5 ms·

General comment and not aimed at this specific instance: Just because you are writing in assembler, does not mean it is going to run faster than the same code

General comment and not aimed at this specific instance: Just because you are writing in assembler, does not mean it is going to run faster than the same code in a compiled language. There has been decades of research and who knows how many man-years of effort that has gone into producing efficient compiled code from C, C++, Fortran etc. Your assembly skills have to be of quite a decent order to beat a modern compiler. BTW: The answer to the question on Stack Overflow by Peter Cordes is a must-read. Brilliant.

Veedrac 9y ago
Quite decent is perhaps a bit pessimistic. Yes, you need to be fairly fluent if you want to write hand-optimised vector code. On the other hand, you only need to know a little assembly to be able to write C code that compiles better; that compilers are frequently not stupid doesn't mean the level below has nothing to teach. I personally find it a shame that there are so many juicy instructions that compilers have no hope of ever using effectively. How often is a compiler smart enough to solve a problem with PDEP/PEXT, for example? Those functions are versatile as heck, but you need to plan for them if you want them to show up.
15155 9y ago
> Your assembly skills have to be of quite a decent order to beat a modern compiler. Start with compiler output, add intrinsics where you can see help is needed, benchmark, repeat. Pathologically-slow ASM is pretty rare from modern compilers in my experience.
- jzwinck 9y ago
  > Pathologically-slow ASM is pretty rare from modern compilers Here are some I've found: https://stackoverflow.com/questions/45496987/gcc-optimizes-fixed-range-based-for-loop-as-if-it-had-longer-variable-length https://stackoverflow.com/questions/45496987/gcc-optimizes-f... (horrific codegen for known-size C++11 loops in member functions, all GCC versions prior to 8 which is not yet released) https://stackoverflow.com/questions/43651923/gcc-fails-to-optimize-aligned-stdarray-like-c-array https://stackoverflow.com/questions/43651923/gcc-fails-to-op... (SIMD opportunity squandered when C++11 features are used) https://stackoverflow.com/questions/42263537/gcc-sometimes-doesnt-inline-stdarrayoperator https://stackoverflow.com/questions/42263537/gcc-sometimes-d... (failure to inline trivial operators) https://stackoverflow.com/questions/26052640/why-does-gcc-implement-isnan-more-efficiently-for-c-cmath-than-c-math-h https://stackoverflow.com/questions/26052640/why-does-gcc-im... (C isnan() not efficient, for many years)
acdha 9y ago
There's a key corollary: you also either have to be doing this as a one-off or committed to maintaining it over time as hardware changes and those cool tricks become moot or even de-optimizations.
- mjw1007 9y ago
  Also as your code changes. It's often not that hard to hand-write a single function a bit better than the compiler. But when you want to make a small change the compiler will rethink the whole function from scratch, and if you want to keep your advantage you may have to do the same.
  - acdha 9y ago
    That's a very important point: this is especially the case when, say, a data structure grows and clever packing or alignment tricks can no longer be used. I certainly remember that being used as an excuse to delay a change because it was much harder to update some complex optimizations — and at least one case where the original developer tested the generic .c file they'd used to develop the algorithm with a current compiler and found it was faster on the latest hardware we were buying.

Veedrac 9y ago

Quite decent is perhaps a bit pessimistic. Yes, you need to be fairly fluent if you want to write hand-optimised vector code. On the other hand, you only need to know a little assembly to be able to write C code that compiles better; that compilers are frequently not stupid doesn't mean the level below has nothing to teach. I personally find it a shame that there are so many juicy instructions that compilers have no hope of ever using effectively. How often is a compiler smart enough to solve a problem with PDEP/PEXT, for example? Those functions are versatile as heck, but you need to plan for them if you want them to show up.

15155 9y ago

> Your assembly skills have to be of quite a decent order to beat a modern compiler. Start with compiler output, add intrinsics where you can see help is needed, benchmark, repeat. Pathologically-slow ASM is pretty rare from modern compilers in my experience.

jzwinck 9y ago
> Pathologically-slow ASM is pretty rare from modern compilers Here are some I've found: https://stackoverflow.com/questions/45496987/gcc-optimizes-fixed-range-based-for-loop-as-if-it-had-longer-variable-length https://stackoverflow.com/questions/45496987/gcc-optimizes-f... (horrific codegen for known-size C++11 loops in member functions, all GCC versions prior to 8 which is not yet released) https://stackoverflow.com/questions/43651923/gcc-fails-to-optimize-aligned-stdarray-like-c-array https://stackoverflow.com/questions/43651923/gcc-fails-to-op... (SIMD opportunity squandered when C++11 features are used) https://stackoverflow.com/questions/42263537/gcc-sometimes-doesnt-inline-stdarrayoperator https://stackoverflow.com/questions/42263537/gcc-sometimes-d... (failure to inline trivial operators) https://stackoverflow.com/questions/26052640/why-does-gcc-implement-isnan-more-efficiently-for-c-cmath-than-c-math-h https://stackoverflow.com/questions/26052640/why-does-gcc-im... (C isnan() not efficient, for many years)

acdha 9y ago

There's a key corollary: you also either have to be doing this as a one-off or committed to maintaining it over time as hardware changes and those cool tricks become moot or even de-optimizations.

mjw1007 9y ago
Also as your code changes. It's often not that hard to hand-write a single function a bit better than the compiler. But when you want to make a small change the compiler will rethink the whole function from scratch, and if you want to keep your advantage you may have to do the same.
- acdha 9y ago
  That's a very important point: this is especially the case when, say, a data structure grows and clever packing or alignment tricks can no longer be used. I certainly remember that being used as an excuse to delay a change because it was much harder to update some complex optimizations — and at least one case where the original developer tested the generic .c file they'd used to develop the algorithm with a current compiler and found it was faster on the latest hardware we were buying.