- Generated by
1.8.1.1
|
GCC Middle and Back End API Reference
|
Data Structures | |
| struct | prop_value_d |
Typedefs | |
| typedef struct prop_value_d | prop_value_t |
Functions | |
| static bool | stmt_may_generate_copy () |
| static prop_value_t * | get_copy_of_val () |
| static tree | valueize_val () |
| static bool | set_copy_of_val () |
| static void | dump_copy_of () |
| static enum ssa_prop_result | copy_prop_visit_assignment () |
| static enum ssa_prop_result | copy_prop_visit_cond_stmt () |
| static enum ssa_prop_result | copy_prop_visit_stmt () |
| static enum ssa_prop_result | copy_prop_visit_phi_node () |
| static void | init_copy_prop () |
| static tree | get_value () |
| static void | fini_copy_prop () |
| static unsigned int | execute_copy_prop () |
| static bool | gate_copy_prop () |
| gimple_opt_pass * | make_pass_copy_prop () |
Variables | |
| static prop_value_t * | copy_of |
| static unsigned | n_copy_of |
| typedef struct prop_value_d prop_value_t |
|
static |
Evaluate the RHS of STMT. If it produces a valid copy, set the LHS value and store the LHS into *RESULT_P.
Straight copy between two SSA names. First, make sure that
we can propagate the RHS into uses of LHS.
|
static |
Visit the GIMPLE_COND STMT. Return SSA_PROP_INTERESTING if it can determine which edge will be taken. Otherwise, return SSA_PROP_VARYING.
The only conditionals that we may be able to compute statically
are predicates involving two SSA_NAMEs. See if we can determine the predicate's value.
We can fold COND and get a useful result only when we have
the same SSA_NAME on both sides of a comparison operator.
References dump_file, and print_gimple_stmt().
Referenced by copy_prop_visit_stmt().
|
static |
Visit PHI node PHI. If all the arguments produce the same value, set it to be the value of the LHS of PHI.
We don't care about values flowing through non-executable
edges. Names that flow through abnormal edges cannot be used to
derive copies. If we didn't visit the definition of arg yet treat it as
UNDEFINED. This also handles PHI arguments that are the
same as lhs. We'll come here again. Avoid copy propagation from an inner into an outer loop.
Otherwise, this may move loop variant variables outside of
their loops and prevent coalescing opportunities. If the
value was loop invariant, it will be hoisted by LICM and
exposed for copy propagation.
??? The value will be always loop invariant.
In loop-closed SSA form do not copy-propagate through
PHI nodes in blocks with a loop exit edge predecessor. If the LHS didn't have a value yet, make it a copy of the
first argument we find. If PHI_VAL and ARG don't have a common copy-of chain, then
this PHI node cannot be a copy operation.
References prop_value_d::value.
|
static |
Evaluate statement STMT. If the statement produces a new output value, return SSA_PROP_INTERESTING and store the SSA_NAME holding the new value in *RESULT_P. If STMT is a conditional branch and we can determine its truth value, set *TAKEN_EDGE_P accordingly. If the new value produced by STMT is varying, return SSA_PROP_VARYING.
If the statement is a copy assignment, evaluate its RHS to
see if the lattice value of its output has changed. See if we can determine which edge goes out of a conditional
jump. Any other kind of statement is not interesting for constant
propagation and, therefore, not worth simulating. The assignment is not a copy operation. Don't visit this
statement again and mark all the definitions in the statement
to be copies of nothing.
References copy_prop_visit_cond_stmt().
|
static |
Dump the copy-of value for variable VAR to FILE.
|
static |
Main entry point to the copy propagator.
PHIS_ONLY is true if we should only consider PHI nodes as generating
copy propagation opportunities.
The algorithm propagates the value COPY-OF using ssa_propagate. For
every variable X_i, COPY-OF(X_i) indicates which variable is X_i created
from. The following example shows how the algorithm proceeds at a
high level:
1 a_24 = x_1
2 a_2 = PHI <a_24, x_1>
3 a_5 = PHI <a_2>
4 x_1 = PHI <x_298, a_5, a_2>
The end result should be that a_2, a_5, a_24 and x_1 are a copy of
x_298. Propagation proceeds as follows.
Visit #1: a_24 is copy-of x_1. Value changed.
Visit #2: a_2 is copy-of x_1. Value changed.
Visit #3: a_5 is copy-of x_1. Value changed.
Visit #4: x_1 is copy-of x_298. Value changed.
Visit #1: a_24 is copy-of x_298. Value changed.
Visit #2: a_2 is copy-of x_298. Value changed.
Visit #3: a_5 is copy-of x_298. Value changed.
Visit #4: x_1 is copy-of x_298. Stable state reached.
When visiting PHI nodes, we only consider arguments that flow
through edges marked executable by the propagation engine. So,
when visiting statement #2 for the first time, we will only look at
the first argument (a_24) and optimistically assume that its value
is the copy of a_24 (x_1).
|
static |
Deallocate memory used in copy propagation and do final substitution.
Set the final copy-of value for each variable by traversing the
copy-of chains. In theory the points-to solution of all members of the
copy chain is their intersection. For now we do not bother
to compute this but only make sure we do not lose points-to
information completely by setting the points-to solution
of the representative to the first solution we find if
it doesn't have one already. Don't do DCE if SCEV is initialized. It would destroy the scev cache.
|
static |
|
inlinestatic |
Return the copy-of value for VAR.
If the variable will never generate a useful copy relation,
make it its own copy.
|
static |
Callback for substitute_and_fold to get at the final copy-of values.
References copy_of, duplicate_ssa_name_ptr_info(), and duplicate_ssa_name_range_info().
|
static |
Initialize structures used for copy propagation.
The only statements that we care about are those that may
generate useful copies. We also need to mark conditional
jumps so that their outgoing edges are added to the work
lists of the propagator.
Avoid copy propagation from an inner into an outer loop.
Otherwise, this may move loop variant variables outside of
their loops and prevent coalescing opportunities. If the
value was loop invariant, it will be hoisted by LICM and
exposed for copy propagation.
??? This doesn't make sense. Since we are iterating over the statements in
BB, not the phi nodes, STMT will always be an
assignment. Mark all the outputs of this statement as not being
the copy of anything.
| gimple_opt_pass* make_pass_copy_prop | ( | ) |
|
inlinestatic |
Set VAL to be the copy of VAR. If that changed return true.
Set FIRST to be the first link in COPY_OF[DEST]. If that
changed, return true.
|
static |
Return true if this statement may generate a useful copy.
If the statement has volatile operands, it won't generate a
useful copy. Statements with loads and/or stores will never generate a useful copy.
Otherwise, the only statements that generate useful copies are
assignments whose RHS is just an SSA name that doesn't flow
through abnormal edges.
|
inlinestatic |
Return the variable VAR is a copy of or VAR if VAR isn't the result of a copy.
References copy_of, operand_equal_p(), and prop_value_d::value.
|
static |
Referenced by get_value(), and valueize_val().
|
static |
1.8.1.1