boostorg/unordered
1
0
Fork 1
mirror of https://github.com/boostorg/unordered.git synced 2025-07-29 10:57:16 +02:00
Code Issues Packages Projects Releases Wiki Activity

polished BOOST_UNORDERED_PREFETCH[_ELEMENTS]

Browse Source
This commit is contained in:
joaquintides
2023-05-28 19:09:14 +02:00
parent 950e640fcf
commit 7aaa2e9452
2 changed files with 26 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/boost/unordered/detail/foa/concurrent_table.hpp
View File

@ -949,7 +949,7 @@ private:
auto mask=pg->match(hash);
if(mask){
auto p=this->arrays.elements+pos*N;
BOOST_UNORDERED_PREFETCH_ELEMENTS(p);
BOOST_UNORDERED_PREFETCH_ELEMENTS(p,N);
auto lck=access(access_mode,pos);
do{
auto n=unchecked_countr_zero(mask);

43
include/boost/unordered/detail/foa/core.hpp
View File

@ -85,33 +85,40 @@
}while(0)
#endif
/* We use BOOST_UNORDERED_PREFETCH[_ELEMENTS] macros rather than proper
* functions because of https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109985
*/
#if defined(BOOST_GCC)||defined(BOOST_CLANG)
#define BOOST_UNORDERED_PREFETCH(p) __builtin_prefetch((const char*)p)
#define BOOST_UNORDERED_PREFETCH(p) __builtin_prefetch((const char*)(p))
#elif defined(BOOST_UNORDERED_SSE2)
#define BOOST_UNORDERED_PREFETCH(p) _mm_prefetch((const char*)p,_MM_HINT_T0)
#define BOOST_UNORDERED_PREFETCH(p) _mm_prefetch((const char*)(p),_MM_HINT_T0)
#else
#define BOOST_UNORDERED_PREFETCH(p)
#define BOOST_UNORDERED_PREFETCH(p) ((void)0)
#endif
/* We have experimentally confirmed that ARM architectures get a higher
* speedup when around the first half of the element slots in a group are
* prefetched, whereas for Intel just the first cache line is best.
* Please report back if you find better tunings for some particular
* architectures.
*/
* speedup when around the first half of the element slots in a group are
* prefetched, whereas for Intel just the first cache line is best.
* Please report back if you find better tunings for some particular
* architectures.
*/
#if BOOST_ARCH_ARM
/* Cache line size can't be known at compile time, so we settle on
* the very frequent value of 64B.
*/
#define BOOST_UNORDERED_PREFETCH_ELEMENTS(p) \
do{ \
constexpr int cache_line=64; \
const char *p0=reinterpret_cast<const char*>(p), \
*p1=p0+sizeof(value_type)*N/2; \
for(;p0<p1;p0+=cache_line)BOOST_UNORDERED_PREFETCH(p0); \
* the very frequent value of 64B.
*/
#define BOOST_UNORDERED_PREFETCH_ELEMENTS(p,N) \
do{ \
auto BOOST_UNORDERED_P=(p); \
constexpr int cache_line=64; \
const char *p0=reinterpret_cast<const char*>(BOOST_UNORDERED_P), \
*p1=p0+sizeof(*BOOST_UNORDERED_P)*(N)/2; \
for(;p0<p1;p0+=cache_line)BOOST_UNORDERED_PREFETCH(p0); \
}while(0)
#else
#define BOOST_UNORDERED_PREFETCH_ELEMENTS(p) BOOST_UNORDERED_PREFETCH(p)
#define BOOST_UNORDERED_PREFETCH_ELEMENTS(p,N) BOOST_UNORDERED_PREFETCH(p)
#endif
#ifdef __has_feature
@ -1471,7 +1478,7 @@ public:
if(mask){
BOOST_UNORDERED_ASSUME(arrays.elements!=nullptr);
auto p=arrays.elements+pos*N;
BOOST_UNORDERED_PREFETCH_ELEMENTS(p);
BOOST_UNORDERED_PREFETCH_ELEMENTS(p,N);
do{
auto n=unchecked_countr_zero(mask);
if(BOOST_LIKELY(bool(pred()(x,key_from(p[n]))))){