This is the mail archive of the java@gcc.gnu.org mailing list for the Java project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Anyone else sees a bootstrap failure (Linux-x86)?


Ranjit Mathew writes:
 > >  > On Solaris 8, I did see an improvement with "-pipe", but it
 > >  > was always a ~5% improvement.
 > >
 > >This is surprising.  You don't say what filesystem you were using.
 > 
 > It's UFS without journalling, AFAICT.
 > 
 > But I still don't fully comprehend why "-pipe" cannot be faster -
 > at some point in time, the OS's filesystem buffers will get full
 > and will have to be flushed out to the disc, incurring I/O
 > penalties, during "normal" compilation.

It depends on how the /tmp filesystem behaves.  It can be designed so
that the creation, use, and subsequent deletion of a temporary file
never results in any I/O activity at all.

 > If the compiler outputs assembly code in one large burst (or a few
 > large bursts) and the assembler reads its input in one large burst
 > (or a few large bursts), the interlocking effect in "-pipe" that
 > you talked about wouldn't be so pronounced.

Right, but the cost of creating, reading, and writing a file in /tmp
is also (or should be, at least) very small.

 > It's not very black and white to me still.

Me neither.  At least on a well-designed OS we know the effect should
be very small.  5% seems to me rather a lot.  Something to do with
Solaris, perhaps.

Andrew.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]