GitLab

To improve performance of the RTS when dynamically linked on x86, I
previously disabled -fPIC for certain critical modules (the GC, and a
few others).  However, build reports suggest that the dynamic linker
on OS X doesn't like this, so I'm disabling this optimsation on that
platform.

'import' syntax is seperate from ':module' syntax

We no longer support building with a compiler that doesn't come with
base 4.

Noticed by Henrique Ferreiro <hferreiro@udc.es>, thanks!

LLVM supports creating pointers in two ways, firstly through
pointer arithmetic (by casting between pointers and ints)
and secondly using the getelementptr instruction. The second way
is preferable as it gives LLVM more information to work with.

This patch changes a lot of pointer related code from the first
method to the getelementptr method.

Instead of calling the C library for these Cmm functions
we use intrinsic functions provided by llvm. LLVM will
then either create a compile time constant if possible, or
use a cpu instruction or as a last resort call the C
library.

At the moment this gives a very slight performance boost of around 1 - 2%.
Future changes to the generated code though so that pointers are kept as
pointers more often instead of being cast to integer types straight away
should hopefully improve the benefit this brings.

Validation fails when validating with PAPI support (i.e. GhcRtsWithPapi  = YES
in validate.mk).  The problem is that the posix symbols are defined by a header
included from papi.h. Compilation then fails because these symbols are
redefined in PosixSource.h. 

This patch adds an undefine for the posix symbols after including papi.h and
before including PosixSource.h. The #undefines are localized to Papi.c since
that is the only case where they are getting defined twice.

 computeLiveness requires the SCCs of blocks to be in reverse dependent
 order, and if they're not it was silently giving bad liveness info, 
 yielding a bad allocation.

 Now it complains, loudly.

 * I've pushed the SPILL and RELOAD instrs down into the
   LiveInstr type to make them easier to work with. 

 * When the graph allocator does a spill cycle it now just
   re-annotates the LiveCmmTops instead of converting them 
   to NatCmmTops and back. 

 * This saves working out the SCCS again, and avoids rewriting
   the SPILL and RELOAD meta instructions into real machine
   instructions.

If there are are unreachable basic blocks in the native code then the 
linear allocator might loop. Detect this case and panic instead.

The problem is that showing SDoc's looks at the static flags global
variables, but those are panics while we are parsing the static flags.
We work around this by explicitly using a fixed prettyprinter style.

Before all the stg registers were simply a bit type or
floating point type but now they can be declared to have
a pointer type to one of these. This will allow various
optimisations in the future in llvm since the type is
more accurate.