QueueImage slow when drawing textures

Programming Help
, ,
1

Hello, I’m trying to use PTB with VPixx to draw 12 frames at 1440Hz using the automatic frame assignment (with QueueImage). I expect the drawing of the frames to take place in less than 0.0083s. When measuring the timings with fill oval, the timings worked as expected. However, when measuring the timings with draw texture (what I need to run my experiment), the timings of every 12th QueueImage take about 0.03s. This is about 4 times slower than it should be. The images I am drawing are small (3 images of about 40KB each), and when I draw just one of these, the timings are still 4 times slower. It seems like the hardware is okay because fill oval works as expected, but I’m not sure why draw texture does not.

Here is a link to a github repository with my sample code.

PsychtoolboxVersion: '3.0.18 - Flavor: beta - Corresponds to SVN Revision 13005

I am doing this on a Mac Mini running Linux

Hardware versions:
H/W path Device Class Description

                          system         Macmini7,1 (System SKU#)

/0 bus Mac-35C5E08120C7EEAF
/0/0 processor Intel(R) Core™ i5-4308U CPU @ 2.
/0/0/2 memory 64KiB L1 cache
/0/0/3 memory 512KiB L2 cache
/0/0/4 memory 3MiB L3 cache
/0/1 memory 64KiB L1 cache
/0/5 memory 8GiB System Memory
/0/5/0 memory 4GiB SODIMM DDR3 Synchronous 1600 M
/0/5/1 memory 4GiB SODIMM DDR3 Synchronous 1600 M
/0/b memory 1MiB BIOS
/0/2fdd memory 1019KiB BIOS
/0/100 bridge Haswell-ULT DRAM Controller
/0/100/2 display Haswell-ULT Integrated Graphics Con
/0/100/3 multimedia Haswell-ULT HD Audio Controller
/0/100/14 bus 8 Series USB xHCI HC
/0/100/14/0 usb1 bus xHCI Host Controller
/0/100/14/0/1 bus Mini DisplayPort to Dual-Link DVI A
/0/100/14/0/2 generic PROPixx
/0/100/14/0/3 bus BRCM20702 Hub
/0/100/14/0/3/3 communication Bluetooth USB Host Controller
/0/100/14/0/4 input IR Receiver
/0/100/14/0/7 bus USB2.0 Hub
/0/100/14/0/7/2 input DELL USB Keyboard
/0/100/14/0/7/3 generic PROPixxCtrl
/0/100/14/0/7/4 input USB Optical Mouse
/0/100/14/1 usb2 bus xHCI Host Controller
/0/100/14/1/3 bus USB3.0 Hub
/0/100/16 communication 8 Series HECI #0
/0/100/1b multimedia 8 Series HD Audio Controller
/0/100/1c bridge 8 Series PCI Express Root Port 1
/0/100/1c.2 bridge 8 Series PCI Express Root Port 3
/0/100/1c.2/0 wlp2s0 network BCM4360 802.11ac Wireless Network A
/0/100/1c.3 bridge 8 Series PCI Express Root Port 4
/0/100/1c.3/0 enp3s0f0 network NetXtreme BCM57766 Gigabit Ethernet
/0/100/1c.3/0.1 generic BCM57765/57785 SDXC/MMC Card Reader
/0/100/1c.4 bridge 8 Series PCI Express Root Port 5
/0/100/1c.4/0 bridge DSL5520 Thunderbolt 2 Bridge [Falco
/0/100/1c.4/0/0 bridge DSL5520 Thunderbolt 2 Bridge [Falco
/0/100/1c.4/0/0/0 generic DSL5520 Thunderbolt 2 NHI [Falcon R
/0/100/1c.4/0/3 bridge DSL5520 Thunderbolt 2 Bridge [Falco
/0/100/1c.4/0/4 bridge DSL5520 Thunderbolt 2 Bridge [Falco
/0/100/1c.4/0/5 bridge DSL5520 Thunderbolt 2 Bridge [Falco
/0/100/1c.4/0/6 bridge DSL5520 Thunderbolt 2 Bridge [Falco
/0/100/1c.5 bridge 8 Series PCI Express Root Port 6
/0/100/1c.5/0 storage Samsung Electronics Co Ltd
/0/100/1f bridge 8 Series LPC Controller
/0/100/1f.3 bus 8 Series SMBus Controller
/0/2 system PnP device PNP0103
/0/3 system PnP device PNP0c02
/0/4 system PnP device PNP0b00
/0/6 system PnP device PNP0c02
/0/7 system PnP device PNP0c01
/0/8 scsi0 storage
/0/8/0.0.0 /dev/sda disk 251GB APPLE SSD SM0256
/0/8/0.0.0/1 volume 200MiB EFI GPT partition
/0/8/0.0.0/2 /dev/sda2 volume 135GiB Darwin/OS X HFS+ partition
/0/8/0.0.0/3 /dev/sda3 volume 619MiB Darwin/OS X HFS+ partition
/0/8/0.0.0/4 /dev/sda4 volume 11GiB Linux swap volume

2

Hello Heather,

observing a spike in processing time every 12th PsychProPixx(‘QueueImage’) call is expected, as every 12 images will be packed into the final image output buffer and flipped in 1440 Hz output mode.

That said, I quickly skimmed your code and there is lots of room for improvements, and your time measurement method is flawed. Because you made the great choice of using Linux as OS in combination with a non-Nvidia graphics card - Intel Haswell in your case, even more optimizations are possible on Linux than on other OS’es.

That said, this is quite a demanding task for a Intel onboard graphics chip from around the year 2014, and normally I’d recommend a more powerful discrete AMD gpu under Linux for this use case. Properly optimized code may work on your hardware setup, on a modern Linux distribution like Ubuntu 22.04.1-LTS or at least 20.04-LTS, but this is not guaranteed.

If you want advice optimizing the code, help PsychPaidSupportAndServices will tell you how to request 30 minutes of paid support.

-mario