runtime: implement fastrand in go

So it could be inlined.

Using bit-tricks it could be implemented without condition
(improved trick version by Minux Ma).

Simple benchmark shows it is faster on i386 and x86_64, though
I don't know will it be faster on other architectures?

benchmark                       old ns/op     new ns/op     delta
BenchmarkFastrand-3             2.79          1.48          -46.95%
BenchmarkFastrandHashiter-3     25.9          24.9          -3.86%

Change-Id: Ie2eb6d0f598c0bb5fac7f6ad0f8b5e3eddaa361b
Reviewed-on: https://go-review.googlesource.com/34782Reviewed-by: Minux Ma <minux@golang.org>
Run-TryBot: Minux Ma <minux@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/runtime/asm_386.s

@@ -1595,20 +1595,6 @@ allsame:
 	MOVL	BX, (AX)
 	RET
 
-TEXT runtime·fastrand(SB), NOSPLIT, $0-4
-	get_tls(CX)
-	MOVL	g(CX), AX
-	MOVL	g_m(AX), AX
-	MOVL	m_fastrand(AX), DX
-	ADDL	DX, DX
-	MOVL	DX, BX
-	XORL	$0x88888eef, DX
-	JPL	2(PC)
-	MOVL	BX, DX
-	MOVL	DX, m_fastrand(AX)
-	MOVL	DX, ret+0(FP)
-	RET
-
 TEXT runtime·return0(SB), NOSPLIT, $0
 	MOVL	$0, AX
 	RET

src/runtime/asm_amd64.s

@@ -2163,19 +2163,6 @@ eqret:
 	MOVB	$0, ret+48(FP)
 	RET
 
-TEXT runtime·fastrand(SB), NOSPLIT, $0-4
-	get_tls(CX)
-	MOVQ	g(CX), AX
-	MOVQ	g_m(AX), AX
-	MOVL	m_fastrand(AX), DX
-	ADDL	DX, DX
-	MOVL	DX, BX
-	XORL	$0x88888eef, DX
-	CMOVLMI	BX, DX
-	MOVL	DX, m_fastrand(AX)
-	MOVL	DX, ret+0(FP)
-	RET
-
 TEXT runtime·return0(SB), NOSPLIT, $0
 	MOVL	$0, AX
 	RET

src/runtime/asm_amd64p32.s

@@ -991,19 +991,6 @@ eqret:
 	MOVB	AX, ret+24(FP)
 	RET
 
-TEXT runtime·fastrand(SB), NOSPLIT, $0-4
-	get_tls(CX)
-	MOVL	g(CX), AX
-	MOVL	g_m(AX), AX
-	MOVL	m_fastrand(AX), DX
-	ADDL	DX, DX
-	MOVL	DX, BX
-	XORL	$0x88888eef, DX
-	CMOVLMI	BX, DX
-	MOVL	DX, m_fastrand(AX)
-	MOVL	DX, ret+0(FP)
-	RET
-
 TEXT runtime·return0(SB), NOSPLIT, $0
 	MOVL	$0, AX
 	RET

src/runtime/asm_arm.s

@@ -971,15 +971,6 @@ _sib_notfound:
 	MOVW	R0, ret+12(FP)
 	RET
 
-TEXT runtime·fastrand(SB),NOSPLIT,$-4-4
-	MOVW	g_m(g), R1
-	MOVW	m_fastrand(R1), R0
-	ADD.S	R0, R0
-	EOR.MI	$0x88888eef, R0
-	MOVW	R0, m_fastrand(R1)
-	MOVW	R0, ret+0(FP)
-	RET
-
 TEXT runtime·return0(SB),NOSPLIT,$0
 	MOVW	$0, R0
 	RET

src/runtime/asm_arm64.s

@@ -959,18 +959,6 @@ equal:
 	MOVB	R0, ret+48(FP)
 	RET
 
-TEXT runtime·fastrand(SB),NOSPLIT,$-8-4
-	MOVD	g_m(g), R1
-	MOVWU	m_fastrand(R1), R0
-	ADD	R0, R0
-	CMPW	$0, R0
-	BGE	notneg
-	EOR	$0x88888eef, R0
-notneg:
-	MOVW	R0, m_fastrand(R1)
-	MOVW	R0, ret+0(FP)
-	RET
-
 TEXT runtime·return0(SB), NOSPLIT, $0
 	MOVW	$0, R0
 	RET

src/runtime/asm_mips64x.s

@@ -831,16 +831,6 @@ notfound:
 	MOVV	R1, ret+24(FP)
 	RET
 
-TEXT runtime·fastrand(SB), NOSPLIT, $0-4
-	MOVV	g_m(g), R2
-	MOVWU	m_fastrand(R2), R1
-	ADDU	R1, R1
-	BGEZ	R1, 2(PC)
-	XOR	$0x88888eef, R1
-	MOVW	R1, m_fastrand(R2)
-	MOVW	R1, ret+0(FP)
-	RET
-
 TEXT runtime·return0(SB), NOSPLIT, $0
 	MOVW	$0, R1
 	RET

src/runtime/asm_mipsx.s

@@ -904,16 +904,6 @@ cmp_ret:
 	MOVW	R8, ret+24(FP)
 	RET
 
-TEXT runtime·fastrand(SB),NOSPLIT,$0-4
-	MOVW	g_m(g), R2
-	MOVW	m_fastrand(R2), R1
-	ADDU	R1, R1
-	BGEZ	R1, 2(PC)
-	XOR	$0x88888eef, R1
-	MOVW	R1, m_fastrand(R2)
-	MOVW	R1, ret+0(FP)
-	RET
-
 TEXT runtime·return0(SB),NOSPLIT,$0
 	MOVW	$0, R1
 	RET

src/runtime/asm_ppc64x.s

@@ -1224,17 +1224,6 @@ TEXT bytes·Compare(SB),NOSPLIT|NOFRAME,$0-56
 	BR      cmpbodyBE<>(SB)
 #endif
 
-TEXT runtime·fastrand(SB), NOSPLIT, $0-4
-	MOVD	g_m(g), R4
-	MOVWZ	m_fastrand(R4), R3
-	ADD	R3, R3
-	CMPW	R3, $0
-	BGE	2(PC)
-	XOR	$0x88888eef, R3
-	MOVW	R3, m_fastrand(R4)
-	MOVW	R3, ret+0(FP)
-	RET
-
 TEXT runtime·return0(SB), NOSPLIT, $0
 	MOVW	$0, R3
 	RET

src/runtime/asm_s390x.s

@@ -851,17 +851,6 @@ TEXT runtime·memeqbodyclc(SB),NOSPLIT|NOFRAME,$0-0
 	CLC	$1, 0(R3), 0(R5)
 	RET
 
-TEXT runtime·fastrand(SB), NOSPLIT, $0-4
-	MOVD	g_m(g), R4
-	MOVWZ	m_fastrand(R4), R3
-	ADD	R3, R3
-	CMPW	R3, $0
-	BGE	2(PC)
-	XOR	$0x88888eef, R3
-	MOVW	R3, m_fastrand(R4)
-	MOVW	R3, ret+0(FP)
-	RET
-
 TEXT bytes·IndexByte(SB),NOSPLIT|NOFRAME,$0-40
 	MOVD	s+0(FP), R3     // s => R3
 	MOVD	s_len+8(FP), R4 // s_len => R4

src/runtime/export_test.go

@@ -245,3 +245,5 @@ func CountPagesInUse() (pagesInUse, counted uintptr) {
 
 	return
 }
+
+func Fastrand() uint32 { return fastrand() }

src/runtime/rand_test.go

+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package runtime_test
+
+import (
+	. "runtime"
+	"testing"
+)
+
+func BenchmarkFastrand(b *testing.B) {
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			Fastrand()
+		}
+	})
+}
+
+func BenchmarkFastrandHashiter(b *testing.B) {
+	var m = make(map[int]int, 10)
+	for i := 0; i < 10; i++ {
+		m[i] = i
+	}
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			for _ = range m {
+				break
+			}
+		}
+	})
+}

src/runtime/stubs.go

@@ -93,8 +93,15 @@ func reflect_memmove(to, from unsafe.Pointer, n uintptr) {
 // exported value for testing
 var hashLoad = loadFactor
 
-// in asm_*.s
-func fastrand() uint32
+//go:nosplit
+func fastrand() uint32 {
+	mp := getg().m
+	fr := mp.fastrand
+	fr <<= 1
+	fr ^= uint32(int32(fr)>>31) & 0x88888eef
+	mp.fastrand = fr
+	return fr
+}
 
 //go:linkname sync_fastrand sync.fastrand
 func sync_fastrand() uint32 { return fastrand() }