How do I read the output of the Pulse Logger?
swpalmer at gmail.com
Tue Mar 15 18:23:32 UTC 2016
In my test I’m running the same application on Windows and Mac, making the same changes to the the scene graph, with the same stylesheets. So I don’t understand the difference I’m seeing.
> On Mar 15, 2016, at 12:33 PM, David Grieve <david.grieve at oracle.com> wrote:
> Sometimes the layout might introduce nodes into the scenegraph. If these new nodes also need laid out, CSS is applied to those nodes since style can affect layout. I would expect CSS overhead to be very small unless there are many new nodes being added to the scene (https://bugs.openjdk.java.net/browse/JDK-8151756 notwithstanding)
> On 3/15/16 12:14 PM, Scott Palmer wrote:
>> Is there a guideline somewhere that explains how to read the output of the Pulse Logger?
>> For example, what do the two times represent in this PULSE line:
>> PULSE: 569 [1459ms:270ms]
>> At first I guessed is that it was PULSE: <current_pulse_#> [<layout time>ms:<time_for_everything_else>]
>> but that doesn’t seem to hold up.
>> What does the first (usually negative) number mean in the Layout Pass line?
>> T23 (-1563 +1563ms): Layout Pass
>> T23 (1 +4ms): Layout Pass
>> I’ve noticed that the output has some notable differences between my Windows machine and my Mac. On the Mac I don’t seem to be getting the same sort of CSS Pass information. I get only one “CSS Pass” per pulse and it is almost always telling me something near 0ms:
>> PULSE: 509 [1404ms:286ms]
>> T23 (-1404 +1404ms): Layout Pass
>> T23 (0 +1ms): CSS Pass
>> T23 (1 +6ms): Layout Pass
>> T23 (7 +4ms): Update bounds
>> T23 (11 +0ms): Waiting for previous rendering
>> T23 (11 +0ms): Copy state to render graph
>> but on Windows I get two CSS Passes, one is usually 0ms, the other makes more sense (I’m investigating a performance issue on a Scene with >10k nodes, so some notable amount of CSS time is expected.):
>> PULSE: 2578 [423ms:3362ms]
>> T35 (-423 +423ms): Layout Pass
>> T35 (0 +0ms): CSS Pass
>> T35 (0 +0ms): Layout Pass
>> T35 (0 +2725ms): CSS Pass
>> T35 (2725 +402ms): Layout Pass
>> T35 (3128 +25ms): Update bounds
>> This seems a bit strange, as I would think Layout and CSS would not be platform specific. Both systems are running 8u72.
>> I’m also finding Layout times on the Mac are higher than I expected - maybe some CSS time has been rolled in to that log line on the Mac?
More information about the openjfx-dev