公开(公告)号：US20150254077A1

公开(公告)日：2015-09-10

申请号：US14601598

申请日：2015-01-21

Applicant: ARM Limited

Inventor： Matthias BOETTCHER ,  Mbou EYOLE-MONONO ,  Giacomo GABRIELLI

IPC: G06F9/30

CPC classification number: G06F9/30036 ,  G06F9/30018 ,  G06F9/30032 ,  G06F9/30072 ,  G06F9/30098 ,  G06F9/3017

Abstract: A data processing apparatus has processing circuitry for processing vector operands from a vector register store in response to vector micro-operations, some of which have control information identifying which data elements of the vector operands are selected for processing. Control circuitry detects vector micro-operations for which the control information specifies that a portion of the vector operand to be processed has no selected elements. If this is the case, then the control circuitry controls the processing circuitry to process a lower latency replacement micro-operation instead of the original micro-operation. This provides better performance than if a branch instruction is used to bypass the micro-operation if there are no selected elements.

Abstract translation: 数据处理装置具有处理电路，用于响应向量微操作来处理来自向量寄存器存储器的向量操作数，其中一些具有标识向量操作数的哪些数据元素被选择用于处理的控制信息。 控制电路检测矢量微操作，其中控制信息指定要处理的矢量操作数的一部分没有选定的元件。 如果是这种情况，则控制电路控制处理电路来处理较低等待时间的替换微操作，而不是原始的微操作。 这提供了比使用分支指令绕过微操作（如果没有选定的元素）更好的性能。

公开(公告)号：US20150254076A1

公开(公告)日：2015-09-10

申请号：US14601431

申请日：2015-01-21

Applicant: ARM LIMITED

Inventor： Matthias BOETTCHER ,  Giacomo GABRIELLI ,  Mbou EYOLE-MONONO

IPC: G06F9/30

CPC classification number: G06F9/30036 ,  G06F9/3001 ,  G06F9/30098

Abstract: A vector scan operation is performed to generate M data elements of a result vector, where each result data element corresponds to a combination of an additional data element with at least some of the data elements of a source vector operand V. The vector scan operation is performed using a plurality of steps, each step comprising one or more combination operations for combining data elements. At least one of the steps includes two or more combination operations performed in parallel. At least two of the steps comprise a combination operation for combining a data element with the additional data element S. This approach enables the vector scan operation to be performed in fewer steps in the case where fewer than M data elements are active, so that the vector scan operation can be performed more quickly.

Abstract translation: 执行矢量扫描操作以产生结果矢量的M个数据元素，其中每个结果数据元素对应于附加数据元素与源向量操作数V的至少一些数据元素的组合。矢量扫描操作是 使用多个步骤执行，每个步骤包括用于组合数据元素的一个或多个组合操作。 至少一个步骤包括并行执行的两个或多个组合操作。 至少两个步骤包括用于将数据元素与附加数据元素S组合的组合操作。在少于M个数据元素有效的情况下，该方法使得能够以更少的步骤执行向量扫描操作，使得 可以更快地执行向量扫描操作。