# US 2012/0260057 A1 · Counter architecture for online dvfs profitability estimation

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2012/0260057 A1
- Document ID: US20120260057A1
- Status: Published application
- Kind: A1
- Application number: 13/516,850
- Filed: 2010-12-10
- Published: 2012-10-11
- Earliest claimed: 2010-05-26
- Assignees: Universiteit Gent
- Inventors: Stijn Eyerman; Lieven Eeckhout
- Web page: https://patentgrep.com/patent/US20120260057A1

## CPC

- G06F 1/3203: Power management, i.e. event-based initiation of a power-saving mode
- G06F 1/324: by lowering clock frequency
- G06F 1/3296: by lowering the supply or operating voltage
- Y02D 10/00: Energy efficient computing, e.g. low power processors, power management or thermal management
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- Y02D: CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE

## Abstract

A counter architecture and a corresponding method are provided for estimating a profitability value of DVFS for a unit of work running on a computing device. The counter architecture and the corresponding method are arranged for dividing total execution time for executing a unit of work on the computing device into a pipelined fraction subject to clock frequency and a non-pipelined fraction due to off-chip memory accesses, and for estimating the DVFS profitability value from the pipelined and the non-pipelined fraction.

## Claims

### Claim at position 1

-13. (canceled)

### Claim 14

A counter architecture for estimating a profitability value of DVFS for a unit of work running on a computing device, the counter architecture comprising:
a computing unit configured to determine from a total execution time for executing the unit of work on the computing device at arbitrary frequency a pipelined fraction and a non-pipelined fraction due to off-chip memory accesses, and
a decision unit that estimates the DVFS profitability value from these two fractions.

### Claim 15 (depends on 14)

The counter architecture according to claim 14, wherein the computing unit is arranged to count the number of cycles of the non-pipelined fraction due to off-chip memory accesses, and to determine the number of cycles of the pipelined fraction from the total execution time and the counted number of cycles of the non-pipelined fraction.

### Claim 16 (depends on 14)

The counter architecture according to claim 14, comprising:
at least one counter that counts the number of off-chip cycles of a unit of work running on a computing device;
a first memory for storing an indication of whether an off-chip memory access is active; and
a second memory configured to keep track of which outstanding off-chip memory access initiated a burst of off-chip memory accesses.

### Claim 17 (depends on 14)

The counter architecture according to claim 14, wherein the computing unit is arranged to count the number of cycles of the pipelined fraction, and to determine the number of cycles of the non-pipelined fraction from the total execution time and the counted number of cycles of the pipelined fraction.

### Claim 18 (depends on 17)

The counter architecture according to claim 17, wherein the computing device comprises a processor and wherein the counter architecture furthermore comprises a counter that determines the pipelined fraction, the counter being arranged to be incremented every processor clock cycle of execution of the unit of work on the processor, except when an instruction cache and/or instruction TLB miss accesses off-chip memory, and when a load/store cache or data TLB miss or I/O operation accesses off-chip memory if this load/store miss or TLB miss or I/O operation is the first in a burst of off-chip load/store cache and/or TLB misses and/or I/O operation.

### Claim 19 (depends on 14)

The counter architecture according to claim 14, wherein the computing device is a processor and wherein the counter architecture comprises at least one of:
a first counter that counts the number of processor clock cycles that an instruction cache miss and/or instruction Translation Lookaside Buffer (TLB) miss accesses off-chip memory; and
a second counter that counts the number of processor clock cycles that a load/store cache and/or TLB miss and/or I/O operation accesses off-chip memory if this memory access is the first in a burst of off-chip load/store cache misses and/or TLB misses and/or I/O operations.

### Claim 20 (depends on 18)

The counter architecture according to claim 18, furthermore comprising
a first memory that stores an indication denoting whether the second counter is actively counting, and
a second memory that stores an indication denoting an entry for the first pending memory access in a burst of memory accesses.

### Claim 21

A method for estimating a DVFS profitability for a unit of work running on a computing device, the method comprising:
dividing total execution time for executing the unit of work on the computing device into a pipelined fraction subject to clock frequency and a non-pipelined fraction due to off-chip memory accesses, and
estimating a DVFS profitability value based on the pipelined fraction and the non-pipelined fraction.

### Claim 22 (depends on 21)

The method according to claim 21 for application in a processor, the method furthermore comprising determining the non-pipelined fraction by counting the number of cycles that an instruction cache miss and/or an instruction TLB miss accesses off-chip memory, and the number of cycles that a load/store cache miss and/or data TLB miss and/or I/O operation accesses off-chip memory, if this load/store miss or data TLB miss or I/O operation is the first in a burst of off-chip load/store cache and/or TLB misses and/or I/O operations.

### Claim 23 (depends on 21)

The method according to claim 21, comprising determining the pipelined fraction by subtracting the non-pipelined fraction from the total execution time.

### Claim 24 (depends on 21)

The method according to claim 21 for application in a processor the method furthermore comprising determining the pipelined fraction by counting every cycle, except when an instruction cache and/or TLB miss accesses off-chip memory, and a load/store cache and/or TLB miss accesses off-chip memory if this load/store miss is the first in a burst of off-chip load/store cache and/or TLB misses.

### Claim 25 (depends on 21)

The method according to claim 21, furthermore comprising determining the non-pipelined fraction by subtracting the pipelined fraction from the total execution time.

### Claim 26 (depends on 21)

The method according to claim 21, furthermore comprising estimating energy consumption based on the estimated execution time.
