[PATCH] b43-asm: Add 3 new virtual instructions.

[francesco.gringoli at ing.unibs.it]

On Nov 12, 2011, at 8:16 PM, Michael Büsch wrote:

> On Sat, 12 Nov 2011 20:03:36 +0100
> francesco.gringoli at ing.unibs.it wrote:
> 
>> On Nov 12, 2011, at 7:05 PM, Michael Büsch wrote:
>> 
>>> On Sat, 12 Nov 2011 18:31:21 +0100
>>> francesco.gringoli at ing.unibs.it wrote:
>>> 
>>>> 	orxh	(r1 << 8) & 0x0100, r2 & ~0x0100, r2
>>> 
>>> This is not really going to fly. If you want this highlevel stuff, you
>>> should port a C compiler to the architecture.
>>> This is assembly. It doesn't know about reg<<imm or similar stuff.
>> Yes, you are right and, in fact, the patched assembler will just accept only what the cpu may execute, I does not pretend to be a C compiler. It's just another way for assembling "or with shift and select" (or jzx), and this way really enhances the readability of the assembly code. Give it a try :-).
>> 
>> If I'm not wrong, it's like "mov 0x1234, r1"
> 
> No.
> It crosses the line where it does (pseudo)operations (like shift, and, or, etc..)
> on non-const (non-immediate) stuff.

> Just do a preprocessor or something like that, that translates your pseudo-insns
> to real insns. Alternatively port a small C compiler.
> 
> I also don't think that this is easier to read for people familiar to the CPU. And
> you have to be familiar to the CPU when writing code for it.
I initially started with a pre-processor but I soon realized that those three instructions were the only one whose readability could have been improved by the preprocessor itself.

> I won't merge this. No way.
Ok, anyway I urge to share with other researchers  a new version of the standard ucode and for the moment I don't have more time to spend on the "preprocessor". I'm sorry but the new ucode source will require the patched assembler. As soon as I have finished some work on the N-PHY firmware I will come back to the preprocessor.

> The "/* duplicate some complex_imm rules to avoid parentheses" part also is not
> really merge-able as-is. yacc knows about operator precedence. If you want operator
> precedence (to get rid of parenthesis), just use this yacc feature to implement this.
Yes, I did it this way because "[reg|mem] << imm" is not assembly (as you say :-) ) and the assembler strictly checks that such syntax is used only with "orxh".

Regards,
-Francesco

> 
> -- 
> Greetings, Michael.