code review 144250043: fmt: make printing of ints 25-35% faster

fmt: make printing of ints 25-35% faster
Inspired by a remark by Leonard Holz, use constants for division

BenchmarkSprintfEmpty           130           132           +1.54%
BenchmarkSprintfString          438           437           -0.23%
BenchmarkSprintfInt             417           414           -0.72%
BenchmarkSprintfIntInt          663           691           +4.22%
BenchmarkSprintfPrefixedInt     791           774           -2.15%
BenchmarkSprintfFloat           701           686           -2.14%
BenchmarkManyArgs               2584          2469          -4.45%
BenchmarkFprintInt              488           357           -26.84%
BenchmarkFprintIntNoAlloc       402           265           -34.08%
BenchmarkScanInts               1244346       1267574       +1.87%
BenchmarkScanRecursiveInt       1748741       1724138       -1.41%

Update issue 3463

[r, 2014-09-19 21:04:43]

Hello golang-codereviews@googlegroups.com,

I'd like you to review this change to
https://code.google.com/p/go/

[josharian, 2014-09-19 23:28:44]

Should this wait until post-freeze?

https://codereview.appspot.com/144250043/diff/1/src/fmt/format.go
File src/fmt/format.go (right):

https://codereview.appspot.com/144250043/diff/1/src/fmt/format.go#newcode208
src/fmt/format.go:208: buf[i] = digits[ua%10]
For cases 10, 8, and 2, switching from

digits[ua%10]

to

byte('0'+ua%10)

should be marginally faster -- no bounds check.

[r, 2014-09-20 18:47:34]

Hello golang-codereviews@googlegroups.com, josharian@gmail.com (cc:
golang-codereviews@googlegroups.com),

Please take another look.

[r, 2014-09-20 18:47:52]

https://codereview.appspot.com/144250043/diff/1/src/fmt/format.go
File src/fmt/format.go (right):

https://codereview.appspot.com/144250043/diff/1/src/fmt/format.go#newcode208
src/fmt/format.go:208: buf[i] = digits[ua%10]
worth another 1.5%.

[josharian, 2014-09-22 18:12:21]

LGTM

https://codereview.appspot.com/144250043/diff/20001/src/fmt/format.go
File src/fmt/format.go (right):

https://codereview.appspot.com/144250043/diff/20001/src/fmt/format.go#newcode202
src/fmt/format.go:202: // use constants for the division and modulo; code is
much better.
Maybe s/code/generated code/

[rsc, 2014-09-22 18:41:16]

https://codereview.appspot.com/144250043/diff/20001/src/fmt/format.go
File src/fmt/format.go (right):

https://codereview.appspot.com/144250043/diff/20001/src/fmt/format.go#newcode215
src/fmt/format.go:215: ua >>= 4
this should be the same as ua /= 16
similarly below

[r, 2014-09-22 18:55:04]

Hello golang-codereviews@googlegroups.com, josharian@gmail.com, rsc@golang.org
(cc: golang-codereviews@googlegroups.com),

Please take another look.

[r, 2014-09-22 18:56:50]

https://codereview.appspot.com/144250043/diff/20001/src/fmt/format.go
File src/fmt/format.go (right):

https://codereview.appspot.com/144250043/diff/20001/src/fmt/format.go#newcode215
src/fmt/format.go:215: ua >>= 4
i prefer it as is, nice and easy to read, and now that base 10 is more complex,
the argument for parallel design is gone.

[rsc, 2014-09-22 18:57:22]

LGTM

yes, with the new %-removal the & and >> are better

[r, 2014-09-22 18:58:09]

Let's not wait until post-freeze because the speedup will offset the
overhead from allocating ints in interfaces, a significant cost in
fmt.

[r, 2014-09-22 18:58:20]

*** Submitted as https://code.google.com/p/go/source/detail?r=f0cbfc3381c4 ***

fmt: make printing of ints 25-35% faster
Inspired by a remark by Leonard Holz, use constants for division

BenchmarkSprintfEmpty           130           132           +1.54%
BenchmarkSprintfString          438           437           -0.23%
BenchmarkSprintfInt             417           414           -0.72%
BenchmarkSprintfIntInt          663           691           +4.22%
BenchmarkSprintfPrefixedInt     791           774           -2.15%
BenchmarkSprintfFloat           701           686           -2.14%
BenchmarkManyArgs               2584          2469          -4.45%
BenchmarkFprintInt              488           357           -26.84%
BenchmarkFprintIntNoAlloc       402           265           -34.08%
BenchmarkScanInts               1244346       1267574       +1.87%
BenchmarkScanRecursiveInt       1748741       1724138       -1.41%

Update issue 3463

LGTM=josharian, rsc
R=golang-codereviews, josharian, rsc
CC=golang-codereviews
https://codereview.appspot.com/144250043