Data processing system, method of operating a data processing system, and medium
By introducing a synchronization unit into the data processing system and using hardware signals to exchange progress messages, the synchronization problem between the producer and the consumer processing unit is solved, the waiting time is reduced, and the efficiency and user experience of the data processing system are improved.
Patent Information
- Application Number
- CN201810593687.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2018-06-11
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2038-06-11
AI Technical Summary
In data processing systems, there is a long waiting time for data synchronization between producer and consumer processing units, especially in low power consumption and mobile devices, resulting in a reduced user experience, especially in gaming and virtual reality use cases .
The synchronization unit is introduced to maintain the data output records of the producer processing unit through hardware signals, and exchange progress messages between the producer and the consumer, so that the consumer processing unit can control the reading of data from the memory according to the received progress message, so as to realize the Synchronization of data output.
It significantly reduces the waiting time between the producer and the consumer processing unit, so that the consumer processing unit can start reading data when the part of the data output is generated without waiting for the entire output to complete, improving the synchronization efficiency of the data processing system and user experience.
Smart Images

Figure CN110580238B_ABST
Abstract
Description
Technical Field
[0001] DATA PROCESSING SYSTEM The present invention relates to data processing systems and in particular to the synchronization of operations of hardware units in a data processing system. Background Art
[0002] In a data processing system, a "producer" processing unit may produce (generate) data outputs that are then used (e.g., processed) by one or more other "consumer" processing units of the data processing system. An example of this would be in a multimedia subsystem, where, for example, a video decoder may decode encoded video data representing a series of video frames to be displayed with one or more other processing units (such as a graphics processing unit), and then process the decoded video frames in a desired manner before those video frames are provided to a display for display.
[0003] Figure 1 An exemplary data processing system including a multimedia subsystem is schematically shown. Figure 1 As shown, data processing system 200 includes a multimedia subsystem in the form of a system on a chip (SOC) 202. The system also typically includes off-chip (main) memory 216, a display device 218, and a camera 220.
[0004] Multimedia subsystem SOC 202 includes a central processing unit (CPU) 204 , a graphics processing unit (GPU) 206 , a video processor 208 , a display controller (display processor) 210 , an interconnect 212 , and a memory controller 214 .
[0005] like Figure 1 As shown, CPU 204, GPU 206, video processor 208, and display controller 210 communicate with each other via interconnect 212 and with memory 216 via interconnect 212 and memory controller 214. Display controller 210 also communicates with display device 218. Camera 220 also communicates with multimedia subsystem SOC 202 via interconnect 212.
[0006] In such Figure 1In the data processing system shown, the video processor 208 may, for example, be operable to decode encoded video data that has been stored in the memory 216, and then store the decoded video data in the memory 216 for subsequent processing by, for example, the GPU 206. The GPU 206 may accordingly store the processed video data in the memory 216, which is then used by the display controller 210 to provide to the display device 218 for display. Thus, in this case, the video processor 208 will act as a producer processing unit that produces frames of decoded video data for consumption by, for example, the GPU 206, while the GPU 206 accordingly acts as a producer processing unit to provide processed video frames for consumption (use) by the display controller 210.
[0007] In such Figure 1 In the illustrated arrangement, a "producer" processing unit typically stores the data it is producing in appropriate memory that is shared with (and accessible to) consumer processing units that will use the data, which then read the data from the memory for use.
[0008] An important aspect of this operation is to synchronize the reading of data from the memory by the consumer processing unit with the writing of data to the memory by the producer processing unit. For example, the consumer processing unit must be controlled to avoid attempting to read data from the memory before the data is stored in the memory by the producer processing unit.
[0009] Such synchronization of operations can be provided by using software "fences" to synchronize the operations of different processing units. These software fences are typically controlled by and operate under the control of a corresponding driver for the processing unit (which is running on the central processing unit of the overall data processing system).
[0010] For example, in the above Figure 1 In the case of the multimedia subsystem in question, when the video processor 208 has finished its decoding of the video frame and has stored the decoded video frame in the memory 216, it can signal an interrupt to the video processor driver executing on the CPU 204, and the video processor driver recognizes the interrupt as indicating that the generation of the video frame is complete, and therefore transmits the event to the driver for the consumer processing unit that will use the decoded video frame, such as for the GPU 206. The driver for, for example, the GPU 206 will receive the message and then trigger, for example, the GPU 206 to process (use) the decoded video frame now present in the memory 216.
[0011] Accordingly, once the GPU 206 has finished processing the decoded video frame and storing the decoded video frame in memory, it will accordingly signal an interrupt to the GPU driver on the CPU 204 indicating that the complete frame has been rendered, and the GPU driver will then recognize this event and accordingly signal (for example) the driver for the display controller (display processor) 210 so that the display controller driver will then trigger the display controller 210 to process the rendered frame from the GPU 206 for display.
[0012] Figure 2 and Figure 3 This operation is illustrated.
[0013] Figure 2 Shown with Figure 1 An exemplary multimedia subsystem stack corresponding to the data processing system and the multimedia subsystem.
[0014] like Figure 2 As shown, for example, an application 30 executed on the CPU 204 of the multimedia subsystem 202 will interact with appropriate multimedia subsystem hardware 31 (including, for example, a graphics processing unit 206, a video processor 208, a display processor 210, and memory 216) via appropriate APIs 32 and corresponding drivers 33 for the hardware units 31. As discussed above, as part of this operation, the communication and control of the hardware processing units 31 will be synchronized by means of software "fences" 34 implemented and realized by the corresponding drivers 33 for the hardware units 31.
[0015] Figure 3 This is illustrated for the case where graphics processor 206 draws a frame, which will then be used by display processor 210 to display the frame on display device 218 .
[0016] like Figure 3 As shown, this operation will first include, after system startup, the initialization of the graphics processor 206 (step 40) and the initialization of the display processor 41 accordingly (step 41). The driver 35 for the graphics processor 206 will then prepare the appropriate commands and data for the graphics processor to draw the desired frame (step 42). As part of this operation, the GPU driver 35 will set a "fence" to identify and signal the completion of the frame (step 43).
[0017] Accordingly, the display processor driver 36 will prepare the appropriate buffer 37 and wait on the "fence" generated by the GPU driver 35 (the display processor driver will not send commands to the display processor hardware until the wait "fence" is signaled). The display processor driver can also set its own "fence" for synchronization if the same buffer is used.
[0018] The graphics processor driver 35 will then issue the appropriate commands and data to the graphics processor hardware 206 which will then draw the frame (step 46) and write the frame into the appropriate buffer 47 in memory.
[0019] When the GPU hardware 206 finishes drawing the frame, it will signal an interrupt to the GPU driver (step 48). The GPU driver will therefore signal the Android sync service (step 50) (which controls the sync "fence") that a "done" fence has occurred (step 49), which will then signal the GPU hardware fence that it has completed (step 51).
[0020] The display processor driver 36 will accordingly be signaled that the graphics processor completion fence has been signaled, and in response thereto, the display processor driver 36 will trigger the display processor hardware 210 to read the completion frame from the buffer 47 and display it on the display (steps 52 and 53).
[0021] Figure 1 , Figure 2 and Figure 3 Arrangements of the type illustrated typically have relatively long latencies. For example, in the multimedia subsystem example described above, the display (consumer) processing unit will only access the completed data output from the graphics (producer) processing unit once the entire output (e.g., a frame) has been completed. This will then result in a latency of one or more frames between, for example, the initial generation of a frame and its display (depending on how many producing and consuming units there are in the overall processing pipeline).
[0022] This latency can be a problem, especially in low-power and mobile devices. For example, longer latency can degrade the user experience, especially in gaming and virtual reality (VR) use cases.
[0023] Applicants therefore believe that there is room for improvement in the synchronization and processing of data output shared between producing and consuming processing units in a data processing system. Summary of the invention
[0024] According to a first aspect of the present invention, there is provided a data processing system, the data processing system comprising:
[0025] one or more producer processing units operable to generate data output;
[0026] one or more consumer processing units operable to use the data output produced by the producer processing unit; and
[0027] a memory for storing data output generated by the one or more producer processing units;
[0028] in:
[0029] The one or more producer processing units each include processing circuitry operable to generate a data output and store the data output in the memory; and
[0030] The one or more consumer processing units each include processing circuitry operable to read a data output from the memory and use the read data output;
[0031] The data processing system further comprises:
[0032] a synchronization unit operable to communicate with the one or more producer processing units and the one or more consumer processing units, and the synchronization unit operable to:
[0033] maintaining a record of data output being generated by a producer processing unit of the data processing system;
[0034] receiving a message from a producer processing unit of the data processing system regarding the progress the producer processing unit has made in generating data output; and
[0035] In response to a message received from a producer processing unit regarding progress the producer processing unit has made in producing data output:
[0036] sending a message to one or more consumer processing units of the data processing system regarding the progress that a producer processing unit has made in producing data output;
[0037] in:
[0038] The one or more producer processing units each further include processing circuitry operable to send a message to the synchronization unit regarding the progress the producer processing unit has made in generating data output; and
[0039] The one or more consumer processing units each further include processing circuitry operable to:
[0040] A data output progress message is received from the synchronization unit and reading of data output from the memory is controlled according to the data output progress message received from the synchronization unit.
[0041] According to a second aspect of the present invention, there is provided a method for operating a data processing system, the data processing system comprising:
[0042] one or more producer processing units operable to generate data output;
[0043] one or more consumer processing units operable to consume data output produced by the producer processing unit;
[0044] a memory for storing data output generated by the one or more producer processing units; and
[0045] a synchronization unit operable to communicate with the one or more producer processing units and the one or more consumer processing units;
[0046] The method includes:
[0047] A producer processing unit of the one or more producer processing units generates a data output and stores the data output in the memory; and
[0048] one or more of the one or more consumer processing units reading a data output from the memory and using the read data output;
[0049] The method further includes:
[0050] The synchronization unit maintains a record of the data output being generated by the producer processing unit;
[0051] the producer processing unit that is producing the data output sends a message to the synchronization unit regarding the progress that the producer processing unit has made in producing the data output;
[0052] the synchronization unit receiving from the producer processing unit the message regarding the progress that the producer processing unit has made in producing the data output, and, in response to the message regarding the progress that the producer processing unit has made in producing the data output from the producer processing unit, sending to one or more consumer processing units of the data processing system a message regarding the progress that the producer processing unit has made in producing the data output; and
[0053] One or more of the consumer processing units receives a data output progress message from the synchronization unit and controls its reading of data output from the memory according to the data output progress message received from the synchronization unit.
[0054] The present invention relates to the situation where a producer processing unit of a data processing system is producing data output to be used by one or more other consumer processing units of the data processing system.
[0055] In the present invention, a synchronization unit is provided which maintains a record of data outputs being produced by producer processing units of the data processing system. The producer processing units update the synchronization unit as they progress in producing the data outputs, and the synchronization unit then signals progress updates to consumer processing units of the data processing system so that they can then use the progress information to control their reading of the data outputs from the memory (e.g. to determine if and when there is data available for them to read).
[0056] Therefore, as will be further discussed below, the present invention uses a hardware unit (synchronization unit) to control the synchronization of production and consumption units in a data processing system, and the production and consumption processing units are synchronized by (hardware) signals (rather than software "fences") to and from the synchronization unit.
[0057] As will be discussed further below, this can be used to significantly reduce latency between producing and consuming hardware units, compared to, for example, arrangements using software "fences". For example, and as will be discussed further below, the present invention can facilitate synchronizing the operation of a producer processing unit and a consumer processing unit on a small portion of a given data output being produced, rather than, for example, the consumer processing unit having to wait until the producer processing unit has produced an entire output (e.g., a frame) before the consumer processing unit can begin to use that output. It can also facilitate more flexible control of the synchronization rate (intervals that trigger synchronization) between the producer processing unit and the consumer processing unit.
[0058] The present invention may be used in any desired and suitable data processing system in which a "producing" processing unit will generate and store data output for use by one or more "consuming" processing units. Examples of data processing systems to which the present invention is particularly applicable include video processing systems, image processing systems, and graphics processing systems.
[0059] One (and each) producer processing unit may accordingly be any suitable and desired processing unit of a data processing system that can generate data output for use by one or more other processing units of the data processing system. In a preferred embodiment, the producer processing unit is one of a video processor (e.g., a video encoder and / or a video decoder), a graphics processor, an image processor, a digital signal processor (DSP), and a central processing unit of the data processing system.
[0060] There can be only one producer processing unit in the data processing system. However, in a preferred embodiment, there are multiple producer processing units. The multiple producer processing units may include, for example, one or more of a video processor, a graphics processor, an image processor, a digital signal processor, and a central processing unit (CPU), and preferably include multiple of them, and preferably include all of them.
[0061] Thus, in a preferred embodiment, the data processing system includes a plurality of producer processing units, each of which operates in the manner of the present invention.
[0062] Each producer processing unit preferably has an associated producer processing unit identifier that identifies the producer processing unit (and preferably uniquely identifies the producer processing unit within the entire set of producer processing units in the data processing system (and that are operable to interact with the synchronization unit in the manner of the present invention)).
[0063] The present invention can be used for all forms of data output that can be provided and / or used by a processing unit of a data processing system. Thus, the data output being produced by a producer processing unit may include any suitable and desired data output that can be used by other processing units of the data processing system. This may depend on, for example, the nature of the producer processing unit.
[0064] The data output may represent (for example, and in preferred embodiments does represent) an array of data elements (such as images or frames), for example, for display. Thus, the data output being generated may include image data, and / or may correspond to one or more images or frames of image data (and in preferred embodiments, this is the case). In preferred embodiments, the data output comprises images or frames generated by a video processor, graphics processor, or image processor.
[0065] One (and each) consumer processing unit may accordingly be any suitable and desired processing unit of the data processing system that can use data output produced by a processing unit of the data processing system.
[0066] In a preferred embodiment, the consumer processing unit is one of a video processor, a graphics processor, an image processor, a digital signal processor (DSP), a display processor (controller), and a central processing unit of a data processing system.
[0067] There may be only a single consumer processing unit, but in a preferred embodiment the data processing system includes a plurality of consumer processing units. The plurality of consumer processing units may include, for example, one or more of a video processor, a graphics processor, an image processor, a digital signal processor, a display processor (controller), and a central processing unit (CPU), and preferably include a plurality of them, and preferably include all of them.
[0068] Where the data processing system comprises a plurality of consumer processing units, then each consumer processing unit is preferably operable in the manner of the present invention.
[0069] Each consumer processing unit preferably has an associated consumer processing unit identifier that identifies the consumer processing unit (and preferably uniquely identifies the consumer processing unit within the entire set of consumer processing units in the data processing system (and that are operable to interact with the synchronization unit in the manner of the present invention)).
[0070] The consumer processing unit can use the data output being produced by the producer processing unit in any suitable and desired manner. This may depend on, for example, the nature of the consumer processing unit and / or the nature of the data output (e.g., as discussed above).
[0071] In one preferred embodiment, the consumer processing unit is a graphics processing unit and performs graphics processing on the data output it is consuming. In another preferred embodiment, the consumer processing unit is a display controller and performs display processing to display the data output it is consuming on a display.
[0072] There may be only one consumer processing unit using the data output, but in a preferred embodiment there may be and preferably are more than one (multiple) consumer processing units using the (same) data output. In the event that multiple consumer processing units are reading and using the same data output, each of the consumer processing units preferably operates in the manner of the present invention (independently of the other consumer processing units).
[0073] Accordingly, in a preferred embodiment, a given consumer processing unit may simultaneously consume (use) multiple data outputs, for example, produced by the same or different producer processing units.
[0074] A given data output will also be capable of being actually produced and consumed (used) by a series of processing units, each operating in the manner of the present invention. Thus in this case, a first producer processing unit in a series of processing units will produce a data output that will then be used by a consumer processing unit, while the consumer processing unit then produces a data output (e.g., a modified version of the initial data output) that is used by another consumer processing unit of the data processing system (and so on, as needed). For example, a video processor can produce a decoded video frame (or sequence of video frames), which is then processed by a graphics processor to produce a modified video frame (or sequence of modified video frames), while one or more modified video frames are then used by a display controller to provide one or more modified video frames to a display for display.
[0075] A processing unit will also be able to act as both a producer processing unit and a consumer processing unit as required. In this case, a processing unit can act as only a producer processing unit or a consumer processing unit at any given time, or it may be the case that a processing unit acts as both a producer processing unit for one or more data outputs and, as required, simultaneously (concurrently) acts as a consumer processing unit for one or more (e.g., other) data outputs. In this case, a processing unit preferably behaves appropriately in the manner of the present invention for each data output it is actively processing.
[0076] Accordingly, a consumer processing unit (or one of these consumer processing units) may optionally be the same unit as one (and the) producer processing unit. Thus, there may be a processing unit that is acting as a producer processing unit and / or a consumer processing unit for data output.
[0077] In a particularly preferred embodiment, operations in accordance with the present invention are performed with respect to multiple data outputs being produced and used simultaneously. In this case, different data outputs can be produced by different producer processing units, or a given producer processing unit can be producing multiple data outputs simultaneously (e.g., where the producer processing unit can support multiple active data channels at any given time), or a combination of both.
[0078] The memory in which the data output is stored (and from which the data output is read) may be any desired and suitable memory of or for the data processing system, such as and preferably a main memory for the processing unit (for example in the case where there is a separate memory system for the processing unit), and / or a main memory of the data processing system shared with other elements, such as a host processor (CPU) of the data processing system.
[0079] The memory may be external to the processing unit. In a preferred embodiment, the memory is an external DRAM.
[0080] The producer processing unit may store its data output in the memory in any suitable and desired manner. This may, for example, depend on the nature of the data processing system and the producer processing unit.
[0081] In a preferred embodiment, one (and each) producer processing unit is operable to store its data output in an allocated area of memory (in a memory "buffer" assigned for the purpose of storing said data output). One (and each) consumer processing unit that will use the data output will then read the data output from the corresponding memory area (buffer). To facilitate this, the producer processing unit and consumer processing unit will therefore be provided with an identification of the memory buffer and / or location in memory that will be used for said data output.
[0082] Other arrangements would of course be possible.
[0083] The synchronization unit of the present invention operable to communicate with producer processing units and consumer processing units may include any suitable and desired hardware unit including appropriate processing circuitry for operating in the manner of the present invention. The synchronization unit should, and preferably does, therefore include appropriate processing circuitry for communicating with one or more producer processing units and with one or more consumer processing units in the desired manner, as well as appropriate processing circuitry for maintaining a record of data output being generated, etc., which the synchronization unit will maintain.
[0084] The synchronization unit will communicate with the producer processing unit and the consumer processing unit and synchronize the production and use of the data output in hardware without any software involvement (i.e., in particular, without any synchronization signals or messages that need to be processed in software) (once the processing of the data output has been triggered by sending one or more appropriate commands to the relevant hardware processing units). Therefore, the synchronization unit will and is operable to automatically and completely synchronize the operation of the producer processing unit and the consumer processing unit in hardware without any software involvement. Accordingly, there will be no software involvement once the processing of the data output has begun.
[0085] Thus, for example, the various progress update (synchronization) messages exchanged between the synchronization unit and the producer and consumer processing units of the present invention are sent, received and processed via hardware communication paths and in hardware, without the need for (and the absence of) any software processing of those messages.
[0086] The synchronization unit maintains a record of the data outputs being generated by the producer processing unit of the data processing system. This record may take any suitable and desired form. It can include a record of a single data output, but in a particularly preferred embodiment, the synchronization unit maintains a record of multiple data outputs being generated by the producer processing unit of the data processing system.
[0087] In a particularly preferred embodiment, the recording of the data output takes the form of a set of multiple record entries ("time slots"), each of the set of record entries having a corresponding entry (time slot) identifier and preferably being unique within the set of entries.
[0088] Each entry (time slot) in the record of data output may preferably (and preferably does) have associated with it (when allocated to a producer processing unit) an identifier for the producer processing unit to which the entry (time slot) has been allocated.
[0089] In a particularly preferred embodiment, the record of data outputs that have been generated by producer processing units of the data processing system maintained by the synchronization unit comprises, for the data outputs for which the record is maintained and preferably for each data output for which the record is maintained, an indication of the progress that the producer processing unit has made in generating said data output. Thus, in a preferred embodiment, each entry (time slot) in the record of data outputs preferably may (and preferably does) have associated with it an indication of the progress that the producer processing unit (to which the data entry (time slot) has been allocated) has made in generating one (and its) data output.
[0090] Therefore, in a particularly preferred embodiment, the record of data output comprises a set of multiple data output record entries (time slots), each of which has associated with it a corresponding entry (time slot) identifier, a corresponding producer processing unit identifier and a corresponding progress indication (preferably only these parameters).
[0091] Thus, where multiple data outputs are being generated and used, then a given (and each) data output will have its own progress record associated with that data output and maintained and used in the manner of the present invention.
[0092] The data output progress indication, for example and preferably associated with a data output log entry (time slot), may take any suitable and desired form.
[0093] In a preferred embodiment, the progress indication is an indication of how far the producer processing unit has progressed in producing the data output (and storing it in the memory). The progress indication can indicate an actual position in the data output that has been written, but most preferably indicates a (relative) position in the data output that the producer processing unit has reached, e.g. relative to a specific, preferably selected, preferably predetermined position in the data output, such as and preferably relative to the start of the data output.
[0094] The progress indication is preferably in the form of a count value. This could be, for example, a count of the number of specific data units (eg groups of scan lines) that have been produced by the producer processing unit and stored in the data output to the memory.
[0095] In a particularly preferred embodiment, the progress indication is in the form of a count of the number of progress "update" messages that the producer processing unit has sent to the synchronization unit (eg, and preferably, since initialization (startup)).
[0096] As will be discussed further below, in a preferred embodiment the producer processing unit is configured to send progress update messages to the synchronization unit at intervals (e.g. and preferably, when a particular portion (sub-portion) of the data output has been generated and stored in memory). The progress indication maintained by the synchronization unit for the data output preferably includes a count of those update messages that have been received from the producer processing unit.
[0097] In this case, the progress indication will accordingly therefore comprise (and be) a count of particular data "portions" of the data output that have been generated by the producer processing unit and stored to the memory. These data portions can comprise any suitable and desired amount of data that can be measured and counted for the data output, such as and preferably, some identifiable and regular amount of data that will be generated by the producer processing unit in generating said data output. In a particularly preferred embodiment, the data portions that are counted comprise groups of scan lines of the data output being generated (e.g., where the data output is an array of data such as an image (such as a frame)). Other arrangements will of course be possible.
[0098] A new count (e.g., and preferably, of a progress update message (data portion)) can be started each time the producer processing unit starts a new data output. However, in a preferred embodiment, the count (for a given record entry (time slot)) is simply a cumulative count, e.g., and preferably, the number of progress update messages that the producer processing unit has sent to the synchronization unit (e.g., and preferably, after initialization), and so is preferably not reset each time a new data output is started.
[0099] Thus in this case the progress indication (progress update message count) will for example be set to zero for the first data output generated by the producer processing unit after having been connected to the synchronization unit, and this counter value will then be increased for each progress update message sent by the producer processing unit when generating a data output, but when a new data output is to be generated and recorded using this producer processing unit data output record entry the initial value of the progress indication (update count) will not be reset but will be the value reached at the end of the previous data output.
[0100] Other arrangements would of course be possible.
[0101] As well as maintaining a record of the data output being produced by the producer processing units, in a particularly preferred embodiment the synchronization unit also maintains a record of the consumer processing units of the data processing system. This may also take any suitable and desired form, but in a preferred embodiment comprises (and preferably only comprises) a list of consumer processing units (that have at least "registered" with the synchronization unit), preferably in the form of (a list of) identifiers identifying the respective consumer processing units. This list preferably enumerates a plurality of consumer processing units.
[0102] The data output records that have been generated by the producer processing unit (and the records of the consumer processing unit (if maintained)) can be stored in any suitable and desired memory of the synchronization unit or accessible to the synchronization unit. In a preferred embodiment, these records are maintained in a memory local to the synchronization unit, such as in one or more caches of the synchronization unit or accessible to the synchronization unit.
[0103] The record of the consumer processing unit (when maintained) is preferably stored inside the synchronization unit. It can be configured (hard-coded) at system build time, for example, or a more flexible arrangement can be used that uses a configuration file that is then read at system startup to initialize the consumer processing unit record, if desired. It will also be possible to allow a consumer processing unit (e.g., its driver) to call the synchronization unit (e.g., its driver) to update the consumer processing unit record, if desired.
[0104] Therefore, the synchronization unit preferably includes and / or has access to local storage in which it stores the data output records or the like that it is maintaining.
[0105] In a particularly preferred embodiment, the synchronization unit includes a controller (control circuit) that is operable to interact (communicate) with producer processing units and consumer units of the data processing system (e.g., and preferably, to exchange messages with those units that are at least actively "connected" to the synchronization unit and preferably "registered" with the synchronization unit), and to maintain and update information in a record of data output being produced by the producer processing unit and a record of the consumer processing unit (when present) (e.g., and preferably, in particular in response to messages received from the producer processing unit and / or the consumer processing unit).
[0106] In a particularly preferred embodiment, the synchronization unit is driven by a driver for the synchronization unit, for example and preferably executed on a CPU (e.g., a host processor) of the data processing system. The synchronization unit driver is preferably operable to (and operates to) configure the synchronization unit hardware for initialization and synchronization, and to interface with drivers for other hardware units (producing and consuming processing units) of the data processing system in order to operate with those units.
[0107] The drivers for the synchronization unit are preferably loaded during system startup and before any "connection" of the synchronization unit to other processing units for operation with those units (so the synchronization unit driver is preferably loaded before the drivers of any other processing units that will be "connected" to the synchronization unit).
[0108] In a particularly preferred embodiment, the producer processing unit and optionally the consumer processing unit of the data processing system is operable to (and is operable to) "register" with the synchronization unit before its operation is synchronized using the synchronization unit. This is preferably done when the data processing system is initialized for operation (e.g. at switch-on / startup).
[0109] In a particularly preferred embodiment, such registration of the processing unit with the synchronization unit is performed by an appropriate exchange of messages between and for the processing unit and the synchronization unit. This is preferably done by means of an appropriate exchange of messages between the corresponding drivers for one or more processing units and the driver for the synchronization unit. For example, and preferably, a driver for a producer processing unit or a consumer processing unit can register with the synchronization unit by sending an appropriate API call to the synchronization unit driver (for example, and preferably, during initialization of the producer processing unit or the consumer processing unit).
[0110] Therefore in a preferred embodiment, the consumer processing unit (and these consumer processing units) is operable to (and preferably operates to) register itself with the synchronization unit (and thereby indicate to the synchronization unit that it should be added to a list of consumer processing units in the synchronization unit), for example and preferably during its initialization.
[0111] To facilitate this operation, the consumer processing unit (e.g., and preferably, a driver for the consumer processing unit) may preferably send a "connect" message to the sync unit (e.g., and preferably, a driver for the sync unit) to "register" the consumer processing unit with the sync unit (so as to "connect" to the sync unit for synchronization purposes). This message preferably indicates an identifier for the processing unit (which will therefore and preferably be recorded in the consumer processing unit record maintained by the sync unit).
[0112] Accordingly, the producer processing unit is preferably operable to request, preferably during initialisation of the producer processing unit, one or more entries (time slots) in a record of data output being produced by the producer processing unit maintained by the synchronisation unit.
[0113] For example, and preferably, depending on how many data outputs the producer processing unit may and / or is likely to and / or desire to produce (simultaneously), the producer processing unit may request from the synchronization unit a single or multiple entries in the producer processing unit record. A driver for the producer processing unit can, for example, determine how many entries will be needed.
[0114] To facilitate this operation, the producer processing unit (e.g., and preferably, a driver for the producer processing unit) may preferably send a "connect" message to the synchronization unit (e.g., and preferably, a driver for the synchronization unit) to request the producer processing unit data output record entry (time slot) from the synchronization unit (in order to "connect" to the synchronization unit for synchronization purposes). This message preferably indicates an identifier for the processing unit (which will therefore and preferably be recorded against the producer processing unit data output record entry in the synchronization unit).
[0115] In response to such a message, the synchronization unit (e.g., and preferably, its driver) preferably returns a message indicating the entry (time slot) that has been allocated to the producer processing unit (e.g., and preferably, in the form of an identifier (entry (time slot) ID) for the allocated record entry, or a message (e.g. a specific, preferably predefined "default" value) indicating that the entry (time slot) in the synchronization unit is not available (i.e., the producer processing unit cannot connect to the synchronization unit for synchronization of its data output).
[0116] Most preferably, in response to such a registration request, the synchronization unit (e.g., preferably its driver) first determines whether there are available entries (time slots) in the producer processing unit data output record (e.g., checks the usage status of each entry (time slot)), and when it finds a free entry (time slot), it allocates the entry (time slot) to the producer processing unit and sets the producer processing unit identifier for the entry (time slot) to the identifier of the requesting producer processing unit, and then returns the relevant entry (time slot) identifier to the producer processing unit (e.g., its driver), so that the producer processing unit knows which producer processing unit record entry (time slot) it has been allocated and will use.
[0117] Where each producer processing unit data output record entry maintained by the synchronization unit has a data output progress indication associated with it, then in a particularly preferred embodiment the synchronization unit also sets the progress indication for the allocated entry (time slot) to a specific, preferably selected, preferably predetermined initialization value, for example and preferably zero.
[0118] Thus, in a particularly preferred embodiment, the method of the present invention comprises (and one or more producer processing units and synchronization units are operable) that the producer processing unit first requests from the synchronization unit an allocation of one or more entries in a producer processing unit data output record maintained by the synchronization unit, and in response to such a request, the synchronization unit allocates one or more entries in the producer processing unit data output record maintained by the synchronization unit to the requesting producer processing unit. In a particularly preferred embodiment of this, the synchronization unit first checks whether there is an available entry in the producer processing unit data output record before allocating the entry to the producer processing unit. As part of this process, the synchronization unit preferably notifies the producer processing unit of the identification of the entry in the producer processing unit data output record to which it has been allocated.
[0119] In a preferred embodiment, the synchronization unit may maintain as many producer processing unit data output record entries (time slots) as are ever likely to be needed by the producer processing units of the data processing system (because the overhead for maintaining records of producer processing unit data outputs is relatively low).
[0120] Accordingly, in a preferred embodiment the producer processing unit may (and does) continuously maintain its allocated one or more entries (time slots) in the producer processing unit data output record maintained by the synchronization unit (i.e., remain "connected" to the synchronization unit) until such time as when the data processing system is, for example, shut down or ceases its current operation.
[0121] However, it would also be possible to have an arrangement where there is a more limited set of entries for producer processing unit data output records, so that, for example, it may be necessary to add and remove producer processing units (and / or data outputs) from the usage list. This could be achieved, for example, by providing an appropriate messaging arrangement for this purpose.
[0122] For example, it would be possible to allow a processing unit to also exchange "disconnect" messages with a synchronization unit, e.g., to allow a processing unit to "disconnect" from a synchronization unit in use (and, e.g., thereby freeing up producer processing unit data output record entries for use by other producer processing units). In this case, the processing unit could, for example, send a "disconnect" message identifying the processing unit that wishes to disconnect, and the synchronization unit would then identify any entries associated with that processing unit and release (clear) those entries accordingly. The synchronization unit could also return a message to the processing unit that has requested the "disconnect" indicating whether the disconnect has been successful (e.g., whether there was an error).
[0123] Once a producer processing unit has registered with the synchronization unit and has been assigned one or more entries in the synchronization unit's producer processing unit data output record, the producer processing unit can (and will) communicate with the synchronization unit to facilitate synchronization with consumer processing units that are using the data output being produced by the producer processing unit.
[0124] As discussed above, a producer processing unit that is producing data output (and, for example and preferably, has "registered" with the synchronization unit) will send one or more messages to the synchronization unit regarding the progress that the producer processing unit has made in producing the data output. These messages will be sent via hardware communications (there is a direct hardware connection between the synchronization unit and all producers and consumers) so that no software is involved in the synchronization message sending and receiving.
[0125] Progress update messages regarding the progress a producer processing unit has made in generating data output may take any suitable and desired form.
[0126] In a particularly preferred embodiment, one (and each) progress update message identifies at least said producer processing unit, for example and preferably by including an identifier for said producer processing unit.
[0127] In a particularly preferred embodiment, one (and each) progress update message also identifies the producer processing unit data output record entry (time slot) to which the message relates (i.e. the specific data output in the record being maintained by the synchronization unit to which the progress update message relates), for example and preferably by including an identifier for said data output record entry (time slot) in the progress update message.
[0128] Where the data output record maintained by the synchronization unit also includes a progress indication for each data output (as discussed above), then one (and each) progress update message sent by the producer processing unit when generating a data output preferably also includes a progress indication (e.g., and preferably, a suitable counter value) indicating the progress the producer processing unit has made in generating the data output.
[0129] Therefore, in a particularly preferred embodiment, the progress update message sent by the producer processing unit includes (and preferably only includes): an indication of the identification of the producer processing unit, an indication of the identification of the producer processing unit record entry (time slot) for the data output, and an updated progress indication (e.g., an updated counter value).
[0130] Accordingly, in a particularly preferred embodiment one (and each) producer processing unit maintains a data output progress record, such as a count, for example and preferably in terms of the number of progress update messages it has sent (e.g. for the data output or as a whole (e.g. since initialization), as discussed above), for one (and individually, for each) corresponding synchronization unit data output record entry to which it has been assigned (which the synchronization unit is tracking), and preferably periodically updates (e.g. increments) the progress record (e.g. counter), for example when it is about to send a new progress update message to the synchronization unit, and includes the updated (incremented) progress record (e.g. counter) value in the progress update message it sends to the synchronization unit.
[0131] The producer processing unit can simply send a single progress update message to the synchronization unit for a (given) data output, for example when it has finished generating (and storing) said data output. However, in a particularly preferred embodiment, the producer processing unit can (and preferably does) send progress updates during the generation of its data output (while it is generating the data output) rather than only when it has finished generating a given data output. Therefore in this case, the producer processing unit will (and preferably does) send progress update messages to the synchronization unit at intervals (and periodically) while it is generating the data output. In other words, the producer processing unit will send update messages after it has generated a corresponding part (small part) of the data output, rather than only when it has completed said data output.
[0132] Thus in this case the producer processing unit will send a number of progress updates, for example and preferably at preferably regular intervals, as the data output is generated. The intervals are preferably measured in terms of the amount (fraction) of data output produced (rather than in terms of eg time).
[0133] The number of progress updates to be provided when generating a data output (the rate at which data output progress updates are sent by the producer processing unit to the synchronization unit when generating the data output) may be selected and set as desired. The number of updates may be fixed, for example, for the producer processing unit and / or for the data processing system as a whole, but in preferred embodiments the number of progress updates to be provided for and per the data output may be configured and set in use. In particularly preferred embodiments, the number of progress updates to be provided for a data output is determined and set by a driver for the producer processing unit.
[0134] The number of progress updates that will be provided for a data output will set the granularity (rate) at which the operation of the producer processing unit and the consumer processing unit are synchronized for the data output as will be discussed further below, and may be selected based on any suitable and desired criteria. In a preferred embodiment, this is based at least in part on the processing rates of the different processing units (producing unit and consuming unit), for example and preferably, in order to seek to maximize the utilization of the processing units in processing a given data output. For example, if the processing unit is relatively slow, the number of progress updates for the data output may be reduced (i.e., the progress update rate will be reduced) compared to an arrangement in which the processing unit is able to process the data output relatively more quickly.
[0135] The number of progress updates can also be based, for example, on the processing that the data output will undergo and / or the manner in which the data output is generated. For example, where the processing that the data output will undergo requires a certain amount (e.g., area) of the data output (e.g., where a filtering "window" will be applied to the data output), it may be appropriate to have relatively few progress update points (i.e., to allow each progress update to relate to a (relatively) large amount (e.g., area) of the data output).
[0136] Similarly, where data output is processed on a sub-region by sub-region basis, such as, for example, might be the case with a tile-based graphics processing unit (GPU), progress updates can be configured and arranged to be performed after each sub-region or after each group of a particular, preferably selected, preferably predefined number of sub-regions have been processed (e.g., generated).
[0137] Other arrangements would of course be possible.
[0138] The number of progress updates to be provided when generating the data output is preferably equally spaced across the data output, i.e. the overall data output is preferably divided into a number of (preferably equal) partitions (small portions) when there are progress updates to be provided when generating the data output, so as to divide the data output into a corresponding number of preferably equally sized partitions, with progress updates then being provided after each such partition has been generated. Thus, for example, in the case of a frame comprising, for example, 160 scan lines and a progress update rate of 10 updates during the generation of a frame, the producer processing unit will therefore provide a synchronization update message to the progress unit every 16 scan lines.
[0139] The effect of this is accordingly that the number of progress updates that will be provided will accordingly define the size of the partitions (small portions) of the data output that are generated between updates (where a larger number of updates results in a progress update being provided after a smaller portion of the data output has been generated and vice versa). Thus, the number of progress updates can be, and in preferred embodiments is, set (selected) based on the size of the data output that is expected to (and will) be generated between successive progress updates. For example, this can be done in order to ensure that enough data of the data output is available to the consumer processing unit for the consumer processing unit to process the data output in the desired manner before a progress update is sent that would trigger the consumer processing unit to process more data output.
[0140] When the synchronization unit receives a progress update message from a producer processing unit, in response to the message, it sends a message to one or more of the consumer processing units regarding the progress that the producer processing unit has made in producing the data output. These messages will again be sent via hardware communications (there is a direct hardware connection between the synchronization unit and all producers and consumers) so that no software is involved in the synchronization message sending and receiving.
[0141] As part of this operation, the synchronization unit preferably first checks that any producer processing unit identifier included with a progress update message from a producer processing unit matches the producer processing unit identifier for any record associated with a producer processing unit data output record entry (time slot) in the data output record being maintained by the synchronization unit. If the identifiers match, the synchronization unit registers the update message as valid and, in response thereto, sends a progress update message to one or more of the consumer processing units. (On the other hand, if the producer processing unit identifier for the progress update message and the indicated producer processing unit data output record entry (time slot) do not match, error operations may be performed, such as to signal an error to the producer processing unit that sent the progress update message).
[0142] Where the synchronization unit maintains a progress indication (such as a counter value) for each data output (for each producer processing unit data output record entry), then the synchronization unit preferably also updates the progress indication (e.g., and preferably, the counter value) for that data output in response to receiving a progress update message from the producer processing unit (and sending messages to one or more of the consumer processing units regarding the progress that the producer processing unit has made in producing that data output).
[0143] In the case where the progress update message from the producer processing unit includes a progress indication (e.g., an updated counter value), the synchronization unit preferably updates the progress indication (e.g., the counter value) of the producer processing unit data output record entry based on (e.g., and preferably) including the progress indication (e.g., the counter value) of the progress update message from the producer processing unit.
[0144] A progress update message sent by a synchronization unit to one or more consumer processing units preferably indicates the data output to which it relates by identifying the producer processing unit data output record entry (time slot) to which the progress update message relates (for example, and preferably, by including an identifier of the record entry (time slot) in the progress update message).
[0145] In the case where the synchronization unit maintains a progress indication (such as a counter value) for each data output (producer processing unit data output record entry), then the progress update message sent by the synchronization unit to the consumer processing unit preferably also indicates (includes) the new updated progress (e.g., and preferably counter) value that has been reached.
[0146] In a preferred embodiment, these consumer progress update messages are broadcast to multiple consumer processing units, and preferably to all consumer processing units in a list of consumer processing units maintained by the synchronization unit (if any), with each consumer processing unit then determining whether the progress update message is for the data output it is processing (and waiting for) from the producer processing unit data output record entry (time slot) identified in the progress update message from the synchronization unit (as will be discussed further below, a consumer processing unit preferably knows the identity of the data output record entry (time slot) relating to the data output it is waiting for (processing) so that it can determine the consumer progress update message that will apply to it).
[0147] Other arrangements, such as a synchronization unit being able to address its consumer progress update messages to specific consumer processing units, would be possible as desired, but would likely require more complex signaling and control operations.
[0148] A (and each) consumer processing unit may control the reading of the data output from the memory in accordance with the data output progress messages received from the synchronization unit in any suitable and desired manner. In a particularly preferred embodiment, the consumer processing unit (and each consumer processing unit) uses these data output progress messages to determine when there is data available for the data output for it to read, and accordingly, trigger the reading of the data output from the memory for processing by the consumer processing unit.
[0149] In a particularly preferred embodiment, one (and each) consumer processing unit uses the data output progress messages from the synchronization unit to ensure that it does not read more data than is available from the data output.
[0150] Most preferably the consumer processing unit uses data output progress messages from the synchronization unit to determine when it is safe to continue reading data for the data output from memory (e.g., and preferably, to determine when there is data stored in memory for the data output so that, for example and preferably, the consumer processing unit can read data for the data output from memory).
[0151] Therefore, when the consumer processing unit identifies that the producer processing unit data output record entry it is waiting for has been updated according to the consumer progress update message sent by the synchronization unit, the consumer processing unit preferably determines whether there is new data (e.g., new data output partition) available for it to read for the data output (e.g., from the buffer). If so, the consumer processing unit preferably reads the new data of the data output (from the buffer) accordingly and processes the data accordingly.
[0152] Accordingly, once the consumer processing unit has read new data of the data output (e.g. the new data output partition) in response to a progress update message sent by the synchronization unit, it preferably then waits for the next progress update message from the synchronization unit for the data output before attempting to read further data of the data output from the memory.
[0153] In a particularly preferred embodiment, as discussed above, the progress update message sent by the synchronization unit to the consumer processing unit also indicates (includes) a progress indication (e.g., and preferably a counter value) indicating the progress (e.g., position) that has been reached by the producer processing unit for the data output, and one (and the) consumer processing unit uses the progress indication (e.g., counter value) included in the progress update message sent by the synchronization unit to control their reading of data of the data output, for example, and preferably, to determine whether the data that the consumer processing unit is waiting for is available for reading for the data output (e.g., from a buffer).
[0154] For example, and preferably, the consumer processing unit can and preferably does use the progress value included in the progress update message to determine which parts of the data output (e.g. scan lines) have been written (to the buffer), and so are available for reading (and processing). This can and preferably is also based on the number of synchronization updates (number of progress update messages) that the producer processing unit will signal to the synchronization unit when producing said data output (and therefore the amount of data output that each progress update "count" actually represents and corresponds to).
[0155] For example, consider a data output with an initial progress (counter) value of n and a number of progress update messages (synchronization points) that the producer processing unit will signal to the synchronization unit when producing the data output is m, then for a data output having a height h (e.g. in terms of the number of scanned rows), in order to determine that the data output from the first row to row h / m-1 is available, the consumer processing unit will need to wait for the counter number to reach n+1 (i.e. for the update message to signal that the progress value is now n+1), and in order for any row of the data output from row h / m to h / m*2-1 to be available, the progress value (counter number) will need to be n+2, and so on.
[0156] Therefore, once the consumer processing unit receives a progress update message with a progress (counter) value, it can and preferably does determine from the progress (counter) value in the progress update message sent by the synchronization unit how much data output has been produced, and therefore determines how much data of that data output it can read and / or whether the data of the data output it is waiting for is now available for it to read for the data output.
[0157] Preferably the consumer processing unit (hardware) records a record progress (counter) value in a progress update message sent by the synchronization unit, and then determines whether the progress (counter) value in the progress update message is large enough to indicate that the data (e.g., rows of the data output) that the consumer processing unit wishes to read has been generated and stored for the data output before it starts reading data of the data output.
[0158] This may also allow a consumer processing unit that for any reason, for example, has "missed" an update message to still be able to determine how much data output has been produced.
[0159] This operation will then be repeated when the producer processing unit next provides a progress update, and so on, until the producer processing unit has generated the entire data output and the consumer processing unit has processed that output.
[0160] In a particularly preferred embodiment, a producer processing unit that is generating a data output associates a producer processing unit data output record entry (time slot) identifier that the producer processing unit is using for the data output with the data output (e.g., with an allocated storage unit (e.g., a buffer) in which the data output will be stored), for example and preferably, so that a consumer processing unit can identify the corresponding producer processing unit data output record entry (time slot) that the synchronization unit is maintaining for the data output. This will then facilitate identification of the data output record entry (time slot) associated with the data output being generated by the producer processing unit, for example and preferably, so that a consumer processing unit can identify a progress update message related to the data output.
[0161] Thus, in a particularly preferred embodiment, when a producer processing unit first begins to generate a data output (e.g., a frame), it assigns to that data output (e.g., a frame) (one of) the producer processing unit data output record entries (time slots) that have been assigned to the producer processing unit so as to identify that producer processing unit record entry (time slot) as being the entry for that data output.
[0162] In a particularly preferred embodiment, the producer processing unit provides this information of the allocated memory area (buffer) for the data output, for example and preferably as or as part of metadata associated with and provided for said data output.
[0163] The consumer processing unit that will read the data output is preferably then able to (and preferably does) read (e.g., from metadata associated with the data output) the producer processing unit data output record entry (time slot) identifier associated with the data output, in order to obtain the producer processing unit data output record entry corresponding to the data output identifier.
[0164] In a particularly preferred embodiment, as well as indicating a producer processing unit data output record entry (time slot) relating to the data output, the producer processing unit also associates with the data output further information that the consumer processing unit can (and preferably does) use to control and synchronize its use of the data output with the operation of the producer processing unit in generating the data output.
[0165] In a particularly preferred embodiment of this, the producer processing unit also associates with the data output an indication of the number of progress update messages (synchronization points) that the producer processing unit will signal to the synchronization unit when generating the data output. Thus, for example, if the producer processing unit is to signal progress updates to the synchronization unit four times during the generation of the data output (e.g., progress updates will be provided when the data output is 25%, 50%, 75% and 100% complete), the producer processing unit will associate the number of progress update points with the data output (e.g., a buffer) so that a consumer processing unit that will use the data output can identify that it should receive four progress updates during the generation of the data output by the producing processing unit output. As discussed above, this can then allow the consumer processing unit to determine what portion of the data output is indicated by a given synchronization progress (count) value.
[0166] In a particularly preferred embodiment, the producer processing unit (before it starts said data output) also associates with the data output it is about to produce a current progress indication (e.g. and preferably a counter value) of the producer processing unit data output record entry (time slot) to be used for that data output. As discussed above, this will then allow, for example, a consumer processing unit to determine, based on this value and the number of progress update messages to be sent by the producer processing unit when producing said data output, what part of the data output a given progress indication (counter) value included in a synchronization message from the synchronization unit corresponds to (indicates), and / or when a final progress update message has been received indicating that the data output has been completed by the producer processing unit.
[0167] As will be appreciated from the above, in a particularly preferred embodiment, a producer processing unit that will generate data outputs for use by one or more consumer processing units will therefore first associate a producer processing unit data output progress record entry with the data output for which it is about to produce an identification of a synchronization unit, which producer processing unit data output progress record entry will be used to monitor the progress of the producer processing unit in producing the data output, preferably together with information regarding the progress (synchronization) update rate that the producer processing unit will provide when producing the data output.
[0168] Accordingly, any consumer processing unit that will use the data output will first read the information associated with the data output (buffer) by the producer processing unit in order to identify the producer processing unit data output record entry in the synchronization unit for that data output (buffer) (and, preferably, the current synchronization update progress value for that data output and / or the number of progress updates that will be provided by the producer processing unit when generating the data output).
[0169] The producer processing unit will then start generating the data output and write it to the designated storage (buffer). Once the producer processing unit has generated the first (small portion) of the data output, after which the progress update will be sent, it will then send a progress update message to the synchronization unit accordingly. The synchronization unit will in turn broadcast the progress update message to the consumer processing units, while the consumer processing units determine whether the progress update message applies to the data output they are processing. If so, the consumer processing unit will then read in the first portion of the data output (from the buffer) and process that portion of the data output accordingly.
[0170] Meanwhile, the producer processing unit will continue to generate data output and write it to the designated storage (buffer). Once the producer processing unit has generated the next portion (small portion) of the data output, after which a progress update will be sent, it will send another progress update message to the synchronization unit, which will then provide the progress update to the consumer processing unit, which will then process the next portion (small portion) of the data output, and so on, until the data output and its processing by the consumer unit has completed.
[0171] In this way, the production and consumption of data outputs may be synchronized by the synchronization unit and at a desired level of granularity within the data outputs (eg, frames).
[0172] Although the present invention has been described above with particular reference to the generation of a single data output by a producer processing unit and the corresponding use of that data output by a consumer processing unit, as will be appreciated by those skilled in the art, these operations may be, and preferably are, performed for each data output generated by a producer processing unit for use by other processing units of the data processing system, and correspondingly, these operations may be performed for and for each producer processing unit of the data processing system that may be generating data outputs for use by other consumer processing units of the data processing system.
[0173] Any one or more or all of the processing units of the present invention may be implemented as processing unit circuits (e.g., in the form of one or more fixed function units (hardware) (processing circuits) and / or in the form of programmable processing circuits that can be programmed to perform desired operations). Similarly, any one or more or all of the processing units and processing unit circuits of the present invention may be provided as separate circuit elements to any one or more of other processing units or processing unit circuits, and / or any one or more or all of these processing units and processing unit circuits may be formed at least in part by shared processing circuits.
[0174] The processing unit and / or the data processing system described in any aspect or embodiment herein may include or may be a system on a chip (SoC), or may form part of a system on a chip (SoC).
[0175] As with specific processing units, the data processing system of the present invention may additionally include any suitable and desired elements and units that the data processing system may include, etc. Therefore, in a preferred embodiment, the data processing system also includes a host (e.g., central) processor. The host processor may, for example, execute an application that needs to be processed by the processing unit of the system data processing system. The host processor may send appropriate commands and data to the processing unit to control them to perform data processing operations and to generate and / or use one or more data outputs required by the application executed on the host processor. To facilitate this, the host processor may execute one or more drivers for the processing unit, and / or may execute one or more compilers for compiling a program to be executed by the programmable execution unit of the processing unit.
[0176] In embodiments, the processing unit or system may include and / or communicate with one or more memories and / or storage devices storing the data described herein and / or storing software for performing the processes described herein. The processing unit or system may include and / or communicate with a display for outputting a display image based on the data.
[0177] The invention may be implemented in any suitable system, such as a suitably configured computer or microprocessor based system. In a preferred embodiment, the invention is implemented in a computer and / or microprocessor based system.
[0178] The various functions of the present invention may be performed in any desired and suitable manner. For example, the steps and functions of the present invention may be implemented in hardware or software as desired. Thus, for example, unless otherwise indicated, the various circuits, functional elements, stages, units and "means" of the present invention may include one or more suitable processors, one or more controllers, functional units, circuits, processing logic, microprocessor arrangements, etc., operable to perform the various steps or functions, etc., such as appropriately dedicated hardware elements (processing circuits) and / or programmable hardware elements (processing circuits) that can be programmed to operate in a desired manner.
[0179] Various steps or functions of the present invention may be replicated and / or performed in parallel on a given processor, etc. Likewise, various processing units, etc. may share processing circuitry, etc. as desired.
[0180] Subject to any hardware necessary to perform the specific steps or functions discussed above, the system may additionally include any one or more or all of the common functional units and the like included in a data processing system.
[0181] In a particularly preferred embodiment, the various functions of the present invention are performed on a single data processing platform that generates and outputs the data streams.
[0182] It will also be appreciated by those skilled in the art that all described embodiments of the present invention may and do include in the embodiments any one or more or all of the features described herein, as appropriate.
[0183] The method according to the invention may be implemented at least in part using software such as a computer program. Therefore, further embodiments of the invention include: computer software that is specifically adapted to perform the method described herein when installed on a data processor; a computer program element that includes computer software code portions for performing the method described herein when the program element is run on a data processor; and a computer program that includes code adapted to perform all steps of one or more methods described herein when the program is run on a data processor. The data processor may be a microprocessor system, a programmable FPGA (field programmable gate array), or the like.
[0184] The invention also extends to a computer software carrier comprising such software which, when used to operate a data processing apparatus or system comprising a data processor, in combination with said data processor causes said apparatus or system to perform the steps of the method of the invention. Such a computer software carrier may be a physical storage medium such as a ROM chip, CD ROM, RAM, flash memory or disk, or may be a signal such as an electronic signal via a wire, an optical signal or a signal such as a radio signal to a satellite.
[0185] It should also be understood that not all steps of the method of the present invention need to be performed by computer software, and therefore in other embodiments, computer software and such software installed on a computer software carrier for performing at least one of the steps of the method stated herein are included.
[0186] The present invention may therefore be suitably implemented as a computer program product for use with a computer system. Such an embodiment may include a series of computer-readable instructions fixed on a tangible non-transitory medium (such as a computer-readable medium, for example a floppy disk, CD, DVD, ROM, RAM, flash memory or hard disk). It can also include a series of computer-readable instructions that can be sent to a computer system via a modem or other interface device through a tangible medium (including but not limited to optical or analog communication lines) or invisibly using wireless technology (including but not limited to microwave, infrared or other transmission technology). The series of computer-readable instructions implements all or part of the functionality previously described herein.
[0187] It will be appreciated by those skilled in the art that these computer readable instructions may be written in any number of programming languages used with any computer architecture or operating system. Furthermore, these instructions may be stored using any memory technology now or in the future including, but not limited to, semiconductor, magnetic, or optical, or sent using any communication technology now or in the future including, but not limited to, optical, infrared, or microwave. It is contemplated that such a computer program product may be distributed as a removable medium with accompanying printed or electronic documentation (e.g., shrink-wrapped software), pre-loaded with a computer system, such as on a system ROM or fixed disk, or distributed over a network (e.g., the Internet or the World Wide Web) from a server or electronic bulletin board. BRIEF DESCRIPTION OF THE DRAWINGS
[0188] Various embodiments of the present invention will now be described, by way of example only, and with reference to the accompanying drawings, in which:
[0189] Figure 1 An exemplary data processing system is shown;
[0190] Figure 2 and Figure 3 Schematically shows Figure 1 Synchronization of operations of producer processing units and consumer processing units in a data processing system;
[0191] Figure 4 A data processing system is shown which implements the present invention;
[0192] Figure 5 Shown in more detail Figure 4 A synchronization unit of an implementation method of the present invention;
[0193] Figure 6 and Figure 7 Schematically shows Figure 4 and Figure 5 Synchronization of operations of the producer processing unit and the consumer processing unit in an implementation manner.
[0194] In the drawings similar reference numerals have been used (where appropriate) for similar features. DETAILED DESCRIPTION
[0195] The accompanying drawings illustrate elements of data processing devices and systems associated with embodiments of the present invention. As will be appreciated by those skilled in the art, there may be other elements of the data processing devices and systems that are not illustrated in the accompanying drawings. It should also be noted here that the accompanying drawings are only schematic, and for example, although the elements shown are schematically shown as separate elements in the accompanying drawings, in reality they may share effective hardware circuits.
[0196] Figure 4 4 shows an embodiment of a data processing system 400 including a multimedia subsystem 402 according to the present invention. Figure 4 As shown, the data processing system 400 and Figure 1 The illustrated data processing system is similar (and therefore accordingly includes a multimedia subsystem 402 in the form of a system on a chip (SoC), wherein the system on a chip includes, among other things, a central processing unit 204, a graphics processing unit 206, a video processor 208, a display controller (processor) 210, an interconnect 212 and a memory controller 214, as well as an off-chip (main) memory 216, a display device 218 and a camera 220).
[0197] However, with Figure 1 Unlike the data processing system shown in FIG. 1 , the data processing system 400 (and in particular the system-on-chip 402) of this embodiment includes a data processing system capable of communicating with the system-on-chip 402 (eg, Figure 4 A hardware "generic" synchronization unit (GSU) 404 for communicating with various hardware units (shown).
[0198] As will be discussed further below, synchronization unit 404 is used to synchronize the operation of hardware units of system-on-chip 402, rather than using software "fences" to do so. In particular, the operation of the hardware units is synchronized by signals from synchronization unit 404 rather than by using software fences.
[0199] Figure 5 The general synchronization unit (GSU) 404 is shown in more detail.
[0200] like Figure 5 As shown, general synchronization unit 404 includes controller (control unit) 500 and maintains a record of data output being generated by producer processing units of data processing system 501, and a record of consumer processing units of data processing system 502. Each of the various hardware units (such as CPU 204, GPU 206, video processor 208, and display processor 210) can be a producer processing unit and / or a consumer processing unit.
[0201] The controller 500 operates to interact with other hardware units that are communicating with (and connected to) the synchronization unit 404 , and maintains (eg, initializes and updates) the producer processing unit data output record 501 and the consumer processing unit record 502 .
[0202] The consumer processing unit record 502 lists the consumer processing (hardware) units that have “registered” with the synchronization unit 404 in the form of a list of corresponding consumer hardware unit identifiers 506 .
[0203] This consumer processing unit record 502 can, for example, be hard-coded when the system is set up, or initialized at system startup (for example according to an appropriate configuration file) by a driver for the synchronization unit, or the consumer processing unit can also or instead use appropriate driver calls to, for example, register with the consumer processing unit record.
[0204] The producer processing unit data output record 501 comprises a series of data output record entries (time slots) 503, each of which has a unique entry (time slot) identifier (within the record). For each entry (time slot) in the producer processing unit data output record, the synchronization unit 404 records the identification 504 of the corresponding producer hardware unit to which the entry (time slot) has been allocated, and maintains a progress counter value 505 indicating the progress that the producer processing unit has made in generating the data output to which the data output record entry (time slot) relates.
[0205] The progress counter value may be, for example, a 64-bit integer and, in the present embodiment, is initialized to 0 and then incrementally increased each time a producer processing unit to which an entry (time slot) has been allocated sends a progress update message as it produces one or more data outputs (thus, the progress counter value is a cumulative count of the number of progress update messages that have been sent for the data output record entry (time slot) and is cumulatively incremented by 1 each time a progress update message for the data output record entry is sent by the producer processing unit). (This will be discussed in more detail below).
[0206] The Generic Synchronization Unit (GSU) 404 is driven by a driver (which runs, for example, in the Linux kernel space) that configures the GSU hardware for initialization and synchronization, and interfaces with drivers of other hardware units to enable those other hardware units to interact and cooperate with the GSU 404.
[0207] Before the GSU 404 is "connected" to other hardware units, the GSU driver is loaded during system startup to initialize the GSU hardware. When loading is complete, the GSU hardware is initialized.
[0208] As will be discussed further below, the GSU 404 is operable to synchronize all hardware units connected to it. The connected hardware units may be data output "producers" or "consumers." The GSU 404 may operate with any and all hardware units of the multimedia subsystem, such as the GPU, video processor, ISP, and display processor.
[0209] In this embodiment, a data output producer processing unit (such as GPU 206 or video processor 208 ) that will generate data output will first “connect” to synchronization unit 404 .
[0210] To do this, the producer processing unit will request one or more producer processing unit data output record entries (time slots) from the synchronization unit 404. This is done by the driver for the producer processing unit signaling the driver for the synchronization unit 404 during initialization of the producer processing unit.
[0211] For example, the driver for a producer processing unit might send a message (function call) of the following form:
[0212] int gsu_connect(int hw_id)
[0213] Wherein “hw_id” specifies the identifier (eg, an integer) of the producer processing unit hardware to be connected to the GSU 404 .
[0214] In response to the gsu_connect() call from the driver of the producer processing unit, the driver for the GSU 404 checks the usage status for each data output record entry (time slot) in the producer processing unit data output record 501. If the driver finds a free entry (time slot), it sets the producer hardware unit ID 504 for the entry (time slot) to the indicated hw_id field in the message from the producer processing unit, and then returns the entry (time slot) identifier for the allocated data output record entry (time slot) to the driver of the producer processing unit.
[0215] If the sync unit driver does not find a free entry (time slot) in the producer processing unit data output record 501, it returns a "not available" value, such as -1, to the driver of the producer processing unit to indicate that situation.
[0216] (Any hardware unit that wishes to connect to the GSU 404 will call this function to get the slot ID before synchronizing with the GSU 404.)
[0217] In this embodiment, there are sufficient producer processing unit data output record entries (time slots) for each of the producer processing units that may be expected to be connected to the synchronization unit 404, so that the producer processing units maintain their entries (time slots) in the producer processing unit data output record 501 maintained by the synchronization unit 404 until the system is shut down.
[0218] However, if desired, it will also or alternatively be possible to allow the producer processing unit to actively "disconnect" from the synchronization unit 404. In this case, the driver for the hardware unit that wishes to disconnect can send a message of the following form to the driver for the synchronization unit 404:
[0219] int gsu_disconnect(int hw_id);
[0220] Where hw_id specifies the hardware ID of the producer processing hardware unit that is to be removed from the synchronization unit.
[0221] In response to such a gsu_disconnect() call from the producer processing unit's driver, the driver for GSU404 will check the hardware identifiers associated with each data output record entry (time slot) in the producer processing unit data output record 501 to find one or more entries that match the hardware ID included in the gsu_disconnect() message.
[0222] If the sync unit driver finds one or more entries with the corresponding hardware ID, it clears the contents of that one or more producer processing unit data output record entries (time slots) and returns a data value (such as 0) to the driver for the hardware unit to indicate that the "disconnect" has been successful. The cleared one or more producer processing unit data output record entries (time slots) then become available for allocation to another one or more producer processing units.
[0223] If the sync unit driver does not find a data output record entry (time slot) with the hardware ID, it returns an "error" value, such as -1, to the driver of the producer processing unit to indicate that an error has occurred.
[0224] Once a producer processing unit (such as graphics processor 206 ) has been connected to synchronization unit 404 in the manner described above, the producer processing unit may then begin producing data outputs, with the production and consumption of these data outputs being synchronized via synchronization unit 404 .
[0225] When a producer processing unit is about to generate a data output, the driver for the producer processing unit will first determine the synchronization strategy to be used for the data output according to the number of progress update messages that the producer processing unit will send to the synchronization unit 404 while generating the data output. For example, the producer processing unit driver may determine that the producer processing unit should send four progress update messages to the synchronization unit 404 while generating the data output (in which case the producer processing unit will send progress update messages after it has completed 25%, 50%, 75%, and 100% of the data output).
[0226] The producer processing unit also maintains a progress counter for (and for each) sync unit data output record 501 entry (time slot) it has been allocated. This progress counter is initialized to 0 when the producer processing unit first connects to the sync unit 404, and is then incremented by one each time the producer processing unit sends a progress update message to the sync unit 404 for that producer processing unit data output record entry.
[0227] The driver for the producer processing unit is then associated with a buffer in which the data output is to be stored (a set of "synchronization data"), the set of "synchronization data" comprising: an identifier of the producer processing unit data output record entry (time slot) being used by the synchronization unit to track the progress of the generation of the data output; the synchronization strategy determined when the data output was generated (in terms of the number of progress update messages to be provided); and the current progress counter value for the producer processing unit data output record entry (time slot) (as discussed above, the progress counter value for the producer processing unit data output record 501 entry is a cumulative count of the number of progress update messages that have been sent for the data output record entry (time slot)).
[0228] (Thus, for example, if the size of the data output buffer to be produced is 160 scan lines, and the determined synchronization strategy is to provide ten progress update messages when producing this data output, then the producer processing unit will send a progress update message to the synchronization unit 404 every 16 scan lines and if the original synchronization point value (count) at the start of the data output was 6, then the end synchronization point value (count) will be 16.)
[0229] A consumer processing unit that is to use the data output produced by the producer processing unit will then read the synchronization data associated with the buffer that will contain the data output so that it knows how to synchronize with the production of the data output by the producer processing unit.
[0230] In particular, one (and each) consumer processing unit that will use the data output will determine, based on the synchronization data, the identity of the producer processing unit data output record entry (time slot) that is being used to track the progress of the generation of the data output, and will also determine, based on the indicated synchronization strategy and the current progress counter value for the data output, how many progress (synchronization) messages it should receive as the data output is being generated, and what portion of the data output each progress (synchronization) message (and progress count) represents and corresponds to (so that it can synchronize its consumption of the data output with the generation of the data output by the producer processing unit).
[0231] The producer processing unit may then begin producing the data output, and as it does so, it will send progress update messages according to the synchronization strategy determined for the data output to the synchronization unit 404. Thus, for example, in the example discussed above, where there will be four progress update messages while the data output is being produced, the producer processing unit will send a progress update message to the synchronization unit 404 when it has completed 25% of the data output, then send a progress update message to the synchronization unit 404 when it has completed 50% of the data output, and so on.
[0232] In this embodiment, the progress update message sent by the producer processing unit when generating data output has the following form:
[0233] {producer_hw_id,slot_id,counter_value}.
[0234] Wherein: producer_hw_id is the hardware ID of the producer processing unit; slot_id is the identifier of the producer processing unit data output record entry (time slot) being used to track the progress of generation of the data output; and counter_value is the new (updated) progress counter value (of the producer processing unit) for the producer processing unit data output record entry (time slot).
[0235] When the synchronization unit 404 receives a progress update message from a producer processing unit, it first checks the producer hardware identification in the progress update message against the producer hardware identification recorded for the indicated data output record entry (time slot). If the producer processing unit hardware identification matches, the progress counter value for the data output record entry is updated (incremented) to the value in the message from the producer processing unit. If the producer hardware identification does not match, the synchronization unit 404 returns an error message to the producer hardware unit.
[0236] As discussed above, progress update (sync) messages are exchanged through hardware communication between the producer hardware and the sync unit without any software involvement.
[0237] When the progress update message from the producer processing unit is determined to be valid, the synchronization unit 404 broadcasts a progress update (synchronization) message of the following form to all consumer processing units in the consumer processing unit record 502:
[0238] {slot_id,counter_value}.
[0239] Where: slot_id is the identifier of the producer processing unit data output record entry (time slot) to which the message relates; and counter_value is the new (updated) progress counter value for said producer processing unit data output record entry (time slot). Again, this progress update (synchronization) message is broadcast and received via hardware communication between the consumer hardware and the synchronization unit without any software involvement.
[0240] In response to such a broadcast message, each consumer processing unit checks (based on the slot_id included in the progress update message broadcast by the synchronization unit 404) whether it is a message for the data output record entry (time slot) for which the consumer processing unit is waiting, and if so, determines accordingly that more data is available for the data output it is processing.
[0241] More specifically, the consumer processing unit in this embodiment uses the counter_value (i.e., the new (updated) progress counter value for the data output) included in the progress update (synchronization) message sent by the synchronization unit 404 to determine what part (how many parts) of the data output have now been written to the buffer by the producer processing unit (based on the indicated number of synchronization updates (the number of progress update messages) that the producer processing unit will signal to the synchronization unit when generating the data output and the indicated initial progress counter value for the data output).
[0242] The consumer processing unit (hardware) records the progress (counter) value in the progress update message sent by the synchronization unit, and then determines whether the progress (counter) value in the progress update message is large enough to indicate that the data (e.g., rows of data output) that the consumer processing unit wants to read has been produced and stored for the data output before it starts reading data for the data output.
[0243] If the consumer processing unit determines that the data it is waiting for data output is now available in the buffer based on the progress counter value included in the progress update message broadcast by the synchronization unit 404, the consumer processing unit determines that it can and therefore does operate to read the required additional data from the buffer and process it accordingly. The consumer processing unit will then wait for the next progress update message, and so on.
[0244] If the progress update message broadcast by the synchronization unit 404 is not for a data output record entry that the consumer processing unit is waiting for, the consumer processing unit simply ignores the progress update message.
[0245] This operation is repeated until the data output has been completely generated and consumed, and is performed and repeated accordingly for any and all data outputs being generated and consumed in the data processing system.
[0246] Figure 6 and Figure 7 The above operation is illustrated when the synchronization unit 404 is used to synchronize the operations between the producer processing unit that is generating the data output and the consumer processing unit that is consuming the data output in the embodiment of the present invention.
[0247] Figure 6 Shown with Figure 2 The exemplary multimedia subsystem stack is similar to the stack shown, but as Figure 6 As shown, instead of using software fences to synchronize operations between producer and consumer processing units, communications and control of the hardware processing units 31 are synchronized with the aid of a general synchronization hardware unit 404 under the control of its corresponding synchronization unit driver 600.
[0248] Figure 7 The synchronization operation using the general synchronization unit 404 in the present embodiment is schematically illustrated where the graphics processing unit 206 is generating a data output, such as a frame, which the display processor 210 will then compose and provide to the display device 218 for display.
[0249] like Figure 7 As shown, this operation will first include the initialization of the general synchronization unit hardware 404 at system startup (step 70). As part of this operation, the synchronization unit driver 600 will reset each producer processing unit data output record entry (time slot) in the record of producer processing unit data output 501 that it is maintaining. It will also reset the list of consumer processing units 502.
[0250] Thereafter, the driver 35 for the graphics processor 206 will exchange messages with the driver 600 for the synchronization unit 404 in order to "connect" to the synchronization unit 404 (step 71) (as discussed above). Figure 7 As shown, and as discussed above, during this process, the graphics processor 206 will be assigned an entry (time slot) in the record 501 of data outputs being maintained by the synchronization unit 404, and accordingly, the controller 500 of the synchronization unit 404 will record the producer hardware identification for the graphics processor 206 against the data output record (time slot) that has been assigned to the graphics processor 206.
[0251] like Figure 7 As shown, in this embodiment, it is assumed that the graphics processor 206 is also registered as a consumer processing unit to the synchronization unit 404, so the driver 600 for the synchronization unit 404 accordingly adds the consumer processing unit identifier of the graphics processor to the list of consumer processing units 502 it is maintaining.
[0252] Then, the corresponding registration process is performed for the display processor 210 (step 72). Figure 7 As shown, it is again assumed here that the display processor 210 is registered with the synchronization unit 404 both as a producer and as a consumer.
[0253] Once graphics processor 206 has been initialized and connected to synchronization unit 404, driver 35 for graphics processor 206 determines the synchronization strategy (number of update messages) to use for the data output (eg, frames for display) it is generating.
[0254] The driver 35 for the graphics processor 206 will then prepare one or more appropriate buffers 37 for storing the data output to be generated (e.g., frames to be displayed). This will include adding 601 to the buffer description (metadata) the identification of the sync unit data output record entry (time slot) for the data output and the number of progress update messages to be sent when the data output is generated (see Figure 6 ). (Step 73).
[0255] The display processor driver 36 then determines from this metadata that the graphics processor driver 35 has associated the buffer in which the data output (frame) is to be stored, the identifier of the data output record entry (time slot) being used for the data output (buffer), and the synchronization strategy that the graphics processor will use for the data output (step 74).
[0256] The graphics processor hardware 206 may then begin drawing the frame and write the frame to the appropriate buffer 76 in memory (step 75). Figure 7As shown, as part of this process, the graphics processor hardware will periodically send progress update messages 80 to the synchronization unit 404 as it reaches each synchronization update point during the generation of the data output (frame) (as determined by the synchronization strategy that has been determined for that frame).
[0257] In response to each progress synchronization update message 80 from the graphics processor hardware 206, the synchronization unit 404 will update the progress count 505 for the data output entry (time slot) and broadcast a corresponding progress synchronization message (as discussed above) to the consumer processing units (step 77).
[0258] The display processor hardware 210 will accordingly receive the broadcasted synchronization update message from the synchronization unit 404 and identify that it is a synchronization progress update message related to the buffer on which it is waiting, and in response thereto, read the new data from the buffer and, for example, use the data to perform composition and provide the data to the display for display (steps 78 and 79).
[0259] The display processor 210 will then wait for the next synchronization update message from the synchronization unit 404 before processing the next portion of the data output (frame), and so on.
[0260] In this manner, synchronization unit 404 operates to synchronize operations between graphics processor 206 that is generating one or more frames for display and display processor 210 that is processing and then displaying those frames, and enables such operations to be synchronized during the generation of a given frame (rather than, for example, only once the frame has been completed by graphics processor 206).
[0261] This embodiment may be used for synchronization of any suitable and desired operations in the multimedia subsystem 402 .
[0262] An example of such an operation would be for an augmented reality display, where the camera 220 captures images and sends them to the GPU 206 for processing, which then sends the processed images to the display processor 210 for composition and presentation.
[0263] When using the general synchronization unit 404, the driver for the GPU 206 can, for example, set 10 synchronization points for each image. In this case, the latency between the camera 220 and the GPU 206 will be 10% of a frame, and the latency between the GPU 206 and the display processor 210 will also be 10% of a frame. Therefore, the total latency is 20% of a frame. (This is in contrast to more conventional synchronization as follows, where the GPU 206 only starts processing the buffer for the image after the image is completely stored in the memory 216, and the display processor 210 only starts processing the processed image in the buffer after the GPU 206 finishes its processing, so that the total latency is at least 2 frames).
[0264] As will be appreciated from the above, the present invention, at least in its preferred embodiments, provides a synchronization apparatus that can reduce latency between corresponding producer processing units and consumer hardware units in a data processing system. This is achieved, at least in the preferred embodiments of the present invention, by using a general synchronization hardware unit to synchronize the operation of the producer processing unit and the consumer processing unit when generating and consuming data outputs rather than using software fences. By using a hardware synchronization unit to synchronize the operation between the producer unit and the consumer unit, the latency between the producer hardware unit and the consumer hardware unit can be reduced, for example, from one frame to one or more scan lines.
Claims
1. A data processing system, the data processing system include: one or more producer processing units operable to generate data output; one or more consumer processing units operable to use data output produced by the producer processing unit; as well as a memory for storing data output generated by the one or more producer processing units; in: The one or more producer processing units each include processing circuitry operable to generate a data output and store the data output in the memory; and The one or more consumer processing units each include processing circuitry operable to read a data output from the memory and use the read data output; The data processing system further comprises: a synchronization unit operable to communicate with the one or more producer processing units and the one or more consumer processing units, and the synchronization unit operable to: maintaining a record of data output being generated by a producer processing unit of the data processing system; receiving a message from a producer processing unit of the data processing system regarding the progress the producer processing unit has made in generating data output; and sending, in response to a message received from a producer processing unit regarding the progress that the producer processing unit has made in generating the data output, to one or more consumer processing units of the data processing system a message regarding the progress that the producer processing unit has made in generating the data output; in: The one or more producer processing units each further include processing circuitry operable to send a message to the synchronization unit regarding the progress the producer processing unit has made in generating data output; and The one or more consumer processing units each further include processing circuitry operable to receive data output progress messages from the synchronization unit and control reading of data output from the memory in accordance with the data output progress messages received from the synchronization unit.
2. The data processing system according to claim 1, in, A record of data output being generated by a producer processing unit of the data processing system maintained by the synchronization unit includes a set of multiple record entries, each entry having a corresponding entry identifier and associating with the entry an identifier of the producer processing unit to which the entry has been assigned.
3. The data processing system according to claim 2, in, Each entry in the record of data outputs also has associated with that entry a progress indication indicating the progress that the producer processing unit to which the data output record entry has been assigned has made in producing the one or more data outputs.
4. The data processing system according to claim 3, in, The progress indication is in the form of a count of the number of progress update messages that have been sent to the synchronization unit by the producer processing unit to which the data output record entry has been assigned.
5. The data processing system according to any one of claims 1 to 4, in, The synchronization unit is further operable to maintain a record of consumer processing units of the data processing system.
6. The data processing system according to any one of claims 1 to 4, in, The producer processing unit and the consumer processing unit of the data processing system are operable to register with the synchronization unit before their operations are synchronized using the synchronization unit.
7. The data processing system according to any one of claims 1 to 4, in, The progress message sent by the producer processing unit includes at least one of: an indication of the identification of the producer processing unit; an indication of the identification of the producer processing unit data output record entry for the data output to which the progress message relates; and an indication of the progress that the producer processing unit has made in generating the data output to which the progress message relates.
8. The data processing system according to any one of claims 1 to 4, in, The producer processing unit is operable to send a plurality of progress messages to the synchronization unit while generating data output.
9. The data processing system according to claim 8, in, The number of progress messages to be provided for a data output is determined and set by the driver for the producer processing unit that is producing the data output.
10. The data processing system according to any one of claims 1 to 4, in, The synchronization unit is operable to broadcast a message to a plurality of consumer processing units regarding the progress that the producer processing unit has made in generating the data output to which the message relates when the synchronization unit receives a progress message from the producer processing unit.
11. The data processing system according to any one of claims 1 to 4, in: The producer processing unit is operable to associate an identification of a producer processing unit data output record for the data output with the data output being produced by the producer processing unit; and The consumer processing unit is operable to read an identification of a producer processing unit data output record associated with a data output to identify the producer processing unit data output record being maintained by the synchronization unit for the data output.
12. A method of operating a data processing system, the data processing system include: one or more producer processing units operable to generate data output; one or more consumer processing units operable to use data output produced by the producer processing unit; a memory for storing data output generated by the one or more producer processing units; as well as a synchronization unit operable to communicate with the one or more producer processing units and the one or more consumer processing units; The method comprises the following steps: A producer processing unit of the one or more producer processing units generates a data output and stores the data output in the memory; and one or more of the one or more consumer processing units reading the data output from the memory and using the read data output; The method further comprises the following steps: The synchronization unit maintains a record of data output being generated by the producer processing unit; the producer processing unit that is producing the data output sends a message to the synchronization unit regarding the progress that the producer processing unit has made in producing the data output; the synchronization unit receiving from the producer processing unit the message regarding the progress that the producer processing unit has made in generating the data output, and, in response to the message from the producer processing unit regarding the progress that the producer processing unit has made in generating the data output, sending to one or more consumer processing units of the data processing system a message regarding the progress that the producer processing unit has made in generating the data output; and One or more of the consumer processing units receives a data output progress message from the synchronization unit, and controls the one or more consumer processing units to read the data output from the memory according to the data output progress message received from the synchronization unit.
13. The method according to claim 12, in, A record of data output being generated by a producer processing unit of the data processing system maintained by the synchronization unit comprises a set of multiple record entries, each entry having a corresponding entry identifier and associating with the entry an identifier of the producer processing unit to which the entry has been assigned.
14. The method according to claim 13, in, Each entry in the record of data outputs also has associated with that entry a progress indication indicating the progress that the producer processing unit to which the data output record entry has been assigned has made in producing one or more data outputs.
15. The method according to claim 14, in, The progress indication is in the form of a count of the number of progress update messages that have been sent to the synchronization unit by the producer processing unit to which the data output record entry has been assigned.
16. The method according to any one of claims 12 to 15, in, The synchronization unit also maintains a record of consumer processing units of the data processing system.
17. The method according to any one of claims 12 to 15, further comprising: The following steps are involved: The producer processing unit first requests from the synchronization unit an allocation of one or more entries in the producer processing unit data output record maintained by the synchronization unit; as well as In response to such a request, the synchronization unit distributes one or more entries in the producer processing unit data output record maintained by the synchronization unit to the producer processing unit.
18. The method according to any one of claims 12 to 15, in, The progress message sent by the producer processing unit includes at least one of: an indication of the identification of the producer processing unit; an indication of the identification of the producer processing unit data output record entry for the data output to which the progress message relates; and an indication of the progress that the producer processing unit has made in generating the data output to which the progress message relates.
19. The method according to any one of claims 12 to 15, The following steps are involved: The producer processing unit sends a plurality of progress messages to the synchronization unit at intervals when generating the data output.
20. The method according to claim 19, wherein The following steps are involved: A driver for the producer processing unit that is producing the data output determines and sets a number of progress messages to be sent for the data output.
21. The method according to any one of claims 12 to 15, The following steps are involved: When the synchronization unit receives a progress message from the producer processing unit, the synchronization unit broadcasts to multiple consumer processing units a message related to the progress that the producer processing unit has achieved in generating the data output, where the data output refers to the data output related to the progress message.
22. The method according to any one of claims 12 to 15, The following steps are involved: the producer processing unit associating an identification of the producer processing unit data output record for the data output with the data output being produced by the producer processing unit; and One or more of the consumer processing units reads an identification of the producer processing unit data output record associated with the data output to identify the producer processing unit data output record that the synchronization unit is maintaining for the data output.
23. A computer readable medium having stored thereon a computer program comprising computer software codes for executing the method of any one of claims 12 to 22 when the computer program is run on a data processor.
