code review 166076: Make printing faster by avoiding mallocs and some other...

Make printing faster by avoiding mallocs and some other advances.
Roughly 33% faster for simple cases, probably more for complex ones.

Before:

mallocs per Sprintf(""): 4
mallocs per Sprintf("xxx"): 6
mallocs per Sprintf("%x"): 10
mallocs per Sprintf("%x %x"): 12

Now:

mallocs per Sprintf(""): 2
mallocs per Sprintf("xxx"): 3
mallocs per Sprintf("%x"): 5
mallocs per Sprintf("%x %x"): 7

Speed improves because of avoiding mallocs and also by sharing a bytes.Buffer
between print.go and format.go rather than copying the data back after each
printed item.

Before:

fmt_test.BenchmarkSprintfEmpty	1000000	      1346 ns/op
fmt_test.BenchmarkSprintfString	500000	      3461 ns/op
fmt_test.BenchmarkSprintfInt	500000	      3671 ns/op

Now:

fmt_test.BenchmarkSprintfEmpty	 2000000	       995 ns/op
fmt_test.BenchmarkSprintfString	 1000000	      2745 ns/op
fmt_test.BenchmarkSprintfInt	 1000000	      2391 ns/op
fmt_test.BenchmarkSprintfIntInt	  500000	      3751 ns/op


I believe there is more to get but this is a good milestone.

[r, 2009-12-06 06:30:45]

Hello rsc (cc: golang-nuts@googlegroups.com),

I'd like you to review the following change.

[hong, 2009-12-06 13:42:13]

http://codereview.appspot.com/166076/diff/1001/13
File src/pkg/fmt/format.go (right):

http://codereview.appspot.com/166076/diff/1001/13#newcode263
src/pkg/fmt/format.go:263: f.padBytes(f.integer(v, 16, true, ldigits));
Make padBytes() take "clear bool". It will reduce caller code.

http://codereview.appspot.com/166076/diff/1001/13#newcode419
src/pkg/fmt/format.go:419: s = "+" + s
String cat seems expensive, we can let pad handle it by adding another boolean
flag. In that case, caller can call pad directly and we don't even need this
function.

[rsc, 2009-12-06 16:22:03]

LGTM

http://codereview.appspot.com/166076/diff/1001/13
File src/pkg/fmt/format.go (right):

http://codereview.appspot.com/166076/diff/1001/13#newcode41
src/pkg/fmt/format.go:41: wid_present	bool;
dregs

widPresent
precPresent

http://codereview.appspot.com/166076/diff/1001/13#newcode103
src/pkg/fmt/format.go:103: if w > 0 {
It bothers me that %100d doesn't work and that
this code is duplicated, since there have been bugs
in it in the past.  I suggest breaking this into

func (f *Fmt) computePad(n int) (leftPad, rightPad int)
func (f *Fmt) writePad(npad int)

where writePad is this if statement but with a for loop.

Then padBytes and pad would shorten to:

left, right := f.computePad(len(b));
if left > 0 {
    f.writePad(left);
}
f.buf.Write(b);
if right > 0 {
    f.writePad(right);
}

http://codereview.appspot.com/166076/diff/1001/14
File src/pkg/fmt/print.go (right):

http://codereview.appspot.com/166076/diff/1001/14#newcode162
src/pkg/fmt/print.go:162: func (p *pp) free()	{ _ = ppFree <- p }
If the buffer is big we should probably drop it on the floor.

b.Reset();
if cap(b.Data()) > 1024 {
    return;
}
_ = ppFree <- p

[r2, 2009-12-06 17:37:45]

there's still lots to do in here, many optimizations, including eliminating
string concatenation. this is just a snapshot.

-rob

[r, 2009-12-06 20:03:58]

*** Submitted as http://code.google.com/p/go/source/detail?r=f967d18ba6d4 ***

Make printing faster by avoiding mallocs and some other advances.
Roughly 33% faster for simple cases, probably more for complex ones.

Before:

mallocs per Sprintf(""): 4
mallocs per Sprintf("xxx"): 6
mallocs per Sprintf("%x"): 10
mallocs per Sprintf("%x %x"): 12

Now:

mallocs per Sprintf(""): 2
mallocs per Sprintf("xxx"): 3
mallocs per Sprintf("%x"): 5
mallocs per Sprintf("%x %x"): 7

Speed improves because of avoiding mallocs and also by sharing a bytes.Buffer
between print.go and format.go rather than copying the data back after each
printed item.

Before:

fmt_test.BenchmarkSprintfEmpty	1000000	      1346 ns/op
fmt_test.BenchmarkSprintfString	500000	      3461 ns/op
fmt_test.BenchmarkSprintfInt	500000	      3671 ns/op

Now:

fmt_test.BenchmarkSprintfEmpty	 2000000	       995 ns/op
fmt_test.BenchmarkSprintfString	 1000000	      2745 ns/op
fmt_test.BenchmarkSprintfInt	 1000000	      2391 ns/op
fmt_test.BenchmarkSprintfIntInt	  500000	      3751 ns/op


I believe there is more to get but this is a good milestone.

R=rsc
CC=golang-dev, hong
http://codereview.appspot.com/166076