Method and apparatus for reducing power consumption of a processing integrated circuit

[0002] The present invention relates generally to integrated circuits that perform data processing such as microprocessors, microcontrollers, logic circuitry, etc., and more particularly to reducing power consumption of such integrated circuits.

[0003] Modern day microprocessors, microcontrollers, logic circuitry, etc., (i.e., processing integrated circuits) consume a large portion of their power budget during AC operation (e.g., when circuitry is active and is processing instructions and/or data). While many power management techniques exist that attempt to reduce power consumption in sections of processing integrated circuits that are not in use (e.g., inactive sections of processing integrated circuits), maintaining a power budget for active sections of such circuitry remains difficult.

[0004] Most conventional techniques for budgeting power consumption of active sections of processing integrated circuits rely on reducing logic toggles through the use of multiplexers, or reducing logic toggles performed by control logic. For example, U.S. Pat. No. 6,054,877, which is hereby incorporated by reference herein in its entirety, discloses a “Transition-Once MUX” that may prevent multiplexer (MUX) output transitions for invalid data and reduce AC power consumption due to such “invalid” MUX output transitions. Likewise, U.S. Pat. No. 6,237,101, which is hereby incorporated by reference herein in its entirety, discloses a technique for reducing power consumption of a microprocessor during decoding of a current instruction by maintaining control signals at the values/levels determined during the decoding of a preceding instruction if it is not necessary to change the values/levels of the control signals to execute the current instruction.

[0005] Another conventional technique for controlling power consumption in portable personal computers is disclosed in U.S. Pat. No. 6,167,524, which is hereby incorporated by reference herein in its entirety. U.S. Pat. No. 6,167,524 discloses, for example, summing expected power consumption values of active execution units to determine whether an expected total power consumption exceeds an optimum battery power efficiency value. If so, the amount of execution unit activity is rationed (e.g., so that power drain remains within a regime of optimum battery/converter efficiency). (Col. 6, lines 47-61). The expected power consumption value for each execution unit may be based on an average toggling power associated within the execution unit as computed via simulation. (Col. 5, line 60-col. 6, line 3).

[0006] These and other power management techniques do not take into account the actual data being operated on during a current clock cycle, and/or do not allow for real-time power balancing, particularly once an instruction is dispatched to an execution unit. Accordingly, maintaining a power budget for active sections of processing integrated circuits remains difficult.

[0007] In accordance with a first aspect of the invention, a method is provided for conserving power in a processing integrated circuit. The method includes the steps of (1) calculating power consumption for executing an instruction and data corresponding to the instruction; and (2) executing the instruction if such execution does not exceed a predetermined power level.

[0008] In accordance with a second aspect of the invention, a method is provided for conserving power in a processing integrated circuit employing a plurality of execution units. The method includes the steps of (1) comparing a total power to be consumed by the processing integrated circuit to a power budget for the processing integrated circuit; and (2) if the total power exceeds the power budget, freezing execution of an instruction by one of the plurality of execution units so as to allow execution of the instruction to continue at a later time from where execution was frozen. Numerous other aspects of the invention are provided, as are systems and apparatus.

[0009] Other objects, features and advantages of the present invention will become more fully apparent from the following detailed description, the appended claims and the accompanying drawings.

[0010] FIG. 1 is a schematic diagram of a novel data processing system configured in accordance with the present invention;

[0011] FIG. 2 is a schematic diagram of an exemplary embodiment of the global power controller of FIG. 1 in communication with an exemplary embodiment of the local power controller of FIG. 1;

[0012] FIG. 3 is a flowchart of an exemplary process for conserving power that may be performed by the global power controller of the inventive data processing system of FIGS. 1 and 2;

[0013] FIGS. 4A and 4B are a flowchart of an exemplary process for conserving power that may be performed by the local power controllers of the execution units of the inventive data processing system of FIGS. 1 and 2;

[0014] FIG. 5 is a more detailed, exemplary embodiment of one of the local power controllers of FIGS. 1 and 2;

[0015] FIG. 6 is a more detailed, exemplary embodiment of the global power controller of FIGS. 1 and 2, that may be employed with the local power controller of FIG. 5;

[0016] FIG. 7 is an exemplary embodiment of logic circuitry of one of the execution units of FIG. 1 useful in explaining the operation of the data processing system of FIG. 1; and

[0017] FIG. 8 is a timing diagram useful in explaining the operation of the execution unit of FIG. 7.

[0018] FIG. 1 is a schematic diagram of a novel data processing system 100 configured in accordance with the present invention. The data processing system 100 includes a memory 102 coupled to an instruction dispatch unit 104 and a plurality of execution units 106a-n. The instruction dispatch unit 104 includes a global power controller 108, and each execution unit 106a-nincludes a local power controller 110a-n. The memory 102, the instruction dispatch unit 104, the execution units 106a-nand the global and/or local power controllers 108, 110a-nare configured to exchange various data (e.g., operands), instructions and/or other control signals as described further below.

[0019] The memory 102 and the instruction dispatch unit 104 may be conventional data processing system components as are known in the art. Exemplary embodiments of the global and local power controllers 108, 110a-nare described below with reference to FIGS. 2-6. An exemplary embodiment of one of the execution units 106a-nis described below with reference to FIGS. 7 and 8.

[0020] FIG. 2 is a schematic diagram of an exemplary embodiment of the global power controller 108 in communication with an exemplary embodiment of the local power controller 110aof FIG. 1. The local power controllers 110b-nmay be similarly configured.

[0021] With reference to FIG. 2, the global power controller 108 includes a register 204 adapted to store the total power budget for the data processing system 100. In one embodiment of the invention, the register 204 comprises a programmable register such as a random access memory (RAM), although another conventional register also may be employed. The total power budget for the data processing system 100 may represent, for example, the power required for maximum power efficiency in a battery operated system as described in previously incorporated U.S. Pat. No. 6,167,254, or any other predetermined power level as is known in the art.

[0022] The local power controller 110aincludes a ROM 206, a RAM 208, a first register 210, a second register 212 and adder circuitry 214. The ROM 206 represents one or more ROMs adapted to store, for each executable instruction, (1) average or otherwise representative values of the power required for the execution unit 106a(FIG. 1) to execute the instruction; and/or (2) “typical” or otherwise representative operands for the instruction type (as described further below). Such average power and typical operands may be based on simulation (e.g., based on running application code with a physical capacitance based simulation tool which determines power consumed per cycle for each execution unit as described in previously incorporated U.S. Pat. No. 6,167,524 or using any other known technique).

[0023] The RAM 208 represents one or more RAMs adapted to store, for each instruction executed by the execution unit 106a, (1) the actual power that was required by the execution unit 106ato execute the instruction; and/or (2) the operands on which the instruction operated. The first register 210 is adapted to store the power budget for the execution unit 106a, and may comprise any conventional register such as a RAM. The power budget may be determined using any conventional technique. The second register 212 is adapted to store the power required to finish execution of a previous instruction that was halted or “frozen” as described further below with reference to FIGS. 3-8. The second register 212 similarly may comprise any conventional register, such as a RAM.

[0024] FIG. 3 is a flowchart of an exemplary process 300 for conserving power that may be performed by the global power controller 108 of the inventive data processing system 100 of FIGS. 1 and 2. With reference to FIG. 3, the process 300 starts in step 301. In step 302, the global power controller 108 computes the total power to be dissipated by the data processing system 100 for the current clock cycle (“Total_Chip_Power”). The Total_Chip_Power is computed based on the actual power to be consumed by each execution unit 106a-nduring the current clock cycle (e.g., by summing the actual power to be consumed by each execution unit 106a-nduring the current clock cycle). The actual power to be consumed by each execution unit 106a-nduring the current clock cycle may be determined and provided to the global power controller 108, for example, as described below with reference to FIG. 4.

[0025] In step 303, the global power controller 108 determines whether the Total_Chip_Power will exceed the power budget for the data processing system 100 (“Chip_Power_Budget”). The Chip_Power_Budget may be stored, for example, in the register 204 of FIG. 2 as previously described. If the Total_Chip_Power will exceed the Chip_Power_Budget, in step 304, the global power controller 108 halts or “freezes” one or more of the execution units 106a-n(e.g., to prevent the Total_Chip_Power from actually exceeding the Chip_Power_Budget during the current cycle). As described below with reference to FIG. 4, each execution unit 106a-nis configured so that it may be halted or “frozen” while an instruction is being executed, and later restarted at the same place where execution was halted. In this manner, the power required to finish execution of an instruction by an execution unit that has been frozen may be consumed at a later time when the data processing system 100 is not over budget. For example, if in step 303 it is determined that the Total_Chip_Power will not exceed the Chip_Power_Budget, in step 305 the global power controller 108 may restart one or more previously frozen execution units 106a-n. The data processing system 100 thus may employ dynamic, real-time power management.

[0026] FIGS. 4A and 4B are a flowchart of an exemplary process 400 for conserving power that may be performed by the local power controllers 110a-nof the execution units 106a-nof the inventive data processing system 100 of FIGS. 1 and 2. For convenience, the process 400 is described with reference to the local power controller 110aof FIGS. 1 and 2. It will be understood that one or more of the local power controllers 110b-nmay perform similar processes.

[0027] With reference to FIGS. 4A and 4B, the process 400 starts in step 401. In step 402, the local power controller 110adetermines whether the execution unit 106ahas received a new instruction from the instruction dispatch unit 104 and is decoding (or has decoded) the new instruction. If so, the process 400 proceeds to step 403; otherwise, the local power controller 110are-checks whether the execution unit 106ahas received a new instruction (step 402).

[0028] In step 403, the local power controller 110adetermines whether the new instruction received by the execution unit 106ahas ever been received previously (e.g., based on the operation code or “opcode” for the instruction). If so, the process 400 proceeds to step 404; otherwise the process 400 proceeds to step 405.

[0029] In step 404, the local power controller 110adetermines a predicted power required to execute the new instruction (“Predicted_Power”) based on the last time the instruction was executed by the execution unit 106a. For example, the RAM 208 (FIG. 2) may store the power previously required to execute the instruction. The local power controller 110aalso determines the operands employed last time the instruction was executed (“Last_Operands”). The Last_Operands also may be stored in the RAM 208. The process 400 then proceeds to step 406.

[0030] If in step 403, the new instruction received by the execution unit 106awas not previously received by the execution unit 106a, in step 405 the local power controller 110adetermines the Predicted_Power and Last_Operands based on simulation data (e.g., stored within the ROM 206 of FIG. 2). For example, the Predicted_Power may be an average power from a simulation, and the Last_Operands may be a typical operand employed during the simulation. Any suitable simulation techniques may be employed, such as those described in previously incorporated U.S. Pat. No. 6,167,524. The process 400 then proceeds to step 406.

[0031] In step 406, the local power controller 110acalculates the actual power required to execute the new instruction (“True_Power”) based on the Predicted_Power. In at least one embodiment of the invention, the True_Power is calculated by multiplying the Predicted_Power by a transition factor (“Transition_Factor”) that represents the results of an XNOR or other XOR-like operation performed on the actual operands associated with the new instruction (“New_Operands”) and the Last_Operands as described further below. In step 407, the local power controller 110astores the True_Power and New_Operands in the RAM 208 for use the next time that the instruction is executed by the execution unit 106a.

[0032] In step 408, the local power controller 110acalculates the available power that the execution unit 106amay use during the current clock cycle (“Avail_Power”). In at least one embodiment of the invention, the Avail_Power equals the power budget of the execution unit 106a(“EX_Power_Budget”, stored, for example, in the register 210 of the local power controller 110a) minus any power that is required to finish an instruction from a previous clock cycle that was frozen by the global power controller 108 or by the local power controller 110a(as described further below).

[0033] In step 409, the local power controller 110adetermines whether the True_Power (the power required by the execution unit 106ato execute the instruction) is greater than the Avail_Power (the power available to the execution unit 106a). That is, the local power controller 110adetermines whether the execution unit 106awould be over its power budget if the new instruction were executed. If not, in step 410 the local power controller 110aallows the execution unit 106ato proceed with execution of the instruction; otherwise, in step 411 the local power controller 110aprevents the execution unit 106afrom executing the instruction (as described below). Execution of the instruction may be halted (frozen) before or during execution. In at least one embodiment, the local power controller 110astores the amount of power required to finish execution of the instruction (e.g., in the register 212 of FIG. 2). This stored power amount may be employed during the calculation of a subsequent Avail_Power for the execution unit 106a(step 408).

[0034] FIG. 5 is a more detailed, exemplary embodiment of the local power controller 110aof FIGS. 1 and 2. With reference to FIG. 5, the local power controller 110aincludes (1) a first register 501 coupled to a RAM 503; (2) a first multiplexer 505 coupled to the RAM 503, to a first ROM 507 and to a multiplier 509; (3) an XOR-like logic circuit 511 coupled to the multiplier 509, to a second multiplexer 513 and to a second register 515; and (4) a second ROM 517 and a third register 519 coupled to the second multiplexer 513. The local power controller 110aalso includes (1) a first adder 521 coupled to the multiplier 509, to a comparator 523 and to a fourth register 525; and (2) a second adder 527 coupled to the comparator 523 and to a fifth register 529. The registers 501, 515, 519, 525 and 529 may comprise one or more conventional registers such as one or more RAMs. The XOR-like logic circuit 511 may comprise any conventional comparison circuitry, such as an XNOR logic circuit, as may the comparator 523. The multiplexers 505 and 513, the multiplier 509 and the adders 521 and 527 may comprise any known, suitable logic circuitry.

[0035] In operation, when an instruction is received by the execution unit 106a, the instruction is stored in the first register 501, and the local power controller 110adetermines a Predicted_Power for execution of the instruction by the execution unit 106aas previously described. In the embodiment of FIG. 5, the Predicted_Power is determined in one of two ways: (1) by using the power that was consumed by the execution unit 106athe last time that the instruction was executed; or (2) by using a power determined by simulation (as previously described). Specifically, the local power controller 110adetermines whether the instruction received by the execution unit 110ahas been received by the execution unit 106apreviously. If so, the local power controller 110adetermines a Predicted_Power for the instruction by using the opcode for the instruction as an address for the first RAM 503 (the power consumed by the execution unit the last time that the instruction was executed being stored in the RAM 503 at the address identified by the opcode); otherwise the Predicted_Power is provided from simulated power data stored in the first ROM 507. The first multiplexer 505 is employed to the select and deliver the appropriate Predicted_Power to the multiplier 509.

[0036] Before, during or after the local power controller 110adetermines the Predicted_Power, the local power controller 110aalso determines the Transition_Factor for the operands that accompany the instruction received the execution unit 106a. The Transition_Factor is computed by performing an XOR-like function, such as an XNOR or other comparison operation, between the operands that accompany the instruction and either the operands employed the last time that the instruction was executed (e.g., Last_Operands stored in the register 519) or typical operands employed during simulation (e.g., Last_Operands stored in the ROM 517). The second multiplexer 513 is employed to select and deliver the appropriate Last_Operands to the XOR-like circuit 511. The XOR-like circuit 511 then outputs the result of its XOR-like operation (the Transition_Factor) to the multiplier 509.

[0037] The mutiplier 509 calculates the actual power required to execute the instruction (“True_Power”) based on the Predicted_Power by multiplying the Predicted_Power by the Transition_Factor. The local power controller 110astores the True_Power in the RAM 503 and the operands for the current instruction in the register 519 for use the next time that the instruction is executed by the execution unit 106a. The multiplier 509 outputs the True_Power to the first adder 521, to the global power controller 108, to the comparator 523 and to the RAM 503.

[0038] The local power controller 110aalso calculates the available power that the execution unit 106amay use during the current clock cycle (Avail_Power) by subtracting any power that is required to finish an instruction from a previous clock cycle that was halted or frozen by the global power controller 108 or by the local power controller 110a(stored in the register 525) from the power budget of the execution unit 106a(stored in the register 529). The subtraction operation is performed via the adder 527, and the resulting Avail_Power is provided to the comparator 523 by the adder 527. If the comparator 523 determines that the True_Power (the power required to perform the current instruction) exceeds the Avail_Power (the power budget of the execution unit, corrected by the power required to complete any instruction execution that was frozen), then the local power controller 110afreezes operation of the execution unit 106a. The local power controller 110amay notify the global power controller 108, and the global power controller 108 may re-assign the instruction to another execution unit. Alternatively, the local power controller 108 may store the amount of power required to finish execution of the instruction within the register 525. If the True_Power does not exceed the Avail_Power, the local power controller 110aallows the execution unit 106ato perform the instruction (and notifies the global power controller 108 of the same). One embodiment for freezing execution of an instruction by an execution unit is described below with reference to FIG. 7.

[0039] The adder 521 determines the actual power that will be dissipated by the execution of the instruction by the execution unit 106a(“Actual Power Dissipated”) by adding the True_Power to the power needed to finish any frozen instruction (stored in register 525). The Actual Power Dissipated is provided to the global power controller 108 as described below with reference to FIG. 6. The local power controllers 110b-nmay be configured similarly to the local power controller 110aof FIG. 5. The control logic required to implement the above functions may be developed by a person of ordinary skill in the relevant art, and is not described in detail herein.

[0040] FIG. 6 is a more detailed, exemplary embodiment of the global power controller 108 of FIGS. 1 and 2, that may be employed with the local power controller 110aof FIG. 5. With reference to FIG. 6, the global power controller 108 includes a first register 601 coupled to a first adder 603 and to a comparator 605. The comparator 605 also is coupled to a second adder 607 and to a selection circuit 609. The register 601, the adders 603 and 607, the comparator 605 and the selection circuit 609 may comprise any known, suitable logic circuitry.

[0041] In operation, the global power controller 108 computes the total power to be dissipated by the data processing system 100 for the current clock cycle by employing the second adder 607 to add together the “Actual Power Dissipated” information received from each local power controller 110a-n(e.g., the power that will be dissipated by each execution unit 106a-nas previously described). This total power represents the Total_Chip_Power for the data processing system 100. The comparator 605 then compares the power budget for the data processing system 100 (Chip_Power_Budget, stored in the register 601) to the Total_Chip_Power. If the Total_Chip_Power is less than the Chip_Power_Budget, the global power controller 108 allows each execution unit 106a-nto execute instructions without interference from the global power controller 108. Any “surplus” power (e.g., the amount by which the Chip_Power_Budget exceeds the Total_Chip_Power) is added to the Chip_Power_Budget via the adder 603 (e.g., increasing the power budget for the data processing system 100 during the next clock cycle). Rather than adding the surplus power to the Chip_Power_Budget, the global power controller 108 may restart one or more of the execution units 106a-nthat have been previously frozen.

[0042] If the Total_Chip_Power is greater than the Chip_Power_Budget (as determined by the comparator 605), then the global power controller 108 (via selection circuit 609) selects one or more execution units 106a-nand freezes execution of instructions by the selected execution units. The selection of the particular execution unit or units to freeze may be based on any relevant factor. In at least one embodiment, the selection circuit 609 selects the appropriate combination of execution units to freeze based on the power to be consumed by each execution unit 106a-nduring the current cycle (e.g., based on the True_Power of each execution unit 106a-n). Other factors include the number of execution units to be frozen, the type of instruction being executed by an execution unit, etc. For example, the global power controller 108 may choose to freeze only speculative execution units (described in previously incorporated U.S. Pat. No. 6,167,524). The control logic required to implement the above functions may be developed by a person of ordinary skill in the relevant art, and is not described in detail herein.

[0043] FIG. 7 is an exemplary embodiment of logic circuitry of the execution unit 106aof FIG. 1 useful in explaining the operation of the data processing system 100 of FIG. 1. It will be understood that other logic circuitry may be employed by any of the execution units 106a-n, and that the logic circuitry of FIG. 7 is merely exemplary.

[0044] With reference to FIG. 7, the execution unit 106aincludes four logic units 702, 704, 706 and 708. Each logic unit may comprise any conventional logic circuitry adapted to process the instructions and operands employed by the data processing system 100. In the embodiment of FIG. 7, the first logic unit 702 consumes about 10% of the total power of the execution unit 106a, the second logic unit 704 consumes about 40% of the total power of the execution unit 106a, the third logic unit 706 consumes about 30% of the total power of the execution unit 106aand the fourth logic unit 708 consumes about 20% of the total power of the execution unit 106a. These percentages are merely exemplary. The local power controller 110amay be pre-programmed with a knowledge of these percentages. Such information may be stored, for example, in a ROM.

[0045] Each logic unit 702-708 is preceded by a “transition-once” buffer or multiplexer (“transition-once logic”) 710, 712, 714 and 716, respectively. Transition-once multiplexers are described in detail in previously incorporated U.S. Pat. No. 6,054,877. Transition-once buffers may be similarly configured (e.g., without the additional logic required for selecting between multiple inputs). Assuming that the transition-once logic 710, 712, 714 and 716 each comprise a transition-once buffer, each transition-once buffer 710-714 is provided with a slow select input and a pass gate/latch (not shown) that holds its value when not selected by the slow select input (e.g., when the slow select input is “inactive”).

[0046] In operation, in response to a “freeze” signal generated by either the local power controller 110a(e.g., in step 411 of process 400 of FIG. 4) or by the global power controller 108 (e.g., in step 304 of process 300 of FIG. 3), the local power controller 110adisables one or more of the slow select inputs of the transition-once logic 710, 712, 714 and 716. For example, if the local power controller 110ais to prevent the execution unit 106afrom consuming any power, the slow select inputs of all transition-once logic 710-716 may be disabled. In this manner, any instruction and operands provided to the execution unit 106aare “held” by the transition-once logic 710. The instruction and operands may be allowed to propagate through the execution unit 106aat any subsequent time by enabling the slow select inputs of the transition-once logic 710-716. The global power controller 108 may be provided with direct control over the slow select inputs of the transaction-once logic 710-716.

[0047] As another example, assume that an instruction and operands have been allowed to propagate through both the logic unit 702 and the logic unit 704 through appropriate selection of the slow select inputs of the transition-once logic 710 and 712. If the local power controller 110athen freezes operation of the execution unit 106a(e.g., in response to a freeze signal from either the local or global power controller 110a, 108), no “active” slow select input is provided to the transition-once logic 714 and 716. This is illustrated, for example, in the timing diagram of FIG. 8 which shows a reference clock signal (CLK), the slow select input applied to the transition-once logic 710 (slow_select A), the slow select input applied to the transition-once logic 712 (slow_select B), the slow select input applied to the transition-once logic 714 (slow_select C) and the slow select input applied to the transition-once logic 716 (slow_select D). With the slow select inputs to the transition-once logic 714 and 716 held inactive, the instruction and operands applied to the execution unit 106apropagate only through the logic units 702 and 704. The resultant output of the logic unit 704 is held by the transition-once logic 714. The processing of the instruction and operands may be continued at a subsequent time by proper activation of the slow select inputs of the transition-once logic 714 and 716. Note that execution of the instruction and operands may start where execution was frozen, without requiring reprocessing through logic units 702 and 704. A significant power savings thereby may be realized.

[0048] In the example above, approximately 50% of the power required for the execution unit 106ato process the instruction and operands has been consumed. In at least one embodiment of the invention, this information is stored in the register 525 of FIG. 5. For example, assume that the total power that the execution unit 106awill consume executing an instruction is 0.2 Watts. If only the slow select inputs of the transition-once logic 710, 712 are active, the local power controller 110amay determine that about 0.1 Watts of power has been consumed by the execution unit 106a, and that an additional 0.1 Watts of power will be required for the execution unit 106ato complete its processing of the instruction.

[0049] It will be understood that through use of the present invention, the power dissipation of the data processing system 100 may be regulated very quickly and accurately as power consumption by instructions and operands may be monitored at every clock cycle and adjusted as desired (e.g., to maintain a steady current drain of a power supply). For example, in at least one embodiment of the invention, the global power controller 108 calculates the total power dissipated by all of the execution units 106a-n(e.g., as provided by the local power controllers 110a-n) and attempts to maintain the total power constant. This may be achieved, for example, by (1) temporarily suspending the dispatching of new instructions to one or more of the execution units 106a-n; and/or (2) temporarily freezing execution of instructions by one or more of the execution units 106a-n(e.g., step 304 in process 300 of FIG. 3). When execution of a halted/frozen instruction resumes, the execution resumes from where execution was halted (e.g., as described previously with reference to FIGS. 7 and 8). Instruction execution thus may be “paused” and restarted at nearly any point during execution.

[0050] As stated above, each local power controller 110a-nmay “predict” the per-cycle power that will be dissipated by each instruction and operands, and may compare the per-cycle power to a power budget for the relevant execution unit 106a-n. In at least one embodiment of the invention, if the power budget for an execution unit will be exceeded by execution of a particular instruction, the relevant local power controller 110a-nremoves the instruction from the execution unit's pipeline. Execution of instructions by the execution unit may be halted (e.g., by inactivating slow select inputs of the execution unit), and the global power controller 108 may be notified that the execution unit cannot execute the instruction in this cycle. The global power controller 108 may reassign the instruction to another execution unit, and may restart the halted execution unit for processing of a subsequent instruction. In one embodiment of the invention, all calculations and/or data manipulations (e.g., calculations/determinations of Predicted_Power, Last_Operands, True_Power, Avail_Power, the various comparisons therebetween, etc.) occur in a single clock cycle so that a decision may be made quickly as to whether to allow an instruction to be executed. In yet another embodiment, each local power controller 110a-nmay only freeze execution of an instruction by an execution unit at the beginning of a clock cycle, while the global power controller 108 may freeze execution of the instruction mid-cycle (e.g., if the global power controller 108 determines that the power budget of the data processing system 100 will be exceeded if execution of the instruction is allowed to continue). Dynamic, real-time power control is provided to the data processing system 100.

[0051] The foregoing description discloses only exemplary embodiments of the invention; modifications of the above disclosed apparatus and method which fall within the scope of the invention will be readily apparent to those of ordinary skill in the art. For example, simulated or actual power consumption data at multiples stages within an execution unit may be stored by the local power controllers 110a-nif desired.

[0052] Accordingly, while the present invention has been disclosed in connection with exemplary embodiments thereof, it should be understood that other embodiments may fall within the spirit and scope of the invention, as defined by the following claims.