RE: Parallel make Order Algorithm

[Martin Dorey]

> I have the feeling that the second algorithm has a bottleneck:

The algorithm you describe is a feature of your makefile(s) rather than make.  
Make doesn't really know about the traditional three phases of a large C 
project that you describe.  It's just as happy driving an assembler, preparing 
ingredients for a cake or generating documentation.

-----Original Message-----
From: address@hidden [mailto:address@hidden On Behalf Of Dilyan Palauzov
Sent: Tuesday, July 14, 2015 05:02
To: address@hidden
Subject: Parallel make Order Algorithm

Hello,

when I run (serial) make, it first compiles the objects files for the 
first library, then links the fist library, then it compiles the object 
files for the second library, then links the second library.  Finally, 
for each executable, it compiles the object files and links.

When I run parallel make -j 4, it first compiles all object files (for 
the libraries and for the executables), then it links all libraries and 
then it links the executables.

I have the feeling that the second algorithm has a bottleneck:

While compiling object files is always possible, linking of programmes 
can start, after  the dependent libraries are linked.  So in the moment, 
when all object files are build, and the single library, on which all 
executables depend, is being linked, parallel make executes only one 
job: linking the library.

I have not measured this with frozen processes, but when I check the 
output of make, and the logic behind, i guess the above scenario can happen.

Can you confirm it?

If yes, I propose also in parallel builds to keep the same logic, as 
with serial builds, with the addition to compile object files, just in 
case no requirements are currently satisfied, to start linkage.  Or with 
other words, first try to build the final leaves (bin, lib, info) and in 
case this cannot be started now for a library, build the first level 
(object files) for the next final target(library) as temporary measure, 
until the requirements for the linkage are satisfied.

Anyway, I am sure that if see the problem exist, you can also find a 
nice solution for it.

Greetings
   Dilian

_______________________________________________
Bug-make mailing list
address@hidden
https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.gnu.org_mailman_listinfo_bug-2Dmake&d=BQICAg&c=DZ-EF4pZfxGSU6MfABwx0g&r=oBMzc8Omr1YTgjig4n4076T3IKL7TuNH9HpVbojD-ms&m=AlBhQEXGb71rWv9skDavAC0iZGD2sFNv1ULX9JOaXFY&s=-BYdicpvPnRuXDkJIjq7H43Mg5-u4hfkQvH0hfk7p5M&e=