Post

GPGPU

GPGPU

在三维的位置看矩阵乘法

matr_pos

一、GPU的基础知识点

一个简易版本的架构区别:

cpu_gpu

GPU索引包含blockIdx和threadIdx,blockIdx是在线程之间共享的

-Warps :The PC is shared; maintain thread mask for Writeback,同一warp内所有线程共用一个程序计数器(PC),因此它们在同一时刻执行相同的指令。由于条件分支或其他条件,GPU会维护一个线程掩码,在执行写回(Writeback)阶段时,只有那些处于活动状态的线程会将计算结果写回到寄存器或内存中。

there are thread groups that share control units,and those are streaming multiprocessors. the core components of our gpu

gpu_archi1 gpu_archi2 gpu_archi3 gpu_archi4 gpu_archi5 gpu_archi6

  1. Raster Engine
  2. ROP
  3. PolyMorph Engine
  4. RT core
  5. TEX unit
  6. Warp Scheduler
  7. Dispatch Unit
  8. SFU
This post is licensed under CC BY 4.0 by the author.