On Friday, 18 July 2014 at 04:40:52 UTC, dennis luehring wrote:
Am 18.07.2014 04:52, schrieb Walter Bright:
On 7/16/2014 7:21 AM, dennis luehring wrote:
can you give an short (working) example code to show the
different resulting
assembler for your for-rewrite example - and what compilers
your using for
testing - only dmd or gdc?
I used dmd.
i sometimes got the feeling that you underestimate the sheer
power of todays clang or gcc optimizers - so partly what
gdc/ldc can do with your code
reminds me of brian schotts exmaple of his sse2 optimized
version of his lexer - the dmd generated was much faster then
the normal version, but gdc/ldc results of the unoptimized
versions are still 50% faster
i understand your focus on dmd - but talking about fast code
and optimizing WITHOUT even trying to compare with other
compiler results is just a little bit strange for someone who
stated speed = money
I think this somewhat misses the point of the example, which I
would say was - in academia-speak - purely illustrative.
The point still stands, which is "unless you understand the
compiler and the architecture, stop trying to second guess
performance on the micro-level"