0x01

摘要： HOW TO OPTIMIZE GEMM 介绍一些常规的优化思路，参考：https://github.com/flame/how-to-optimize-gemm/wiki baseline /* Create macros so that the matrices are stored in co 阅读全文

摘要： 问题背景： https://airflow-megengine.iap.hh-d.brainpp.cn/log?dag_id=megbrain-release&task_id=prebuild-cu111&execution_date=2022-10-08T06%3A06%3A51%2B00%3A0 阅读全文

摘要： 2022.Q3 沟通 with wangbiao 晋升milestone: Q3: 怎么做目标切分 怎么当一个owner Q4: 要独立当一个owner 2023.Q1: 要学习带人做项目 kernel优化路线 cuda c -> tensor core -> cutlass -> tvm Q3 p 阅读全文

摘要： 03 Feb 2023 : 最近一周，重新梳理了dnn上rrconv的codegen代码，dnn上rrconv fprop全部test跑通。dnn rrconv dgrad不能通过，部分case计算错误。rrconv cutlass dgrad全部通过。 2.2号来了，先检查dgrad的codege 阅读全文

摘要： https://developer.nvidia.com/blog/cutlass-linear-algebra-cuda/ Efficient Matrix Multiplication on GPUs 计算密集度 = (时间复杂度/空间复杂度) = O(N^3)/O(N^2) = O(N) // 阅读全文

摘要： #include <vector> #include <utility> // 对于 std::move #include <type_traits> #include <iostream> using namespace std; template<typename T> struct A { A 阅读全文

摘要： problem1 How many bytes is the program? For the above x86 assembly code, how many bytes of instructions need to be fetched if x = 0x01020304 and n = 5 阅读全文

摘要： the report finished in first time the report finished in first time 3.4 Note how the mix of different types of instructions vary between benchmarks. R 阅读全文

摘要：  阅读全文

摘要： 参考: code: https://github.com/mit-pdos/xv6-riscv book: https://pdos.csail.mit.edu/6.828/2021/xv6/book-riscv-rev2.pdf note: https://mit-public-courses-c 阅读全文