Use AVX512_VPOPCNTDQ for the population-count kernels

go.mod

@@ -9,6 +9,7 @@ require (
 	github.com/google/uuid v1.6.0
 	github.com/mschoch/smat v0.2.0
 	github.com/stretchr/testify v1.11.1
+	golang.org/x/sys v0.30.0
 )
 
 require (

go.sum

@@ -18,6 +18,8 @@ golang.org/x/mod v0.23.0 h1:Zb7khfcRGKk+kqfxFaP5tZqCnDZMjC5VtUBs87Hr6QM=
 golang.org/x/mod v0.23.0/go.mod h1:6SkKJ3Xj0I0BrPOZoBy3bdMptDDU9oJrpohJ3eWZ1fY=
 golang.org/x/sync v0.11.0 h1:GGz8+XQP4FvTTrjZPzNKTMFtSXH80RAzG+5ghFPgK9w=
 golang.org/x/sync v0.11.0/go.mod h1:Czt+wKu1gCyEFDUtn0jG5QVvpJ6rzVqr5aXyt9drQfk=
+golang.org/x/sys v0.30.0 h1:QjkSwP/36a20jFYWkSue1YwXzLmsV5Gfq7Eiy72C1uc=
+golang.org/x/sys v0.30.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA=
 golang.org/x/text v0.22.0 h1:bofq7m3/HAFvbF51jz3Q9wLg3jkvSPuiZu/pD1XwgtM=
 golang.org/x/text v0.22.0/go.mod h1:YRoo4H8PVmsu+E3Ou7cqLVH8oXWIHVoX0jqUWALQhfY=
 golang.org/x/tools v0.30.0 h1:BgcpHewrV5AUp2G9MebG4XPFI1E2W41zU1SaqVA9vJY=

popcnt_avx2_amd64.go

@@ -32,34 +32,49 @@ func _popcntXorSliceAVX2(s, m []uint64) uint64
 var useAVX2 = _hasAVX2()
 
 func popcntSlice(s []uint64) uint64 {
+	if useAVX512Popcnt {
+		return popcntSliceAVX512(s)
+	}
 	if useAVX2 {
 		return _popcntSliceAVX2(s)
 	}
 	return popcntSliceGo(s)
 }
 
 func popcntMaskSlice(s, m []uint64) uint64 {
+	if useAVX512Popcnt {
+		return popcntMaskSliceAVX512(s, m)
+	}
 	if useAVX2 {
 		return _popcntMaskSliceAVX2(s, m)
 	}
 	return popcntMaskSliceGo(s, m)
 }
 
 func popcntAndSlice(s, m []uint64) uint64 {
+	if useAVX512Popcnt {
+		return popcntAndSliceAVX512(s, m)
+	}
 	if useAVX2 {
 		return _popcntAndSliceAVX2(s, m)
 	}
 	return popcntAndSliceGo(s, m)
 }
 
 func popcntOrSlice(s, m []uint64) uint64 {
+	if useAVX512Popcnt {
+		return popcntOrSliceAVX512(s, m)
+	}
 	if useAVX2 {
 		return _popcntOrSliceAVX2(s, m)
 	}
 	return popcntOrSliceGo(s, m)
 }
 
 func popcntXorSlice(s, m []uint64) uint64 {
+	if useAVX512Popcnt {
+		return popcntXorSliceAVX512(s, m)
+	}
 	if useAVX2 {
 		return _popcntXorSliceAVX2(s, m)
 	}

popcnt_avx512_amd64.go

@@ -0,0 +1,31 @@
+//go:build amd64 && !appengine
+// +build amd64,!appengine
+
+package roaring
+
+import "golang.org/x/sys/cpu"
+
+// The functions below are implemented in popcnt_avx512_amd64.s using
+// AVX512_VPOPCNTDQ. They are used in preference to the AVX2 kernels when the
+// CPU provides VPOPCNTQ (see useAVX512Popcnt).
+
+//go:noescape
+func popcntSliceAVX512(s []uint64) uint64
+
+//go:noescape
+func popcntMaskSliceAVX512(s, m []uint64) uint64
+
+//go:noescape
+func popcntAndSliceAVX512(s, m []uint64) uint64
+
+//go:noescape
+func popcntOrSliceAVX512(s, m []uint64) uint64
+
+//go:noescape
+func popcntXorSliceAVX512(s, m []uint64) uint64
+
+// useAVX512Popcnt selects the AVX-512 implementations when the running CPU has
+// AVX512_VPOPCNTDQ. x/sys/cpu verifies that the operating system saves the
+// opmask and ZMM state, and honors GODEBUG=cpu.avx512vpopcntdq=off so the path
+// can be disabled at run time. Evaluated once at package initialization.
+var useAVX512Popcnt = cpu.X86.HasAVX512VPOPCNTDQ