Re: [PATCH] Better __ashlDI3, __ashrDI3 and __lshrDI3 functions, plus fixed __bswapsi2 function

["Stefan Kanthak" <stefan.kanthak@nexgo.de>]

Jeff Law <law@redhat.com> wrote:

[...]

> By understanding how your proposed changes affect other processors, you
> can write better changes that are more likely to get included. 
> Furthermore you can focus efforts on things that matter more in the real
> world. DImode shifts in libgcc are _not_ useful to try and optimize on
> x86_64 as it has instructions to implement 64 bit shifts. DImode shifts
> in libgcc are not useful to try and optimize on i686 as the compiler can
> synthesize them on i686. DImode shifts can't be easily synthesized on
> other targets and on those targets we call the routines in libgcc2. 
> Similarly for other routines you find in libgcc2.

What makes you think that my patches addressed only i386 and AMD64?

Again: from the absence of __addDI3/__subDI3 in libgcc2.[ch] I had reason
to assume that GCC synthesizes "double-word" addition/subtraction on all
processors, not just on x86.
Since "double-word" comparision and shifts are likewise simple operations
I further assumed that GCC synthesizes them too on all processors.

What's the fundamental difference between subtraction and comparision?
Why does GCC generate calls to __[u]cmpDI2 for a simple comparision
instead to synthesize it?
And: as shown in libgcc2.c, "double-word" shifts can easily be synthesized
using "single-word" shifts plus logical OR on any target.
I expected GCC to synthesize these operations on non-x86 processors too,
just like "double-word" addition and subtraction.

A possible/reasonable explanation would be code size, i.e. if the synthesized
instructions need significantly more memory than the function call (including
the argument setup of course).

Stefan