On Sun, 22 Mar 2009 22:09:53 -0400, Robert Jacques <sandf...@jhu.edu>
wrote:
On Sun, 22 Mar 2009 21:55:04 -0400, Brian <digitalm...@brianguertin.com>
wrote:
"Windows and Linux Not Well Prepared For Multicore Chips"
http://developers.slashdot.org/article.pl?sid=09/03/22/193205&from=rss
I wouldn't completely agree that operating systems aren't ready for such
things.
One of the comments reads: "This is the kind of the the compiler could
do
just fine ... by isolating parts of the code in which there are no
dependencies in the data-flow, and which could therefore run in
parallel..."
This sounds like what "pure" functions are going to be for, right?
Yes and No. No, what they are talking about is a parallelizing compiler.
i.e. one that performs automatic parallelization / code analysis, which
for the most part doesn't work because it has no grantees. And yes, pure
is all about giving the complier the grantees it needs to actually
automatically parallelize the code.
Okay, next time I should read TFA before posting. What the author is
talking about isn't pure or parallelizing compilers. What he's actually
talking about is adding parallel instructions to the CPU. (The focus was
on Itanium (which exchanged out-of-order execution for compiler controlled
parallel instructions), though it could easily apply to the next SSE
instruction set (specifically the one coming with larrabee) which
primarily adds conditional tests and an execution mask. These allow for
SIMD flow control (and have been in GPUs for a while) )
