公开(公告)号：US08635483B2

公开(公告)日：2014-01-21

申请号：US13079842

申请日：2011-04-05

申请人： Emrah Acar ,  Pradip Bose ,  Bishop C. Brock ,  Alper Buyuktosunoglu ,  Michael S. Floyd ,  Maria L. Pesantez ,  Gregory S. Still

发明人： Emrah Acar ,  Pradip Bose ,  Bishop C. Brock ,  Alper Buyuktosunoglu ,  Michael S. Floyd ,  Maria L. Pesantez ,  Gregory S. Still

IPC分类号： G06F1/00 ,  G06F1/26 ,  G06F15/00 ,  G01R21/00 ,  G01R19/00

CPC分类号： G06F1/3243 ,  Y02D10/152

摘要： A mechanism is provided for automatically tuning power proxy architectures. Based on the set of conditions related to an application being executed on a microprocessor core, a weight factor to use for each activity in a set of activities being monitored for the microprocessor core is identified, thereby forming a set of weight factors. A power usage estimate value is generated using the set of activities and the set of weight factors. A determination is made as to whether the power usage estimate value is greater than a power proxy threshold value identifying a maximum power usage for the microprocessor core. Responsive to the power usage estimate value being greater than the power proxy threshold value, a set of signals is sent to one or more on-chip actuators in the power proxy unit associated with the microprocessor core and a set of operational parameters associated with the component are adjusted.

摘要翻译： 提供了一种自动调整电源代理架构的机制。 基于与在微处理器核心上执行的应用有关的条件集合，识别用于针对微处理器核心的一组活动中的每个活动的权重因子，从而形成一组权重因子。 使用一组活动和一组权重因子生成用电量估计值。 确定功率使用估计值是否大于识别微处理器核的最大功率使用的功率代理阈值。 响应于功率使用估计值大于功率代理阈值，一组信号被发送到与微处理器核心相关联的功率代理单元中的一个或多个片上致动器，以及与组件相关联的一组操作参数 被调整。

公开(公告)号：US20120260117A1

公开(公告)日：2012-10-11

申请号：US13079842

申请日：2011-04-05

申请人： Emrah Acar ,  Pradip Bose ,  Bishop C. Brock ,  Alper Buyuktosunoglu ,  Michael S. Floyd ,  Maria L. Pesantez ,  Gregory S. Still

发明人： Emrah Acar ,  Pradip Bose ,  Bishop C. Brock ,  Alper Buyuktosunoglu ,  Michael S. Floyd ,  Maria L. Pesantez ,  Gregory S. Still

IPC分类号： G06F1/28

CPC分类号： G06F1/3243 ,  Y02D10/152

摘要： A mechanism is provided for automatically tuning power proxy architectures. Based on the set of conditions related to an application being executed on a microprocessor core, a weight factor to use for each activity in a set of activities being monitored for the microprocessor core is identified, thereby forming a set of weight factors. A power usage estimate value is generated using the set of activities and the set of weight factors. A determination is made as to whether the power usage estimate value is greater than a power proxy threshold value identifying a maximum power usage for the microprocessor core. Responsive to the power usage estimate value being greater than the power proxy threshold value, a set of signals is sent to one or more on-chip actuators in the power proxy unit associated with the microprocessor core and a set of operational parameters associated with the component are adjusted.

摘要翻译： 提供了一种自动调整电源代理架构的机制。 基于与在微处理器核心上执行的应用有关的一组条件，识别用于针对微处理器核心的一组活动中的每个活动的权重因子，从而形成一组权重因子。 使用一组活动和一组权重因子生成用电量估计值。 确定功率使用估计值是否大于识别微处理器核的最大功率使用的功率代理阈值。 响应于功率使用估计值大于功率代理阈值，一组信号被发送到与微处理器核心相关联的功率代理单元中的一个或多个片上致动器，以及与该组件相关联的一组操作参数 被调整。

公开(公告)号：US20100268930A1

公开(公告)日：2010-10-21

申请号：US12749179

申请日：2010-03-29

申请人： Pradip Bose ,  Alper Buyuktosunoglu ,  Michael S. Floyd ,  Maria L. Pesantez

发明人： Pradip Bose ,  Alper Buyuktosunoglu ,  Michael S. Floyd ,  Maria L. Pesantez

IPC分类号： G06F9/00 ,  G06F1/00

CPC分类号： G06F1/3203 ,  G06F11/3062 ,  G06F11/3096 ,  Y02D10/126

摘要： The embodiments provide an assigned counter of a first set of counters and stores a value for an activity of a set of activities forming a set of stored values. The value comprises the count multiplied by a weight factor specific to the activity. A power manager manages the first set of counters, receives a set of activities to be monitored for a unit, groups the portion into subsets based on at least one of a frequency of occurrence of each activity and power consumption for each activity, sums the stored values corresponding to each activity in each subset to reach a total value for each subset, multiplies the total value of each subset by factor corresponding to the subset to form a scaled value for each subset, and sums the scaled value of each subset to form a power usage value.

摘要翻译： 这些实施例提供了第一组计数器的分配的计数器，并且存储形成一组存储值的一组活动的活动的值。 该值包括计数乘以活动特有的权重因子。 功率管理器管理第一组计数器，接收一组要监视单元的活动，基于每个活动的发生频率和每个活动的功耗中的至少一个将该部分分组成子集，将所存储的 对应于每个子集中的每个活动的值以达到每个子集的总值，将每个子集的总值乘以与该子集对应的因子，以形成每个子集的缩放值，并且将每个子集的缩放值相加以形成 电力使用价值。

公开(公告)号：US20140195996A1

公开(公告)日：2014-07-10

申请号：US13526252

申请日：2012-06-18

申请人： Pradip Bose ,  Alper Buyuktosunoglu ,  John A. Darringer ,  Moinuddin K. Qureshi ,  Jeonghee Shin

发明人： Pradip Bose ,  Alper Buyuktosunoglu ,  John A. Darringer ,  Moinuddin K. Qureshi ,  Jeonghee Shin

IPC分类号： G06F17/50

CPC分类号： G06F17/5072 ,  G06F17/5036

摘要： A dynamic system coupled with “pre-Silicon” design methodologies and “post-Silicon” current optimizing programming methodologies to improve and optimize current delivery into a chip, which is limited by the physical properties of the connections (e.g., Controlled Collapse Chip Connection or C4s). The mechanism consists of measuring or estimating power consumption at a certain granularity within a chip, converting the power information into C4 current information using a method, and triggering throttling mechanisms (including token based throttling) where applicable to limit the current delivery per C4 beyond pre-established limits or periods. Design aids are used to allocate C4s throughout the chip based on the current delivery requirements. The system coupled with design and programming methodologies improve and optimize current delivery is extendable to connections across layers in a multilayer 3D chip stack.

摘要翻译： 与“前置硅”设计方法和“后硅”当前优化编程方法相结合的动态系统，以改进和优化当前传输到芯片的传输，这受到连接的物理特性的限制（例如，受控崩溃芯片连接或 C4s）。 该机制包括测量或估计芯片内特定粒度的功耗，使用一种方法将功率信息转换为C4当前信息，并在适用的情况下触发限流机制（包括基于令牌的限制），以限制每个C4超过预先的当前传输 - 建立的限制或期限。 设计辅助工具用于根据当前交付要求在整个芯片上分配C4。 该系统与设计和编程方法相结合，改进并优化电流传递可扩展到多层3D芯片堆叠中的层间连接。

公开(公告)号：US20140082574A1

公开(公告)日：2014-03-20

申请号：US13526153

申请日：2012-06-18

申请人： Pradip Bose ,  Alper Buyuktosunoglu ,  John A. Darringer ,  Moinuddin K. Qureshi ,  Jeonghee Shin

发明人： Pradip Bose ,  Alper Buyuktosunoglu ,  John A. Darringer ,  Moinuddin K. Qureshi ,  Jeonghee Shin

IPC分类号： G06F17/50

CPC分类号： G06F17/5054 ,  G06F9/455 ,  G06F9/46 ,  G06F9/5044 ,  G06F15/177 ,  G06F17/5022 ,  G06F17/5045 ,  G06F2212/1041 ,  G06F2217/78

摘要： A dynamic system coupled with “pre-Silicon” design methodologies and “post-Silicon” current optimizing programming methodologies to improve and optimize current delivery into a chip, which is limited by the physical properties of the connections (e.g., Controlled Collapse Chip Connection or C4s). The mechanism consists of measuring or estimating power consumption at a certain granularity within a chip, converting the power information into C4 current information using a method, and triggering throttling mechanisms (including token based throttling) where applicable to limit the current delivery per C4 beyond pre-established limits or periods. Design aids are used to allocate C4s throughout the chip based on the current delivery requirements. The system coupled with design and programming methodologies improve and optimize current delivery is extendable to connections across layers in a multilayer 3D chip stack.

摘要翻译： 与“前置硅”设计方法和“后硅”当前优化编程方法相结合的动态系统，以改进和优化当前传输到芯片的传输，这受到连接的物理特性的限制（例如，受控崩溃芯片连接或 C4s）。 该机制包括测量或估计芯片内特定粒度的功耗，使用一种方法将功率信息转换为C4当前信息，并在适用的情况下触发限流机制（包括基于令牌的限制），以限制每个C4超过预先的当前传输 - 建立的限制或期限。 设计辅助工具用于根据当前交付要求在整个芯片上分配C4。 该系统与设计和编程方法相结合，改进并优化电流传递可扩展到多层3D芯片堆叠中的层间连接。

公开(公告)号：US08527994B2

公开(公告)日：2013-09-03

申请号：US13024781

申请日：2011-02-10

申请人： Pradip Bose ,  Alper Buyuktosunoglu ,  Niti Madan

发明人： Pradip Bose ,  Alper Buyuktosunoglu ,  Niti Madan

IPC分类号： G06F9/46

CPC分类号： G06F11/3058 ,  G06F1/3206 ,  G06F1/3234 ,  G06F9/5094 ,  Y02D10/22

摘要： A mechanism is provided for guarded, multi-metric resource control. Monitoring is performed for an intended action to address a negative condition from a resource manager in a plurality of resource managers in the data processing system. Responsive to receiving the intended action, a determination is made as to whether the intended action will cause an additional negative condition within the data processing system. Responsive to determining that the intended action will cause the additional negative condition within the data processing system, at least one alternative action is identified to be implemented in the data processing system that addresses the negative condition while not causing any additional negative condition. The at least one alternative action is then implemented in the data processing system.

摘要翻译： 提供了一种用于保护的多度量资源控制的机制。 执行针对数据处理系统中的多个资源管理器中的资源管理器处理负面条件的预期动作的监视。 响应于接收预期的动作，确定预期动作是否将在数据处理系统内引起额外的负面情况。 响应于确定预期动作将导致数据处理系统内的附加负面条件，至少一个备选动作被识别为在不引起任何附加负面条件的情况下处理负条件的数据处理系统中实现。 然后在数据处理系统中实现至少一个备选动作。

公开(公告)号：US08001394B2

公开(公告)日：2011-08-16

申请号：US12023536

申请日：2008-01-31

申请人： Pradip Bose ,  Alper Buyuktosunoglu ,  Michael Stephen Floyd

发明人： Pradip Bose ,  Alper Buyuktosunoglu ,  Michael Stephen Floyd

IPC分类号： G06F1/26 ,  G06F1/32

CPC分类号： G06F1/3203 ,  G06F1/3243 ,  G06F1/3287 ,  Y02D10/152 ,  Y02D10/171

摘要： A computer-implemented method and a system for managing power in a multi-core microprocessor are provided. A power management control microarchitecture in a chiplet translates a first command comprising a power setting. A chiplet comprises a processor core and associated memory cache. The power management control microarchitecture comprises power mode registers, power mode adjusters, translators, and microarchitectural power management techniques. The power management control microarchitecture sets microarchitectural power management techniques according to the power setting. The global power management controller issues the first command. The global power management controller may reside either on or off of the microprocessor. The global power management controller issues commands either directly for a specific chiplet out of the plurality of chiplets or to the plurality of chiplets and the control slave bus translates the command into sub-commands dedicated to specific chiplets within the plurality of chiplets. Each chiplet may be set to separate power levels.

摘要翻译： 提供了一种计算机实现的方法和用于管理多核微处理器中的电力的系统。 小电流中的电源管理控制微体系结构转换包括功率设置的第一命令。 小巧包括处理器核心和相关联的存储器高速缓存。 功率管理控制微体系结构包括功率模式寄存器，功率模式调节器，转换器和微架构电源管理技术。 电源管理控制微架构根据功率设置设置微体系结构电源管理技术。 全球电源管理控制器发出第一个命令。 全局功率管理控制器可以驻留在微处理器上或者关闭。 全局功率管理控制器直接针对多个小芯片中的特定小时或多个小芯片发出命令，并且控制从总线将命令转换为专用于多个小芯片内的特定小芯片的子命令。 每个小穗可以设置为分开的功率水平。

公开(公告)号：US20110040994A1

公开(公告)日：2011-02-17

申请号：US12539941

申请日：2009-08-12

申请人： Jayanta Basak ,  Pradip Bose ,  Alper Buyuktosunoglu ,  Anita Lungu

发明人： Jayanta Basak ,  Pradip Bose ,  Alper Buyuktosunoglu ,  Anita Lungu

IPC分类号： G06F1/26

CPC分类号： G06F1/3203 ,  G06F11/3466 ,  Y02D10/34

摘要： A mechanism is provided for two-level guarded predictive power gating of a set of units within the data processing system. A success determines whether a unit within the set of units is power gated during a monitoring interval. If the unit is power gated, the success monitor determines whether a count of idle cycles for the unit is below a breakeven point. If the count is above the breakeven point, the success monitor increments a success efficiency counter. If the count is below the breakeven point, the success monitor determines whether the unit needs to be woke up. If the unit needs to be woke up, the success monitor increments a harmful efficiency counter. If the value of the harmful efficiency counter is less than the value from the success efficiency counter, the success monitor enables power gating for the unit via a first-level power-gating mechanism.

摘要翻译： 提供了一种用于数据处理系统内的一组单元的两级保护预测能力门控的机制。 一个成功的确定在一个监视间隔期间该单元组内的一个单元是否电源门控。 如果该单元是电源门控，则成功监视器确定该单元的空闲周期计数是否低于盈亏平衡点。 如果计数高于盈亏平衡点，则成功监视器会增加成功效率计数器。 如果计数低于盈亏平衡点，则成功监视器确定该单元是否需要醒来。 如果单位需要醒来，成功监测器会增加一个有害的效率计数器。 如果有害效率计数器的值小于成功效率计数器的值，则成功监视器可通过一级电源门控机构对单元进行功率门控。

公开(公告)号：US20100268975A1

公开(公告)日：2010-10-21

申请号：US12424161

申请日：2009-04-15

申请人： Pradip Bose ,  Alper Buyuktosunoglu ,  Michael Stephen Floyd

发明人： Pradip Bose ,  Alper Buyuktosunoglu ,  Michael Stephen Floyd

IPC分类号： G06F1/26

CPC分类号： G06F11/348 ,  G06F11/3006 ,  G06F11/3024 ,  G06F11/3062 ,  G06F2201/88 ,  Y02D10/34

摘要： A method for estimating power consumption within a multi-core microprocessor chip is provided. An authorized user selects a set of activities to be monitored. A value for each activity of the set of activities is stored in a separate counter of a set of counters, forming a set of stored values. The value comprises the count multiplied by a weight factor specific to the activity. The set of activities are grouped into subsets. The stored values corresponding to each activity in each subset are summed, forming a total value for each subset. The total value of each subset is multiplied by a factor corresponding to the subset, forming a scaled value for each subset. The scaled value of each subset is summed, forming a power usage value. A power manager adjusts the operational parameters of the unit based on a comparison of the power usage value to a threshold value.

摘要翻译： 提供了一种用于估计多核微处理器芯片内的功耗的方法。 授权用户选择要监视的一组活动。 一组活动的每个活动的值存储在一组计数器的单独计数器中，形成一组存储的值。 该值包括计数乘以活动特有的权重因子。 该组活动被分组成子集。 将对应于每个子集中的每个活动的存储值相加，形成每个子集的总值。 每个子集的总值乘以与子集对应的因子，形成每个子集的缩放值。 将每个子集的缩放值相加，形成功率使用值。 功率管理器基于功率使用值与阈值的比较来调整单元的操作参数。

公开(公告)号：US20100262808A1

公开(公告)日：2010-10-14

申请号：US12420143

申请日：2009-04-08

申请人： Pradip Bose ,  Alper Buyuktosunoglu ,  Michael Stephen Floyd ,  Dung Quoc Nguyen ,  Bruce Joseph Ronchetti

发明人： Pradip Bose ,  Alper Buyuktosunoglu ,  Michael Stephen Floyd ,  Dung Quoc Nguyen ,  Bruce Joseph Ronchetti

IPC分类号： G06F9/30

CPC分类号： G06F9/3861 ,  G06F9/3824 ,  G06F9/3836 ,  G06F11/348 ,  G06F2201/88 ,  Y02D10/34

摘要： The illustrative embodiments described herein provide a computer-implemented method, apparatus, and a system for managing instructions. A load/store unit receives a first instruction at a port. The load/store unit rejects the first instruction in response to determining that the first instruction has a first reject condition. Then, the instruction sequencing unit activates a first bit in response to the load/store unit rejection the first instruction. The instruction sequencing unit blocks the first instruction from reissue while the first bit is activated. The processor unit determines a class of rejection of the first instruction. The instruction sequencing unit starts a timer. The length of the timer is based on the class of rejection of the first instruction. The instruction sequencing unit resets the first bit in response to the timer expiring. The instruction sequencing unit allows the first instruction to become eligible for reissue in response to resetting the first bit.

摘要翻译： 本文描述的说明性实施例提供了一种计算机实现的方法，装置和用于管理指令的系统。 加载/存储单元在端口接收第一条指令。 响应于确定第一指令具有第一拒绝条件，加载/存储单元拒绝第一指令。 然后，指令排序单元响应于加载/存储单元来激活第一位以拒绝第一指令。 当第一位被激活时，指令排序单元阻止重新发行的第一条指令。 处理器单元确定第一指令的拒绝类。 指令排序单元启动定时器。 定时器的长度取决于第一条指令的拒绝类型。 指令排序单元重置响应定时器超时的第一位。 响应于重置第一位，指令排序单元允许第一指令变得有资格重新发行。