DriverDetails: Set fixed versions for old Adreno bugs - #14819
Conversation
For use with DriverDetails.
I tested if the following bugs were still present on different Android devices running Adreno drivers 494.0, 676.0, and 800.70, and all of them seem to have been fixed at some point: * BUG_BROKEN_BUFFER_STREAM * BUG_BROKEN_NEGATED_BOOLEAN * BUG_BROKEN_D32F_CLEAR * BUG_BROKEN_REVERSED_DEPTH_RANGE * BUG_SLOW_CACHED_READBACK_MEMORY I set the fixed version to the oldest driver I had that didn't have the bug, so the fixed version is an upper bound rather than the exact driver that fixed the problem.
I found this useful to figure out what exact version the driver is reporting, which I needed to know in order to edit m_known_bugs.
|
I suggest adding a todo: note stating that the exact version that fixed the bug may be older. |
| BUG_BROKEN_CLEAR_LOADOP_RENDERPASS, -1.0, -1.0, true}, | ||
| {API_VULKAN, OS_ALL, VENDOR_QUALCOMM, DRIVER_QUALCOMM, Family::UNKNOWN, BUG_BROKEN_D32F_CLEAR, | ||
| -1.0, -1.0, true}, | ||
| -1.0, 0x801EE000, true}, |
There was a problem hiding this comment.
In the text it says 494, should it match here?
There was a problem hiding this comment.
Also, what does 0x801EE000 represent?
There was a problem hiding this comment.
Vulkan driverVersions are formatted in a specific way. If you look at the definition of VK_API_VERSION_MINOR as a reference, the 0x1EE part of this hex number represents a minor version of 494. (Qualcomm has the major version set to 0.)
Do you mean in DriverDetails.cpp or in DriverDetails.h? Because I already noted this down in DriverDetails.h for the entries that had a version line. |
|
FifoCI detected that this change impacts graphical rendering. Here are the behavior differences detected by the system: Detected differences
|
I tested if the following bugs were still present on different Android devices running Adreno drivers 494.0, 676.0, and 800.70, and all of them seem to have been fixed at some point:
I set the fixed version to the oldest driver I had that didn't have the bug, so the fixed version is an upper bound rather than the exact driver that fixed the problem.