bit/rotate.h>
#include <__concepts/arithmetic.h>
#include <__config>
+#include <__type_traits/is_unsigned_integer.h>
#include <limits>
#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
@@ -35,10 +36,11 @@ _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_ct
return __builtin_ctzll(__x);
}
-#if _LIBCPP_STD_VER >= 20
+#if _LIBCPP_STD_VER >= 17
-template <__libcpp_unsigned_integer _Tp>
-_LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr int countr_zero(_Tp __t) noexcept {
+template <typename _Tp>
+_LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __countr_zero(_Tp __t) _NOEXCEPT {
+ static_assert(__libcpp_is_unsigned_integer<_Tp>::value, "__countr_zero requires an unsigned integer type");
if (__t == 0)
return numeric_limits<_Tp>::digits;
@@ -59,6 +61,15 @@ _LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr int countr_zero(_Tp __t) n
}
}
+#endif
+
+#if _LIBCPP_STD_VER >= 20
+
+template <__libcpp_unsigned_integer _Tp>
+_LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr int countr_zero(_Tp __t) noexcept {
+ return std::__countr_zero(__t);
+}
+
template <__libcpp_unsigned_integer _Tp>
_LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr int countr_one(_Tp __t) noexcept {
return __t != numeric_limits<_Tp>::max() ? std::countr_zero(static_cast<_Tp>(~__t)) : numeric_limits<_Tp>::digits;
diff --git a/libcxx/include/__numeric/gcd_lcm.h b/libcxx/include/__numeric/gcd_lcm.h
index 3e9c244f25c285..bcf2fbe688445e 100644
--- a/libcxx/include/__numeric/gcd_lcm.h
+++ b/libcxx/include/__numeric/gcd_lcm.h
@@ -10,7 +10,9 @@
#ifndef _LIBCPP___NUMERIC_GCD_LCM_H
#define _LIBCPP___NUMERIC_GCD_LCM_H
+#include <__algorithm/min.h>
#include <__assert>
+#include <__bit/countr.h>
#include <__config>
#include <__type_traits/common_type.h>
#include <__type_traits/is_integral.h>
@@ -50,9 +52,25 @@ struct __ct_abs<_Result, _Source, false> {
};
template <class _Tp>
-_LIBCPP_CONSTEXPR _LIBCPP_HIDDEN _Tp __gcd(_Tp __m, _Tp __n) {
+_LIBCPP_CONSTEXPR _LIBCPP_HIDDEN _Tp __gcd(_Tp __a, _Tp __b) {
static_assert((!is_signed<_Tp>::value), "");
- return __n == 0 ? __m : std::__gcd<_Tp>(__n, __m % __n);
+ if (__a == 0)
+ return __b;
+ if (__b == 0)
+ return __a;
+
+ int __az = std::__countr_zero(__a);
+ int __bz = std::__countr_zero(__b);
+ int __shift = std::min(__az, __bz);
+ __b >>= __bz;
+ while (__a != 0) {
+ __a >>= __az;
+ _Tp __absdiff = __a > __b ? __a - __b : __b - __a;
+ __b = std::min(__a, __b);
+ __a = __absdiff;
+ __az = std::__countr_zero(__absdiff);
+ }
+ return __b << __shift;
}
template <class _Tp, class _Up>
diff --git a/libcxx/test/libcxx/transitive_includes/cxx03.csv b/libcxx/test/libcxx/transitive_includes/cxx03.csv
index e709628ce2311b..731513a005d7fc 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx03.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx03.csv
@@ -549,6 +549,7 @@ numeric cstddef
numeric cstdint
numeric execution
numeric functional
+numeric initializer_list
numeric iterator
numeric limits
numeric new
diff --git a/libcxx/test/libcxx/transitive_includes/cxx11.csv b/libcxx/test/libcxx/transitive_includes/cxx11.csv
index d3ea6ed97367e4..91e70f3f7185fe 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx11.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx11.csv
@@ -554,6 +554,7 @@ numeric cstddef
numeric cstdint
numeric execution
numeric functional
+numeric initializer_list
numeric iterator
numeric limits
numeric new
diff --git a/libcxx/test/libcxx/transitive_includes/cxx14.csv b/libcxx/test/libcxx/transitive_includes/cxx14.csv
index 3e56017bfb537b..40f1f8483104a7 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx14.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx14.csv
@@ -556,6 +556,7 @@ numeric cstddef
numeric cstdint
numeric execution
numeric functional
+numeric initializer_list
numeric iterator
numeric limits
numeric new
diff --git a/libcxx/test/libcxx/transitive_includes/cxx17.csv b/libcxx/test/libcxx/transitive_includes/cxx17.csv
index 3e56017bfb537b..40f1f8483104a7 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx17.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx17.csv
@@ -556,6 +556,7 @@ numeric cstddef
numeric cstdint
numeric execution
numeric functional
+numeric initializer_list
numeric iterator
numeric limits
numeric new
diff --git a/libcxx/test/libcxx/transitive_includes/cxx20.csv b/libcxx/test/libcxx/transitive_includes/cxx20.csv
index 0c5b9721a22551..2ae88dbea585c9 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx20.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx20.csv
@@ -561,6 +561,7 @@ numeric cstddef
numeric cstdint
numeric execution
numeric functional
+numeric initializer_list
numeric iterator
numeric limits
numeric new
diff --git a/libcxx/test/libcxx/transitive_includes/cxx26.csv b/libcxx/test/libcxx/transitive_includes/cxx26.csv
index 0d8c3fa21b2f37..e3d325169a3fa8 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx26.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx26.csv
@@ -173,6 +173,29 @@ experimental/simd limits
experimental/type_traits initializer_list
experimental/type_traits type_traits
experimental/utility utility
+experimental/vector experimental/memory_resource
+experimental/vector vector
+ext/hash_map algorithm
+ext/hash_map cmath
+ext/hash_map cstddef
+ext/hash_map cstdint
+ext/hash_map cstring
+ext/hash_map functional
+ext/hash_map initializer_list
+ext/hash_map limits
+ext/hash_map new
+ext/hash_map stdexcept
+ext/hash_map string
+ext/hash_set algorithm
+ext/hash_set cmath
+ext/hash_set cstddef
+ext/hash_set cstdint
+ext/hash_set cstring
+ext/hash_set functional
+ext/hash_set initializer_list
+ext/hash_set limits
+ext/hash_set new
+ext/hash_set string
filesystem compare
filesystem cstddef
filesystem cstdint
diff --git a/libcxx/test/std/numerics/numeric.ops/numeric.ops.gcd/gcd.pass.cpp b/libcxx/test/std/numerics/numeric.ops/numeric.ops.gcd/gcd.pass.cpp
index 831c226f9c8ea1..5bcede1bf5d307 100644
--- a/libcxx/test/std/numerics/numeric.ops/numeric.ops.gcd/gcd.pass.cpp
+++ b/libcxx/test/std/numerics/numeric.ops/numeric.ops.gcd/gcd.pass.cpp
@@ -17,7 +17,9 @@
#include <cassert>
#include <climits>
#include <cstdint>
+#include <random>
#include <type_traits>
+#include <iostream>
#include "test_macros.h"
@@ -48,6 +50,64 @@ constexpr bool test0(int in1, int in2, int out)
return true;
}
+template <typename T>
+T basic_gcd_(T m, T n) {
+ return n == 0 ? m : basic_gcd_<T>(n, m % n);
+}
+
+template <typename T>
+T basic_gcd(T m, T n) {
+ using Tp = std::make_unsigned_t<T>;
+ return basic_gcd_(static_cast<Tp>(m < 0 ? -m : m), static_cast<Tp>(n < 0 ? -n : n));
+}
+
+template <typename Input>
+void do_fuzzy_tests() {
+ std::mt19937 gen(1938);
+ std::uniform_int_distribution<Input> distrib;
+
+ constexpr int nb_rounds = 10000;
+ for (int i = 0; i < nb_rounds; ++i) {
+ Input n = distrib(gen);
+ Input m = distrib(gen);
+ assert(std::gcd(n, m) == basic_gcd(n, m));
+ }
+}
+
+template <typename Input>
+void do_limit_tests() {
+ Input inputs[] = {
+ std::numeric_limits<Input>::min(),
+ std::numeric_limits<Input>::max(),
+ 0,
+ 1,
+ 2,
+ 3,
+ 4,
+ 5,
+ 6,
+ 7,
+ 8,
+ 9,
+ 10,
+ (Input)-1,
+ (Input)-2,
+ (Input)-3,
+ (Input)-4,
+ (Input)-5,
+ (Input)-6,
+ (Input)-7,
+ (Input)-8,
+ (Input)-9,
+ (Input)-10,
+ };
+
+ for (auto n : inputs) {
+ for (auto m : inputs) {
+ assert(std::gcd(n, m) == basic_gcd(n, m));
+ }
+ }
+}
template <typename Input1, typename Input2 = Input1>
constexpr bool do_test(int = 0)
@@ -143,5 +203,23 @@ int main(int argc, char**)
assert(res == 2);
}
- return 0;
+ do_fuzzy_tests<std::int8_t>();
+ do_fuzzy_tests<std::int16_t>();
+ do_fuzzy_tests<std::int32_t>();
+ do_fuzzy_tests<std::int64_t>();
+ do_fuzzy_tests<std::uint8_t>();
+ do_fuzzy_tests<std::uint16_t>();
+ do_fuzzy_tests<std::uint32_t>();
+ do_fuzzy_tests<std::uint64_t>();
+
+ do_limit_tests<std::int8_t>();
+ do_limit_tests<std::int16_t>();
+ do_limit_tests<std::int32_t>();
+ do_limit_tests<std::int64_t>();
+ do_limit_tests<std::uint8_t>();
+ do_limit_tests<std::uint16_t>();
+ do_limit_tests<std::uint32_t>();
+ do_limit_tests<std::uint64_t>();
+
+ return 0;
}
More information about the libcxx-commits
mailing list