The warning arises from the removal of the barrier before peephole
optimization. An operand of the barrier is now kept by a new be_Keep
which is scheduled directly after the operand.

No functional change.

[r28460]

This simplifies the code and removes some divergencies between the CSE and
combo.

[r28459]

[r28458]

[r28457]

[r28454]

[r28453]

[r28447]

Difference of idx for Calls with no order relation is stable,
but may lead to circular dependencies. Fixed by looking at
the heights first.

Fixes backend/transform_loop.c (with -O0)

[r28436]

[r28435]

If a shift operation has a dependency operand (64bit shifts),
then instruction selection inserts a dependency edge between them,
so the scheduler orders them correctly. However, if the operand
is not used, the scheduler is confused about the dependency edge, which
is never resolved.

Fixes opt/longlong.c testcase

[r28434]

Respect block dominance in localopt. Node a may not be live in b's block.

[r28430]

[r28426]

Without parsing the text of a ASM node it is not possible to decide whether two
ASM nodes are congruent. Don't even try it.

This fixes testsuite/opt/comboattrs.c

[r28422]

[r28421]

Extend the fix_adds_and_subs pass after strength reduction, which
inserts Convs to make the verifier happy. In case the index
variable is inverted (i=1-i) on each loop iteration, the optimization
may yield a SubP(ConstIu,PhiP), which is rejected by the verifier.

[r28409]

Some iterations must be done backwards.

[r28401]

Introduce a get_irn_resmode() and use it at all places in combo where the
part of the opcode is meaned.

[r28400]

The backend sometimes creates ProjM with a temporarily broken pn,
so the verifier must not check them.

[r28392]

[r28391]

[r28387]

reworked ia32 condition code handling, remove backend Bound support (really someone should lower it before the backend

[r28386]

This optimization is unsafe and breaks Load-Store-optimization.

It is not enough that both loads are connected by the memory edge,
if one Load has an exception-edge, they cannot be parallelized, because
this would destroy exception order.
load-store-opt handle this case and many others, but expect all
Loads in order yet.
opt_parallelize_mem() does this optimization in a wider range.

[r28385]

All leaders of a partition must have the same opcode, which is the pair
of <irn_opcode, ir_mode> in Firm.

[r28384]

[r28383]

[r28382]

[r28381]

[r28379]

[r28378]