|Publication number||US20060288244 A1|
|Application number||US 11/511,911|
|Publication date||Dec 21, 2006|
|Filing date||Aug 28, 2006|
|Priority date||Jun 26, 2002|
|Also published as||US7100060, US20040003309|
|Publication number||11511911, 511911, US 2006/0288244 A1, US 2006/288244 A1, US 20060288244 A1, US 20060288244A1, US 2006288244 A1, US 2006288244A1, US-A1-20060288244, US-A1-2006288244, US2006/0288244A1, US2006/288244A1, US20060288244 A1, US20060288244A1, US2006288244 A1, US2006288244A1|
|Inventors||Zhong-ning Cai, Chee Lim|
|Original Assignee||Cai Zhong-Ning, Lim Chee H|
|Export Citation||BiBTeX, EndNote, RefMan|
|Referenced by (4), Classifications (36)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application is a continuation of prior application Ser. No. 10/184,557, filed Jun. 26, 2002.
The present disclosure pertains to the field of processing systems, and particularly the use of a secondary processing resource to execute instructions under some conditions.
2. Description of Related Art
Several techniques are presently used to control temperature and power consumption of electronic components such as processors. Typically, maintaining a temperature of a component at an acceptable level is important to avoid damaging the component as well as to ensure safety. One way to control or reduce temperature is to limit power consumption. Additionally, with the popularity of mobile computing and communications devices, limiting power consumption to preserve battery life is an increasingly important goal as well. Thus, power conservation may be advantageous to limit heat generation, to preserve battery power, or both.
Dynamic clock disabling is one prior art technique to reduce power consumption. Dynamic clock disabling is the temporary or intermittent stopping of the clock or clocks of a processor. During the period in which the clock is stopped, clearly less power is consumed; however, no work can be performed if all of the clocks are stopped. In some cases, a reduced level of functionality may be provided by periodically stopping clocks; however, during “on” periods large and power-hungry high performance logic structures are used. Another variation is to recognize when there is no work to be done, and to stop the clocks at that point in time. Another variation is to stop clocks to a particular functional unit (e.g., a floating point unit) when that unit is idle. However, when a unit or processor is idled, no work is accomplished.
Dynamic frequency scaling is the change of processing frequency, typically effectuated by altering a clock frequency of the processor. While reduction of operating frequency decreases power proportionately, dynamic frequency scaling may in some cases require that a phase locked loop re-acquire lock, which can be a relatively time consuming proposition. Moreover, dynamic frequency scaling also still keeps large power-hungry structures active.
The present invention is illustrated by way of example and not limitation in the Figures of the accompanying drawings.
The following description discloses a processing system having asymmetric secondary processing resources. In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. It will be appreciated, however, by one skilled in the art that the invention may be practiced without such specific details. In other instances, control structures and gate level circuits have not been shown in detail in order not to obscure the invention. Those of ordinary skill in the art, with the included descriptions, will be able to implement appropriate logic circuits without undue experimentation.
Distribution of processing activity across asymmetric resources may be advantageous in a number of ways. Some embodiments may allow new low power but ‘awake’ modes that might not otherwise be possible. Some embodiments provide a secondary processing resource without the large area increase associated with fully duplicating resources. Moreover, some embodiments can provide a processor that operates to reduce power consumption or reduce temperature without requiring other system hardware or software changes, although hardware and/or software changes may prove advantageous in some embodiments.
For example, in the embodiment of
The secondary resource 120 also performs the function f(x); however, the secondary resource is asymmetric with respect to the primary resource in terms of its throughput, size, and power consumption. The secondary resource may be asymmetric with respect to any one or more of these characteristics, but they typically are correlated and all change together. Because the secondary resource need not achieve as high of throughput as the primary resource 110, the secondary resource 120 is typically smaller in size and consumes less power than the primary resource 110. Thus, as indicated in
While the primary and secondary resources of
The embodiment of
Various permutations of reduced power modes and processing resources may be used. For example, additional increments of processing resources may be provided between the highest and lowest power resources, with different combinations of resources being active in order to maintain some reasonable throughput on a reasonable power budget under the circumstances. For example, N different copies of a particular functional unit or function may be provided, each having a different power consumption and a different throughput. A power aware scheduler may schedule instruction dispatch not only seeking to minimize throughput, but also to stay within a certain total power consumption, or based on a current die temperature. Thus, various ones of the N resources may be selected for execution of a particular function at a particular time based on the power allotted or the current thermal environment.
In one embodiment, the processor 100 is a thermally aware microprocessor. The thermally aware microprocessor may scale back its power consumption yet still continue processing at a steady pace by switching to its lower throughput set of resources. In some embodiments, the processor switches to its lower throughput resources with little disruption by draining instructions utilizing the high performance and high power resources, and then initiating execution using the secondary resource. Such switching may be performed without stopping a clock or re-synchronizing a phase locked loop in some cases. Moreover, by using such hardware mechanisms, the thermally aware microprocessor may achieve power conservation and/or cooling without external intervention and without any software support in some embodiments. Furthermore, with multiple resources spreading out heat generation, less expensive heat dissipation technologies may be used to cool the processor 100 in some cases.
A prior art processor may be more prone to overheating since a single set of resources may be continuously active. As a result of an overheating event, a disruptive cooling period (e.g., stopping or reducing frequency of clocks) may be triggered according to prior art techniques, causing the processor to operate very slowly and/or need to re-lock phase locked loops. Thus, an overheating period may degrade performance significantly. In a thermally aware processor, the addition of secondary processing resources may consume a small amount of processor area (e.g., area on a silicon die in some embodiments); however, the additional area may be performance justified by allowing the processor to operate more coolly and avoid disruptive cooling periods. For example, in one embodiment, five percent of die area may be dedicated to providing functionally duplicative structures that allow a lower yet continuous execution rate. Clocks may be gated to the large functionally duplicated structures, thereby eliminating the need to change frequency and re-synchronize a phase locked loop. Thus, overall the obtained energy to performance ratio may readily justify the expense in consumed die area in some cases.
Moreover, transistor reliability decreases and leakage current increases as temperature rises. These generally negative effects may be mitigated if lower processor temperatures are obtained by spreading processing to the secondary processing resources to reduce heat concentration. Since leakage current becomes a larger concern as device geometries shrink, keeping device temperature low should continue to be an important factor in reducing leakage current. Therefore, it may be desirable to locate the functionally duplicative processing resources far away from the resources which they are functionally duplicating to maximize thermal de-coupling. The desire to thermally de-couple these resources, however, typically will need to be tempered by the long signal lines (and hence delay) that might be introduced in embodiments in which the functionally duplicated resources interact substantially with a set of shared resources.
While a general purpose microprocessor may be one beneficiary of the use of asymmetric secondary resources to perform processing in some cases, other types of devices may benefit as well. For example, the processor 100 may be any type of processor such as a graphics processor, a network processor, a communications processor, a system-on-a-chip processor, an embedded processor, a digital signal processor or any other known or otherwise available component which performs processing. Moreover, other electronic components that generate heat and are capable of operating at different throughput and power levels may likewise benefit from using a secondary asymmetric resource at some times.
As a more specific example,
In the embodiment of
Under normal operation (e.g., the die temperature is below a given threshold), the wide superscalar pipelines provide the instructions for the execution units, just like a typical high-performance microprocessor. The Operating System (OS) may schedule some “easy” tasks to the secondary pipeline as well in some embodiments. Once the temperature exceeds a selected threshold, or if the processor is switched to a low power mobile mode, the cooler secondary pipeline is used exclusively, and the primary pipeline is disabled (e.g., by clock gating).
If either one of these conditions is true (or if any condition is true which justifies switching to only the in-order pipeline), then a fetch stall is generated to stall fetching of instructions as indicated in block 315. If instructions remain in the out-of-order pipeline (as determined in block 320), the instructions are executed and retired as indicated in block 330. The fetch unit may remain stalled until all the remaining instructions in the out-of-order pipeline are executed and retired. Once this process is completed, the fetch unit 205 is restarted, and instructions are fetched to the in-order pipeline as indicated in block 325.
In some embodiments, the fetch unit 205 may also be a great consumer of power when operating at full speed. In one embodiment, the fetch unit 205 may be split into primary and secondary resources to alleviate the high power consumption issue by utilizing the secondary resource at times. In another embodiment, the fetch unit 205 may be operated in a low power mode when the overall low power mode that directs execution to the secondary pipeline is entered. For example, the clocks to the fetch unit 205 may be gated a portion of the time or periodically.
In some embodiments, both cores may be fully compatible (i.e., fully decode and execute an entire instruction set). In other embodiments, the secondary core 520 may only be capable of processing a subset of the entire instruction set. In such a case, a programmer or compiler may be responsible for ensuring that tasks to be executed by the secondary core 520 do not include unsupported instructions. In one embodiment, the secondary core may not include support for instructions such as floating point or single instruction multiple data (SIMD) instructions. The switching logic 530 may detect such instructions and force a switch back to the main core 510 if they occur. However, it may be possible to run various minimal connectivity programs (e.g., email or other messaging programs) without using some complex or compute-intensive instructions. In such cases, it may be advantageous to have a core that executes only a subset of an instruction set and to carefully architect routines to run on that core in a low power consumption mode.
In one embodiment, the main memory 695 of
In one embodiment, the main memory 695 of
A typical hardware design may go through various stages, from creation to simulation to fabrication. Data representing a design may represent the design in a number of manners. First, as is useful in simulations, the hardware may be represented using a hardware description language or another functional description language Additionally, a circuit level model with logic and/or transistor gates may be produced at some stages of the design process. Furthermore, most designs, at some stage, reach a level of data representing the physical placement of various devices in the hardware model. In the case where conventional semiconductor fabrication techniques are used, the data representing the hardware model may be the data specifying the presence or absence of various features on different mask layers for masks used to produce the integrated circuit. In any representation of the design, the data may be stored in any form of a machine readable medium. In a software design, the design typically remains on a machine readable medium. An optical or electrical wave modulated or otherwise generated to transmit such information, a memory, or a magnetic or optical storage such as a disc may be the machine readable medium. Any of these mediums may “carry” the design information. A processing device utilizing disclosed techniques may be represented in these or other manners and accordingly may be carried in various media.
Thus, a processing system having asymmetric secondary processing resources is disclosed. While certain exemplary embodiments have been described and shown in the accompanying drawings, it is to be understood that such embodiments are merely illustrative of and not restrictive on the broad invention, and that this invention not be limited to the specific constructions and arrangements shown and described, since various other modifications may occur to those ordinarily skilled in the art upon studying this disclosure.
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US8073669 *||Aug 21, 2007||Dec 6, 2011||International Business Machines Corporation||Method and apparatus for detecting clock gating opportunities in a pipelined electronic circuit design|
|US8244515 *||Dec 31, 2008||Aug 14, 2012||International Business Machines Corporation||Structure for detecting clock gating opportunities in a pipelined electronic circuit design|
|US9069553 *||Sep 5, 2012||Jun 30, 2015||Marvell World Trade Ltd.||Switching tasks between heterogeneous cores|
|US20130061237 *||Mar 7, 2013||Ofer Zaarur||Switching Tasks Between Heterogeneous Cores|
|U.S. Classification||713/300, 712/E09.069, 712/E09.063, 712/E09.071|
|International Classification||G06F1/20, G06F1/00, G06F1/32|
|Cooperative Classification||G06F1/206, Y02B60/1275, G06F9/4893, G06F1/3203, Y02B60/162, G06F9/3828, G06F9/3877, G06F9/3885, G06F9/4856, G06F9/3802, G06F1/329, G06F1/3293, G06F1/3287, G06F9/3869, Y02B60/144, Y02B60/121, Y02B60/1282|
|European Classification||G06F9/38D1B, G06F9/38B, G06F9/48C4S2, G06F9/38S, G06F1/32P5S, G06F1/32P5T, G06F1/32P5U, G06F1/32P, G06F1/20T, G06F9/38T, G06F9/38P2, G06F9/48C4P2|