Re: [PATCH 2/6] b43: N-PHY: add RSSI functions: poll and set 2055 vcm

[Larry Finger <Larry.Finger@lwfinger.net>]

On 01/10/2010 04:32 PM, Michael Buesch wrote:
> On Sunday 10 January 2010 23:13:20 Rafał Miłecki wrote:
>> +		buf[0] += (s8)(((s[0] & 0x3F) << 2) >> 2);
>> +		buf[1] += (s8)((((s[0] >> 8) & 0x3F) << 2) >> 2);
>> +		buf[2] += (s8)(((s[1] & 0x3F) << 2) >> 2);
>> +		buf[3] += (s8)((((s[1] >> 8) & 0x3F) << 2) >> 2);
> 
> I suggest buf[3] += (s8)((((s[1] >> 8) & 0x3F) << 2) >> 2) << 2) >> 2) << 2) >> 2) << 2) >> 2) << 2) >> 2) << 2) >> 2) << 2) >> 2)
> ;)
> No, seriously, why shift left and then shift right? Is this a translation error?
> I _guess_ it's some mistranslation of the sign extension going on.
> Or alternatively a compiler going insane on sign extension.
> 
> The question is: Do we want these integers to be signextended or not?
> 
> What we currently do is this:
> buf[3] += (s8)((s[1] >> 8) & 0x3F);
> 
> which will always result in a positive 8bit integer, as far as I can see.
> Which smells fishy.

We do want sign extension of the signed 6-bit quantities into an 8-bit word. The
translation is correct. Do you know of a better way to extend the signs?

Larry