[libc++] Simplify the implementation of std::sort a bit #104902
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
philnik777
force-pushed
the
simplify_sort
branch
2 times, most recently
from
58a8d95 to
d90d328
Compare
philnik777
added
the
libc++
|
@llvm/pr-subscribers-libcxx Author: Nikolas Klauser (philnik777) ChangesPatch is 21.80 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/104902.diff 8 Files Affected:
diff --git a/libcxx/include/__algorithm/comp.h b/libcxx/include/__algorithm/comp.h
index 1f38f5d2d99b43..ab3c598418828a 100644
--- a/libcxx/include/__algorithm/comp.h
+++ b/libcxx/include/__algorithm/comp.h
@@ -42,6 +42,9 @@ struct __less<void, void> {
}
};
+template <class _Tp>
+inline const bool __desugars_to_v<__less_tag, __less<>, _Tp, _Tp> = true;
+
template <class _Tp>
inline const bool __desugars_to_v<__totally_ordered_less_tag, __less<>, _Tp, _Tp> = is_integral<_Tp>::value;
diff --git a/libcxx/include/__algorithm/ranges_minmax.h b/libcxx/include/__algorithm/ranges_minmax.h
index 9b8551d2213400..1b43b1e19cdec9 100644
--- a/libcxx/include/__algorithm/ranges_minmax.h
+++ b/libcxx/include/__algorithm/ranges_minmax.h
@@ -88,7 +88,7 @@ struct __minmax {
// vectorize the code.
if constexpr (contiguous_range<_Range> && is_integral_v<_ValueT> &&
__is_cheap_to_copy<_ValueT> & __is_identity<_Proj>::value &&
- __desugars_to_v<__totally_ordered_less_tag, _Comp, _ValueT, _ValueT>) {
+ __desugars_to_v<__less_tag, _Comp, _ValueT, _ValueT>) {
minmax_result<_ValueT> __result = {__r[0], __r[0]};
for (auto __e : __r) {
if (__e < __result.min)
diff --git a/libcxx/include/__algorithm/sort.h b/libcxx/include/__algorithm/sort.h
index 07b5814639e9e4..77ff1533c79492 100644
--- a/libcxx/include/__algorithm/sort.h
+++ b/libcxx/include/__algorithm/sort.h
@@ -27,9 +27,11 @@
#include <__functional/ranges_operations.h>
#include <__iterator/iterator_traits.h>
#include <__type_traits/conditional.h>
+#include <__type_traits/desugars_to.h>
#include <__type_traits/disjunction.h>
#include <__type_traits/is_arithmetic.h>
#include <__type_traits/is_constant_evaluated.h>
+#include <__type_traits/is_trivially_copyable.h>
#include <__utility/move.h>
#include <__utility/pair.h>
#include <climits>
@@ -44,110 +46,11 @@ _LIBCPP_PUSH_MACROS
_LIBCPP_BEGIN_NAMESPACE_STD
-// stable, 2-3 compares, 0-2 swaps
-
-template <class _AlgPolicy, class _Compare, class _ForwardIterator>
-_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 unsigned
-__sort3(_ForwardIterator __x, _ForwardIterator __y, _ForwardIterator __z, _Compare __c) {
- using _Ops = _IterOps<_AlgPolicy>;
-
- unsigned __r = 0;
- if (!__c(*__y, *__x)) // if x <= y
- {
- if (!__c(*__z, *__y)) // if y <= z
- return __r; // x <= y && y <= z
- // x <= y && y > z
- _Ops::iter_swap(__y, __z); // x <= z && y < z
- __r = 1;
- if (__c(*__y, *__x)) // if x > y
- {
- _Ops::iter_swap(__x, __y); // x < y && y <= z
- __r = 2;
- }
- return __r; // x <= y && y < z
- }
- if (__c(*__z, *__y)) // x > y, if y > z
- {
- _Ops::iter_swap(__x, __z); // x < y && y < z
- __r = 1;
- return __r;
- }
- _Ops::iter_swap(__x, __y); // x > y && y <= z
- __r = 1; // x < y && x <= z
- if (__c(*__z, *__y)) // if y > z
- {
- _Ops::iter_swap(__y, __z); // x <= y && y < z
- __r = 2;
- }
- return __r;
-} // x <= y && y <= z
-
-// stable, 3-6 compares, 0-5 swaps
-
-template <class _AlgPolicy, class _Compare, class _ForwardIterator>
-_LIBCPP_HIDE_FROM_ABI void
-__sort4(_ForwardIterator __x1, _ForwardIterator __x2, _ForwardIterator __x3, _ForwardIterator __x4, _Compare __c) {
- using _Ops = _IterOps<_AlgPolicy>;
- std::__sort3<_AlgPolicy, _Compare>(__x1, __x2, __x3, __c);
- if (__c(*__x4, *__x3)) {
- _Ops::iter_swap(__x3, __x4);
- if (__c(*__x3, *__x2)) {
- _Ops::iter_swap(__x2, __x3);
- if (__c(*__x2, *__x1)) {
- _Ops::iter_swap(__x1, __x2);
- }
- }
- }
-}
-
-// stable, 4-10 compares, 0-9 swaps
-
-template <class _AlgPolicy, class _Comp, class _ForwardIterator>
-_LIBCPP_HIDE_FROM_ABI void
-__sort5(_ForwardIterator __x1,
- _ForwardIterator __x2,
- _ForwardIterator __x3,
- _ForwardIterator __x4,
- _ForwardIterator __x5,
- _Comp __comp) {
- using _Ops = _IterOps<_AlgPolicy>;
-
- std::__sort4<_AlgPolicy, _Comp>(__x1, __x2, __x3, __x4, __comp);
- if (__comp(*__x5, *__x4)) {
- _Ops::iter_swap(__x4, __x5);
- if (__comp(*__x4, *__x3)) {
- _Ops::iter_swap(__x3, __x4);
- if (__comp(*__x3, *__x2)) {
- _Ops::iter_swap(__x2, __x3);
- if (__comp(*__x2, *__x1)) {
- _Ops::iter_swap(__x1, __x2);
- }
- }
- }
- }
-}
-
-// The comparator being simple is a prerequisite for using the branchless optimization.
-template <class _Tp>
-struct __is_simple_comparator : false_type {};
-template <>
-struct __is_simple_comparator<__less<>&> : true_type {};
-template <class _Tp>
-struct __is_simple_comparator<less<_Tp>&> : true_type {};
-template <class _Tp>
-struct __is_simple_comparator<greater<_Tp>&> : true_type {};
-#if _LIBCPP_STD_VER >= 20
-template <>
-struct __is_simple_comparator<ranges::less&> : true_type {};
-template <>
-struct __is_simple_comparator<ranges::greater&> : true_type {};
-#endif
-
template <class _Compare, class _Iter, class _Tp = typename iterator_traits<_Iter>::value_type>
-using __use_branchless_sort =
- integral_constant<bool,
- __libcpp_is_contiguous_iterator<_Iter>::value && sizeof(_Tp) <= sizeof(void*) &&
- is_arithmetic<_Tp>::value && __is_simple_comparator<_Compare>::value>;
+inline const bool __use_branchless_sort =
+ __libcpp_is_contiguous_iterator<_Iter>::value && __is_cheap_to_copy<_Tp> && is_arithmetic<_Tp>::value &&
+ (__desugars_to_v<__less_tag, __remove_cvref_t<_Compare>, _Tp, _Tp> ||
+ __desugars_to_v<__greater_tag, __remove_cvref_t<_Compare>, _Tp, _Tp>);
namespace __detail {
@@ -158,59 +61,88 @@ enum { __block_size = sizeof(uint64_t) * 8 };
// Ensures that __c(*__x, *__y) is true by swapping *__x and *__y if necessary.
template <class _Compare, class _RandomAccessIterator>
-inline _LIBCPP_HIDE_FROM_ABI void __cond_swap(_RandomAccessIterator __x, _RandomAccessIterator __y, _Compare __c) {
+inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 bool
+__cond_swap(_RandomAccessIterator __x, _RandomAccessIterator __y, _Compare __c) {
// Note: this function behaves correctly even with proxy iterators (because it relies on `value_type`).
using value_type = typename iterator_traits<_RandomAccessIterator>::value_type;
bool __r = __c(*__x, *__y);
value_type __tmp = __r ? *__x : *__y;
*__y = __r ? *__y : *__x;
*__x = __tmp;
+ return !__r;
}
// Ensures that *__x, *__y and *__z are ordered according to the comparator __c,
// under the assumption that *__y and *__z are already ordered.
template <class _Compare, class _RandomAccessIterator>
-inline _LIBCPP_HIDE_FROM_ABI void
+inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 bool
__partially_sorted_swap(_RandomAccessIterator __x, _RandomAccessIterator __y, _RandomAccessIterator __z, _Compare __c) {
// Note: this function behaves correctly even with proxy iterators (because it relies on `value_type`).
using value_type = typename iterator_traits<_RandomAccessIterator>::value_type;
- bool __r = __c(*__z, *__x);
- value_type __tmp = __r ? *__z : *__x;
- *__z = __r ? *__x : *__z;
- __r = __c(__tmp, *__y);
- *__x = __r ? *__x : *__y;
- *__y = __r ? *__y : __tmp;
+ bool __r1 = __c(*__z, *__x);
+ value_type __tmp = __r1 ? *__z : *__x;
+ *__z = __r1 ? *__x : *__z;
+ bool __r2 = __c(__tmp, *__y);
+ *__x = __r2 ? *__x : *__y;
+ *__y = __r2 ? *__y : __tmp;
+ return !__r1 || !__r2;
}
+// stable, 2-3 compares, 0-2 swaps
+
template <class,
class _Compare,
class _RandomAccessIterator,
- __enable_if_t<__use_branchless_sort<_Compare, _RandomAccessIterator>::value, int> = 0>
-inline _LIBCPP_HIDE_FROM_ABI void __sort3_maybe_branchless(
- _RandomAccessIterator __x1, _RandomAccessIterator __x2, _RandomAccessIterator __x3, _Compare __c) {
- std::__cond_swap<_Compare>(__x2, __x3, __c);
- std::__partially_sorted_swap<_Compare>(__x1, __x2, __x3, __c);
+ __enable_if_t<__use_branchless_sort<_Compare, _RandomAccessIterator>, int> = 0>
+inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 bool
+__sort3(_RandomAccessIterator __x1, _RandomAccessIterator __x2, _RandomAccessIterator __x3, _Compare __c) {
+ bool __swapped1 = std::__cond_swap<_Compare>(__x2, __x3, __c);
+ bool __swapped2 = std::__partially_sorted_swap<_Compare>(__x1, __x2, __x3, __c);
+ return __swapped1 || __swapped2;
}
template <class _AlgPolicy,
class _Compare,
class _RandomAccessIterator,
- __enable_if_t<!__use_branchless_sort<_Compare, _RandomAccessIterator>::value, int> = 0>
-inline _LIBCPP_HIDE_FROM_ABI void __sort3_maybe_branchless(
- _RandomAccessIterator __x1, _RandomAccessIterator __x2, _RandomAccessIterator __x3, _Compare __c) {
- std::__sort3<_AlgPolicy, _Compare>(__x1, __x2, __x3, __c);
-}
+ __enable_if_t<!__use_branchless_sort<_Compare, _RandomAccessIterator>, int> = 0>
+inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 bool
+__sort3(_RandomAccessIterator __x, _RandomAccessIterator __y, _RandomAccessIterator __z, _Compare __c) {
+ using _Ops = _IterOps<_AlgPolicy>;
+
+ if (!__c(*__y, *__x)) // if x <= y
+ {
+ if (!__c(*__z, *__y)) // if y <= z
+ return false; // x <= y && y <= z
+ // x <= y && y > z
+ _Ops::iter_swap(__y, __z); // x <= z && y < z
+ if (__c(*__y, *__x)) // if x > y
+ _Ops::iter_swap(__x, __y); // x < y && y <= z
+ return true; // x <= y && y < z
+ }
+ if (__c(*__z, *__y)) // x > y, if y > z
+ {
+ _Ops::iter_swap(__x, __z); // x < y && y < z
+ return true;
+ }
+ _Ops::iter_swap(__x, __y); // x > y && y <= z
+ // x < y && x <= z
+ if (__c(*__z, *__y)) // if y > z
+ _Ops::iter_swap(__y, __z); // x <= y && y < z
+ return true;
+} // x <= y && y <= z
+
+// stable, 3-6 compares, 0-5 swaps
template <class,
class _Compare,
class _RandomAccessIterator,
- __enable_if_t<__use_branchless_sort<_Compare, _RandomAccessIterator>::value, int> = 0>
-inline _LIBCPP_HIDE_FROM_ABI void __sort4_maybe_branchless(
- _RandomAccessIterator __x1,
- _RandomAccessIterator __x2,
- _RandomAccessIterator __x3,
- _RandomAccessIterator __x4,
- _Compare __c) {
+ __enable_if_t<__use_branchless_sort<_Compare, _RandomAccessIterator>, int> = 0>
+inline _LIBCPP_HIDE_FROM_ABI void
+__sort4(_RandomAccessIterator __x1,
+ _RandomAccessIterator __x2,
+ _RandomAccessIterator __x3,
+ _RandomAccessIterator __x4,
+ _Compare __c) {
std::__cond_swap<_Compare>(__x1, __x3, __c);
std::__cond_swap<_Compare>(__x2, __x4, __c);
std::__cond_swap<_Compare>(__x1, __x2, __c);
@@ -221,27 +153,39 @@ inline _LIBCPP_HIDE_FROM_ABI void __sort4_maybe_branchless(
template <class _AlgPolicy,
class _Compare,
class _RandomAccessIterator,
- __enable_if_t<!__use_branchless_sort<_Compare, _RandomAccessIterator>::value, int> = 0>
-inline _LIBCPP_HIDE_FROM_ABI void __sort4_maybe_branchless(
- _RandomAccessIterator __x1,
- _RandomAccessIterator __x2,
- _RandomAccessIterator __x3,
- _RandomAccessIterator __x4,
- _Compare __c) {
- std::__sort4<_AlgPolicy, _Compare>(__x1, __x2, __x3, __x4, __c);
+ __enable_if_t<!__use_branchless_sort<_Compare, _RandomAccessIterator>, int> = 0>
+inline _LIBCPP_HIDE_FROM_ABI void
+__sort4(_RandomAccessIterator __x1,
+ _RandomAccessIterator __x2,
+ _RandomAccessIterator __x3,
+ _RandomAccessIterator __x4,
+ _Compare __c) {
+ using _Ops = _IterOps<_AlgPolicy>;
+ std::__sort3<_AlgPolicy, _Compare>(__x1, __x2, __x3, __c);
+ if (__c(*__x4, *__x3)) {
+ _Ops::iter_swap(__x3, __x4);
+ if (__c(*__x3, *__x2)) {
+ _Ops::iter_swap(__x2, __x3);
+ if (__c(*__x2, *__x1)) {
+ _Ops::iter_swap(__x1, __x2);
+ }
+ }
+ }
}
+// stable, 4-10 compares, 0-9 swaps
+
template <class _AlgPolicy,
class _Compare,
class _RandomAccessIterator,
- __enable_if_t<__use_branchless_sort<_Compare, _RandomAccessIterator>::value, int> = 0>
-inline _LIBCPP_HIDE_FROM_ABI void __sort5_maybe_branchless(
- _RandomAccessIterator __x1,
- _RandomAccessIterator __x2,
- _RandomAccessIterator __x3,
- _RandomAccessIterator __x4,
- _RandomAccessIterator __x5,
- _Compare __c) {
+ __enable_if_t<__use_branchless_sort<_Compare, _RandomAccessIterator>, int> = 0>
+inline _LIBCPP_HIDE_FROM_ABI void
+__sort5(_RandomAccessIterator __x1,
+ _RandomAccessIterator __x2,
+ _RandomAccessIterator __x3,
+ _RandomAccessIterator __x4,
+ _RandomAccessIterator __x5,
+ _Compare __c) {
std::__cond_swap<_Compare>(__x1, __x2, __c);
std::__cond_swap<_Compare>(__x4, __x5, __c);
std::__partially_sorted_swap<_Compare>(__x3, __x4, __x5, __c);
@@ -253,16 +197,29 @@ inline _LIBCPP_HIDE_FROM_ABI void __sort5_maybe_branchless(
template <class _AlgPolicy,
class _Compare,
class _RandomAccessIterator,
- __enable_if_t<!__use_branchless_sort<_Compare, _RandomAccessIterator>::value, int> = 0>
-inline _LIBCPP_HIDE_FROM_ABI void __sort5_maybe_branchless(
- _RandomAccessIterator __x1,
- _RandomAccessIterator __x2,
- _RandomAccessIterator __x3,
- _RandomAccessIterator __x4,
- _RandomAccessIterator __x5,
- _Compare __c) {
- std::__sort5<_AlgPolicy, _Compare, _RandomAccessIterator>(
- std::move(__x1), std::move(__x2), std::move(__x3), std::move(__x4), std::move(__x5), __c);
+ __enable_if_t<!__use_branchless_sort<_Compare, _RandomAccessIterator>, int> = 0>
+inline _LIBCPP_HIDE_FROM_ABI void
+__sort5(_RandomAccessIterator __x1,
+ _RandomAccessIterator __x2,
+ _RandomAccessIterator __x3,
+ _RandomAccessIterator __x4,
+ _RandomAccessIterator __x5,
+ _Compare __comp) {
+ using _Ops = _IterOps<_AlgPolicy>;
+
+ std::__sort4<_AlgPolicy, _Compare>(__x1, __x2, __x3, __x4, __comp);
+ if (__comp(*__x5, *__x4)) {
+ _Ops::iter_swap(__x4, __x5);
+ if (__comp(*__x4, *__x3)) {
+ _Ops::iter_swap(__x3, __x4);
+ if (__comp(*__x3, *__x2)) {
+ _Ops::iter_swap(__x2, __x3);
+ if (__comp(*__x2, *__x1)) {
+ _Ops::iter_swap(__x1, __x2);
+ }
+ }
+ }
+ }
}
// Assumes size > 0
@@ -352,14 +309,14 @@ __insertion_sort_incomplete(_RandomAccessIterator __first, _RandomAccessIterator
_Ops::iter_swap(__first, __last);
return true;
case 3:
- std::__sort3_maybe_branchless<_AlgPolicy, _Comp>(__first, __first + difference_type(1), --__last, __comp);
+ std::__sort3<_AlgPolicy, _Comp>(__first, __first + difference_type(1), --__last, __comp);
return true;
case 4:
- std::__sort4_maybe_branchless<_AlgPolicy, _Comp>(
+ std::__sort4<_AlgPolicy, _Comp>(
__first, __first + difference_type(1), __first + difference_type(2), --__last, __comp);
return true;
case 5:
- std::__sort5_maybe_branchless<_AlgPolicy, _Comp>(
+ std::__sort5<_AlgPolicy, _Comp>(
__first,
__first + difference_type(1),
__first + difference_type(2),
@@ -370,7 +327,7 @@ __insertion_sort_incomplete(_RandomAccessIterator __first, _RandomAccessIterator
}
typedef typename iterator_traits<_RandomAccessIterator>::value_type value_type;
_RandomAccessIterator __j = __first + difference_type(2);
- std::__sort3_maybe_branchless<_AlgPolicy, _Comp>(__first, __first + difference_type(1), __j, __comp);
+ std::__sort3<_AlgPolicy, _Comp>(__first, __first + difference_type(1), __j, __comp);
const unsigned __limit = 8;
unsigned __count = 0;
for (_RandomAccessIterator __i = __j + difference_type(1); __i != __last; ++__i) {
@@ -777,14 +734,14 @@ void __introsort(_RandomAccessIterator __first,
_Ops::iter_swap(__first, __last);
return;
case 3:
- std::__sort3_maybe_branchless<_AlgPolicy, _Compare>(__first, __first + difference_type(1), --__last, __comp);
+ std::__sort3<_AlgPolicy, _Compare>(__first, __first + difference_type(1), --__last, __comp);
return;
case 4:
- std::__sort4_maybe_branchless<_AlgPolicy, _Compare>(
+ std::__sort4<_AlgPolicy, _Compare>(
__first, __first + difference_type(1), __first + difference_type(2), --__last, __comp);
return;
case 5:
- std::__sort5_maybe_branchless<_AlgPolicy, _Compare>(
+ std::__sort5<_AlgPolicy, _Compare>(
__first,
__first + difference_type(1),
__first + difference_type(2),
@@ -925,10 +882,8 @@ __sort_dispatch(_RandomAccessIterator __first, _RandomAccessIterator __last, _Co
// Only use bitset partitioning for arithmetic types. We should also check
// that the default comparator is in use so that we are sure that there are no
// branches in the comparator.
- std::__introsort<_AlgPolicy,
- _Comp&,
- _RandomAccessIterator,
- __use_branchless_sort<_Comp, _RandomAccessIterator>::value>(__first, __last, __comp, __depth_limit);
+ std::__introsort<_AlgPolicy, _Comp&, _RandomAccessIterator, __use_branchless_sort<_Comp, _RandomAccessIterator> >(
+ __first, __last, __comp, __depth_limit);
}
template <class _Type, class... _Options>
diff --git a/libcxx/include/__functional/operations.h b/libcxx/include/__functional/operations.h
index 6022bd679ed3e3..67d9da289aead3 100644
--- a/libcxx/include/__functional/operations.h
+++ b/libcxx/include/__functional/operations.h
@@ -362,6 +362,9 @@ struct _LIBCPP_TEMPLATE_VIS less : __binary_function<_Tp, _Tp, bool> {
};
_LIBCPP_CTAD_SUPPORTED_FOR_TYPE(less);
+template <class _Tp>
+inline const bool __desugars_to_v<__less_tag, less<_Tp>, _Tp, _Tp> = true;
+
template <class _Tp>
inline const bool __desugars_to_v<__totally_ordered_less_tag, less<_Tp>, _Tp, _Tp> = is_integral<_Tp>::value;
@@ -377,6 +380,9 @@ struct _LIBCPP_TEMPLATE_VIS less<void> {
typedef void is_transparent;
};
+template <class _Tp, class _Up>
+inline const bool __desugars_to_v<__less_tag, less<>, _Tp, _Up> = true;
+
template <class _Tp>
inline const bool __desugars_to_v<__totally_ordered_less_tag, less<>, _Tp, _Tp> = is_integral<_Tp>::value;
#endif
@@ -446,6 +452,9 @@ struct _LIBCPP_TEMPLATE_VIS greater : __binary_function<_Tp, _Tp, bool> {
};
_LIBCPP_CTAD_SUPPORTED_FOR_TYPE(greater);
+template <class _Tp>
+inline const bool __desugars_to_v<__greater_tag, greater<_Tp>, _Tp, _Tp> = true;
+
#if _LIBCPP_STD_VER >= 14
template <>
struct _LIBCPP_TEMPLATE_VIS greater<void> {
@@ -457,6 +466,9 @@ struct _LIBCPP_TEMPLATE_VIS greater<void> {
}
typedef void is_transparent;
};
+
+template <class _Tp, class _Up>
+inline const bool __desugars_to_v<__greater_tag, greater<>, _Tp, _Up> = true;
#endif
// Logical operations
diff --git a/libcxx/include/__functional/ranges_operations.h b/libcxx/include/__functional/ranges_operations.h
index f023d765a6c8ab..c8210193257c03 100644
--- a/libcxx/include/__functional/ranges_operations.h
+++ b/libcxx/include/__functional/ranges_operations.h
@@ -102,6 +102,9 @@ inline const bool __desugars_to_v<__equal_tag, ranges::equal_to, _Tp, _Up> = tru
template <class _Tp, class _Up>
inline const bool __desugars_to_v<__totally_ordered_less_tag, ranges::less, _Tp, _Up> = true;
+template <class _Tp, class _Up>
+inline const bool __desugars_to_v<__greater_tag, ranges::greater, _Tp, _Up> = true;
+
#endif // _LIBCPP_STD_VER >= 20
_LIBCPP_END_NAMESPACE_STD
diff --git a/libcxx/include/__type_traits/desugars_to.h b/libcxx/include/__type_traits/desugars_to.h
index b0ce7c414e5d77..d74830de9e0d7c 100644
--- a/libcxx/include/__type_traits/desugars_to.h
+++ b/libcxx/include/__type_traits/desugars_to.h
@@ -25,6 +25,12 @@ struct __equal_tag {};
// syntactically, the operation is equivalent to calling `a + b`
struct __plus_tag {};
+// syntactically, the operation is equivalent to calling `a < b`
+struct __less_tag {};
+
+// syntactically, the operation is equivalent to calling `a > b`
+struct __greater_tag {};
+
// syntactically, the operation is equivalent to calling `a < b`, and these expressions
// have to be true for any `a` and `b`:
// - `...
[truncated]
|
ldionne
approved these changes
Aug 22, 2024
philnik777
force-pushed
the
simplify_sort
branch
from
d90d328 to
7fd3793
Compare
philnik777
force-pushed
the
simplify_sort
branch
from
7fd3793 to
de412b1
Compare
|
We've observed ~20% regressions when sorting small collections (e.g. 300 elements or fewer). A reproducer from the team that made this observation should hopefully arrive in the near future. Should we roll this back and re-apply once the regression is addressed, or would it be better to prioritise a forward fix? |
|
@cjdb Since this isn't a correctness issue I'd rather fix forward. I've had someone claim a performance regression and them revert just for me having to press for information one too many times. If there is a fundamental problem with this code (which I doubt) we can still revert (parts of) it. |
|
Regarding the performance regression, here is a rough example of what we are seeing internally. https://godbolt.org/z/6zqEa4xEP |
|
In addition to @joanahalili's comment: the regression varies from 5% to 25% depending on the platform. And there's also a much more significant improvement for larger vector sizes (5000+) - around 1.3-2x speedup, but the code is used with smaller vector sizes in prod, which makes the improvement irrelevant to us. |
|
@philnik777 I would suggest reverting this and then splitting it into a few sub-commits. There were a few orthogonal changes in this patch that we could split out to isolate what's causing the regression. |
|
If there is a consensus to revert this, can we please do it ASAP? Our internal releases are blocked because of unclear performance implications. |
ldionne
added a commit
that referenced
this pull request
Sep 20, 2024
)" This reverts commit d4ffccf, which caused a performance regression that needs to be investigated further.
|
I reverted for now. I'll sync up with @philnik777 . |
|
Thank you! |
To provide more context. The situation is more nuanced. I am also from Google and was one of the people who developed new std::sort implementation back in 2022/2023. The patch is correct. It fixed a problem which was introduced (oh my) on Feb 1, 2023 here. Note that template <class _Tp>
struct __is_simple_comparator : false_type {};
template <>
struct __is_simple_comparator<__less<>&> : true_type {};
template <class _Tp>
struct __is_simple_comparator<less<_Tp>&> : true_type {};
template <class _Tp>
struct __is_simple_comparator<greater<_Tp>&> : true_type {};
#if _LIBCPP_STD_VER >= 20
template <>
struct __is_simple_comparator<ranges::less&> : true_type {};
template <>
struct __is_simple_comparator<ranges::greater&> : true_type {};
#endifSee godbolt for Disabling branch free version of bitset partitioning is not the desired behavior (which was from Feb 1, 2023). Branch free version was developed when we changed std::sort implementation and that was the whole point of doing so. This patch fixes it but regressions of 200-300 elements are real as floats and doubles are slightly harder to move branch free. I checked the same benchmarks on integers and they are fine for <100 elements. Let's move forward with this patch. We checked and actually biggest clients of double and float sorting are kind of big ones. If everybody here begins to be concerned with <200 elements sorts for floats, we can disable is_arithmetic |
|
Or should we have a fancier dispatching mechanism instead? Like |
|
We can but I would recommend enabling previous behavior first and then we can refine the threshold. Also in practice, microbenchmarks can lie as the memory is hot and even for low hundred elements we can see improvements because we remove branches. For example, when the first version with the branch free partitioning was rolled out at Google, we saw 10-15% improvement fleetwide even though microbenchmarks were saying much more. Unfortunately, we missed the regression from the last year by accident. :( |
|
@philnik777, please go ahead and resubmit the patch, it no longer blocks us and it fixes a long time bug. Thank you a lot for the work |
NoumanAmir657
pushed a commit
to NoumanAmir657/llvm-project
that referenced
this pull request
Nov 4, 2024
…m#104902)" (llvm#114023) This reverts commit ef44e46. The patch was originally reverted because it was deemed to introduce a performance regression for small inputs, however it also fixed a previous performance regression for larger inputs. So overall, this patch is desirable.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This does a few things to canonicalize the library a bit. Specifically
__desugars_to_vinstead of the custom__is_simple_comparator__use_branchless_sortan inline variable_maybe_branchlessversions of the__sortNfunctions and overload based on whether we can do branchless sorting instead.