Pybullet Debug Text increases render time 323x

[ghost]

Hi,

I'm noticed that using p.addUserDebugText() makes my rendering (using p.connect(p.GUI)) extremely slow. Is there a way to speed this up? Is there perhaps an issue with my system (running a 2060 super, more details in terminal output below).

Attached is a zip file with a script to reproduce the issue.
test_text_rendering.zip

I have also tried turning all of the following on and off, without any change on the order of magnitude that adding the text makes.

self.client.configureDebugVisualizer(p.COV_ENABLE_WIREFRAME,1)
self.client.configureDebugVisualizer(p.COV_ENABLE_SHADOWS, 0)
self.client.configureDebugVisualizer(p.COV_ENABLE_GUI, 0)
self.client.configureDebugVisualizer(p.COV_ENABLE_TINY_RENDERER, 0)

Here is my terminal output from running the script:

python test_text_rendering.py
pybullet build time: Dec 23 2020 01:47:29
startThreads creating 1 threads.
starting thread 0
started thread 0 
argc=2
argv[0] = --unused
argv[1] = --start_demo_name=Physics Server
ExampleBrowserThreadFunc started
X11 functions dynamically loaded using dlopen/dlsym OK!
X11 functions dynamically loaded using dlopen/dlsym OK!
Creating context
Created GL 3.3 context
Direct GLX rendering context obtained
Making context current
GL_VENDOR=NVIDIA Corporation
GL_RENDERER=GeForce RTX 2060 SUPER/PCIe/SSE2
GL_VERSION=3.3.0 NVIDIA 460.32.03
GL_SHADING_LANGUAGE_VERSION=3.30 NVIDIA via Cg compiler
pthread_getconcurrency()=0
Version = 3.3.0 NVIDIA 460.32.03
Vendor = NVIDIA Corporation
Renderer = GeForce RTX 2060 SUPER/PCIe/SSE2
b3Printf: Selected demo: Physics Server
startThreads creating 1 threads.
starting thread 0
started thread 0 
MotionThreadFunc thread started
ven = NVIDIA Corporation
ven = NVIDIA Corporation

####################################################################################################
Average render time per sim step WITHOUT text: 4.35662269592285e-05 s
Average render time per sim step WITH text: 0.014121932983398434 s
Increase in render time with text: 0.014078366756439205 s
323x Increase
####################################################################################################

numActiveThreads = 0
stopping threads
Thread with taskId 0 exiting
Thread TERMINATED
destroy semaphore
semaphore destroyed
destroy main semaphore
main semaphore destroyed
finished
numActiveThreads = 0
btShutDownExampleBrowser stopping threads
Thread with taskId 0 exiting
Thread TERMINATED
destroy semaphore
semaphore destroyed
destroy main semaphore
main semaphore destroyed

[rohit-kumar-j]

@jcoholich
I have had a similar issue. It seems like the script re-creates, re-renders, and re-destroys the text in each loop. Hence, the performance drop:

for _ in range(iters):
    p.stepSimulation()
    
for _ in range(iters):
    p.stepSimulation()
    p.addUserDebugText(....)

Solution:
See debugDrawItems.py
This script treats the userDebugItem as a child of the robot, also this is done not in each loop, but at the beginning of the script.

kuka = p.loadURDF("kuka_iiwa/model.urdf")
p.addUserDebugText("tip", [0, 0, 0.1],
                   textColorRGB=[1, 0, 0],
                   textSize=1.5,
                   parentObjectUniqueId=kuka,    <----
                   parentLinkIndex=6)            <----
loop():...

Scores:

####################################################################################################
Average render time per sim step WITHOUT text: 5.0001144409179674e-05 s
Average render time per sim step WITH text: 5.000352859497071e-05 s
Increase in render time with text: 2.3841857910337854e-09 s
0x Increase
####################################################################################################
Average render time per sim step WITHOUT text: 0.00010002613067626954 s
Average render time per sim step WITH text: 9.004831314086913e-05 s
Increase in render time with text: -9.97781753540041e-06 s
-0x Increase
####################################################################################################