Dynamic enabling and disabling of simd units in graphics processors

Abstract

The present invention describes a system and method for increasing performance in a graphics processing unit. Embodiments achieve power savings in graphics processing units by dynamically activating / deactivating individual SIMDs in shader compositors containing multiple SIMD units. Dynamic on-the-fly disabling and enabling of individual SIMDs provides flexibility to achieve desired performance and power levels in a given processing application. In this way, embodiments enable optimal use of graphics processing units. Embodiments of the present invention also implement SIMD dynamic texture (eg medium texture) clock gating in the shader compositor. The embodiment provides a clock mechanism on demand and closes unused logical clock trees to reduce switching power. In this way, embodiments enhance clock gating to save more switching power for the duration when the SIMD is idle (or not assigned any work).

Description

technical field [0001] Embodiments of the invention generally relate to the control of a single instruction multiple data (SIMD) unit in a graphics processing unit (GPU). Background technique [0002] The arithmetic unit of the graphics processor includes an arithmetic logic unit (ALU) or an arithmetic unit configured to perform integer, logical, and single / double precision floating point operations. [0003] Graphics processors may include arrays of arithmetic units known as shader cores. A shader core includes a shader pipeline (SP). For higher performance, multiple SPs can be configured to work together as stacked SPs. SPs can be configured as Quad Arrays (QP) and SIMD. All shader units per SIMD can conditionally execute the same ALU instruction on different sets of origin, vertex, or pixel values. In this way, SIMD provides arithmetic processing power to the graphics processor. [0004] In conventional GPUs with stacked SPs, SIMD can be statically enabled or disable...

AI Technical Summary

Problems solved by technology

However, traditional methods continuously and actively provide clock signals to these SIMD

This method increases the power consumption of the graphics processor and is inefficient

Images