[Fwd: Re: [PATCH] MIPS32 DSP instructions again]

[David Daney]

Paul Koning wrote:
>>>>>>"Dominic" == Dominic Sweetman <dom@mips.com> writes:
> 
> 
>  Dominic> Nigel forwarded me your comments on why an assembler should
>  Dominic> be passed a flag and on the basis of that flag reject
>  Dominic> mnemonics from the DSP ASE:
> 
>  Ian> I disagree.  I think it is appropriate and useful for the
>  Ian> assembler to be able to correctly accept or reject instructions
>  Ian> based on the command line options and .set options.
> 
>  Dominic> And Paul...
> 
>  >> Agreed.
> 
>  Dominic> Which comes down to this: under what circumstances might it
>  Dominic> be helpful for the assembler to reject a piece of code which
>  Dominic> it could have assembled?
> 
> If you tell the compiler that it should generate code for, say, R3000,
> but by some oversight you stuck in some embedded asm code with opcodes
> not valid on that architecture, you want the assembler to tell you
> that.  It's far bettter to get a compile time error than a crash at
> runtime that may be hard to reproduce.
> 
> In other words when I tell the assembler "I'm writing code for
> architecture X" I want the assembler to do TWO things:
> 
> 1. Accept all constructs that are valid in X.
> 2. Reject all constructs that are NOT valid in X.
> 

I was going to say the same thing, but you did it first.

Speaking as a person who works with MIPS32 without the DSP ASE every 
day, I would would prefer not to have to find the problems at runtime.

Just my $0.02

David Daney