Re: Top 10 most common hard skills listed on resumes...

On 28/08/2024 07:02, Bonita Montero wrote:

Am 27.08.2024 um 20:54 schrieb David Brown:
 

And you were completely wrong when you said that.  Perhaps in /your/ field of programming you are correct - but you are ignoring the rest
of the world.

 You've to omit nearly the whole standard libary to avoid exceptions.
F.e. everything that has an allocator may throw bad_alloc.

And in small systems programming, much of the standard library /is/ omitted.  When you are doing real-time or critical software, you avoid dynamic memory like the plague.  The standard container classes (other than std::array<>) are inappropriate for most low-level or small-systems programming.
The standard C++ libraries for gcc and clang will not throw exceptions if you have "-fno-exceptions" enabled - on allocation failure, they will go straight to abort().  You can do better by using custom allocators - alternatively, you can use more appropriate container classes that are suitable for such systems.

 

Often that is correct.  Often it is /not/ correct.

 It's correct nearly every time since usually you've more than a dozen
instructions in a virtual function.

No, it is not - function calls when the compiler cannot see, or cannot use, the definition of the called function can have significant overheads well above the cost of a dozen instructions.  Virtual function calls add to that due to an extra layer of indirection.  And many virtual functions are very small, simply reading or writing a variable.
So any claims about how much of an overhead you have are meaningless - all that can be said is that they are sometimes relevant, sometimes not.

 

For every one of your favourite big x86 chips sold, there will be a hundred small microcontrollers - none of which has branch prediction.

 Even the simple Cortex CPUs have a branch prediciton.
 

No, they don't.  The Cortex-M cores do not have branch prediction.  Some have static speculative prefetch of branch targets from flash to reduce the delay on branch, but no more than that.  These are not out-of-order processors, they have no speculative execution, and only the Cortex-M7 is superscalar (slightly).  This is, of course, a benefit - the cores are intended to be small and low power, and to have consistent execution times rather than maximal average throughput.
(I am not even sure why you thought branch prediction was relevant here.)