jyapayne/Nim
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

count 64 bits at a time instead of 8 (#12159)

Browse source 
* count 64 bits at a time

* spacing

* only do 64bts on x86

* add amd64

* use while
This commit is contained in:
Andreas Rumpf 2019-09-12 08:20:53 +02:00 • committed by GitHub
commit eb241f4241
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 9 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

12
lib/system/sets.nim
View file

@ -21,7 +21,7 @@ proc countBits32(n: uint32): int {.compilerproc.} =
v = (v and 0x33333333) + ((v shr 2) and 0x33333333) v = (v and 0x33333333) + ((v shr 2) and 0x33333333)
result = (((v + (v shr 4) and 0xF0F0F0F) * 0x1010101) shr 24).int result = (((v + (v shr 4) and 0xF0F0F0F) * 0x1010101) shr 24).int
proc countBits64(n: uint64): int {.compilerproc.} = proc countBits64(n: uint64): int {.compilerproc, inline.} =
# generic formula is from: https://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel # generic formula is from: https://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel
var v = uint64(n) var v = uint64(n)
v = v - ((v shr 1'u64) and 0x5555555555555555'u64) v = v - ((v shr 1'u64) and 0x5555555555555555'u64)
@ -30,6 +30,12 @@ proc countBits64(n: uint64): int {.compilerproc.} =
result = ((v * 0x0101010101010101'u64) shr 56'u64).int result = ((v * 0x0101010101010101'u64) shr 56'u64).int
proc cardSet(s: NimSet, len: int): int {.compilerproc, inline.} = proc cardSet(s: NimSet, len: int): int {.compilerproc, inline.} =
for i in 0..<len: var i = 0
if likely(s[i] == 0): continue when defined(x86) or defined(amd64):
while i < len - 8:
inc(result, countBits64((cast[ptr uint64](s[i].unsafeAddr))[]))
inc(i, 8)
while i < len:
inc(result, countBits32(uint32(s[i]))) inc(result, countBits32(uint32(s[i])))
inc(i, 1)