coding spirit

User Tools

Site Tools

You are here: start » development » Performance with C++
development:performance

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

development:performance [2025/10/30] – created hayatidevelopment:performance [Unknown date] (current) – external edit (Unknown date) 127.0.0.1
Line 1: Line 1:
 +~~NOTRANS~~
 +{{indexmenu_n>600}}
 +
 +====== Performance with C++ ======
 +
 +===== IEEE-754 Float Numbers =====
 +
 +as written in [[development:numeric_math#ieee-754_float_numbers|Numeric / Math / Linear Algebra]], calculation with denormals or non-numbers slows down performance - even when not signalled. \\
 +it might be interesting to abort a calculation, e.g. a matrix/vector multiplication, with first occurence of NaN - or with one of the other conditions ..
 +
 +===== Optimization Remarks from Compiler =====
 +
 +[[https://ofekshilon.com/|Ofek Shilon]] held the talk [[https://www.youtube.com/watch?v=qmEsx4MbKoc|Optimization Remarks - Helping the Compiler to Generate Better Code]] at [[https://cppcon.org/|CppCon 2022]]. Here his [[https://github.com/OfekShilon/CppCon2022/blob/main/Presentations/Optimization-Remarks.pdf|PDF slides]].
 +Here also slides of a sligthly older talk on the same topic: https://www.slideshare.net/ofekshilon/optview2-c-on-sea-2022
 +
 +references from his slides:
 +
 +  * llvm-opt-report
 +    * https://reviews.llvm.org/D25262
 +    * https://github.com/llvm/llvm-project/tree/main/llvm/tools/llvm-opt-report
 +    * https://llvm.org/docs/Remarks.html
 +    * debian package ''llvm-14-tools''
 +    * clang compiler switch ''-fsave-optimization-record''
 +  * opt-viewer / optview2
 +    * https://github.com/llvm/llvm-project/tree/main/llvm/tools/opt-viewer  (origin)
 +    * https://github.com/OfekShilon/optview2  produces much better output
 +    * ''opt-viewer.py --output-dir <htmls folder> --source-dir <repo> <yamls folder>''
 +  * //Clobbered by store//
 +    * use compiler specific keyword ''%%__restrict%%'' or ''%%HEDLEY_RESTRICT%%''
 +      * https://github.com/nemequ/Hedley
 +  * //Clobbered by call// or //load//
 +    * function attribute ''%%__attribute__((const))%%'' or ''%%HEDLEY_CONST%%''
 +      * ''const'' functions are not using any saved/global state; thus free of side-effects
 +      * compiler can remove repeated invokations
 +      * ''const'' is stronger than ''pure''
 +    * function attribute ''%%__attribute__((pure))%%'' or ''%%HEDLEY_PURE%%''
 +      * see https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html for ''pure'' and ''const''
 +      * hedley doc is much easier to understand / discriminate:
 +        * https://nemequ.github.io/hedley/api-reference.html#HEDLEY_PURE
 +        * https://nemequ.github.io/hedley/api-reference.html#HEDLEY_CONST
 +    * function parameter attribute ''%%__attribute__((noescape))%%'' or ''%%HEDLEY_NO_ESCAPE%%''
 +      * see https://clang.llvm.org/docs/AttributeReference.html#noescape
 +  * Talk //Compiler-assisted performance analysis// of Adam Nemet
 +    * https://www.youtube.com/watch?v=qq0q1hfzidg
 +
 +===== Sites, Blogs and Books =====
 +
 +  * Denis Bakhalov - EasyPerf!
 +    * https://easyperf.net/
 +  * Johnny’s Software Lab
 +    * https://johnnysswlab.com/