Question about generic\time.c 2.4.17

Question about generic\time.c 2.4.17

[Nick Zajerko-McKee]

Hi,

I'm working on a new 4Kc platform and was looking at the
arch\mips\mips-boards\generic\time.c sources.  Can someone explain to me
the function of do_fast_gettimeoffset(), especially the do_div64_32()
assembler routine?  One of the requirements I have will be not modify
the timer resolution for my platform to something in the msec range w/o
disturbing the underlying jiffie setup found in linux.

Thanks.

Re: Question about generic\time.c 2.4.17

[Maciej W. Rozycki]

On 23 Jul 2002, Nick Zajerko-McKee wrote:

> I'm working on a new 4Kc platform and was looking at the
> arch\mips\mips-boards\generic\time.c sources.  Can someone explain to me
> the function of do_fast_gettimeoffset(), especially the do_div64_32()
> assembler routine?  One of the requirements I have will be not modify
> the timer resolution for my platform to something in the msec range w/o
> disturbing the underlying jiffie setup found in linux.

 That's a traditional double-precision division, i.e. in this case it's a
64-bit dividend by a 32-bit divisor with a 32-bit quotient and a 32-bit
remainder (hmm, the code should be obvious).  It isn't used in the file,
though. 

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: Question about generic\time.c 2.4.17

[Jun Sun]

Nick Zajerko-McKee wrote:
> Hi,
> 
> I'm working on a new 4Kc platform and was looking at the
> arch\mips\mips-boards\generic\time.c sources.  Can someone explain to me
> the function of do_fast_gettimeoffset(), especially the do_div64_32()
> assembler routine?  One of the requirements I have will be not modify
> the timer resolution for my platform to something in the msec range w/o
> disturbing the underlying jiffie setup found in linux.
> 

do_div64_32() emulate 64bit division with 32bit registers/values. 
do_fast_gettimeoffset() gives the intra-jiffy time offset, which yields higher 
timer resolution.

Take a look of /Documentation/mips/time.README.

Jun

RE: Question about generic\time.c 2.4.17

[Zajerko-McKee, Nick]

Thanks for the reply.   No, the code wasn't too obvious.  I went through the
gas info page to try to understand the inline assembler options + see mips
run.  I believe the code is used in the MIPS32 condition, which is what mode
I'm building for...  

so the result is res = (high |low)/ base ?

What had me confused was high and low are also modified as part of the
function.

-----Original Message-----
From: Maciej W. Rozycki [mailto:macro@ds2.pg.gda.pl]
Sent: Wednesday, July 24, 2002 10:38 AM
To: Nick Zajerko-McKee
Cc: linux-mips@oss.sgi.com
Subject: Re: Question about generic\time.c 2.4.17


On 23 Jul 2002, Nick Zajerko-McKee wrote:

> I'm working on a new 4Kc platform and was looking at the
> arch\mips\mips-boards\generic\time.c sources.  Can someone explain to me
> the function of do_fast_gettimeoffset(), especially the do_div64_32()
> assembler routine?  One of the requirements I have will be not modify
> the timer resolution for my platform to something in the msec range w/o
> disturbing the underlying jiffie setup found in linux.

 That's a traditional double-precision division, i.e. in this case it's a
64-bit dividend by a 32-bit divisor with a 32-bit quotient and a 32-bit
remainder (hmm, the code should be obvious).  It isn't used in the file,
though. 

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

RE: Question about generic\time.c 2.4.17

[Maciej W. Rozycki]

On Wed, 24 Jul 2002, Zajerko-McKee, Nick wrote:

> Thanks for the reply.   No, the code wasn't too obvious.  I went through the
> gas info page to try to understand the inline assembler options + see mips
> run.  I believe the code is used in the MIPS32 condition, which is what mode
> I'm building for...  

 For the 32-bit mode, not necessarily on a MIPS32 processor. 

> so the result is res = (high |low)/ base ?

 Strictly speaking, res = (high:low) / base and the result is (high:low) %
base.  That's a macro, hence a bit weird semantics (two results are
actually provided), but it makes the use easier.  A few architectures
provide such an operation in hardware.

> What had me confused was high and low are also modified as part of the
> function.

 That's just how the algorithm works.  These are local variables anyway.

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +