11 years, 7 months ago
(2013-02-14 07:13:24 UTC)
#1
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
File
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/command_line
(right):
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/command_line:1:
python bench/bench_graph_svg.py -d
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/raw-bench-data
-r -150 -f -150 -x 1024 -y 768 -l Title -m 25th -o
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-actual/graph.xhtml
Toplevel comments:
My goal with this CL is to create an easy way to test speedups of
bench_graph_svg.py, while making sure that the results are unchanged. (or at
least any changes to the results will be flagged)
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
File
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/graph.xhtml
(right):
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/graph.xhtml:1:
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
I captured this output from my local run... given the same input files, I think
it should always generate this same output.
I took a quick look at the graph and it seemed reasonable, but I didn't dig into
it too far.
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
File
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/tarballs/README.sh
(right):
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/tarballs/README.sh:15:
FILES="bench_r${REV}_data_skp_device_bitmap_multi_2_mode_tile_256_256_timeIndividualTiles
bench_r${REV}_data_skp_device_bitmap_multi_3_mode_tile_256_256_timeIndividualTiles
bench_r${REV}_data_skp_device_bitmap_multi_4_mode_tile_256_256_timeIndividualTiles"
With this small dataset (captured when we were recording all the per-tile and
per-iteration numbers), it takes 39 seconds to run bench_graph_svg.py! So there
is hopefully room for improvement.
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/tarballs/README.sh:31:
tar --create --gzip --file ${TARBALL_DIR}/r${REV}.tgz bench_r${REV}_*
I stored the tarred up bench data, rather than the original data files, to save
space in the repository. It takes up >90% less space this way.
If anyone thinks it would be better to store the raw bench data instead, I can
certainly make that change.
LGTM. Thanks for setting this up! https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/graph.xhtml File tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/graph.xhtml (right): https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/graph.xhtml#newcode1 tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/graph.xhtml:1: <!DOCTYPE html PUBLIC ...
11 years, 7 months ago
(2013-02-14 15:09:03 UTC)
#2
LGTM. Thanks for setting this up!
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
File
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/graph.xhtml
(right):
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/output-expected/graph.xhtml:1:
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
I also believe the output is deterministic for reasonable inputs.
On 2013/02/14 07:13:24, epoger wrote:
> I captured this output from my local run... given the same input files, I
think
> it should always generate this same output.
>
> I took a quick look at the graph and it seemed reasonable, but I didn't dig
into
> it too far.
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
File
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/tarballs/README.sh
(right):
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/tarballs/README.sh:15:
FILES="bench_r${REV}_data_skp_device_bitmap_multi_2_mode_tile_256_256_timeIndividualTiles
bench_r${REV}_data_skp_device_bitmap_multi_3_mode_tile_256_256_timeIndividualTiles
bench_r${REV}_data_skp_device_bitmap_multi_4_mode_tile_256_256_timeIndividualTiles"
I always suspect it's the regex matching that takes lots of time.
On 2013/02/14 07:13:24, epoger wrote:
> With this small dataset (captured when we were recording all the per-tile and
> per-iteration numbers), it takes 39 seconds to run bench_graph_svg.py! So
there
> is hopefully room for improvement.
https://codereview.appspot.com/7319044/diff/1/tools/tests/benchgraphs/Skia_Sh...
tools/tests/benchgraphs/Skia_Shuttle_Ubuntu12_ATI5770_Float_Bench_32/tarballs/README.sh:31:
tar --create --gzip --file ${TARBALL_DIR}/r${REV}.tgz bench_r${REV}_*
I think it's good to store .tgz instead.
On 2013/02/14 07:13:24, epoger wrote:
> I stored the tarred up bench data, rather than the original data files, to
save
> space in the repository. It takes up >90% less space this way.
>
> If anyone thinks it would be better to store the raw bench data instead, I can
> certainly make that change.
Issue 7319044: tools/tests/run.sh: add self-test for bench_graph_svg.py
(Closed)
Created 11 years, 7 months ago by epoger
Modified 11 years, 7 months ago
Reviewers: benchen
Base URL: http://skia.googlecode.com/svn/trunk/
Comments: 7