• For EXA solid fill, only solid plane masks and only GXcopy raster-op operations are
accelerated.
• For EXA copy operation, the raster-op operations (GXandInverted, GXnor,
GXorReverse, GXorInverted, GXnand ) are not accelerated.
• EXA composite allows for many options and combinations of source/mask/target for
rendering. Commonly used EXA composite operations are accelerated.
The following types of EXA composite operations are accelerated:
• Composite operations for source/target drawables containing at least 640 pixels. If
less than 640 pixels, the composite path falls to software.
• Simple source composite operations are used when source/target drawables contain
more than 1024x1024 pixels (operations with mask not supported).
• Constant source (with or without alpha mask) composite with target.
• Repeating pattern source (with or without alpha mask) composite with target.
• Only these blending functions: SOURCE, OVER, IN, IN-REVERSE, OUT-
REVERSE, and ADD (some of these need to support the accelerated component-
alpha blending).
• In general, the following types of less commonly used EXA composite operations are
not accelerated:
• Transformed (meaning scaled, rotated) sources and masks.
• Gradient sources.
• Alpha masks with repeating patterns.
The implementation handles all pixmap allocation for X through the EXA callback
interface. A first attempt is made to allocate the memory where it can be accessed by a
physical GPU address. This attempt may fail if there is insufficient GPU accessible
memory remaining, but it can also fail when the bits per pixel, which are being requested
for the pixmap, are less than 8. If the attempt to allocate from the GPU accessible
memory fails, the memory is allocated from the system. If the pixmap memory is
allocated from the system, this pixmap cannot be involved in GPU accelerated option.
The number of pitch bytes used to access the pixmap memory may be different
depending on whether it was allocated from GPU accessible memory or from the system.
Once the memory for X pixmap has been allocated, no matter it is from GPU accessible
memory or from the system, the pixmap is locked and can never migrate to other type of
memory. Pixmap migration from GPU accessible memory to system memory is not
necessary since a system virtual address is always available for GPU accessible memory.
Pixmap migration from system memory to GPU accessible memory is not currently
implemented, but would only help in situations where there was insufficient GPU
accessible memory at initial allocation. More memory, however, becomes available
(through de-allocation) at a later time.
Software Operation
i.MX 6Solo/6DualLite Linux Reference Manual, Rev. L3.0.35_4.1.0, 09/2013
126
Freescale Semiconductor, Inc.
