added benchmark for assembly vs native popcount

This commit is contained in:
Todd Gruben 2015-12-22 09:34:26 -06:00
parent c62cda7d17
commit 8cbf77a11d
6 changed files with 102 additions and 7 deletions

View file

@ -93,6 +93,7 @@ func (m *Main) Run(args ...string) error {
return err
} else if addr == "" {
return errors.New("port must be specified in config host")
}
// Set up profiling.
@ -110,7 +111,7 @@ func (m *Main) Run(args ...string) error {
cluster := m.Config.PilosaCluster()
// Create index to store fragments.
index := pilosa.NewIndex()
index := pilosa.NewIndex("/tmp/")
// Create executor for executing queries.
e := pilosa.NewExecutor(index)

View file

@ -3,7 +3,7 @@ package pilosa
// bit population count, take from
// https://code.google.com/p/go/issues/detail?id=4988#c11
// credit: https://code.google.com/u/arnehormann/
func popcount(x uint64) (n uint64) {
func popcntGo(x uint64) (n uint64) {
x -= (x >> 1) & 0x5555555555555555
x = (x>>2)&0x3333333333333333 + x&0x3333333333333333
x += x >> 4
@ -15,7 +15,7 @@ func popcount(x uint64) (n uint64) {
func popcntSliceGo(s []uint64) uint64 {
cnt := uint64(0)
for _, x := range s {
cnt += popcount(x)
cnt += popcntGo(x)
}
return cnt
}
@ -23,7 +23,7 @@ func popcntSliceGo(s []uint64) uint64 {
func popcntMaskSliceGo(s, m []uint64) uint64 {
cnt := uint64(0)
for i := range s {
cnt += popcount(s[i] &^ m[i])
cnt += popcntGo(s[i] &^ m[i])
}
return cnt
}
@ -31,7 +31,7 @@ func popcntMaskSliceGo(s, m []uint64) uint64 {
func popcntAndSliceGo(s, m []uint64) uint64 {
cnt := uint64(0)
for i := range s {
cnt += popcount(s[i] & m[i])
cnt += popcntGo(s[i] & m[i])
}
return cnt
}
@ -39,7 +39,7 @@ func popcntAndSliceGo(s, m []uint64) uint64 {
func popcntOrSliceGo(s, m []uint64) uint64 {
cnt := uint64(0)
for i := range s {
cnt += popcount(s[i] | m[i])
cnt += popcntGo(s[i] | m[i])
}
return cnt
}
@ -47,7 +47,7 @@ func popcntOrSliceGo(s, m []uint64) uint64 {
func popcntXorSliceGo(s, m []uint64) uint64 {
cnt := uint64(0)
for i := range s {
cnt += popcount(s[i] ^ m[i])
cnt += popcntGo(s[i] ^ m[i])
}
return cnt
}

View file

@ -100,3 +100,10 @@ LOOP popcntXorSliceLoop
popcntXorSliceEnd:
MOVQ AX, ret+48(FP)
RET
TEXT ·popcntAsm(SB),4,$0-16
MOVQ x+0(FP), DX
POPCNTQ_DX_DX
MOVQ DX, ret+8(FP)
RET

View file

@ -28,6 +28,9 @@ func popcntOrSliceAsm(s, m []uint64) uint64
func popcntXorSliceAsm(s, m []uint64) uint64
//go:noescape
func popcntAsm(x uint64)uint64
func popcntSlice(s []uint64) uint64 {
if useAsm {
return popcntSliceAsm(s)
@ -62,3 +65,11 @@ func popcntXorSlice(s, m []uint64) uint64 {
}
return popcntXorSliceGo(s, m)
}
func popcnt(x uint64) uint64{
if useAsm {
return popcntAsm(x)
}
return popcntGo(x)
}

View file

@ -7,3 +7,4 @@ func popcntMaskSlice(s, m []uint64) uint64 { return popcntMaskSliceGo(s, m) }
func popcntAndSlice(s, m []uint64) uint64 { return popcntAndSliceGo(s, m) }
func popcntOrSlice(s, m []uint64) uint64 { return popcntOrSliceGo(s, m) }
func popcntXorSlice(s, m []uint64) uint64 { return popcntXorSliceGo(s, m) }
func popcnt(s uint64) uint64 { return popcntGo(s)}

75
popcnt_test.go Normal file
View file

@ -0,0 +1,75 @@
package pilosa
import "testing"
func BenchmarkPopcntAsm(b *testing.B) {
// run the Fib function b.N times
for n := 0; n < b.N; n++ {
popcntAsm(0xdeadbeef)
}
}
func BenchmarkPopcntGo(b *testing.B) {
// run the Fib function b.N times
for n := 0; n < b.N; n++ {
popcntGo(0xdeadbeef)
}
}
func getData() []uint64 {
return []uint64{
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
0xdeadbeef,
}
}
func BenchmarkPopcntSliceGo(b *testing.B) {
d := getData()
for n := 0; n < b.N; n++ {
popcntSliceGo(d)
}
}
func BenchmarkPopcntSliceAsm(b *testing.B) {
d := getData()
for n := 0; n < b.N; n++ {
popcntSliceAsm(d)
}
}
func BenchmarkPopcntSlice(b *testing.B) {
d := getData()
for n := 0; n < b.N; n++ {
popcntSlice(d)
}
}