Re: Using local storage pools...

[Georg Bauhaus <rm-host.bauhaus@maps.futureapps.de>]

On 2/27/11 9:18 AM, Pascal Obry wrote:
> Le 27/02/2011 05:16, anon@att.net a écrit :
>> 1.2 The GNAT Compiler
>> ...
>> "In order to bridge the semantic gap between Ada and C, several GCC code
>> generation routines have been extended, and others added, so that the
>> burden of translation is also assumed by GIGI and GCC whenever it is
>> awkward or inefficient to perform the expansion in the front-end. For
>> example, there are code generation actions for exceptions, variant parts
>> and accesses to unconstrained types. As a matter of GCC policy, the code
>> generator is extended only when the extension is likely to be of benefit
>> to more than one language."
>>
>> The first sentence suggest the conversion is from Ada to C. And extending
>
> Wrong again. GCC is GNU Compiler Collection it is not a C compiler. So GCC code generation means just that, the middle-end and backend of GCC (shared with the C, C++ frontends and others) has been extended.
>
> Pascal.

I understand one claim is that GCC's infrastructure
had originally been made with sequential {}-languages in mind?
So that GCC's infrastructure won't be best for a concurrent language
such as Ada.
Consequently, that GNAT might actually profit from any changes
to GCC infrastructure in case they'll change it in order to support
the new C++ with its concurrency libraries.

Is it true at all that GCC's internals are biased towards "sequential"?
Hasn't even the C part grown with POSIX threads and corresponding
needs?  Grown with GCC being used as the system compiler of a major
commercial server operating system (GNU/Linux) written in C and
being a multitasking OS?

Is it true at all that GCC's infrastructure is unaware of parallelism?
GCC supports many of the primitive "concurrency instructions"
like read-and-increment or test-and-set.  So I don't think it is
true that today's GCC infrastructure neglects concurrency.

Is it likely that Apple could have developed Grand Central
Dispatch for the Objective-C part of GCC if using its backend
had been in the way ofthe concurrency goal?
They are switching to LLVM and CLang, though.

Curiously, the parallelism built-ins are readily available
to languages such as C and, to a lesser extent, to Ada.
Fortunately, recent GNAT makes some of these available, like the
SIMD instructions  for processors made by AMD and Intel.