dmd codegen improvements

[Dmitry Olshansky via Digitalmars-d]

On 18-Aug-2015 13:45, Walter Bright wrote:
> Martin ran some benchmarks recently that showed that ddmd compiled with
> dmd was about 30% slower than when compiled with gdc/ldc. This seems to
> be fairly typical.
>
> I'm interested in ways to reduce that gap.

..

> 2. instruction selection patterns like should one generate:
>
>      SETC AL
>      MOVZ EAX,AL
>
> or:
>      SBB EAX
>      NEG EAX
>

See section "Problematic instructions" here:
http://www.agner.org/optimize/optimizing_assembly.pdf

And some invaluable material on each CPU specifics for all x86 from 
Pentium to Haswell and AMD from K6 toBuldozer:

http://www.agner.org/optimize/microarchitecture.pdf

Hope this helps.

-- 
Dmitry Olshansky