Buffer queue management for multiple processes.

<u style="single">DESCRIPTION</u>

0001This invention relates generally to data processing and, more specifically, to buffer queue management in such systems.

0002In data processing systems, the management of buffer queues is required in many areas. For instance, commands relating to an independently executing process will be temporarily stored in buffer queues. Therefore, the supplier of the commands need not provide the commands synchronised with the execution of commands by the independent processing device. However, it must be ensured that the buffer storing these commands temporarily does not overflow or, in a system optimally providing commands, that the buffer queue does not become empty. Traditionally, two techniques have been used to manage buffers. One technique requires that the sender of commands to the buffer query the status of the buffer. In a data processing system where the buffer is tightly coupled to the remainder of the system, this may not be a problem. However, if the buffer is located on a communications bus, the querying of status includes a bus overhead. A second technique is to fill the buffer and wait for the buffer to empty before it is filled again.

0003One example of an embodiment using the full buffer technique is contained in a paper entitled "Virtual Graphics" by Voorhies, Kirk and Lathrop, published in <u style="single">Computer Graphics</u>, Vol. 22, No. 4, August, 1988. This reference illustrates a first-in/first-out (FIFO) buffer that provides interrupts when the buffer is nearly full by providing an interrupt each time a command is presented to the buffer. This forces the process providing these commands to wait until the buffer is only half full. While this process does regulate the number of commands placed into the buffer, it also generates an excessive number of interrupts.

0004It is the object of the present invention to provide a buffer manager that minimises the number of interrupts required while optimally providing entries to the buffer.

0005According to the invention we provide a data processing system comprising a plurality of processors each for executing a process, at least one of which processors includes a buffer, and means for transferring information between processors characterised by a buffer manager in at least one of said processors for regulating process access and controlling words input by an executing one of the processes to said buffer, said buffer manager including a regulator for regulating a number of words from an executing first process for loading into the buffer, means for loading the number of words, means for loading a synchronisation word when a change in access to a second process is to occur, and means for halting the loading of words from the first process and providing for the loading of words from the second process when said synchronisation word is accessed by the independently executing processor.

0006We further provide in a data processing system having a plurality of processors, and a buffer for storing words such as commands requiring temporary storage during execution of a process, a method for buffer management for controlling process access and controlling words input by an executing one of the processors to a buffer of an independently executing processor, characterised by the steps of loading words into said buffer as they are produced by the executing processor, loading a synchronisation word when a change in process access is to occur, and halting the loading of the words from the current executing processor and providing for the loading of words produced by the new process when said synchronisation word is accessed by the subsequently executing processor.

0007In order that the invention may be well understood a preferred embodiment thereof will now be described with reference to the accompanying drawings, in which: <ul id="ul0001" list-style="none"><li>Figure 1 is a block diagram of a computer system with a graphics display;</li><li>Figure 2 is a symbolic diagram of a first-in/first-out buffer;</li><li>Figure 3 is a block diagram of the system processor illustrating the software modules contained therein;</li><li>Figure 4 is a flow chart of the initialisation routine for an applications process;</li><li>Figure 5 is a flow chart for the applications process graphics module;</li><li>Figure 6 is a flow chart of an initialisation routine for the rendering context manager software module;</li><li>Figure 7 is a flow chart illustrating the graphics adapter monitoring routine of the rendering context manager process;</li><li>Figure 8 is a flow chart illustrating the context switch handler of the rendering context manager process module;</li><li>Figure 9 is a flow chart illustrating a second embodiment of the application process graphics module; and</li><li>Figure 10 is a flow chart illustrating a third embodiment of the application process graphics module.</li></ul>

Best Mode for Carrying Out the Invention

0008This invention provides the ability for multiple processes executing in a system processor to access a single display adapter to display information on a graphics display. The invention makes it possible for these multiple application modules to coexist in a manner where each module operates independently of the other applications modules even though they are all sharing the single graphics adapter.

0009Figure 1 is a block diagram of the system illustrating the system processor 10 connected to an input/output bus 12. The input/output bus 12 is further connected to a display adapter 14. The display adapter is connected to a graphics display 22. The display adapter 14 further includes a command list first-in/first-out buffer 16, graphics logic 18 and a display buffer 20. Commands contained in the buffer queue 16 are executed in the graphics logic 18 and provide display data to the buffer 20 for display on the graphics display 22. The operation of the graphics adapter 14 is known in the art. The invention relates to the management of the command queue 16 by multiple processes in the system processor 10. Therefore, this invention can be used with any display adapter that has a single first-in/first-out command queue such as queue 16 in Figure 1. One such display adapter is the IBM High Performance 24-Bit 3D Graphics Processor.

0010A symbolic diagram of the first-in/first-out command queue buffer 16 is illustrated in Figure 2. This queue 16 includes multiple locations for storing commands. The command threshold size is illustrated by arrow 38. In the preferred embodiment, midway in the queue 16, a location is designated as the reload threshold 34. At this point, in the preferred embodiment, the queue 16 would be half full if all locations above the threshold 34 are empty. The number of these locations is the free increment size represented by the arrow 32. When the command queue 16 is completely filled, all of the locations from the bottom up to the full threshold 30 are occupied. As the commands are accessed on a first-in/first-out basis, the locations starting at the bottom location are emptied. When the reload threshold location 34 has been emptied, an interrupt is raised in the preferred embodiment. Furthermore, in the preferred embodiment, when the last command at the empty threshold location 36 has been executed, an empty threshold interrupt is provided. These two interrupts, the reload threshold interrupt and the empty threshold interrupt, are the only hardware signals required by the graphics adapter 14 (Figure 1) by this invention.

0011Figure 3 is a block diagram of the system processor 10 illustrating the software modules contained for execution. These software modules consist of applications software processes A 50, B 52 and C 53. These are software application routines that are being run internally in the system processor that, according to this embodiment, require access to the graphics display to provide output information to the operator. In this invention, each of the processes 50, 52 and 54 are treated as independently executing processes. In other words, each process 50, 52 and 54 is executed independently of the existence of the other processes. Each process thinks that it is the only process that is accessing the graphics adapter. Therefore, these processes, 50, 52 and 54, do not address the problems associated with the management of the graphics adapter by multiple processes. The process dispatcher 56 is provided to manage the execution of the application processes 50, 52 and 54 in the system processor 10. Process dispatcher 56 determines which of the application processes 50, 52 and 54 are to be executed at any time within the system processor 10. The rendering context manager 58 manages the access of each of the application processes 50, 52 and 54 to the graphics adapter 14. Operation of the rendering context manager 58 provides for the independent operations of the application processes 50, 52 and 54.

0012In Figure 4, a flow chart illustrating the initialisation routine for the application processes (such as processes 50, 52 and 54 in Figure 3) is illustrated. In step 100, a counter, used by the application when writing to the display adapter 14, is set to the full threshold size of the first-in/first-out command queue 16 (Figure 2). Lastly, in step 104, the address of the counter C is provided to the rendering context manager module 58 (Figure 3).

0013Figure 5 is a flow chart illustrating the graphics module that is contained in the application processes 50, 52 and 54 (Figure 3). In step 110, the process first computes the size N of a task for providing a drawing or attribute order. a drawing order is a command which causes an object to become visible on the graphics display 22 (Figure 1). An attribute order sets up conditions under which the drawing orders are executed; for example, an order to change the colour of all lines to be drawn is an attribute order. The distinction between the two categories is not important for the purposes of this invention; rather, the importance is that all commands are processed via the first-in/first-out command queue.

0014In step 120, the size of the order is compared to the the free increment size (I). The free increment size is equivalent to the number of locations illustrated by arrow 32 in Figure 2. This is the number of locations between the full threshold 30 and the reload threshold 34 locations in the queue 16. The independent operation of multiple processes 50, 52, 54 under the method of the invention depends upon the replacement of C by the value I in step 130 of Figure 5. Therefore, no order can have a size which exceed the value I. It is not anticipated that users of the graphics subsystem will generate the error exit of step 112.

0015In step 122, the value N is then compared to the stored value of C, which is the counter used by the application in Figures 4 and 5. If the value N is less than the value of C, the implication is that insufficient capacity in the buffer exists to add N more words. Thus, the process proceeds to step 128 to check the first-in/first-out buffer 16 status of reload threshold. This step begins the synchronisation of the software counter C with the hardware queue status. In step 128, if the queue depth is below the reload threshold, then process proceeds to step 130 to replace C with the value I. a replacement is shown in the flowchart, because there are at least I locations free in the buffer, and because a replace is non-interruptible in the best mode of the invention. However, if the system processor permits other non-interruptible operations with very low overhead, the addition of C+I could be substituted for the value of I. The process then continues to step 124 to provide the commands to the queue 16.

0016Returning to step 128, if the process has determined that the queue 16 is not below the reload threshold, the process proceeds to step 132 which accesses the rendering context manager module 58, which effectively places the applications module in a wait or sleep state. Actually, the process dispatch module 56 stops the application's process and proceeds to another application's process while the rendering context manager 58 waits for the queue 16 to drop below the reload threshold. The application does not issue a request to sleep; instead, it calls a routine within the kernel process, the rendering context manager 58, which manages the sleep process. In step 134, when the hardware condition has been met, i.e., the threshold has been crossed, an interrupt is received by the rendering context manager 58. The rendering context manager 58, knowing the state of the display adapter, overwrites the value of C with the known size of available locations. The rendering context manager 58 then provides an indication to the process dispatcher 56 to activate the application process, which then proceeds to step 124 to provide the write commands to the queue 16.

0017Figure 6 is a flow chart illustrating the single step, step 150, in the initialisation routine for the rendering context manager 58. In step 150, the address of counter C is stored together with the software process I.D. into a table. This table includes the critical numbers F and I (queue size and reload threshold increment size).

0018In Figure 7, the rendering context manager module that monitors the threshold interrupt is illustrated. In step 159, the rendering context manager looks up the address of counter C for the process which has called the rendering context manager. In step 160, the rendering context manager checks the status of the queue 16 reload threshold. In this step, a determination is made as to whether the queue 16 is below the reload threshold. This is done in advance of waiting on the interrupt, as an optimisation of performance. If so, the process proceeds to step 165. If not, the process proceeds to step 162 to put the current process in a wait or sleep state by calling the process dispatcher 56. When the threshold interrupt is received, the process dispatcher 56 reactives the application process which had made the kernek subroutine call to the rendering context manager 58. Program execution resumes in the rendering context manager 58 at step 164, which then continues. Proceeding to step 165, the rendering context manager then replaces the content of C from its graphics adapter table with the parameter I. a replacement is shown in the flowchart, because there are at least I locations free in the buffer, and because a replace is non-interruptible in the best mode of the invention. However, if the system processor permits other non-interruptible operations with very low overhead, the addition of C+I could be substituted for the value of I.

0019In step 166, a test is made as to whether the queue 16 has reached the empty status. This test is another optimisation to performance. In a preemptible system, such as is used on the best embodiment, many other activities could have taken place in the system between steps 162 and 164. Therefore, a significant increase in the value C is possible (from I to F) at the cost of one additional test of the hardware. If the queue is empty the full value F of the queue is used to replace C in step 168. The process then exits this routine. Control is returned to the application process, which was at step 132 of Figure 5 and which resumes at step 134 in Figure 5.

0020In Figure 8, the rendering context manager module that provides for context switching is illustrated in a flow chart. Context switching is the ability of one process to access the display adapter after another process has accessed the display adapter. Since only one process is active in the system processor 10 it would seem that only one active applications module could access the display adapter 14. However, since the display adapter 14 includes the command queue 16, it is possible for an applications process such as applications process 50 to provide many commands to the command queue 16 and then be deactivated (or put to sleep) while a second applications process 52 is activated. The rendering context manager software provides for the second process 52 to also access the display adapter 14 in a manner that is independent of the previous access by the applications process 50. In other words, the new applications process 52 is not concerned with what commands exist on the command queue 16 in the display adapter 14. In Figure 8, in step 170, the rendering context manager prepares the new process context for loading. This means that the elements of that applications processes context (access to the graphics adapter) are provided to the graphics adapter. Examples of context include the set of colours used to draw objects, the set of matrices used to map coordinates, or the definition of the active region in use in a windowing system.

0021In step 172, the process determines if the queue 16 is below the reload threshold. If so, the process proceeds to step 178. If not, the process proceeds to step 174 where the rendering context manager 58 is put in a wait state until the threshold interrupt is received. The rendering context manager must wait until the queue 16 drops below the reload threshold in order to place the synchronisation token onto the command queue. In step 176 when the threshold interrupt is received, the rendering context manager is reactivated. The rendering context manager process then proceeds to step 178 to place a synchronisation token on the queue 16. The synchronisation token indicates that a new command list is to be loaded for a new applications process. In step 180, the rendering context manager goes into a wait state waiting for a posting to indicate required service from the display adapter 14. Then upon the occurrence of some posting, such as the empty interrupt, a timer interrupt (indicating the error condition) or the desired posting that the synchronisation token has been read from the queue 16 and processed by the graphics logic 18, the rendering context manager is reactivated in step 182. In step 184, the old context (the context data for the applications process that is being removed from the display adapter 14) is stored and the new context is provided to the display adapter 14. In step 186, the contents of counter C of the new process, associated with the newly loaded context, is replaced with the value F, the full threshold size. The module is then exited.

0022Figure 9 is a flow chart of a second embodiment of the graphics process module that regulates the number of commands written into the buffer. In Figure 9, in step 200, the process computes the number of commands N to be drawn. In step 202, a decision is made as to whether N is less than or equal to I, where I is the reload increment defined previously. If N is greater than I, an error condition has occurred and the program exits in step 204. If N is less than or equal to I, the process proceeds to step 206 to determine if N is less than C, where C is the count for that process. If the answer is yes, then the process proceeds to step 210 to determine if the depth of the buffer is less than the reload threshold level. If yes, in step 212 it is determined whether the depth is empty. In other words, it is determined whether the buffer is empty. If so, the process proceeds to step 218 where the counter is set to F, the total number of commands for the buffer. If the depth is not equal to empty, the process proceeds to step 214 to set the count equal to I, which is the number of commands when the buffer is full minus the number of threshold commands. a replacement is shown in the flowchart, because there are at least I locations free in the buffer, and because a replacement is non-interruptible in the best mode of the invention. However, if the system processor permits other non-interruptible operations with very low overhead, the addition of C+I could be substituted for the value of I for step 214.

0023Returning to step 210, if the depth is not less than the reload threshold, the rendering context manager program is called and the graphics process is placed in a wait state until the reload threshold interrupt has been received. Then, in step 220, after this reload threshold interrupt has been received, the rendering context manager returns the graphics process to an active status and proceeds to step 208. In step 208, the counter is recomputed to be the count minus the N commands being loaded and the module is exited.

0024Figure 10 is a third embodiment of the graphics application command loading module. In step 250, the number of commands N is computed. In step 252, it is determined whether N is less than or equal to I. If not, then in step 254 a temporary count is set that is equal to N-I and then N is set equal to I. The effect of this is to place only I commands at a time into the FIFO, with TEMP holding the unwritten residue each pass. The process then proceeds to step 258.

0025In step 252, if N is less than or equal to I, the process proceeds to step 256 where the temporary parameter is set to zero. TEMP is set to zero to indicate that no additional looping is required.

0026Proceeding to step 258, it is determined whether or not N is less than C. If N is less than C then a determination is made as to whether the depth is less than the reload threshold in step 264. If so, then in step 266, it is determined whether the depth is equal to the empty state or whether the buffer is empty. If so, in step 270, C is set equal to F, the total number of commands in the buffer. The process then proceeds to step 260. Returning to step 266, if the depth is not equal to empty, then the process proceeds to step 268 to set C equal to I. The process then proceeds to step 260. Returning to step 264, if the depth is not less than the reload threshold, then the process proceeds to step 272 to call the rendering context manager and entering a wait state waiting for the reload threshold interrupt to be returned. When this interrupt is returned, the process proceeds to step 274 and then proceeds to step 260 where C is then adjusted to C-N and the N words are stored into the command. Then in step 262, it is determined if the temporary count is set to zero. If not, the process returns to step 251, in which N is restored, and thence to step 252 to repeat, since a residue of unwritten commands remain. If the temporary count is zero, the process is exited.

0027Although the invention has been described with reference to this specific embodiment, this description is not meant to be construed in a limiting sense. Various modifications of the disclosed embodiment, as well as other embodiments of the invention, will become apparent to persons skilled in the art upon reference to the description of this invention. It is, therefore, contemplated that the appended claims will cover any such modifications or embodiments as fall within the true scope of the invention.