Branchless Programming in C++ Fedor G Pikus Chief ScientistBranchless Computing 3 PLAN ● Efficiency and performance ● Understanding the hardware and using it efficiently – Computing resources of loop unrolling ● Conditional code vs efficiency ● Optimizing conditional code ● Branchless programmingWHAT CAN BRANCHLESS OPTIMIZATIONS DO? f(bool b, unsigned long x, unsigned long& s) {if (b) s +=x;} += v1[i]*v2[i]; }Branchless Computing 8 COMPUTING RESOURCES OF A CPU unsigned long v1[N], v2[N]; unsigned long a = 0; for (size_t i = 0; i < N; ++i) { a += v1[i]*v2[i]; }Branchless Computing 9 COMPUTING

性能优化 之 无分支编程 Branchless Programming by 彭于斌（ @archibate ） 两种代码写法：分支 vs 三目运算符 两种使用方式：排序 vs 不排序 测试结果（均为 gcc -O3 ） 测试结果可视化 图表比较：分支 vs 无分支 分支 无分支 0 0.01 0.02 0.03 耗时（越低越好） 乱序 有序 • 传统的分支方法实现的

ls_not_halted_cyc:uperf record –g –p 48Branchless binary search 49 template ForwardIt branchless_lower_bound(ForwardIt first, ForwardIt last, const generate CMOV first += comp(first[half], value) * (length - half); length = half; } return first; }Branchless binary search 50HW counters with libpapi 51 #include "papipp.h“ // see https://github (events.get().counter() / (double)events.get().counter()) << std::endl;Branchless binary search 52Binary search – memory access 53Linear search 54Principle #4: “Simplicity is

may make sense, but generally, we should avoid that. Later in the book, there is a section about branchless code where we will present a procedure that actually may get faster by using such a trick. Characters

may make sense, but generally, we should avoid that. Later in the book, there is a section about branchless code where we will present a procedure that actually may get faster by using such a trick. Characters

may make sense, but generally, we should avoid that. Later in the book, there is a section about branchless code where we will present a procedure that actually may get faster by using such a trick. 78

may make sense, but generally, we should avoid that. Later in the book, there is a section about branchless code where we will present a procedure that actually may get faster by using such a trick. Characters

may make sense, but generally, we should avoid that. Later in the book, there is a section about branchless code where we will present a procedure that actually may get faster by using such a trick. 78

may make sense, but generally, we should avoid that. Later in the book, there is a section about branchless code where we will present a procedure that actually may get faster by using such a trick. Characters

may make sense, but generally, we should avoid that. Later in the book, there is a section about branchless code where we will present a procedure that actually may get faster by using such a trick. Characters