Bug 97619

Created attachment 126256 [details]
Output of sudo lspci -vv for one machine with the described problem.

Because this one is hard to describe, it has taken a long time to find anyone else with the problem, but here it is:
http://askubuntu.com/questions/783358/visual-glitches-when-rotating-display-ati-radeon

The problem appeared after upgrading to XUbuntu 16.04 on 2 32-bit Pentium 4 machines. Any display that is rotated becomes unintelligible because it is not delivered to the display properly, and I cannot send a screen shot because the screen shot records what should be on the screen, not what is displayed. On this machine I have 2 monitors, one 1920x1080 horizontal, and one to the right, 1280x1024 rotated 90 degrees clockwise with the picture must be rotated left. The horizontal monitor displays normally, the picture on the rotated one is divided into 128 vertical bands, 8 pixels wide, and the vertical bands appear to be divided into 40 segments 32 pixels high. The vertical bands do not always contain information that belongs with the bands adjacent to them, but possibly with the bands one band away on each side (reminds me of an interlace pattern). The vertical blocks containing information that might belong to blocks above and below them are also alternated with blocks containing other information, and appear to be stretched vertically (2x?). Knowing what the background picture is, I recognized 8x32 pixel pieces, but the picture is so fractured as to be unrecognizable. If I open a window on the rotated display, it is so fractured and stretched that I cannot tell where the top of the window is, and if I slide an open window from the left desktop to the rotated one so the "x" (in the close window button) appears on the rotated screen, the "x" first appears on the second 8 pixel vertical band from the left. If I slowly move the window down, the "X" moves down twice as fast and skips 32 pixel vertical blocks on the way down. When it leaves the bottom of the screen, it appears at the top in the left most 8 pixel band and moves down without skipping blocks (but still at twice the speed the windows is moving down). As soon as any part of the window appears on the rotated screen, moving the window is no longer smooth. The mouse does NOT behave strangely on the rotated screen, and if a window opens on the rotated screen, and I move the mouse slowly down the screen until it changes shape (as it crosses the top edge) and then changes back, I can click and drag the window to the left display where I can work with it. The mouse arrow does NOT move down the rotated display twice as fast as it would on the left display, so when it changes shape as I cross onto a window, any recognizable elements of that window are twice as far down the display as the mouse arrow. As I said, a screen shot shows the rotated monitor as it would be if the driver were working properly.

This is not a hardware problem; the other machine only has one display, but when I use xrandr to rotate it, the display fractures as described above (for all --rotate options but "normal"). Inverting the display does not change the orientation of the 8x32 bit blocks; they still look like interlacing with the 8-bit wide bands parallel to the normal top/bottom of the display. Both machines work fine with Windows XP and worked fine until 16.04 was installed. A 64 bit installation with an NVIDIA video card does not have the problem.

I recognize that this may be a bug for the people at ATI/Radeon, but have no idea where to submit it; and I'm guessing bugs.freedesktop.org knows. I'm not even sure there's any point in providing the results of sudo lspci -vv, because since 'Print Screen' works, this has all the signs of being a problem with how the virtual desktop in memory is being sent to the display. However, for what it's worth, I've attached a copy.