Agner's CPU blog

I have written a series of five manuals describing everything you need to know about optimizing code for x86 and x86-64 family microprocessors, including optimization advice for C++ and assembly language, details about the microarchitecture and instruction timings of Intel, AMD and VIA processors, and details about different compilers and calling conventions.

1. Optimizing software in C++: An optimization guide for Windows, Linux and Mac platforms
2. Optimizing subroutines in assembly language: An optimization guide for x86 platforms
3. The microarchitecture of Intel, AMD and VIA CPUs: An optimization guide for assembly programmers and compiler makers
4. Instruction tables: Lists of instruction latencies, throughputs and micro-operation breakdowns for Intel, AMD and VIA CPUs
5. Calling conventions for different C++ compilers and operating systems

These five manuals are continuously maintained and updated whenever I have new information to provide. You can always find the latest version at https://www.agner.org/optimize/#manuals

Could you provide EPUB, MOBI, or AZW3 versions of the manuals?

Sorry, no. It is too much work to maintain multiple versions of the manuals. I think that almost all E-book readers support the pdf format.

Hi Agner,

I was reading your book, Optimization C++, many years ago.
Today I wanted to find arithmetic hacks and I wasn't able to do that.
I checked random older versions on internet archive.


I remember hacks such as:

• An int value is treated as a vector of bytes (simulation SIMD in plain C)
• Add number if it is bigger than other number without if.

Please, cloud you hint a version of the book having such stuff?

Sorry,



I found what I was looking for

https://graphics.stanford.edu/~seander/bithacks.html

Over the years these papers got mixed in my head.

Doing vector operations in general purpose registers is explained in my assembly manual chapter 13.9.

The book: Hacker's Delight by Henry S. Warren (2003) also has a lot of useful tricks.

In the 5. Calling conventions volume, chapter 7, could you perhaps make it more clear how C++ empty objects are passed as a function argument? I know there's a "has now data" row in the table, but even with that it's not clear. I did some experiments. It seems that on x86_64 Linux, the empty object is just ignored, it doesn't take space on either the stack or registers. In x86_64 Windows, it seems one of the argument registers is reserved for the empty object (though I haven't checked whether the register is zeroed or what happens for arguments after the four registers are exhausted).

Maybe you already know, maybe not ... in manual no. 3 "The microarchitecture of Intel, AMD and VIA CPUsAn optimization guide for assembly programmers and compiler makers" Last updated 2020-03-08 there are two times chapters 3.1 - 3.6

Is there a correct place for suggestions/comments on the optimization manuals?
Found a small typo on optimizing_assembly.pdf page 113:
"AVX512IFMA. Fused multiply-and-add on 52-bit integers."
That was meant to be 32-bit integers, I suppose.

AVX512IFMA is indeed multiplying 52 bit integers.