New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.
Starred by 6 users
Status: New
Owner:
Cc:
Area: ----
Priority: Medium
Type: Defect



Sign in to add a comment
Perspective view testing
Project Member Reported by edisonn@google.com, Sep 20 2013 Back to list
FYI:

I created a CL to apply perspective on all GMs (run it without pdf, so far pdf does not supports perspective so it asserts)

there are 2 gms that are crashing, and one that does not show the same results for 8888 and GPU (see inverse_path)

I am thinking of a "cheap" way to test perspective: GM + Perspective should be very close to "Image from gm" + perspective (and ignore the bounds)

The results of the GM+Perspective should be at least once manually analyzed

Edi
 
Comment 1 by epoger@google.com, Sep 23 2013
Status: New
Edi, I'm afraid your report provides more HINTS at information than information itself.

1. Please provide a link to your CL

2. Which 2 gms are crashing?  Please paste in or attach the output that leads you to believe they are crashing.

3. What do the inverse_path results look like?  You've already run it, so can you please attach the 8888 and GPU images to this bug?

4. Can you please re-explain what you mean by "GM + Perspective should be very close to "Image from gm" + perspective (and ignore the bounds)"?  What exactly are you proposing?
Project Member Comment 2 by edisonn@google.com, Sep 23 2013
1)
https://codereview.chromium.org/23587029/

2)
Crashing:
gm/tilemodes_scaled.cpp
gm/convexpaths.cpp

3) attached

4) If I run the gm normally, and we apply perspective on the resulted image, it should be similar with the image outputed if i run the gm with the perspective flag
inverse_paths_8888.png
15.4 KB View Download
inverse_paths_gpu.png
15.4 KB View Download
Project Member Comment 3 by bsalo...@google.com, Sep 23 2013
Cc: bsalo...@google.com
It appears that they draw differently in GPU vs 8888 without perspective as well.
Comment 4 by epoger@google.com, Sep 23 2013
Fleshing out comment 3:

1. Go to https://skia-autogen.googlecode.com/svn/gm-actual/view.html
2. Select Test-Ubuntu12-ShuttleA-ATI5770-x86-Release from the drop-down
3. Search for inverse_paths
3a. 8888 output: http://chromium-skia-gm.commondatastorage.googleapis.com/gm/bitmap-64bitMD5/inverse_paths/5665381282859861524.png
3b. gpu output: http://chromium-skia-gm.commondatastorage.googleapis.com/gm/bitmap-64bitMD5/inverse_paths/946773628625952905.png

As Brian noted, the 8888 and gpu outputs look different (in the same way that the ones Edi attached do), even without perspective applied.
Project Member Comment 5 by edisonn@google.com, Oct 24 2013
update, there are 3 gms that crash
scaled_tilemodes
conexpath
clipped-bitmap

the diffs are expected, but the crashes need to be fixed

to repro the crash run the gm with --forcePerspectiveMatrix, as the flag is checked in
Project Member Comment 6 by edisonn@google.com, Oct 31 2013
and xfermodes3
Project Member Comment 7 by hcm@google.com, Mar 19 2014
Cc: bungeman@google.com
Comment 8 by epoger@google.com, Aug 18 2014
Cc: -epoger@google.com
Project Member Comment 9 by hcm@google.com, Dec 7 2015
Labels: Hotlist-Fixit
Sign in to add a comment