GitLab

If we are cross-compiling, we assume that if timer_create exists then
it works.

All supported bootstrapping compilers now have --info output,
so we can use that unconditionally.

We no longer actually use ITIMER_VIRTUAL anyway

In particular, avoid referring the user to GHC.Prim; they generally
ought to use GHC.Exts instead.

It's not really a primop, and GHC.Prim doesn't export it. It has a
definition in GHC.Magic.

This reverts the compiler parts of

    commit 7b594a5d
    Author: David Terei <davidterei@gmail.com>
    Date:   Mon Nov 21 12:05:18 2011 -0800

    Remove registerised code for dead architectures: mips, ia64, alpha,
    hppa1, m68k

In particular, we want to know whether bewareLoadStoreAlignment should
return True or False for them.

It also reverts

    commit 3fc68b5c
    Author: Simon Marlow <marlowsd@gmail.com>
    Date:   Wed Jan 4 11:44:02 2012 +0000

    Remove missing archs (mipseb, mipsel, alpha) (#5734)

    It doesn't hurt to map these to ArchUnknown since we don't need to
    know anything specific about them, and adding them would be a pain
    (there are a bunch of places where we have to case-match on all the
    arches to avoid warnings).

When offsetof is defined, we use that. This avoids "variably
modified at file scope" warnings/errors with recent gccs.

All uses of it cast the result anyway. However, DeriveConstants needs
it to not include the cast, as (void *) casts can't be used in constant
expressions.

nus in #ghc ran into a problem where the wrong filepath was being
stripped.

The Slow calling convention passes the closure in R1, but we were
ignoring this and hoping it would work, which it often did.  However,
this bug seems to have been the cause of #7192, because the
graph-colouring allocator is more sensitive to having correct liveness
information on jumps.

Fixes build on OS X

DeriveConstants.hs works in a cross-compilation-friendly way. Rather
than running a C program that prints out the constants, we just compile
a C file which has the constants are encoded in symbol sizes. We then
parse the output of 'nm' to find out what the constants are.

Based on work by Gabor Greif <ggreif@gmail.com>.

This fixes a CmmLint complaint when doing proc-point splitting.

This removes the OldCmm data type and the CmmCvt pass that converts
new Cmm to OldCmm.  The backends (NCGs, LLVM and C) have all been
converted to consume new Cmm.

The main difference between the two data types is that conditional
branches in new Cmm have both true/false successors, whereas in OldCmm
the false case was a fallthrough.  To generate slightly better code we
occasionally need to invert a conditional to ensure that the
branch-not-taken becomes a fallthrough; this was previously done in
CmmCvt, and it is now done in CmmContFlowOpt.

We could go further and use the Hoopl Block representation for native
code, which would mean that we could use Hoopl's postorderDfs and
analyses for native code, but for now I've left it as is, using the
old ListGraph representation for native code.

Note that PRIdPTR is considered as linux-ism so it's not available on platforms
like Solaris, although some other free Unix(-like) OSes apparently supports
it too.

Based on a patch from markwright in #3072.