|
(11) | EP 0 192 139 A2 |
| (12) | EUROPEAN PATENT APPLICATION |
|
|
|
|
|||||||||||||||||||||||||||
| (54) | Frame buffer memory controller |
| (57) A frame buffer memory controller (24) allows rapid image updating while maintaining
screen refresh data flow rate. One frame buffer memory controller (24) controls one
or more pixel depth columns comprising one or more frame buffer memory chips (22)
per pixel. Each frame buffer memory controller (24) listens on a display processor
bus (26) for read, write or read-modify-write commands addressed to a pixel, or memory
chip, under its control. Such commands, along with the associated addresses and data,
are stored in a first-in, first-out (FIFO) buffer (35) for execution during the first
free memory cycle. |
Background of the Invention
Field of the Invention.
[0001] The present invention relates to frame buffer memory systems for raster displays, and more particularly to a frame buffer memory controller for allowing rapid picture updating while maintaining screen refresh data flow rate.
Description of the Prior Art.
[0002] Raster scan, frame buffer displays have become increasingly popular as the price of semiconductor memory has decreased. The image to be displayed is represented in a large memory that saves a digital representation of the intensity and/or color of each picture element, or pixel, on the screen. By properly recording the data in the memory an arbitrary image can be displayed, making the display hardware insensitive to image content. The frame buffer memory is equipped with hardware to generate a video signal to refresh the display and with a memory port to allow a host computer or display processor to change the frame buffer memory in order to change the image being displayed.
[0003] Interactive graphics applications require rapid changes to the displayed image, which in turn require rapid changes to the frame buffer memory. Although the speed of the host processor and display processor is clearly important to high performance, so also are the properties of the memory system, such as update bandwidth, i.e., the rate at which the host processor or data processor may access each pixel. For a given memory technology the implicit geometry of frame buffer memory access can affect this rate. Conventional pixel memory systems arrange words of memory so that a single memory cycle provides access to sixteen, twenty, thirty-two, or some other fixed number of pixels in a horizontal scan line on a display. Other systems use arrays of pixels, such as 4x4, 4x5, 8x8, etc. for each frame buffer word.
[0004] A conventional frame buffer memory writes pixels along a horizontal line very rapidly, but is slow for most other directions. For a frame buffer memory with a 16 pixel wide by 1 pixel high word, let the average time to do a memory write be T seconds (including the delay caused by interspersed display refresh reads). Then horizontal lines can be written at a rate as high as 16/T pixels per second. Since the beginning and end of the lines will generally not lie on word boundaries, the actual rate will be less than 16/T on average. Now consider a vertical line, or any line steeper than 45 degrees. Every pixel written will lie in a different word. The pixel rate is thus 1/T pixels per second. Averaging the pixel drawing rate over all vector angles and ignoring the end effects gives roughly 1.36/T pixels per second. Frame buffer memories with words covering a rectangular array of pixels improve on the average pixel writing rate. For a frame buffer memory with a 4 pixel wide by 4 pixel high word and average writing time T, except for the beginning and end of lines, 4 pixels can be written on each memory cycle independent of the orientation of the line. The pixel writing rate thus approaches 4/T )pixels per second.
[0005] What is desired is a means for speeding up the process of updating the image in frame buffer memory, i.e., increasing update bandwidth.
Summary of the Invention
[0006] Accordingly, the present invention provides a Frame buffer memory controller which allows rapid Lmage updating while maintaining screen refresh data flow rate. One frame buffer memory controller controls one or more pixel depth columns comprising one or more frame buffer memory chips per pixel. Each frame buffer memory controller listens on a display processor bus for read, write or read-modify-write commands addressed to a pixel, or memory chip, under its control. Such commands, along with the associated addresses and data, are stored in a first-in, first-out (FIFO) buffer for execution during the first free memory cycle. The result is, for example, a line drawing throughput approaching n times higher where n is the number of pixels per memory word.
[0007] Objects, advantages and novel features of the present invention will be apparent from the following detailed description when read in conjunction with the appended claims and attached drawing.
Brief Description of the Drawing
[0008]
Fig. 1 is a block diagram view of a frame buffer memory according to the present invention.
Figs. 2A through 2D are examples of various geometries for frame buffer memory access.
Fig. 3 is a block diagram view of a frame buffer memory controller for the frame buffer memory of Fig. 1.
Fig. 4 is an illustration of the operation of the present invention.
Description of the Preferred Embodiment
[0009] Referring now to Fig. 1 a frame buffer memory 20 is shown generally. The frame buffer memory 20 has a plurality of memory devices 22, typically random access memories (RAM), each RAM corresponding to one plane of one pixel bit in a display or frame buffer, word as shown in Figs. 2A through 2D. For a conventional memory geometry as shown in Fig. 2A the number of RAMs 22 is lx16xn where n is the number of planes (number of bits per pixel). A 1024x1024 raster display would require such RAMs to have a 64K capacity.
[0010] For each pixel or small group of pixels within the display word a frame buffer memory controller 24 serves as an interface with a display processor bus 26. Each frame buffer memory controller 24 recognizes addresses from the display processor bus 26 which pertain to the RAMs 22 under its control. The associated commands/data are then routed to the appropriate RAM 22 and location within that RAM corresponding to the particular display word via address bus 28 and/or data bus 30 from the frame buffer memory controller 24. Data to be displayed is transferred via the frame buffer memory controller 24 to a digital-to-analog converter 32 for conversion to video data. The example shown for Fig. 1 is illustrative of the configuration for a 16-pixel display word of 4-bits, or planes, per pixel. Inserted between the FBMCs 24 and DAC 32 may be a color map 33 for determining the color of each pixel.
[0011] Fig. 1 shows the pixel data for display refresh coming from the FBMCs 24. However, video RAMs 22, such as the TMS 4161 by Texas Instruments Inc., Dallas, Texas, that have integral video shift registers, could be used equally well. In either case the color map 33 or other logic may be inserted in the pixel data path before the DAC 32. As shown the pixel data outputs of the FBMCs 24 are bussed together onto one set of wires (one wire per plane). Under this scheme each FBMC 24 drives the pixel data bus for one pixel clock, then tristates its outputs to allow the next FBMC to send its pixel data. This is practical for low resolution displays with slow pixel clocks. For higher resolution displays the pixel data outputs from each FBMC are not bussed, but are connected to video shift registers as in conventional frame buffer memories with one shift register per plane. Each shift register receives one bit, or as many bits as there are pixels in the display word controlled by the FBMC 24, from each FBMC. Other variations are possible, such as incorporating the shift registers within the FBMCs 24 so that the pixel data output from each FBMC shifts into the next one, with the final output connecting to the color map 33 or DAC 32.
[0012] The frame buffer memory controller 24 is shown in some detail in Fig. 3. An address recognizer circuit 34 is connected to the display controller bus 26 to determine those operations directed to the pixels controlled by that particular frame buffer memory controller. The recognized address is stored in a portion 36 of a FIFO buffer 35 when an associated command is received. During a screen refresh cycle data is read from the RAMs 22, transferred via a latch 38 to a second FIFO buffer 40, and thence to an output buffer 42 to provide the appropriate pixel data. Display refresh timing is provided by a video timing circuit 44 and the pixels are addressed sequentially via a refresh address counter 46. An address multiplexer 48 passes the addresses from the refresh address counter 46 to the RAMs 22 during the refresh cycle, and passes the addresses from the first FIFO buffer 36 to the RAMs 22 for image changes during free memory cycles. The address recognizer circuit 34 may be replaced by an external address decoder 37 on the display processor bus 26 which routes the command/data/addresses to the appropriate FBMCs 24.
[0013] When a command/data word on the display bus 26 is recognized as applying to the RAMs 22 associated with the particular frame buffer memory controller 24 by the address recognizer circuit 34, the command/data information is stored in respective portions 50,52 of the FIFO buffer 35 along with the address information. At the first free memory cycle the appropriate RAM 22 is accessed according to the first-in address in the address buffer 36, and the associated command in the command buffer 50 is executed, data in the write buffer 52 being written into the RAM, or data being read from the RAM and either stored in the read buffer 54 for transfer to the display processor via the display bus 26 or for modification and rewrite into the RAM via a read-modify-write (RMW) logic circuit 56.
[0014] In operation the frame buffer memory 20 can constantly receive commands/data via the display bus 26 from the display processor even while pixel data is being read from the memory units 22 to refresh the display screen.
[0015] The command/data information together with RAM address is stored in the FIFO buffer 35 and acted upon at the first free memory cycle. Thus, the image on the display is rapidly updated since the information for each pixel can be transmitted to the frame buffer memory via the frame buffer memory controllers while the display screen is being refreshed. Additionally, performance is enhanced by the fact that all n FMBCs 24 can be writing pixels simultaneously even when the pixels being written are not on a single horizontal line.
[0016] Given the same memory technology and display refresh overhead (same average write time of T seconds) and the same word size (16 pixels) a frame buffer memory 20 as shown in Fig. 1 can write pixels at a rate approaching 16/T per second. This speed improvement is obtained by dividing the frame buffer memory 20 into 16 separate pieces, one for each pixel in the frame buffer word. Each piece of the memory 20 has separate address and data lines controlled by a separate frame buffer memory controller 24. Thus memory accesses are no longer limited to fixed frame buffer words. At any moment each of the 16 FBMCs 24 may be writing a pixel to a different frame buffer word.
[0017] To achieve this performance improvement, each frame buffer memory controller 24 has a FIFO buffer 35 for commands, address and data coming from the display processor in addition to the separate address and data lines to the RAMS 22 controlled by it. This FIFO buffer 35 serves two purposes. First, it allows continued receiving of commands from the display processor while the frame buffer memory 20 is busy with display refresh reads. This advantage becomes negligible if video RAMs are used. The second and more important purpose is to allow multiple display processor commands to address the same frame buffer memory controller 24 (same one of the 16 pieces of frame buffer memory 20) without waiting for each memory cycle to finish before sending the next command. On average the display processor commands will address each piece of the memory (each frame buffer memory controller 24) at an equal rate. Over short periods of time, however, one or a few FBMCs 24 may receive most of the commands. The FIFO buffer 35 smoothes out this short term unevenness allowing all pieces of memory to keep busy most of the time.
[0018] Consider an example as shown in Fig. 4 with a 16 pixel wide by 1 pixel high frame buffer word. Further consider the operation of drawing a line into this frame buffer memory with slope of 3 (for every one pixel to the right the line goes 3 pixels up). Let the line run from 100,200 to 150,50. The first two pixel write commands address frame buffer memory controller (FBMC) number 5, the next three pixel writes address FBMC number 6, and so on. If the pixel write commands are sent at the rate of 16/T per second, each FBMC just finishes its last write when it is again addressed with three more pixel write commands.
[0019] The above example requires the FBMCs 24 to have a FIFO buffer 35 of length 3 or larger to run at the full 16/T rate. For stepper lines (closer to vertical) the required FIFO buffer size is larger. Assume some reasonable FIFO buffer size of 32 words per FBMC 24. Then long lines with slope much over 32 fill the FIFO buffer 35 of one FBMC 24, requiring the display processor to wait until some of the write commands have been executed before continuing. Short vertical lines do not cause this problem unless several lines in succession address the same FBMC 24. Thus a few cases, such as long vertical lines, run slower than 16/T pixels per second. For typical pictures this fraction is very small, and can be made to approach zero by increasing the FIFO buffer size. For random short vectors this example system yields roughly 13/T to 14/T pixels per second with its 32 word FIFO buffer 35.
[0020] All performance estimations are proportional to 1/T. If video RAM's are used where display refresh overhead is minimal, then T is simply the memory cycle time. If writes only are being done, then T is the write cycle time. If read-modify-writes are being done, then T is the read-modify-write cycle time. If normal RAMs 22 are used in the frame buffer memory 20 then performance is decreased by the percentage of time used for the display refresh reads. If refresh reads take 50% of the RAMs' time, then T is increased by a factor of two and performance decreases by a factor of two. In either case, however, the relative increase in performance obtained by using FBMCs 24 is the same.
[0021] Other display word configurations work equally well, such as a 4 by 4 pixel array or any other size as shown in Figs. 2b and 2c. In fact, the 4 by 4 pixel array has a slight advantage in that vertical lines would address 4 different FBMCs 24. It therefore requires four times as long a vertical line to fill the FIFO buffer 35 and cause the display process to wait (for a given FIFO buffer size).
[0022] Also the conventional organization may be mixed with the FBMC concept to give a combination implementation. For example, a 20 pixel frame buffer word may be organized as 10 pixels wide by 2 pixels high with 5 FBMCs 24 each controlling a 2 by 2 pixel square within the word. This reduces the number of FBMCs 24 required from 20 to 5. For drawing vectors, each FBMC 24 can write 2 pixels at a time from its 2 by 2 array. Since there are 5 FBMCs 24 writing independently, the writing performance approaches 5*2/T = 10/T pixels per second. If all 20 FBMCs 24 were used the rate would have approached 20/T. If 10/T pixels per second is sufficient to keep up with the display processor, however, cost can be saved by using only 5 FBMCs 24 instead of 20.
1. A frame buffer memory characterized by
a plurality of memory units (22), each of said memory units corresponding to one bit of a pixel within a frame buffer word;
a plurality of frame buffer memory controllers (24), each of said frame buffer memory controllers providing access to a subset of said memory units on a first-in, first-out basis; and means for addressing each of said frame buffer memory controllers (24) from a display processor via a display bus (26).
2. A frame buffer memory as recited in claim 1, characterized in that said frame buffer
memory controllers (24) each comprise: means (34) for recognizing information on said
display bus (26) directed to said subset of memory units controlled by said frame
buffer memory controller (24); means (35) for preserving said information for execution
on a first-in, first-out as available basis; and means (38,40,42) for transferring
data to said memory units (22) from said preserving means (35).
3. A frame buffer memory as recited in claim 1 or 2, characterized in that said subset
comprises n of said memory units (22) where n is the number of bits for each said
pixel, each bit corresponding to a plane.
4. A frame buffer memory as recited in one of claims 1 to 3, characterized in that
said subset comprises m x n of said memory units (22) where m is the number of said
pixels controlled by said frame buffer memory controller (24) and n is the number
of bits for each said pixel, each bit corresponding to a plane.
5. A frame buffer memory controller, characterized by:
means (34) for recognizing information on a display bus (26), said information being addressed to a subset of a plurality of memory units (22);
means (35) for temporarily storing said information on a first-in, first-out basis; and
means (38,40,42) for transferring said information from said storing means to said subset.
6. A frame buffer memory controller as recited in claim 5, characterized by means
for addressing said subset to transfer said information to a video display on a non-interference
basis with said transferring means.