Electrical computers and digital data processing systems: input/ – Interrupt processing – Interrupt inhibiting or masking
Reexamination Certificate
2000-06-15
2003-09-30
Thai, Xuan M. (Department: 2181)
Electrical computers and digital data processing systems: input/
Interrupt processing
Interrupt inhibiting or masking
C710S049000
Reexamination Certificate
active
06629180
ABSTRACT:
The invention relates to a method of managing interrupts coming from peripherals associated with a digital signal processor (DSP) that uses a cache memory when executing real-time tasks.
A particular application of the invention lies in voice processing for telephone transmission, for example, such as that used by the global system for mobile communication (GSM), but also in processing and transmitting images in real time.
BACKGROUND OF THE INVENTION
Certain definitions are given below.
A digital signal processor (DSP) is a powerful computational unit suitable for performing tasks in real time, i.e. within a certain fixed time interval. Such a DSP is thus used to compress and decompress voice during a GSM telephone call, for example, or to perform speech recognition, or to process any digital signal such as an image, or some other signal.
Real-time tasks are generally subdivided into samples. For example, in the GSM application, voice is decompressed on the basis of samples taken over 20 ms for transmission in packets.
A DSP advantageously makes use of a cache memory.
Such a memory is configured to copy automatically the memory which is being read by the processor. It has the characteristic of being small but fast.
When the processor performs a loop in its program, and requests the same memory location on several occasions, the cache memory responds more quickly than the external main memory and gives the processor a copy of the previously used data.
The cache memory operates automatically and thus gives the impression of the external main memory being very fast.
This memory is transparent to the processor and hides the external main memory but not the inputs/outputs. Thus, when an interrupt appears, the cache memory takes account thereof.
An interrupt serves to interrupt the normal flow of the program to indicate that urgent processing needs to be performed. The arrival of an interrupt diverts the processor to an interrupt handler which saves the execution context of the current program and processes the interrupt.
Like the main program, interrupt execution also makes use of the cache memory. The cache memory is thus partially or completely emptied of the current data and program code so as to be filled with data and code for the interrupt to be processed.
When the DSP returns to the program for executing the real-time task, the cache memory consequently needs to be refilled with program data and code.
Such reloading is necessarily expensive in time. Unfortunately, time is particularly precious in a real-time application.
Until now, DSPs have used their own memory to maintain their real-time constraints. In addition, they have been subjected to few interrupts, essentially only to an interrupt indicating that the next voice sample is ready.
Nowadays, the tasks conferred on DSPs are being developed as options are added to main apparatuses.
Thus, ever more varied tasks such as hands-free operation, speech recognition, graphics processing, etc., are being added to the main task of voice processing.
Since the DSP is a powerful computational machine, it can also be used for graphics processing, such as displaying an image from the Internet, for example.
Some of these functions, such as image display or graphics functions require the DSP to access the external main memory. Furthermore, the code executed by the DSP can advantageously be stored in the external main memory, thereby enabling said code to be updated more easily. It is therefore essential to use the cache memory which is loaded with the code and data accessed by the DSP in the main memory to enable processing to be performed more quickly.
In addition to real-time processing, such as voice compression, and processing which is not real time, such as graphics functions, the DSP must also perform processing that corresponds to interrupts. For example, with Internet telephony, the DSP must simultaneously process voice compression and decompression and also the interrupts that come from the serial line that delivers a continuous stream of bytes from the modem.
On each interrupt, the code and the data of the interrupt replace some or all of the code or data associated with the real-time processing in the cache memory. Thus, after each interrupt, the DSP must reload its cache memory with the code and data relating to the program for executing the real-time task. This necessarily gives rise to significant slowing down, particularly when the number of interrupts is high, and real-time processing is then no longer guaranteed.
Several solutions have already been proposed in the prior art for guaranteeing real-time processing of the main task.
A first solution consists in masking interrupts throughout the time the main task is being executed by the DSP.
Such a solution requires peripherals to be dimensioned so that they have local memory capable of storing all the data relating to interrupts throughout the masking time. For example, a communications peripheral must store in local memory the data emitted or received during the interrupt masking time.
If the real-time task that the DSP is to perform is known, for example voice compression, it is possible to determine in advance the time required to execute this task and thus to determine the maximum interrupt masking time. It is then possible to dimension the peripherals so that they can accommodate response latency corresponding to the masking time. Nevertheless, when multimedia tasks are performed, it becomes necessary to dimension the peripherals for the largest conceivable amount of multimedia processing, so the resulting systems are expensive.
If it is decided not to store data in peripherals during the interrupt masking time, then hardware mechanisms must be provided that give direct access to the main memory so as to enable the data streams relating to the peripherals to be stored in buffer memories or directly in main memories. Such a solution can be envisaged, but it makes the system complex and expensive, having peripherals that are more powerful than is necessary.
A second solution proposed in the prior art consists in locking the cache memory so as to prevent it losing its contents.
Thus, after the first trip round the main program loop of the DSP, the cache memory is locked. When interrupts occur, they are processed without using the cache memory.
Unfortunately, that solution presents numerous drawbacks.
Firstly, in the event of the DSP processing an algorithm that contains conditional tests subdividing the program into a plurality of processing branches it can be difficult to load the main program into the cache memory before it becomes locked; in addition, locking the cache memory amounts to transforming it into a local memory that is loaded manually. This loses the advantages associated with a cache memory.
Secondly, interrupt processing is slowed down by being prevented from using the cache memory.
Furthermore, in the context of a wireless application (such as GSM), access to the external memory is expensive in terms of energy and thus has an impact on the battery charge life of the apparatus.
That solution of locking the cache memory during processing of the real-time task is therefore not optimal.
A third solution consists in using a memory management unit (MMU).
The MMU specifies which data can be cached, such as memory data, and which data is unsuitable for caching, such as inputs/outputs, and which data can be modified by a microcontroller. The MMU can require the code and data associated with interrupts to be non-cachable. Thus, during execution of an interrupt, the associated data is considered as being temporary and is not put into the cache memory.
That solution suffers from a major drawback because interrupts do not benefit from the advantages of a cache memory, and that can cause processing to be slowed down considerably, with the direct consequence of increasing the overall time required for the DSP to process the main task.
Another method proposed in the prior art consists in overdimensioning the system so that the DSP can maintain real time even if the ca
Attimont Luc
Bodin Jannick
Alcatel
Sughrue & Mion, PLLC
Thai Xuan M.
LandOfFree
Method of performing a task in real time by a digital signal... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of performing a task in real time by a digital signal..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of performing a task in real time by a digital signal... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3092719