Integrated Functions (Continued)
4.5.1 Display FIFO
The compression logic has the ability to insert a program-
mable number of "static" frames, during which time dirty
bits are ignored and the valid bits are read to determine
whether a line should be retrieved from the frame buffer or
compressed display buffer. The less frequently the dirty
bits are sampled, the more frequently lines will be
retrieved from the compressed display buffer. This allows
a programmable screen image update rate (as opposed to
refresh rate). Generally, an update rate of 30 frames per
second is adequate for displaying most types of data,
including real-time video. If a flat panel display is used that
has a slow response time, such as 100 ms, the image
need not be updated faster than ten frames per second,
since the panel could not display changes beyond that
rate.
The display controller contains a large (64x64 bit) FIFO for
queuing up display data from the memory controller as
required for output to the screen. The memory controller
must arbitrate between display controller requests and
other requests for memory access from the microproces-
sor core, L1 cache controller, and the graphics pipeline.
Display data is required in real time, making it the highest
priority in the system. Without efficient memory manage-
ment, system performance would suffer dramatically due
to the constant display-refresh requests from the display
controller. The large size of the display FIFO is desirable
so that the FIFO may primarily be loaded during times
when there is no other request pending to the DRAM con-
troller which allows the memory controller to stay in page
mode for a longer period of time when servicing the dis-
play FIFO. When a priority request from the cache or
graphics pipeline occurs, if the display FIFO has enough
data queued up, the DRAM controller can immediately
service the request without concern that the display FIFO
will underflow. If the display FIFO is below a programma-
ble threshold, a high-priority request will be sent to the
DRAM controller, which will take precedence over any
other requests that are pending.
The compression algorithm used in the GXLV processor
commonly achieves compression ratios between 10:1 and
20:1, depending on the nature of the display data. This
high level of compression provides higher system perfor-
mance by reducing typical latency for normal system
memory access, higher graphics performance by increas-
ing available drawing bandwidth to the DRAM array, and
much lower power consumption by significantly reducing
the number of off-chip DRAM accesses required for
refreshing the display. These advantages become even
more pronounced as display resolution, color depth, and
refresh rate are increased and as the size of the installed
DRAM increases.
The display FIFO is 64 bits wide to accommodate high-
speed burst read operations from the DRAM controller at
maximum memory bandwidth. In addition to the normal
pixel data stream, the display FIFO also queues up cursor
patterns.
As uncompressed lines are fed to the display, they will be
compressed and stored in an on-chip compressed line
buffer (64x32 bits). Lines will not be written back to the
compressed display buffer in the DRAM unless a valid
compression has resulted, so there is no penalty for
pathological frame buffer images where the compression
algorithm breaks down.
4.5.2 Compression Technology
To reduce the system memory contention caused by the
display refresh, the display controller contains compres-
sion and decompression logic for compressing the frame
buffer image in real time as it is sent to the display. It com-
bines this compressed display buffer into the extra off-
screen memory within the graphics memory aperture.
Coherency of the compressed display buffer is maintained
by use of dirty and valid bits for each line. The dirty and
valid RAM is contained on-chip for maximum efficiency.
Whenever a line has been validly compressed, it will be
retrieved from the compressed display buffer for all future
accesses until the line becomes dirty again. Dirty lines will
be retrieved from the normal uncompressed frame buffer.
Revision 1.1
135
www.national.com
NSC [ National Semiconductor ]
