Re: [Boost-build] Suggestions for additional features / options for compiler toolsets: special instructions and floating point

From: John Pavel (jrp_at_[hidden])
Date: 2007-12-01 17:43:40

On 1 Dec 2007, at 18:34, Vladimir Prus wrote:

> On Friday 16 November 2007 21:44:04 John Pavel wrote:
>> Just as it is possible to invoke, eg, async exceptions by means of
>>
>> project
>>  : requirements
>>    <toolset>msvc:<asynch-exceptions>on
>>  ;
>>
>> it would be helpful to be able to set a few other flags, notably for
>> different types of special instruction and floating point handling:
>>
>>
>> More specifically, could we invoke the use of SSE instructions?
>>
>> In gcc, this is done by the -mssse, -mssse2, etc, flags.  gcc 4.3.0  
>> also
>> provides
>> . Support for SSSE3 built-in functions and code generation are  
>> available via
>> -mssse3 (which also comes with the core2 architecture)
>> . Support for SSE4.1 built-in functions and code generation are  
>> available
>> via -msse4.1.
>> . Support for SSE4.2 built-in functions and code generation are  
>> available
>> via -msse4.2.
>>
>> (see http://gcc.gnu.org/gcc-4.3/changes.html)
>>
>> VC++ has the /arch option (see
>> http://msdn2.microsoft.com/en-gb/library/7t5yh4fd(VS.90).aspx)  
>> Intel has
>> similar.
>>
>>
>> Could we also invoke different types of floating point handling:
>>
>> For recent versions of VC++ (and Intel compilers) we have /fp: 
>> [precise |
>> except[-] | fast | strict ] (see
>> http://msdn2.microsoft.com/en-gb/library/e7s85ffb(VS.90).aspx)
>>
>> Gcc has a similar, if more finely grained, set of options.
>>
>> These options would be even more useful if combined with the  
>> ability to time
>> the running of tests, as it would allow the user to test the impact  
>> of using
>> faster v more accurate fp handling.
>
> I'm not sure. Are those options present in other compilers, in  
> basically
> same form. It's probably not worth adding a feature if it's easy to
> accomplish using explicit options, and there are just two compilers  
> of interest.

They are also available in the Intel compilers.  They ought to be  
properly integrated with the cpu/architecture options that are being  
discussed in a parallel thread.