go/test/codegen/mathbits.go
Josh Bleecher Snyder 54dbab5221 cmd/compile: optimize TrailingZeros(8|16) on amd64
Introduce Ctz8 and Ctz16 ops and provide optimized lowerings for them.
amd64 only for this CL, although it wouldn't surprise me
if other architectures also admit of optimized lowerings.

name               old time/op  new time/op  delta
TrailingZeros8-8   1.33ns ± 6%  0.84ns ± 3%  -36.90%  (p=0.000 n=20+20)
TrailingZeros16-8  1.26ns ± 5%  0.84ns ± 5%  -33.50%  (p=0.000 n=20+18)

Code:

func f8(x uint8)   { z = bits.TrailingZeros8(x) }
func f16(x uint16) { z = bits.TrailingZeros16(x) }

Before:

"".f8 STEXT nosplit size=34 args=0x8 locals=0x0
	0x0000 00000 (x.go:7)	TEXT	"".f8(SB), NOSPLIT, $0-8
	0x0000 00000 (x.go:7)	FUNCDATA	$0, gclocals·2a5305abe05176240e61b8620e19a815(SB)
	0x0000 00000 (x.go:7)	FUNCDATA	$1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
	0x0000 00000 (x.go:7)	MOVBLZX	"".x+8(SP), AX
	0x0005 00005 (x.go:7)	MOVBLZX	AL, AX
	0x0008 00008 (x.go:7)	BTSQ	$8, AX
	0x000d 00013 (x.go:7)	BSFQ	AX, AX
	0x0011 00017 (x.go:7)	MOVL	$64, CX
	0x0016 00022 (x.go:7)	CMOVQEQ	CX, AX
	0x001a 00026 (x.go:7)	MOVQ	AX, "".z(SB)
	0x0021 00033 (x.go:7)	RET

"".f16 STEXT nosplit size=34 args=0x8 locals=0x0
	0x0000 00000 (x.go:8)	TEXT	"".f16(SB), NOSPLIT, $0-8
	0x0000 00000 (x.go:8)	FUNCDATA	$0, gclocals·2a5305abe05176240e61b8620e19a815(SB)
	0x0000 00000 (x.go:8)	FUNCDATA	$1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
	0x0000 00000 (x.go:8)	MOVWLZX	"".x+8(SP), AX
	0x0005 00005 (x.go:8)	MOVWLZX	AX, AX
	0x0008 00008 (x.go:8)	BTSQ	$16, AX
	0x000d 00013 (x.go:8)	BSFQ	AX, AX
	0x0011 00017 (x.go:8)	MOVL	$64, CX
	0x0016 00022 (x.go:8)	CMOVQEQ	CX, AX
	0x001a 00026 (x.go:8)	MOVQ	AX, "".z(SB)
	0x0021 00033 (x.go:8)	RET

After:

"".f8 STEXT nosplit size=20 args=0x8 locals=0x0
	0x0000 00000 (x.go:7)	TEXT	"".f8(SB), NOSPLIT, $0-8
	0x0000 00000 (x.go:7)	FUNCDATA	$0, gclocals·2a5305abe05176240e61b8620e19a815(SB)
	0x0000 00000 (x.go:7)	FUNCDATA	$1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
	0x0000 00000 (x.go:7)	MOVBLZX	"".x+8(SP), AX
	0x0005 00005 (x.go:7)	BTSL	$8, AX
	0x0009 00009 (x.go:7)	BSFL	AX, AX
	0x000c 00012 (x.go:7)	MOVQ	AX, "".z(SB)
	0x0013 00019 (x.go:7)	RET

"".f16 STEXT nosplit size=20 args=0x8 locals=0x0
	0x0000 00000 (x.go:8)	TEXT	"".f16(SB), NOSPLIT, $0-8
	0x0000 00000 (x.go:8)	FUNCDATA	$0, gclocals·2a5305abe05176240e61b8620e19a815(SB)
	0x0000 00000 (x.go:8)	FUNCDATA	$1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
	0x0000 00000 (x.go:8)	MOVWLZX	"".x+8(SP), AX
	0x0005 00005 (x.go:8)	BTSL	$16, AX
	0x0009 00009 (x.go:8)	BSFL	AX, AX
	0x000c 00012 (x.go:8)	MOVQ	AX, "".z(SB)
	0x0013 00019 (x.go:8)	RET

Change-Id: I0551e357348de2b724737d569afd6ac9f5c3aa11
Reviewed-on: https://go-review.googlesource.com/108940
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Reviewed-by: Keith Randall <khr@golang.org>
2018-04-25 21:33:52 +00:00

218 lines
4.1 KiB
Go

// asmcheck
// Copyright 2018 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 codegen
import "math/bits"
// ----------------------- //
// bits.LeadingZeros //
// ----------------------- //
func LeadingZeros(n uint) int {
// amd64:"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.LeadingZeros(n)
}
func LeadingZeros64(n uint64) int {
// amd64:"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.LeadingZeros64(n)
}
func LeadingZeros32(n uint32) int {
// amd64:"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.LeadingZeros32(n)
}
func LeadingZeros16(n uint16) int {
// amd64:"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.LeadingZeros16(n)
}
func LeadingZeros8(n uint8) int {
// amd64 LeadingZeros8 not intrinsified (see ssa.go)
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.LeadingZeros8(n)
}
// --------------- //
// bits.Len* //
// --------------- //
func Len(n uint) int {
// amd64:"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.Len(n)
}
func Len64(n uint64) int {
// amd64:"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.Len64(n)
}
func Len32(n uint32) int {
// amd64:"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.Len32(n)
}
func Len16(n uint16) int {
// amd64:"BSRQ"
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.Len16(n)
}
func Len8(n uint8) int {
// amd64 Len8 not intrisified (see ssa.go)
// s390x:"FLOGR"
// arm:"CLZ" arm64:"CLZ"
// mips:"CLZ"
return bits.Len8(n)
}
// -------------------- //
// bits.OnesCount //
// -------------------- //
func OnesCount(n uint) int {
// amd64:"POPCNTQ",".*support_popcnt"
// arm64:"VCNT","VUADDLV"
return bits.OnesCount(n)
}
func OnesCount64(n uint64) int {
// amd64:"POPCNTQ",".*support_popcnt"
// arm64:"VCNT","VUADDLV"
return bits.OnesCount64(n)
}
func OnesCount32(n uint32) int {
// amd64:"POPCNTL",".*support_popcnt"
// arm64:"VCNT","VUADDLV"
return bits.OnesCount32(n)
}
func OnesCount16(n uint16) int {
// amd64:"POPCNTL",".*support_popcnt"
// arm64:"VCNT","VUADDLV"
return bits.OnesCount16(n)
}
// ----------------------- //
// bits.ReverseBytes //
// ----------------------- //
func ReverseBytes(n uint) uint {
// amd64:"BSWAPQ"
// s390x:"MOVDBR"
// arm64:"REV"
return bits.ReverseBytes(n)
}
func ReverseBytes64(n uint64) uint64 {
// amd64:"BSWAPQ"
// s390x:"MOVDBR"
// arm64:"REV"
return bits.ReverseBytes64(n)
}
func ReverseBytes32(n uint32) uint32 {
// amd64:"BSWAPL"
// s390x:"MOVWBR"
// arm64:"REVW"
return bits.ReverseBytes32(n)
}
func ReverseBytes16(n uint16) uint16 {
// amd64:"ROLW"
return bits.ReverseBytes16(n)
}
// --------------------- //
// bits.RotateLeft //
// --------------------- //
func RotateLeft64(n uint64) uint64 {
// amd64:"ROLQ"
// arm64:"ROR"
// ppc64:"ROTL"
return bits.RotateLeft64(n, 37)
}
func RotateLeft32(n uint32) uint32 {
// amd64:"ROLL" 386:"ROLL"
// arm64:"RORW"
// ppc64:"ROTLW"
return bits.RotateLeft32(n, 9)
}
func RotateLeft16(n uint16) uint16 {
// amd64:"ROLW" 386:"ROLW"
return bits.RotateLeft16(n, 5)
}
func RotateLeft8(n uint8) uint8 {
// amd64:"ROLB" 386:"ROLB"
return bits.RotateLeft8(n, 5)
}
// ------------------------ //
// bits.TrailingZeros //
// ------------------------ //
func TrailingZeros(n uint) int {
// amd64:"BSFQ","MOVL\t\\$64","CMOVQEQ"
// s390x:"FLOGR"
return bits.TrailingZeros(n)
}
func TrailingZeros64(n uint64) int {
// amd64:"BSFQ","MOVL\t\\$64","CMOVQEQ"
// s390x:"FLOGR"
return bits.TrailingZeros64(n)
}
func TrailingZeros32(n uint32) int {
// amd64:"BTSQ\\t\\$32","BSFQ"
// s390x:"FLOGR","MOVWZ"
return bits.TrailingZeros32(n)
}
func TrailingZeros16(n uint16) int {
// amd64:"BSFL","BTSL\\t\\$16"
// s390x:"FLOGR","OR\t\\$65536"
return bits.TrailingZeros16(n)
}
func TrailingZeros8(n uint8) int {
// amd64:"BSFL","BTSL\\t\\$8"
// s390x:"FLOGR","OR\t\\$256"
return bits.TrailingZeros8(n)
}