Iris Xe Graphics Memory Leak 64bit calling Direct2D BeginDraw EndDraw

[CaptBridgeport]

I believe I found another issue with Iris Xe Graphics.  After investigating a very bad leak, I narrowed it down to BeginDraw and EndDraw.  No other graphics driver, and all 32 bit has no issue.  Only the Iris Xe driver serving Direct2D graphics to a 64 bit exe has the issue.   Is this a known issue?  This is much worse an issue than my previous find https://community.intel.com/t5/Graphics/Iris-Xe-Help-Emulator-Most-common-failure-create-D3Device/m-p/1295201#M98580  as I haven't found a way to avoid the leak.  

[DeividA_Intel]

Hello CaptBridgeport,  

Thank you for posting on the Intel® communities.   

In order to better assist you, please provide the following:  

1. Run the Intel® System Support Utility (Intel® SSU) to gather more details about the system.  

· Download the Intel® SSU and save the application on your computer: https://downloadcenter.intel.com/download/25293/Intel-System-Support-Utility-for-Windows-  

· Open the application, check the "Everything" checkbox, and click "Scan" to see the system and device information. The Intel® SSU defaults to the "Summary View" on the output screen following the scan. Click the menu where it says "Summary" to change to "Detailed View".  

· To save your scan, click Next and click Save.  

2. Can you share more details of the issue?

3. What troubleshooting steps have you takes so far?

4. Is this issue recent? If so, when did start?

5. Can you take pictures or videos of the issue?

Regards,    

Deivid A. 

Intel Customer Support Technician 

[CaptBridgeport]

Hello

Thank you for responding.  I've already donated way too much of my time because of Intel.  If  you are truly wanting to research this leak, which you should, as it's horrific, 

1) download our ProEssentials v9 Pro evaluation edition (no hassle download)

2) Compile our Visual Studio demo project in the VC subfolder (its defaults to x86 32 bit and uses PEGRP32G.DLL found in syswow64).  

3) Compile the demo project for Win64/x64 (uses our PEGRP64G.DLL found in system32) 

Place all these files on a flash drive, keeping the 32 bit and 64 bit portions separate.  

This should take 5 minutes, compiling, and copying the files to  a flash drive so you can test on a few systems, Iris Xe and Non-Iris Xe. 

Run and test 32bit example 115 of the demo (test on any chip/system that uses Iris Xe Graphics, no leak, or on any system 32 bit, no leak) 

Test the 64 bit exe and DLL on any system other than one that uses Iris Xe Graphics (no leak.)   Test 64bit on Iris Xe system HUGE LEAK. 

The technology here is a Direct3D rendering loop, where Direct2D is called via  

CreateDxgiSurfaceRenderTarget as shown on MS's site  https://docs.microsoft.com/en-us/windows/win32/Direct2D/direct2d-and-direct3d-interoperation-overview

Again, the leak is horrible. And very ease to prove to yourselves with our demo project.  Or create your own simple 64 bit Direct3D render loop that calls CreateDxgiSurfaceRenderTarget to add a few 2d lines.  My experiments saw creating a simple solid black brush and drawing 10 lines (grid lines) easily showed the leak.  Example 115 draws grid lines, draws a few D2D polygons (grid area and legends) , and outputs some Direct2D Text grid line labels, and it leaks horribly only on 64 bit exes and Iris Xe Graphics.  

I did a quick search for CreateDxgiSurfaceRenderTarget on Github and found   https://github.com/microsoft/Windows-classic-samples/blob/master/Samples/Win7Samples/multimedia/Direct2D/DXGISample/DxgiSample.cpp    There may be others but this does look promising it might replicate the issue.    Maybe look at this once you confirm above there is an issue.   Or find other examples of CreateDxgiSurfaceRenderTarget combined with Direct3D device. 

Please research and fix asap.  This is a critical issue and should be given highest priority.  It's likely an easy bug to find given the 32 bit Iris Xe driver does not have an issue and the 64 bit driver does. 

[DeividA_Intel]

Hello CaptBridgeport, 

Thanks for the information provided.

We will try to investigate further this issue, however, to know better your system I will appreciate it if you help me with the following:

1. Attached the report from the Intel® System Support Utility,

- https://downloadcenter.intel.com/download/25293/Intel-System-Support-Utility-for-Windows-

2. Is this issue recent? If so, when did start?

3. Can you take pictures or videos of the issue?

Regards,   

Deivid A. 

Intel Customer Support Technician

[CaptBridgeport]

Here you go,  please  promptly investigate and fix.  All Iris systems behave this way along with the other driver issues. I  had to buy a random refurbished $499 Acer i5 with Iris Xe Graphics to debug your hardware/software.  But my customer had a new $1200 HP i7.  I assume tons of these systems are being returned.  Intel will be lucky if this does not blow up into a class action lawsuit if a fix doesn't show soon. Pass this on to your manager and his manager and on up to the top. 

[n_scott_pearson]

I do so enjoy the backwoods lawyers who come here and threatening lawsuits if they do not receive instant satisfaction - especially when they are people that don't even have a warranty from Intel! You have an Acer laptop? Then your warranty is from Acer. The only entity you could sue is Acer. Same goes for HP. So do us all a favor and can the lawsuit threats; they accomplish nothing other than making you look like a petulant child having a tantrum - regardless, this is against the rules of conduct for this community.

Off my soapbox now.

...S

[CaptBridgeport]

Its just a common sense warning to make right what's supposed to be right, and not a shoddy product being sold as a complete product.  Iris Xe is very impressive if Intel can get the driver fixed.  

[n_scott_pearson]

That is most certainly an implied threat. I have already reported you to the administrator.

...S

[CaptBridgeport]

The more publicity this gets the better. I hope HP and Acer read this.  I already requested this issue be sent to the top.  Obviously you are biased and defensive, and obviously you do not comprehend that REAL financial harm is happening to those that placed TRUST in Intel.  I would think Intel would be apologetic and put 100 engineers on this issue and squash it.  I would think Intel should apologize for your rudeness and threats to me, tattle tailing on me to the administrator.  You should be banned from this community, I'm helping, donating my time and money, you are threatening and mocking me. 

[RonaldM_Intel]

Hi @CaptBridgeport

Thank you for the time you have taken reporting this issue to us.

I haven't seen any reports or issues with memory leaks on Iris Xe coming from any major application or from Windows 10, so I would like to have a better understanding of the impact this issue has before I can raise it with our devs with a proper priority classification.

Previously you listed steps to reproduce the issue involving the application "ProEssentials v9 Pro". Is there a way to reproduce the issue following a standard use case scenario? (e.g. something a user would do or encounter on a regular basis) You see where I want to go with this right? if the issue can be easily reproduced on scenarios that are likely to occur in common workloads then I can push it internally with higher priority, but if it hasn't show up yet on any application that is commercially available, or it is difficult to reproduce on regular day-to-day scenarios then it is going to be difficult to push it to the top of other critical issues our devs may be working on.

Once again, thank you for taking the time to engage with our Community.

Best Regards,

Ronald M.

[CaptBridgeport]

Google "Iris Xe Graphics Problem" and you will see company after company telling customers to disable Graphics Hardware Acceleration etc.  All since Dec 2020.   So your customers are likely expecting the driver to be improved,  and not spending time trying to report it.  

[CaptBridgeport]

Our product is a graphics rendering component.  Our customers buy our component to place into their products.  Our customer's  products are then sold to more customers.  And even our customer's customers may distribute to their/more customers/users.   So you see, the situation is critical.   So you see, if there is an issue, it looks bad for many customers however the blame is out of our hands. It's all Intel's blame because the issue is the driver though our customers and us are inherently to blame for something out of our control.  This all leads to  much defamation of character and many engineers trying to debug what ultimately can not be fixed. 

This issue is very common to our customers and so on.  Real-Time graphics and speed and quality of rendering are all our strengths.   And thus our customer's strengths.  

I wrapped up a demo within one zipped file.  All source code is identical between 32 and 64 bit.   There is Pedemo32 and PeDemo64.  Start them both side by side, monitor memory on both in Resource Monitor.  Within the 32 bit demo, use left list box to select demo115.  within the 64 bit demo, do the same, select demo 115.  So all you need to do is unzip onto a flash driver and test on a few system, one Iris Xe and one not Iris Xe. 

All I'm asking for is one minute for one Top Lead engineer on the Iris Xe Driver team to try this test.  And then all I want is a response that a) yes something is wrong with our 64 bit driver, or b) no, nothing is wrong with the 64 bit driver.   Again, run on anything other than Iris Xe you will not see a leak.  Only Iris Xe.    Please have an engineer look at this to give me some feedback. 

I believe the issue is related to a Direct3D render loop that calls  CreateDxgiSurfaceRenderTarget to add Direct2D content to the Surface.  Sort of like a Direct2D headsup display, but in this case the grid, titles, labels, bound box.  The data/graphics being real-time updated is Direct3D.  In our code, if we temp comment out the Direct2D content and CreateDxgiSurfaceRenderTarget, no leak 64 bit.  This is why I believe it's related to this call and feature. 

I wanted to add, if you see the Intel System Support utility output I provided above, you do see at the top ...  (Note in this case PEDemo was compiled as 64 bit.  )

Windows Error Reporting:
+++ WER0 +++:
Fault bucket 2190798233257984861, type 5
Event Name: RADAR_PRE_LEAK_64
Response: Not available
Cab Id: 0

Problem signature:
P1: PEDemo.exe
P2: 9.7.0.0
P3: 10.0.19042.2.0.0

along with all the other crash reports which all are unique to this Iris Xe driver.  

And one more note, I looked through our customer data base and Intel has purchased about 20 licenses of ProEssentials over the last 15 years.  So this problem may even show impacting some of Intel's own internal software.  

[DeividA_Intel]

Hello CaptBridgeport, 

Thank you for the information provided 

I will proceed to check the issue internally and post back soon with more details, however, bear in mind that I may request again some of the information that you thought was not necessary. 

Best regards,  

Deivid A.  

Intel Customer Support Technician 

[DeividA_Intel · Accepted solution]

Hello CaptBridgeport, 

I would like to let you know that we tried to replicate the issue with the latest driver "30.0.100.9684" and we did not face the same issue. We tried a couple of times just to corroborate results using different laptops and even a Tiger Lake reference board.

My recommendation is to try with our latest driver (30.0.100.9684) and check if the issue persists:

- https://downloadcenter.intel.com/download/30579/Gr-ficos-Intel-controladores-DCH-Windows

Best regards,  

Deivid A.  

Intel Customer Support Technician 

[CaptBridgeport]

Hello

Success, this latest driver dated 7/14/21 does resolve the 64 bit memory leak.  

Thank you.   Hopefully 64 bit shader storage is coming soon along with fixing  pd3dDevice->CheckMultisampleQualityLevels to report supported sample count correctly.  

Best regards...