brooks@lll-crg.ARPA (Eugene D. Brooks III) (03/17/85)
For those who think that aligment is not important on a VAX I suggest your
actually trying a code without alignment before makeing any claims that it
is not important. The very simple example
for(i = 0; i < NTIMES; i += 1) {
fa = fb + fc;
da = db + dc;
ia = ib + ic;
}
Where the variables f, d, i are floats, doubles and ints in static data
has a speed reduction of a factor of two if you take out the .align pseudos
and padd in some bytes to break alignment. Alignmet if very important
as far as performance is concerned.
For those who want to read structures from files using a single read
the best course to follow is write the stuff into the file using the
same structure. The padding for alignment is automatically taken care of.
If the padding cant be tolerated then the most portable course of action
is to read each data item into the structure seperately. This will save you
a lot of headaches when moving your code from one machine to another.ron@brl-tgr.ARPA (Ron Natalie <ron>) (03/19/85)
> For those who think that aligment is not important on a VAX I suggest your > actually trying a code without alignment before makeing any claims that it > is not important. The very simple example > You're twisting what I was saying. What I was complaining about was that there was no feature for defeating it even on machines like the VAX where the hardware will support it easily. It would be nice to read structures from other machines (like PDP-11's) sometime. -Ron
ndiamond@watdaisy.UUCP (Norman Diamond) (03/20/85)
> For those who want to read structures from files using a single read > the best course to follow is write the stuff into the file using the > same structure. The padding for alignment is automatically taken care of. > If the padding cant be tolerated then the most portable course of action > is to read each data item into the structure seperately. This will save you > a lot of headaches when moving your code from one machine to another. So if your data files are supposed to be accessible by programs in any other language, or by programs in the same language compiled by a different compiler -- on the same machine and/or on other machines -- then you should read each data item separately. If you're writing a network interface, don't dare read N bytes of header into a structure. Read each byte separately. Hope your code can respond in time. If you have a bunch of old files and a new compiler, don't just copy your files -- convert them and convert the padding. How can one source program adapt a structure to different padding for the output file than for the input file? By coding in assembly? Yes, padding really has some disadvantages. There should be an OPTION to disable it. -- Norman Diamond UUCP: {decvax|utzoo|ihnp4|allegra}!watmath!watdaisy!ndiamond CSNET: ndiamond%watdaisy@waterloo.csnet ARPA: ndiamond%watdaisy%waterloo.csnet@csnet-relay.arpa "Opinions are those of the keyboard, and do not reflect on me or higher-ups."
mwm@ucbtopaz.CC.Berkeley.ARPA (03/21/85)
In article <9339@brl-tgr.ARPA> ron@brl-tgr.ARPA (Ron Natalie <ron>) writes: >> For those who think that aligment is not important on a VAX I suggest your >You're twisting what I was saying. What I was complaining about was that >there was no feature for defeating it even on machines like the VAX where >the hardware will support it easily. It would be nice to read structures >from other machines (like PDP-11's) sometime. Ron is right, C needs a "packed" struct type. I.e. - static packed struct gezornin { <objects which will not have alignment> ; } ; There: by careful choice of wording, I've managed to insure that such a feature will *never* appear in C :-). <mike
alexis@reed.UUCP (Alexis Dimitriadis) (03/24/85)
In article <9339@brl-tgr.ARPA> ron@brl-tgr.ARPA (Ron Natalie <ron>) writes: > What I was complaining about was that >there was no feature for defeating it even on machines like the VAX where >the hardware will support it easily. It would be nice to read structures >from other machines (like PDP-11's) sometime. > >-Ron So C needs something like the Pascal keyword `packed'. (hee hee) -- _______________________________________________ alexis @ reed ...ihnp4!{harvard|tektronix}!reed ...decvax!tektronix!reed ...teneron!reed
ndiamond@watdaisy.UUCP (Norman Diamond) (03/26/85)
*** ALIGN THIS MESS WITH YOUR REPLACEMENT ***
--
Norman Diamond
UUCP: {decvax|utzoo|ihnp4|allegra}!watmath!watdaisy!ndiamond
CSNET: ndiamond%watdaisy@waterloo.csnet
ARPA: ndiamond%watdaisy%waterloo.csnet@csnet-relay.arpa
"Opinions are those of the keyboard, and do not reflect on me or higher-ups."brooks@lll-crg.ARPA (Eugene D. Brooks III) (03/27/85)
> >the hardware will support it easily. It would be nice to read structures > >from other machines (like PDP-11's) sometime. > So C needs something like the Pascal keyword `packed'. (hee hee) A packed structure type would only give portability between a few machines. Although I would never want a machine with a 36 bit word they do exist and some of them have C compilers. The portability problem is best handled with a subroutine or macro to read the structure. Using a 'packed' structure introduces a speed penalty when the structure is used elsewhere and a hole host of other troubles just for the sake of giving limited I/O portability. I can't see any sense in this. This is against the philosophy of the language. The language has no I/O at all. Lets keep the I/O in routines where it belongs. If you want to read a structure from a pdp11 write a routine or a macro. It appears to me that packed structures are even jumping the gun. Does the C language even define the order in which the elements of a structure are placed in memory? The chapter on structures in K&R only states that the structure is a grouping of data items and I failed to find any statements about ordering of the elements. Am I mistaken about this?