commit 9384c46598f70b326111a4b43debb6635ffd234a
Author: Barak Michener <me@barakmich.com>
Date:   Mon Nov 15 16:12:34 2021 -0800

    countlines in go asm, v1

diff --git a/benchmark_test.go b/benchmark_test.go
new file mode 100644
index 0000000..cf668e7
--- /dev/null
+++ b/benchmark_test.go
@@ -0,0 +1,61 @@
+package countlines
+
+import (
+	"math/rand"
+	"testing"
+)
+
+type size struct {
+	name string
+	l    int
+}
+
+var sizes = []size{
+	{"32", 32},
+	{"128", 128},
+	{"1K", 1 * 1024},
+	{"16K", 16 * 1024},
+	{"128K", 128 * 1024},
+	{"1M", 1024 * 1024},
+	{"16M", 16 * 1024 * 1024},
+	{"128M", 128 * 1024 * 1024},
+	{"512M", 512 * 1024 * 1024},
+}
+
+func randRead64(s []uint64) {
+	for i := range s {
+		s[i] = uint64(rand.Int63())
+	}
+}
+
+func BenchmarkCountNewlines(b *testing.B) {
+	for _, size := range sizes {
+		b.Run(size.name, func(b *testing.B) {
+			s := make([]byte, size.l)
+			rand.Read(s)
+
+			b.SetBytes(int64(size.l))
+			b.ResetTimer()
+
+			for i := 0; i < b.N; i++ {
+				CountNewlines(s)
+			}
+		})
+	}
+}
+
+func BenchmarkCountNewlinesGo(b *testing.B) {
+	for _, size := range sizes {
+		b.Run(size.name, func(b *testing.B) {
+			s := make([]byte, size.l)
+			rand.Read(s)
+
+			b.SetBytes(int64(size.l))
+			b.ResetTimer()
+
+			for i := 0; i < b.N; i++ {
+				countNewlinesGo(s)
+			}
+		})
+	}
+}
diff --git a/countlines_amd64.go b/countlines_amd64.go
new file mode 100644
index 0000000..30a4c1e
--- /dev/null
+++ b/countlines_amd64.go
@@ -0,0 +1,14 @@
+//go:build amd64 && !gccgo && !appengine
+
+package countlines
+
+func CountNewlines(s []byte) uint64 {
+	if len(s) == 0 {
+		return 0
+	}
+
+	return countNewlinesASM(&s[0], uint64(len(s)))
+}
+
+//go:noescape
+func countNewlinesASM(src *byte, len uint64) (ret uint64)
diff --git a/countlines_amd64.s b/countlines_amd64.s
new file mode 100644
index 0000000..306ac06
--- /dev/null
+++ b/countlines_amd64.s
@@ -0,0 +1,86 @@
+// +build amd64,!gccgo,!appengine
+
+#include "textflag.h"
+
+TEXT ·countNewlinesASM(SB),NOSPLIT,$0
+	MOVQ src+0(FP), SI
+	MOVQ len+8(FP), BX
+
+	XORQ AX, AX
+	XORQ DX, DX
+	XORPD X0, X0
+	XORPD X2, X2
+	XORPD X3, X3
+
+	MOVQ $0x0A0A0A0A0A0A0A0A, R10 // prep '\n'
+	PINSRQ $1, R10, X1
+	PINSRQ $0, R10, X1
+
+	CMPQ BX, $16
+	JB tail
+
+	CMPQ BX, $64
+	JB loop
+
+bigloop:
+	VMOVDQU -16(SI)(BX*1),  X11
+	VMOVDQU -32(SI)(BX*1),  X10
+	VMOVDQU -48(SI)(BX*1), X9
+	VMOVDQU -64(SI)(BX*1), X8
+	
+	VPCMPEQB X11, X1, X11
+	VPCMPEQB X10, X1, X10
+	VPCMPEQB X9, X1, X9
+	VPCMPEQB X8, X1, X8
+
+	VPADDB X0, X11, X2
+	VPADDB X2, X10, X2
+	VPADDB X2, X9, X2
+	VPADDB X2, X8, X2
+	PSIGNB X2, X2
+	VPSADBW X2, X0, X2
+	VPADDQ X2, X3, X3
+
+	SUBQ $64, BX
+	JZ ret
+
+	CMPQ BX, $64
+	JAE bigloop
+
+	CMPQ BX, $8
+	JB tail
+
+loop:
+	VMOVDQU -16(SI)(BX*1), X11
+
+	VPCMPEQB X11, X1, X11
+	VPADDB X0, X11, X2
+	PSIGNB X2, X2
+	VPSADBW X2, X0, X2
+	VPADDQ X2, X3, X3
+
+	SUBQ $16, BX
+	JZ ret
+
+	CMPQ BX, $16
+	JAE loop
+
+tail:
+	MOVB -1(SI)(BX*1), DX
+	CMPB DX, $0x0A
+	JNZ next
+
+	INCQ AX
+next:
+
+	SUBQ $1, BX
+	JNZ tail
+
+
+ret:
+	PEXTRQ $0, X3, CX
+	PEXTRQ $1, X3, DX
+	ADDQ CX, AX
+	ADDQ DX, AX
+	MOVQ AX, ret+16(FP)
+	RET
diff --git a/countlines_generic.go b/countlines_generic.go
new file mode 100644
index 0000000..aad9afe
--- /dev/null
+++ b/countlines_generic.go
@@ -0,0 +1,8 @@
+//go:build !amd64 || gccgo || appengine
+
+package countlines
+
+// CountBytes function counts number of non-zero bits in slice of 8bit unsigned integers.
+func CountNewlines(s []byte) uint64 {
+	return countNewlinesGo(s)
+}
diff --git a/countlines_go.go b/countlines_go.go
new file mode 100644
index 0000000..824a65a
--- /dev/null
+++ b/countlines_go.go
@@ -0,0 +1,14 @@
+package countlines
+
+//func countNewlinesGo(s []byte) uint64 {
+//return uint64(bytes.Count(s, []byte{'\n'}))
+//}
+
+func countNewlinesGo(s []byte) (out uint64) {
+	for _, x := range s {
+		if x == '\n' {
+			out += 1
+		}
+	}
+	return
+}
diff --git a/countlines_test.go b/countlines_test.go
new file mode 100644
index 0000000..ff1eaa8
--- /dev/null
+++ b/countlines_test.go
@@ -0,0 +1,60 @@
+package countlines
+
+import (
+	"testing"
+	"testing/quick"
+)
+
+type testVector struct {
+	n uint64
+	b []byte
+}
+
+func repTestVector(expected uint64, s []byte, repeatTimes int) testVector {
+	b := make([]byte, repeatTimes*len(s))
+	for i := 0; i < repeatTimes; i++ {
+		for j, v := range s {
+			b[i*len(s)+j] = v
+		}
+	}
+	return testVector{expected * uint64(repeatTimes), b}
+}
+
+var testVectors = []testVector{
+	repTestVector(1, []byte{0xa, 0x5e, 0x74, 0x15, 0x4e, 0xf3, 0xeb, 0xa6, 0x66, 0x83, 0x78, 0xfc, 0xfe, 0xd, 0x3e, 0xbd, 0xa8, 0x57, 0x93, 0x9e, 0x2b, 0x3d, 0xed, 0x99, 0xc9, 0xf9, 0x81, 0x10, 0x7f, 0xb0, 0xb0,
+		0xad, 0x1e, 0x2a, 0x84, 0xd0}, 1),
+	repTestVector(1, []byte{0x01, 0x02, 0x03, 0x0A}, 128),
+	repTestVector(0, []byte{0xf5, 0xc, 0x36, 0x9e, 0x86, 0xca, 0xf9}, 1),
+	repTestVector(0, []byte{0xf1, 0x35, 0xe5, 0xa3, 0x3c, 0x9f, 0x2c, 0x93, 0xbd, 0x72, 0xcf, 0x95, 0x16, 0x34, 0x37, 0xc5, 0xfd, 0xe4, 0x5d, 0x75, 0xb8, 0x2f, 0x5f, 0x53, 0x19, 0x2d, 0x6, 0xc3, 0xdb, 0x6d, 0xd4,
+		0xb5, 0xc0, 0x24, 0x95, 0x8e, 0x8d, 0x76, 0x20, 0xc5, 0x2b, 0x92, 0xc0, 0xa1, 0x3d, 0xee}, 1),
+}
+
+func testCountNewlines(t *testing.T, count func(s []byte) uint64) {
+	for _, tc := range testVectors {
+		if n := count(tc.b); n != tc.n {
+			t.Errorf("Expected %d, got %d", tc.n, n)
+		}
+	}
+}
+
+func TestCountNewlines(t *testing.T) {
+	testCountNewlines(t, CountNewlines)
+}
+
+func TestCountNewlinesGo(t *testing.T) {
+	testCountNewlines(t, countNewlinesGo)
+}
+
+func TestCountNewlinesCompare(t *testing.T) {
+	for _, tc := range testVectors {
+		if a, b := CountNewlines(tc.b), countNewlinesGo(tc.b); a != b {
+			t.Errorf("CountNewlines(%[1]v) = %[2]d; countNewlinesGo(%[1]v) = %[3]d", tc.b, a, b)
+		}
+	}
+
+	if err := quick.CheckEqual(countNewlinesGo, CountNewlines, &quick.Config{
+		MaxCountScale: 1000,
+	}); err != nil {
+		t.Error(err)
+	}
+}
diff --git a/go.mod b/go.mod
new file mode 100644
index 0000000..575c697
--- /dev/null
+++ b/go.mod
@@ -0,0 +1,5 @@
+module git.barakmich.com/barak/go-countlines
+
+go 1.17
+
+require github.com/klauspost/cpuid v1.3.1 // indirect
diff --git a/go.sum b/go.sum
new file mode 100644
index 0000000..f568b65
--- /dev/null
+++ b/go.sum
@@ -0,0 +1,2 @@
+github.com/klauspost/cpuid v1.3.1 h1:5JNjFYYQrZeKRJ0734q51WCEEn2huer72Dc7K+R/b6s=
+github.com/klauspost/cpuid v1.3.1/go.mod h1:bYW4mA6ZgKPob1/Dlai2LviZJO7KGI3uoWLd42rAQw4=