
Philips Semiconductors
Image Co-Processor
File: icp.fm5, modified 7/26/99
PRELIMINARY INFORMATION
13-3
Scaling and filtering of the two dimensional image re-
quires two passes of the image data through the filter,
one for vertical and one for horizontal. Scaling an image
and sending it to the PCI bus requires three transfers of
the image over the SDRAM bus: one transfer to read the
image for vertical filtering, one transfer to write the fil-
tered data back, and one transfer to read the image for
horizontal filtering and output to the PCI bus. This means
an average of SDRAM bus bandwidth of 3 x 50 = 150
megabytes/second for the 1024 x 768 image case de-
scribed above, assuming a scaling factor of 1.0. A larger
or smaller scaling factor means that either the input or
output image will be smaller than 1024 x 768. The band-
widths required are determined by the larger of the two
images, input or output. This is because all input pixels
must be scanned to generate all the output pixels. Scal-
ing and filtering the image back to the SDRAM requires
an additional transfer to write the horizontally filtered im-
age back to SDRAM.
13.2.3
Image Size and Scaling
Image sizes in the TM1100 have a nominal range of 16
x 16 to 1024 x 768. Sizes smaller than 16 x 16 are pos-
sible, but are too small to be recognizable images. Imag-
es larger than 1024 x 768 (up to 64K x 64K) are possible
but cannot be processed in real time. They also require
larger SDRAM size to support them. Scaling factors have
a nominal range of 1/4 (down scaling by 4) to 4 (upscal-
ing by 4). Larger up and down scaling factors are possi-
ble, up to 1000 and beyond; however, very large upscal-
ing factors result in a large magnification of a few pixels,
and very large down scaling factors give only a few pixels
as a result.
13.3
INTERFACE
The Image Co-Processor block has no TM1100 external
pins. It interfaces internally to the SDRAM Data Highway
and the PCI output.
13.4
DATA FORMATS
The Image Co-Processor block accepts input and over-
lay image data to generate output image data. The ICP
accommodates a variety of formats for the input, overlay
and output data. These image data formats define the re-
lationship between the Y, U and V or the R, G, and B
components of the image as they are stored in memory.
The ICP accepts input image data in planar format,
where the Y, U and V components are in separate tables
in SDRAM. The various input image data formats differ
in the position of the U and V components relative to the
Y component and the amount of U and V data relative to
the Y data.
In all modes except the YUV to RGB conversion modes,
each ICP operation processes one Y, U or V image com-
ponent. Three separate commands are required to pro-
cess all three components of an image. Since each com-
ponent is scaled and filtered separately, the calling
software defines the image format and format conversion
by how it scales each component.
In the YUV to RGB conversion to PCI output or SDRAM
output mode, each output pixel is a combination of RGB
or YUV components as defined by the output format. The
YUV input data and the RGB or YUV overlay data are
combined by the ICP hardware pixel by pixel to form the
RGB or YUV output pixels. Because all three YUV com-
ponents are simultaneously woven together to create
each output pixel, the ICP hardware must know the im-
age data format in SDRAM, defined as how the compo-
nents of the image data are to be found and combined.
In the YUV to RGB conversion mode, the ICP accepts
the following input data formats: YUV 4:2:2 co-sited,
YUV 4:2:2 interspersed and YUV 4:2:0. In the YUV to
RGB conversion mode, the ICP also accepts image
overlay data when PCI output is specified. The ICP ac-
cepts image overlay data in several combined formats:
RGB-24+
α, RGB15+α and YUV 4:2:2+α. In this mode,
the ICP generates RGB or YUV output data in several
RGB and YUV formats. These formats are compatible
with a wide variety of PCI frame buffers.
13.4.1
Image Input Formats
The ICP image input formats define the relative positions
of the Y component and the U and V components of the
input image pixel data. There are three input formats to
the ICP: 4:2:2 co-sited, 4:2:2 interspersed, and 4:2:0 in-
terspersed. The 4:2:2 formats have 2 U and 2 V pixels for
every 4 Y pixels, so the ratio of Y to U or V is 2:1. The
4:2:0 format has 1 U and 1 V pixel for every 4 Y pixels,
so the ratio of Y to U or V is 4:1. The input formats are
given below. The input formats have a significant impact
on the 2 dimensional scaling operation.
13.4.1.1
YUV 4:2:2 Co-Sited
In the YUV 4:2:2 co-sited format, the U and V pixels co-
incide with the Y pixel on every other pixel, as shown in
13.4.1.2
YUV 4:2:2 Interspersed
In the YUV 4:2:2 interspersed format, the U and V pixels
lie between the Y pixels on every other pixel of the hori-
13.4.1.3
YUV 4:2:0 XY Interspersed
In the YUV 4:2:0 interspersed format, the U and V pixels
lie between the Y pixels on every other pixel of the hori-
13.4.1.4
YUV 4:1:1 Co-Sited
In the YUV 4:1:1 co-sited format, the U and V pixels co-
incide with the Y pixel on every fourth pixel, as shown in