1
0
mirror of https://github.com/golang/go synced 2024-11-26 23:01:23 -07:00

cmd/gc: optimize memclr of slices and arrays

Recognize loops of the form

for i := range a {
	a[i] = zero
}

in which the evaluation of a is free from side effects.
Replace these loops with calls to memclr.
This occurs in the stdlib in 18 places.

The motivating example is clearing a byte slice:

benchmark                old ns/op     new ns/op     delta
BenchmarkGoMemclr5       3.31          3.26          -1.51%
BenchmarkGoMemclr16      13.7          3.28          -76.06%
BenchmarkGoMemclr64      50.8          4.14          -91.85%
BenchmarkGoMemclr256     157           6.02          -96.17%

Update #5373.

Change-Id: I99d3e6f5f268e8c6499b7e661df46403e5eb83e4
Reviewed-on: https://go-review.googlesource.com/2520
Reviewed-by: Keith Randall <khr@golang.org>
This commit is contained in:
Josh Bleecher Snyder 2015-01-07 17:44:49 -08:00
parent 1d0c7792f1
commit f03c9202c4
7 changed files with 155 additions and 1 deletions

View File

@ -126,6 +126,7 @@ char *runtimeimport =
"func @\"\".makeslice (@\"\".typ·2 *byte, @\"\".nel·3 int64, @\"\".cap·4 int64) (@\"\".ary·1 []any)\n" "func @\"\".makeslice (@\"\".typ·2 *byte, @\"\".nel·3 int64, @\"\".cap·4 int64) (@\"\".ary·1 []any)\n"
"func @\"\".growslice (@\"\".typ·2 *byte, @\"\".old·3 []any, @\"\".n·4 int64) (@\"\".ary·1 []any)\n" "func @\"\".growslice (@\"\".typ·2 *byte, @\"\".old·3 []any, @\"\".n·4 int64) (@\"\".ary·1 []any)\n"
"func @\"\".memmove (@\"\".to·1 *any, @\"\".frm·2 *any, @\"\".length·3 uintptr)\n" "func @\"\".memmove (@\"\".to·1 *any, @\"\".frm·2 *any, @\"\".length·3 uintptr)\n"
"func @\"\".memclr (@\"\".ptr·1 *byte, @\"\".length·2 uintptr)\n"
"func @\"\".memequal (@\"\".x·2 *any, @\"\".y·3 *any, @\"\".size·4 uintptr) (? bool)\n" "func @\"\".memequal (@\"\".x·2 *any, @\"\".y·3 *any, @\"\".size·4 uintptr) (? bool)\n"
"func @\"\".memequal8 (@\"\".x·2 *any, @\"\".y·3 *any) (? bool)\n" "func @\"\".memequal8 (@\"\".x·2 *any, @\"\".y·3 *any) (? bool)\n"
"func @\"\".memequal16 (@\"\".x·2 *any, @\"\".y·3 *any) (? bool)\n" "func @\"\".memequal16 (@\"\".x·2 *any, @\"\".y·3 *any) (? bool)\n"

View File

@ -1451,6 +1451,7 @@ void walkswitch(Node *sw);
* typecheck.c * typecheck.c
*/ */
int islvalue(Node *n); int islvalue(Node *n);
int samesafeexpr(Node *l, Node *r);
Node* typecheck(Node **np, int top); Node* typecheck(Node **np, int top);
void typechecklist(NodeList *l, int top); void typechecklist(NodeList *l, int top);
Node* typecheckdef(Node *n); Node* typecheckdef(Node *n);

View File

@ -144,6 +144,72 @@ walkrange(Node *n)
fatal("walkrange"); fatal("walkrange");
case TARRAY: case TARRAY:
// Lower n into runtime·memclr if possible, for
// fast zeroing of slices and arrays (issue 5373).
// Look for instances of
//
// for i := range a {
// a[i] = zero
// }
//
// in which the evaluation of a is side-effect-free.
if(!debug['N'])
if(!flag_race)
if(v1 != N)
if(v2 == N)
if(n->nbody != nil)
if(n->nbody->n != N) // at least one statement in body
if(n->nbody->next == nil) { // at most one statement in body
tmp = n->nbody->n; // first statement of body
if(tmp->op == OAS)
if(tmp->left->op == OINDEX)
if(samesafeexpr(tmp->left->left, a))
if(samesafeexpr(tmp->left->right, v1))
if(t->type->width > 0)
if(iszero(tmp->right)) {
// Convert to
// if len(a) != 0 {
// hp = &a[0]
// hn = len(a)*sizeof(elem(a))
// memclr(hp, hn)
// i = len(a) - 1
// }
n->op = OIF;
n->nbody = nil;
n->ntest = nod(ONE, nod(OLEN, a, N), nodintconst(0));
n->nincr = nil;
// hp = &a[0]
hp = temp(ptrto(types[TUINT8]));
tmp = nod(OINDEX, a, nodintconst(0));
tmp->bounded = 1;
tmp = nod(OADDR, tmp, N);
tmp = nod(OCONVNOP, tmp, N);
n->nbody = list(n->nbody, nod(OAS, hp, tmp));
// hn = len(a) * sizeof(elem(a))
hn = temp(types[TUINTPTR]);
tmp = nod(OLEN, a, N);
tmp = nod(OMUL, tmp, nodintconst(t->type->width));
tmp = conv(tmp, types[TUINTPTR]);
n->nbody = list(n->nbody, nod(OAS, hn, tmp));
// memclr(hp, hn)
fn = mkcall("memclr", T, nil, hp, hn);
n->nbody = list(n->nbody, fn);
// i = len(a) - 1
v1 = nod(OAS, v1, nod(OSUB, nod(OLEN, a, N), nodintconst(1)));
n->nbody = list(n->nbody, v1);
typecheck(&n->ntest, Erv);
typechecklist(n->nbody, Etop);
walkstmt(&n);
lineno = lno;
return;
}
}
// orderstmt arranged for a copy of the array/slice variable if needed. // orderstmt arranged for a copy of the array/slice variable if needed.
ha = a; ha = a;
hv1 = temp(types[TINT]); hv1 = temp(types[TINT]);

View File

@ -160,6 +160,7 @@ func block()
func makeslice(typ *byte, nel int64, cap int64) (ary []any) func makeslice(typ *byte, nel int64, cap int64) (ary []any)
func growslice(typ *byte, old []any, n int64) (ary []any) func growslice(typ *byte, old []any, n int64) (ary []any)
func memmove(to *any, frm *any, length uintptr) func memmove(to *any, frm *any, length uintptr)
func memclr(ptr *byte, length uintptr)
func memequal(x, y *any, size uintptr) bool func memequal(x, y *any, size uintptr) bool
func memequal8(x, y *any) bool func memequal8(x, y *any) bool

View File

@ -2831,7 +2831,7 @@ checkassignlist(NodeList *l)
// Check whether l and r are the same side effect-free expression, // Check whether l and r are the same side effect-free expression,
// so that it is safe to reuse one instead of computing both. // so that it is safe to reuse one instead of computing both.
static int int
samesafeexpr(Node *l, Node *r) samesafeexpr(Node *l, Node *r)
{ {
if(l->op != r->op || !eqtype(l->type, r->type)) if(l->op != r->op || !eqtype(l->type, r->type))

View File

@ -162,6 +162,20 @@ func BenchmarkMemclr256(b *testing.B) { bmMemclr(b, 256) }
func BenchmarkMemclr4096(b *testing.B) { bmMemclr(b, 4096) } func BenchmarkMemclr4096(b *testing.B) { bmMemclr(b, 4096) }
func BenchmarkMemclr65536(b *testing.B) { bmMemclr(b, 65536) } func BenchmarkMemclr65536(b *testing.B) { bmMemclr(b, 65536) }
func bmGoMemclr(b *testing.B, n int) {
x := make([]byte, n)
b.SetBytes(int64(n))
for i := 0; i < b.N; i++ {
for j := range x {
x[j] = 0
}
}
}
func BenchmarkGoMemclr5(b *testing.B) { bmGoMemclr(b, 5) }
func BenchmarkGoMemclr16(b *testing.B) { bmGoMemclr(b, 16) }
func BenchmarkGoMemclr64(b *testing.B) { bmGoMemclr(b, 64) }
func BenchmarkGoMemclr256(b *testing.B) { bmGoMemclr(b, 256) }
func BenchmarkClearFat8(b *testing.B) { func BenchmarkClearFat8(b *testing.B) {
for i := 0; i < b.N; i++ { for i := 0; i < b.N; i++ {
var x [8 / 4]uint32 var x [8 / 4]uint32

View File

@ -0,0 +1,71 @@
// run
// Copyright 2015 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.
// Ensure that zeroing range loops have the requisite side-effects.
package main
import (
"fmt"
"os"
)
func check(n int) {
// When n == 0, i is untouched by the range loop.
// Picking an initial value of -1 for i makes the
// "want" calculation below correct in all cases.
i := -1
s := make([]byte, n)
for i = range s {
s[i] = 0
}
if want := n - 1; i != want {
fmt.Printf("index after range with side-effect = %d want %d\n", i, want)
os.Exit(1)
}
i = n + 1
// i is shadowed here, so its value should be unchanged.
for i := range s {
s[i] = 0
}
if want := n + 1; i != want {
fmt.Printf("index after range without side-effect = %d want %d\n", i, want)
os.Exit(1)
}
// Index variable whose evaluation has side-effects
var x int
f := func() int {
x++
return 0
}
var a [1]int
for a[f()] = range s {
s[a[f()]] = 0
}
if want := n * 2; x != want {
fmt.Printf("index function calls = %d want %d\n", x, want)
os.Exit(1)
}
// Range expression whose evaluation has side-effects
x = 0
b := [1][]byte{s}
for i := range b[f()] {
b[f()][i] = 0
}
if want := n + 1; x != n+1 {
fmt.Printf("range expr function calls = %d want %d\n", x, want)
os.Exit(1)
}
}
func main() {
check(0)
check(1)
check(15)
}