sdw@hpsad.HP.COM (Steve Warwick) (02/07/91)
I just got done talking to a vendor of Digital Signal processors who mentioned that their company is involved with a group trying to define something called "numerical C" which defines a set of special types and operatiors which are supposedly tuned to typical DSP/Vector processing applications. I was troubled by this in that it appears that language constructs available in C++ provedes syntax which could cover these cases, such as fractional integer arithmetic, and vectorized expressions. In addition, such modules could be supported on non-DSP platforms given a suitable supprt library. My question is: are there members of the C++ community aware of this other standardization effort, and can point out how the two efforts are unable to be combined... is there some fundamental weakness in the ability of a C++ compiler to optimise for the DSP environment?
steve@taumet.com (Stephen Clamage) (02/09/91)
sdw@hpsad.HP.COM (Steve Warwick) writes: >I just got done talking to a vendor of Digital Signal processors >who mentioned that their company is involved with a group trying >to define something called "numerical C" ... >Are there members of the C++ community aware of this >other standardization effort, and can point out how the two efforts >are unable to be combined... is there some fundamental weakness in the >ability of a C++ compiler to optimise for the DSP environment? Members of the ANSI C committe and of the Numerical C Extensions Group sit on the ANSI C++ committee. So, yes, members of the C++ community are aware of other standards efforts. The C++ committee has no desire to introduce gratuitous incompatibilities with C or with numerical requirements. On the other hand, it does not seem appropriate for the committee to tailor language features to the peculiarities of any one machine architecture. A feature which works nicely with machine ABC may not be implementable on machine DEF, and may be the reverse of a desirable construct on machine GHI. And how desireable will this feature appear, which all language implementations must support, when machine ABC is no longer manufactured? -- Steve Clamage, TauMetric Corp, steve@taumet.com
jimad@microsoft.UUCP (Jim ADCOCK) (02/12/91)
In article <1920001@hpsad.HP.COM> sdw@hpsad.HP.COM (Steve Warwick) writes: |I just got done talking to a vendor of Digital Signal processors |who mentioned that their company is involved with a group trying |to define something called "numerical C" which defines a set of |special types and operatiors which are supposedly tuned to |typical DSP/Vector processing applications. | | I was troubled by this in that it appears that language constructs |available in C++ provedes syntax which could cover these cases, such |as fractional integer arithmetic, and vectorized expressions. In |addition, such modules could be supported on non-DSP platforms given a |suitable supprt library. I agree, I share this concern. See, I believe, December "Journal C Language Translation" for a discussion of "restricted" pointer types [better know as 'son of noalias'] A better solution, I believe, is for C++ to clean up references, and then to define clearly what exactly is being aliased by various types of pointers and references.