zeroshade commented on code in PR #13768:
URL: https://github.com/apache/arrow/pull/13768#discussion_r937040390
##########
go/arrow/bitutil/bitmaps.go:
##########
@@ -422,3 +423,81 @@ func CopyBitmap(src []byte, srcOffset, length int, dst
[]byte, dstOffset int) {
dst[nbytes-1] &= ^trailMask
dst[nbytes-1] |= lastData & trailMask
}
+
+type bitOp struct {
+ opWord func(uint64, uint64) uint64
+ opByte func(byte, byte) byte
+}
+
+var (
+ bitAndOp = bitOp{
+ opWord: func(l, r uint64) uint64 { return l & r },
+ opByte: func(l, r byte) byte { return l & r },
+ }
+ bitOrOp = bitOp{
+ opWord: func(l, r uint64) uint64 { return l | r },
+ opByte: func(l, r byte) byte { return l | r },
+ }
+)
+
+func alignedBitmapOp(op bitOp, left, right []byte, lOffset, rOffset int64, out
[]byte, outOffset int64, length int64) {
+ debug.Assert(lOffset%8 == rOffset%8, "aligned bitmap op called with
unaligned offsets")
+ debug.Assert(lOffset%8 == outOffset%8, "aligned bitmap op called with
unaligned output offset")
+
+ nbytes := BytesForBits(length + lOffset%8)
+ left = left[lOffset/8:]
+ right = right[rOffset/8:]
+ out = out[outOffset/8:]
+ for i := int64(0); i < nbytes; i++ {
+ out[i] = op.opByte(left[i], right[i])
+ }
+}
+
+func unalignedBitmapOp(op bitOp, left, right []byte, lOffset, rOffset int64,
out []byte, outOffset int64, length int64) {
+ leftRdr := NewBitmapWordReader(left, int(lOffset), int(length))
+ rightRdr := NewBitmapWordReader(right, int(rOffset), int(length))
+ writer := NewBitmapWordWriter(out, int(outOffset), int(length))
+
+ for nwords := leftRdr.Words(); nwords > 0; nwords-- {
+ writer.PutNextWord(op.opWord(leftRdr.NextWord(),
rightRdr.NextWord()))
Review Comment:
Well this is much better...
```
goos: linux
goarch: amd64
pkg: github.com/apache/arrow/go/v10/arrow/bitutil
cpu: 12th Gen Intel(R) Core(TM) i7-12700H
BenchmarkBitmapAnd/nbytes=32768/0-20 1110499 1078
ns/op 60797.80 MB/s 0 B/op 0 allocs/op
BenchmarkBitmapAnd/nbytes=32768/1-20 29881 39699
ns/op 1650.82 MB/s 192 B/op 3 allocs/op
BenchmarkBitmapAnd/nbytes=32768/2-20 30028 39396
ns/op 1663.50 MB/s 192 B/op 3 allocs/op
BenchmarkBitmapAnd/nbytes=131072/0-20 254810 4698
ns/op 55795.91 MB/s 0 B/op 0 allocs/op
BenchmarkBitmapAnd/nbytes=131072/1-20 7418 159083
ns/op 1647.85 MB/s 192 B/op 3 allocs/op
BenchmarkBitmapAnd/nbytes=131072/2-20 7378 157388
ns/op 1665.59 MB/s 192 B/op 3 allocs/op
```
:smile: Though it looks like i should make a card to look into similarly
speeding up the unaligned cases (by way of making the bitmapword reader and
writer faster)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
