MirrorHub/construct
0
0
Fork 0
mirror of https://github.com/matrix-construct/construct synced 2024-11-19 16:30:52 +01:00
Code Issues Releases Wiki Activity

ircd::simd: Portabilities for clz on various x86.

Browse source
This commit is contained in:
Jason Volk 2020-07-17 06:39:52 -07:00
parent 6adc71151d
commit 291038a014
1 changed files with 32 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

34
include/ircd/simd/simd.h
View file

@ -43,6 +43,15 @@ __attribute__((target("lzcnt")))
ircd::simd::clz(const T a) ircd::simd::clz(const T a)
noexcept noexcept
{ {
constexpr auto uses_bsr
{
#ifndef __LZCNT__
true
#else
false
#endif
};
uint ret(0), i(0); do uint ret(0), i(0); do
{ {
const auto mask const auto mask
@ -50,10 +59,16 @@ noexcept
boolmask(uint(ret == sizeof_lane<T>() * 8 * i)) boolmask(uint(ret == sizeof_lane<T>() * 8 * i))
}; };
if constexpr(sizeof_lane<T>() <= sizeof(u16)) if constexpr(sizeof_lane<T>() <= sizeof(u16) && uses_bsr)
ret += (15 - __lzcnt16(__builtin_bswap16(a[i++]))) & mask;
else if constexpr(sizeof_lane<T>() <= sizeof(u16))
ret += __lzcnt16(__builtin_bswap16(a[i++])) & mask; ret += __lzcnt16(__builtin_bswap16(a[i++])) & mask;
else if constexpr(sizeof_lane<T>() <= sizeof(u32) && uses_bsr)
ret += (31 - __lzcnt32(__builtin_bswap32(a[i++]))) & mask;
else if constexpr(sizeof_lane<T>() <= sizeof(u32)) else if constexpr(sizeof_lane<T>() <= sizeof(u32))
ret += __lzcnt32(__builtin_bswap32(a[i++])) & mask; ret += __lzcnt32(__builtin_bswap32(a[i++])) & mask;
else if constexpr(uses_bsr)
ret += (63 - __lzcnt64(__builtin_bswap64(a[i++]))) & mask;
else else
ret += __lzcnt64(__builtin_bswap64(a[i++])) & mask; ret += __lzcnt64(__builtin_bswap64(a[i++])) & mask;
} }
@ -70,12 +85,27 @@ __attribute__((target("lzcnt")))
ircd::simd::ctz(const T a) ircd::simd::ctz(const T a)
noexcept noexcept
{ {
constexpr auto uses_bsr
{
#ifndef __LZCNT__
true
#else
false
#endif
};
uint ret(0), i(lanes<T>()), mask(-1U); do uint ret(0), i(lanes<T>()), mask(-1U); do
{ {
if constexpr(sizeof_lane<T>() <= sizeof(u16)) if constexpr(sizeof_lane<T>() <= sizeof(u16) && uses_bsr)
ret += (15 - __lzcnt16(a[--i])) & mask;
else if constexpr(sizeof_lane<T>() <= sizeof(u16))
ret += __lzcnt16(a[--i]) & mask; ret += __lzcnt16(a[--i]) & mask;
else if constexpr(sizeof_lane<T>() <= sizeof(u32) && uses_bsr)
ret += (31 - __lzcnt32(a[--i])) & mask;
else if constexpr(sizeof_lane<T>() <= sizeof(u32)) else if constexpr(sizeof_lane<T>() <= sizeof(u32))
ret += __lzcnt32(a[--i]) & mask; ret += __lzcnt32(a[--i]) & mask;
else if constexpr(uses_bsr)
ret += (63 - __lzcnt64(a[--i])) & mask;
else else
ret += __lzcnt64(a[--i]) & mask; ret += __lzcnt64(a[--i]) & mask;