code review 6625061: test/bench/shootout: match gcc architecture to GOARCH

test/bench/shootout: match gcc architecture to GOARCH

If we're benchmarking 8g, use gcc -m32.
If we're benchmarking 6g, use gcc -m64.

[rsc, 2012-10-07 19:32:07]

Hello golang-dev@googlegroups.com,

I'd like you to review this change to
https://go.googlecode.com/hg/

[bradfitz, 2012-10-07 19:34:29]

LGTM

On Sun, Oct 7, 2012 at 12:32 PM, <rsc@golang.org> wrote:

> Reviewers: golang-dev_googlegroups.com,
>
> Message:
> Hello golang-dev@googlegroups.com,
>
> I'd like you to review this change to
> https://go.googlecode.com/hg/
>
>
> Description:
> test/bench/shootout: match gcc architecture to GOARCH
>
> If we're benchmarking 8g, use gcc -m32.
> If we're benchmarking 6g, use gcc -m64.
>
> Please review this at
http://codereview.appspot.com/**6625061/<http://codereview.appspot.com/6625061/>
>
> Affected files:
>   M test/bench/shootout/timing.sh
>
>
> Index: test/bench/shootout/timing.sh
> ==============================**==============================**=======
> --- a/test/bench/shootout/timing.**sh
> +++ b/test/bench/shootout/timing.**sh
> @@ -10,6 +10,14 @@
>  GC="go tool ${O}g"
>  LD="go tool ${O}l"
>
> +gccm=""
> +case "$O" in
> +8)
> +       gccm=-m32;;
> +6)
> +       gccm=-m64;;
> +esac
> +
>  PATH=.:$PATH
>
>  havegccgo=false
> @@ -78,7 +86,7 @@
>
>  fasta() {
>         runonly echo 'fasta -n 25000000'
> -       run 'gcc -O2 fasta.c' a.out 25000000
> +       run "gcc $gccm -O2 fasta.c" a.out 25000000
>         run 'gccgo -O2 fasta.go' a.out -n 25000000      #commented out
> until WriteString is in bufio
>         run 'gc fasta' $O.out -n 25000000
>         run 'gc_B fasta' $O.out -n 25000000
> @@ -88,7 +96,7 @@
>         runonly gcc -O2 fasta.c
>         runonly a.out 25000000 > x
>         runonly echo 'reverse-complement < output-of-fasta-25000000'
> -       run 'gcc -O2 reverse-complement.c' a.out < x
> +       run "gcc $gccm -O2 reverse-complement.c" a.out < x
>         run 'gccgo -O2 reverse-complement.go' a.out < x
>         run 'gc reverse-complement' $O.out < x
>         run 'gc_B reverse-complement' $O.out < x
> @@ -97,7 +105,7 @@
>
>  nbody() {
>         runonly echo 'nbody -n 50000000'
> -       run 'gcc -O2 nbody.c -lm' a.out 50000000
> +       run "gcc $gccm -O2 nbody.c -lm" a.out 50000000
>         run 'gccgo -O2 nbody.go' a.out -n 50000000
>         run 'gc nbody' $O.out -n 50000000
>         run 'gc_B nbody' $O.out -n 50000000
> @@ -105,7 +113,7 @@
>
>  binarytree() {
>         runonly echo 'binary-tree 15 # too slow to use 20'
> -       run 'gcc -O2 binary-tree.c -lm' a.out 15
> +       run "gcc $gccm -O2 binary-tree.c -lm" a.out 15
>         run 'gccgo -O2 binary-tree.go' a.out -n 15
>         run 'gccgo -O2 binary-tree-freelist.go' a.out -n 15
>         run 'gc binary-tree' $O.out -n 15
> @@ -114,7 +122,7 @@
>
>  fannkuch() {
>         runonly echo 'fannkuch 12'
> -       run 'gcc -O2 fannkuch.c' a.out 12
> +       run "gcc $gccm -O2 fannkuch.c" a.out 12
>         run 'gccgo -O2 fannkuch.go' a.out -n 12
>         run 'gccgo -O2 fannkuch-parallel.go' a.out -n 12
>         run 'gc fannkuch' $O.out -n 12
> @@ -126,7 +134,7 @@
>         runonly gcc -O2 fasta.c
>         runonly a.out 100000 > x
>         runonly echo 'regex-dna 100000'
> -       run 'gcc -O2 regex-dna.c -lpcre' a.out <x
> +       run "gcc $gccm -O2 regex-dna.c -lpcre" a.out <x
>         run 'gccgo -O2 regex-dna.go' a.out <x
>         run 'gccgo -O2 regex-dna-parallel.go' a.out <x
>         run 'gc regex-dna' $O.out <x
> @@ -137,7 +145,7 @@
>
>  spectralnorm() {
>         runonly echo 'spectral-norm 5500'
> -       run 'gcc -O2 spectral-norm.c -lm' a.out 5500
> +       run "gcc $gccm -O2 spectral-norm.c -lm" a.out 5500
>         run 'gccgo -O2 spectral-norm.go' a.out -n 5500
>         run 'gc spectral-norm' $O.out -n 5500
>         run 'gc_B spectral-norm' $O.out -n 5500
> @@ -160,7 +168,7 @@
>
>  mandelbrot() {
>         runonly echo 'mandelbrot 16000'
> -       run 'gcc -O2 mandelbrot.c' a.out 16000
> +       run "gcc $gccm -O2 mandelbrot.c" a.out 16000
>         run 'gccgo -O2 mandelbrot.go' a.out -n 16000
>         run 'gc mandelbrot' $O.out -n 16000
>         run 'gc_B mandelbrot' $O.out -n 16000
> @@ -168,7 +176,7 @@
>
>  meteor() {
>         runonly echo 'meteor 2098'
> -       run 'gcc -O2 meteor-contest.c' a.out 2098
> +       run "gcc $gccm -O2 meteor-contest.c" a.out 2098
>         run 'gccgo -O2 meteor-contest.go' a.out -n 2098
>         run 'gc meteor-contest' $O.out -n 2098
>         run 'gc_B  meteor-contest' $O.out -n 2098
> @@ -176,7 +184,7 @@
>
>  pidigits() {
>         runonly echo 'pidigits 10000'
> -       run 'gcc -O2 pidigits.c -lgmp' a.out 10000
> +       run "gcc $gccm -O2 pidigits.c -lgmp" a.out 10000
>         run 'gccgo -O2 pidigits.go' a.out -n 10000
>         run 'gc pidigits' $O.out -n 10000
>         run 'gc_B  pidigits' $O.out -n 10000
> @@ -184,14 +192,14 @@
>
>  threadring() {
>         runonly echo 'threadring 50000000'
> -       run 'gcc -O2 threadring.c -lpthread' a.out 50000000
> +       run "gcc $gccm -O2 threadring.c -lpthread" a.out 50000000
>         run 'gccgo -O2 threadring.go' a.out -n 50000000
>         run 'gc threadring' $O.out -n 50000000
>  }
>
>  chameneos() {
>         runonly echo 'chameneos 6000000'
> -       run 'gcc -O2 chameneosredux.c -lpthread' a.out 6000000
> +       run "gcc $gccm -O2 chameneosredux.c -lpthread" a.out 6000000
>         run 'gccgo -O2 chameneosredux.go' a.out 6000000
>         run 'gc chameneosredux' $O.out 6000000
>  }
>
>
>

[minux1, 2012-10-07 19:36:01]

as 5g/5c use ARM instructions, it makes sense to force gcc
to do the same.

http://codereview.appspot.com/6625061/diff/4001/test/bench/shootout/timing.sh
File test/bench/shootout/timing.sh (right):

http://codereview.appspot.com/6625061/diff/4001/test/bench/shootout/timing.sh...
test/bench/shootout/timing.sh:18: gccm=-m64;;
5)
   gccm=-marm;;

[remyoudompheng, 2012-10-07 19:37:23]

Should we start a timing.log for GOARCH=386 with some chosen measure points?

[rsc, 2012-10-07 19:38:22]

On Sun, Oct 7, 2012 at 3:37 PM,  <remyoudompheng@gmail.com> wrote:
> Should we start a timing.log for GOARCH=386 with some chosen measure
> points?

I don't think we need to do that, but I do think we should copy one or
two of the FP tests into test/bench/go1.

Russ

[rsc, 2012-10-07 19:49:58]

*** Submitted as http://code.google.com/p/go/source/detail?r=d2bdff3cc2c3 ***

test/bench/shootout: match gcc architecture to GOARCH

If we're benchmarking 8g, use gcc -m32.
If we're benchmarking 6g, use gcc -m64.

R=golang-dev, bradfitz, minux.ma, remyoudompheng
CC=golang-dev
http://codereview.appspot.com/6625061