The next update to the manuals
is finally here. The most important additions are:
- Test results for AMD Bulldozer (manual 2, 3, 4).
- Shared objects in Unix systems are inefficient because of
position-independent code, symbol interposition, global offset table (GOT)
and procedure linkage table (PLT). All references to public symbols (and
even some references to local symbols) require a table lookup, which is a
complete waste of time if we don't need the symbol interposition feature,
which we rarely do. The updated C++ manual gives advice on how avoid these
time-consuming complications for shared objects (*.so) in Linux, BSD and Mac
systems (manual 1).
- Updated advice on memory copying, with description of false memory
dependence (manual 2).
- Methods for integer division by a constant revised, with more references
and support in the asmlib
library (manual 2).
- Chapter on vectorization updated and revised, including AVX (manual 1).
- SSE4.2 string instructions (manual 2).
- Small modifications in chapter on multithreading (manual 1).
- The instruction tables are now available both as a .pdf file (link)
and as a spreadsheet (link)
because several people have requested this (manual 4).
|
Optimization manuals updated - Agner Fog - 2009-09-27
Optimization manuals updated - Agner Fog - 2009-09-29