[libcxx-commits] [libcxx] [libc++] Implement std::gcd using the binary version (PR #77747)
via libcxx-commits
libcxx-commits at lists.llvm.org
Thu Jan 11 02:29:35 PST 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-libcxx
Author: None (serge-sans-paille)
<details>
<summary>Changes</summary>
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
---
Full diff: https://github.com/llvm/llvm-project/pull/77747.diff
4 Files Affected:
- (modified) libcxx/include/__bit/countr.h (+8-3)
- (modified) libcxx/include/__numeric/gcd_lcm.h (+21-3)
- (modified) libcxx/test/libcxx/transitive_includes/cxx26.csv (+23)
- (modified) libcxx/test/std/numerics/numeric.ops/numeric.ops.gcd/gcd.pass.cpp (+26)
``````````diff
diff --git a/libcxx/include/__bit/countr.h b/libcxx/include/__bit/countr.h
index 0cc679f87a99d9..e47b8245be3cea 100644
--- a/libcxx/include/__bit/countr.h
+++ b/libcxx/include/__bit/countr.h
@@ -35,10 +35,8 @@ _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR int __libcpp_ct
return __builtin_ctzll(__x);
}
-#if _LIBCPP_STD_VER >= 20
-
template <__libcpp_unsigned_integer _Tp>
-_LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr int countr_zero(_Tp __t) noexcept {
+_LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr int __countr_zero(_Tp __t) noexcept {
if (__t == 0)
return numeric_limits<_Tp>::digits;
@@ -59,6 +57,13 @@ _LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr int countr_zero(_Tp __t) n
}
}
+#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 __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..55199547fa4db2 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) {
- static_assert((!is_signed<_Tp>::value), "");
- return __n == 0 ? __m : std::__gcd<_Tp>(__n, __m % __n);
+_LIBCPP_CONSTEXPR _LIBCPP_HIDDEN _Tp __gcd(_Tp __a, _Tp __b) {
+ static_assert((!is_signed<_Tp>::value), "");
+ 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/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..0435575bd8759a 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
@@ -17,6 +17,7 @@
#include <cassert>
#include <climits>
#include <cstdint>
+#include <random>
#include <type_traits>
#include "test_macros.h"
@@ -48,6 +49,26 @@ 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 Input>
+void do_fuzzy_tests() {
+
+ std::random_device rd;
+ std::mt19937 gen(rd());
+ 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 Input1, typename Input2 = Input1>
constexpr bool do_test(int = 0)
@@ -143,5 +164,10 @@ int main(int argc, char**)
assert(res == 2);
}
+ do_fuzzy_tests<std::uint8_t>();
+ do_fuzzy_tests<std::uint16_t>();
+ do_fuzzy_tests<std::uint32_t>();
+ do_fuzzy_tests<std::uint64_t>();
+
return 0;
}
``````````
</details>
https://github.com/llvm/llvm-project/pull/77747
More information about the libcxx-commits
mailing list