1
0
mirror of https://github.com/golang/go synced 2024-11-24 13:20:15 -07:00
go/test/codegen/bits.go
Keith Randall b211fe0058 cmd/compile: remove bit operations that modify memory directly
These operations (BT{S,R,C}{Q,L}modify) are quite a bit slower than
other ways of doing the same thing.

Without the BTxmodify operations, there are two fallback ways the compiler
performs these operations: AND/OR/XOR operations directly on memory, or
load-BTx-write sequences. The compiler kinda chooses one arbitrarily
depending on rewrite rule application order. Currently, it uses
load-BTx-write for the Const benchmarks and AND/OR/XOR directly to memory
for the non-Const benchmarks. TBD, someone might investigate which of
the two fallback strategies is really better. For now, they are both
better than BTx ops.

name              old time/op  new time/op  delta
BitSet-8          1.09µs ± 2%  0.64µs ± 5%  -41.60%  (p=0.000 n=9+10)
BitClear-8        1.15µs ± 3%  0.68µs ± 6%  -41.00%  (p=0.000 n=10+10)
BitToggle-8       1.18µs ± 4%  0.73µs ± 2%  -38.36%  (p=0.000 n=10+8)
BitSetConst-8     37.0ns ± 7%  25.8ns ± 2%  -30.24%  (p=0.000 n=10+10)
BitClearConst-8   30.7ns ± 2%  25.0ns ±12%  -18.46%  (p=0.000 n=10+10)
BitToggleConst-8  36.9ns ± 1%  23.8ns ± 3%  -35.46%  (p=0.000 n=9+10)

Fixes #45790
Update #45242

Change-Id: Ie33a72dc139f261af82db15d446cd0855afb4e59
Reviewed-on: https://go-review.googlesource.com/c/go/+/318149
Trust: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Ben Shi <powerman1st@163.com>
2021-05-08 03:27:59 +00:00

358 lines
5.8 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
/************************************
* 64-bit instructions
************************************/
func bitcheck64_constleft(a uint64) (n int) {
// amd64:"BTQ\t[$]63"
if a&(1<<63) != 0 {
return 1
}
// amd64:"BTQ\t[$]60"
if a&(1<<60) != 0 {
return 1
}
// amd64:"BTL\t[$]0"
if a&(1<<0) != 0 {
return 1
}
return 0
}
func bitcheck64_constright(a [8]uint64) (n int) {
// amd64:"BTQ\t[$]63"
if (a[0]>>63)&1 != 0 {
return 1
}
// amd64:"BTQ\t[$]63"
if a[1]>>63 != 0 {
return 1
}
// amd64:"BTQ\t[$]63"
if a[2]>>63 == 0 {
return 1
}
// amd64:"BTQ\t[$]60"
if (a[3]>>60)&1 == 0 {
return 1
}
// amd64:"BTL\t[$]1"
if (a[4]>>1)&1 == 0 {
return 1
}
// amd64:"BTL\t[$]0"
if (a[5]>>0)&1 == 0 {
return 1
}
// amd64:"BTL\t[$]7"
if (a[6]>>5)&4 == 0 {
return 1
}
return 0
}
func bitcheck64_var(a, b uint64) (n int) {
// amd64:"BTQ"
if a&(1<<(b&63)) != 0 {
return 1
}
// amd64:"BTQ",-"BT.\t[$]0"
if (b>>(a&63))&1 != 0 {
return 1
}
return 0
}
func bitcheck64_mask(a uint64) (n int) {
// amd64:"BTQ\t[$]63"
if a&0x8000000000000000 != 0 {
return 1
}
// amd64:"BTQ\t[$]59"
if a&0x800000000000000 != 0 {
return 1
}
// amd64:"BTL\t[$]0"
if a&0x1 != 0 {
return 1
}
return 0
}
func biton64(a, b uint64) (n uint64) {
// amd64:"BTSQ"
n += b | (1 << (a & 63))
// amd64:"BTSQ\t[$]63"
n += a | (1 << 63)
// amd64:"BTSQ\t[$]60"
n += a | (1 << 60)
// amd64:"ORQ\t[$]1"
n += a | (1 << 0)
return n
}
func bitoff64(a, b uint64) (n uint64) {
// amd64:"BTRQ"
n += b &^ (1 << (a & 63))
// amd64:"BTRQ\t[$]63"
n += a &^ (1 << 63)
// amd64:"BTRQ\t[$]60"
n += a &^ (1 << 60)
// amd64:"ANDQ\t[$]-2"
n += a &^ (1 << 0)
return n
}
func bitcompl64(a, b uint64) (n uint64) {
// amd64:"BTCQ"
n += b ^ (1 << (a & 63))
// amd64:"BTCQ\t[$]63"
n += a ^ (1 << 63)
// amd64:"BTCQ\t[$]60"
n += a ^ (1 << 60)
// amd64:"XORQ\t[$]1"
n += a ^ (1 << 0)
return n
}
/************************************
* 32-bit instructions
************************************/
func bitcheck32_constleft(a uint32) (n int) {
// amd64:"BTL\t[$]31"
if a&(1<<31) != 0 {
return 1
}
// amd64:"BTL\t[$]28"
if a&(1<<28) != 0 {
return 1
}
// amd64:"BTL\t[$]0"
if a&(1<<0) != 0 {
return 1
}
return 0
}
func bitcheck32_constright(a [8]uint32) (n int) {
// amd64:"BTL\t[$]31"
if (a[0]>>31)&1 != 0 {
return 1
}
// amd64:"BTL\t[$]31"
if a[1]>>31 != 0 {
return 1
}
// amd64:"BTL\t[$]31"
if a[2]>>31 == 0 {
return 1
}
// amd64:"BTL\t[$]28"
if (a[3]>>28)&1 == 0 {
return 1
}
// amd64:"BTL\t[$]1"
if (a[4]>>1)&1 == 0 {
return 1
}
// amd64:"BTL\t[$]0"
if (a[5]>>0)&1 == 0 {
return 1
}
// amd64:"BTL\t[$]7"
if (a[6]>>5)&4 == 0 {
return 1
}
return 0
}
func bitcheck32_var(a, b uint32) (n int) {
// amd64:"BTL"
if a&(1<<(b&31)) != 0 {
return 1
}
// amd64:"BTL",-"BT.\t[$]0"
if (b>>(a&31))&1 != 0 {
return 1
}
return 0
}
func bitcheck32_mask(a uint32) (n int) {
// amd64:"BTL\t[$]31"
if a&0x80000000 != 0 {
return 1
}
// amd64:"BTL\t[$]27"
if a&0x8000000 != 0 {
return 1
}
// amd64:"BTL\t[$]0"
if a&0x1 != 0 {
return 1
}
return 0
}
func biton32(a, b uint32) (n uint32) {
// amd64:"BTSL"
n += b | (1 << (a & 31))
// amd64:"BTSL\t[$]31"
n += a | (1 << 31)
// amd64:"BTSL\t[$]28"
n += a | (1 << 28)
// amd64:"ORL\t[$]1"
n += a | (1 << 0)
return n
}
func bitoff32(a, b uint32) (n uint32) {
// amd64:"BTRL"
n += b &^ (1 << (a & 31))
// amd64:"BTRL\t[$]31"
n += a &^ (1 << 31)
// amd64:"BTRL\t[$]28"
n += a &^ (1 << 28)
// amd64:"ANDL\t[$]-2"
n += a &^ (1 << 0)
return n
}
func bitcompl32(a, b uint32) (n uint32) {
// amd64:"BTCL"
n += b ^ (1 << (a & 31))
// amd64:"BTCL\t[$]31"
n += a ^ (1 << 31)
// amd64:"BTCL\t[$]28"
n += a ^ (1 << 28)
// amd64:"XORL\t[$]1"
n += a ^ (1 << 0)
return n
}
// check direct operation on memory with constant and shifted constant sources
func bitOpOnMem(a []uint32, b, c, d uint32) {
// amd64:`ANDL\s[$]200,\s\([A-Z][A-Z0-9]+\)`
a[0] &= 200
// amd64:`ORL\s[$]220,\s4\([A-Z][A-Z0-9]+\)`
a[1] |= 220
// amd64:`XORL\s[$]240,\s8\([A-Z][A-Z0-9]+\)`
a[2] ^= 240
}
func bitcheckMostNegative(b uint8) bool {
// amd64:"TESTB"
return b&0x80 == 0x80
}
// Check AND masking on arm64 (Issue #19857)
func and_mask_1(a uint64) uint64 {
// arm64:`AND\t`
return a & ((1 << 63) - 1)
}
func and_mask_2(a uint64) uint64 {
// arm64:`AND\t`
return a & (1 << 63)
}
func and_mask_3(a, b uint32) (uint32, uint32) {
// arm/7:`BIC`,-`AND`
a &= 0xffffaaaa
// arm/7:`BFC`,-`AND`,-`BIC`
b &= 0xffc003ff
return a, b
}
// Check generation of arm64 BIC/EON/ORN instructions
func op_bic(x, y uint32) uint32 {
// arm64:`BIC\t`,-`AND`
return x &^ y
}
func op_eon(x, y, z uint32, a []uint32, n, m uint64) uint64 {
// arm64:`EON\t`,-`EOR`,-`MVN`
a[0] = x ^ (y ^ 0xffffffff)
// arm64:`EON\t`,-`EOR`,-`MVN`
a[1] = ^(y ^ z)
// arm64:`EON\t`,-`XOR`
a[2] = x ^ ^z
// arm64:`EON\t`,-`EOR`,-`MVN`
return n ^ (m ^ 0xffffffffffffffff)
}
func op_orn(x, y uint32) uint32 {
// arm64:`ORN\t`,-`ORR`
return x | ^y
}
// check bitsets
func bitSetPowerOf2Test(x int) bool {
// amd64:"BTL\t[$]3"
return x&8 == 8
}
func bitSetTest(x int) bool {
// amd64:"ANDQ\t[$]9, AX"
// amd64:"CMPQ\tAX, [$]9"
return x&9 == 9
}
// mask contiguous one bits
func cont1Mask64U(x uint64) uint64 {
// s390x:"RISBGZ\t[$]16, [$]47, [$]0,"
return x & 0x0000ffffffff0000
}
// mask contiguous zero bits
func cont0Mask64U(x uint64) uint64 {
// s390x:"RISBGZ\t[$]48, [$]15, [$]0,"
return x & 0xffff00000000ffff
}
func issue44228a(a []int64, i int) bool {
// amd64: "BTQ", -"SHL"
return a[i>>6]&(1<<(i&63)) != 0
}
func issue44228b(a []int32, i int) bool {
// amd64: "BTL", -"SHL"
return a[i>>5]&(1<<(i&31)) != 0
}