| Age | Commit message (Collapse) | Author |
|---|
|
|
|
Due to a bug in Dialyzer, unknown types have been introduced.
|
|
hipe_regalloc_loop considers SpillLimit to be an inclusive lower bound,
the allocators considered it to be an exclusive lower bound. The
allocators are changed to also consider it an inclusive lower bound.
This caused the register allocators to occasionally spill the first
"unspillable" temporary. This caused a failure in a newly added
assertion when hipe-compiling dets_v9 on x86.
|
|
These pseudo instructions are added to all backends and allow spill slot
to spill slot move coalescing in a clean way.
They have regular move semantics, but contain an additional scratch
register to be used if both source and destination are spilled, and can
not be move coalesced.
Additionally, a register allocator callback
Target:is_spill_move(Instr, Context) is added which allows the spill
slot allocators to check for these instructions and try to coalesce the
spill slots the two temporaries are allocated to.
|
|
hipe_range_split is a complex live range splitter, more sophisticated
thatn hipe_restore_reuse, but still targeted specifically at temporaries
forced onto stack by being live over call instructions.
hipe_range_split partitions the control flow graph at call instructions,
like hipe_regalloc_prepass. Splitting decisions are made on a per
partition and per temporary basis.
There are three different ways in which hipe_range_split may choose to
split a temporary in a program partition:
* Mode1: Spill the temp before calls, and restore it after them
* Mode2: Spill the temp after definitions, restore it after calls
* Mode3: Spill the temp after definitions, restore it before uses
To pick which of these should be used for each temp×partiton pair,
hipe_range_split uses a cost function. The cost is simply the sum of the
cost of all expected stack accesses, and the cost for an individual
stack access is based on the probability weight of the basic block that
it resides in. This biases the range splitter so that it attempts moving
stack accesses from a functions hot path to the cold path.
hipe_bb_weights is used to compute the probability weights.
mode3 is effectively the same as what hipe_restore_reuse does. Because
of this, hipe_restore_reuse reuses the analysis pass of
hipe_restore_reuse in order to compute the minimal needed set of spills
and restores. The reason mode3 was introduced to hipe_range_split rather
than simply composing it with hipe_restore_reuse (by running both) is
that such a composition resulted in poor register allocation results due
to insufficiently strong move coalescing in the register allocator.
The cost function heuristic has a couple of tuning knobs:
* {range_split_min_gain, Gain} (default: 1.1, range: [0.0, inf))
The minimum proportional improvement that the cost of all stack
accesses to a temp must display in order for that temp to be split.
* {range_split_mode1_fudge, Factor} (default: 1.1, range: [0.0, inf))
Costs for mode1 are multiplied by this factor in order to discourage
it when it provides marginal benefits. The justification is that
mode1 causes temps to be live for longest, thus leading to higher
register pressure.
* {range_split_weight_power, Factor} (default: 2, range: (0.0, inf))
Adjusts how much effect the basic block weights have on the cost of a
stack access. A stack access in a block with weight 1.0 has cost 1.0,
a stack access in a block with weight 0.01 has cost 1/Factor.
Additionally, the option range_split_weights chooses whether the basic
block weights are used at all.
In the case that the input is very big, hipe_range_split automatically
falls back to hipe_restore_reuse only in order to keep compile times
under control. Note that this is not only because of hipe_range_split
being slow, but also due to the resulting program being slow to register
allocate, and is not as partitionable by hipe_regalloc_prepass.
hipe_restore_reuse, on the other hand, does not affect the programs
partitionability.
The hipe_range_split pass is controlled by a new option ra_range_split.
ra_range_split is added to o2, and ra_restore_reuse is disabled in o2.
|
|
Adds a new register allocator callback
Target:branch_preds(Instr, Context) which, for a control flow
instruction Instr, returns a list of tuples {Target, Probability} for
each label name Target that Instr may branch to. Probability is a float
between 0.0 and 1.0 and corresponds to the predicted probability that
control flow branches to the corresponding target. The probabilities may
sum to at most 1.0 (rounding errors aside). Note that a sum less than
1.0 is valid.
|
|
hipe_restore_reuse is a simplistic range splitter that splits temps that
are forced onto the stack by being live over call instructions. In
particular, it attempts to avoid cases where there are several accesses
to such stack allocated temps in straight-line code, uninterrupted by
any calls. In order to achieve this it splits temps between just before
the first access(es) and just after the last access(es) in such
straight-line code groups.
The hipe_restore_reuse pass is controlled by a new option
ra_restore_reuse.
ra_restore_reuse is added to o1.
|
|
In addition to the temporary name rewriting that hipe_regalloc_prepass
does, range splitters also need to be able to insert move instructions,
as well as inserting new basic blocks in the control flow graph. The
following four callbacks are added for that purpose:
* Target:mk_move(Src, Dst, Context)
Returns a move instruction from the temporary (not just register
number) Src to Dst.
* Target:mk_goto(Label, Context)
Returns a unconditional control flow instruction that branches to the
label with name Label.
* Target:redirect_jmp(Instr, ToOld, ToNew, Context)
Modifies the control flow instruction Instr so that any control flow
that would go to a label with name ToOld instead goes to the label
with name ToNew.
* Target:new_label(Context)
Returns a fresh label name that does not belong to any existing block
in the current function, and is to be used to create a new basic
block in the control flow graph by calling Target:update_bb/4 with
this new name.
|
|
|
|
|
|
|
|
|
|
This allows us to pass around the context data that
hipe_regalloc_prepass needs cleanly, without using process dictionary or
parameterised modules (like it was previous to this change).
|
|
This is sound because the liveness data structure only stores liveness
info at basic block boundaries, and the rewrites that happen in
TargetSpecific:check_and_rewrite/2 preserves all existing definitions
and uses, and all new liveness intervals, belonging to newly introduced
temporaries, are always local to a basic block, and thus do not show up
in the liveout or livein sets for the basic block.
|
|
The division into an initial pass that may introduce temps, and
following passes that must not forces us to make the same heuristic
decision during each of these passes. Thus, the splitting heuristic
can't be based on the number of temporaries -- at least without
excluding temporaries above SpillLimit.
|
|
If temps introduced by hipe_regalloc_prepass end up above SpillLimit,
the register allocators will not spill them. This constraint is
unnecessarily limiting the allocators and might theoretically lead to
unallocatable programs (more temps above SpillLimit alive at a time than
there are physical registers).
|
|
|
|
These will not only be useful for hipe_regalloc_prepass, but also, after
the introduction of a mk_move/2 (or similar) callback, for the purpose
of range splitting.
Since the substitution needed to case over all the instructions, a new
module, hipe_ppc_subst, was introduced to the ppc backend.
|
|
These will not only be useful for hipe_regalloc_prepass, but also, after
the introduction of a mk_move/2 (or similar) callback, for the purpose
of range splitting.
Since the substitution needed to case over all the instructions, a new
module, hipe_sparc_subst, was introduced to the sparc backend.
|
|
These will not only be useful for hipe_regalloc_prepass, but also, after
the introduction of a mk_move/2 (or similar) callback, for the purpose
of range splitting.
Since the substitution needed to case over all the instructions, a new
module, hipe_arm_subst, was introduced to the arm backend.
|
|
These will not only be useful for hipe_regalloc_prepass, but also, after
the introduction of a mk_move/2 (or similar) callback, for the purpose
of range splitting.
Since the substitution needed to case over all the instructions, a new
module, hipe_x86_subst, was introduced to the x86 backend.
Due to differences in the 'jtab' field of a #jmp_switch{} between x86
and amd64, it regrettably needed to be duplicated to hipe_amd64_subst.
|
|
Now that all backends do register allocation on a CFG directly and
define the defun_to_cfg/1 callback as the identity function, it can be
removed.
|
|
As the just_as_good_as assertion was loosened with the `NowRegs >=
CheckRegs` check, it no longer verified that hipe_regalloc_prepass had
not incorrectly labeled a temp as unallocatable. We add that behaviour
back.
|
|
|
|
|
|
hipe_regalloc_prepass speeds up register allocation by spilling any temp
that is live over a call (which clobbers all register).
In order to detect these, a new function was added to the target
interface; defines_all_alloc/1, that takes an instruction and returns a
boolean.
|
|
Now, there will only ever be a single Linear->CFG conversion, just after
lowering from RTL, and only ever a single CFG->Linear conversion, just
before the finalise pass. Both of these now happen in hipe_sparc_main.
|
|
Now, there will only ever be a single Linear->CFG conversion, just after
lowering from RTL, and only ever a single CFG->Linear conversion, just
before the finalise pass. Both of these now happen in hipe_ppc_main.
|
|
Now, there will only ever be a single Linear->CFG conversion, just after
lowering from RTL, and only ever a single CFG->Linear conversion, just
before the finalise pass. Both of these now happen in hipe_arm_main.
|
|
For x86, additionally reuse liveness from float LSRA for the GP LSRA.
|
|
Most x86 passes were either linearise(pass(to_cfg(Code))) or trivially
rewritable to process a CFG. This saves a great deal of time and memory
churn when compiling large programs.
Now, there will only ever be a single Linear->CFG conversion, just after
lowering from RTL, and only ever a single CFG->Linear conversion, just
before the finalise pass. Both of these now happen in hipe_x86_main.
|
|
There is little point offering LSRA for x86 if we're still going to call
hipe_graph_coloring_regalloc for the floats. In particular, all
allocators except LSRA allocates an N^2 interference matrix, making them
unusable for really large functions.
|
|
The 'array' module is highly optimised for the hipe_vectors use-case,
and seems to perform slightly better than the gb_trees implementation.
Also, we remove the completely unnecessary hipe_vectors.hrl header.
|
|
and correct the name of another, erroneously spelt, option in the process.
|
|
|
|
|
|
|
|
The types array(), dict(), digraph(), gb_set(), gb_tree(), queue(),
set(), and tid() have been deprecated. They will be removed in OTP 18.0.
Instead the types array:array(), dict:dict(), digraph:graph(),
gb_set:set(), gb_tree:tree(), queue:queue(), sets:set(), and ets:tid()
can be used. (Note: it has always been necessary to use ets:tid().)
It is allowed in OTP 17.0 to locally re-define the types array(), dict(),
and so on.
New types array:array/1, dict:dict/2, gb_sets:set/1, gb_trees:tree/2,
queue:queue/1, and sets:set/1 have been added.
|
|
|
|
|
|
|
|
|
|
|
|
OTP-10106
OTP-10107
|
|
|
|
|
|
|
|
|
|
|
|
|
