/* Global common subexpression elimination/Partial redundancy elimination
|
/* Global common subexpression elimination/Partial redundancy elimination
|
and global constant/copy propagation for GNU compiler.
|
and global constant/copy propagation for GNU compiler.
|
Copyright (C) 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2007
|
Copyright (C) 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2007
|
Free Software Foundation, Inc.
|
Free Software Foundation, Inc.
|
|
|
This file is part of GCC.
|
This file is part of GCC.
|
|
|
GCC is free software; you can redistribute it and/or modify it under
|
GCC is free software; you can redistribute it and/or modify it under
|
the terms of the GNU General Public License as published by the Free
|
the terms of the GNU General Public License as published by the Free
|
Software Foundation; either version 3, or (at your option) any later
|
Software Foundation; either version 3, or (at your option) any later
|
version.
|
version.
|
|
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
for more details.
|
for more details.
|
|
|
You should have received a copy of the GNU General Public License
|
You should have received a copy of the GNU General Public License
|
along with GCC; see the file COPYING3. If not see
|
along with GCC; see the file COPYING3. If not see
|
<http://www.gnu.org/licenses/>. */
|
<http://www.gnu.org/licenses/>. */
|
|
|
/* TODO
|
/* TODO
|
- reordering of memory allocation and freeing to be more space efficient
|
- reordering of memory allocation and freeing to be more space efficient
|
- do rough calc of how many regs are needed in each block, and a rough
|
- do rough calc of how many regs are needed in each block, and a rough
|
calc of how many regs are available in each class and use that to
|
calc of how many regs are available in each class and use that to
|
throttle back the code in cases where RTX_COST is minimal.
|
throttle back the code in cases where RTX_COST is minimal.
|
- a store to the same address as a load does not kill the load if the
|
- a store to the same address as a load does not kill the load if the
|
source of the store is also the destination of the load. Handling this
|
source of the store is also the destination of the load. Handling this
|
allows more load motion, particularly out of loops.
|
allows more load motion, particularly out of loops.
|
- ability to realloc sbitmap vectors would allow one initial computation
|
- ability to realloc sbitmap vectors would allow one initial computation
|
of reg_set_in_block with only subsequent additions, rather than
|
of reg_set_in_block with only subsequent additions, rather than
|
recomputing it for each pass
|
recomputing it for each pass
|
|
|
*/
|
*/
|
|
|
/* References searched while implementing this.
|
/* References searched while implementing this.
|
|
|
Compilers Principles, Techniques and Tools
|
Compilers Principles, Techniques and Tools
|
Aho, Sethi, Ullman
|
Aho, Sethi, Ullman
|
Addison-Wesley, 1988
|
Addison-Wesley, 1988
|
|
|
Global Optimization by Suppression of Partial Redundancies
|
Global Optimization by Suppression of Partial Redundancies
|
E. Morel, C. Renvoise
|
E. Morel, C. Renvoise
|
communications of the acm, Vol. 22, Num. 2, Feb. 1979
|
communications of the acm, Vol. 22, Num. 2, Feb. 1979
|
|
|
A Portable Machine-Independent Global Optimizer - Design and Measurements
|
A Portable Machine-Independent Global Optimizer - Design and Measurements
|
Frederick Chow
|
Frederick Chow
|
Stanford Ph.D. thesis, Dec. 1983
|
Stanford Ph.D. thesis, Dec. 1983
|
|
|
A Fast Algorithm for Code Movement Optimization
|
A Fast Algorithm for Code Movement Optimization
|
D.M. Dhamdhere
|
D.M. Dhamdhere
|
SIGPLAN Notices, Vol. 23, Num. 10, Oct. 1988
|
SIGPLAN Notices, Vol. 23, Num. 10, Oct. 1988
|
|
|
A Solution to a Problem with Morel and Renvoise's
|
A Solution to a Problem with Morel and Renvoise's
|
Global Optimization by Suppression of Partial Redundancies
|
Global Optimization by Suppression of Partial Redundancies
|
K-H Drechsler, M.P. Stadel
|
K-H Drechsler, M.P. Stadel
|
ACM TOPLAS, Vol. 10, Num. 4, Oct. 1988
|
ACM TOPLAS, Vol. 10, Num. 4, Oct. 1988
|
|
|
Practical Adaptation of the Global Optimization
|
Practical Adaptation of the Global Optimization
|
Algorithm of Morel and Renvoise
|
Algorithm of Morel and Renvoise
|
D.M. Dhamdhere
|
D.M. Dhamdhere
|
ACM TOPLAS, Vol. 13, Num. 2. Apr. 1991
|
ACM TOPLAS, Vol. 13, Num. 2. Apr. 1991
|
|
|
Efficiently Computing Static Single Assignment Form and the Control
|
Efficiently Computing Static Single Assignment Form and the Control
|
Dependence Graph
|
Dependence Graph
|
R. Cytron, J. Ferrante, B.K. Rosen, M.N. Wegman, and F.K. Zadeck
|
R. Cytron, J. Ferrante, B.K. Rosen, M.N. Wegman, and F.K. Zadeck
|
ACM TOPLAS, Vol. 13, Num. 4, Oct. 1991
|
ACM TOPLAS, Vol. 13, Num. 4, Oct. 1991
|
|
|
Lazy Code Motion
|
Lazy Code Motion
|
J. Knoop, O. Ruthing, B. Steffen
|
J. Knoop, O. Ruthing, B. Steffen
|
ACM SIGPLAN Notices Vol. 27, Num. 7, Jul. 1992, '92 Conference on PLDI
|
ACM SIGPLAN Notices Vol. 27, Num. 7, Jul. 1992, '92 Conference on PLDI
|
|
|
What's In a Region? Or Computing Control Dependence Regions in Near-Linear
|
What's In a Region? Or Computing Control Dependence Regions in Near-Linear
|
Time for Reducible Flow Control
|
Time for Reducible Flow Control
|
Thomas Ball
|
Thomas Ball
|
ACM Letters on Programming Languages and Systems,
|
ACM Letters on Programming Languages and Systems,
|
Vol. 2, Num. 1-4, Mar-Dec 1993
|
Vol. 2, Num. 1-4, Mar-Dec 1993
|
|
|
An Efficient Representation for Sparse Sets
|
An Efficient Representation for Sparse Sets
|
Preston Briggs, Linda Torczon
|
Preston Briggs, Linda Torczon
|
ACM Letters on Programming Languages and Systems,
|
ACM Letters on Programming Languages and Systems,
|
Vol. 2, Num. 1-4, Mar-Dec 1993
|
Vol. 2, Num. 1-4, Mar-Dec 1993
|
|
|
A Variation of Knoop, Ruthing, and Steffen's Lazy Code Motion
|
A Variation of Knoop, Ruthing, and Steffen's Lazy Code Motion
|
K-H Drechsler, M.P. Stadel
|
K-H Drechsler, M.P. Stadel
|
ACM SIGPLAN Notices, Vol. 28, Num. 5, May 1993
|
ACM SIGPLAN Notices, Vol. 28, Num. 5, May 1993
|
|
|
Partial Dead Code Elimination
|
Partial Dead Code Elimination
|
J. Knoop, O. Ruthing, B. Steffen
|
J. Knoop, O. Ruthing, B. Steffen
|
ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994
|
ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994
|
|
|
Effective Partial Redundancy Elimination
|
Effective Partial Redundancy Elimination
|
P. Briggs, K.D. Cooper
|
P. Briggs, K.D. Cooper
|
ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994
|
ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994
|
|
|
The Program Structure Tree: Computing Control Regions in Linear Time
|
The Program Structure Tree: Computing Control Regions in Linear Time
|
R. Johnson, D. Pearson, K. Pingali
|
R. Johnson, D. Pearson, K. Pingali
|
ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994
|
ACM SIGPLAN Notices, Vol. 29, Num. 6, Jun. 1994
|
|
|
Optimal Code Motion: Theory and Practice
|
Optimal Code Motion: Theory and Practice
|
J. Knoop, O. Ruthing, B. Steffen
|
J. Knoop, O. Ruthing, B. Steffen
|
ACM TOPLAS, Vol. 16, Num. 4, Jul. 1994
|
ACM TOPLAS, Vol. 16, Num. 4, Jul. 1994
|
|
|
The power of assignment motion
|
The power of assignment motion
|
J. Knoop, O. Ruthing, B. Steffen
|
J. Knoop, O. Ruthing, B. Steffen
|
ACM SIGPLAN Notices Vol. 30, Num. 6, Jun. 1995, '95 Conference on PLDI
|
ACM SIGPLAN Notices Vol. 30, Num. 6, Jun. 1995, '95 Conference on PLDI
|
|
|
Global code motion / global value numbering
|
Global code motion / global value numbering
|
C. Click
|
C. Click
|
ACM SIGPLAN Notices Vol. 30, Num. 6, Jun. 1995, '95 Conference on PLDI
|
ACM SIGPLAN Notices Vol. 30, Num. 6, Jun. 1995, '95 Conference on PLDI
|
|
|
Value Driven Redundancy Elimination
|
Value Driven Redundancy Elimination
|
L.T. Simpson
|
L.T. Simpson
|
Rice University Ph.D. thesis, Apr. 1996
|
Rice University Ph.D. thesis, Apr. 1996
|
|
|
Value Numbering
|
Value Numbering
|
L.T. Simpson
|
L.T. Simpson
|
Massively Scalar Compiler Project, Rice University, Sep. 1996
|
Massively Scalar Compiler Project, Rice University, Sep. 1996
|
|
|
High Performance Compilers for Parallel Computing
|
High Performance Compilers for Parallel Computing
|
Michael Wolfe
|
Michael Wolfe
|
Addison-Wesley, 1996
|
Addison-Wesley, 1996
|
|
|
Advanced Compiler Design and Implementation
|
Advanced Compiler Design and Implementation
|
Steven Muchnick
|
Steven Muchnick
|
Morgan Kaufmann, 1997
|
Morgan Kaufmann, 1997
|
|
|
Building an Optimizing Compiler
|
Building an Optimizing Compiler
|
Robert Morgan
|
Robert Morgan
|
Digital Press, 1998
|
Digital Press, 1998
|
|
|
People wishing to speed up the code here should read:
|
People wishing to speed up the code here should read:
|
Elimination Algorithms for Data Flow Analysis
|
Elimination Algorithms for Data Flow Analysis
|
B.G. Ryder, M.C. Paull
|
B.G. Ryder, M.C. Paull
|
ACM Computing Surveys, Vol. 18, Num. 3, Sep. 1986
|
ACM Computing Surveys, Vol. 18, Num. 3, Sep. 1986
|
|
|
How to Analyze Large Programs Efficiently and Informatively
|
How to Analyze Large Programs Efficiently and Informatively
|
D.M. Dhamdhere, B.K. Rosen, F.K. Zadeck
|
D.M. Dhamdhere, B.K. Rosen, F.K. Zadeck
|
ACM SIGPLAN Notices Vol. 27, Num. 7, Jul. 1992, '92 Conference on PLDI
|
ACM SIGPLAN Notices Vol. 27, Num. 7, Jul. 1992, '92 Conference on PLDI
|
|
|
People wishing to do something different can find various possibilities
|
People wishing to do something different can find various possibilities
|
in the above papers and elsewhere.
|
in the above papers and elsewhere.
|
*/
|
*/
|
|
|
#include "config.h"
|
#include "config.h"
|
#include "system.h"
|
#include "system.h"
|
#include "coretypes.h"
|
#include "coretypes.h"
|
#include "tm.h"
|
#include "tm.h"
|
#include "toplev.h"
|
#include "toplev.h"
|
|
|
#include "rtl.h"
|
#include "rtl.h"
|
#include "tree.h"
|
#include "tree.h"
|
#include "tm_p.h"
|
#include "tm_p.h"
|
#include "regs.h"
|
#include "regs.h"
|
#include "hard-reg-set.h"
|
#include "hard-reg-set.h"
|
#include "flags.h"
|
#include "flags.h"
|
#include "real.h"
|
#include "real.h"
|
#include "insn-config.h"
|
#include "insn-config.h"
|
#include "recog.h"
|
#include "recog.h"
|
#include "basic-block.h"
|
#include "basic-block.h"
|
#include "output.h"
|
#include "output.h"
|
#include "function.h"
|
#include "function.h"
|
#include "expr.h"
|
#include "expr.h"
|
#include "except.h"
|
#include "except.h"
|
#include "ggc.h"
|
#include "ggc.h"
|
#include "params.h"
|
#include "params.h"
|
#include "cselib.h"
|
#include "cselib.h"
|
#include "intl.h"
|
#include "intl.h"
|
#include "obstack.h"
|
#include "obstack.h"
|
#include "timevar.h"
|
#include "timevar.h"
|
#include "tree-pass.h"
|
#include "tree-pass.h"
|
#include "hashtab.h"
|
#include "hashtab.h"
|
|
|
/* Propagate flow information through back edges and thus enable PRE's
|
/* Propagate flow information through back edges and thus enable PRE's
|
moving loop invariant calculations out of loops.
|
moving loop invariant calculations out of loops.
|
|
|
Originally this tended to create worse overall code, but several
|
Originally this tended to create worse overall code, but several
|
improvements during the development of PRE seem to have made following
|
improvements during the development of PRE seem to have made following
|
back edges generally a win.
|
back edges generally a win.
|
|
|
Note much of the loop invariant code motion done here would normally
|
Note much of the loop invariant code motion done here would normally
|
be done by loop.c, which has more heuristics for when to move invariants
|
be done by loop.c, which has more heuristics for when to move invariants
|
out of loops. At some point we might need to move some of those
|
out of loops. At some point we might need to move some of those
|
heuristics into gcse.c. */
|
heuristics into gcse.c. */
|
|
|
/* We support GCSE via Partial Redundancy Elimination. PRE optimizations
|
/* We support GCSE via Partial Redundancy Elimination. PRE optimizations
|
are a superset of those done by GCSE.
|
are a superset of those done by GCSE.
|
|
|
We perform the following steps:
|
We perform the following steps:
|
|
|
1) Compute basic block information.
|
1) Compute basic block information.
|
|
|
2) Compute table of places where registers are set.
|
2) Compute table of places where registers are set.
|
|
|
3) Perform copy/constant propagation.
|
3) Perform copy/constant propagation.
|
|
|
4) Perform global cse using lazy code motion if not optimizing
|
4) Perform global cse using lazy code motion if not optimizing
|
for size, or code hoisting if we are.
|
for size, or code hoisting if we are.
|
|
|
5) Perform another pass of copy/constant propagation.
|
5) Perform another pass of copy/constant propagation.
|
|
|
Two passes of copy/constant propagation are done because the first one
|
Two passes of copy/constant propagation are done because the first one
|
enables more GCSE and the second one helps to clean up the copies that
|
enables more GCSE and the second one helps to clean up the copies that
|
GCSE creates. This is needed more for PRE than for Classic because Classic
|
GCSE creates. This is needed more for PRE than for Classic because Classic
|
GCSE will try to use an existing register containing the common
|
GCSE will try to use an existing register containing the common
|
subexpression rather than create a new one. This is harder to do for PRE
|
subexpression rather than create a new one. This is harder to do for PRE
|
because of the code motion (which Classic GCSE doesn't do).
|
because of the code motion (which Classic GCSE doesn't do).
|
|
|
Expressions we are interested in GCSE-ing are of the form
|
Expressions we are interested in GCSE-ing are of the form
|
(set (pseudo-reg) (expression)).
|
(set (pseudo-reg) (expression)).
|
Function want_to_gcse_p says what these are.
|
Function want_to_gcse_p says what these are.
|
|
|
PRE handles moving invariant expressions out of loops (by treating them as
|
PRE handles moving invariant expressions out of loops (by treating them as
|
partially redundant).
|
partially redundant).
|
|
|
Eventually it would be nice to replace cse.c/gcse.c with SSA (static single
|
Eventually it would be nice to replace cse.c/gcse.c with SSA (static single
|
assignment) based GVN (global value numbering). L. T. Simpson's paper
|
assignment) based GVN (global value numbering). L. T. Simpson's paper
|
(Rice University) on value numbering is a useful reference for this.
|
(Rice University) on value numbering is a useful reference for this.
|
|
|
**********************
|
**********************
|
|
|
We used to support multiple passes but there are diminishing returns in
|
We used to support multiple passes but there are diminishing returns in
|
doing so. The first pass usually makes 90% of the changes that are doable.
|
doing so. The first pass usually makes 90% of the changes that are doable.
|
A second pass can make a few more changes made possible by the first pass.
|
A second pass can make a few more changes made possible by the first pass.
|
Experiments show any further passes don't make enough changes to justify
|
Experiments show any further passes don't make enough changes to justify
|
the expense.
|
the expense.
|
|
|
A study of spec92 using an unlimited number of passes:
|
A study of spec92 using an unlimited number of passes:
|
[1 pass] = 1208 substitutions, [2] = 577, [3] = 202, [4] = 192, [5] = 83,
|
[1 pass] = 1208 substitutions, [2] = 577, [3] = 202, [4] = 192, [5] = 83,
|
[6] = 34, [7] = 17, [8] = 9, [9] = 4, [10] = 4, [11] = 2,
|
[6] = 34, [7] = 17, [8] = 9, [9] = 4, [10] = 4, [11] = 2,
|
[12] = 2, [13] = 1, [15] = 1, [16] = 2, [41] = 1
|
[12] = 2, [13] = 1, [15] = 1, [16] = 2, [41] = 1
|
|
|
It was found doing copy propagation between each pass enables further
|
It was found doing copy propagation between each pass enables further
|
substitutions.
|
substitutions.
|
|
|
PRE is quite expensive in complicated functions because the DFA can take
|
PRE is quite expensive in complicated functions because the DFA can take
|
a while to converge. Hence we only perform one pass. The parameter
|
a while to converge. Hence we only perform one pass. The parameter
|
max-gcse-passes can be modified if one wants to experiment.
|
max-gcse-passes can be modified if one wants to experiment.
|
|
|
**********************
|
**********************
|
|
|
The steps for PRE are:
|
The steps for PRE are:
|
|
|
1) Build the hash table of expressions we wish to GCSE (expr_hash_table).
|
1) Build the hash table of expressions we wish to GCSE (expr_hash_table).
|
|
|
2) Perform the data flow analysis for PRE.
|
2) Perform the data flow analysis for PRE.
|
|
|
3) Delete the redundant instructions
|
3) Delete the redundant instructions
|
|
|
4) Insert the required copies [if any] that make the partially
|
4) Insert the required copies [if any] that make the partially
|
redundant instructions fully redundant.
|
redundant instructions fully redundant.
|
|
|
5) For other reaching expressions, insert an instruction to copy the value
|
5) For other reaching expressions, insert an instruction to copy the value
|
to a newly created pseudo that will reach the redundant instruction.
|
to a newly created pseudo that will reach the redundant instruction.
|
|
|
The deletion is done first so that when we do insertions we
|
The deletion is done first so that when we do insertions we
|
know which pseudo reg to use.
|
know which pseudo reg to use.
|
|
|
Various papers have argued that PRE DFA is expensive (O(n^2)) and others
|
Various papers have argued that PRE DFA is expensive (O(n^2)) and others
|
argue it is not. The number of iterations for the algorithm to converge
|
argue it is not. The number of iterations for the algorithm to converge
|
is typically 2-4 so I don't view it as that expensive (relatively speaking).
|
is typically 2-4 so I don't view it as that expensive (relatively speaking).
|
|
|
PRE GCSE depends heavily on the second CSE pass to clean up the copies
|
PRE GCSE depends heavily on the second CSE pass to clean up the copies
|
we create. To make an expression reach the place where it's redundant,
|
we create. To make an expression reach the place where it's redundant,
|
the result of the expression is copied to a new register, and the redundant
|
the result of the expression is copied to a new register, and the redundant
|
expression is deleted by replacing it with this new register. Classic GCSE
|
expression is deleted by replacing it with this new register. Classic GCSE
|
doesn't have this problem as much as it computes the reaching defs of
|
doesn't have this problem as much as it computes the reaching defs of
|
each register in each block and thus can try to use an existing
|
each register in each block and thus can try to use an existing
|
register. */
|
register. */
|
|
|
/* GCSE global vars. */
|
/* GCSE global vars. */
|
|
|
/* Note whether or not we should run jump optimization after gcse. We
|
/* Note whether or not we should run jump optimization after gcse. We
|
want to do this for two cases.
|
want to do this for two cases.
|
|
|
* If we changed any jumps via cprop.
|
* If we changed any jumps via cprop.
|
|
|
* If we added any labels via edge splitting. */
|
* If we added any labels via edge splitting. */
|
static int run_jump_opt_after_gcse;
|
static int run_jump_opt_after_gcse;
|
|
|
/* An obstack for our working variables. */
|
/* An obstack for our working variables. */
|
static struct obstack gcse_obstack;
|
static struct obstack gcse_obstack;
|
|
|
struct reg_use {rtx reg_rtx; };
|
struct reg_use {rtx reg_rtx; };
|
|
|
/* Hash table of expressions. */
|
/* Hash table of expressions. */
|
|
|
struct expr
|
struct expr
|
{
|
{
|
/* The expression (SET_SRC for expressions, PATTERN for assignments). */
|
/* The expression (SET_SRC for expressions, PATTERN for assignments). */
|
rtx expr;
|
rtx expr;
|
/* Index in the available expression bitmaps. */
|
/* Index in the available expression bitmaps. */
|
int bitmap_index;
|
int bitmap_index;
|
/* Next entry with the same hash. */
|
/* Next entry with the same hash. */
|
struct expr *next_same_hash;
|
struct expr *next_same_hash;
|
/* List of anticipatable occurrences in basic blocks in the function.
|
/* List of anticipatable occurrences in basic blocks in the function.
|
An "anticipatable occurrence" is one that is the first occurrence in the
|
An "anticipatable occurrence" is one that is the first occurrence in the
|
basic block, the operands are not modified in the basic block prior
|
basic block, the operands are not modified in the basic block prior
|
to the occurrence and the output is not used between the start of
|
to the occurrence and the output is not used between the start of
|
the block and the occurrence. */
|
the block and the occurrence. */
|
struct occr *antic_occr;
|
struct occr *antic_occr;
|
/* List of available occurrence in basic blocks in the function.
|
/* List of available occurrence in basic blocks in the function.
|
An "available occurrence" is one that is the last occurrence in the
|
An "available occurrence" is one that is the last occurrence in the
|
basic block and the operands are not modified by following statements in
|
basic block and the operands are not modified by following statements in
|
the basic block [including this insn]. */
|
the basic block [including this insn]. */
|
struct occr *avail_occr;
|
struct occr *avail_occr;
|
/* Non-null if the computation is PRE redundant.
|
/* Non-null if the computation is PRE redundant.
|
The value is the newly created pseudo-reg to record a copy of the
|
The value is the newly created pseudo-reg to record a copy of the
|
expression in all the places that reach the redundant copy. */
|
expression in all the places that reach the redundant copy. */
|
rtx reaching_reg;
|
rtx reaching_reg;
|
};
|
};
|
|
|
/* Occurrence of an expression.
|
/* Occurrence of an expression.
|
There is one per basic block. If a pattern appears more than once the
|
There is one per basic block. If a pattern appears more than once the
|
last appearance is used [or first for anticipatable expressions]. */
|
last appearance is used [or first for anticipatable expressions]. */
|
|
|
struct occr
|
struct occr
|
{
|
{
|
/* Next occurrence of this expression. */
|
/* Next occurrence of this expression. */
|
struct occr *next;
|
struct occr *next;
|
/* The insn that computes the expression. */
|
/* The insn that computes the expression. */
|
rtx insn;
|
rtx insn;
|
/* Nonzero if this [anticipatable] occurrence has been deleted. */
|
/* Nonzero if this [anticipatable] occurrence has been deleted. */
|
char deleted_p;
|
char deleted_p;
|
/* Nonzero if this [available] occurrence has been copied to
|
/* Nonzero if this [available] occurrence has been copied to
|
reaching_reg. */
|
reaching_reg. */
|
/* ??? This is mutually exclusive with deleted_p, so they could share
|
/* ??? This is mutually exclusive with deleted_p, so they could share
|
the same byte. */
|
the same byte. */
|
char copied_p;
|
char copied_p;
|
};
|
};
|
|
|
/* Expression and copy propagation hash tables.
|
/* Expression and copy propagation hash tables.
|
Each hash table is an array of buckets.
|
Each hash table is an array of buckets.
|
??? It is known that if it were an array of entries, structure elements
|
??? It is known that if it were an array of entries, structure elements
|
`next_same_hash' and `bitmap_index' wouldn't be necessary. However, it is
|
`next_same_hash' and `bitmap_index' wouldn't be necessary. However, it is
|
not clear whether in the final analysis a sufficient amount of memory would
|
not clear whether in the final analysis a sufficient amount of memory would
|
be saved as the size of the available expression bitmaps would be larger
|
be saved as the size of the available expression bitmaps would be larger
|
[one could build a mapping table without holes afterwards though].
|
[one could build a mapping table without holes afterwards though].
|
Someday I'll perform the computation and figure it out. */
|
Someday I'll perform the computation and figure it out. */
|
|
|
struct hash_table
|
struct hash_table
|
{
|
{
|
/* The table itself.
|
/* The table itself.
|
This is an array of `expr_hash_table_size' elements. */
|
This is an array of `expr_hash_table_size' elements. */
|
struct expr **table;
|
struct expr **table;
|
|
|
/* Size of the hash table, in elements. */
|
/* Size of the hash table, in elements. */
|
unsigned int size;
|
unsigned int size;
|
|
|
/* Number of hash table elements. */
|
/* Number of hash table elements. */
|
unsigned int n_elems;
|
unsigned int n_elems;
|
|
|
/* Whether the table is expression of copy propagation one. */
|
/* Whether the table is expression of copy propagation one. */
|
int set_p;
|
int set_p;
|
};
|
};
|
|
|
/* Expression hash table. */
|
/* Expression hash table. */
|
static struct hash_table expr_hash_table;
|
static struct hash_table expr_hash_table;
|
|
|
/* Copy propagation hash table. */
|
/* Copy propagation hash table. */
|
static struct hash_table set_hash_table;
|
static struct hash_table set_hash_table;
|
|
|
/* Mapping of uids to cuids.
|
/* Mapping of uids to cuids.
|
Only real insns get cuids. */
|
Only real insns get cuids. */
|
static int *uid_cuid;
|
static int *uid_cuid;
|
|
|
/* Highest UID in UID_CUID. */
|
/* Highest UID in UID_CUID. */
|
static int max_uid;
|
static int max_uid;
|
|
|
/* Get the cuid of an insn. */
|
/* Get the cuid of an insn. */
|
#ifdef ENABLE_CHECKING
|
#ifdef ENABLE_CHECKING
|
#define INSN_CUID(INSN) \
|
#define INSN_CUID(INSN) \
|
(gcc_assert (INSN_UID (INSN) <= max_uid), uid_cuid[INSN_UID (INSN)])
|
(gcc_assert (INSN_UID (INSN) <= max_uid), uid_cuid[INSN_UID (INSN)])
|
#else
|
#else
|
#define INSN_CUID(INSN) (uid_cuid[INSN_UID (INSN)])
|
#define INSN_CUID(INSN) (uid_cuid[INSN_UID (INSN)])
|
#endif
|
#endif
|
|
|
/* Number of cuids. */
|
/* Number of cuids. */
|
static int max_cuid;
|
static int max_cuid;
|
|
|
/* Mapping of cuids to insns. */
|
/* Mapping of cuids to insns. */
|
static rtx *cuid_insn;
|
static rtx *cuid_insn;
|
|
|
/* Get insn from cuid. */
|
/* Get insn from cuid. */
|
#define CUID_INSN(CUID) (cuid_insn[CUID])
|
#define CUID_INSN(CUID) (cuid_insn[CUID])
|
|
|
/* Maximum register number in function prior to doing gcse + 1.
|
/* Maximum register number in function prior to doing gcse + 1.
|
Registers created during this pass have regno >= max_gcse_regno.
|
Registers created during this pass have regno >= max_gcse_regno.
|
This is named with "gcse" to not collide with global of same name. */
|
This is named with "gcse" to not collide with global of same name. */
|
static unsigned int max_gcse_regno;
|
static unsigned int max_gcse_regno;
|
|
|
/* Table of registers that are modified.
|
/* Table of registers that are modified.
|
|
|
For each register, each element is a list of places where the pseudo-reg
|
For each register, each element is a list of places where the pseudo-reg
|
is set.
|
is set.
|
|
|
For simplicity, GCSE is done on sets of pseudo-regs only. PRE GCSE only
|
For simplicity, GCSE is done on sets of pseudo-regs only. PRE GCSE only
|
requires knowledge of which blocks kill which regs [and thus could use
|
requires knowledge of which blocks kill which regs [and thus could use
|
a bitmap instead of the lists `reg_set_table' uses].
|
a bitmap instead of the lists `reg_set_table' uses].
|
|
|
`reg_set_table' and could be turned into an array of bitmaps (num-bbs x
|
`reg_set_table' and could be turned into an array of bitmaps (num-bbs x
|
num-regs) [however perhaps it may be useful to keep the data as is]. One
|
num-regs) [however perhaps it may be useful to keep the data as is]. One
|
advantage of recording things this way is that `reg_set_table' is fairly
|
advantage of recording things this way is that `reg_set_table' is fairly
|
sparse with respect to pseudo regs but for hard regs could be fairly dense
|
sparse with respect to pseudo regs but for hard regs could be fairly dense
|
[relatively speaking]. And recording sets of pseudo-regs in lists speeds
|
[relatively speaking]. And recording sets of pseudo-regs in lists speeds
|
up functions like compute_transp since in the case of pseudo-regs we only
|
up functions like compute_transp since in the case of pseudo-regs we only
|
need to iterate over the number of times a pseudo-reg is set, not over the
|
need to iterate over the number of times a pseudo-reg is set, not over the
|
number of basic blocks [clearly there is a bit of a slow down in the cases
|
number of basic blocks [clearly there is a bit of a slow down in the cases
|
where a pseudo is set more than once in a block, however it is believed
|
where a pseudo is set more than once in a block, however it is believed
|
that the net effect is to speed things up]. This isn't done for hard-regs
|
that the net effect is to speed things up]. This isn't done for hard-regs
|
because recording call-clobbered hard-regs in `reg_set_table' at each
|
because recording call-clobbered hard-regs in `reg_set_table' at each
|
function call can consume a fair bit of memory, and iterating over
|
function call can consume a fair bit of memory, and iterating over
|
hard-regs stored this way in compute_transp will be more expensive. */
|
hard-regs stored this way in compute_transp will be more expensive. */
|
|
|
typedef struct reg_set
|
typedef struct reg_set
|
{
|
{
|
/* The next setting of this register. */
|
/* The next setting of this register. */
|
struct reg_set *next;
|
struct reg_set *next;
|
/* The index of the block where it was set. */
|
/* The index of the block where it was set. */
|
int bb_index;
|
int bb_index;
|
} reg_set;
|
} reg_set;
|
|
|
static reg_set **reg_set_table;
|
static reg_set **reg_set_table;
|
|
|
/* Size of `reg_set_table'.
|
/* Size of `reg_set_table'.
|
The table starts out at max_gcse_regno + slop, and is enlarged as
|
The table starts out at max_gcse_regno + slop, and is enlarged as
|
necessary. */
|
necessary. */
|
static int reg_set_table_size;
|
static int reg_set_table_size;
|
|
|
/* Amount to grow `reg_set_table' by when it's full. */
|
/* Amount to grow `reg_set_table' by when it's full. */
|
#define REG_SET_TABLE_SLOP 100
|
#define REG_SET_TABLE_SLOP 100
|
|
|
/* This is a list of expressions which are MEMs and will be used by load
|
/* This is a list of expressions which are MEMs and will be used by load
|
or store motion.
|
or store motion.
|
Load motion tracks MEMs which aren't killed by
|
Load motion tracks MEMs which aren't killed by
|
anything except itself. (i.e., loads and stores to a single location).
|
anything except itself. (i.e., loads and stores to a single location).
|
We can then allow movement of these MEM refs with a little special
|
We can then allow movement of these MEM refs with a little special
|
allowance. (all stores copy the same value to the reaching reg used
|
allowance. (all stores copy the same value to the reaching reg used
|
for the loads). This means all values used to store into memory must have
|
for the loads). This means all values used to store into memory must have
|
no side effects so we can re-issue the setter value.
|
no side effects so we can re-issue the setter value.
|
Store Motion uses this structure as an expression table to track stores
|
Store Motion uses this structure as an expression table to track stores
|
which look interesting, and might be moveable towards the exit block. */
|
which look interesting, and might be moveable towards the exit block. */
|
|
|
struct ls_expr
|
struct ls_expr
|
{
|
{
|
struct expr * expr; /* Gcse expression reference for LM. */
|
struct expr * expr; /* Gcse expression reference for LM. */
|
rtx pattern; /* Pattern of this mem. */
|
rtx pattern; /* Pattern of this mem. */
|
rtx pattern_regs; /* List of registers mentioned by the mem. */
|
rtx pattern_regs; /* List of registers mentioned by the mem. */
|
rtx loads; /* INSN list of loads seen. */
|
rtx loads; /* INSN list of loads seen. */
|
rtx stores; /* INSN list of stores seen. */
|
rtx stores; /* INSN list of stores seen. */
|
struct ls_expr * next; /* Next in the list. */
|
struct ls_expr * next; /* Next in the list. */
|
int invalid; /* Invalid for some reason. */
|
int invalid; /* Invalid for some reason. */
|
int index; /* If it maps to a bitmap index. */
|
int index; /* If it maps to a bitmap index. */
|
unsigned int hash_index; /* Index when in a hash table. */
|
unsigned int hash_index; /* Index when in a hash table. */
|
rtx reaching_reg; /* Register to use when re-writing. */
|
rtx reaching_reg; /* Register to use when re-writing. */
|
};
|
};
|
|
|
/* Array of implicit set patterns indexed by basic block index. */
|
/* Array of implicit set patterns indexed by basic block index. */
|
static rtx *implicit_sets;
|
static rtx *implicit_sets;
|
|
|
/* Head of the list of load/store memory refs. */
|
/* Head of the list of load/store memory refs. */
|
static struct ls_expr * pre_ldst_mems = NULL;
|
static struct ls_expr * pre_ldst_mems = NULL;
|
|
|
/* Hashtable for the load/store memory refs. */
|
/* Hashtable for the load/store memory refs. */
|
static htab_t pre_ldst_table = NULL;
|
static htab_t pre_ldst_table = NULL;
|
|
|
/* Bitmap containing one bit for each register in the program.
|
/* Bitmap containing one bit for each register in the program.
|
Used when performing GCSE to track which registers have been set since
|
Used when performing GCSE to track which registers have been set since
|
the start of the basic block. */
|
the start of the basic block. */
|
static regset reg_set_bitmap;
|
static regset reg_set_bitmap;
|
|
|
/* For each block, a bitmap of registers set in the block.
|
/* For each block, a bitmap of registers set in the block.
|
This is used by compute_transp.
|
This is used by compute_transp.
|
It is computed during hash table computation and not by compute_sets
|
It is computed during hash table computation and not by compute_sets
|
as it includes registers added since the last pass (or between cprop and
|
as it includes registers added since the last pass (or between cprop and
|
gcse) and it's currently not easy to realloc sbitmap vectors. */
|
gcse) and it's currently not easy to realloc sbitmap vectors. */
|
static sbitmap *reg_set_in_block;
|
static sbitmap *reg_set_in_block;
|
|
|
/* Array, indexed by basic block number for a list of insns which modify
|
/* Array, indexed by basic block number for a list of insns which modify
|
memory within that block. */
|
memory within that block. */
|
static rtx * modify_mem_list;
|
static rtx * modify_mem_list;
|
static bitmap modify_mem_list_set;
|
static bitmap modify_mem_list_set;
|
|
|
/* This array parallels modify_mem_list, but is kept canonicalized. */
|
/* This array parallels modify_mem_list, but is kept canonicalized. */
|
static rtx * canon_modify_mem_list;
|
static rtx * canon_modify_mem_list;
|
|
|
/* Bitmap indexed by block numbers to record which blocks contain
|
/* Bitmap indexed by block numbers to record which blocks contain
|
function calls. */
|
function calls. */
|
static bitmap blocks_with_calls;
|
static bitmap blocks_with_calls;
|
|
|
/* Various variables for statistics gathering. */
|
/* Various variables for statistics gathering. */
|
|
|
/* Memory used in a pass.
|
/* Memory used in a pass.
|
This isn't intended to be absolutely precise. Its intent is only
|
This isn't intended to be absolutely precise. Its intent is only
|
to keep an eye on memory usage. */
|
to keep an eye on memory usage. */
|
static int bytes_used;
|
static int bytes_used;
|
|
|
/* GCSE substitutions made. */
|
/* GCSE substitutions made. */
|
static int gcse_subst_count;
|
static int gcse_subst_count;
|
/* Number of copy instructions created. */
|
/* Number of copy instructions created. */
|
static int gcse_create_count;
|
static int gcse_create_count;
|
/* Number of local constants propagated. */
|
/* Number of local constants propagated. */
|
static int local_const_prop_count;
|
static int local_const_prop_count;
|
/* Number of local copies propagated. */
|
/* Number of local copies propagated. */
|
static int local_copy_prop_count;
|
static int local_copy_prop_count;
|
/* Number of global constants propagated. */
|
/* Number of global constants propagated. */
|
static int global_const_prop_count;
|
static int global_const_prop_count;
|
/* Number of global copies propagated. */
|
/* Number of global copies propagated. */
|
static int global_copy_prop_count;
|
static int global_copy_prop_count;
|
|
|
/* For available exprs */
|
/* For available exprs */
|
static sbitmap *ae_kill, *ae_gen;
|
static sbitmap *ae_kill, *ae_gen;
|
|
|
static void compute_can_copy (void);
|
static void compute_can_copy (void);
|
static void *gmalloc (size_t) ATTRIBUTE_MALLOC;
|
static void *gmalloc (size_t) ATTRIBUTE_MALLOC;
|
static void *gcalloc (size_t, size_t) ATTRIBUTE_MALLOC;
|
static void *gcalloc (size_t, size_t) ATTRIBUTE_MALLOC;
|
static void *grealloc (void *, size_t);
|
static void *grealloc (void *, size_t);
|
static void *gcse_alloc (unsigned long);
|
static void *gcse_alloc (unsigned long);
|
static void alloc_gcse_mem (void);
|
static void alloc_gcse_mem (void);
|
static void free_gcse_mem (void);
|
static void free_gcse_mem (void);
|
static void alloc_reg_set_mem (int);
|
static void alloc_reg_set_mem (int);
|
static void free_reg_set_mem (void);
|
static void free_reg_set_mem (void);
|
static void record_one_set (int, rtx);
|
static void record_one_set (int, rtx);
|
static void record_set_info (rtx, rtx, void *);
|
static void record_set_info (rtx, rtx, void *);
|
static void compute_sets (void);
|
static void compute_sets (void);
|
static void hash_scan_insn (rtx, struct hash_table *, int);
|
static void hash_scan_insn (rtx, struct hash_table *, int);
|
static void hash_scan_set (rtx, rtx, struct hash_table *);
|
static void hash_scan_set (rtx, rtx, struct hash_table *);
|
static void hash_scan_clobber (rtx, rtx, struct hash_table *);
|
static void hash_scan_clobber (rtx, rtx, struct hash_table *);
|
static void hash_scan_call (rtx, rtx, struct hash_table *);
|
static void hash_scan_call (rtx, rtx, struct hash_table *);
|
static int want_to_gcse_p (rtx);
|
static int want_to_gcse_p (rtx);
|
static bool can_assign_to_reg_p (rtx);
|
static bool can_assign_to_reg_p (rtx);
|
static bool gcse_constant_p (rtx);
|
static bool gcse_constant_p (rtx);
|
static int oprs_unchanged_p (rtx, rtx, int);
|
static int oprs_unchanged_p (rtx, rtx, int);
|
static int oprs_anticipatable_p (rtx, rtx);
|
static int oprs_anticipatable_p (rtx, rtx);
|
static int oprs_available_p (rtx, rtx);
|
static int oprs_available_p (rtx, rtx);
|
static void insert_expr_in_table (rtx, enum machine_mode, rtx, int, int,
|
static void insert_expr_in_table (rtx, enum machine_mode, rtx, int, int,
|
struct hash_table *);
|
struct hash_table *);
|
static void insert_set_in_table (rtx, rtx, struct hash_table *);
|
static void insert_set_in_table (rtx, rtx, struct hash_table *);
|
static unsigned int hash_expr (rtx, enum machine_mode, int *, int);
|
static unsigned int hash_expr (rtx, enum machine_mode, int *, int);
|
static unsigned int hash_set (int, int);
|
static unsigned int hash_set (int, int);
|
static int expr_equiv_p (rtx, rtx);
|
static int expr_equiv_p (rtx, rtx);
|
static void record_last_reg_set_info (rtx, int);
|
static void record_last_reg_set_info (rtx, int);
|
static void record_last_mem_set_info (rtx);
|
static void record_last_mem_set_info (rtx);
|
static void record_last_set_info (rtx, rtx, void *);
|
static void record_last_set_info (rtx, rtx, void *);
|
static void compute_hash_table (struct hash_table *);
|
static void compute_hash_table (struct hash_table *);
|
static void alloc_hash_table (int, struct hash_table *, int);
|
static void alloc_hash_table (int, struct hash_table *, int);
|
static void free_hash_table (struct hash_table *);
|
static void free_hash_table (struct hash_table *);
|
static void compute_hash_table_work (struct hash_table *);
|
static void compute_hash_table_work (struct hash_table *);
|
static void dump_hash_table (FILE *, const char *, struct hash_table *);
|
static void dump_hash_table (FILE *, const char *, struct hash_table *);
|
static struct expr *lookup_set (unsigned int, struct hash_table *);
|
static struct expr *lookup_set (unsigned int, struct hash_table *);
|
static struct expr *next_set (unsigned int, struct expr *);
|
static struct expr *next_set (unsigned int, struct expr *);
|
static void reset_opr_set_tables (void);
|
static void reset_opr_set_tables (void);
|
static int oprs_not_set_p (rtx, rtx);
|
static int oprs_not_set_p (rtx, rtx);
|
static void mark_call (rtx);
|
static void mark_call (rtx);
|
static void mark_set (rtx, rtx);
|
static void mark_set (rtx, rtx);
|
static void mark_clobber (rtx, rtx);
|
static void mark_clobber (rtx, rtx);
|
static void mark_oprs_set (rtx);
|
static void mark_oprs_set (rtx);
|
static void alloc_cprop_mem (int, int);
|
static void alloc_cprop_mem (int, int);
|
static void free_cprop_mem (void);
|
static void free_cprop_mem (void);
|
static void compute_transp (rtx, int, sbitmap *, int);
|
static void compute_transp (rtx, int, sbitmap *, int);
|
static void compute_transpout (void);
|
static void compute_transpout (void);
|
static void compute_local_properties (sbitmap *, sbitmap *, sbitmap *,
|
static void compute_local_properties (sbitmap *, sbitmap *, sbitmap *,
|
struct hash_table *);
|
struct hash_table *);
|
static void compute_cprop_data (void);
|
static void compute_cprop_data (void);
|
static void find_used_regs (rtx *, void *);
|
static void find_used_regs (rtx *, void *);
|
static int try_replace_reg (rtx, rtx, rtx);
|
static int try_replace_reg (rtx, rtx, rtx);
|
static struct expr *find_avail_set (int, rtx);
|
static struct expr *find_avail_set (int, rtx);
|
static int cprop_jump (basic_block, rtx, rtx, rtx, rtx);
|
static int cprop_jump (basic_block, rtx, rtx, rtx, rtx);
|
static void mems_conflict_for_gcse_p (rtx, rtx, void *);
|
static void mems_conflict_for_gcse_p (rtx, rtx, void *);
|
static int load_killed_in_block_p (basic_block, int, rtx, int);
|
static int load_killed_in_block_p (basic_block, int, rtx, int);
|
static void canon_list_insert (rtx, rtx, void *);
|
static void canon_list_insert (rtx, rtx, void *);
|
static int cprop_insn (rtx, int);
|
static int cprop_insn (rtx, int);
|
static int cprop (int);
|
static int cprop (int);
|
static void find_implicit_sets (void);
|
static void find_implicit_sets (void);
|
static int one_cprop_pass (int, bool, bool);
|
static int one_cprop_pass (int, bool, bool);
|
static bool constprop_register (rtx, rtx, rtx, bool);
|
static bool constprop_register (rtx, rtx, rtx, bool);
|
static struct expr *find_bypass_set (int, int);
|
static struct expr *find_bypass_set (int, int);
|
static bool reg_killed_on_edge (rtx, edge);
|
static bool reg_killed_on_edge (rtx, edge);
|
static int bypass_block (basic_block, rtx, rtx);
|
static int bypass_block (basic_block, rtx, rtx);
|
static int bypass_conditional_jumps (void);
|
static int bypass_conditional_jumps (void);
|
static void alloc_pre_mem (int, int);
|
static void alloc_pre_mem (int, int);
|
static void free_pre_mem (void);
|
static void free_pre_mem (void);
|
static void compute_pre_data (void);
|
static void compute_pre_data (void);
|
static int pre_expr_reaches_here_p (basic_block, struct expr *,
|
static int pre_expr_reaches_here_p (basic_block, struct expr *,
|
basic_block);
|
basic_block);
|
static void insert_insn_end_bb (struct expr *, basic_block, int);
|
static void insert_insn_end_bb (struct expr *, basic_block, int);
|
static void pre_insert_copy_insn (struct expr *, rtx);
|
static void pre_insert_copy_insn (struct expr *, rtx);
|
static void pre_insert_copies (void);
|
static void pre_insert_copies (void);
|
static int pre_delete (void);
|
static int pre_delete (void);
|
static int pre_gcse (void);
|
static int pre_gcse (void);
|
static int one_pre_gcse_pass (int);
|
static int one_pre_gcse_pass (int);
|
static void add_label_notes (rtx, rtx);
|
static void add_label_notes (rtx, rtx);
|
static void alloc_code_hoist_mem (int, int);
|
static void alloc_code_hoist_mem (int, int);
|
static void free_code_hoist_mem (void);
|
static void free_code_hoist_mem (void);
|
static void compute_code_hoist_vbeinout (void);
|
static void compute_code_hoist_vbeinout (void);
|
static void compute_code_hoist_data (void);
|
static void compute_code_hoist_data (void);
|
static int hoist_expr_reaches_here_p (basic_block, int, basic_block, char *);
|
static int hoist_expr_reaches_here_p (basic_block, int, basic_block, char *);
|
static void hoist_code (void);
|
static void hoist_code (void);
|
static int one_code_hoisting_pass (void);
|
static int one_code_hoisting_pass (void);
|
static rtx process_insert_insn (struct expr *);
|
static rtx process_insert_insn (struct expr *);
|
static int pre_edge_insert (struct edge_list *, struct expr **);
|
static int pre_edge_insert (struct edge_list *, struct expr **);
|
static int pre_expr_reaches_here_p_work (basic_block, struct expr *,
|
static int pre_expr_reaches_here_p_work (basic_block, struct expr *,
|
basic_block, char *);
|
basic_block, char *);
|
static struct ls_expr * ldst_entry (rtx);
|
static struct ls_expr * ldst_entry (rtx);
|
static void free_ldst_entry (struct ls_expr *);
|
static void free_ldst_entry (struct ls_expr *);
|
static void free_ldst_mems (void);
|
static void free_ldst_mems (void);
|
static void print_ldst_list (FILE *);
|
static void print_ldst_list (FILE *);
|
static struct ls_expr * find_rtx_in_ldst (rtx);
|
static struct ls_expr * find_rtx_in_ldst (rtx);
|
static int enumerate_ldsts (void);
|
static int enumerate_ldsts (void);
|
static inline struct ls_expr * first_ls_expr (void);
|
static inline struct ls_expr * first_ls_expr (void);
|
static inline struct ls_expr * next_ls_expr (struct ls_expr *);
|
static inline struct ls_expr * next_ls_expr (struct ls_expr *);
|
static int simple_mem (rtx);
|
static int simple_mem (rtx);
|
static void invalidate_any_buried_refs (rtx);
|
static void invalidate_any_buried_refs (rtx);
|
static void compute_ld_motion_mems (void);
|
static void compute_ld_motion_mems (void);
|
static void trim_ld_motion_mems (void);
|
static void trim_ld_motion_mems (void);
|
static void update_ld_motion_stores (struct expr *);
|
static void update_ld_motion_stores (struct expr *);
|
static void reg_set_info (rtx, rtx, void *);
|
static void reg_set_info (rtx, rtx, void *);
|
static void reg_clear_last_set (rtx, rtx, void *);
|
static void reg_clear_last_set (rtx, rtx, void *);
|
static bool store_ops_ok (rtx, int *);
|
static bool store_ops_ok (rtx, int *);
|
static rtx extract_mentioned_regs (rtx);
|
static rtx extract_mentioned_regs (rtx);
|
static rtx extract_mentioned_regs_helper (rtx, rtx);
|
static rtx extract_mentioned_regs_helper (rtx, rtx);
|
static void find_moveable_store (rtx, int *, int *);
|
static void find_moveable_store (rtx, int *, int *);
|
static int compute_store_table (void);
|
static int compute_store_table (void);
|
static bool load_kills_store (rtx, rtx, int);
|
static bool load_kills_store (rtx, rtx, int);
|
static bool find_loads (rtx, rtx, int);
|
static bool find_loads (rtx, rtx, int);
|
static bool store_killed_in_insn (rtx, rtx, rtx, int);
|
static bool store_killed_in_insn (rtx, rtx, rtx, int);
|
static bool store_killed_after (rtx, rtx, rtx, basic_block, int *, rtx *);
|
static bool store_killed_after (rtx, rtx, rtx, basic_block, int *, rtx *);
|
static bool store_killed_before (rtx, rtx, rtx, basic_block, int *);
|
static bool store_killed_before (rtx, rtx, rtx, basic_block, int *);
|
static void build_store_vectors (void);
|
static void build_store_vectors (void);
|
static void insert_insn_start_bb (rtx, basic_block);
|
static void insert_insn_start_bb (rtx, basic_block);
|
static int insert_store (struct ls_expr *, edge);
|
static int insert_store (struct ls_expr *, edge);
|
static void remove_reachable_equiv_notes (basic_block, struct ls_expr *);
|
static void remove_reachable_equiv_notes (basic_block, struct ls_expr *);
|
static void replace_store_insn (rtx, rtx, basic_block, struct ls_expr *);
|
static void replace_store_insn (rtx, rtx, basic_block, struct ls_expr *);
|
static void delete_store (struct ls_expr *, basic_block);
|
static void delete_store (struct ls_expr *, basic_block);
|
static void free_store_memory (void);
|
static void free_store_memory (void);
|
static void store_motion (void);
|
static void store_motion (void);
|
static void free_insn_expr_list_list (rtx *);
|
static void free_insn_expr_list_list (rtx *);
|
static void clear_modify_mem_tables (void);
|
static void clear_modify_mem_tables (void);
|
static void free_modify_mem_tables (void);
|
static void free_modify_mem_tables (void);
|
static rtx gcse_emit_move_after (rtx, rtx, rtx);
|
static rtx gcse_emit_move_after (rtx, rtx, rtx);
|
static void local_cprop_find_used_regs (rtx *, void *);
|
static void local_cprop_find_used_regs (rtx *, void *);
|
static bool do_local_cprop (rtx, rtx, bool, rtx*);
|
static bool do_local_cprop (rtx, rtx, bool, rtx*);
|
static bool adjust_libcall_notes (rtx, rtx, rtx, rtx*);
|
static bool adjust_libcall_notes (rtx, rtx, rtx, rtx*);
|
static void local_cprop_pass (bool);
|
static void local_cprop_pass (bool);
|
static bool is_too_expensive (const char *);
|
static bool is_too_expensive (const char *);
|
|
|
|
|
/* Entry point for global common subexpression elimination.
|
/* Entry point for global common subexpression elimination.
|
F is the first instruction in the function. Return nonzero if a
|
F is the first instruction in the function. Return nonzero if a
|
change is mode. */
|
change is mode. */
|
|
|
static int
|
static int
|
gcse_main (rtx f ATTRIBUTE_UNUSED)
|
gcse_main (rtx f ATTRIBUTE_UNUSED)
|
{
|
{
|
int changed, pass;
|
int changed, pass;
|
/* Bytes used at start of pass. */
|
/* Bytes used at start of pass. */
|
int initial_bytes_used;
|
int initial_bytes_used;
|
/* Maximum number of bytes used by a pass. */
|
/* Maximum number of bytes used by a pass. */
|
int max_pass_bytes;
|
int max_pass_bytes;
|
/* Point to release obstack data from for each pass. */
|
/* Point to release obstack data from for each pass. */
|
char *gcse_obstack_bottom;
|
char *gcse_obstack_bottom;
|
|
|
/* We do not construct an accurate cfg in functions which call
|
/* We do not construct an accurate cfg in functions which call
|
setjmp, so just punt to be safe. */
|
setjmp, so just punt to be safe. */
|
if (current_function_calls_setjmp)
|
if (current_function_calls_setjmp)
|
return 0;
|
return 0;
|
|
|
/* Assume that we do not need to run jump optimizations after gcse. */
|
/* Assume that we do not need to run jump optimizations after gcse. */
|
run_jump_opt_after_gcse = 0;
|
run_jump_opt_after_gcse = 0;
|
|
|
/* Identify the basic block information for this function, including
|
/* Identify the basic block information for this function, including
|
successors and predecessors. */
|
successors and predecessors. */
|
max_gcse_regno = max_reg_num ();
|
max_gcse_regno = max_reg_num ();
|
|
|
if (dump_file)
|
if (dump_file)
|
dump_flow_info (dump_file, dump_flags);
|
dump_flow_info (dump_file, dump_flags);
|
|
|
/* Return if there's nothing to do, or it is too expensive. */
|
/* Return if there's nothing to do, or it is too expensive. */
|
if (n_basic_blocks <= NUM_FIXED_BLOCKS + 1
|
if (n_basic_blocks <= NUM_FIXED_BLOCKS + 1
|
|| is_too_expensive (_("GCSE disabled")))
|
|| is_too_expensive (_("GCSE disabled")))
|
return 0;
|
return 0;
|
|
|
gcc_obstack_init (&gcse_obstack);
|
gcc_obstack_init (&gcse_obstack);
|
bytes_used = 0;
|
bytes_used = 0;
|
|
|
/* We need alias. */
|
/* We need alias. */
|
init_alias_analysis ();
|
init_alias_analysis ();
|
/* Record where pseudo-registers are set. This data is kept accurate
|
/* Record where pseudo-registers are set. This data is kept accurate
|
during each pass. ??? We could also record hard-reg information here
|
during each pass. ??? We could also record hard-reg information here
|
[since it's unchanging], however it is currently done during hash table
|
[since it's unchanging], however it is currently done during hash table
|
computation.
|
computation.
|
|
|
It may be tempting to compute MEM set information here too, but MEM sets
|
It may be tempting to compute MEM set information here too, but MEM sets
|
will be subject to code motion one day and thus we need to compute
|
will be subject to code motion one day and thus we need to compute
|
information about memory sets when we build the hash tables. */
|
information about memory sets when we build the hash tables. */
|
|
|
alloc_reg_set_mem (max_gcse_regno);
|
alloc_reg_set_mem (max_gcse_regno);
|
compute_sets ();
|
compute_sets ();
|
|
|
pass = 0;
|
pass = 0;
|
initial_bytes_used = bytes_used;
|
initial_bytes_used = bytes_used;
|
max_pass_bytes = 0;
|
max_pass_bytes = 0;
|
gcse_obstack_bottom = gcse_alloc (1);
|
gcse_obstack_bottom = gcse_alloc (1);
|
changed = 1;
|
changed = 1;
|
while (changed && pass < MAX_GCSE_PASSES)
|
while (changed && pass < MAX_GCSE_PASSES)
|
{
|
{
|
changed = 0;
|
changed = 0;
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "GCSE pass %d\n\n", pass + 1);
|
fprintf (dump_file, "GCSE pass %d\n\n", pass + 1);
|
|
|
/* Initialize bytes_used to the space for the pred/succ lists,
|
/* Initialize bytes_used to the space for the pred/succ lists,
|
and the reg_set_table data. */
|
and the reg_set_table data. */
|
bytes_used = initial_bytes_used;
|
bytes_used = initial_bytes_used;
|
|
|
/* Each pass may create new registers, so recalculate each time. */
|
/* Each pass may create new registers, so recalculate each time. */
|
max_gcse_regno = max_reg_num ();
|
max_gcse_regno = max_reg_num ();
|
|
|
alloc_gcse_mem ();
|
alloc_gcse_mem ();
|
|
|
/* Don't allow constant propagation to modify jumps
|
/* Don't allow constant propagation to modify jumps
|
during this pass. */
|
during this pass. */
|
timevar_push (TV_CPROP1);
|
timevar_push (TV_CPROP1);
|
changed = one_cprop_pass (pass + 1, false, false);
|
changed = one_cprop_pass (pass + 1, false, false);
|
timevar_pop (TV_CPROP1);
|
timevar_pop (TV_CPROP1);
|
|
|
if (optimize_size)
|
if (optimize_size)
|
/* Do nothing. */ ;
|
/* Do nothing. */ ;
|
else
|
else
|
{
|
{
|
timevar_push (TV_PRE);
|
timevar_push (TV_PRE);
|
changed |= one_pre_gcse_pass (pass + 1);
|
changed |= one_pre_gcse_pass (pass + 1);
|
/* We may have just created new basic blocks. Release and
|
/* We may have just created new basic blocks. Release and
|
recompute various things which are sized on the number of
|
recompute various things which are sized on the number of
|
basic blocks. */
|
basic blocks. */
|
if (changed)
|
if (changed)
|
{
|
{
|
free_modify_mem_tables ();
|
free_modify_mem_tables ();
|
modify_mem_list = gcalloc (last_basic_block, sizeof (rtx));
|
modify_mem_list = gcalloc (last_basic_block, sizeof (rtx));
|
canon_modify_mem_list = gcalloc (last_basic_block, sizeof (rtx));
|
canon_modify_mem_list = gcalloc (last_basic_block, sizeof (rtx));
|
}
|
}
|
free_reg_set_mem ();
|
free_reg_set_mem ();
|
alloc_reg_set_mem (max_reg_num ());
|
alloc_reg_set_mem (max_reg_num ());
|
compute_sets ();
|
compute_sets ();
|
run_jump_opt_after_gcse = 1;
|
run_jump_opt_after_gcse = 1;
|
timevar_pop (TV_PRE);
|
timevar_pop (TV_PRE);
|
}
|
}
|
|
|
if (max_pass_bytes < bytes_used)
|
if (max_pass_bytes < bytes_used)
|
max_pass_bytes = bytes_used;
|
max_pass_bytes = bytes_used;
|
|
|
/* Free up memory, then reallocate for code hoisting. We can
|
/* Free up memory, then reallocate for code hoisting. We can
|
not re-use the existing allocated memory because the tables
|
not re-use the existing allocated memory because the tables
|
will not have info for the insns or registers created by
|
will not have info for the insns or registers created by
|
partial redundancy elimination. */
|
partial redundancy elimination. */
|
free_gcse_mem ();
|
free_gcse_mem ();
|
|
|
/* It does not make sense to run code hoisting unless we are optimizing
|
/* It does not make sense to run code hoisting unless we are optimizing
|
for code size -- it rarely makes programs faster, and can make
|
for code size -- it rarely makes programs faster, and can make
|
them bigger if we did partial redundancy elimination (when optimizing
|
them bigger if we did partial redundancy elimination (when optimizing
|
for space, we don't run the partial redundancy algorithms). */
|
for space, we don't run the partial redundancy algorithms). */
|
if (optimize_size)
|
if (optimize_size)
|
{
|
{
|
timevar_push (TV_HOIST);
|
timevar_push (TV_HOIST);
|
max_gcse_regno = max_reg_num ();
|
max_gcse_regno = max_reg_num ();
|
alloc_gcse_mem ();
|
alloc_gcse_mem ();
|
changed |= one_code_hoisting_pass ();
|
changed |= one_code_hoisting_pass ();
|
free_gcse_mem ();
|
free_gcse_mem ();
|
|
|
if (max_pass_bytes < bytes_used)
|
if (max_pass_bytes < bytes_used)
|
max_pass_bytes = bytes_used;
|
max_pass_bytes = bytes_used;
|
timevar_pop (TV_HOIST);
|
timevar_pop (TV_HOIST);
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
fflush (dump_file);
|
fflush (dump_file);
|
}
|
}
|
|
|
obstack_free (&gcse_obstack, gcse_obstack_bottom);
|
obstack_free (&gcse_obstack, gcse_obstack_bottom);
|
pass++;
|
pass++;
|
}
|
}
|
|
|
/* Do one last pass of copy propagation, including cprop into
|
/* Do one last pass of copy propagation, including cprop into
|
conditional jumps. */
|
conditional jumps. */
|
|
|
max_gcse_regno = max_reg_num ();
|
max_gcse_regno = max_reg_num ();
|
alloc_gcse_mem ();
|
alloc_gcse_mem ();
|
/* This time, go ahead and allow cprop to alter jumps. */
|
/* This time, go ahead and allow cprop to alter jumps. */
|
timevar_push (TV_CPROP2);
|
timevar_push (TV_CPROP2);
|
one_cprop_pass (pass + 1, true, false);
|
one_cprop_pass (pass + 1, true, false);
|
timevar_pop (TV_CPROP2);
|
timevar_pop (TV_CPROP2);
|
free_gcse_mem ();
|
free_gcse_mem ();
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "GCSE of %s: %d basic blocks, ",
|
fprintf (dump_file, "GCSE of %s: %d basic blocks, ",
|
current_function_name (), n_basic_blocks);
|
current_function_name (), n_basic_blocks);
|
fprintf (dump_file, "%d pass%s, %d bytes\n\n",
|
fprintf (dump_file, "%d pass%s, %d bytes\n\n",
|
pass, pass > 1 ? "es" : "", max_pass_bytes);
|
pass, pass > 1 ? "es" : "", max_pass_bytes);
|
}
|
}
|
|
|
obstack_free (&gcse_obstack, NULL);
|
obstack_free (&gcse_obstack, NULL);
|
free_reg_set_mem ();
|
free_reg_set_mem ();
|
|
|
/* We are finished with alias. */
|
/* We are finished with alias. */
|
end_alias_analysis ();
|
end_alias_analysis ();
|
allocate_reg_info (max_reg_num (), FALSE, FALSE);
|
allocate_reg_info (max_reg_num (), FALSE, FALSE);
|
|
|
if (!optimize_size && flag_gcse_sm)
|
if (!optimize_size && flag_gcse_sm)
|
{
|
{
|
timevar_push (TV_LSM);
|
timevar_push (TV_LSM);
|
store_motion ();
|
store_motion ();
|
timevar_pop (TV_LSM);
|
timevar_pop (TV_LSM);
|
}
|
}
|
|
|
/* Record where pseudo-registers are set. */
|
/* Record where pseudo-registers are set. */
|
return run_jump_opt_after_gcse;
|
return run_jump_opt_after_gcse;
|
}
|
}
|
|
|
/* Misc. utilities. */
|
/* Misc. utilities. */
|
|
|
/* Nonzero for each mode that supports (set (reg) (reg)).
|
/* Nonzero for each mode that supports (set (reg) (reg)).
|
This is trivially true for integer and floating point values.
|
This is trivially true for integer and floating point values.
|
It may or may not be true for condition codes. */
|
It may or may not be true for condition codes. */
|
static char can_copy[(int) NUM_MACHINE_MODES];
|
static char can_copy[(int) NUM_MACHINE_MODES];
|
|
|
/* Compute which modes support reg/reg copy operations. */
|
/* Compute which modes support reg/reg copy operations. */
|
|
|
static void
|
static void
|
compute_can_copy (void)
|
compute_can_copy (void)
|
{
|
{
|
int i;
|
int i;
|
#ifndef AVOID_CCMODE_COPIES
|
#ifndef AVOID_CCMODE_COPIES
|
rtx reg, insn;
|
rtx reg, insn;
|
#endif
|
#endif
|
memset (can_copy, 0, NUM_MACHINE_MODES);
|
memset (can_copy, 0, NUM_MACHINE_MODES);
|
|
|
start_sequence ();
|
start_sequence ();
|
for (i = 0; i < NUM_MACHINE_MODES; i++)
|
for (i = 0; i < NUM_MACHINE_MODES; i++)
|
if (GET_MODE_CLASS (i) == MODE_CC)
|
if (GET_MODE_CLASS (i) == MODE_CC)
|
{
|
{
|
#ifdef AVOID_CCMODE_COPIES
|
#ifdef AVOID_CCMODE_COPIES
|
can_copy[i] = 0;
|
can_copy[i] = 0;
|
#else
|
#else
|
reg = gen_rtx_REG ((enum machine_mode) i, LAST_VIRTUAL_REGISTER + 1);
|
reg = gen_rtx_REG ((enum machine_mode) i, LAST_VIRTUAL_REGISTER + 1);
|
insn = emit_insn (gen_rtx_SET (VOIDmode, reg, reg));
|
insn = emit_insn (gen_rtx_SET (VOIDmode, reg, reg));
|
if (recog (PATTERN (insn), insn, NULL) >= 0)
|
if (recog (PATTERN (insn), insn, NULL) >= 0)
|
can_copy[i] = 1;
|
can_copy[i] = 1;
|
#endif
|
#endif
|
}
|
}
|
else
|
else
|
can_copy[i] = 1;
|
can_copy[i] = 1;
|
|
|
end_sequence ();
|
end_sequence ();
|
}
|
}
|
|
|
/* Returns whether the mode supports reg/reg copy operations. */
|
/* Returns whether the mode supports reg/reg copy operations. */
|
|
|
bool
|
bool
|
can_copy_p (enum machine_mode mode)
|
can_copy_p (enum machine_mode mode)
|
{
|
{
|
static bool can_copy_init_p = false;
|
static bool can_copy_init_p = false;
|
|
|
if (! can_copy_init_p)
|
if (! can_copy_init_p)
|
{
|
{
|
compute_can_copy ();
|
compute_can_copy ();
|
can_copy_init_p = true;
|
can_copy_init_p = true;
|
}
|
}
|
|
|
return can_copy[mode] != 0;
|
return can_copy[mode] != 0;
|
}
|
}
|
|
|
/* Cover function to xmalloc to record bytes allocated. */
|
/* Cover function to xmalloc to record bytes allocated. */
|
|
|
static void *
|
static void *
|
gmalloc (size_t size)
|
gmalloc (size_t size)
|
{
|
{
|
bytes_used += size;
|
bytes_used += size;
|
return xmalloc (size);
|
return xmalloc (size);
|
}
|
}
|
|
|
/* Cover function to xcalloc to record bytes allocated. */
|
/* Cover function to xcalloc to record bytes allocated. */
|
|
|
static void *
|
static void *
|
gcalloc (size_t nelem, size_t elsize)
|
gcalloc (size_t nelem, size_t elsize)
|
{
|
{
|
bytes_used += nelem * elsize;
|
bytes_used += nelem * elsize;
|
return xcalloc (nelem, elsize);
|
return xcalloc (nelem, elsize);
|
}
|
}
|
|
|
/* Cover function to xrealloc.
|
/* Cover function to xrealloc.
|
We don't record the additional size since we don't know it.
|
We don't record the additional size since we don't know it.
|
It won't affect memory usage stats much anyway. */
|
It won't affect memory usage stats much anyway. */
|
|
|
static void *
|
static void *
|
grealloc (void *ptr, size_t size)
|
grealloc (void *ptr, size_t size)
|
{
|
{
|
return xrealloc (ptr, size);
|
return xrealloc (ptr, size);
|
}
|
}
|
|
|
/* Cover function to obstack_alloc. */
|
/* Cover function to obstack_alloc. */
|
|
|
static void *
|
static void *
|
gcse_alloc (unsigned long size)
|
gcse_alloc (unsigned long size)
|
{
|
{
|
bytes_used += size;
|
bytes_used += size;
|
return obstack_alloc (&gcse_obstack, size);
|
return obstack_alloc (&gcse_obstack, size);
|
}
|
}
|
|
|
/* Allocate memory for the cuid mapping array,
|
/* Allocate memory for the cuid mapping array,
|
and reg/memory set tracking tables.
|
and reg/memory set tracking tables.
|
|
|
This is called at the start of each pass. */
|
This is called at the start of each pass. */
|
|
|
static void
|
static void
|
alloc_gcse_mem (void)
|
alloc_gcse_mem (void)
|
{
|
{
|
int i;
|
int i;
|
basic_block bb;
|
basic_block bb;
|
rtx insn;
|
rtx insn;
|
|
|
/* Find the largest UID and create a mapping from UIDs to CUIDs.
|
/* Find the largest UID and create a mapping from UIDs to CUIDs.
|
CUIDs are like UIDs except they increase monotonically, have no gaps,
|
CUIDs are like UIDs except they increase monotonically, have no gaps,
|
and only apply to real insns.
|
and only apply to real insns.
|
(Actually, there are gaps, for insn that are not inside a basic block.
|
(Actually, there are gaps, for insn that are not inside a basic block.
|
but we should never see those anyway, so this is OK.) */
|
but we should never see those anyway, so this is OK.) */
|
|
|
max_uid = get_max_uid ();
|
max_uid = get_max_uid ();
|
uid_cuid = gcalloc (max_uid + 1, sizeof (int));
|
uid_cuid = gcalloc (max_uid + 1, sizeof (int));
|
i = 0;
|
i = 0;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
{
|
{
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
uid_cuid[INSN_UID (insn)] = i++;
|
uid_cuid[INSN_UID (insn)] = i++;
|
else
|
else
|
uid_cuid[INSN_UID (insn)] = i;
|
uid_cuid[INSN_UID (insn)] = i;
|
}
|
}
|
|
|
/* Create a table mapping cuids to insns. */
|
/* Create a table mapping cuids to insns. */
|
|
|
max_cuid = i;
|
max_cuid = i;
|
cuid_insn = gcalloc (max_cuid + 1, sizeof (rtx));
|
cuid_insn = gcalloc (max_cuid + 1, sizeof (rtx));
|
i = 0;
|
i = 0;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
CUID_INSN (i++) = insn;
|
CUID_INSN (i++) = insn;
|
|
|
/* Allocate vars to track sets of regs. */
|
/* Allocate vars to track sets of regs. */
|
reg_set_bitmap = BITMAP_ALLOC (NULL);
|
reg_set_bitmap = BITMAP_ALLOC (NULL);
|
|
|
/* Allocate vars to track sets of regs, memory per block. */
|
/* Allocate vars to track sets of regs, memory per block. */
|
reg_set_in_block = sbitmap_vector_alloc (last_basic_block, max_gcse_regno);
|
reg_set_in_block = sbitmap_vector_alloc (last_basic_block, max_gcse_regno);
|
/* Allocate array to keep a list of insns which modify memory in each
|
/* Allocate array to keep a list of insns which modify memory in each
|
basic block. */
|
basic block. */
|
modify_mem_list = gcalloc (last_basic_block, sizeof (rtx));
|
modify_mem_list = gcalloc (last_basic_block, sizeof (rtx));
|
canon_modify_mem_list = gcalloc (last_basic_block, sizeof (rtx));
|
canon_modify_mem_list = gcalloc (last_basic_block, sizeof (rtx));
|
modify_mem_list_set = BITMAP_ALLOC (NULL);
|
modify_mem_list_set = BITMAP_ALLOC (NULL);
|
blocks_with_calls = BITMAP_ALLOC (NULL);
|
blocks_with_calls = BITMAP_ALLOC (NULL);
|
}
|
}
|
|
|
/* Free memory allocated by alloc_gcse_mem. */
|
/* Free memory allocated by alloc_gcse_mem. */
|
|
|
static void
|
static void
|
free_gcse_mem (void)
|
free_gcse_mem (void)
|
{
|
{
|
free (uid_cuid);
|
free (uid_cuid);
|
free (cuid_insn);
|
free (cuid_insn);
|
|
|
BITMAP_FREE (reg_set_bitmap);
|
BITMAP_FREE (reg_set_bitmap);
|
|
|
sbitmap_vector_free (reg_set_in_block);
|
sbitmap_vector_free (reg_set_in_block);
|
free_modify_mem_tables ();
|
free_modify_mem_tables ();
|
BITMAP_FREE (modify_mem_list_set);
|
BITMAP_FREE (modify_mem_list_set);
|
BITMAP_FREE (blocks_with_calls);
|
BITMAP_FREE (blocks_with_calls);
|
}
|
}
|
|
|
/* Compute the local properties of each recorded expression.
|
/* Compute the local properties of each recorded expression.
|
|
|
Local properties are those that are defined by the block, irrespective of
|
Local properties are those that are defined by the block, irrespective of
|
other blocks.
|
other blocks.
|
|
|
An expression is transparent in a block if its operands are not modified
|
An expression is transparent in a block if its operands are not modified
|
in the block.
|
in the block.
|
|
|
An expression is computed (locally available) in a block if it is computed
|
An expression is computed (locally available) in a block if it is computed
|
at least once and expression would contain the same value if the
|
at least once and expression would contain the same value if the
|
computation was moved to the end of the block.
|
computation was moved to the end of the block.
|
|
|
An expression is locally anticipatable in a block if it is computed at
|
An expression is locally anticipatable in a block if it is computed at
|
least once and expression would contain the same value if the computation
|
least once and expression would contain the same value if the computation
|
was moved to the beginning of the block.
|
was moved to the beginning of the block.
|
|
|
We call this routine for cprop, pre and code hoisting. They all compute
|
We call this routine for cprop, pre and code hoisting. They all compute
|
basically the same information and thus can easily share this code.
|
basically the same information and thus can easily share this code.
|
|
|
TRANSP, COMP, and ANTLOC are destination sbitmaps for recording local
|
TRANSP, COMP, and ANTLOC are destination sbitmaps for recording local
|
properties. If NULL, then it is not necessary to compute or record that
|
properties. If NULL, then it is not necessary to compute or record that
|
particular property.
|
particular property.
|
|
|
TABLE controls which hash table to look at. If it is set hash table,
|
TABLE controls which hash table to look at. If it is set hash table,
|
additionally, TRANSP is computed as ~TRANSP, since this is really cprop's
|
additionally, TRANSP is computed as ~TRANSP, since this is really cprop's
|
ABSALTERED. */
|
ABSALTERED. */
|
|
|
static void
|
static void
|
compute_local_properties (sbitmap *transp, sbitmap *comp, sbitmap *antloc,
|
compute_local_properties (sbitmap *transp, sbitmap *comp, sbitmap *antloc,
|
struct hash_table *table)
|
struct hash_table *table)
|
{
|
{
|
unsigned int i;
|
unsigned int i;
|
|
|
/* Initialize any bitmaps that were passed in. */
|
/* Initialize any bitmaps that were passed in. */
|
if (transp)
|
if (transp)
|
{
|
{
|
if (table->set_p)
|
if (table->set_p)
|
sbitmap_vector_zero (transp, last_basic_block);
|
sbitmap_vector_zero (transp, last_basic_block);
|
else
|
else
|
sbitmap_vector_ones (transp, last_basic_block);
|
sbitmap_vector_ones (transp, last_basic_block);
|
}
|
}
|
|
|
if (comp)
|
if (comp)
|
sbitmap_vector_zero (comp, last_basic_block);
|
sbitmap_vector_zero (comp, last_basic_block);
|
if (antloc)
|
if (antloc)
|
sbitmap_vector_zero (antloc, last_basic_block);
|
sbitmap_vector_zero (antloc, last_basic_block);
|
|
|
for (i = 0; i < table->size; i++)
|
for (i = 0; i < table->size; i++)
|
{
|
{
|
struct expr *expr;
|
struct expr *expr;
|
|
|
for (expr = table->table[i]; expr != NULL; expr = expr->next_same_hash)
|
for (expr = table->table[i]; expr != NULL; expr = expr->next_same_hash)
|
{
|
{
|
int indx = expr->bitmap_index;
|
int indx = expr->bitmap_index;
|
struct occr *occr;
|
struct occr *occr;
|
|
|
/* The expression is transparent in this block if it is not killed.
|
/* The expression is transparent in this block if it is not killed.
|
We start by assuming all are transparent [none are killed], and
|
We start by assuming all are transparent [none are killed], and
|
then reset the bits for those that are. */
|
then reset the bits for those that are. */
|
if (transp)
|
if (transp)
|
compute_transp (expr->expr, indx, transp, table->set_p);
|
compute_transp (expr->expr, indx, transp, table->set_p);
|
|
|
/* The occurrences recorded in antic_occr are exactly those that
|
/* The occurrences recorded in antic_occr are exactly those that
|
we want to set to nonzero in ANTLOC. */
|
we want to set to nonzero in ANTLOC. */
|
if (antloc)
|
if (antloc)
|
for (occr = expr->antic_occr; occr != NULL; occr = occr->next)
|
for (occr = expr->antic_occr; occr != NULL; occr = occr->next)
|
{
|
{
|
SET_BIT (antloc[BLOCK_NUM (occr->insn)], indx);
|
SET_BIT (antloc[BLOCK_NUM (occr->insn)], indx);
|
|
|
/* While we're scanning the table, this is a good place to
|
/* While we're scanning the table, this is a good place to
|
initialize this. */
|
initialize this. */
|
occr->deleted_p = 0;
|
occr->deleted_p = 0;
|
}
|
}
|
|
|
/* The occurrences recorded in avail_occr are exactly those that
|
/* The occurrences recorded in avail_occr are exactly those that
|
we want to set to nonzero in COMP. */
|
we want to set to nonzero in COMP. */
|
if (comp)
|
if (comp)
|
for (occr = expr->avail_occr; occr != NULL; occr = occr->next)
|
for (occr = expr->avail_occr; occr != NULL; occr = occr->next)
|
{
|
{
|
SET_BIT (comp[BLOCK_NUM (occr->insn)], indx);
|
SET_BIT (comp[BLOCK_NUM (occr->insn)], indx);
|
|
|
/* While we're scanning the table, this is a good place to
|
/* While we're scanning the table, this is a good place to
|
initialize this. */
|
initialize this. */
|
occr->copied_p = 0;
|
occr->copied_p = 0;
|
}
|
}
|
|
|
/* While we're scanning the table, this is a good place to
|
/* While we're scanning the table, this is a good place to
|
initialize this. */
|
initialize this. */
|
expr->reaching_reg = 0;
|
expr->reaching_reg = 0;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Register set information.
|
/* Register set information.
|
|
|
`reg_set_table' records where each register is set or otherwise
|
`reg_set_table' records where each register is set or otherwise
|
modified. */
|
modified. */
|
|
|
static struct obstack reg_set_obstack;
|
static struct obstack reg_set_obstack;
|
|
|
static void
|
static void
|
alloc_reg_set_mem (int n_regs)
|
alloc_reg_set_mem (int n_regs)
|
{
|
{
|
reg_set_table_size = n_regs + REG_SET_TABLE_SLOP;
|
reg_set_table_size = n_regs + REG_SET_TABLE_SLOP;
|
reg_set_table = gcalloc (reg_set_table_size, sizeof (struct reg_set *));
|
reg_set_table = gcalloc (reg_set_table_size, sizeof (struct reg_set *));
|
|
|
gcc_obstack_init (®_set_obstack);
|
gcc_obstack_init (®_set_obstack);
|
}
|
}
|
|
|
static void
|
static void
|
free_reg_set_mem (void)
|
free_reg_set_mem (void)
|
{
|
{
|
free (reg_set_table);
|
free (reg_set_table);
|
obstack_free (®_set_obstack, NULL);
|
obstack_free (®_set_obstack, NULL);
|
}
|
}
|
|
|
/* Record REGNO in the reg_set table. */
|
/* Record REGNO in the reg_set table. */
|
|
|
static void
|
static void
|
record_one_set (int regno, rtx insn)
|
record_one_set (int regno, rtx insn)
|
{
|
{
|
/* Allocate a new reg_set element and link it onto the list. */
|
/* Allocate a new reg_set element and link it onto the list. */
|
struct reg_set *new_reg_info;
|
struct reg_set *new_reg_info;
|
|
|
/* If the table isn't big enough, enlarge it. */
|
/* If the table isn't big enough, enlarge it. */
|
if (regno >= reg_set_table_size)
|
if (regno >= reg_set_table_size)
|
{
|
{
|
int new_size = regno + REG_SET_TABLE_SLOP;
|
int new_size = regno + REG_SET_TABLE_SLOP;
|
|
|
reg_set_table = grealloc (reg_set_table,
|
reg_set_table = grealloc (reg_set_table,
|
new_size * sizeof (struct reg_set *));
|
new_size * sizeof (struct reg_set *));
|
memset (reg_set_table + reg_set_table_size, 0,
|
memset (reg_set_table + reg_set_table_size, 0,
|
(new_size - reg_set_table_size) * sizeof (struct reg_set *));
|
(new_size - reg_set_table_size) * sizeof (struct reg_set *));
|
reg_set_table_size = new_size;
|
reg_set_table_size = new_size;
|
}
|
}
|
|
|
new_reg_info = obstack_alloc (®_set_obstack, sizeof (struct reg_set));
|
new_reg_info = obstack_alloc (®_set_obstack, sizeof (struct reg_set));
|
bytes_used += sizeof (struct reg_set);
|
bytes_used += sizeof (struct reg_set);
|
new_reg_info->bb_index = BLOCK_NUM (insn);
|
new_reg_info->bb_index = BLOCK_NUM (insn);
|
new_reg_info->next = reg_set_table[regno];
|
new_reg_info->next = reg_set_table[regno];
|
reg_set_table[regno] = new_reg_info;
|
reg_set_table[regno] = new_reg_info;
|
}
|
}
|
|
|
/* Called from compute_sets via note_stores to handle one SET or CLOBBER in
|
/* Called from compute_sets via note_stores to handle one SET or CLOBBER in
|
an insn. The DATA is really the instruction in which the SET is
|
an insn. The DATA is really the instruction in which the SET is
|
occurring. */
|
occurring. */
|
|
|
static void
|
static void
|
record_set_info (rtx dest, rtx setter ATTRIBUTE_UNUSED, void *data)
|
record_set_info (rtx dest, rtx setter ATTRIBUTE_UNUSED, void *data)
|
{
|
{
|
rtx record_set_insn = (rtx) data;
|
rtx record_set_insn = (rtx) data;
|
|
|
if (REG_P (dest) && REGNO (dest) >= FIRST_PSEUDO_REGISTER)
|
if (REG_P (dest) && REGNO (dest) >= FIRST_PSEUDO_REGISTER)
|
record_one_set (REGNO (dest), record_set_insn);
|
record_one_set (REGNO (dest), record_set_insn);
|
}
|
}
|
|
|
/* Scan the function and record each set of each pseudo-register.
|
/* Scan the function and record each set of each pseudo-register.
|
|
|
This is called once, at the start of the gcse pass. See the comments for
|
This is called once, at the start of the gcse pass. See the comments for
|
`reg_set_table' for further documentation. */
|
`reg_set_table' for further documentation. */
|
|
|
static void
|
static void
|
compute_sets (void)
|
compute_sets (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
rtx insn;
|
rtx insn;
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
note_stores (PATTERN (insn), record_set_info, insn);
|
note_stores (PATTERN (insn), record_set_info, insn);
|
}
|
}
|
|
|
/* Hash table support. */
|
/* Hash table support. */
|
|
|
struct reg_avail_info
|
struct reg_avail_info
|
{
|
{
|
basic_block last_bb;
|
basic_block last_bb;
|
int first_set;
|
int first_set;
|
int last_set;
|
int last_set;
|
};
|
};
|
|
|
static struct reg_avail_info *reg_avail_info;
|
static struct reg_avail_info *reg_avail_info;
|
static basic_block current_bb;
|
static basic_block current_bb;
|
|
|
|
|
/* See whether X, the source of a set, is something we want to consider for
|
/* See whether X, the source of a set, is something we want to consider for
|
GCSE. */
|
GCSE. */
|
|
|
static int
|
static int
|
want_to_gcse_p (rtx x)
|
want_to_gcse_p (rtx x)
|
{
|
{
|
#ifdef STACK_REGS
|
#ifdef STACK_REGS
|
/* On register stack architectures, don't GCSE constants from the
|
/* On register stack architectures, don't GCSE constants from the
|
constant pool, as the benefits are often swamped by the overhead
|
constant pool, as the benefits are often swamped by the overhead
|
of shuffling the register stack between basic blocks. */
|
of shuffling the register stack between basic blocks. */
|
if (IS_STACK_MODE (GET_MODE (x)))
|
if (IS_STACK_MODE (GET_MODE (x)))
|
x = avoid_constant_pool_reference (x);
|
x = avoid_constant_pool_reference (x);
|
#endif
|
#endif
|
|
|
switch (GET_CODE (x))
|
switch (GET_CODE (x))
|
{
|
{
|
case REG:
|
case REG:
|
case SUBREG:
|
case SUBREG:
|
case CONST_INT:
|
case CONST_INT:
|
case CONST_DOUBLE:
|
case CONST_DOUBLE:
|
case CONST_VECTOR:
|
case CONST_VECTOR:
|
case CALL:
|
case CALL:
|
return 0;
|
return 0;
|
|
|
default:
|
default:
|
return can_assign_to_reg_p (x);
|
return can_assign_to_reg_p (x);
|
}
|
}
|
}
|
}
|
|
|
/* Used internally by can_assign_to_reg_p. */
|
/* Used internally by can_assign_to_reg_p. */
|
|
|
static GTY(()) rtx test_insn;
|
static GTY(()) rtx test_insn;
|
|
|
/* Return true if we can assign X to a pseudo register. */
|
/* Return true if we can assign X to a pseudo register. */
|
|
|
static bool
|
static bool
|
can_assign_to_reg_p (rtx x)
|
can_assign_to_reg_p (rtx x)
|
{
|
{
|
int num_clobbers = 0;
|
int num_clobbers = 0;
|
int icode;
|
int icode;
|
|
|
/* If this is a valid operand, we are OK. If it's VOIDmode, we aren't. */
|
/* If this is a valid operand, we are OK. If it's VOIDmode, we aren't. */
|
if (general_operand (x, GET_MODE (x)))
|
if (general_operand (x, GET_MODE (x)))
|
return 1;
|
return 1;
|
else if (GET_MODE (x) == VOIDmode)
|
else if (GET_MODE (x) == VOIDmode)
|
return 0;
|
return 0;
|
|
|
/* Otherwise, check if we can make a valid insn from it. First initialize
|
/* Otherwise, check if we can make a valid insn from it. First initialize
|
our test insn if we haven't already. */
|
our test insn if we haven't already. */
|
if (test_insn == 0)
|
if (test_insn == 0)
|
{
|
{
|
test_insn
|
test_insn
|
= make_insn_raw (gen_rtx_SET (VOIDmode,
|
= make_insn_raw (gen_rtx_SET (VOIDmode,
|
gen_rtx_REG (word_mode,
|
gen_rtx_REG (word_mode,
|
FIRST_PSEUDO_REGISTER * 2),
|
FIRST_PSEUDO_REGISTER * 2),
|
const0_rtx));
|
const0_rtx));
|
NEXT_INSN (test_insn) = PREV_INSN (test_insn) = 0;
|
NEXT_INSN (test_insn) = PREV_INSN (test_insn) = 0;
|
}
|
}
|
|
|
/* Now make an insn like the one we would make when GCSE'ing and see if
|
/* Now make an insn like the one we would make when GCSE'ing and see if
|
valid. */
|
valid. */
|
PUT_MODE (SET_DEST (PATTERN (test_insn)), GET_MODE (x));
|
PUT_MODE (SET_DEST (PATTERN (test_insn)), GET_MODE (x));
|
SET_SRC (PATTERN (test_insn)) = x;
|
SET_SRC (PATTERN (test_insn)) = x;
|
return ((icode = recog (PATTERN (test_insn), test_insn, &num_clobbers)) >= 0
|
return ((icode = recog (PATTERN (test_insn), test_insn, &num_clobbers)) >= 0
|
&& (num_clobbers == 0 || ! added_clobbers_hard_reg_p (icode)));
|
&& (num_clobbers == 0 || ! added_clobbers_hard_reg_p (icode)));
|
}
|
}
|
|
|
/* Return nonzero if the operands of expression X are unchanged from the
|
/* Return nonzero if the operands of expression X are unchanged from the
|
start of INSN's basic block up to but not including INSN (if AVAIL_P == 0),
|
start of INSN's basic block up to but not including INSN (if AVAIL_P == 0),
|
or from INSN to the end of INSN's basic block (if AVAIL_P != 0). */
|
or from INSN to the end of INSN's basic block (if AVAIL_P != 0). */
|
|
|
static int
|
static int
|
oprs_unchanged_p (rtx x, rtx insn, int avail_p)
|
oprs_unchanged_p (rtx x, rtx insn, int avail_p)
|
{
|
{
|
int i, j;
|
int i, j;
|
enum rtx_code code;
|
enum rtx_code code;
|
const char *fmt;
|
const char *fmt;
|
|
|
if (x == 0)
|
if (x == 0)
|
return 1;
|
return 1;
|
|
|
code = GET_CODE (x);
|
code = GET_CODE (x);
|
switch (code)
|
switch (code)
|
{
|
{
|
case REG:
|
case REG:
|
{
|
{
|
struct reg_avail_info *info = ®_avail_info[REGNO (x)];
|
struct reg_avail_info *info = ®_avail_info[REGNO (x)];
|
|
|
if (info->last_bb != current_bb)
|
if (info->last_bb != current_bb)
|
return 1;
|
return 1;
|
if (avail_p)
|
if (avail_p)
|
return info->last_set < INSN_CUID (insn);
|
return info->last_set < INSN_CUID (insn);
|
else
|
else
|
return info->first_set >= INSN_CUID (insn);
|
return info->first_set >= INSN_CUID (insn);
|
}
|
}
|
|
|
case MEM:
|
case MEM:
|
if (load_killed_in_block_p (current_bb, INSN_CUID (insn),
|
if (load_killed_in_block_p (current_bb, INSN_CUID (insn),
|
x, avail_p))
|
x, avail_p))
|
return 0;
|
return 0;
|
else
|
else
|
return oprs_unchanged_p (XEXP (x, 0), insn, avail_p);
|
return oprs_unchanged_p (XEXP (x, 0), insn, avail_p);
|
|
|
case PRE_DEC:
|
case PRE_DEC:
|
case PRE_INC:
|
case PRE_INC:
|
case POST_DEC:
|
case POST_DEC:
|
case POST_INC:
|
case POST_INC:
|
case PRE_MODIFY:
|
case PRE_MODIFY:
|
case POST_MODIFY:
|
case POST_MODIFY:
|
return 0;
|
return 0;
|
|
|
case PC:
|
case PC:
|
case CC0: /*FIXME*/
|
case CC0: /*FIXME*/
|
case CONST:
|
case CONST:
|
case CONST_INT:
|
case CONST_INT:
|
case CONST_DOUBLE:
|
case CONST_DOUBLE:
|
case CONST_VECTOR:
|
case CONST_VECTOR:
|
case SYMBOL_REF:
|
case SYMBOL_REF:
|
case LABEL_REF:
|
case LABEL_REF:
|
case ADDR_VEC:
|
case ADDR_VEC:
|
case ADDR_DIFF_VEC:
|
case ADDR_DIFF_VEC:
|
return 1;
|
return 1;
|
|
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
|
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
{
|
{
|
if (fmt[i] == 'e')
|
if (fmt[i] == 'e')
|
{
|
{
|
/* If we are about to do the last recursive call needed at this
|
/* If we are about to do the last recursive call needed at this
|
level, change it into iteration. This function is called enough
|
level, change it into iteration. This function is called enough
|
to be worth it. */
|
to be worth it. */
|
if (i == 0)
|
if (i == 0)
|
return oprs_unchanged_p (XEXP (x, i), insn, avail_p);
|
return oprs_unchanged_p (XEXP (x, i), insn, avail_p);
|
|
|
else if (! oprs_unchanged_p (XEXP (x, i), insn, avail_p))
|
else if (! oprs_unchanged_p (XEXP (x, i), insn, avail_p))
|
return 0;
|
return 0;
|
}
|
}
|
else if (fmt[i] == 'E')
|
else if (fmt[i] == 'E')
|
for (j = 0; j < XVECLEN (x, i); j++)
|
for (j = 0; j < XVECLEN (x, i); j++)
|
if (! oprs_unchanged_p (XVECEXP (x, i, j), insn, avail_p))
|
if (! oprs_unchanged_p (XVECEXP (x, i, j), insn, avail_p))
|
return 0;
|
return 0;
|
}
|
}
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Used for communication between mems_conflict_for_gcse_p and
|
/* Used for communication between mems_conflict_for_gcse_p and
|
load_killed_in_block_p. Nonzero if mems_conflict_for_gcse_p finds a
|
load_killed_in_block_p. Nonzero if mems_conflict_for_gcse_p finds a
|
conflict between two memory references. */
|
conflict between two memory references. */
|
static int gcse_mems_conflict_p;
|
static int gcse_mems_conflict_p;
|
|
|
/* Used for communication between mems_conflict_for_gcse_p and
|
/* Used for communication between mems_conflict_for_gcse_p and
|
load_killed_in_block_p. A memory reference for a load instruction,
|
load_killed_in_block_p. A memory reference for a load instruction,
|
mems_conflict_for_gcse_p will see if a memory store conflicts with
|
mems_conflict_for_gcse_p will see if a memory store conflicts with
|
this memory load. */
|
this memory load. */
|
static rtx gcse_mem_operand;
|
static rtx gcse_mem_operand;
|
|
|
/* DEST is the output of an instruction. If it is a memory reference, and
|
/* DEST is the output of an instruction. If it is a memory reference, and
|
possibly conflicts with the load found in gcse_mem_operand, then set
|
possibly conflicts with the load found in gcse_mem_operand, then set
|
gcse_mems_conflict_p to a nonzero value. */
|
gcse_mems_conflict_p to a nonzero value. */
|
|
|
static void
|
static void
|
mems_conflict_for_gcse_p (rtx dest, rtx setter ATTRIBUTE_UNUSED,
|
mems_conflict_for_gcse_p (rtx dest, rtx setter ATTRIBUTE_UNUSED,
|
void *data ATTRIBUTE_UNUSED)
|
void *data ATTRIBUTE_UNUSED)
|
{
|
{
|
while (GET_CODE (dest) == SUBREG
|
while (GET_CODE (dest) == SUBREG
|
|| GET_CODE (dest) == ZERO_EXTRACT
|
|| GET_CODE (dest) == ZERO_EXTRACT
|
|| GET_CODE (dest) == STRICT_LOW_PART)
|
|| GET_CODE (dest) == STRICT_LOW_PART)
|
dest = XEXP (dest, 0);
|
dest = XEXP (dest, 0);
|
|
|
/* If DEST is not a MEM, then it will not conflict with the load. Note
|
/* If DEST is not a MEM, then it will not conflict with the load. Note
|
that function calls are assumed to clobber memory, but are handled
|
that function calls are assumed to clobber memory, but are handled
|
elsewhere. */
|
elsewhere. */
|
if (! MEM_P (dest))
|
if (! MEM_P (dest))
|
return;
|
return;
|
|
|
/* If we are setting a MEM in our list of specially recognized MEMs,
|
/* If we are setting a MEM in our list of specially recognized MEMs,
|
don't mark as killed this time. */
|
don't mark as killed this time. */
|
|
|
if (expr_equiv_p (dest, gcse_mem_operand) && pre_ldst_mems != NULL)
|
if (expr_equiv_p (dest, gcse_mem_operand) && pre_ldst_mems != NULL)
|
{
|
{
|
if (!find_rtx_in_ldst (dest))
|
if (!find_rtx_in_ldst (dest))
|
gcse_mems_conflict_p = 1;
|
gcse_mems_conflict_p = 1;
|
return;
|
return;
|
}
|
}
|
|
|
if (true_dependence (dest, GET_MODE (dest), gcse_mem_operand,
|
if (true_dependence (dest, GET_MODE (dest), gcse_mem_operand,
|
rtx_addr_varies_p))
|
rtx_addr_varies_p))
|
gcse_mems_conflict_p = 1;
|
gcse_mems_conflict_p = 1;
|
}
|
}
|
|
|
/* Return nonzero if the expression in X (a memory reference) is killed
|
/* Return nonzero if the expression in X (a memory reference) is killed
|
in block BB before or after the insn with the CUID in UID_LIMIT.
|
in block BB before or after the insn with the CUID in UID_LIMIT.
|
AVAIL_P is nonzero for kills after UID_LIMIT, and zero for kills
|
AVAIL_P is nonzero for kills after UID_LIMIT, and zero for kills
|
before UID_LIMIT.
|
before UID_LIMIT.
|
|
|
To check the entire block, set UID_LIMIT to max_uid + 1 and
|
To check the entire block, set UID_LIMIT to max_uid + 1 and
|
AVAIL_P to 0. */
|
AVAIL_P to 0. */
|
|
|
static int
|
static int
|
load_killed_in_block_p (basic_block bb, int uid_limit, rtx x, int avail_p)
|
load_killed_in_block_p (basic_block bb, int uid_limit, rtx x, int avail_p)
|
{
|
{
|
rtx list_entry = modify_mem_list[bb->index];
|
rtx list_entry = modify_mem_list[bb->index];
|
|
|
/* If this is a readonly then we aren't going to be changing it. */
|
/* If this is a readonly then we aren't going to be changing it. */
|
if (MEM_READONLY_P (x))
|
if (MEM_READONLY_P (x))
|
return 0;
|
return 0;
|
|
|
while (list_entry)
|
while (list_entry)
|
{
|
{
|
rtx setter;
|
rtx setter;
|
/* Ignore entries in the list that do not apply. */
|
/* Ignore entries in the list that do not apply. */
|
if ((avail_p
|
if ((avail_p
|
&& INSN_CUID (XEXP (list_entry, 0)) < uid_limit)
|
&& INSN_CUID (XEXP (list_entry, 0)) < uid_limit)
|
|| (! avail_p
|
|| (! avail_p
|
&& INSN_CUID (XEXP (list_entry, 0)) > uid_limit))
|
&& INSN_CUID (XEXP (list_entry, 0)) > uid_limit))
|
{
|
{
|
list_entry = XEXP (list_entry, 1);
|
list_entry = XEXP (list_entry, 1);
|
continue;
|
continue;
|
}
|
}
|
|
|
setter = XEXP (list_entry, 0);
|
setter = XEXP (list_entry, 0);
|
|
|
/* If SETTER is a call everything is clobbered. Note that calls
|
/* If SETTER is a call everything is clobbered. Note that calls
|
to pure functions are never put on the list, so we need not
|
to pure functions are never put on the list, so we need not
|
worry about them. */
|
worry about them. */
|
if (CALL_P (setter))
|
if (CALL_P (setter))
|
return 1;
|
return 1;
|
|
|
/* SETTER must be an INSN of some kind that sets memory. Call
|
/* SETTER must be an INSN of some kind that sets memory. Call
|
note_stores to examine each hunk of memory that is modified.
|
note_stores to examine each hunk of memory that is modified.
|
|
|
The note_stores interface is pretty limited, so we have to
|
The note_stores interface is pretty limited, so we have to
|
communicate via global variables. Yuk. */
|
communicate via global variables. Yuk. */
|
gcse_mem_operand = x;
|
gcse_mem_operand = x;
|
gcse_mems_conflict_p = 0;
|
gcse_mems_conflict_p = 0;
|
note_stores (PATTERN (setter), mems_conflict_for_gcse_p, NULL);
|
note_stores (PATTERN (setter), mems_conflict_for_gcse_p, NULL);
|
if (gcse_mems_conflict_p)
|
if (gcse_mems_conflict_p)
|
return 1;
|
return 1;
|
list_entry = XEXP (list_entry, 1);
|
list_entry = XEXP (list_entry, 1);
|
}
|
}
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* Return nonzero if the operands of expression X are unchanged from
|
/* Return nonzero if the operands of expression X are unchanged from
|
the start of INSN's basic block up to but not including INSN. */
|
the start of INSN's basic block up to but not including INSN. */
|
|
|
static int
|
static int
|
oprs_anticipatable_p (rtx x, rtx insn)
|
oprs_anticipatable_p (rtx x, rtx insn)
|
{
|
{
|
return oprs_unchanged_p (x, insn, 0);
|
return oprs_unchanged_p (x, insn, 0);
|
}
|
}
|
|
|
/* Return nonzero if the operands of expression X are unchanged from
|
/* Return nonzero if the operands of expression X are unchanged from
|
INSN to the end of INSN's basic block. */
|
INSN to the end of INSN's basic block. */
|
|
|
static int
|
static int
|
oprs_available_p (rtx x, rtx insn)
|
oprs_available_p (rtx x, rtx insn)
|
{
|
{
|
return oprs_unchanged_p (x, insn, 1);
|
return oprs_unchanged_p (x, insn, 1);
|
}
|
}
|
|
|
/* Hash expression X.
|
/* Hash expression X.
|
|
|
MODE is only used if X is a CONST_INT. DO_NOT_RECORD_P is a boolean
|
MODE is only used if X is a CONST_INT. DO_NOT_RECORD_P is a boolean
|
indicating if a volatile operand is found or if the expression contains
|
indicating if a volatile operand is found or if the expression contains
|
something we don't want to insert in the table. HASH_TABLE_SIZE is
|
something we don't want to insert in the table. HASH_TABLE_SIZE is
|
the current size of the hash table to be probed. */
|
the current size of the hash table to be probed. */
|
|
|
static unsigned int
|
static unsigned int
|
hash_expr (rtx x, enum machine_mode mode, int *do_not_record_p,
|
hash_expr (rtx x, enum machine_mode mode, int *do_not_record_p,
|
int hash_table_size)
|
int hash_table_size)
|
{
|
{
|
unsigned int hash;
|
unsigned int hash;
|
|
|
*do_not_record_p = 0;
|
*do_not_record_p = 0;
|
|
|
hash = hash_rtx (x, mode, do_not_record_p,
|
hash = hash_rtx (x, mode, do_not_record_p,
|
NULL, /*have_reg_qty=*/false);
|
NULL, /*have_reg_qty=*/false);
|
return hash % hash_table_size;
|
return hash % hash_table_size;
|
}
|
}
|
|
|
/* Hash a set of register REGNO.
|
/* Hash a set of register REGNO.
|
|
|
Sets are hashed on the register that is set. This simplifies the PRE copy
|
Sets are hashed on the register that is set. This simplifies the PRE copy
|
propagation code.
|
propagation code.
|
|
|
??? May need to make things more elaborate. Later, as necessary. */
|
??? May need to make things more elaborate. Later, as necessary. */
|
|
|
static unsigned int
|
static unsigned int
|
hash_set (int regno, int hash_table_size)
|
hash_set (int regno, int hash_table_size)
|
{
|
{
|
unsigned int hash;
|
unsigned int hash;
|
|
|
hash = regno;
|
hash = regno;
|
return hash % hash_table_size;
|
return hash % hash_table_size;
|
}
|
}
|
|
|
/* Return nonzero if exp1 is equivalent to exp2. */
|
/* Return nonzero if exp1 is equivalent to exp2. */
|
|
|
static int
|
static int
|
expr_equiv_p (rtx x, rtx y)
|
expr_equiv_p (rtx x, rtx y)
|
{
|
{
|
return exp_equiv_p (x, y, 0, true);
|
return exp_equiv_p (x, y, 0, true);
|
}
|
}
|
|
|
/* Insert expression X in INSN in the hash TABLE.
|
/* Insert expression X in INSN in the hash TABLE.
|
If it is already present, record it as the last occurrence in INSN's
|
If it is already present, record it as the last occurrence in INSN's
|
basic block.
|
basic block.
|
|
|
MODE is the mode of the value X is being stored into.
|
MODE is the mode of the value X is being stored into.
|
It is only used if X is a CONST_INT.
|
It is only used if X is a CONST_INT.
|
|
|
ANTIC_P is nonzero if X is an anticipatable expression.
|
ANTIC_P is nonzero if X is an anticipatable expression.
|
AVAIL_P is nonzero if X is an available expression. */
|
AVAIL_P is nonzero if X is an available expression. */
|
|
|
static void
|
static void
|
insert_expr_in_table (rtx x, enum machine_mode mode, rtx insn, int antic_p,
|
insert_expr_in_table (rtx x, enum machine_mode mode, rtx insn, int antic_p,
|
int avail_p, struct hash_table *table)
|
int avail_p, struct hash_table *table)
|
{
|
{
|
int found, do_not_record_p;
|
int found, do_not_record_p;
|
unsigned int hash;
|
unsigned int hash;
|
struct expr *cur_expr, *last_expr = NULL;
|
struct expr *cur_expr, *last_expr = NULL;
|
struct occr *antic_occr, *avail_occr;
|
struct occr *antic_occr, *avail_occr;
|
|
|
hash = hash_expr (x, mode, &do_not_record_p, table->size);
|
hash = hash_expr (x, mode, &do_not_record_p, table->size);
|
|
|
/* Do not insert expression in table if it contains volatile operands,
|
/* Do not insert expression in table if it contains volatile operands,
|
or if hash_expr determines the expression is something we don't want
|
or if hash_expr determines the expression is something we don't want
|
to or can't handle. */
|
to or can't handle. */
|
if (do_not_record_p)
|
if (do_not_record_p)
|
return;
|
return;
|
|
|
cur_expr = table->table[hash];
|
cur_expr = table->table[hash];
|
found = 0;
|
found = 0;
|
|
|
while (cur_expr && 0 == (found = expr_equiv_p (cur_expr->expr, x)))
|
while (cur_expr && 0 == (found = expr_equiv_p (cur_expr->expr, x)))
|
{
|
{
|
/* If the expression isn't found, save a pointer to the end of
|
/* If the expression isn't found, save a pointer to the end of
|
the list. */
|
the list. */
|
last_expr = cur_expr;
|
last_expr = cur_expr;
|
cur_expr = cur_expr->next_same_hash;
|
cur_expr = cur_expr->next_same_hash;
|
}
|
}
|
|
|
if (! found)
|
if (! found)
|
{
|
{
|
cur_expr = gcse_alloc (sizeof (struct expr));
|
cur_expr = gcse_alloc (sizeof (struct expr));
|
bytes_used += sizeof (struct expr);
|
bytes_used += sizeof (struct expr);
|
if (table->table[hash] == NULL)
|
if (table->table[hash] == NULL)
|
/* This is the first pattern that hashed to this index. */
|
/* This is the first pattern that hashed to this index. */
|
table->table[hash] = cur_expr;
|
table->table[hash] = cur_expr;
|
else
|
else
|
/* Add EXPR to end of this hash chain. */
|
/* Add EXPR to end of this hash chain. */
|
last_expr->next_same_hash = cur_expr;
|
last_expr->next_same_hash = cur_expr;
|
|
|
/* Set the fields of the expr element. */
|
/* Set the fields of the expr element. */
|
cur_expr->expr = x;
|
cur_expr->expr = x;
|
cur_expr->bitmap_index = table->n_elems++;
|
cur_expr->bitmap_index = table->n_elems++;
|
cur_expr->next_same_hash = NULL;
|
cur_expr->next_same_hash = NULL;
|
cur_expr->antic_occr = NULL;
|
cur_expr->antic_occr = NULL;
|
cur_expr->avail_occr = NULL;
|
cur_expr->avail_occr = NULL;
|
}
|
}
|
|
|
/* Now record the occurrence(s). */
|
/* Now record the occurrence(s). */
|
if (antic_p)
|
if (antic_p)
|
{
|
{
|
antic_occr = cur_expr->antic_occr;
|
antic_occr = cur_expr->antic_occr;
|
|
|
if (antic_occr && BLOCK_NUM (antic_occr->insn) != BLOCK_NUM (insn))
|
if (antic_occr && BLOCK_NUM (antic_occr->insn) != BLOCK_NUM (insn))
|
antic_occr = NULL;
|
antic_occr = NULL;
|
|
|
if (antic_occr)
|
if (antic_occr)
|
/* Found another instance of the expression in the same basic block.
|
/* Found another instance of the expression in the same basic block.
|
Prefer the currently recorded one. We want the first one in the
|
Prefer the currently recorded one. We want the first one in the
|
block and the block is scanned from start to end. */
|
block and the block is scanned from start to end. */
|
; /* nothing to do */
|
; /* nothing to do */
|
else
|
else
|
{
|
{
|
/* First occurrence of this expression in this basic block. */
|
/* First occurrence of this expression in this basic block. */
|
antic_occr = gcse_alloc (sizeof (struct occr));
|
antic_occr = gcse_alloc (sizeof (struct occr));
|
bytes_used += sizeof (struct occr);
|
bytes_used += sizeof (struct occr);
|
antic_occr->insn = insn;
|
antic_occr->insn = insn;
|
antic_occr->next = cur_expr->antic_occr;
|
antic_occr->next = cur_expr->antic_occr;
|
antic_occr->deleted_p = 0;
|
antic_occr->deleted_p = 0;
|
cur_expr->antic_occr = antic_occr;
|
cur_expr->antic_occr = antic_occr;
|
}
|
}
|
}
|
}
|
|
|
if (avail_p)
|
if (avail_p)
|
{
|
{
|
avail_occr = cur_expr->avail_occr;
|
avail_occr = cur_expr->avail_occr;
|
|
|
if (avail_occr && BLOCK_NUM (avail_occr->insn) == BLOCK_NUM (insn))
|
if (avail_occr && BLOCK_NUM (avail_occr->insn) == BLOCK_NUM (insn))
|
{
|
{
|
/* Found another instance of the expression in the same basic block.
|
/* Found another instance of the expression in the same basic block.
|
Prefer this occurrence to the currently recorded one. We want
|
Prefer this occurrence to the currently recorded one. We want
|
the last one in the block and the block is scanned from start
|
the last one in the block and the block is scanned from start
|
to end. */
|
to end. */
|
avail_occr->insn = insn;
|
avail_occr->insn = insn;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* First occurrence of this expression in this basic block. */
|
/* First occurrence of this expression in this basic block. */
|
avail_occr = gcse_alloc (sizeof (struct occr));
|
avail_occr = gcse_alloc (sizeof (struct occr));
|
bytes_used += sizeof (struct occr);
|
bytes_used += sizeof (struct occr);
|
avail_occr->insn = insn;
|
avail_occr->insn = insn;
|
avail_occr->next = cur_expr->avail_occr;
|
avail_occr->next = cur_expr->avail_occr;
|
avail_occr->deleted_p = 0;
|
avail_occr->deleted_p = 0;
|
cur_expr->avail_occr = avail_occr;
|
cur_expr->avail_occr = avail_occr;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Insert pattern X in INSN in the hash table.
|
/* Insert pattern X in INSN in the hash table.
|
X is a SET of a reg to either another reg or a constant.
|
X is a SET of a reg to either another reg or a constant.
|
If it is already present, record it as the last occurrence in INSN's
|
If it is already present, record it as the last occurrence in INSN's
|
basic block. */
|
basic block. */
|
|
|
static void
|
static void
|
insert_set_in_table (rtx x, rtx insn, struct hash_table *table)
|
insert_set_in_table (rtx x, rtx insn, struct hash_table *table)
|
{
|
{
|
int found;
|
int found;
|
unsigned int hash;
|
unsigned int hash;
|
struct expr *cur_expr, *last_expr = NULL;
|
struct expr *cur_expr, *last_expr = NULL;
|
struct occr *cur_occr;
|
struct occr *cur_occr;
|
|
|
gcc_assert (GET_CODE (x) == SET && REG_P (SET_DEST (x)));
|
gcc_assert (GET_CODE (x) == SET && REG_P (SET_DEST (x)));
|
|
|
hash = hash_set (REGNO (SET_DEST (x)), table->size);
|
hash = hash_set (REGNO (SET_DEST (x)), table->size);
|
|
|
cur_expr = table->table[hash];
|
cur_expr = table->table[hash];
|
found = 0;
|
found = 0;
|
|
|
while (cur_expr && 0 == (found = expr_equiv_p (cur_expr->expr, x)))
|
while (cur_expr && 0 == (found = expr_equiv_p (cur_expr->expr, x)))
|
{
|
{
|
/* If the expression isn't found, save a pointer to the end of
|
/* If the expression isn't found, save a pointer to the end of
|
the list. */
|
the list. */
|
last_expr = cur_expr;
|
last_expr = cur_expr;
|
cur_expr = cur_expr->next_same_hash;
|
cur_expr = cur_expr->next_same_hash;
|
}
|
}
|
|
|
if (! found)
|
if (! found)
|
{
|
{
|
cur_expr = gcse_alloc (sizeof (struct expr));
|
cur_expr = gcse_alloc (sizeof (struct expr));
|
bytes_used += sizeof (struct expr);
|
bytes_used += sizeof (struct expr);
|
if (table->table[hash] == NULL)
|
if (table->table[hash] == NULL)
|
/* This is the first pattern that hashed to this index. */
|
/* This is the first pattern that hashed to this index. */
|
table->table[hash] = cur_expr;
|
table->table[hash] = cur_expr;
|
else
|
else
|
/* Add EXPR to end of this hash chain. */
|
/* Add EXPR to end of this hash chain. */
|
last_expr->next_same_hash = cur_expr;
|
last_expr->next_same_hash = cur_expr;
|
|
|
/* Set the fields of the expr element.
|
/* Set the fields of the expr element.
|
We must copy X because it can be modified when copy propagation is
|
We must copy X because it can be modified when copy propagation is
|
performed on its operands. */
|
performed on its operands. */
|
cur_expr->expr = copy_rtx (x);
|
cur_expr->expr = copy_rtx (x);
|
cur_expr->bitmap_index = table->n_elems++;
|
cur_expr->bitmap_index = table->n_elems++;
|
cur_expr->next_same_hash = NULL;
|
cur_expr->next_same_hash = NULL;
|
cur_expr->antic_occr = NULL;
|
cur_expr->antic_occr = NULL;
|
cur_expr->avail_occr = NULL;
|
cur_expr->avail_occr = NULL;
|
}
|
}
|
|
|
/* Now record the occurrence. */
|
/* Now record the occurrence. */
|
cur_occr = cur_expr->avail_occr;
|
cur_occr = cur_expr->avail_occr;
|
|
|
if (cur_occr && BLOCK_NUM (cur_occr->insn) == BLOCK_NUM (insn))
|
if (cur_occr && BLOCK_NUM (cur_occr->insn) == BLOCK_NUM (insn))
|
{
|
{
|
/* Found another instance of the expression in the same basic block.
|
/* Found another instance of the expression in the same basic block.
|
Prefer this occurrence to the currently recorded one. We want
|
Prefer this occurrence to the currently recorded one. We want
|
the last one in the block and the block is scanned from start
|
the last one in the block and the block is scanned from start
|
to end. */
|
to end. */
|
cur_occr->insn = insn;
|
cur_occr->insn = insn;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* First occurrence of this expression in this basic block. */
|
/* First occurrence of this expression in this basic block. */
|
cur_occr = gcse_alloc (sizeof (struct occr));
|
cur_occr = gcse_alloc (sizeof (struct occr));
|
bytes_used += sizeof (struct occr);
|
bytes_used += sizeof (struct occr);
|
|
|
cur_occr->insn = insn;
|
cur_occr->insn = insn;
|
cur_occr->next = cur_expr->avail_occr;
|
cur_occr->next = cur_expr->avail_occr;
|
cur_occr->deleted_p = 0;
|
cur_occr->deleted_p = 0;
|
cur_expr->avail_occr = cur_occr;
|
cur_expr->avail_occr = cur_occr;
|
}
|
}
|
}
|
}
|
|
|
/* Determine whether the rtx X should be treated as a constant for
|
/* Determine whether the rtx X should be treated as a constant for
|
the purposes of GCSE's constant propagation. */
|
the purposes of GCSE's constant propagation. */
|
|
|
static bool
|
static bool
|
gcse_constant_p (rtx x)
|
gcse_constant_p (rtx x)
|
{
|
{
|
/* Consider a COMPARE of two integers constant. */
|
/* Consider a COMPARE of two integers constant. */
|
if (GET_CODE (x) == COMPARE
|
if (GET_CODE (x) == COMPARE
|
&& GET_CODE (XEXP (x, 0)) == CONST_INT
|
&& GET_CODE (XEXP (x, 0)) == CONST_INT
|
&& GET_CODE (XEXP (x, 1)) == CONST_INT)
|
&& GET_CODE (XEXP (x, 1)) == CONST_INT)
|
return true;
|
return true;
|
|
|
/* Consider a COMPARE of the same registers is a constant
|
/* Consider a COMPARE of the same registers is a constant
|
if they are not floating point registers. */
|
if they are not floating point registers. */
|
if (GET_CODE(x) == COMPARE
|
if (GET_CODE(x) == COMPARE
|
&& REG_P (XEXP (x, 0)) && REG_P (XEXP (x, 1))
|
&& REG_P (XEXP (x, 0)) && REG_P (XEXP (x, 1))
|
&& REGNO (XEXP (x, 0)) == REGNO (XEXP (x, 1))
|
&& REGNO (XEXP (x, 0)) == REGNO (XEXP (x, 1))
|
&& ! FLOAT_MODE_P (GET_MODE (XEXP (x, 0)))
|
&& ! FLOAT_MODE_P (GET_MODE (XEXP (x, 0)))
|
&& ! FLOAT_MODE_P (GET_MODE (XEXP (x, 1))))
|
&& ! FLOAT_MODE_P (GET_MODE (XEXP (x, 1))))
|
return true;
|
return true;
|
|
|
return CONSTANT_P (x);
|
return CONSTANT_P (x);
|
}
|
}
|
|
|
/* Scan pattern PAT of INSN and add an entry to the hash TABLE (set or
|
/* Scan pattern PAT of INSN and add an entry to the hash TABLE (set or
|
expression one). */
|
expression one). */
|
|
|
static void
|
static void
|
hash_scan_set (rtx pat, rtx insn, struct hash_table *table)
|
hash_scan_set (rtx pat, rtx insn, struct hash_table *table)
|
{
|
{
|
rtx src = SET_SRC (pat);
|
rtx src = SET_SRC (pat);
|
rtx dest = SET_DEST (pat);
|
rtx dest = SET_DEST (pat);
|
rtx note;
|
rtx note;
|
|
|
if (GET_CODE (src) == CALL)
|
if (GET_CODE (src) == CALL)
|
hash_scan_call (src, insn, table);
|
hash_scan_call (src, insn, table);
|
|
|
else if (REG_P (dest))
|
else if (REG_P (dest))
|
{
|
{
|
unsigned int regno = REGNO (dest);
|
unsigned int regno = REGNO (dest);
|
rtx tmp;
|
rtx tmp;
|
|
|
/* See if a REG_NOTE shows this equivalent to a simpler expression.
|
/* See if a REG_NOTE shows this equivalent to a simpler expression.
|
This allows us to do a single GCSE pass and still eliminate
|
This allows us to do a single GCSE pass and still eliminate
|
redundant constants, addresses or other expressions that are
|
redundant constants, addresses or other expressions that are
|
constructed with multiple instructions. */
|
constructed with multiple instructions. */
|
note = find_reg_equal_equiv_note (insn);
|
note = find_reg_equal_equiv_note (insn);
|
if (note != 0
|
if (note != 0
|
&& (table->set_p
|
&& (table->set_p
|
? gcse_constant_p (XEXP (note, 0))
|
? gcse_constant_p (XEXP (note, 0))
|
: want_to_gcse_p (XEXP (note, 0))))
|
: want_to_gcse_p (XEXP (note, 0))))
|
src = XEXP (note, 0), pat = gen_rtx_SET (VOIDmode, dest, src);
|
src = XEXP (note, 0), pat = gen_rtx_SET (VOIDmode, dest, src);
|
|
|
/* Only record sets of pseudo-regs in the hash table. */
|
/* Only record sets of pseudo-regs in the hash table. */
|
if (! table->set_p
|
if (! table->set_p
|
&& regno >= FIRST_PSEUDO_REGISTER
|
&& regno >= FIRST_PSEUDO_REGISTER
|
/* Don't GCSE something if we can't do a reg/reg copy. */
|
/* Don't GCSE something if we can't do a reg/reg copy. */
|
&& can_copy_p (GET_MODE (dest))
|
&& can_copy_p (GET_MODE (dest))
|
/* GCSE commonly inserts instruction after the insn. We can't
|
/* GCSE commonly inserts instruction after the insn. We can't
|
do that easily for EH_REGION notes so disable GCSE on these
|
do that easily for EH_REGION notes so disable GCSE on these
|
for now. */
|
for now. */
|
&& !find_reg_note (insn, REG_EH_REGION, NULL_RTX)
|
&& !find_reg_note (insn, REG_EH_REGION, NULL_RTX)
|
/* Is SET_SRC something we want to gcse? */
|
/* Is SET_SRC something we want to gcse? */
|
&& want_to_gcse_p (src)
|
&& want_to_gcse_p (src)
|
/* Don't CSE a nop. */
|
/* Don't CSE a nop. */
|
&& ! set_noop_p (pat)
|
&& ! set_noop_p (pat)
|
/* Don't GCSE if it has attached REG_EQUIV note.
|
/* Don't GCSE if it has attached REG_EQUIV note.
|
At this point this only function parameters should have
|
At this point this only function parameters should have
|
REG_EQUIV notes and if the argument slot is used somewhere
|
REG_EQUIV notes and if the argument slot is used somewhere
|
explicitly, it means address of parameter has been taken,
|
explicitly, it means address of parameter has been taken,
|
so we should not extend the lifetime of the pseudo. */
|
so we should not extend the lifetime of the pseudo. */
|
&& (note == NULL_RTX || ! MEM_P (XEXP (note, 0))))
|
&& (note == NULL_RTX || ! MEM_P (XEXP (note, 0))))
|
{
|
{
|
/* An expression is not anticipatable if its operands are
|
/* An expression is not anticipatable if its operands are
|
modified before this insn or if this is not the only SET in
|
modified before this insn or if this is not the only SET in
|
this insn. */
|
this insn. */
|
int antic_p = oprs_anticipatable_p (src, insn) && single_set (insn);
|
int antic_p = oprs_anticipatable_p (src, insn) && single_set (insn);
|
/* An expression is not available if its operands are
|
/* An expression is not available if its operands are
|
subsequently modified, including this insn. It's also not
|
subsequently modified, including this insn. It's also not
|
available if this is a branch, because we can't insert
|
available if this is a branch, because we can't insert
|
a set after the branch. */
|
a set after the branch. */
|
int avail_p = (oprs_available_p (src, insn)
|
int avail_p = (oprs_available_p (src, insn)
|
&& ! JUMP_P (insn));
|
&& ! JUMP_P (insn));
|
|
|
insert_expr_in_table (src, GET_MODE (dest), insn, antic_p, avail_p, table);
|
insert_expr_in_table (src, GET_MODE (dest), insn, antic_p, avail_p, table);
|
}
|
}
|
|
|
/* Record sets for constant/copy propagation. */
|
/* Record sets for constant/copy propagation. */
|
else if (table->set_p
|
else if (table->set_p
|
&& regno >= FIRST_PSEUDO_REGISTER
|
&& regno >= FIRST_PSEUDO_REGISTER
|
&& ((REG_P (src)
|
&& ((REG_P (src)
|
&& REGNO (src) >= FIRST_PSEUDO_REGISTER
|
&& REGNO (src) >= FIRST_PSEUDO_REGISTER
|
&& can_copy_p (GET_MODE (dest))
|
&& can_copy_p (GET_MODE (dest))
|
&& REGNO (src) != regno)
|
&& REGNO (src) != regno)
|
|| gcse_constant_p (src))
|
|| gcse_constant_p (src))
|
/* A copy is not available if its src or dest is subsequently
|
/* A copy is not available if its src or dest is subsequently
|
modified. Here we want to search from INSN+1 on, but
|
modified. Here we want to search from INSN+1 on, but
|
oprs_available_p searches from INSN on. */
|
oprs_available_p searches from INSN on. */
|
&& (insn == BB_END (BLOCK_FOR_INSN (insn))
|
&& (insn == BB_END (BLOCK_FOR_INSN (insn))
|
|| ((tmp = next_nonnote_insn (insn)) != NULL_RTX
|
|| ((tmp = next_nonnote_insn (insn)) != NULL_RTX
|
&& oprs_available_p (pat, tmp))))
|
&& oprs_available_p (pat, tmp))))
|
insert_set_in_table (pat, insn, table);
|
insert_set_in_table (pat, insn, table);
|
}
|
}
|
/* In case of store we want to consider the memory value as available in
|
/* In case of store we want to consider the memory value as available in
|
the REG stored in that memory. This makes it possible to remove
|
the REG stored in that memory. This makes it possible to remove
|
redundant loads from due to stores to the same location. */
|
redundant loads from due to stores to the same location. */
|
else if (flag_gcse_las && REG_P (src) && MEM_P (dest))
|
else if (flag_gcse_las && REG_P (src) && MEM_P (dest))
|
{
|
{
|
unsigned int regno = REGNO (src);
|
unsigned int regno = REGNO (src);
|
|
|
/* Do not do this for constant/copy propagation. */
|
/* Do not do this for constant/copy propagation. */
|
if (! table->set_p
|
if (! table->set_p
|
/* Only record sets of pseudo-regs in the hash table. */
|
/* Only record sets of pseudo-regs in the hash table. */
|
&& regno >= FIRST_PSEUDO_REGISTER
|
&& regno >= FIRST_PSEUDO_REGISTER
|
/* Don't GCSE something if we can't do a reg/reg copy. */
|
/* Don't GCSE something if we can't do a reg/reg copy. */
|
&& can_copy_p (GET_MODE (src))
|
&& can_copy_p (GET_MODE (src))
|
/* GCSE commonly inserts instruction after the insn. We can't
|
/* GCSE commonly inserts instruction after the insn. We can't
|
do that easily for EH_REGION notes so disable GCSE on these
|
do that easily for EH_REGION notes so disable GCSE on these
|
for now. */
|
for now. */
|
&& ! find_reg_note (insn, REG_EH_REGION, NULL_RTX)
|
&& ! find_reg_note (insn, REG_EH_REGION, NULL_RTX)
|
/* Is SET_DEST something we want to gcse? */
|
/* Is SET_DEST something we want to gcse? */
|
&& want_to_gcse_p (dest)
|
&& want_to_gcse_p (dest)
|
/* Don't CSE a nop. */
|
/* Don't CSE a nop. */
|
&& ! set_noop_p (pat)
|
&& ! set_noop_p (pat)
|
/* Don't GCSE if it has attached REG_EQUIV note.
|
/* Don't GCSE if it has attached REG_EQUIV note.
|
At this point this only function parameters should have
|
At this point this only function parameters should have
|
REG_EQUIV notes and if the argument slot is used somewhere
|
REG_EQUIV notes and if the argument slot is used somewhere
|
explicitly, it means address of parameter has been taken,
|
explicitly, it means address of parameter has been taken,
|
so we should not extend the lifetime of the pseudo. */
|
so we should not extend the lifetime of the pseudo. */
|
&& ((note = find_reg_note (insn, REG_EQUIV, NULL_RTX)) == 0
|
&& ((note = find_reg_note (insn, REG_EQUIV, NULL_RTX)) == 0
|
|| ! MEM_P (XEXP (note, 0))))
|
|| ! MEM_P (XEXP (note, 0))))
|
{
|
{
|
/* Stores are never anticipatable. */
|
/* Stores are never anticipatable. */
|
int antic_p = 0;
|
int antic_p = 0;
|
/* An expression is not available if its operands are
|
/* An expression is not available if its operands are
|
subsequently modified, including this insn. It's also not
|
subsequently modified, including this insn. It's also not
|
available if this is a branch, because we can't insert
|
available if this is a branch, because we can't insert
|
a set after the branch. */
|
a set after the branch. */
|
int avail_p = oprs_available_p (dest, insn)
|
int avail_p = oprs_available_p (dest, insn)
|
&& ! JUMP_P (insn);
|
&& ! JUMP_P (insn);
|
|
|
/* Record the memory expression (DEST) in the hash table. */
|
/* Record the memory expression (DEST) in the hash table. */
|
insert_expr_in_table (dest, GET_MODE (dest), insn,
|
insert_expr_in_table (dest, GET_MODE (dest), insn,
|
antic_p, avail_p, table);
|
antic_p, avail_p, table);
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
static void
|
static void
|
hash_scan_clobber (rtx x ATTRIBUTE_UNUSED, rtx insn ATTRIBUTE_UNUSED,
|
hash_scan_clobber (rtx x ATTRIBUTE_UNUSED, rtx insn ATTRIBUTE_UNUSED,
|
struct hash_table *table ATTRIBUTE_UNUSED)
|
struct hash_table *table ATTRIBUTE_UNUSED)
|
{
|
{
|
/* Currently nothing to do. */
|
/* Currently nothing to do. */
|
}
|
}
|
|
|
static void
|
static void
|
hash_scan_call (rtx x ATTRIBUTE_UNUSED, rtx insn ATTRIBUTE_UNUSED,
|
hash_scan_call (rtx x ATTRIBUTE_UNUSED, rtx insn ATTRIBUTE_UNUSED,
|
struct hash_table *table ATTRIBUTE_UNUSED)
|
struct hash_table *table ATTRIBUTE_UNUSED)
|
{
|
{
|
/* Currently nothing to do. */
|
/* Currently nothing to do. */
|
}
|
}
|
|
|
/* Process INSN and add hash table entries as appropriate.
|
/* Process INSN and add hash table entries as appropriate.
|
|
|
Only available expressions that set a single pseudo-reg are recorded.
|
Only available expressions that set a single pseudo-reg are recorded.
|
|
|
Single sets in a PARALLEL could be handled, but it's an extra complication
|
Single sets in a PARALLEL could be handled, but it's an extra complication
|
that isn't dealt with right now. The trick is handling the CLOBBERs that
|
that isn't dealt with right now. The trick is handling the CLOBBERs that
|
are also in the PARALLEL. Later.
|
are also in the PARALLEL. Later.
|
|
|
If SET_P is nonzero, this is for the assignment hash table,
|
If SET_P is nonzero, this is for the assignment hash table,
|
otherwise it is for the expression hash table.
|
otherwise it is for the expression hash table.
|
If IN_LIBCALL_BLOCK nonzero, we are in a libcall block, and should
|
If IN_LIBCALL_BLOCK nonzero, we are in a libcall block, and should
|
not record any expressions. */
|
not record any expressions. */
|
|
|
static void
|
static void
|
hash_scan_insn (rtx insn, struct hash_table *table, int in_libcall_block)
|
hash_scan_insn (rtx insn, struct hash_table *table, int in_libcall_block)
|
{
|
{
|
rtx pat = PATTERN (insn);
|
rtx pat = PATTERN (insn);
|
int i;
|
int i;
|
|
|
if (in_libcall_block)
|
if (in_libcall_block)
|
return;
|
return;
|
|
|
/* Pick out the sets of INSN and for other forms of instructions record
|
/* Pick out the sets of INSN and for other forms of instructions record
|
what's been modified. */
|
what's been modified. */
|
|
|
if (GET_CODE (pat) == SET)
|
if (GET_CODE (pat) == SET)
|
hash_scan_set (pat, insn, table);
|
hash_scan_set (pat, insn, table);
|
else if (GET_CODE (pat) == PARALLEL)
|
else if (GET_CODE (pat) == PARALLEL)
|
for (i = 0; i < XVECLEN (pat, 0); i++)
|
for (i = 0; i < XVECLEN (pat, 0); i++)
|
{
|
{
|
rtx x = XVECEXP (pat, 0, i);
|
rtx x = XVECEXP (pat, 0, i);
|
|
|
if (GET_CODE (x) == SET)
|
if (GET_CODE (x) == SET)
|
hash_scan_set (x, insn, table);
|
hash_scan_set (x, insn, table);
|
else if (GET_CODE (x) == CLOBBER)
|
else if (GET_CODE (x) == CLOBBER)
|
hash_scan_clobber (x, insn, table);
|
hash_scan_clobber (x, insn, table);
|
else if (GET_CODE (x) == CALL)
|
else if (GET_CODE (x) == CALL)
|
hash_scan_call (x, insn, table);
|
hash_scan_call (x, insn, table);
|
}
|
}
|
|
|
else if (GET_CODE (pat) == CLOBBER)
|
else if (GET_CODE (pat) == CLOBBER)
|
hash_scan_clobber (pat, insn, table);
|
hash_scan_clobber (pat, insn, table);
|
else if (GET_CODE (pat) == CALL)
|
else if (GET_CODE (pat) == CALL)
|
hash_scan_call (pat, insn, table);
|
hash_scan_call (pat, insn, table);
|
}
|
}
|
|
|
static void
|
static void
|
dump_hash_table (FILE *file, const char *name, struct hash_table *table)
|
dump_hash_table (FILE *file, const char *name, struct hash_table *table)
|
{
|
{
|
int i;
|
int i;
|
/* Flattened out table, so it's printed in proper order. */
|
/* Flattened out table, so it's printed in proper order. */
|
struct expr **flat_table;
|
struct expr **flat_table;
|
unsigned int *hash_val;
|
unsigned int *hash_val;
|
struct expr *expr;
|
struct expr *expr;
|
|
|
flat_table = xcalloc (table->n_elems, sizeof (struct expr *));
|
flat_table = xcalloc (table->n_elems, sizeof (struct expr *));
|
hash_val = xmalloc (table->n_elems * sizeof (unsigned int));
|
hash_val = xmalloc (table->n_elems * sizeof (unsigned int));
|
|
|
for (i = 0; i < (int) table->size; i++)
|
for (i = 0; i < (int) table->size; i++)
|
for (expr = table->table[i]; expr != NULL; expr = expr->next_same_hash)
|
for (expr = table->table[i]; expr != NULL; expr = expr->next_same_hash)
|
{
|
{
|
flat_table[expr->bitmap_index] = expr;
|
flat_table[expr->bitmap_index] = expr;
|
hash_val[expr->bitmap_index] = i;
|
hash_val[expr->bitmap_index] = i;
|
}
|
}
|
|
|
fprintf (file, "%s hash table (%d buckets, %d entries)\n",
|
fprintf (file, "%s hash table (%d buckets, %d entries)\n",
|
name, table->size, table->n_elems);
|
name, table->size, table->n_elems);
|
|
|
for (i = 0; i < (int) table->n_elems; i++)
|
for (i = 0; i < (int) table->n_elems; i++)
|
if (flat_table[i] != 0)
|
if (flat_table[i] != 0)
|
{
|
{
|
expr = flat_table[i];
|
expr = flat_table[i];
|
fprintf (file, "Index %d (hash value %d)\n ",
|
fprintf (file, "Index %d (hash value %d)\n ",
|
expr->bitmap_index, hash_val[i]);
|
expr->bitmap_index, hash_val[i]);
|
print_rtl (file, expr->expr);
|
print_rtl (file, expr->expr);
|
fprintf (file, "\n");
|
fprintf (file, "\n");
|
}
|
}
|
|
|
fprintf (file, "\n");
|
fprintf (file, "\n");
|
|
|
free (flat_table);
|
free (flat_table);
|
free (hash_val);
|
free (hash_val);
|
}
|
}
|
|
|
/* Record register first/last/block set information for REGNO in INSN.
|
/* Record register first/last/block set information for REGNO in INSN.
|
|
|
first_set records the first place in the block where the register
|
first_set records the first place in the block where the register
|
is set and is used to compute "anticipatability".
|
is set and is used to compute "anticipatability".
|
|
|
last_set records the last place in the block where the register
|
last_set records the last place in the block where the register
|
is set and is used to compute "availability".
|
is set and is used to compute "availability".
|
|
|
last_bb records the block for which first_set and last_set are
|
last_bb records the block for which first_set and last_set are
|
valid, as a quick test to invalidate them.
|
valid, as a quick test to invalidate them.
|
|
|
reg_set_in_block records whether the register is set in the block
|
reg_set_in_block records whether the register is set in the block
|
and is used to compute "transparency". */
|
and is used to compute "transparency". */
|
|
|
static void
|
static void
|
record_last_reg_set_info (rtx insn, int regno)
|
record_last_reg_set_info (rtx insn, int regno)
|
{
|
{
|
struct reg_avail_info *info = ®_avail_info[regno];
|
struct reg_avail_info *info = ®_avail_info[regno];
|
int cuid = INSN_CUID (insn);
|
int cuid = INSN_CUID (insn);
|
|
|
info->last_set = cuid;
|
info->last_set = cuid;
|
if (info->last_bb != current_bb)
|
if (info->last_bb != current_bb)
|
{
|
{
|
info->last_bb = current_bb;
|
info->last_bb = current_bb;
|
info->first_set = cuid;
|
info->first_set = cuid;
|
SET_BIT (reg_set_in_block[current_bb->index], regno);
|
SET_BIT (reg_set_in_block[current_bb->index], regno);
|
}
|
}
|
}
|
}
|
|
|
|
|
/* Record all of the canonicalized MEMs of record_last_mem_set_info's insn.
|
/* Record all of the canonicalized MEMs of record_last_mem_set_info's insn.
|
Note we store a pair of elements in the list, so they have to be
|
Note we store a pair of elements in the list, so they have to be
|
taken off pairwise. */
|
taken off pairwise. */
|
|
|
static void
|
static void
|
canon_list_insert (rtx dest ATTRIBUTE_UNUSED, rtx unused1 ATTRIBUTE_UNUSED,
|
canon_list_insert (rtx dest ATTRIBUTE_UNUSED, rtx unused1 ATTRIBUTE_UNUSED,
|
void * v_insn)
|
void * v_insn)
|
{
|
{
|
rtx dest_addr, insn;
|
rtx dest_addr, insn;
|
int bb;
|
int bb;
|
|
|
while (GET_CODE (dest) == SUBREG
|
while (GET_CODE (dest) == SUBREG
|
|| GET_CODE (dest) == ZERO_EXTRACT
|
|| GET_CODE (dest) == ZERO_EXTRACT
|
|| GET_CODE (dest) == STRICT_LOW_PART)
|
|| GET_CODE (dest) == STRICT_LOW_PART)
|
dest = XEXP (dest, 0);
|
dest = XEXP (dest, 0);
|
|
|
/* If DEST is not a MEM, then it will not conflict with a load. Note
|
/* If DEST is not a MEM, then it will not conflict with a load. Note
|
that function calls are assumed to clobber memory, but are handled
|
that function calls are assumed to clobber memory, but are handled
|
elsewhere. */
|
elsewhere. */
|
|
|
if (! MEM_P (dest))
|
if (! MEM_P (dest))
|
return;
|
return;
|
|
|
dest_addr = get_addr (XEXP (dest, 0));
|
dest_addr = get_addr (XEXP (dest, 0));
|
dest_addr = canon_rtx (dest_addr);
|
dest_addr = canon_rtx (dest_addr);
|
insn = (rtx) v_insn;
|
insn = (rtx) v_insn;
|
bb = BLOCK_NUM (insn);
|
bb = BLOCK_NUM (insn);
|
|
|
canon_modify_mem_list[bb] =
|
canon_modify_mem_list[bb] =
|
alloc_EXPR_LIST (VOIDmode, dest_addr, canon_modify_mem_list[bb]);
|
alloc_EXPR_LIST (VOIDmode, dest_addr, canon_modify_mem_list[bb]);
|
canon_modify_mem_list[bb] =
|
canon_modify_mem_list[bb] =
|
alloc_EXPR_LIST (VOIDmode, dest, canon_modify_mem_list[bb]);
|
alloc_EXPR_LIST (VOIDmode, dest, canon_modify_mem_list[bb]);
|
}
|
}
|
|
|
/* Record memory modification information for INSN. We do not actually care
|
/* Record memory modification information for INSN. We do not actually care
|
about the memory location(s) that are set, or even how they are set (consider
|
about the memory location(s) that are set, or even how they are set (consider
|
a CALL_INSN). We merely need to record which insns modify memory. */
|
a CALL_INSN). We merely need to record which insns modify memory. */
|
|
|
static void
|
static void
|
record_last_mem_set_info (rtx insn)
|
record_last_mem_set_info (rtx insn)
|
{
|
{
|
int bb = BLOCK_NUM (insn);
|
int bb = BLOCK_NUM (insn);
|
|
|
/* load_killed_in_block_p will handle the case of calls clobbering
|
/* load_killed_in_block_p will handle the case of calls clobbering
|
everything. */
|
everything. */
|
modify_mem_list[bb] = alloc_INSN_LIST (insn, modify_mem_list[bb]);
|
modify_mem_list[bb] = alloc_INSN_LIST (insn, modify_mem_list[bb]);
|
bitmap_set_bit (modify_mem_list_set, bb);
|
bitmap_set_bit (modify_mem_list_set, bb);
|
|
|
if (CALL_P (insn))
|
if (CALL_P (insn))
|
{
|
{
|
/* Note that traversals of this loop (other than for free-ing)
|
/* Note that traversals of this loop (other than for free-ing)
|
will break after encountering a CALL_INSN. So, there's no
|
will break after encountering a CALL_INSN. So, there's no
|
need to insert a pair of items, as canon_list_insert does. */
|
need to insert a pair of items, as canon_list_insert does. */
|
canon_modify_mem_list[bb] =
|
canon_modify_mem_list[bb] =
|
alloc_INSN_LIST (insn, canon_modify_mem_list[bb]);
|
alloc_INSN_LIST (insn, canon_modify_mem_list[bb]);
|
bitmap_set_bit (blocks_with_calls, bb);
|
bitmap_set_bit (blocks_with_calls, bb);
|
}
|
}
|
else
|
else
|
note_stores (PATTERN (insn), canon_list_insert, (void*) insn);
|
note_stores (PATTERN (insn), canon_list_insert, (void*) insn);
|
}
|
}
|
|
|
/* Called from compute_hash_table via note_stores to handle one
|
/* Called from compute_hash_table via note_stores to handle one
|
SET or CLOBBER in an insn. DATA is really the instruction in which
|
SET or CLOBBER in an insn. DATA is really the instruction in which
|
the SET is taking place. */
|
the SET is taking place. */
|
|
|
static void
|
static void
|
record_last_set_info (rtx dest, rtx setter ATTRIBUTE_UNUSED, void *data)
|
record_last_set_info (rtx dest, rtx setter ATTRIBUTE_UNUSED, void *data)
|
{
|
{
|
rtx last_set_insn = (rtx) data;
|
rtx last_set_insn = (rtx) data;
|
|
|
if (GET_CODE (dest) == SUBREG)
|
if (GET_CODE (dest) == SUBREG)
|
dest = SUBREG_REG (dest);
|
dest = SUBREG_REG (dest);
|
|
|
if (REG_P (dest))
|
if (REG_P (dest))
|
record_last_reg_set_info (last_set_insn, REGNO (dest));
|
record_last_reg_set_info (last_set_insn, REGNO (dest));
|
else if (MEM_P (dest)
|
else if (MEM_P (dest)
|
/* Ignore pushes, they clobber nothing. */
|
/* Ignore pushes, they clobber nothing. */
|
&& ! push_operand (dest, GET_MODE (dest)))
|
&& ! push_operand (dest, GET_MODE (dest)))
|
record_last_mem_set_info (last_set_insn);
|
record_last_mem_set_info (last_set_insn);
|
}
|
}
|
|
|
/* Top level function to create an expression or assignment hash table.
|
/* Top level function to create an expression or assignment hash table.
|
|
|
Expression entries are placed in the hash table if
|
Expression entries are placed in the hash table if
|
- they are of the form (set (pseudo-reg) src),
|
- they are of the form (set (pseudo-reg) src),
|
- src is something we want to perform GCSE on,
|
- src is something we want to perform GCSE on,
|
- none of the operands are subsequently modified in the block
|
- none of the operands are subsequently modified in the block
|
|
|
Assignment entries are placed in the hash table if
|
Assignment entries are placed in the hash table if
|
- they are of the form (set (pseudo-reg) src),
|
- they are of the form (set (pseudo-reg) src),
|
- src is something we want to perform const/copy propagation on,
|
- src is something we want to perform const/copy propagation on,
|
- none of the operands or target are subsequently modified in the block
|
- none of the operands or target are subsequently modified in the block
|
|
|
Currently src must be a pseudo-reg or a const_int.
|
Currently src must be a pseudo-reg or a const_int.
|
|
|
TABLE is the table computed. */
|
TABLE is the table computed. */
|
|
|
static void
|
static void
|
compute_hash_table_work (struct hash_table *table)
|
compute_hash_table_work (struct hash_table *table)
|
{
|
{
|
unsigned int i;
|
unsigned int i;
|
|
|
/* While we compute the hash table we also compute a bit array of which
|
/* While we compute the hash table we also compute a bit array of which
|
registers are set in which blocks.
|
registers are set in which blocks.
|
??? This isn't needed during const/copy propagation, but it's cheap to
|
??? This isn't needed during const/copy propagation, but it's cheap to
|
compute. Later. */
|
compute. Later. */
|
sbitmap_vector_zero (reg_set_in_block, last_basic_block);
|
sbitmap_vector_zero (reg_set_in_block, last_basic_block);
|
|
|
/* re-Cache any INSN_LIST nodes we have allocated. */
|
/* re-Cache any INSN_LIST nodes we have allocated. */
|
clear_modify_mem_tables ();
|
clear_modify_mem_tables ();
|
/* Some working arrays used to track first and last set in each block. */
|
/* Some working arrays used to track first and last set in each block. */
|
reg_avail_info = gmalloc (max_gcse_regno * sizeof (struct reg_avail_info));
|
reg_avail_info = gmalloc (max_gcse_regno * sizeof (struct reg_avail_info));
|
|
|
for (i = 0; i < max_gcse_regno; ++i)
|
for (i = 0; i < max_gcse_regno; ++i)
|
reg_avail_info[i].last_bb = NULL;
|
reg_avail_info[i].last_bb = NULL;
|
|
|
FOR_EACH_BB (current_bb)
|
FOR_EACH_BB (current_bb)
|
{
|
{
|
rtx insn;
|
rtx insn;
|
unsigned int regno;
|
unsigned int regno;
|
int in_libcall_block;
|
int in_libcall_block;
|
|
|
/* First pass over the instructions records information used to
|
/* First pass over the instructions records information used to
|
determine when registers and memory are first and last set.
|
determine when registers and memory are first and last set.
|
??? hard-reg reg_set_in_block computation
|
??? hard-reg reg_set_in_block computation
|
could be moved to compute_sets since they currently don't change. */
|
could be moved to compute_sets since they currently don't change. */
|
|
|
FOR_BB_INSNS (current_bb, insn)
|
FOR_BB_INSNS (current_bb, insn)
|
{
|
{
|
if (! INSN_P (insn))
|
if (! INSN_P (insn))
|
continue;
|
continue;
|
|
|
if (CALL_P (insn))
|
if (CALL_P (insn))
|
{
|
{
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
|
record_last_reg_set_info (insn, regno);
|
record_last_reg_set_info (insn, regno);
|
|
|
mark_call (insn);
|
mark_call (insn);
|
}
|
}
|
|
|
note_stores (PATTERN (insn), record_last_set_info, insn);
|
note_stores (PATTERN (insn), record_last_set_info, insn);
|
}
|
}
|
|
|
/* Insert implicit sets in the hash table. */
|
/* Insert implicit sets in the hash table. */
|
if (table->set_p
|
if (table->set_p
|
&& implicit_sets[current_bb->index] != NULL_RTX)
|
&& implicit_sets[current_bb->index] != NULL_RTX)
|
hash_scan_set (implicit_sets[current_bb->index],
|
hash_scan_set (implicit_sets[current_bb->index],
|
BB_HEAD (current_bb), table);
|
BB_HEAD (current_bb), table);
|
|
|
/* The next pass builds the hash table. */
|
/* The next pass builds the hash table. */
|
in_libcall_block = 0;
|
in_libcall_block = 0;
|
FOR_BB_INSNS (current_bb, insn)
|
FOR_BB_INSNS (current_bb, insn)
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
{
|
{
|
if (find_reg_note (insn, REG_LIBCALL, NULL_RTX))
|
if (find_reg_note (insn, REG_LIBCALL, NULL_RTX))
|
in_libcall_block = 1;
|
in_libcall_block = 1;
|
else if (table->set_p && find_reg_note (insn, REG_RETVAL, NULL_RTX))
|
else if (table->set_p && find_reg_note (insn, REG_RETVAL, NULL_RTX))
|
in_libcall_block = 0;
|
in_libcall_block = 0;
|
hash_scan_insn (insn, table, in_libcall_block);
|
hash_scan_insn (insn, table, in_libcall_block);
|
if (!table->set_p && find_reg_note (insn, REG_RETVAL, NULL_RTX))
|
if (!table->set_p && find_reg_note (insn, REG_RETVAL, NULL_RTX))
|
in_libcall_block = 0;
|
in_libcall_block = 0;
|
}
|
}
|
}
|
}
|
|
|
free (reg_avail_info);
|
free (reg_avail_info);
|
reg_avail_info = NULL;
|
reg_avail_info = NULL;
|
}
|
}
|
|
|
/* Allocate space for the set/expr hash TABLE.
|
/* Allocate space for the set/expr hash TABLE.
|
N_INSNS is the number of instructions in the function.
|
N_INSNS is the number of instructions in the function.
|
It is used to determine the number of buckets to use.
|
It is used to determine the number of buckets to use.
|
SET_P determines whether set or expression table will
|
SET_P determines whether set or expression table will
|
be created. */
|
be created. */
|
|
|
static void
|
static void
|
alloc_hash_table (int n_insns, struct hash_table *table, int set_p)
|
alloc_hash_table (int n_insns, struct hash_table *table, int set_p)
|
{
|
{
|
int n;
|
int n;
|
|
|
table->size = n_insns / 4;
|
table->size = n_insns / 4;
|
if (table->size < 11)
|
if (table->size < 11)
|
table->size = 11;
|
table->size = 11;
|
|
|
/* Attempt to maintain efficient use of hash table.
|
/* Attempt to maintain efficient use of hash table.
|
Making it an odd number is simplest for now.
|
Making it an odd number is simplest for now.
|
??? Later take some measurements. */
|
??? Later take some measurements. */
|
table->size |= 1;
|
table->size |= 1;
|
n = table->size * sizeof (struct expr *);
|
n = table->size * sizeof (struct expr *);
|
table->table = gmalloc (n);
|
table->table = gmalloc (n);
|
table->set_p = set_p;
|
table->set_p = set_p;
|
}
|
}
|
|
|
/* Free things allocated by alloc_hash_table. */
|
/* Free things allocated by alloc_hash_table. */
|
|
|
static void
|
static void
|
free_hash_table (struct hash_table *table)
|
free_hash_table (struct hash_table *table)
|
{
|
{
|
free (table->table);
|
free (table->table);
|
}
|
}
|
|
|
/* Compute the hash TABLE for doing copy/const propagation or
|
/* Compute the hash TABLE for doing copy/const propagation or
|
expression hash table. */
|
expression hash table. */
|
|
|
static void
|
static void
|
compute_hash_table (struct hash_table *table)
|
compute_hash_table (struct hash_table *table)
|
{
|
{
|
/* Initialize count of number of entries in hash table. */
|
/* Initialize count of number of entries in hash table. */
|
table->n_elems = 0;
|
table->n_elems = 0;
|
memset (table->table, 0, table->size * sizeof (struct expr *));
|
memset (table->table, 0, table->size * sizeof (struct expr *));
|
|
|
compute_hash_table_work (table);
|
compute_hash_table_work (table);
|
}
|
}
|
|
|
/* Expression tracking support. */
|
/* Expression tracking support. */
|
|
|
/* Lookup REGNO in the set TABLE. The result is a pointer to the
|
/* Lookup REGNO in the set TABLE. The result is a pointer to the
|
table entry, or NULL if not found. */
|
table entry, or NULL if not found. */
|
|
|
static struct expr *
|
static struct expr *
|
lookup_set (unsigned int regno, struct hash_table *table)
|
lookup_set (unsigned int regno, struct hash_table *table)
|
{
|
{
|
unsigned int hash = hash_set (regno, table->size);
|
unsigned int hash = hash_set (regno, table->size);
|
struct expr *expr;
|
struct expr *expr;
|
|
|
expr = table->table[hash];
|
expr = table->table[hash];
|
|
|
while (expr && REGNO (SET_DEST (expr->expr)) != regno)
|
while (expr && REGNO (SET_DEST (expr->expr)) != regno)
|
expr = expr->next_same_hash;
|
expr = expr->next_same_hash;
|
|
|
return expr;
|
return expr;
|
}
|
}
|
|
|
/* Return the next entry for REGNO in list EXPR. */
|
/* Return the next entry for REGNO in list EXPR. */
|
|
|
static struct expr *
|
static struct expr *
|
next_set (unsigned int regno, struct expr *expr)
|
next_set (unsigned int regno, struct expr *expr)
|
{
|
{
|
do
|
do
|
expr = expr->next_same_hash;
|
expr = expr->next_same_hash;
|
while (expr && REGNO (SET_DEST (expr->expr)) != regno);
|
while (expr && REGNO (SET_DEST (expr->expr)) != regno);
|
|
|
return expr;
|
return expr;
|
}
|
}
|
|
|
/* Like free_INSN_LIST_list or free_EXPR_LIST_list, except that the node
|
/* Like free_INSN_LIST_list or free_EXPR_LIST_list, except that the node
|
types may be mixed. */
|
types may be mixed. */
|
|
|
static void
|
static void
|
free_insn_expr_list_list (rtx *listp)
|
free_insn_expr_list_list (rtx *listp)
|
{
|
{
|
rtx list, next;
|
rtx list, next;
|
|
|
for (list = *listp; list ; list = next)
|
for (list = *listp; list ; list = next)
|
{
|
{
|
next = XEXP (list, 1);
|
next = XEXP (list, 1);
|
if (GET_CODE (list) == EXPR_LIST)
|
if (GET_CODE (list) == EXPR_LIST)
|
free_EXPR_LIST_node (list);
|
free_EXPR_LIST_node (list);
|
else
|
else
|
free_INSN_LIST_node (list);
|
free_INSN_LIST_node (list);
|
}
|
}
|
|
|
*listp = NULL;
|
*listp = NULL;
|
}
|
}
|
|
|
/* Clear canon_modify_mem_list and modify_mem_list tables. */
|
/* Clear canon_modify_mem_list and modify_mem_list tables. */
|
static void
|
static void
|
clear_modify_mem_tables (void)
|
clear_modify_mem_tables (void)
|
{
|
{
|
unsigned i;
|
unsigned i;
|
bitmap_iterator bi;
|
bitmap_iterator bi;
|
|
|
EXECUTE_IF_SET_IN_BITMAP (modify_mem_list_set, 0, i, bi)
|
EXECUTE_IF_SET_IN_BITMAP (modify_mem_list_set, 0, i, bi)
|
{
|
{
|
free_INSN_LIST_list (modify_mem_list + i);
|
free_INSN_LIST_list (modify_mem_list + i);
|
free_insn_expr_list_list (canon_modify_mem_list + i);
|
free_insn_expr_list_list (canon_modify_mem_list + i);
|
}
|
}
|
bitmap_clear (modify_mem_list_set);
|
bitmap_clear (modify_mem_list_set);
|
bitmap_clear (blocks_with_calls);
|
bitmap_clear (blocks_with_calls);
|
}
|
}
|
|
|
/* Release memory used by modify_mem_list_set. */
|
/* Release memory used by modify_mem_list_set. */
|
|
|
static void
|
static void
|
free_modify_mem_tables (void)
|
free_modify_mem_tables (void)
|
{
|
{
|
clear_modify_mem_tables ();
|
clear_modify_mem_tables ();
|
free (modify_mem_list);
|
free (modify_mem_list);
|
free (canon_modify_mem_list);
|
free (canon_modify_mem_list);
|
modify_mem_list = 0;
|
modify_mem_list = 0;
|
canon_modify_mem_list = 0;
|
canon_modify_mem_list = 0;
|
}
|
}
|
|
|
/* Reset tables used to keep track of what's still available [since the
|
/* Reset tables used to keep track of what's still available [since the
|
start of the block]. */
|
start of the block]. */
|
|
|
static void
|
static void
|
reset_opr_set_tables (void)
|
reset_opr_set_tables (void)
|
{
|
{
|
/* Maintain a bitmap of which regs have been set since beginning of
|
/* Maintain a bitmap of which regs have been set since beginning of
|
the block. */
|
the block. */
|
CLEAR_REG_SET (reg_set_bitmap);
|
CLEAR_REG_SET (reg_set_bitmap);
|
|
|
/* Also keep a record of the last instruction to modify memory.
|
/* Also keep a record of the last instruction to modify memory.
|
For now this is very trivial, we only record whether any memory
|
For now this is very trivial, we only record whether any memory
|
location has been modified. */
|
location has been modified. */
|
clear_modify_mem_tables ();
|
clear_modify_mem_tables ();
|
}
|
}
|
|
|
/* Return nonzero if the operands of X are not set before INSN in
|
/* Return nonzero if the operands of X are not set before INSN in
|
INSN's basic block. */
|
INSN's basic block. */
|
|
|
static int
|
static int
|
oprs_not_set_p (rtx x, rtx insn)
|
oprs_not_set_p (rtx x, rtx insn)
|
{
|
{
|
int i, j;
|
int i, j;
|
enum rtx_code code;
|
enum rtx_code code;
|
const char *fmt;
|
const char *fmt;
|
|
|
if (x == 0)
|
if (x == 0)
|
return 1;
|
return 1;
|
|
|
code = GET_CODE (x);
|
code = GET_CODE (x);
|
switch (code)
|
switch (code)
|
{
|
{
|
case PC:
|
case PC:
|
case CC0:
|
case CC0:
|
case CONST:
|
case CONST:
|
case CONST_INT:
|
case CONST_INT:
|
case CONST_DOUBLE:
|
case CONST_DOUBLE:
|
case CONST_VECTOR:
|
case CONST_VECTOR:
|
case SYMBOL_REF:
|
case SYMBOL_REF:
|
case LABEL_REF:
|
case LABEL_REF:
|
case ADDR_VEC:
|
case ADDR_VEC:
|
case ADDR_DIFF_VEC:
|
case ADDR_DIFF_VEC:
|
return 1;
|
return 1;
|
|
|
case MEM:
|
case MEM:
|
if (load_killed_in_block_p (BLOCK_FOR_INSN (insn),
|
if (load_killed_in_block_p (BLOCK_FOR_INSN (insn),
|
INSN_CUID (insn), x, 0))
|
INSN_CUID (insn), x, 0))
|
return 0;
|
return 0;
|
else
|
else
|
return oprs_not_set_p (XEXP (x, 0), insn);
|
return oprs_not_set_p (XEXP (x, 0), insn);
|
|
|
case REG:
|
case REG:
|
return ! REGNO_REG_SET_P (reg_set_bitmap, REGNO (x));
|
return ! REGNO_REG_SET_P (reg_set_bitmap, REGNO (x));
|
|
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
|
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
{
|
{
|
if (fmt[i] == 'e')
|
if (fmt[i] == 'e')
|
{
|
{
|
/* If we are about to do the last recursive call
|
/* If we are about to do the last recursive call
|
needed at this level, change it into iteration.
|
needed at this level, change it into iteration.
|
This function is called enough to be worth it. */
|
This function is called enough to be worth it. */
|
if (i == 0)
|
if (i == 0)
|
return oprs_not_set_p (XEXP (x, i), insn);
|
return oprs_not_set_p (XEXP (x, i), insn);
|
|
|
if (! oprs_not_set_p (XEXP (x, i), insn))
|
if (! oprs_not_set_p (XEXP (x, i), insn))
|
return 0;
|
return 0;
|
}
|
}
|
else if (fmt[i] == 'E')
|
else if (fmt[i] == 'E')
|
for (j = 0; j < XVECLEN (x, i); j++)
|
for (j = 0; j < XVECLEN (x, i); j++)
|
if (! oprs_not_set_p (XVECEXP (x, i, j), insn))
|
if (! oprs_not_set_p (XVECEXP (x, i, j), insn))
|
return 0;
|
return 0;
|
}
|
}
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Mark things set by a CALL. */
|
/* Mark things set by a CALL. */
|
|
|
static void
|
static void
|
mark_call (rtx insn)
|
mark_call (rtx insn)
|
{
|
{
|
if (! CONST_OR_PURE_CALL_P (insn))
|
if (! CONST_OR_PURE_CALL_P (insn))
|
record_last_mem_set_info (insn);
|
record_last_mem_set_info (insn);
|
}
|
}
|
|
|
/* Mark things set by a SET. */
|
/* Mark things set by a SET. */
|
|
|
static void
|
static void
|
mark_set (rtx pat, rtx insn)
|
mark_set (rtx pat, rtx insn)
|
{
|
{
|
rtx dest = SET_DEST (pat);
|
rtx dest = SET_DEST (pat);
|
|
|
while (GET_CODE (dest) == SUBREG
|
while (GET_CODE (dest) == SUBREG
|
|| GET_CODE (dest) == ZERO_EXTRACT
|
|| GET_CODE (dest) == ZERO_EXTRACT
|
|| GET_CODE (dest) == STRICT_LOW_PART)
|
|| GET_CODE (dest) == STRICT_LOW_PART)
|
dest = XEXP (dest, 0);
|
dest = XEXP (dest, 0);
|
|
|
if (REG_P (dest))
|
if (REG_P (dest))
|
SET_REGNO_REG_SET (reg_set_bitmap, REGNO (dest));
|
SET_REGNO_REG_SET (reg_set_bitmap, REGNO (dest));
|
else if (MEM_P (dest))
|
else if (MEM_P (dest))
|
record_last_mem_set_info (insn);
|
record_last_mem_set_info (insn);
|
|
|
if (GET_CODE (SET_SRC (pat)) == CALL)
|
if (GET_CODE (SET_SRC (pat)) == CALL)
|
mark_call (insn);
|
mark_call (insn);
|
}
|
}
|
|
|
/* Record things set by a CLOBBER. */
|
/* Record things set by a CLOBBER. */
|
|
|
static void
|
static void
|
mark_clobber (rtx pat, rtx insn)
|
mark_clobber (rtx pat, rtx insn)
|
{
|
{
|
rtx clob = XEXP (pat, 0);
|
rtx clob = XEXP (pat, 0);
|
|
|
while (GET_CODE (clob) == SUBREG || GET_CODE (clob) == STRICT_LOW_PART)
|
while (GET_CODE (clob) == SUBREG || GET_CODE (clob) == STRICT_LOW_PART)
|
clob = XEXP (clob, 0);
|
clob = XEXP (clob, 0);
|
|
|
if (REG_P (clob))
|
if (REG_P (clob))
|
SET_REGNO_REG_SET (reg_set_bitmap, REGNO (clob));
|
SET_REGNO_REG_SET (reg_set_bitmap, REGNO (clob));
|
else
|
else
|
record_last_mem_set_info (insn);
|
record_last_mem_set_info (insn);
|
}
|
}
|
|
|
/* Record things set by INSN.
|
/* Record things set by INSN.
|
This data is used by oprs_not_set_p. */
|
This data is used by oprs_not_set_p. */
|
|
|
static void
|
static void
|
mark_oprs_set (rtx insn)
|
mark_oprs_set (rtx insn)
|
{
|
{
|
rtx pat = PATTERN (insn);
|
rtx pat = PATTERN (insn);
|
int i;
|
int i;
|
|
|
if (GET_CODE (pat) == SET)
|
if (GET_CODE (pat) == SET)
|
mark_set (pat, insn);
|
mark_set (pat, insn);
|
else if (GET_CODE (pat) == PARALLEL)
|
else if (GET_CODE (pat) == PARALLEL)
|
for (i = 0; i < XVECLEN (pat, 0); i++)
|
for (i = 0; i < XVECLEN (pat, 0); i++)
|
{
|
{
|
rtx x = XVECEXP (pat, 0, i);
|
rtx x = XVECEXP (pat, 0, i);
|
|
|
if (GET_CODE (x) == SET)
|
if (GET_CODE (x) == SET)
|
mark_set (x, insn);
|
mark_set (x, insn);
|
else if (GET_CODE (x) == CLOBBER)
|
else if (GET_CODE (x) == CLOBBER)
|
mark_clobber (x, insn);
|
mark_clobber (x, insn);
|
else if (GET_CODE (x) == CALL)
|
else if (GET_CODE (x) == CALL)
|
mark_call (insn);
|
mark_call (insn);
|
}
|
}
|
|
|
else if (GET_CODE (pat) == CLOBBER)
|
else if (GET_CODE (pat) == CLOBBER)
|
mark_clobber (pat, insn);
|
mark_clobber (pat, insn);
|
else if (GET_CODE (pat) == CALL)
|
else if (GET_CODE (pat) == CALL)
|
mark_call (insn);
|
mark_call (insn);
|
}
|
}
|
|
|
|
|
/* Compute copy/constant propagation working variables. */
|
/* Compute copy/constant propagation working variables. */
|
|
|
/* Local properties of assignments. */
|
/* Local properties of assignments. */
|
static sbitmap *cprop_pavloc;
|
static sbitmap *cprop_pavloc;
|
static sbitmap *cprop_absaltered;
|
static sbitmap *cprop_absaltered;
|
|
|
/* Global properties of assignments (computed from the local properties). */
|
/* Global properties of assignments (computed from the local properties). */
|
static sbitmap *cprop_avin;
|
static sbitmap *cprop_avin;
|
static sbitmap *cprop_avout;
|
static sbitmap *cprop_avout;
|
|
|
/* Allocate vars used for copy/const propagation. N_BLOCKS is the number of
|
/* Allocate vars used for copy/const propagation. N_BLOCKS is the number of
|
basic blocks. N_SETS is the number of sets. */
|
basic blocks. N_SETS is the number of sets. */
|
|
|
static void
|
static void
|
alloc_cprop_mem (int n_blocks, int n_sets)
|
alloc_cprop_mem (int n_blocks, int n_sets)
|
{
|
{
|
cprop_pavloc = sbitmap_vector_alloc (n_blocks, n_sets);
|
cprop_pavloc = sbitmap_vector_alloc (n_blocks, n_sets);
|
cprop_absaltered = sbitmap_vector_alloc (n_blocks, n_sets);
|
cprop_absaltered = sbitmap_vector_alloc (n_blocks, n_sets);
|
|
|
cprop_avin = sbitmap_vector_alloc (n_blocks, n_sets);
|
cprop_avin = sbitmap_vector_alloc (n_blocks, n_sets);
|
cprop_avout = sbitmap_vector_alloc (n_blocks, n_sets);
|
cprop_avout = sbitmap_vector_alloc (n_blocks, n_sets);
|
}
|
}
|
|
|
/* Free vars used by copy/const propagation. */
|
/* Free vars used by copy/const propagation. */
|
|
|
static void
|
static void
|
free_cprop_mem (void)
|
free_cprop_mem (void)
|
{
|
{
|
sbitmap_vector_free (cprop_pavloc);
|
sbitmap_vector_free (cprop_pavloc);
|
sbitmap_vector_free (cprop_absaltered);
|
sbitmap_vector_free (cprop_absaltered);
|
sbitmap_vector_free (cprop_avin);
|
sbitmap_vector_free (cprop_avin);
|
sbitmap_vector_free (cprop_avout);
|
sbitmap_vector_free (cprop_avout);
|
}
|
}
|
|
|
/* For each block, compute whether X is transparent. X is either an
|
/* For each block, compute whether X is transparent. X is either an
|
expression or an assignment [though we don't care which, for this context
|
expression or an assignment [though we don't care which, for this context
|
an assignment is treated as an expression]. For each block where an
|
an assignment is treated as an expression]. For each block where an
|
element of X is modified, set (SET_P == 1) or reset (SET_P == 0) the INDX
|
element of X is modified, set (SET_P == 1) or reset (SET_P == 0) the INDX
|
bit in BMAP. */
|
bit in BMAP. */
|
|
|
static void
|
static void
|
compute_transp (rtx x, int indx, sbitmap *bmap, int set_p)
|
compute_transp (rtx x, int indx, sbitmap *bmap, int set_p)
|
{
|
{
|
int i, j;
|
int i, j;
|
basic_block bb;
|
basic_block bb;
|
enum rtx_code code;
|
enum rtx_code code;
|
reg_set *r;
|
reg_set *r;
|
const char *fmt;
|
const char *fmt;
|
|
|
/* repeat is used to turn tail-recursion into iteration since GCC
|
/* repeat is used to turn tail-recursion into iteration since GCC
|
can't do it when there's no return value. */
|
can't do it when there's no return value. */
|
repeat:
|
repeat:
|
|
|
if (x == 0)
|
if (x == 0)
|
return;
|
return;
|
|
|
code = GET_CODE (x);
|
code = GET_CODE (x);
|
switch (code)
|
switch (code)
|
{
|
{
|
case REG:
|
case REG:
|
if (set_p)
|
if (set_p)
|
{
|
{
|
if (REGNO (x) < FIRST_PSEUDO_REGISTER)
|
if (REGNO (x) < FIRST_PSEUDO_REGISTER)
|
{
|
{
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
if (TEST_BIT (reg_set_in_block[bb->index], REGNO (x)))
|
if (TEST_BIT (reg_set_in_block[bb->index], REGNO (x)))
|
SET_BIT (bmap[bb->index], indx);
|
SET_BIT (bmap[bb->index], indx);
|
}
|
}
|
else
|
else
|
{
|
{
|
for (r = reg_set_table[REGNO (x)]; r != NULL; r = r->next)
|
for (r = reg_set_table[REGNO (x)]; r != NULL; r = r->next)
|
SET_BIT (bmap[r->bb_index], indx);
|
SET_BIT (bmap[r->bb_index], indx);
|
}
|
}
|
}
|
}
|
else
|
else
|
{
|
{
|
if (REGNO (x) < FIRST_PSEUDO_REGISTER)
|
if (REGNO (x) < FIRST_PSEUDO_REGISTER)
|
{
|
{
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
if (TEST_BIT (reg_set_in_block[bb->index], REGNO (x)))
|
if (TEST_BIT (reg_set_in_block[bb->index], REGNO (x)))
|
RESET_BIT (bmap[bb->index], indx);
|
RESET_BIT (bmap[bb->index], indx);
|
}
|
}
|
else
|
else
|
{
|
{
|
for (r = reg_set_table[REGNO (x)]; r != NULL; r = r->next)
|
for (r = reg_set_table[REGNO (x)]; r != NULL; r = r->next)
|
RESET_BIT (bmap[r->bb_index], indx);
|
RESET_BIT (bmap[r->bb_index], indx);
|
}
|
}
|
}
|
}
|
|
|
return;
|
return;
|
|
|
case MEM:
|
case MEM:
|
if (! MEM_READONLY_P (x))
|
if (! MEM_READONLY_P (x))
|
{
|
{
|
bitmap_iterator bi;
|
bitmap_iterator bi;
|
unsigned bb_index;
|
unsigned bb_index;
|
|
|
/* First handle all the blocks with calls. We don't need to
|
/* First handle all the blocks with calls. We don't need to
|
do any list walking for them. */
|
do any list walking for them. */
|
EXECUTE_IF_SET_IN_BITMAP (blocks_with_calls, 0, bb_index, bi)
|
EXECUTE_IF_SET_IN_BITMAP (blocks_with_calls, 0, bb_index, bi)
|
{
|
{
|
if (set_p)
|
if (set_p)
|
SET_BIT (bmap[bb_index], indx);
|
SET_BIT (bmap[bb_index], indx);
|
else
|
else
|
RESET_BIT (bmap[bb_index], indx);
|
RESET_BIT (bmap[bb_index], indx);
|
}
|
}
|
|
|
/* Now iterate over the blocks which have memory modifications
|
/* Now iterate over the blocks which have memory modifications
|
but which do not have any calls. */
|
but which do not have any calls. */
|
EXECUTE_IF_AND_COMPL_IN_BITMAP (modify_mem_list_set,
|
EXECUTE_IF_AND_COMPL_IN_BITMAP (modify_mem_list_set,
|
blocks_with_calls,
|
blocks_with_calls,
|
0, bb_index, bi)
|
0, bb_index, bi)
|
{
|
{
|
rtx list_entry = canon_modify_mem_list[bb_index];
|
rtx list_entry = canon_modify_mem_list[bb_index];
|
|
|
while (list_entry)
|
while (list_entry)
|
{
|
{
|
rtx dest, dest_addr;
|
rtx dest, dest_addr;
|
|
|
/* LIST_ENTRY must be an INSN of some kind that sets memory.
|
/* LIST_ENTRY must be an INSN of some kind that sets memory.
|
Examine each hunk of memory that is modified. */
|
Examine each hunk of memory that is modified. */
|
|
|
dest = XEXP (list_entry, 0);
|
dest = XEXP (list_entry, 0);
|
list_entry = XEXP (list_entry, 1);
|
list_entry = XEXP (list_entry, 1);
|
dest_addr = XEXP (list_entry, 0);
|
dest_addr = XEXP (list_entry, 0);
|
|
|
if (canon_true_dependence (dest, GET_MODE (dest), dest_addr,
|
if (canon_true_dependence (dest, GET_MODE (dest), dest_addr,
|
x, rtx_addr_varies_p))
|
x, rtx_addr_varies_p))
|
{
|
{
|
if (set_p)
|
if (set_p)
|
SET_BIT (bmap[bb_index], indx);
|
SET_BIT (bmap[bb_index], indx);
|
else
|
else
|
RESET_BIT (bmap[bb_index], indx);
|
RESET_BIT (bmap[bb_index], indx);
|
break;
|
break;
|
}
|
}
|
list_entry = XEXP (list_entry, 1);
|
list_entry = XEXP (list_entry, 1);
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
x = XEXP (x, 0);
|
x = XEXP (x, 0);
|
goto repeat;
|
goto repeat;
|
|
|
case PC:
|
case PC:
|
case CC0: /*FIXME*/
|
case CC0: /*FIXME*/
|
case CONST:
|
case CONST:
|
case CONST_INT:
|
case CONST_INT:
|
case CONST_DOUBLE:
|
case CONST_DOUBLE:
|
case CONST_VECTOR:
|
case CONST_VECTOR:
|
case SYMBOL_REF:
|
case SYMBOL_REF:
|
case LABEL_REF:
|
case LABEL_REF:
|
case ADDR_VEC:
|
case ADDR_VEC:
|
case ADDR_DIFF_VEC:
|
case ADDR_DIFF_VEC:
|
return;
|
return;
|
|
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
|
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
{
|
{
|
if (fmt[i] == 'e')
|
if (fmt[i] == 'e')
|
{
|
{
|
/* If we are about to do the last recursive call
|
/* If we are about to do the last recursive call
|
needed at this level, change it into iteration.
|
needed at this level, change it into iteration.
|
This function is called enough to be worth it. */
|
This function is called enough to be worth it. */
|
if (i == 0)
|
if (i == 0)
|
{
|
{
|
x = XEXP (x, i);
|
x = XEXP (x, i);
|
goto repeat;
|
goto repeat;
|
}
|
}
|
|
|
compute_transp (XEXP (x, i), indx, bmap, set_p);
|
compute_transp (XEXP (x, i), indx, bmap, set_p);
|
}
|
}
|
else if (fmt[i] == 'E')
|
else if (fmt[i] == 'E')
|
for (j = 0; j < XVECLEN (x, i); j++)
|
for (j = 0; j < XVECLEN (x, i); j++)
|
compute_transp (XVECEXP (x, i, j), indx, bmap, set_p);
|
compute_transp (XVECEXP (x, i, j), indx, bmap, set_p);
|
}
|
}
|
}
|
}
|
|
|
/* Top level routine to do the dataflow analysis needed by copy/const
|
/* Top level routine to do the dataflow analysis needed by copy/const
|
propagation. */
|
propagation. */
|
|
|
static void
|
static void
|
compute_cprop_data (void)
|
compute_cprop_data (void)
|
{
|
{
|
compute_local_properties (cprop_absaltered, cprop_pavloc, NULL, &set_hash_table);
|
compute_local_properties (cprop_absaltered, cprop_pavloc, NULL, &set_hash_table);
|
compute_available (cprop_pavloc, cprop_absaltered,
|
compute_available (cprop_pavloc, cprop_absaltered,
|
cprop_avout, cprop_avin);
|
cprop_avout, cprop_avin);
|
}
|
}
|
|
|
/* Copy/constant propagation. */
|
/* Copy/constant propagation. */
|
|
|
/* Maximum number of register uses in an insn that we handle. */
|
/* Maximum number of register uses in an insn that we handle. */
|
#define MAX_USES 8
|
#define MAX_USES 8
|
|
|
/* Table of uses found in an insn.
|
/* Table of uses found in an insn.
|
Allocated statically to avoid alloc/free complexity and overhead. */
|
Allocated statically to avoid alloc/free complexity and overhead. */
|
static struct reg_use reg_use_table[MAX_USES];
|
static struct reg_use reg_use_table[MAX_USES];
|
|
|
/* Index into `reg_use_table' while building it. */
|
/* Index into `reg_use_table' while building it. */
|
static int reg_use_count;
|
static int reg_use_count;
|
|
|
/* Set up a list of register numbers used in INSN. The found uses are stored
|
/* Set up a list of register numbers used in INSN. The found uses are stored
|
in `reg_use_table'. `reg_use_count' is initialized to zero before entry,
|
in `reg_use_table'. `reg_use_count' is initialized to zero before entry,
|
and contains the number of uses in the table upon exit.
|
and contains the number of uses in the table upon exit.
|
|
|
??? If a register appears multiple times we will record it multiple times.
|
??? If a register appears multiple times we will record it multiple times.
|
This doesn't hurt anything but it will slow things down. */
|
This doesn't hurt anything but it will slow things down. */
|
|
|
static void
|
static void
|
find_used_regs (rtx *xptr, void *data ATTRIBUTE_UNUSED)
|
find_used_regs (rtx *xptr, void *data ATTRIBUTE_UNUSED)
|
{
|
{
|
int i, j;
|
int i, j;
|
enum rtx_code code;
|
enum rtx_code code;
|
const char *fmt;
|
const char *fmt;
|
rtx x = *xptr;
|
rtx x = *xptr;
|
|
|
/* repeat is used to turn tail-recursion into iteration since GCC
|
/* repeat is used to turn tail-recursion into iteration since GCC
|
can't do it when there's no return value. */
|
can't do it when there's no return value. */
|
repeat:
|
repeat:
|
if (x == 0)
|
if (x == 0)
|
return;
|
return;
|
|
|
code = GET_CODE (x);
|
code = GET_CODE (x);
|
if (REG_P (x))
|
if (REG_P (x))
|
{
|
{
|
if (reg_use_count == MAX_USES)
|
if (reg_use_count == MAX_USES)
|
return;
|
return;
|
|
|
reg_use_table[reg_use_count].reg_rtx = x;
|
reg_use_table[reg_use_count].reg_rtx = x;
|
reg_use_count++;
|
reg_use_count++;
|
}
|
}
|
|
|
/* Recursively scan the operands of this expression. */
|
/* Recursively scan the operands of this expression. */
|
|
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
{
|
{
|
if (fmt[i] == 'e')
|
if (fmt[i] == 'e')
|
{
|
{
|
/* If we are about to do the last recursive call
|
/* If we are about to do the last recursive call
|
needed at this level, change it into iteration.
|
needed at this level, change it into iteration.
|
This function is called enough to be worth it. */
|
This function is called enough to be worth it. */
|
if (i == 0)
|
if (i == 0)
|
{
|
{
|
x = XEXP (x, 0);
|
x = XEXP (x, 0);
|
goto repeat;
|
goto repeat;
|
}
|
}
|
|
|
find_used_regs (&XEXP (x, i), data);
|
find_used_regs (&XEXP (x, i), data);
|
}
|
}
|
else if (fmt[i] == 'E')
|
else if (fmt[i] == 'E')
|
for (j = 0; j < XVECLEN (x, i); j++)
|
for (j = 0; j < XVECLEN (x, i); j++)
|
find_used_regs (&XVECEXP (x, i, j), data);
|
find_used_regs (&XVECEXP (x, i, j), data);
|
}
|
}
|
}
|
}
|
|
|
/* Try to replace all non-SET_DEST occurrences of FROM in INSN with TO.
|
/* Try to replace all non-SET_DEST occurrences of FROM in INSN with TO.
|
Returns nonzero is successful. */
|
Returns nonzero is successful. */
|
|
|
static int
|
static int
|
try_replace_reg (rtx from, rtx to, rtx insn)
|
try_replace_reg (rtx from, rtx to, rtx insn)
|
{
|
{
|
rtx note = find_reg_equal_equiv_note (insn);
|
rtx note = find_reg_equal_equiv_note (insn);
|
rtx src = 0;
|
rtx src = 0;
|
int success = 0;
|
int success = 0;
|
rtx set = single_set (insn);
|
rtx set = single_set (insn);
|
|
|
validate_replace_src_group (from, to, insn);
|
validate_replace_src_group (from, to, insn);
|
if (num_changes_pending () && apply_change_group ())
|
if (num_changes_pending () && apply_change_group ())
|
success = 1;
|
success = 1;
|
|
|
/* Try to simplify SET_SRC if we have substituted a constant. */
|
/* Try to simplify SET_SRC if we have substituted a constant. */
|
if (success && set && CONSTANT_P (to))
|
if (success && set && CONSTANT_P (to))
|
{
|
{
|
src = simplify_rtx (SET_SRC (set));
|
src = simplify_rtx (SET_SRC (set));
|
|
|
if (src)
|
if (src)
|
validate_change (insn, &SET_SRC (set), src, 0);
|
validate_change (insn, &SET_SRC (set), src, 0);
|
}
|
}
|
|
|
/* If there is already a REG_EQUAL note, update the expression in it
|
/* If there is already a REG_EQUAL note, update the expression in it
|
with our replacement. */
|
with our replacement. */
|
if (note != 0 && REG_NOTE_KIND (note) == REG_EQUAL)
|
if (note != 0 && REG_NOTE_KIND (note) == REG_EQUAL)
|
XEXP (note, 0) = simplify_replace_rtx (XEXP (note, 0), from, to);
|
XEXP (note, 0) = simplify_replace_rtx (XEXP (note, 0), from, to);
|
|
|
if (!success && set && reg_mentioned_p (from, SET_SRC (set)))
|
if (!success && set && reg_mentioned_p (from, SET_SRC (set)))
|
{
|
{
|
/* If above failed and this is a single set, try to simplify the source of
|
/* If above failed and this is a single set, try to simplify the source of
|
the set given our substitution. We could perhaps try this for multiple
|
the set given our substitution. We could perhaps try this for multiple
|
SETs, but it probably won't buy us anything. */
|
SETs, but it probably won't buy us anything. */
|
src = simplify_replace_rtx (SET_SRC (set), from, to);
|
src = simplify_replace_rtx (SET_SRC (set), from, to);
|
|
|
if (!rtx_equal_p (src, SET_SRC (set))
|
if (!rtx_equal_p (src, SET_SRC (set))
|
&& validate_change (insn, &SET_SRC (set), src, 0))
|
&& validate_change (insn, &SET_SRC (set), src, 0))
|
success = 1;
|
success = 1;
|
|
|
/* If we've failed to do replacement, have a single SET, don't already
|
/* If we've failed to do replacement, have a single SET, don't already
|
have a note, and have no special SET, add a REG_EQUAL note to not
|
have a note, and have no special SET, add a REG_EQUAL note to not
|
lose information. */
|
lose information. */
|
if (!success && note == 0 && set != 0
|
if (!success && note == 0 && set != 0
|
&& GET_CODE (SET_DEST (set)) != ZERO_EXTRACT
|
&& GET_CODE (SET_DEST (set)) != ZERO_EXTRACT
|
&& GET_CODE (SET_DEST (set)) != STRICT_LOW_PART)
|
&& GET_CODE (SET_DEST (set)) != STRICT_LOW_PART)
|
note = set_unique_reg_note (insn, REG_EQUAL, copy_rtx (src));
|
note = set_unique_reg_note (insn, REG_EQUAL, copy_rtx (src));
|
}
|
}
|
|
|
/* REG_EQUAL may get simplified into register.
|
/* REG_EQUAL may get simplified into register.
|
We don't allow that. Remove that note. This code ought
|
We don't allow that. Remove that note. This code ought
|
not to happen, because previous code ought to synthesize
|
not to happen, because previous code ought to synthesize
|
reg-reg move, but be on the safe side. */
|
reg-reg move, but be on the safe side. */
|
if (note && REG_NOTE_KIND (note) == REG_EQUAL && REG_P (XEXP (note, 0)))
|
if (note && REG_NOTE_KIND (note) == REG_EQUAL && REG_P (XEXP (note, 0)))
|
remove_note (insn, note);
|
remove_note (insn, note);
|
|
|
return success;
|
return success;
|
}
|
}
|
|
|
/* Find a set of REGNOs that are available on entry to INSN's block. Returns
|
/* Find a set of REGNOs that are available on entry to INSN's block. Returns
|
NULL no such set is found. */
|
NULL no such set is found. */
|
|
|
static struct expr *
|
static struct expr *
|
find_avail_set (int regno, rtx insn)
|
find_avail_set (int regno, rtx insn)
|
{
|
{
|
/* SET1 contains the last set found that can be returned to the caller for
|
/* SET1 contains the last set found that can be returned to the caller for
|
use in a substitution. */
|
use in a substitution. */
|
struct expr *set1 = 0;
|
struct expr *set1 = 0;
|
|
|
/* Loops are not possible here. To get a loop we would need two sets
|
/* Loops are not possible here. To get a loop we would need two sets
|
available at the start of the block containing INSN. i.e. we would
|
available at the start of the block containing INSN. i.e. we would
|
need two sets like this available at the start of the block:
|
need two sets like this available at the start of the block:
|
|
|
(set (reg X) (reg Y))
|
(set (reg X) (reg Y))
|
(set (reg Y) (reg X))
|
(set (reg Y) (reg X))
|
|
|
This can not happen since the set of (reg Y) would have killed the
|
This can not happen since the set of (reg Y) would have killed the
|
set of (reg X) making it unavailable at the start of this block. */
|
set of (reg X) making it unavailable at the start of this block. */
|
while (1)
|
while (1)
|
{
|
{
|
rtx src;
|
rtx src;
|
struct expr *set = lookup_set (regno, &set_hash_table);
|
struct expr *set = lookup_set (regno, &set_hash_table);
|
|
|
/* Find a set that is available at the start of the block
|
/* Find a set that is available at the start of the block
|
which contains INSN. */
|
which contains INSN. */
|
while (set)
|
while (set)
|
{
|
{
|
if (TEST_BIT (cprop_avin[BLOCK_NUM (insn)], set->bitmap_index))
|
if (TEST_BIT (cprop_avin[BLOCK_NUM (insn)], set->bitmap_index))
|
break;
|
break;
|
set = next_set (regno, set);
|
set = next_set (regno, set);
|
}
|
}
|
|
|
/* If no available set was found we've reached the end of the
|
/* If no available set was found we've reached the end of the
|
(possibly empty) copy chain. */
|
(possibly empty) copy chain. */
|
if (set == 0)
|
if (set == 0)
|
break;
|
break;
|
|
|
gcc_assert (GET_CODE (set->expr) == SET);
|
gcc_assert (GET_CODE (set->expr) == SET);
|
|
|
src = SET_SRC (set->expr);
|
src = SET_SRC (set->expr);
|
|
|
/* We know the set is available.
|
/* We know the set is available.
|
Now check that SRC is ANTLOC (i.e. none of the source operands
|
Now check that SRC is ANTLOC (i.e. none of the source operands
|
have changed since the start of the block).
|
have changed since the start of the block).
|
|
|
If the source operand changed, we may still use it for the next
|
If the source operand changed, we may still use it for the next
|
iteration of this loop, but we may not use it for substitutions. */
|
iteration of this loop, but we may not use it for substitutions. */
|
|
|
if (gcse_constant_p (src) || oprs_not_set_p (src, insn))
|
if (gcse_constant_p (src) || oprs_not_set_p (src, insn))
|
set1 = set;
|
set1 = set;
|
|
|
/* If the source of the set is anything except a register, then
|
/* If the source of the set is anything except a register, then
|
we have reached the end of the copy chain. */
|
we have reached the end of the copy chain. */
|
if (! REG_P (src))
|
if (! REG_P (src))
|
break;
|
break;
|
|
|
/* Follow the copy chain, i.e. start another iteration of the loop
|
/* Follow the copy chain, i.e. start another iteration of the loop
|
and see if we have an available copy into SRC. */
|
and see if we have an available copy into SRC. */
|
regno = REGNO (src);
|
regno = REGNO (src);
|
}
|
}
|
|
|
/* SET1 holds the last set that was available and anticipatable at
|
/* SET1 holds the last set that was available and anticipatable at
|
INSN. */
|
INSN. */
|
return set1;
|
return set1;
|
}
|
}
|
|
|
/* Subroutine of cprop_insn that tries to propagate constants into
|
/* Subroutine of cprop_insn that tries to propagate constants into
|
JUMP_INSNS. JUMP must be a conditional jump. If SETCC is non-NULL
|
JUMP_INSNS. JUMP must be a conditional jump. If SETCC is non-NULL
|
it is the instruction that immediately precedes JUMP, and must be a
|
it is the instruction that immediately precedes JUMP, and must be a
|
single SET of a register. FROM is what we will try to replace,
|
single SET of a register. FROM is what we will try to replace,
|
SRC is the constant we will try to substitute for it. Returns nonzero
|
SRC is the constant we will try to substitute for it. Returns nonzero
|
if a change was made. */
|
if a change was made. */
|
|
|
static int
|
static int
|
cprop_jump (basic_block bb, rtx setcc, rtx jump, rtx from, rtx src)
|
cprop_jump (basic_block bb, rtx setcc, rtx jump, rtx from, rtx src)
|
{
|
{
|
rtx new, set_src, note_src;
|
rtx new, set_src, note_src;
|
rtx set = pc_set (jump);
|
rtx set = pc_set (jump);
|
rtx note = find_reg_equal_equiv_note (jump);
|
rtx note = find_reg_equal_equiv_note (jump);
|
|
|
if (note)
|
if (note)
|
{
|
{
|
note_src = XEXP (note, 0);
|
note_src = XEXP (note, 0);
|
if (GET_CODE (note_src) == EXPR_LIST)
|
if (GET_CODE (note_src) == EXPR_LIST)
|
note_src = NULL_RTX;
|
note_src = NULL_RTX;
|
}
|
}
|
else note_src = NULL_RTX;
|
else note_src = NULL_RTX;
|
|
|
/* Prefer REG_EQUAL notes except those containing EXPR_LISTs. */
|
/* Prefer REG_EQUAL notes except those containing EXPR_LISTs. */
|
set_src = note_src ? note_src : SET_SRC (set);
|
set_src = note_src ? note_src : SET_SRC (set);
|
|
|
/* First substitute the SETCC condition into the JUMP instruction,
|
/* First substitute the SETCC condition into the JUMP instruction,
|
then substitute that given values into this expanded JUMP. */
|
then substitute that given values into this expanded JUMP. */
|
if (setcc != NULL_RTX
|
if (setcc != NULL_RTX
|
&& !modified_between_p (from, setcc, jump)
|
&& !modified_between_p (from, setcc, jump)
|
&& !modified_between_p (src, setcc, jump))
|
&& !modified_between_p (src, setcc, jump))
|
{
|
{
|
rtx setcc_src;
|
rtx setcc_src;
|
rtx setcc_set = single_set (setcc);
|
rtx setcc_set = single_set (setcc);
|
rtx setcc_note = find_reg_equal_equiv_note (setcc);
|
rtx setcc_note = find_reg_equal_equiv_note (setcc);
|
setcc_src = (setcc_note && GET_CODE (XEXP (setcc_note, 0)) != EXPR_LIST)
|
setcc_src = (setcc_note && GET_CODE (XEXP (setcc_note, 0)) != EXPR_LIST)
|
? XEXP (setcc_note, 0) : SET_SRC (setcc_set);
|
? XEXP (setcc_note, 0) : SET_SRC (setcc_set);
|
set_src = simplify_replace_rtx (set_src, SET_DEST (setcc_set),
|
set_src = simplify_replace_rtx (set_src, SET_DEST (setcc_set),
|
setcc_src);
|
setcc_src);
|
}
|
}
|
else
|
else
|
setcc = NULL_RTX;
|
setcc = NULL_RTX;
|
|
|
new = simplify_replace_rtx (set_src, from, src);
|
new = simplify_replace_rtx (set_src, from, src);
|
|
|
/* If no simplification can be made, then try the next register. */
|
/* If no simplification can be made, then try the next register. */
|
if (rtx_equal_p (new, SET_SRC (set)))
|
if (rtx_equal_p (new, SET_SRC (set)))
|
return 0;
|
return 0;
|
|
|
/* If this is now a no-op delete it, otherwise this must be a valid insn. */
|
/* If this is now a no-op delete it, otherwise this must be a valid insn. */
|
if (new == pc_rtx)
|
if (new == pc_rtx)
|
delete_insn (jump);
|
delete_insn (jump);
|
else
|
else
|
{
|
{
|
/* Ensure the value computed inside the jump insn to be equivalent
|
/* Ensure the value computed inside the jump insn to be equivalent
|
to one computed by setcc. */
|
to one computed by setcc. */
|
if (setcc && modified_in_p (new, setcc))
|
if (setcc && modified_in_p (new, setcc))
|
return 0;
|
return 0;
|
if (! validate_change (jump, &SET_SRC (set), new, 0))
|
if (! validate_change (jump, &SET_SRC (set), new, 0))
|
{
|
{
|
/* When (some) constants are not valid in a comparison, and there
|
/* When (some) constants are not valid in a comparison, and there
|
are two registers to be replaced by constants before the entire
|
are two registers to be replaced by constants before the entire
|
comparison can be folded into a constant, we need to keep
|
comparison can be folded into a constant, we need to keep
|
intermediate information in REG_EQUAL notes. For targets with
|
intermediate information in REG_EQUAL notes. For targets with
|
separate compare insns, such notes are added by try_replace_reg.
|
separate compare insns, such notes are added by try_replace_reg.
|
When we have a combined compare-and-branch instruction, however,
|
When we have a combined compare-and-branch instruction, however,
|
we need to attach a note to the branch itself to make this
|
we need to attach a note to the branch itself to make this
|
optimization work. */
|
optimization work. */
|
|
|
if (!rtx_equal_p (new, note_src))
|
if (!rtx_equal_p (new, note_src))
|
set_unique_reg_note (jump, REG_EQUAL, copy_rtx (new));
|
set_unique_reg_note (jump, REG_EQUAL, copy_rtx (new));
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* Remove REG_EQUAL note after simplification. */
|
/* Remove REG_EQUAL note after simplification. */
|
if (note_src)
|
if (note_src)
|
remove_note (jump, note);
|
remove_note (jump, note);
|
|
|
/* If this has turned into an unconditional jump,
|
/* If this has turned into an unconditional jump,
|
then put a barrier after it so that the unreachable
|
then put a barrier after it so that the unreachable
|
code will be deleted. */
|
code will be deleted. */
|
if (GET_CODE (SET_SRC (set)) == LABEL_REF)
|
if (GET_CODE (SET_SRC (set)) == LABEL_REF)
|
emit_barrier_after (jump);
|
emit_barrier_after (jump);
|
}
|
}
|
|
|
#ifdef HAVE_cc0
|
#ifdef HAVE_cc0
|
/* Delete the cc0 setter. */
|
/* Delete the cc0 setter. */
|
if (setcc != NULL && CC0_P (SET_DEST (single_set (setcc))))
|
if (setcc != NULL && CC0_P (SET_DEST (single_set (setcc))))
|
delete_insn (setcc);
|
delete_insn (setcc);
|
#endif
|
#endif
|
|
|
run_jump_opt_after_gcse = 1;
|
run_jump_opt_after_gcse = 1;
|
|
|
global_const_prop_count++;
|
global_const_prop_count++;
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
{
|
{
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"GLOBAL CONST-PROP: Replacing reg %d in jump_insn %d with constant ",
|
"GLOBAL CONST-PROP: Replacing reg %d in jump_insn %d with constant ",
|
REGNO (from), INSN_UID (jump));
|
REGNO (from), INSN_UID (jump));
|
print_rtl (dump_file, src);
|
print_rtl (dump_file, src);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
purge_dead_edges (bb);
|
purge_dead_edges (bb);
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
static bool
|
static bool
|
constprop_register (rtx insn, rtx from, rtx to, bool alter_jumps)
|
constprop_register (rtx insn, rtx from, rtx to, bool alter_jumps)
|
{
|
{
|
rtx sset;
|
rtx sset;
|
|
|
/* Check for reg or cc0 setting instructions followed by
|
/* Check for reg or cc0 setting instructions followed by
|
conditional branch instructions first. */
|
conditional branch instructions first. */
|
if (alter_jumps
|
if (alter_jumps
|
&& (sset = single_set (insn)) != NULL
|
&& (sset = single_set (insn)) != NULL
|
&& NEXT_INSN (insn)
|
&& NEXT_INSN (insn)
|
&& any_condjump_p (NEXT_INSN (insn)) && onlyjump_p (NEXT_INSN (insn)))
|
&& any_condjump_p (NEXT_INSN (insn)) && onlyjump_p (NEXT_INSN (insn)))
|
{
|
{
|
rtx dest = SET_DEST (sset);
|
rtx dest = SET_DEST (sset);
|
if ((REG_P (dest) || CC0_P (dest))
|
if ((REG_P (dest) || CC0_P (dest))
|
&& cprop_jump (BLOCK_FOR_INSN (insn), insn, NEXT_INSN (insn), from, to))
|
&& cprop_jump (BLOCK_FOR_INSN (insn), insn, NEXT_INSN (insn), from, to))
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Handle normal insns next. */
|
/* Handle normal insns next. */
|
if (NONJUMP_INSN_P (insn)
|
if (NONJUMP_INSN_P (insn)
|
&& try_replace_reg (from, to, insn))
|
&& try_replace_reg (from, to, insn))
|
return 1;
|
return 1;
|
|
|
/* Try to propagate a CONST_INT into a conditional jump.
|
/* Try to propagate a CONST_INT into a conditional jump.
|
We're pretty specific about what we will handle in this
|
We're pretty specific about what we will handle in this
|
code, we can extend this as necessary over time.
|
code, we can extend this as necessary over time.
|
|
|
Right now the insn in question must look like
|
Right now the insn in question must look like
|
(set (pc) (if_then_else ...)) */
|
(set (pc) (if_then_else ...)) */
|
else if (alter_jumps && any_condjump_p (insn) && onlyjump_p (insn))
|
else if (alter_jumps && any_condjump_p (insn) && onlyjump_p (insn))
|
return cprop_jump (BLOCK_FOR_INSN (insn), NULL, insn, from, to);
|
return cprop_jump (BLOCK_FOR_INSN (insn), NULL, insn, from, to);
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* Perform constant and copy propagation on INSN.
|
/* Perform constant and copy propagation on INSN.
|
The result is nonzero if a change was made. */
|
The result is nonzero if a change was made. */
|
|
|
static int
|
static int
|
cprop_insn (rtx insn, int alter_jumps)
|
cprop_insn (rtx insn, int alter_jumps)
|
{
|
{
|
struct reg_use *reg_used;
|
struct reg_use *reg_used;
|
int changed = 0;
|
int changed = 0;
|
rtx note;
|
rtx note;
|
|
|
if (!INSN_P (insn))
|
if (!INSN_P (insn))
|
return 0;
|
return 0;
|
|
|
reg_use_count = 0;
|
reg_use_count = 0;
|
note_uses (&PATTERN (insn), find_used_regs, NULL);
|
note_uses (&PATTERN (insn), find_used_regs, NULL);
|
|
|
note = find_reg_equal_equiv_note (insn);
|
note = find_reg_equal_equiv_note (insn);
|
|
|
/* We may win even when propagating constants into notes. */
|
/* We may win even when propagating constants into notes. */
|
if (note)
|
if (note)
|
find_used_regs (&XEXP (note, 0), NULL);
|
find_used_regs (&XEXP (note, 0), NULL);
|
|
|
for (reg_used = ®_use_table[0]; reg_use_count > 0;
|
for (reg_used = ®_use_table[0]; reg_use_count > 0;
|
reg_used++, reg_use_count--)
|
reg_used++, reg_use_count--)
|
{
|
{
|
unsigned int regno = REGNO (reg_used->reg_rtx);
|
unsigned int regno = REGNO (reg_used->reg_rtx);
|
rtx pat, src;
|
rtx pat, src;
|
struct expr *set;
|
struct expr *set;
|
|
|
/* Ignore registers created by GCSE.
|
/* Ignore registers created by GCSE.
|
We do this because ... */
|
We do this because ... */
|
if (regno >= max_gcse_regno)
|
if (regno >= max_gcse_regno)
|
continue;
|
continue;
|
|
|
/* If the register has already been set in this block, there's
|
/* If the register has already been set in this block, there's
|
nothing we can do. */
|
nothing we can do. */
|
if (! oprs_not_set_p (reg_used->reg_rtx, insn))
|
if (! oprs_not_set_p (reg_used->reg_rtx, insn))
|
continue;
|
continue;
|
|
|
/* Find an assignment that sets reg_used and is available
|
/* Find an assignment that sets reg_used and is available
|
at the start of the block. */
|
at the start of the block. */
|
set = find_avail_set (regno, insn);
|
set = find_avail_set (regno, insn);
|
if (! set)
|
if (! set)
|
continue;
|
continue;
|
|
|
pat = set->expr;
|
pat = set->expr;
|
/* ??? We might be able to handle PARALLELs. Later. */
|
/* ??? We might be able to handle PARALLELs. Later. */
|
gcc_assert (GET_CODE (pat) == SET);
|
gcc_assert (GET_CODE (pat) == SET);
|
|
|
src = SET_SRC (pat);
|
src = SET_SRC (pat);
|
|
|
/* Constant propagation. */
|
/* Constant propagation. */
|
if (gcse_constant_p (src))
|
if (gcse_constant_p (src))
|
{
|
{
|
if (constprop_register (insn, reg_used->reg_rtx, src, alter_jumps))
|
if (constprop_register (insn, reg_used->reg_rtx, src, alter_jumps))
|
{
|
{
|
changed = 1;
|
changed = 1;
|
global_const_prop_count++;
|
global_const_prop_count++;
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
{
|
{
|
fprintf (dump_file, "GLOBAL CONST-PROP: Replacing reg %d in ", regno);
|
fprintf (dump_file, "GLOBAL CONST-PROP: Replacing reg %d in ", regno);
|
fprintf (dump_file, "insn %d with constant ", INSN_UID (insn));
|
fprintf (dump_file, "insn %d with constant ", INSN_UID (insn));
|
print_rtl (dump_file, src);
|
print_rtl (dump_file, src);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
if (INSN_DELETED_P (insn))
|
if (INSN_DELETED_P (insn))
|
return 1;
|
return 1;
|
}
|
}
|
}
|
}
|
else if (REG_P (src)
|
else if (REG_P (src)
|
&& REGNO (src) >= FIRST_PSEUDO_REGISTER
|
&& REGNO (src) >= FIRST_PSEUDO_REGISTER
|
&& REGNO (src) != regno)
|
&& REGNO (src) != regno)
|
{
|
{
|
if (try_replace_reg (reg_used->reg_rtx, src, insn))
|
if (try_replace_reg (reg_used->reg_rtx, src, insn))
|
{
|
{
|
changed = 1;
|
changed = 1;
|
global_copy_prop_count++;
|
global_copy_prop_count++;
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
{
|
{
|
fprintf (dump_file, "GLOBAL COPY-PROP: Replacing reg %d in insn %d",
|
fprintf (dump_file, "GLOBAL COPY-PROP: Replacing reg %d in insn %d",
|
regno, INSN_UID (insn));
|
regno, INSN_UID (insn));
|
fprintf (dump_file, " with reg %d\n", REGNO (src));
|
fprintf (dump_file, " with reg %d\n", REGNO (src));
|
}
|
}
|
|
|
/* The original insn setting reg_used may or may not now be
|
/* The original insn setting reg_used may or may not now be
|
deletable. We leave the deletion to flow. */
|
deletable. We leave the deletion to flow. */
|
/* FIXME: If it turns out that the insn isn't deletable,
|
/* FIXME: If it turns out that the insn isn't deletable,
|
then we may have unnecessarily extended register lifetimes
|
then we may have unnecessarily extended register lifetimes
|
and made things worse. */
|
and made things worse. */
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* Like find_used_regs, but avoid recording uses that appear in
|
/* Like find_used_regs, but avoid recording uses that appear in
|
input-output contexts such as zero_extract or pre_dec. This
|
input-output contexts such as zero_extract or pre_dec. This
|
restricts the cases we consider to those for which local cprop
|
restricts the cases we consider to those for which local cprop
|
can legitimately make replacements. */
|
can legitimately make replacements. */
|
|
|
static void
|
static void
|
local_cprop_find_used_regs (rtx *xptr, void *data)
|
local_cprop_find_used_regs (rtx *xptr, void *data)
|
{
|
{
|
rtx x = *xptr;
|
rtx x = *xptr;
|
|
|
if (x == 0)
|
if (x == 0)
|
return;
|
return;
|
|
|
switch (GET_CODE (x))
|
switch (GET_CODE (x))
|
{
|
{
|
case ZERO_EXTRACT:
|
case ZERO_EXTRACT:
|
case SIGN_EXTRACT:
|
case SIGN_EXTRACT:
|
case STRICT_LOW_PART:
|
case STRICT_LOW_PART:
|
return;
|
return;
|
|
|
case PRE_DEC:
|
case PRE_DEC:
|
case PRE_INC:
|
case PRE_INC:
|
case POST_DEC:
|
case POST_DEC:
|
case POST_INC:
|
case POST_INC:
|
case PRE_MODIFY:
|
case PRE_MODIFY:
|
case POST_MODIFY:
|
case POST_MODIFY:
|
/* Can only legitimately appear this early in the context of
|
/* Can only legitimately appear this early in the context of
|
stack pushes for function arguments, but handle all of the
|
stack pushes for function arguments, but handle all of the
|
codes nonetheless. */
|
codes nonetheless. */
|
return;
|
return;
|
|
|
case SUBREG:
|
case SUBREG:
|
/* Setting a subreg of a register larger than word_mode leaves
|
/* Setting a subreg of a register larger than word_mode leaves
|
the non-written words unchanged. */
|
the non-written words unchanged. */
|
if (GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (x))) > BITS_PER_WORD)
|
if (GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (x))) > BITS_PER_WORD)
|
return;
|
return;
|
break;
|
break;
|
|
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
|
|
find_used_regs (xptr, data);
|
find_used_regs (xptr, data);
|
}
|
}
|
|
|
/* LIBCALL_SP is a zero-terminated array of insns at the end of a libcall;
|
/* LIBCALL_SP is a zero-terminated array of insns at the end of a libcall;
|
their REG_EQUAL notes need updating. */
|
their REG_EQUAL notes need updating. */
|
|
|
static bool
|
static bool
|
do_local_cprop (rtx x, rtx insn, bool alter_jumps, rtx *libcall_sp)
|
do_local_cprop (rtx x, rtx insn, bool alter_jumps, rtx *libcall_sp)
|
{
|
{
|
rtx newreg = NULL, newcnst = NULL;
|
rtx newreg = NULL, newcnst = NULL;
|
|
|
/* Rule out USE instructions and ASM statements as we don't want to
|
/* Rule out USE instructions and ASM statements as we don't want to
|
change the hard registers mentioned. */
|
change the hard registers mentioned. */
|
if (REG_P (x)
|
if (REG_P (x)
|
&& (REGNO (x) >= FIRST_PSEUDO_REGISTER
|
&& (REGNO (x) >= FIRST_PSEUDO_REGISTER
|
|| (GET_CODE (PATTERN (insn)) != USE
|
|| (GET_CODE (PATTERN (insn)) != USE
|
&& asm_noperands (PATTERN (insn)) < 0)))
|
&& asm_noperands (PATTERN (insn)) < 0)))
|
{
|
{
|
cselib_val *val = cselib_lookup (x, GET_MODE (x), 0);
|
cselib_val *val = cselib_lookup (x, GET_MODE (x), 0);
|
struct elt_loc_list *l;
|
struct elt_loc_list *l;
|
|
|
if (!val)
|
if (!val)
|
return false;
|
return false;
|
for (l = val->locs; l; l = l->next)
|
for (l = val->locs; l; l = l->next)
|
{
|
{
|
rtx this_rtx = l->loc;
|
rtx this_rtx = l->loc;
|
rtx note;
|
rtx note;
|
|
|
/* Don't CSE non-constant values out of libcall blocks. */
|
/* Don't CSE non-constant values out of libcall blocks. */
|
if (l->in_libcall && ! CONSTANT_P (this_rtx))
|
if (l->in_libcall && ! CONSTANT_P (this_rtx))
|
continue;
|
continue;
|
|
|
if (gcse_constant_p (this_rtx))
|
if (gcse_constant_p (this_rtx))
|
newcnst = this_rtx;
|
newcnst = this_rtx;
|
if (REG_P (this_rtx) && REGNO (this_rtx) >= FIRST_PSEUDO_REGISTER
|
if (REG_P (this_rtx) && REGNO (this_rtx) >= FIRST_PSEUDO_REGISTER
|
/* Don't copy propagate if it has attached REG_EQUIV note.
|
/* Don't copy propagate if it has attached REG_EQUIV note.
|
At this point this only function parameters should have
|
At this point this only function parameters should have
|
REG_EQUIV notes and if the argument slot is used somewhere
|
REG_EQUIV notes and if the argument slot is used somewhere
|
explicitly, it means address of parameter has been taken,
|
explicitly, it means address of parameter has been taken,
|
so we should not extend the lifetime of the pseudo. */
|
so we should not extend the lifetime of the pseudo. */
|
&& (!(note = find_reg_note (l->setting_insn, REG_EQUIV, NULL_RTX))
|
&& (!(note = find_reg_note (l->setting_insn, REG_EQUIV, NULL_RTX))
|
|| ! MEM_P (XEXP (note, 0))))
|
|| ! MEM_P (XEXP (note, 0))))
|
newreg = this_rtx;
|
newreg = this_rtx;
|
}
|
}
|
if (newcnst && constprop_register (insn, x, newcnst, alter_jumps))
|
if (newcnst && constprop_register (insn, x, newcnst, alter_jumps))
|
{
|
{
|
/* If we find a case where we can't fix the retval REG_EQUAL notes
|
/* If we find a case where we can't fix the retval REG_EQUAL notes
|
match the new register, we either have to abandon this replacement
|
match the new register, we either have to abandon this replacement
|
or fix delete_trivially_dead_insns to preserve the setting insn,
|
or fix delete_trivially_dead_insns to preserve the setting insn,
|
or make it delete the REG_EUAQL note, and fix up all passes that
|
or make it delete the REG_EUAQL note, and fix up all passes that
|
require the REG_EQUAL note there. */
|
require the REG_EQUAL note there. */
|
bool adjusted;
|
bool adjusted;
|
|
|
adjusted = adjust_libcall_notes (x, newcnst, insn, libcall_sp);
|
adjusted = adjust_libcall_notes (x, newcnst, insn, libcall_sp);
|
gcc_assert (adjusted);
|
gcc_assert (adjusted);
|
|
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
{
|
{
|
fprintf (dump_file, "LOCAL CONST-PROP: Replacing reg %d in ",
|
fprintf (dump_file, "LOCAL CONST-PROP: Replacing reg %d in ",
|
REGNO (x));
|
REGNO (x));
|
fprintf (dump_file, "insn %d with constant ",
|
fprintf (dump_file, "insn %d with constant ",
|
INSN_UID (insn));
|
INSN_UID (insn));
|
print_rtl (dump_file, newcnst);
|
print_rtl (dump_file, newcnst);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
local_const_prop_count++;
|
local_const_prop_count++;
|
return true;
|
return true;
|
}
|
}
|
else if (newreg && newreg != x && try_replace_reg (x, newreg, insn))
|
else if (newreg && newreg != x && try_replace_reg (x, newreg, insn))
|
{
|
{
|
adjust_libcall_notes (x, newreg, insn, libcall_sp);
|
adjust_libcall_notes (x, newreg, insn, libcall_sp);
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
{
|
{
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"LOCAL COPY-PROP: Replacing reg %d in insn %d",
|
"LOCAL COPY-PROP: Replacing reg %d in insn %d",
|
REGNO (x), INSN_UID (insn));
|
REGNO (x), INSN_UID (insn));
|
fprintf (dump_file, " with reg %d\n", REGNO (newreg));
|
fprintf (dump_file, " with reg %d\n", REGNO (newreg));
|
}
|
}
|
local_copy_prop_count++;
|
local_copy_prop_count++;
|
return true;
|
return true;
|
}
|
}
|
}
|
}
|
return false;
|
return false;
|
}
|
}
|
|
|
/* LIBCALL_SP is a zero-terminated array of insns at the end of a libcall;
|
/* LIBCALL_SP is a zero-terminated array of insns at the end of a libcall;
|
their REG_EQUAL notes need updating to reflect that OLDREG has been
|
their REG_EQUAL notes need updating to reflect that OLDREG has been
|
replaced with NEWVAL in INSN. Return true if all substitutions could
|
replaced with NEWVAL in INSN. Return true if all substitutions could
|
be made. */
|
be made. */
|
static bool
|
static bool
|
adjust_libcall_notes (rtx oldreg, rtx newval, rtx insn, rtx *libcall_sp)
|
adjust_libcall_notes (rtx oldreg, rtx newval, rtx insn, rtx *libcall_sp)
|
{
|
{
|
rtx end;
|
rtx end;
|
|
|
while ((end = *libcall_sp++))
|
while ((end = *libcall_sp++))
|
{
|
{
|
rtx note = find_reg_equal_equiv_note (end);
|
rtx note = find_reg_equal_equiv_note (end);
|
|
|
if (! note)
|
if (! note)
|
continue;
|
continue;
|
|
|
if (REG_P (newval))
|
if (REG_P (newval))
|
{
|
{
|
if (reg_set_between_p (newval, PREV_INSN (insn), end))
|
if (reg_set_between_p (newval, PREV_INSN (insn), end))
|
{
|
{
|
do
|
do
|
{
|
{
|
note = find_reg_equal_equiv_note (end);
|
note = find_reg_equal_equiv_note (end);
|
if (! note)
|
if (! note)
|
continue;
|
continue;
|
if (reg_mentioned_p (newval, XEXP (note, 0)))
|
if (reg_mentioned_p (newval, XEXP (note, 0)))
|
return false;
|
return false;
|
}
|
}
|
while ((end = *libcall_sp++));
|
while ((end = *libcall_sp++));
|
return true;
|
return true;
|
}
|
}
|
}
|
}
|
XEXP (note, 0) = simplify_replace_rtx (XEXP (note, 0), oldreg, newval);
|
XEXP (note, 0) = simplify_replace_rtx (XEXP (note, 0), oldreg, newval);
|
insn = end;
|
insn = end;
|
}
|
}
|
return true;
|
return true;
|
}
|
}
|
|
|
#define MAX_NESTED_LIBCALLS 9
|
#define MAX_NESTED_LIBCALLS 9
|
|
|
/* Do local const/copy propagation (i.e. within each basic block).
|
/* Do local const/copy propagation (i.e. within each basic block).
|
If ALTER_JUMPS is true, allow propagating into jump insns, which
|
If ALTER_JUMPS is true, allow propagating into jump insns, which
|
could modify the CFG. */
|
could modify the CFG. */
|
|
|
static void
|
static void
|
local_cprop_pass (bool alter_jumps)
|
local_cprop_pass (bool alter_jumps)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
rtx insn;
|
rtx insn;
|
struct reg_use *reg_used;
|
struct reg_use *reg_used;
|
rtx libcall_stack[MAX_NESTED_LIBCALLS + 1], *libcall_sp;
|
rtx libcall_stack[MAX_NESTED_LIBCALLS + 1], *libcall_sp;
|
bool changed = false;
|
bool changed = false;
|
|
|
cselib_init (false);
|
cselib_init (false);
|
libcall_sp = &libcall_stack[MAX_NESTED_LIBCALLS];
|
libcall_sp = &libcall_stack[MAX_NESTED_LIBCALLS];
|
*libcall_sp = 0;
|
*libcall_sp = 0;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
{
|
{
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
{
|
{
|
rtx note = find_reg_note (insn, REG_LIBCALL, NULL_RTX);
|
rtx note = find_reg_note (insn, REG_LIBCALL, NULL_RTX);
|
|
|
if (note)
|
if (note)
|
{
|
{
|
gcc_assert (libcall_sp != libcall_stack);
|
gcc_assert (libcall_sp != libcall_stack);
|
*--libcall_sp = XEXP (note, 0);
|
*--libcall_sp = XEXP (note, 0);
|
}
|
}
|
note = find_reg_note (insn, REG_RETVAL, NULL_RTX);
|
note = find_reg_note (insn, REG_RETVAL, NULL_RTX);
|
if (note)
|
if (note)
|
libcall_sp++;
|
libcall_sp++;
|
note = find_reg_equal_equiv_note (insn);
|
note = find_reg_equal_equiv_note (insn);
|
do
|
do
|
{
|
{
|
reg_use_count = 0;
|
reg_use_count = 0;
|
note_uses (&PATTERN (insn), local_cprop_find_used_regs,
|
note_uses (&PATTERN (insn), local_cprop_find_used_regs,
|
NULL);
|
NULL);
|
if (note)
|
if (note)
|
local_cprop_find_used_regs (&XEXP (note, 0), NULL);
|
local_cprop_find_used_regs (&XEXP (note, 0), NULL);
|
|
|
for (reg_used = ®_use_table[0]; reg_use_count > 0;
|
for (reg_used = ®_use_table[0]; reg_use_count > 0;
|
reg_used++, reg_use_count--)
|
reg_used++, reg_use_count--)
|
if (do_local_cprop (reg_used->reg_rtx, insn, alter_jumps,
|
if (do_local_cprop (reg_used->reg_rtx, insn, alter_jumps,
|
libcall_sp))
|
libcall_sp))
|
{
|
{
|
changed = true;
|
changed = true;
|
break;
|
break;
|
}
|
}
|
if (INSN_DELETED_P (insn))
|
if (INSN_DELETED_P (insn))
|
break;
|
break;
|
}
|
}
|
while (reg_use_count);
|
while (reg_use_count);
|
}
|
}
|
cselib_process_insn (insn);
|
cselib_process_insn (insn);
|
}
|
}
|
|
|
/* Forget everything at the end of a basic block. Make sure we are
|
/* Forget everything at the end of a basic block. Make sure we are
|
not inside a libcall, they should never cross basic blocks. */
|
not inside a libcall, they should never cross basic blocks. */
|
cselib_clear_table ();
|
cselib_clear_table ();
|
gcc_assert (libcall_sp == &libcall_stack[MAX_NESTED_LIBCALLS]);
|
gcc_assert (libcall_sp == &libcall_stack[MAX_NESTED_LIBCALLS]);
|
}
|
}
|
|
|
cselib_finish ();
|
cselib_finish ();
|
|
|
/* Global analysis may get into infinite loops for unreachable blocks. */
|
/* Global analysis may get into infinite loops for unreachable blocks. */
|
if (changed && alter_jumps)
|
if (changed && alter_jumps)
|
{
|
{
|
delete_unreachable_blocks ();
|
delete_unreachable_blocks ();
|
free_reg_set_mem ();
|
free_reg_set_mem ();
|
alloc_reg_set_mem (max_reg_num ());
|
alloc_reg_set_mem (max_reg_num ());
|
compute_sets ();
|
compute_sets ();
|
}
|
}
|
}
|
}
|
|
|
/* Forward propagate copies. This includes copies and constants. Return
|
/* Forward propagate copies. This includes copies and constants. Return
|
nonzero if a change was made. */
|
nonzero if a change was made. */
|
|
|
static int
|
static int
|
cprop (int alter_jumps)
|
cprop (int alter_jumps)
|
{
|
{
|
int changed;
|
int changed;
|
basic_block bb;
|
basic_block bb;
|
rtx insn;
|
rtx insn;
|
|
|
/* Note we start at block 1. */
|
/* Note we start at block 1. */
|
if (ENTRY_BLOCK_PTR->next_bb == EXIT_BLOCK_PTR)
|
if (ENTRY_BLOCK_PTR->next_bb == EXIT_BLOCK_PTR)
|
{
|
{
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
return 0;
|
return 0;
|
}
|
}
|
|
|
changed = 0;
|
changed = 0;
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR->next_bb->next_bb, EXIT_BLOCK_PTR, next_bb)
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR->next_bb->next_bb, EXIT_BLOCK_PTR, next_bb)
|
{
|
{
|
/* Reset tables used to keep track of what's still valid [since the
|
/* Reset tables used to keep track of what's still valid [since the
|
start of the block]. */
|
start of the block]. */
|
reset_opr_set_tables ();
|
reset_opr_set_tables ();
|
|
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
{
|
{
|
changed |= cprop_insn (insn, alter_jumps);
|
changed |= cprop_insn (insn, alter_jumps);
|
|
|
/* Keep track of everything modified by this insn. */
|
/* Keep track of everything modified by this insn. */
|
/* ??? Need to be careful w.r.t. mods done to INSN. Don't
|
/* ??? Need to be careful w.r.t. mods done to INSN. Don't
|
call mark_oprs_set if we turned the insn into a NOTE. */
|
call mark_oprs_set if we turned the insn into a NOTE. */
|
if (! NOTE_P (insn))
|
if (! NOTE_P (insn))
|
mark_oprs_set (insn);
|
mark_oprs_set (insn);
|
}
|
}
|
}
|
}
|
|
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
|
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* Similar to get_condition, only the resulting condition must be
|
/* Similar to get_condition, only the resulting condition must be
|
valid at JUMP, instead of at EARLIEST.
|
valid at JUMP, instead of at EARLIEST.
|
|
|
This differs from noce_get_condition in ifcvt.c in that we prefer not to
|
This differs from noce_get_condition in ifcvt.c in that we prefer not to
|
settle for the condition variable in the jump instruction being integral.
|
settle for the condition variable in the jump instruction being integral.
|
We prefer to be able to record the value of a user variable, rather than
|
We prefer to be able to record the value of a user variable, rather than
|
the value of a temporary used in a condition. This could be solved by
|
the value of a temporary used in a condition. This could be solved by
|
recording the value of *every* register scanned by canonicalize_condition,
|
recording the value of *every* register scanned by canonicalize_condition,
|
but this would require some code reorganization. */
|
but this would require some code reorganization. */
|
|
|
rtx
|
rtx
|
fis_get_condition (rtx jump)
|
fis_get_condition (rtx jump)
|
{
|
{
|
return get_condition (jump, NULL, false, true);
|
return get_condition (jump, NULL, false, true);
|
}
|
}
|
|
|
/* Check the comparison COND to see if we can safely form an implicit set from
|
/* Check the comparison COND to see if we can safely form an implicit set from
|
it. COND is either an EQ or NE comparison. */
|
it. COND is either an EQ or NE comparison. */
|
|
|
static bool
|
static bool
|
implicit_set_cond_p (rtx cond)
|
implicit_set_cond_p (rtx cond)
|
{
|
{
|
enum machine_mode mode = GET_MODE (XEXP (cond, 0));
|
enum machine_mode mode = GET_MODE (XEXP (cond, 0));
|
rtx cst = XEXP (cond, 1);
|
rtx cst = XEXP (cond, 1);
|
|
|
/* We can't perform this optimization if either operand might be or might
|
/* We can't perform this optimization if either operand might be or might
|
contain a signed zero. */
|
contain a signed zero. */
|
if (HONOR_SIGNED_ZEROS (mode))
|
if (HONOR_SIGNED_ZEROS (mode))
|
{
|
{
|
/* It is sufficient to check if CST is or contains a zero. We must
|
/* It is sufficient to check if CST is or contains a zero. We must
|
handle float, complex, and vector. If any subpart is a zero, then
|
handle float, complex, and vector. If any subpart is a zero, then
|
the optimization can't be performed. */
|
the optimization can't be performed. */
|
/* ??? The complex and vector checks are not implemented yet. We just
|
/* ??? The complex and vector checks are not implemented yet. We just
|
always return zero for them. */
|
always return zero for them. */
|
if (GET_CODE (cst) == CONST_DOUBLE)
|
if (GET_CODE (cst) == CONST_DOUBLE)
|
{
|
{
|
REAL_VALUE_TYPE d;
|
REAL_VALUE_TYPE d;
|
REAL_VALUE_FROM_CONST_DOUBLE (d, cst);
|
REAL_VALUE_FROM_CONST_DOUBLE (d, cst);
|
if (REAL_VALUES_EQUAL (d, dconst0))
|
if (REAL_VALUES_EQUAL (d, dconst0))
|
return 0;
|
return 0;
|
}
|
}
|
else
|
else
|
return 0;
|
return 0;
|
}
|
}
|
|
|
return gcse_constant_p (cst);
|
return gcse_constant_p (cst);
|
}
|
}
|
|
|
/* Find the implicit sets of a function. An "implicit set" is a constraint
|
/* Find the implicit sets of a function. An "implicit set" is a constraint
|
on the value of a variable, implied by a conditional jump. For example,
|
on the value of a variable, implied by a conditional jump. For example,
|
following "if (x == 2)", the then branch may be optimized as though the
|
following "if (x == 2)", the then branch may be optimized as though the
|
conditional performed an "explicit set", in this example, "x = 2". This
|
conditional performed an "explicit set", in this example, "x = 2". This
|
function records the set patterns that are implicit at the start of each
|
function records the set patterns that are implicit at the start of each
|
basic block. */
|
basic block. */
|
|
|
static void
|
static void
|
find_implicit_sets (void)
|
find_implicit_sets (void)
|
{
|
{
|
basic_block bb, dest;
|
basic_block bb, dest;
|
unsigned int count;
|
unsigned int count;
|
rtx cond, new;
|
rtx cond, new;
|
|
|
count = 0;
|
count = 0;
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
/* Check for more than one successor. */
|
/* Check for more than one successor. */
|
if (EDGE_COUNT (bb->succs) > 1)
|
if (EDGE_COUNT (bb->succs) > 1)
|
{
|
{
|
cond = fis_get_condition (BB_END (bb));
|
cond = fis_get_condition (BB_END (bb));
|
|
|
if (cond
|
if (cond
|
&& (GET_CODE (cond) == EQ || GET_CODE (cond) == NE)
|
&& (GET_CODE (cond) == EQ || GET_CODE (cond) == NE)
|
&& REG_P (XEXP (cond, 0))
|
&& REG_P (XEXP (cond, 0))
|
&& REGNO (XEXP (cond, 0)) >= FIRST_PSEUDO_REGISTER
|
&& REGNO (XEXP (cond, 0)) >= FIRST_PSEUDO_REGISTER
|
&& implicit_set_cond_p (cond))
|
&& implicit_set_cond_p (cond))
|
{
|
{
|
dest = GET_CODE (cond) == EQ ? BRANCH_EDGE (bb)->dest
|
dest = GET_CODE (cond) == EQ ? BRANCH_EDGE (bb)->dest
|
: FALLTHRU_EDGE (bb)->dest;
|
: FALLTHRU_EDGE (bb)->dest;
|
|
|
if (dest && single_pred_p (dest)
|
if (dest && single_pred_p (dest)
|
&& dest != EXIT_BLOCK_PTR)
|
&& dest != EXIT_BLOCK_PTR)
|
{
|
{
|
new = gen_rtx_SET (VOIDmode, XEXP (cond, 0),
|
new = gen_rtx_SET (VOIDmode, XEXP (cond, 0),
|
XEXP (cond, 1));
|
XEXP (cond, 1));
|
implicit_sets[dest->index] = new;
|
implicit_sets[dest->index] = new;
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf(dump_file, "Implicit set of reg %d in ",
|
fprintf(dump_file, "Implicit set of reg %d in ",
|
REGNO (XEXP (cond, 0)));
|
REGNO (XEXP (cond, 0)));
|
fprintf(dump_file, "basic block %d\n", dest->index);
|
fprintf(dump_file, "basic block %d\n", dest->index);
|
}
|
}
|
count++;
|
count++;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Found %d implicit sets\n", count);
|
fprintf (dump_file, "Found %d implicit sets\n", count);
|
}
|
}
|
|
|
/* Perform one copy/constant propagation pass.
|
/* Perform one copy/constant propagation pass.
|
PASS is the pass count. If CPROP_JUMPS is true, perform constant
|
PASS is the pass count. If CPROP_JUMPS is true, perform constant
|
propagation into conditional jumps. If BYPASS_JUMPS is true,
|
propagation into conditional jumps. If BYPASS_JUMPS is true,
|
perform conditional jump bypassing optimizations. */
|
perform conditional jump bypassing optimizations. */
|
|
|
static int
|
static int
|
one_cprop_pass (int pass, bool cprop_jumps, bool bypass_jumps)
|
one_cprop_pass (int pass, bool cprop_jumps, bool bypass_jumps)
|
{
|
{
|
int changed = 0;
|
int changed = 0;
|
|
|
global_const_prop_count = local_const_prop_count = 0;
|
global_const_prop_count = local_const_prop_count = 0;
|
global_copy_prop_count = local_copy_prop_count = 0;
|
global_copy_prop_count = local_copy_prop_count = 0;
|
|
|
local_cprop_pass (cprop_jumps);
|
local_cprop_pass (cprop_jumps);
|
|
|
/* Determine implicit sets. */
|
/* Determine implicit sets. */
|
implicit_sets = XCNEWVEC (rtx, last_basic_block);
|
implicit_sets = XCNEWVEC (rtx, last_basic_block);
|
find_implicit_sets ();
|
find_implicit_sets ();
|
|
|
alloc_hash_table (max_cuid, &set_hash_table, 1);
|
alloc_hash_table (max_cuid, &set_hash_table, 1);
|
compute_hash_table (&set_hash_table);
|
compute_hash_table (&set_hash_table);
|
|
|
/* Free implicit_sets before peak usage. */
|
/* Free implicit_sets before peak usage. */
|
free (implicit_sets);
|
free (implicit_sets);
|
implicit_sets = NULL;
|
implicit_sets = NULL;
|
|
|
if (dump_file)
|
if (dump_file)
|
dump_hash_table (dump_file, "SET", &set_hash_table);
|
dump_hash_table (dump_file, "SET", &set_hash_table);
|
if (set_hash_table.n_elems > 0)
|
if (set_hash_table.n_elems > 0)
|
{
|
{
|
alloc_cprop_mem (last_basic_block, set_hash_table.n_elems);
|
alloc_cprop_mem (last_basic_block, set_hash_table.n_elems);
|
compute_cprop_data ();
|
compute_cprop_data ();
|
changed = cprop (cprop_jumps);
|
changed = cprop (cprop_jumps);
|
if (bypass_jumps)
|
if (bypass_jumps)
|
changed |= bypass_conditional_jumps ();
|
changed |= bypass_conditional_jumps ();
|
free_cprop_mem ();
|
free_cprop_mem ();
|
}
|
}
|
|
|
free_hash_table (&set_hash_table);
|
free_hash_table (&set_hash_table);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "CPROP of %s, pass %d: %d bytes needed, ",
|
fprintf (dump_file, "CPROP of %s, pass %d: %d bytes needed, ",
|
current_function_name (), pass, bytes_used);
|
current_function_name (), pass, bytes_used);
|
fprintf (dump_file, "%d local const props, %d local copy props, ",
|
fprintf (dump_file, "%d local const props, %d local copy props, ",
|
local_const_prop_count, local_copy_prop_count);
|
local_const_prop_count, local_copy_prop_count);
|
fprintf (dump_file, "%d global const props, %d global copy props\n\n",
|
fprintf (dump_file, "%d global const props, %d global copy props\n\n",
|
global_const_prop_count, global_copy_prop_count);
|
global_const_prop_count, global_copy_prop_count);
|
}
|
}
|
/* Global analysis may get into infinite loops for unreachable blocks. */
|
/* Global analysis may get into infinite loops for unreachable blocks. */
|
if (changed && cprop_jumps)
|
if (changed && cprop_jumps)
|
delete_unreachable_blocks ();
|
delete_unreachable_blocks ();
|
|
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* Bypass conditional jumps. */
|
/* Bypass conditional jumps. */
|
|
|
/* The value of last_basic_block at the beginning of the jump_bypass
|
/* The value of last_basic_block at the beginning of the jump_bypass
|
pass. The use of redirect_edge_and_branch_force may introduce new
|
pass. The use of redirect_edge_and_branch_force may introduce new
|
basic blocks, but the data flow analysis is only valid for basic
|
basic blocks, but the data flow analysis is only valid for basic
|
block indices less than bypass_last_basic_block. */
|
block indices less than bypass_last_basic_block. */
|
|
|
static int bypass_last_basic_block;
|
static int bypass_last_basic_block;
|
|
|
/* Find a set of REGNO to a constant that is available at the end of basic
|
/* Find a set of REGNO to a constant that is available at the end of basic
|
block BB. Returns NULL if no such set is found. Based heavily upon
|
block BB. Returns NULL if no such set is found. Based heavily upon
|
find_avail_set. */
|
find_avail_set. */
|
|
|
static struct expr *
|
static struct expr *
|
find_bypass_set (int regno, int bb)
|
find_bypass_set (int regno, int bb)
|
{
|
{
|
struct expr *result = 0;
|
struct expr *result = 0;
|
|
|
for (;;)
|
for (;;)
|
{
|
{
|
rtx src;
|
rtx src;
|
struct expr *set = lookup_set (regno, &set_hash_table);
|
struct expr *set = lookup_set (regno, &set_hash_table);
|
|
|
while (set)
|
while (set)
|
{
|
{
|
if (TEST_BIT (cprop_avout[bb], set->bitmap_index))
|
if (TEST_BIT (cprop_avout[bb], set->bitmap_index))
|
break;
|
break;
|
set = next_set (regno, set);
|
set = next_set (regno, set);
|
}
|
}
|
|
|
if (set == 0)
|
if (set == 0)
|
break;
|
break;
|
|
|
gcc_assert (GET_CODE (set->expr) == SET);
|
gcc_assert (GET_CODE (set->expr) == SET);
|
|
|
src = SET_SRC (set->expr);
|
src = SET_SRC (set->expr);
|
if (gcse_constant_p (src))
|
if (gcse_constant_p (src))
|
result = set;
|
result = set;
|
|
|
if (! REG_P (src))
|
if (! REG_P (src))
|
break;
|
break;
|
|
|
regno = REGNO (src);
|
regno = REGNO (src);
|
}
|
}
|
return result;
|
return result;
|
}
|
}
|
|
|
|
|
/* Subroutine of bypass_block that checks whether a pseudo is killed by
|
/* Subroutine of bypass_block that checks whether a pseudo is killed by
|
any of the instructions inserted on an edge. Jump bypassing places
|
any of the instructions inserted on an edge. Jump bypassing places
|
condition code setters on CFG edges using insert_insn_on_edge. This
|
condition code setters on CFG edges using insert_insn_on_edge. This
|
function is required to check that our data flow analysis is still
|
function is required to check that our data flow analysis is still
|
valid prior to commit_edge_insertions. */
|
valid prior to commit_edge_insertions. */
|
|
|
static bool
|
static bool
|
reg_killed_on_edge (rtx reg, edge e)
|
reg_killed_on_edge (rtx reg, edge e)
|
{
|
{
|
rtx insn;
|
rtx insn;
|
|
|
for (insn = e->insns.r; insn; insn = NEXT_INSN (insn))
|
for (insn = e->insns.r; insn; insn = NEXT_INSN (insn))
|
if (INSN_P (insn) && reg_set_p (reg, insn))
|
if (INSN_P (insn) && reg_set_p (reg, insn))
|
return true;
|
return true;
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Subroutine of bypass_conditional_jumps that attempts to bypass the given
|
/* Subroutine of bypass_conditional_jumps that attempts to bypass the given
|
basic block BB which has more than one predecessor. If not NULL, SETCC
|
basic block BB which has more than one predecessor. If not NULL, SETCC
|
is the first instruction of BB, which is immediately followed by JUMP_INSN
|
is the first instruction of BB, which is immediately followed by JUMP_INSN
|
JUMP. Otherwise, SETCC is NULL, and JUMP is the first insn of BB.
|
JUMP. Otherwise, SETCC is NULL, and JUMP is the first insn of BB.
|
Returns nonzero if a change was made.
|
Returns nonzero if a change was made.
|
|
|
During the jump bypassing pass, we may place copies of SETCC instructions
|
During the jump bypassing pass, we may place copies of SETCC instructions
|
on CFG edges. The following routine must be careful to pay attention to
|
on CFG edges. The following routine must be careful to pay attention to
|
these inserted insns when performing its transformations. */
|
these inserted insns when performing its transformations. */
|
|
|
static int
|
static int
|
bypass_block (basic_block bb, rtx setcc, rtx jump)
|
bypass_block (basic_block bb, rtx setcc, rtx jump)
|
{
|
{
|
rtx insn, note;
|
rtx insn, note;
|
edge e, edest;
|
edge e, edest;
|
int i, change;
|
int i, change;
|
int may_be_loop_header;
|
int may_be_loop_header;
|
unsigned removed_p;
|
unsigned removed_p;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
insn = (setcc != NULL) ? setcc : jump;
|
insn = (setcc != NULL) ? setcc : jump;
|
|
|
/* Determine set of register uses in INSN. */
|
/* Determine set of register uses in INSN. */
|
reg_use_count = 0;
|
reg_use_count = 0;
|
note_uses (&PATTERN (insn), find_used_regs, NULL);
|
note_uses (&PATTERN (insn), find_used_regs, NULL);
|
note = find_reg_equal_equiv_note (insn);
|
note = find_reg_equal_equiv_note (insn);
|
if (note)
|
if (note)
|
find_used_regs (&XEXP (note, 0), NULL);
|
find_used_regs (&XEXP (note, 0), NULL);
|
|
|
may_be_loop_header = false;
|
may_be_loop_header = false;
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
if (e->flags & EDGE_DFS_BACK)
|
if (e->flags & EDGE_DFS_BACK)
|
{
|
{
|
may_be_loop_header = true;
|
may_be_loop_header = true;
|
break;
|
break;
|
}
|
}
|
|
|
change = 0;
|
change = 0;
|
for (ei = ei_start (bb->preds); (e = ei_safe_edge (ei)); )
|
for (ei = ei_start (bb->preds); (e = ei_safe_edge (ei)); )
|
{
|
{
|
removed_p = 0;
|
removed_p = 0;
|
|
|
if (e->flags & EDGE_COMPLEX)
|
if (e->flags & EDGE_COMPLEX)
|
{
|
{
|
ei_next (&ei);
|
ei_next (&ei);
|
continue;
|
continue;
|
}
|
}
|
|
|
/* We can't redirect edges from new basic blocks. */
|
/* We can't redirect edges from new basic blocks. */
|
if (e->src->index >= bypass_last_basic_block)
|
if (e->src->index >= bypass_last_basic_block)
|
{
|
{
|
ei_next (&ei);
|
ei_next (&ei);
|
continue;
|
continue;
|
}
|
}
|
|
|
/* The irreducible loops created by redirecting of edges entering the
|
/* The irreducible loops created by redirecting of edges entering the
|
loop from outside would decrease effectiveness of some of the following
|
loop from outside would decrease effectiveness of some of the following
|
optimizations, so prevent this. */
|
optimizations, so prevent this. */
|
if (may_be_loop_header
|
if (may_be_loop_header
|
&& !(e->flags & EDGE_DFS_BACK))
|
&& !(e->flags & EDGE_DFS_BACK))
|
{
|
{
|
ei_next (&ei);
|
ei_next (&ei);
|
continue;
|
continue;
|
}
|
}
|
|
|
for (i = 0; i < reg_use_count; i++)
|
for (i = 0; i < reg_use_count; i++)
|
{
|
{
|
struct reg_use *reg_used = ®_use_table[i];
|
struct reg_use *reg_used = ®_use_table[i];
|
unsigned int regno = REGNO (reg_used->reg_rtx);
|
unsigned int regno = REGNO (reg_used->reg_rtx);
|
basic_block dest, old_dest;
|
basic_block dest, old_dest;
|
struct expr *set;
|
struct expr *set;
|
rtx src, new;
|
rtx src, new;
|
|
|
if (regno >= max_gcse_regno)
|
if (regno >= max_gcse_regno)
|
continue;
|
continue;
|
|
|
set = find_bypass_set (regno, e->src->index);
|
set = find_bypass_set (regno, e->src->index);
|
|
|
if (! set)
|
if (! set)
|
continue;
|
continue;
|
|
|
/* Check the data flow is valid after edge insertions. */
|
/* Check the data flow is valid after edge insertions. */
|
if (e->insns.r && reg_killed_on_edge (reg_used->reg_rtx, e))
|
if (e->insns.r && reg_killed_on_edge (reg_used->reg_rtx, e))
|
continue;
|
continue;
|
|
|
src = SET_SRC (pc_set (jump));
|
src = SET_SRC (pc_set (jump));
|
|
|
if (setcc != NULL)
|
if (setcc != NULL)
|
src = simplify_replace_rtx (src,
|
src = simplify_replace_rtx (src,
|
SET_DEST (PATTERN (setcc)),
|
SET_DEST (PATTERN (setcc)),
|
SET_SRC (PATTERN (setcc)));
|
SET_SRC (PATTERN (setcc)));
|
|
|
new = simplify_replace_rtx (src, reg_used->reg_rtx,
|
new = simplify_replace_rtx (src, reg_used->reg_rtx,
|
SET_SRC (set->expr));
|
SET_SRC (set->expr));
|
|
|
/* Jump bypassing may have already placed instructions on
|
/* Jump bypassing may have already placed instructions on
|
edges of the CFG. We can't bypass an outgoing edge that
|
edges of the CFG. We can't bypass an outgoing edge that
|
has instructions associated with it, as these insns won't
|
has instructions associated with it, as these insns won't
|
get executed if the incoming edge is redirected. */
|
get executed if the incoming edge is redirected. */
|
|
|
if (new == pc_rtx)
|
if (new == pc_rtx)
|
{
|
{
|
edest = FALLTHRU_EDGE (bb);
|
edest = FALLTHRU_EDGE (bb);
|
dest = edest->insns.r ? NULL : edest->dest;
|
dest = edest->insns.r ? NULL : edest->dest;
|
}
|
}
|
else if (GET_CODE (new) == LABEL_REF)
|
else if (GET_CODE (new) == LABEL_REF)
|
{
|
{
|
dest = BLOCK_FOR_INSN (XEXP (new, 0));
|
dest = BLOCK_FOR_INSN (XEXP (new, 0));
|
/* Don't bypass edges containing instructions. */
|
/* Don't bypass edges containing instructions. */
|
edest = find_edge (bb, dest);
|
edest = find_edge (bb, dest);
|
if (edest && edest->insns.r)
|
if (edest && edest->insns.r)
|
dest = NULL;
|
dest = NULL;
|
}
|
}
|
else
|
else
|
dest = NULL;
|
dest = NULL;
|
|
|
/* Avoid unification of the edge with other edges from original
|
/* Avoid unification of the edge with other edges from original
|
branch. We would end up emitting the instruction on "both"
|
branch. We would end up emitting the instruction on "both"
|
edges. */
|
edges. */
|
|
|
if (dest && setcc && !CC0_P (SET_DEST (PATTERN (setcc)))
|
if (dest && setcc && !CC0_P (SET_DEST (PATTERN (setcc)))
|
&& find_edge (e->src, dest))
|
&& find_edge (e->src, dest))
|
dest = NULL;
|
dest = NULL;
|
|
|
old_dest = e->dest;
|
old_dest = e->dest;
|
if (dest != NULL
|
if (dest != NULL
|
&& dest != old_dest
|
&& dest != old_dest
|
&& dest != EXIT_BLOCK_PTR)
|
&& dest != EXIT_BLOCK_PTR)
|
{
|
{
|
redirect_edge_and_branch_force (e, dest);
|
redirect_edge_and_branch_force (e, dest);
|
|
|
/* Copy the register setter to the redirected edge.
|
/* Copy the register setter to the redirected edge.
|
Don't copy CC0 setters, as CC0 is dead after jump. */
|
Don't copy CC0 setters, as CC0 is dead after jump. */
|
if (setcc)
|
if (setcc)
|
{
|
{
|
rtx pat = PATTERN (setcc);
|
rtx pat = PATTERN (setcc);
|
if (!CC0_P (SET_DEST (pat)))
|
if (!CC0_P (SET_DEST (pat)))
|
insert_insn_on_edge (copy_insn (pat), e);
|
insert_insn_on_edge (copy_insn (pat), e);
|
}
|
}
|
|
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
{
|
{
|
fprintf (dump_file, "JUMP-BYPASS: Proved reg %d "
|
fprintf (dump_file, "JUMP-BYPASS: Proved reg %d "
|
"in jump_insn %d equals constant ",
|
"in jump_insn %d equals constant ",
|
regno, INSN_UID (jump));
|
regno, INSN_UID (jump));
|
print_rtl (dump_file, SET_SRC (set->expr));
|
print_rtl (dump_file, SET_SRC (set->expr));
|
fprintf (dump_file, "\nBypass edge from %d->%d to %d\n",
|
fprintf (dump_file, "\nBypass edge from %d->%d to %d\n",
|
e->src->index, old_dest->index, dest->index);
|
e->src->index, old_dest->index, dest->index);
|
}
|
}
|
change = 1;
|
change = 1;
|
removed_p = 1;
|
removed_p = 1;
|
break;
|
break;
|
}
|
}
|
}
|
}
|
if (!removed_p)
|
if (!removed_p)
|
ei_next (&ei);
|
ei_next (&ei);
|
}
|
}
|
return change;
|
return change;
|
}
|
}
|
|
|
/* Find basic blocks with more than one predecessor that only contain a
|
/* Find basic blocks with more than one predecessor that only contain a
|
single conditional jump. If the result of the comparison is known at
|
single conditional jump. If the result of the comparison is known at
|
compile-time from any incoming edge, redirect that edge to the
|
compile-time from any incoming edge, redirect that edge to the
|
appropriate target. Returns nonzero if a change was made.
|
appropriate target. Returns nonzero if a change was made.
|
|
|
This function is now mis-named, because we also handle indirect jumps. */
|
This function is now mis-named, because we also handle indirect jumps. */
|
|
|
static int
|
static int
|
bypass_conditional_jumps (void)
|
bypass_conditional_jumps (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
int changed;
|
int changed;
|
rtx setcc;
|
rtx setcc;
|
rtx insn;
|
rtx insn;
|
rtx dest;
|
rtx dest;
|
|
|
/* Note we start at block 1. */
|
/* Note we start at block 1. */
|
if (ENTRY_BLOCK_PTR->next_bb == EXIT_BLOCK_PTR)
|
if (ENTRY_BLOCK_PTR->next_bb == EXIT_BLOCK_PTR)
|
return 0;
|
return 0;
|
|
|
bypass_last_basic_block = last_basic_block;
|
bypass_last_basic_block = last_basic_block;
|
mark_dfs_back_edges ();
|
mark_dfs_back_edges ();
|
|
|
changed = 0;
|
changed = 0;
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR->next_bb->next_bb,
|
FOR_BB_BETWEEN (bb, ENTRY_BLOCK_PTR->next_bb->next_bb,
|
EXIT_BLOCK_PTR, next_bb)
|
EXIT_BLOCK_PTR, next_bb)
|
{
|
{
|
/* Check for more than one predecessor. */
|
/* Check for more than one predecessor. */
|
if (!single_pred_p (bb))
|
if (!single_pred_p (bb))
|
{
|
{
|
setcc = NULL_RTX;
|
setcc = NULL_RTX;
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
if (NONJUMP_INSN_P (insn))
|
if (NONJUMP_INSN_P (insn))
|
{
|
{
|
if (setcc)
|
if (setcc)
|
break;
|
break;
|
if (GET_CODE (PATTERN (insn)) != SET)
|
if (GET_CODE (PATTERN (insn)) != SET)
|
break;
|
break;
|
|
|
dest = SET_DEST (PATTERN (insn));
|
dest = SET_DEST (PATTERN (insn));
|
if (REG_P (dest) || CC0_P (dest))
|
if (REG_P (dest) || CC0_P (dest))
|
setcc = insn;
|
setcc = insn;
|
else
|
else
|
break;
|
break;
|
}
|
}
|
else if (JUMP_P (insn))
|
else if (JUMP_P (insn))
|
{
|
{
|
if ((any_condjump_p (insn) || computed_jump_p (insn))
|
if ((any_condjump_p (insn) || computed_jump_p (insn))
|
&& onlyjump_p (insn))
|
&& onlyjump_p (insn))
|
changed |= bypass_block (bb, setcc, insn);
|
changed |= bypass_block (bb, setcc, insn);
|
break;
|
break;
|
}
|
}
|
else if (INSN_P (insn))
|
else if (INSN_P (insn))
|
break;
|
break;
|
}
|
}
|
}
|
}
|
|
|
/* If we bypassed any register setting insns, we inserted a
|
/* If we bypassed any register setting insns, we inserted a
|
copy on the redirected edge. These need to be committed. */
|
copy on the redirected edge. These need to be committed. */
|
if (changed)
|
if (changed)
|
commit_edge_insertions();
|
commit_edge_insertions();
|
|
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* Compute PRE+LCM working variables. */
|
/* Compute PRE+LCM working variables. */
|
|
|
/* Local properties of expressions. */
|
/* Local properties of expressions. */
|
/* Nonzero for expressions that are transparent in the block. */
|
/* Nonzero for expressions that are transparent in the block. */
|
static sbitmap *transp;
|
static sbitmap *transp;
|
|
|
/* Nonzero for expressions that are transparent at the end of the block.
|
/* Nonzero for expressions that are transparent at the end of the block.
|
This is only zero for expressions killed by abnormal critical edge
|
This is only zero for expressions killed by abnormal critical edge
|
created by a calls. */
|
created by a calls. */
|
static sbitmap *transpout;
|
static sbitmap *transpout;
|
|
|
/* Nonzero for expressions that are computed (available) in the block. */
|
/* Nonzero for expressions that are computed (available) in the block. */
|
static sbitmap *comp;
|
static sbitmap *comp;
|
|
|
/* Nonzero for expressions that are locally anticipatable in the block. */
|
/* Nonzero for expressions that are locally anticipatable in the block. */
|
static sbitmap *antloc;
|
static sbitmap *antloc;
|
|
|
/* Nonzero for expressions where this block is an optimal computation
|
/* Nonzero for expressions where this block is an optimal computation
|
point. */
|
point. */
|
static sbitmap *pre_optimal;
|
static sbitmap *pre_optimal;
|
|
|
/* Nonzero for expressions which are redundant in a particular block. */
|
/* Nonzero for expressions which are redundant in a particular block. */
|
static sbitmap *pre_redundant;
|
static sbitmap *pre_redundant;
|
|
|
/* Nonzero for expressions which should be inserted on a specific edge. */
|
/* Nonzero for expressions which should be inserted on a specific edge. */
|
static sbitmap *pre_insert_map;
|
static sbitmap *pre_insert_map;
|
|
|
/* Nonzero for expressions which should be deleted in a specific block. */
|
/* Nonzero for expressions which should be deleted in a specific block. */
|
static sbitmap *pre_delete_map;
|
static sbitmap *pre_delete_map;
|
|
|
/* Contains the edge_list returned by pre_edge_lcm. */
|
/* Contains the edge_list returned by pre_edge_lcm. */
|
static struct edge_list *edge_list;
|
static struct edge_list *edge_list;
|
|
|
/* Redundant insns. */
|
/* Redundant insns. */
|
static sbitmap pre_redundant_insns;
|
static sbitmap pre_redundant_insns;
|
|
|
/* Allocate vars used for PRE analysis. */
|
/* Allocate vars used for PRE analysis. */
|
|
|
static void
|
static void
|
alloc_pre_mem (int n_blocks, int n_exprs)
|
alloc_pre_mem (int n_blocks, int n_exprs)
|
{
|
{
|
transp = sbitmap_vector_alloc (n_blocks, n_exprs);
|
transp = sbitmap_vector_alloc (n_blocks, n_exprs);
|
comp = sbitmap_vector_alloc (n_blocks, n_exprs);
|
comp = sbitmap_vector_alloc (n_blocks, n_exprs);
|
antloc = sbitmap_vector_alloc (n_blocks, n_exprs);
|
antloc = sbitmap_vector_alloc (n_blocks, n_exprs);
|
|
|
pre_optimal = NULL;
|
pre_optimal = NULL;
|
pre_redundant = NULL;
|
pre_redundant = NULL;
|
pre_insert_map = NULL;
|
pre_insert_map = NULL;
|
pre_delete_map = NULL;
|
pre_delete_map = NULL;
|
ae_kill = sbitmap_vector_alloc (n_blocks, n_exprs);
|
ae_kill = sbitmap_vector_alloc (n_blocks, n_exprs);
|
|
|
/* pre_insert and pre_delete are allocated later. */
|
/* pre_insert and pre_delete are allocated later. */
|
}
|
}
|
|
|
/* Free vars used for PRE analysis. */
|
/* Free vars used for PRE analysis. */
|
|
|
static void
|
static void
|
free_pre_mem (void)
|
free_pre_mem (void)
|
{
|
{
|
sbitmap_vector_free (transp);
|
sbitmap_vector_free (transp);
|
sbitmap_vector_free (comp);
|
sbitmap_vector_free (comp);
|
|
|
/* ANTLOC and AE_KILL are freed just after pre_lcm finishes. */
|
/* ANTLOC and AE_KILL are freed just after pre_lcm finishes. */
|
|
|
if (pre_optimal)
|
if (pre_optimal)
|
sbitmap_vector_free (pre_optimal);
|
sbitmap_vector_free (pre_optimal);
|
if (pre_redundant)
|
if (pre_redundant)
|
sbitmap_vector_free (pre_redundant);
|
sbitmap_vector_free (pre_redundant);
|
if (pre_insert_map)
|
if (pre_insert_map)
|
sbitmap_vector_free (pre_insert_map);
|
sbitmap_vector_free (pre_insert_map);
|
if (pre_delete_map)
|
if (pre_delete_map)
|
sbitmap_vector_free (pre_delete_map);
|
sbitmap_vector_free (pre_delete_map);
|
|
|
transp = comp = NULL;
|
transp = comp = NULL;
|
pre_optimal = pre_redundant = pre_insert_map = pre_delete_map = NULL;
|
pre_optimal = pre_redundant = pre_insert_map = pre_delete_map = NULL;
|
}
|
}
|
|
|
/* Top level routine to do the dataflow analysis needed by PRE. */
|
/* Top level routine to do the dataflow analysis needed by PRE. */
|
|
|
static void
|
static void
|
compute_pre_data (void)
|
compute_pre_data (void)
|
{
|
{
|
sbitmap trapping_expr;
|
sbitmap trapping_expr;
|
basic_block bb;
|
basic_block bb;
|
unsigned int ui;
|
unsigned int ui;
|
|
|
compute_local_properties (transp, comp, antloc, &expr_hash_table);
|
compute_local_properties (transp, comp, antloc, &expr_hash_table);
|
sbitmap_vector_zero (ae_kill, last_basic_block);
|
sbitmap_vector_zero (ae_kill, last_basic_block);
|
|
|
/* Collect expressions which might trap. */
|
/* Collect expressions which might trap. */
|
trapping_expr = sbitmap_alloc (expr_hash_table.n_elems);
|
trapping_expr = sbitmap_alloc (expr_hash_table.n_elems);
|
sbitmap_zero (trapping_expr);
|
sbitmap_zero (trapping_expr);
|
for (ui = 0; ui < expr_hash_table.size; ui++)
|
for (ui = 0; ui < expr_hash_table.size; ui++)
|
{
|
{
|
struct expr *e;
|
struct expr *e;
|
for (e = expr_hash_table.table[ui]; e != NULL; e = e->next_same_hash)
|
for (e = expr_hash_table.table[ui]; e != NULL; e = e->next_same_hash)
|
if (may_trap_p (e->expr))
|
if (may_trap_p (e->expr))
|
SET_BIT (trapping_expr, e->bitmap_index);
|
SET_BIT (trapping_expr, e->bitmap_index);
|
}
|
}
|
|
|
/* Compute ae_kill for each basic block using:
|
/* Compute ae_kill for each basic block using:
|
|
|
~(TRANSP | COMP)
|
~(TRANSP | COMP)
|
*/
|
*/
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
edge e;
|
edge e;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
/* If the current block is the destination of an abnormal edge, we
|
/* If the current block is the destination of an abnormal edge, we
|
kill all trapping expressions because we won't be able to properly
|
kill all trapping expressions because we won't be able to properly
|
place the instruction on the edge. So make them neither
|
place the instruction on the edge. So make them neither
|
anticipatable nor transparent. This is fairly conservative. */
|
anticipatable nor transparent. This is fairly conservative. */
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
FOR_EACH_EDGE (e, ei, bb->preds)
|
if (e->flags & EDGE_ABNORMAL)
|
if (e->flags & EDGE_ABNORMAL)
|
{
|
{
|
sbitmap_difference (antloc[bb->index], antloc[bb->index], trapping_expr);
|
sbitmap_difference (antloc[bb->index], antloc[bb->index], trapping_expr);
|
sbitmap_difference (transp[bb->index], transp[bb->index], trapping_expr);
|
sbitmap_difference (transp[bb->index], transp[bb->index], trapping_expr);
|
break;
|
break;
|
}
|
}
|
|
|
sbitmap_a_or_b (ae_kill[bb->index], transp[bb->index], comp[bb->index]);
|
sbitmap_a_or_b (ae_kill[bb->index], transp[bb->index], comp[bb->index]);
|
sbitmap_not (ae_kill[bb->index], ae_kill[bb->index]);
|
sbitmap_not (ae_kill[bb->index], ae_kill[bb->index]);
|
}
|
}
|
|
|
edge_list = pre_edge_lcm (expr_hash_table.n_elems, transp, comp, antloc,
|
edge_list = pre_edge_lcm (expr_hash_table.n_elems, transp, comp, antloc,
|
ae_kill, &pre_insert_map, &pre_delete_map);
|
ae_kill, &pre_insert_map, &pre_delete_map);
|
sbitmap_vector_free (antloc);
|
sbitmap_vector_free (antloc);
|
antloc = NULL;
|
antloc = NULL;
|
sbitmap_vector_free (ae_kill);
|
sbitmap_vector_free (ae_kill);
|
ae_kill = NULL;
|
ae_kill = NULL;
|
sbitmap_free (trapping_expr);
|
sbitmap_free (trapping_expr);
|
}
|
}
|
|
|
/* PRE utilities */
|
/* PRE utilities */
|
|
|
/* Return nonzero if an occurrence of expression EXPR in OCCR_BB would reach
|
/* Return nonzero if an occurrence of expression EXPR in OCCR_BB would reach
|
block BB.
|
block BB.
|
|
|
VISITED is a pointer to a working buffer for tracking which BB's have
|
VISITED is a pointer to a working buffer for tracking which BB's have
|
been visited. It is NULL for the top-level call.
|
been visited. It is NULL for the top-level call.
|
|
|
We treat reaching expressions that go through blocks containing the same
|
We treat reaching expressions that go through blocks containing the same
|
reaching expression as "not reaching". E.g. if EXPR is generated in blocks
|
reaching expression as "not reaching". E.g. if EXPR is generated in blocks
|
2 and 3, INSN is in block 4, and 2->3->4, we treat the expression in block
|
2 and 3, INSN is in block 4, and 2->3->4, we treat the expression in block
|
2 as not reaching. The intent is to improve the probability of finding
|
2 as not reaching. The intent is to improve the probability of finding
|
only one reaching expression and to reduce register lifetimes by picking
|
only one reaching expression and to reduce register lifetimes by picking
|
the closest such expression. */
|
the closest such expression. */
|
|
|
static int
|
static int
|
pre_expr_reaches_here_p_work (basic_block occr_bb, struct expr *expr, basic_block bb, char *visited)
|
pre_expr_reaches_here_p_work (basic_block occr_bb, struct expr *expr, basic_block bb, char *visited)
|
{
|
{
|
edge pred;
|
edge pred;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
FOR_EACH_EDGE (pred, ei, bb->preds)
|
FOR_EACH_EDGE (pred, ei, bb->preds)
|
{
|
{
|
basic_block pred_bb = pred->src;
|
basic_block pred_bb = pred->src;
|
|
|
if (pred->src == ENTRY_BLOCK_PTR
|
if (pred->src == ENTRY_BLOCK_PTR
|
/* Has predecessor has already been visited? */
|
/* Has predecessor has already been visited? */
|
|| visited[pred_bb->index])
|
|| visited[pred_bb->index])
|
;/* Nothing to do. */
|
;/* Nothing to do. */
|
|
|
/* Does this predecessor generate this expression? */
|
/* Does this predecessor generate this expression? */
|
else if (TEST_BIT (comp[pred_bb->index], expr->bitmap_index))
|
else if (TEST_BIT (comp[pred_bb->index], expr->bitmap_index))
|
{
|
{
|
/* Is this the occurrence we're looking for?
|
/* Is this the occurrence we're looking for?
|
Note that there's only one generating occurrence per block
|
Note that there's only one generating occurrence per block
|
so we just need to check the block number. */
|
so we just need to check the block number. */
|
if (occr_bb == pred_bb)
|
if (occr_bb == pred_bb)
|
return 1;
|
return 1;
|
|
|
visited[pred_bb->index] = 1;
|
visited[pred_bb->index] = 1;
|
}
|
}
|
/* Ignore this predecessor if it kills the expression. */
|
/* Ignore this predecessor if it kills the expression. */
|
else if (! TEST_BIT (transp[pred_bb->index], expr->bitmap_index))
|
else if (! TEST_BIT (transp[pred_bb->index], expr->bitmap_index))
|
visited[pred_bb->index] = 1;
|
visited[pred_bb->index] = 1;
|
|
|
/* Neither gen nor kill. */
|
/* Neither gen nor kill. */
|
else
|
else
|
{
|
{
|
visited[pred_bb->index] = 1;
|
visited[pred_bb->index] = 1;
|
if (pre_expr_reaches_here_p_work (occr_bb, expr, pred_bb, visited))
|
if (pre_expr_reaches_here_p_work (occr_bb, expr, pred_bb, visited))
|
return 1;
|
return 1;
|
}
|
}
|
}
|
}
|
|
|
/* All paths have been checked. */
|
/* All paths have been checked. */
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* The wrapper for pre_expr_reaches_here_work that ensures that any
|
/* The wrapper for pre_expr_reaches_here_work that ensures that any
|
memory allocated for that function is returned. */
|
memory allocated for that function is returned. */
|
|
|
static int
|
static int
|
pre_expr_reaches_here_p (basic_block occr_bb, struct expr *expr, basic_block bb)
|
pre_expr_reaches_here_p (basic_block occr_bb, struct expr *expr, basic_block bb)
|
{
|
{
|
int rval;
|
int rval;
|
char *visited = XCNEWVEC (char, last_basic_block);
|
char *visited = XCNEWVEC (char, last_basic_block);
|
|
|
rval = pre_expr_reaches_here_p_work (occr_bb, expr, bb, visited);
|
rval = pre_expr_reaches_here_p_work (occr_bb, expr, bb, visited);
|
|
|
free (visited);
|
free (visited);
|
return rval;
|
return rval;
|
}
|
}
|
|
|
|
|
/* Given an expr, generate RTL which we can insert at the end of a BB,
|
/* Given an expr, generate RTL which we can insert at the end of a BB,
|
or on an edge. Set the block number of any insns generated to
|
or on an edge. Set the block number of any insns generated to
|
the value of BB. */
|
the value of BB. */
|
|
|
static rtx
|
static rtx
|
process_insert_insn (struct expr *expr)
|
process_insert_insn (struct expr *expr)
|
{
|
{
|
rtx reg = expr->reaching_reg;
|
rtx reg = expr->reaching_reg;
|
rtx exp = copy_rtx (expr->expr);
|
rtx exp = copy_rtx (expr->expr);
|
rtx pat;
|
rtx pat;
|
|
|
start_sequence ();
|
start_sequence ();
|
|
|
/* If the expression is something that's an operand, like a constant,
|
/* If the expression is something that's an operand, like a constant,
|
just copy it to a register. */
|
just copy it to a register. */
|
if (general_operand (exp, GET_MODE (reg)))
|
if (general_operand (exp, GET_MODE (reg)))
|
emit_move_insn (reg, exp);
|
emit_move_insn (reg, exp);
|
|
|
/* Otherwise, make a new insn to compute this expression and make sure the
|
/* Otherwise, make a new insn to compute this expression and make sure the
|
insn will be recognized (this also adds any needed CLOBBERs). Copy the
|
insn will be recognized (this also adds any needed CLOBBERs). Copy the
|
expression to make sure we don't have any sharing issues. */
|
expression to make sure we don't have any sharing issues. */
|
else
|
else
|
{
|
{
|
rtx insn = emit_insn (gen_rtx_SET (VOIDmode, reg, exp));
|
rtx insn = emit_insn (gen_rtx_SET (VOIDmode, reg, exp));
|
|
|
if (insn_invalid_p (insn))
|
if (insn_invalid_p (insn))
|
gcc_unreachable ();
|
gcc_unreachable ();
|
}
|
}
|
|
|
|
|
pat = get_insns ();
|
pat = get_insns ();
|
end_sequence ();
|
end_sequence ();
|
|
|
return pat;
|
return pat;
|
}
|
}
|
|
|
/* Add EXPR to the end of basic block BB.
|
/* Add EXPR to the end of basic block BB.
|
|
|
This is used by both the PRE and code hoisting.
|
This is used by both the PRE and code hoisting.
|
|
|
For PRE, we want to verify that the expr is either transparent
|
For PRE, we want to verify that the expr is either transparent
|
or locally anticipatable in the target block. This check makes
|
or locally anticipatable in the target block. This check makes
|
no sense for code hoisting. */
|
no sense for code hoisting. */
|
|
|
static void
|
static void
|
insert_insn_end_bb (struct expr *expr, basic_block bb, int pre)
|
insert_insn_end_bb (struct expr *expr, basic_block bb, int pre)
|
{
|
{
|
rtx insn = BB_END (bb);
|
rtx insn = BB_END (bb);
|
rtx new_insn;
|
rtx new_insn;
|
rtx reg = expr->reaching_reg;
|
rtx reg = expr->reaching_reg;
|
int regno = REGNO (reg);
|
int regno = REGNO (reg);
|
rtx pat, pat_end;
|
rtx pat, pat_end;
|
|
|
pat = process_insert_insn (expr);
|
pat = process_insert_insn (expr);
|
gcc_assert (pat && INSN_P (pat));
|
gcc_assert (pat && INSN_P (pat));
|
|
|
pat_end = pat;
|
pat_end = pat;
|
while (NEXT_INSN (pat_end) != NULL_RTX)
|
while (NEXT_INSN (pat_end) != NULL_RTX)
|
pat_end = NEXT_INSN (pat_end);
|
pat_end = NEXT_INSN (pat_end);
|
|
|
/* If the last insn is a jump, insert EXPR in front [taking care to
|
/* If the last insn is a jump, insert EXPR in front [taking care to
|
handle cc0, etc. properly]. Similarly we need to care trapping
|
handle cc0, etc. properly]. Similarly we need to care trapping
|
instructions in presence of non-call exceptions. */
|
instructions in presence of non-call exceptions. */
|
|
|
if (JUMP_P (insn)
|
if (JUMP_P (insn)
|
|| (NONJUMP_INSN_P (insn)
|
|| (NONJUMP_INSN_P (insn)
|
&& (!single_succ_p (bb)
|
&& (!single_succ_p (bb)
|
|| single_succ_edge (bb)->flags & EDGE_ABNORMAL)))
|
|| single_succ_edge (bb)->flags & EDGE_ABNORMAL)))
|
{
|
{
|
#ifdef HAVE_cc0
|
#ifdef HAVE_cc0
|
rtx note;
|
rtx note;
|
#endif
|
#endif
|
/* It should always be the case that we can put these instructions
|
/* It should always be the case that we can put these instructions
|
anywhere in the basic block with performing PRE optimizations.
|
anywhere in the basic block with performing PRE optimizations.
|
Check this. */
|
Check this. */
|
gcc_assert (!NONJUMP_INSN_P (insn) || !pre
|
gcc_assert (!NONJUMP_INSN_P (insn) || !pre
|
|| TEST_BIT (antloc[bb->index], expr->bitmap_index)
|
|| TEST_BIT (antloc[bb->index], expr->bitmap_index)
|
|| TEST_BIT (transp[bb->index], expr->bitmap_index));
|
|| TEST_BIT (transp[bb->index], expr->bitmap_index));
|
|
|
/* If this is a jump table, then we can't insert stuff here. Since
|
/* If this is a jump table, then we can't insert stuff here. Since
|
we know the previous real insn must be the tablejump, we insert
|
we know the previous real insn must be the tablejump, we insert
|
the new instruction just before the tablejump. */
|
the new instruction just before the tablejump. */
|
if (GET_CODE (PATTERN (insn)) == ADDR_VEC
|
if (GET_CODE (PATTERN (insn)) == ADDR_VEC
|
|| GET_CODE (PATTERN (insn)) == ADDR_DIFF_VEC)
|
|| GET_CODE (PATTERN (insn)) == ADDR_DIFF_VEC)
|
insn = prev_real_insn (insn);
|
insn = prev_real_insn (insn);
|
|
|
#ifdef HAVE_cc0
|
#ifdef HAVE_cc0
|
/* FIXME: 'twould be nice to call prev_cc0_setter here but it aborts
|
/* FIXME: 'twould be nice to call prev_cc0_setter here but it aborts
|
if cc0 isn't set. */
|
if cc0 isn't set. */
|
note = find_reg_note (insn, REG_CC_SETTER, NULL_RTX);
|
note = find_reg_note (insn, REG_CC_SETTER, NULL_RTX);
|
if (note)
|
if (note)
|
insn = XEXP (note, 0);
|
insn = XEXP (note, 0);
|
else
|
else
|
{
|
{
|
rtx maybe_cc0_setter = prev_nonnote_insn (insn);
|
rtx maybe_cc0_setter = prev_nonnote_insn (insn);
|
if (maybe_cc0_setter
|
if (maybe_cc0_setter
|
&& INSN_P (maybe_cc0_setter)
|
&& INSN_P (maybe_cc0_setter)
|
&& sets_cc0_p (PATTERN (maybe_cc0_setter)))
|
&& sets_cc0_p (PATTERN (maybe_cc0_setter)))
|
insn = maybe_cc0_setter;
|
insn = maybe_cc0_setter;
|
}
|
}
|
#endif
|
#endif
|
/* FIXME: What if something in cc0/jump uses value set in new insn? */
|
/* FIXME: What if something in cc0/jump uses value set in new insn? */
|
new_insn = emit_insn_before_noloc (pat, insn);
|
new_insn = emit_insn_before_noloc (pat, insn);
|
}
|
}
|
|
|
/* Likewise if the last insn is a call, as will happen in the presence
|
/* Likewise if the last insn is a call, as will happen in the presence
|
of exception handling. */
|
of exception handling. */
|
else if (CALL_P (insn)
|
else if (CALL_P (insn)
|
&& (!single_succ_p (bb)
|
&& (!single_succ_p (bb)
|
|| single_succ_edge (bb)->flags & EDGE_ABNORMAL))
|
|| single_succ_edge (bb)->flags & EDGE_ABNORMAL))
|
{
|
{
|
/* Keeping in mind SMALL_REGISTER_CLASSES and parameters in registers,
|
/* Keeping in mind SMALL_REGISTER_CLASSES and parameters in registers,
|
we search backward and place the instructions before the first
|
we search backward and place the instructions before the first
|
parameter is loaded. Do this for everyone for consistency and a
|
parameter is loaded. Do this for everyone for consistency and a
|
presumption that we'll get better code elsewhere as well.
|
presumption that we'll get better code elsewhere as well.
|
|
|
It should always be the case that we can put these instructions
|
It should always be the case that we can put these instructions
|
anywhere in the basic block with performing PRE optimizations.
|
anywhere in the basic block with performing PRE optimizations.
|
Check this. */
|
Check this. */
|
|
|
gcc_assert (!pre
|
gcc_assert (!pre
|
|| TEST_BIT (antloc[bb->index], expr->bitmap_index)
|
|| TEST_BIT (antloc[bb->index], expr->bitmap_index)
|
|| TEST_BIT (transp[bb->index], expr->bitmap_index));
|
|| TEST_BIT (transp[bb->index], expr->bitmap_index));
|
|
|
/* Since different machines initialize their parameter registers
|
/* Since different machines initialize their parameter registers
|
in different orders, assume nothing. Collect the set of all
|
in different orders, assume nothing. Collect the set of all
|
parameter registers. */
|
parameter registers. */
|
insn = find_first_parameter_load (insn, BB_HEAD (bb));
|
insn = find_first_parameter_load (insn, BB_HEAD (bb));
|
|
|
/* If we found all the parameter loads, then we want to insert
|
/* If we found all the parameter loads, then we want to insert
|
before the first parameter load.
|
before the first parameter load.
|
|
|
If we did not find all the parameter loads, then we might have
|
If we did not find all the parameter loads, then we might have
|
stopped on the head of the block, which could be a CODE_LABEL.
|
stopped on the head of the block, which could be a CODE_LABEL.
|
If we inserted before the CODE_LABEL, then we would be putting
|
If we inserted before the CODE_LABEL, then we would be putting
|
the insn in the wrong basic block. In that case, put the insn
|
the insn in the wrong basic block. In that case, put the insn
|
after the CODE_LABEL. Also, respect NOTE_INSN_BASIC_BLOCK. */
|
after the CODE_LABEL. Also, respect NOTE_INSN_BASIC_BLOCK. */
|
while (LABEL_P (insn)
|
while (LABEL_P (insn)
|
|| NOTE_INSN_BASIC_BLOCK_P (insn))
|
|| NOTE_INSN_BASIC_BLOCK_P (insn))
|
insn = NEXT_INSN (insn);
|
insn = NEXT_INSN (insn);
|
|
|
new_insn = emit_insn_before_noloc (pat, insn);
|
new_insn = emit_insn_before_noloc (pat, insn);
|
}
|
}
|
else
|
else
|
new_insn = emit_insn_after_noloc (pat, insn);
|
new_insn = emit_insn_after_noloc (pat, insn);
|
|
|
while (1)
|
while (1)
|
{
|
{
|
if (INSN_P (pat))
|
if (INSN_P (pat))
|
{
|
{
|
add_label_notes (PATTERN (pat), new_insn);
|
add_label_notes (PATTERN (pat), new_insn);
|
note_stores (PATTERN (pat), record_set_info, pat);
|
note_stores (PATTERN (pat), record_set_info, pat);
|
}
|
}
|
if (pat == pat_end)
|
if (pat == pat_end)
|
break;
|
break;
|
pat = NEXT_INSN (pat);
|
pat = NEXT_INSN (pat);
|
}
|
}
|
|
|
gcse_create_count++;
|
gcse_create_count++;
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "PRE/HOIST: end of bb %d, insn %d, ",
|
fprintf (dump_file, "PRE/HOIST: end of bb %d, insn %d, ",
|
bb->index, INSN_UID (new_insn));
|
bb->index, INSN_UID (new_insn));
|
fprintf (dump_file, "copying expression %d to reg %d\n",
|
fprintf (dump_file, "copying expression %d to reg %d\n",
|
expr->bitmap_index, regno);
|
expr->bitmap_index, regno);
|
}
|
}
|
}
|
}
|
|
|
/* Insert partially redundant expressions on edges in the CFG to make
|
/* Insert partially redundant expressions on edges in the CFG to make
|
the expressions fully redundant. */
|
the expressions fully redundant. */
|
|
|
static int
|
static int
|
pre_edge_insert (struct edge_list *edge_list, struct expr **index_map)
|
pre_edge_insert (struct edge_list *edge_list, struct expr **index_map)
|
{
|
{
|
int e, i, j, num_edges, set_size, did_insert = 0;
|
int e, i, j, num_edges, set_size, did_insert = 0;
|
sbitmap *inserted;
|
sbitmap *inserted;
|
|
|
/* Where PRE_INSERT_MAP is nonzero, we add the expression on that edge
|
/* Where PRE_INSERT_MAP is nonzero, we add the expression on that edge
|
if it reaches any of the deleted expressions. */
|
if it reaches any of the deleted expressions. */
|
|
|
set_size = pre_insert_map[0]->size;
|
set_size = pre_insert_map[0]->size;
|
num_edges = NUM_EDGES (edge_list);
|
num_edges = NUM_EDGES (edge_list);
|
inserted = sbitmap_vector_alloc (num_edges, expr_hash_table.n_elems);
|
inserted = sbitmap_vector_alloc (num_edges, expr_hash_table.n_elems);
|
sbitmap_vector_zero (inserted, num_edges);
|
sbitmap_vector_zero (inserted, num_edges);
|
|
|
for (e = 0; e < num_edges; e++)
|
for (e = 0; e < num_edges; e++)
|
{
|
{
|
int indx;
|
int indx;
|
basic_block bb = INDEX_EDGE_PRED_BB (edge_list, e);
|
basic_block bb = INDEX_EDGE_PRED_BB (edge_list, e);
|
|
|
for (i = indx = 0; i < set_size; i++, indx += SBITMAP_ELT_BITS)
|
for (i = indx = 0; i < set_size; i++, indx += SBITMAP_ELT_BITS)
|
{
|
{
|
SBITMAP_ELT_TYPE insert = pre_insert_map[e]->elms[i];
|
SBITMAP_ELT_TYPE insert = pre_insert_map[e]->elms[i];
|
|
|
for (j = indx; insert && j < (int) expr_hash_table.n_elems; j++, insert >>= 1)
|
for (j = indx; insert && j < (int) expr_hash_table.n_elems; j++, insert >>= 1)
|
if ((insert & 1) != 0 && index_map[j]->reaching_reg != NULL_RTX)
|
if ((insert & 1) != 0 && index_map[j]->reaching_reg != NULL_RTX)
|
{
|
{
|
struct expr *expr = index_map[j];
|
struct expr *expr = index_map[j];
|
struct occr *occr;
|
struct occr *occr;
|
|
|
/* Now look at each deleted occurrence of this expression. */
|
/* Now look at each deleted occurrence of this expression. */
|
for (occr = expr->antic_occr; occr != NULL; occr = occr->next)
|
for (occr = expr->antic_occr; occr != NULL; occr = occr->next)
|
{
|
{
|
if (! occr->deleted_p)
|
if (! occr->deleted_p)
|
continue;
|
continue;
|
|
|
/* Insert this expression on this edge if it would
|
/* Insert this expression on this edge if it would
|
reach the deleted occurrence in BB. */
|
reach the deleted occurrence in BB. */
|
if (!TEST_BIT (inserted[e], j))
|
if (!TEST_BIT (inserted[e], j))
|
{
|
{
|
rtx insn;
|
rtx insn;
|
edge eg = INDEX_EDGE (edge_list, e);
|
edge eg = INDEX_EDGE (edge_list, e);
|
|
|
/* We can't insert anything on an abnormal and
|
/* We can't insert anything on an abnormal and
|
critical edge, so we insert the insn at the end of
|
critical edge, so we insert the insn at the end of
|
the previous block. There are several alternatives
|
the previous block. There are several alternatives
|
detailed in Morgans book P277 (sec 10.5) for
|
detailed in Morgans book P277 (sec 10.5) for
|
handling this situation. This one is easiest for
|
handling this situation. This one is easiest for
|
now. */
|
now. */
|
|
|
if (eg->flags & EDGE_ABNORMAL)
|
if (eg->flags & EDGE_ABNORMAL)
|
insert_insn_end_bb (index_map[j], bb, 0);
|
insert_insn_end_bb (index_map[j], bb, 0);
|
else
|
else
|
{
|
{
|
insn = process_insert_insn (index_map[j]);
|
insn = process_insert_insn (index_map[j]);
|
insert_insn_on_edge (insn, eg);
|
insert_insn_on_edge (insn, eg);
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "PRE/HOIST: edge (%d,%d), ",
|
fprintf (dump_file, "PRE/HOIST: edge (%d,%d), ",
|
bb->index,
|
bb->index,
|
INDEX_EDGE_SUCC_BB (edge_list, e)->index);
|
INDEX_EDGE_SUCC_BB (edge_list, e)->index);
|
fprintf (dump_file, "copy expression %d\n",
|
fprintf (dump_file, "copy expression %d\n",
|
expr->bitmap_index);
|
expr->bitmap_index);
|
}
|
}
|
|
|
update_ld_motion_stores (expr);
|
update_ld_motion_stores (expr);
|
SET_BIT (inserted[e], j);
|
SET_BIT (inserted[e], j);
|
did_insert = 1;
|
did_insert = 1;
|
gcse_create_count++;
|
gcse_create_count++;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
sbitmap_vector_free (inserted);
|
sbitmap_vector_free (inserted);
|
return did_insert;
|
return did_insert;
|
}
|
}
|
|
|
/* Copy the result of EXPR->EXPR generated by INSN to EXPR->REACHING_REG.
|
/* Copy the result of EXPR->EXPR generated by INSN to EXPR->REACHING_REG.
|
Given "old_reg <- expr" (INSN), instead of adding after it
|
Given "old_reg <- expr" (INSN), instead of adding after it
|
reaching_reg <- old_reg
|
reaching_reg <- old_reg
|
it's better to do the following:
|
it's better to do the following:
|
reaching_reg <- expr
|
reaching_reg <- expr
|
old_reg <- reaching_reg
|
old_reg <- reaching_reg
|
because this way copy propagation can discover additional PRE
|
because this way copy propagation can discover additional PRE
|
opportunities. But if this fails, we try the old way.
|
opportunities. But if this fails, we try the old way.
|
When "expr" is a store, i.e.
|
When "expr" is a store, i.e.
|
given "MEM <- old_reg", instead of adding after it
|
given "MEM <- old_reg", instead of adding after it
|
reaching_reg <- old_reg
|
reaching_reg <- old_reg
|
it's better to add it before as follows:
|
it's better to add it before as follows:
|
reaching_reg <- old_reg
|
reaching_reg <- old_reg
|
MEM <- reaching_reg. */
|
MEM <- reaching_reg. */
|
|
|
static void
|
static void
|
pre_insert_copy_insn (struct expr *expr, rtx insn)
|
pre_insert_copy_insn (struct expr *expr, rtx insn)
|
{
|
{
|
rtx reg = expr->reaching_reg;
|
rtx reg = expr->reaching_reg;
|
int regno = REGNO (reg);
|
int regno = REGNO (reg);
|
int indx = expr->bitmap_index;
|
int indx = expr->bitmap_index;
|
rtx pat = PATTERN (insn);
|
rtx pat = PATTERN (insn);
|
rtx set, first_set, new_insn;
|
rtx set, first_set, new_insn;
|
rtx old_reg;
|
rtx old_reg;
|
int i;
|
int i;
|
|
|
/* This block matches the logic in hash_scan_insn. */
|
/* This block matches the logic in hash_scan_insn. */
|
switch (GET_CODE (pat))
|
switch (GET_CODE (pat))
|
{
|
{
|
case SET:
|
case SET:
|
set = pat;
|
set = pat;
|
break;
|
break;
|
|
|
case PARALLEL:
|
case PARALLEL:
|
/* Search through the parallel looking for the set whose
|
/* Search through the parallel looking for the set whose
|
source was the expression that we're interested in. */
|
source was the expression that we're interested in. */
|
first_set = NULL_RTX;
|
first_set = NULL_RTX;
|
set = NULL_RTX;
|
set = NULL_RTX;
|
for (i = 0; i < XVECLEN (pat, 0); i++)
|
for (i = 0; i < XVECLEN (pat, 0); i++)
|
{
|
{
|
rtx x = XVECEXP (pat, 0, i);
|
rtx x = XVECEXP (pat, 0, i);
|
if (GET_CODE (x) == SET)
|
if (GET_CODE (x) == SET)
|
{
|
{
|
/* If the source was a REG_EQUAL or REG_EQUIV note, we
|
/* If the source was a REG_EQUAL or REG_EQUIV note, we
|
may not find an equivalent expression, but in this
|
may not find an equivalent expression, but in this
|
case the PARALLEL will have a single set. */
|
case the PARALLEL will have a single set. */
|
if (first_set == NULL_RTX)
|
if (first_set == NULL_RTX)
|
first_set = x;
|
first_set = x;
|
if (expr_equiv_p (SET_SRC (x), expr->expr))
|
if (expr_equiv_p (SET_SRC (x), expr->expr))
|
{
|
{
|
set = x;
|
set = x;
|
break;
|
break;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
gcc_assert (first_set);
|
gcc_assert (first_set);
|
if (set == NULL_RTX)
|
if (set == NULL_RTX)
|
set = first_set;
|
set = first_set;
|
break;
|
break;
|
|
|
default:
|
default:
|
gcc_unreachable ();
|
gcc_unreachable ();
|
}
|
}
|
|
|
if (REG_P (SET_DEST (set)))
|
if (REG_P (SET_DEST (set)))
|
{
|
{
|
old_reg = SET_DEST (set);
|
old_reg = SET_DEST (set);
|
/* Check if we can modify the set destination in the original insn. */
|
/* Check if we can modify the set destination in the original insn. */
|
if (validate_change (insn, &SET_DEST (set), reg, 0))
|
if (validate_change (insn, &SET_DEST (set), reg, 0))
|
{
|
{
|
new_insn = gen_move_insn (old_reg, reg);
|
new_insn = gen_move_insn (old_reg, reg);
|
new_insn = emit_insn_after (new_insn, insn);
|
new_insn = emit_insn_after (new_insn, insn);
|
|
|
/* Keep register set table up to date. */
|
/* Keep register set table up to date. */
|
record_one_set (regno, insn);
|
record_one_set (regno, insn);
|
}
|
}
|
else
|
else
|
{
|
{
|
new_insn = gen_move_insn (reg, old_reg);
|
new_insn = gen_move_insn (reg, old_reg);
|
new_insn = emit_insn_after (new_insn, insn);
|
new_insn = emit_insn_after (new_insn, insn);
|
|
|
/* Keep register set table up to date. */
|
/* Keep register set table up to date. */
|
record_one_set (regno, new_insn);
|
record_one_set (regno, new_insn);
|
}
|
}
|
}
|
}
|
else /* This is possible only in case of a store to memory. */
|
else /* This is possible only in case of a store to memory. */
|
{
|
{
|
old_reg = SET_SRC (set);
|
old_reg = SET_SRC (set);
|
new_insn = gen_move_insn (reg, old_reg);
|
new_insn = gen_move_insn (reg, old_reg);
|
|
|
/* Check if we can modify the set source in the original insn. */
|
/* Check if we can modify the set source in the original insn. */
|
if (validate_change (insn, &SET_SRC (set), reg, 0))
|
if (validate_change (insn, &SET_SRC (set), reg, 0))
|
new_insn = emit_insn_before (new_insn, insn);
|
new_insn = emit_insn_before (new_insn, insn);
|
else
|
else
|
new_insn = emit_insn_after (new_insn, insn);
|
new_insn = emit_insn_after (new_insn, insn);
|
|
|
/* Keep register set table up to date. */
|
/* Keep register set table up to date. */
|
record_one_set (regno, new_insn);
|
record_one_set (regno, new_insn);
|
}
|
}
|
|
|
gcse_create_count++;
|
gcse_create_count++;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"PRE: bb %d, insn %d, copy expression %d in insn %d to reg %d\n",
|
"PRE: bb %d, insn %d, copy expression %d in insn %d to reg %d\n",
|
BLOCK_NUM (insn), INSN_UID (new_insn), indx,
|
BLOCK_NUM (insn), INSN_UID (new_insn), indx,
|
INSN_UID (insn), regno);
|
INSN_UID (insn), regno);
|
}
|
}
|
|
|
/* Copy available expressions that reach the redundant expression
|
/* Copy available expressions that reach the redundant expression
|
to `reaching_reg'. */
|
to `reaching_reg'. */
|
|
|
static void
|
static void
|
pre_insert_copies (void)
|
pre_insert_copies (void)
|
{
|
{
|
unsigned int i, added_copy;
|
unsigned int i, added_copy;
|
struct expr *expr;
|
struct expr *expr;
|
struct occr *occr;
|
struct occr *occr;
|
struct occr *avail;
|
struct occr *avail;
|
|
|
/* For each available expression in the table, copy the result to
|
/* For each available expression in the table, copy the result to
|
`reaching_reg' if the expression reaches a deleted one.
|
`reaching_reg' if the expression reaches a deleted one.
|
|
|
??? The current algorithm is rather brute force.
|
??? The current algorithm is rather brute force.
|
Need to do some profiling. */
|
Need to do some profiling. */
|
|
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (expr = expr_hash_table.table[i]; expr != NULL; expr = expr->next_same_hash)
|
for (expr = expr_hash_table.table[i]; expr != NULL; expr = expr->next_same_hash)
|
{
|
{
|
/* If the basic block isn't reachable, PPOUT will be TRUE. However,
|
/* If the basic block isn't reachable, PPOUT will be TRUE. However,
|
we don't want to insert a copy here because the expression may not
|
we don't want to insert a copy here because the expression may not
|
really be redundant. So only insert an insn if the expression was
|
really be redundant. So only insert an insn if the expression was
|
deleted. This test also avoids further processing if the
|
deleted. This test also avoids further processing if the
|
expression wasn't deleted anywhere. */
|
expression wasn't deleted anywhere. */
|
if (expr->reaching_reg == NULL)
|
if (expr->reaching_reg == NULL)
|
continue;
|
continue;
|
|
|
/* Set when we add a copy for that expression. */
|
/* Set when we add a copy for that expression. */
|
added_copy = 0;
|
added_copy = 0;
|
|
|
for (occr = expr->antic_occr; occr != NULL; occr = occr->next)
|
for (occr = expr->antic_occr; occr != NULL; occr = occr->next)
|
{
|
{
|
if (! occr->deleted_p)
|
if (! occr->deleted_p)
|
continue;
|
continue;
|
|
|
for (avail = expr->avail_occr; avail != NULL; avail = avail->next)
|
for (avail = expr->avail_occr; avail != NULL; avail = avail->next)
|
{
|
{
|
rtx insn = avail->insn;
|
rtx insn = avail->insn;
|
|
|
/* No need to handle this one if handled already. */
|
/* No need to handle this one if handled already. */
|
if (avail->copied_p)
|
if (avail->copied_p)
|
continue;
|
continue;
|
|
|
/* Don't handle this one if it's a redundant one. */
|
/* Don't handle this one if it's a redundant one. */
|
if (TEST_BIT (pre_redundant_insns, INSN_CUID (insn)))
|
if (TEST_BIT (pre_redundant_insns, INSN_CUID (insn)))
|
continue;
|
continue;
|
|
|
/* Or if the expression doesn't reach the deleted one. */
|
/* Or if the expression doesn't reach the deleted one. */
|
if (! pre_expr_reaches_here_p (BLOCK_FOR_INSN (avail->insn),
|
if (! pre_expr_reaches_here_p (BLOCK_FOR_INSN (avail->insn),
|
expr,
|
expr,
|
BLOCK_FOR_INSN (occr->insn)))
|
BLOCK_FOR_INSN (occr->insn)))
|
continue;
|
continue;
|
|
|
added_copy = 1;
|
added_copy = 1;
|
|
|
/* Copy the result of avail to reaching_reg. */
|
/* Copy the result of avail to reaching_reg. */
|
pre_insert_copy_insn (expr, insn);
|
pre_insert_copy_insn (expr, insn);
|
avail->copied_p = 1;
|
avail->copied_p = 1;
|
}
|
}
|
}
|
}
|
|
|
if (added_copy)
|
if (added_copy)
|
update_ld_motion_stores (expr);
|
update_ld_motion_stores (expr);
|
}
|
}
|
}
|
}
|
|
|
/* Emit move from SRC to DEST noting the equivalence with expression computed
|
/* Emit move from SRC to DEST noting the equivalence with expression computed
|
in INSN. */
|
in INSN. */
|
static rtx
|
static rtx
|
gcse_emit_move_after (rtx src, rtx dest, rtx insn)
|
gcse_emit_move_after (rtx src, rtx dest, rtx insn)
|
{
|
{
|
rtx new;
|
rtx new;
|
rtx set = single_set (insn), set2;
|
rtx set = single_set (insn), set2;
|
rtx note;
|
rtx note;
|
rtx eqv;
|
rtx eqv;
|
|
|
/* This should never fail since we're creating a reg->reg copy
|
/* This should never fail since we're creating a reg->reg copy
|
we've verified to be valid. */
|
we've verified to be valid. */
|
|
|
new = emit_insn_after (gen_move_insn (dest, src), insn);
|
new = emit_insn_after (gen_move_insn (dest, src), insn);
|
|
|
/* Note the equivalence for local CSE pass. */
|
/* Note the equivalence for local CSE pass. */
|
set2 = single_set (new);
|
set2 = single_set (new);
|
if (!set2 || !rtx_equal_p (SET_DEST (set2), dest))
|
if (!set2 || !rtx_equal_p (SET_DEST (set2), dest))
|
return new;
|
return new;
|
if ((note = find_reg_equal_equiv_note (insn)))
|
if ((note = find_reg_equal_equiv_note (insn)))
|
eqv = XEXP (note, 0);
|
eqv = XEXP (note, 0);
|
else
|
else
|
eqv = SET_SRC (set);
|
eqv = SET_SRC (set);
|
|
|
set_unique_reg_note (new, REG_EQUAL, copy_insn_1 (eqv));
|
set_unique_reg_note (new, REG_EQUAL, copy_insn_1 (eqv));
|
|
|
return new;
|
return new;
|
}
|
}
|
|
|
/* Delete redundant computations.
|
/* Delete redundant computations.
|
Deletion is done by changing the insn to copy the `reaching_reg' of
|
Deletion is done by changing the insn to copy the `reaching_reg' of
|
the expression into the result of the SET. It is left to later passes
|
the expression into the result of the SET. It is left to later passes
|
(cprop, cse2, flow, combine, regmove) to propagate the copy or eliminate it.
|
(cprop, cse2, flow, combine, regmove) to propagate the copy or eliminate it.
|
|
|
Returns nonzero if a change is made. */
|
Returns nonzero if a change is made. */
|
|
|
static int
|
static int
|
pre_delete (void)
|
pre_delete (void)
|
{
|
{
|
unsigned int i;
|
unsigned int i;
|
int changed;
|
int changed;
|
struct expr *expr;
|
struct expr *expr;
|
struct occr *occr;
|
struct occr *occr;
|
|
|
changed = 0;
|
changed = 0;
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (expr = expr_hash_table.table[i];
|
for (expr = expr_hash_table.table[i];
|
expr != NULL;
|
expr != NULL;
|
expr = expr->next_same_hash)
|
expr = expr->next_same_hash)
|
{
|
{
|
int indx = expr->bitmap_index;
|
int indx = expr->bitmap_index;
|
|
|
/* We only need to search antic_occr since we require
|
/* We only need to search antic_occr since we require
|
ANTLOC != 0. */
|
ANTLOC != 0. */
|
|
|
for (occr = expr->antic_occr; occr != NULL; occr = occr->next)
|
for (occr = expr->antic_occr; occr != NULL; occr = occr->next)
|
{
|
{
|
rtx insn = occr->insn;
|
rtx insn = occr->insn;
|
rtx set;
|
rtx set;
|
basic_block bb = BLOCK_FOR_INSN (insn);
|
basic_block bb = BLOCK_FOR_INSN (insn);
|
|
|
/* We only delete insns that have a single_set. */
|
/* We only delete insns that have a single_set. */
|
if (TEST_BIT (pre_delete_map[bb->index], indx)
|
if (TEST_BIT (pre_delete_map[bb->index], indx)
|
&& (set = single_set (insn)) != 0)
|
&& (set = single_set (insn)) != 0)
|
{
|
{
|
/* Create a pseudo-reg to store the result of reaching
|
/* Create a pseudo-reg to store the result of reaching
|
expressions into. Get the mode for the new pseudo from
|
expressions into. Get the mode for the new pseudo from
|
the mode of the original destination pseudo. */
|
the mode of the original destination pseudo. */
|
if (expr->reaching_reg == NULL)
|
if (expr->reaching_reg == NULL)
|
expr->reaching_reg
|
expr->reaching_reg
|
= gen_reg_rtx (GET_MODE (SET_DEST (set)));
|
= gen_reg_rtx (GET_MODE (SET_DEST (set)));
|
|
|
gcse_emit_move_after (expr->reaching_reg, SET_DEST (set), insn);
|
gcse_emit_move_after (expr->reaching_reg, SET_DEST (set), insn);
|
delete_insn (insn);
|
delete_insn (insn);
|
occr->deleted_p = 1;
|
occr->deleted_p = 1;
|
SET_BIT (pre_redundant_insns, INSN_CUID (insn));
|
SET_BIT (pre_redundant_insns, INSN_CUID (insn));
|
changed = 1;
|
changed = 1;
|
gcse_subst_count++;
|
gcse_subst_count++;
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"PRE: redundant insn %d (expression %d) in ",
|
"PRE: redundant insn %d (expression %d) in ",
|
INSN_UID (insn), indx);
|
INSN_UID (insn), indx);
|
fprintf (dump_file, "bb %d, reaching reg is %d\n",
|
fprintf (dump_file, "bb %d, reaching reg is %d\n",
|
bb->index, REGNO (expr->reaching_reg));
|
bb->index, REGNO (expr->reaching_reg));
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* Perform GCSE optimizations using PRE.
|
/* Perform GCSE optimizations using PRE.
|
This is called by one_pre_gcse_pass after all the dataflow analysis
|
This is called by one_pre_gcse_pass after all the dataflow analysis
|
has been done.
|
has been done.
|
|
|
This is based on the original Morel-Renvoise paper Fred Chow's thesis, and
|
This is based on the original Morel-Renvoise paper Fred Chow's thesis, and
|
lazy code motion from Knoop, Ruthing and Steffen as described in Advanced
|
lazy code motion from Knoop, Ruthing and Steffen as described in Advanced
|
Compiler Design and Implementation.
|
Compiler Design and Implementation.
|
|
|
??? A new pseudo reg is created to hold the reaching expression. The nice
|
??? A new pseudo reg is created to hold the reaching expression. The nice
|
thing about the classical approach is that it would try to use an existing
|
thing about the classical approach is that it would try to use an existing
|
reg. If the register can't be adequately optimized [i.e. we introduce
|
reg. If the register can't be adequately optimized [i.e. we introduce
|
reload problems], one could add a pass here to propagate the new register
|
reload problems], one could add a pass here to propagate the new register
|
through the block.
|
through the block.
|
|
|
??? We don't handle single sets in PARALLELs because we're [currently] not
|
??? We don't handle single sets in PARALLELs because we're [currently] not
|
able to copy the rest of the parallel when we insert copies to create full
|
able to copy the rest of the parallel when we insert copies to create full
|
redundancies from partial redundancies. However, there's no reason why we
|
redundancies from partial redundancies. However, there's no reason why we
|
can't handle PARALLELs in the cases where there are no partial
|
can't handle PARALLELs in the cases where there are no partial
|
redundancies. */
|
redundancies. */
|
|
|
static int
|
static int
|
pre_gcse (void)
|
pre_gcse (void)
|
{
|
{
|
unsigned int i;
|
unsigned int i;
|
int did_insert, changed;
|
int did_insert, changed;
|
struct expr **index_map;
|
struct expr **index_map;
|
struct expr *expr;
|
struct expr *expr;
|
|
|
/* Compute a mapping from expression number (`bitmap_index') to
|
/* Compute a mapping from expression number (`bitmap_index') to
|
hash table entry. */
|
hash table entry. */
|
|
|
index_map = XCNEWVEC (struct expr *, expr_hash_table.n_elems);
|
index_map = XCNEWVEC (struct expr *, expr_hash_table.n_elems);
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (expr = expr_hash_table.table[i]; expr != NULL; expr = expr->next_same_hash)
|
for (expr = expr_hash_table.table[i]; expr != NULL; expr = expr->next_same_hash)
|
index_map[expr->bitmap_index] = expr;
|
index_map[expr->bitmap_index] = expr;
|
|
|
/* Reset bitmap used to track which insns are redundant. */
|
/* Reset bitmap used to track which insns are redundant. */
|
pre_redundant_insns = sbitmap_alloc (max_cuid);
|
pre_redundant_insns = sbitmap_alloc (max_cuid);
|
sbitmap_zero (pre_redundant_insns);
|
sbitmap_zero (pre_redundant_insns);
|
|
|
/* Delete the redundant insns first so that
|
/* Delete the redundant insns first so that
|
- we know what register to use for the new insns and for the other
|
- we know what register to use for the new insns and for the other
|
ones with reaching expressions
|
ones with reaching expressions
|
- we know which insns are redundant when we go to create copies */
|
- we know which insns are redundant when we go to create copies */
|
|
|
changed = pre_delete ();
|
changed = pre_delete ();
|
|
|
did_insert = pre_edge_insert (edge_list, index_map);
|
did_insert = pre_edge_insert (edge_list, index_map);
|
|
|
/* In other places with reaching expressions, copy the expression to the
|
/* In other places with reaching expressions, copy the expression to the
|
specially allocated pseudo-reg that reaches the redundant expr. */
|
specially allocated pseudo-reg that reaches the redundant expr. */
|
pre_insert_copies ();
|
pre_insert_copies ();
|
if (did_insert)
|
if (did_insert)
|
{
|
{
|
commit_edge_insertions ();
|
commit_edge_insertions ();
|
changed = 1;
|
changed = 1;
|
}
|
}
|
|
|
free (index_map);
|
free (index_map);
|
sbitmap_free (pre_redundant_insns);
|
sbitmap_free (pre_redundant_insns);
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* Top level routine to perform one PRE GCSE pass.
|
/* Top level routine to perform one PRE GCSE pass.
|
|
|
Return nonzero if a change was made. */
|
Return nonzero if a change was made. */
|
|
|
static int
|
static int
|
one_pre_gcse_pass (int pass)
|
one_pre_gcse_pass (int pass)
|
{
|
{
|
int changed = 0;
|
int changed = 0;
|
|
|
gcse_subst_count = 0;
|
gcse_subst_count = 0;
|
gcse_create_count = 0;
|
gcse_create_count = 0;
|
|
|
alloc_hash_table (max_cuid, &expr_hash_table, 0);
|
alloc_hash_table (max_cuid, &expr_hash_table, 0);
|
add_noreturn_fake_exit_edges ();
|
add_noreturn_fake_exit_edges ();
|
if (flag_gcse_lm)
|
if (flag_gcse_lm)
|
compute_ld_motion_mems ();
|
compute_ld_motion_mems ();
|
|
|
compute_hash_table (&expr_hash_table);
|
compute_hash_table (&expr_hash_table);
|
trim_ld_motion_mems ();
|
trim_ld_motion_mems ();
|
if (dump_file)
|
if (dump_file)
|
dump_hash_table (dump_file, "Expression", &expr_hash_table);
|
dump_hash_table (dump_file, "Expression", &expr_hash_table);
|
|
|
if (expr_hash_table.n_elems > 0)
|
if (expr_hash_table.n_elems > 0)
|
{
|
{
|
alloc_pre_mem (last_basic_block, expr_hash_table.n_elems);
|
alloc_pre_mem (last_basic_block, expr_hash_table.n_elems);
|
compute_pre_data ();
|
compute_pre_data ();
|
changed |= pre_gcse ();
|
changed |= pre_gcse ();
|
free_edge_list (edge_list);
|
free_edge_list (edge_list);
|
free_pre_mem ();
|
free_pre_mem ();
|
}
|
}
|
|
|
free_ldst_mems ();
|
free_ldst_mems ();
|
remove_fake_exit_edges ();
|
remove_fake_exit_edges ();
|
free_hash_table (&expr_hash_table);
|
free_hash_table (&expr_hash_table);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "\nPRE GCSE of %s, pass %d: %d bytes needed, ",
|
fprintf (dump_file, "\nPRE GCSE of %s, pass %d: %d bytes needed, ",
|
current_function_name (), pass, bytes_used);
|
current_function_name (), pass, bytes_used);
|
fprintf (dump_file, "%d substs, %d insns created\n",
|
fprintf (dump_file, "%d substs, %d insns created\n",
|
gcse_subst_count, gcse_create_count);
|
gcse_subst_count, gcse_create_count);
|
}
|
}
|
|
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* If X contains any LABEL_REF's, add REG_LABEL notes for them to INSN.
|
/* If X contains any LABEL_REF's, add REG_LABEL notes for them to INSN.
|
If notes are added to an insn which references a CODE_LABEL, the
|
If notes are added to an insn which references a CODE_LABEL, the
|
LABEL_NUSES count is incremented. We have to add REG_LABEL notes,
|
LABEL_NUSES count is incremented. We have to add REG_LABEL notes,
|
because the following loop optimization pass requires them. */
|
because the following loop optimization pass requires them. */
|
|
|
/* ??? If there was a jump optimization pass after gcse and before loop,
|
/* ??? If there was a jump optimization pass after gcse and before loop,
|
then we would not need to do this here, because jump would add the
|
then we would not need to do this here, because jump would add the
|
necessary REG_LABEL notes. */
|
necessary REG_LABEL notes. */
|
|
|
static void
|
static void
|
add_label_notes (rtx x, rtx insn)
|
add_label_notes (rtx x, rtx insn)
|
{
|
{
|
enum rtx_code code = GET_CODE (x);
|
enum rtx_code code = GET_CODE (x);
|
int i, j;
|
int i, j;
|
const char *fmt;
|
const char *fmt;
|
|
|
if (code == LABEL_REF && !LABEL_REF_NONLOCAL_P (x))
|
if (code == LABEL_REF && !LABEL_REF_NONLOCAL_P (x))
|
{
|
{
|
/* This code used to ignore labels that referred to dispatch tables to
|
/* This code used to ignore labels that referred to dispatch tables to
|
avoid flow generating (slightly) worse code.
|
avoid flow generating (slightly) worse code.
|
|
|
We no longer ignore such label references (see LABEL_REF handling in
|
We no longer ignore such label references (see LABEL_REF handling in
|
mark_jump_label for additional information). */
|
mark_jump_label for additional information). */
|
|
|
REG_NOTES (insn) = gen_rtx_INSN_LIST (REG_LABEL, XEXP (x, 0),
|
REG_NOTES (insn) = gen_rtx_INSN_LIST (REG_LABEL, XEXP (x, 0),
|
REG_NOTES (insn));
|
REG_NOTES (insn));
|
if (LABEL_P (XEXP (x, 0)))
|
if (LABEL_P (XEXP (x, 0)))
|
LABEL_NUSES (XEXP (x, 0))++;
|
LABEL_NUSES (XEXP (x, 0))++;
|
return;
|
return;
|
}
|
}
|
|
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
for (i = GET_RTX_LENGTH (code) - 1, fmt = GET_RTX_FORMAT (code); i >= 0; i--)
|
{
|
{
|
if (fmt[i] == 'e')
|
if (fmt[i] == 'e')
|
add_label_notes (XEXP (x, i), insn);
|
add_label_notes (XEXP (x, i), insn);
|
else if (fmt[i] == 'E')
|
else if (fmt[i] == 'E')
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
add_label_notes (XVECEXP (x, i, j), insn);
|
add_label_notes (XVECEXP (x, i, j), insn);
|
}
|
}
|
}
|
}
|
|
|
/* Compute transparent outgoing information for each block.
|
/* Compute transparent outgoing information for each block.
|
|
|
An expression is transparent to an edge unless it is killed by
|
An expression is transparent to an edge unless it is killed by
|
the edge itself. This can only happen with abnormal control flow,
|
the edge itself. This can only happen with abnormal control flow,
|
when the edge is traversed through a call. This happens with
|
when the edge is traversed through a call. This happens with
|
non-local labels and exceptions.
|
non-local labels and exceptions.
|
|
|
This would not be necessary if we split the edge. While this is
|
This would not be necessary if we split the edge. While this is
|
normally impossible for abnormal critical edges, with some effort
|
normally impossible for abnormal critical edges, with some effort
|
it should be possible with exception handling, since we still have
|
it should be possible with exception handling, since we still have
|
control over which handler should be invoked. But due to increased
|
control over which handler should be invoked. But due to increased
|
EH table sizes, this may not be worthwhile. */
|
EH table sizes, this may not be worthwhile. */
|
|
|
static void
|
static void
|
compute_transpout (void)
|
compute_transpout (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
unsigned int i;
|
unsigned int i;
|
struct expr *expr;
|
struct expr *expr;
|
|
|
sbitmap_vector_ones (transpout, last_basic_block);
|
sbitmap_vector_ones (transpout, last_basic_block);
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
/* Note that flow inserted a nop a the end of basic blocks that
|
/* Note that flow inserted a nop a the end of basic blocks that
|
end in call instructions for reasons other than abnormal
|
end in call instructions for reasons other than abnormal
|
control flow. */
|
control flow. */
|
if (! CALL_P (BB_END (bb)))
|
if (! CALL_P (BB_END (bb)))
|
continue;
|
continue;
|
|
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (expr = expr_hash_table.table[i]; expr ; expr = expr->next_same_hash)
|
for (expr = expr_hash_table.table[i]; expr ; expr = expr->next_same_hash)
|
if (MEM_P (expr->expr))
|
if (MEM_P (expr->expr))
|
{
|
{
|
if (GET_CODE (XEXP (expr->expr, 0)) == SYMBOL_REF
|
if (GET_CODE (XEXP (expr->expr, 0)) == SYMBOL_REF
|
&& CONSTANT_POOL_ADDRESS_P (XEXP (expr->expr, 0)))
|
&& CONSTANT_POOL_ADDRESS_P (XEXP (expr->expr, 0)))
|
continue;
|
continue;
|
|
|
/* ??? Optimally, we would use interprocedural alias
|
/* ??? Optimally, we would use interprocedural alias
|
analysis to determine if this mem is actually killed
|
analysis to determine if this mem is actually killed
|
by this call. */
|
by this call. */
|
RESET_BIT (transpout[bb->index], expr->bitmap_index);
|
RESET_BIT (transpout[bb->index], expr->bitmap_index);
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Code Hoisting variables and subroutines. */
|
/* Code Hoisting variables and subroutines. */
|
|
|
/* Very busy expressions. */
|
/* Very busy expressions. */
|
static sbitmap *hoist_vbein;
|
static sbitmap *hoist_vbein;
|
static sbitmap *hoist_vbeout;
|
static sbitmap *hoist_vbeout;
|
|
|
/* Hoistable expressions. */
|
/* Hoistable expressions. */
|
static sbitmap *hoist_exprs;
|
static sbitmap *hoist_exprs;
|
|
|
/* ??? We could compute post dominators and run this algorithm in
|
/* ??? We could compute post dominators and run this algorithm in
|
reverse to perform tail merging, doing so would probably be
|
reverse to perform tail merging, doing so would probably be
|
more effective than the tail merging code in jump.c.
|
more effective than the tail merging code in jump.c.
|
|
|
It's unclear if tail merging could be run in parallel with
|
It's unclear if tail merging could be run in parallel with
|
code hoisting. It would be nice. */
|
code hoisting. It would be nice. */
|
|
|
/* Allocate vars used for code hoisting analysis. */
|
/* Allocate vars used for code hoisting analysis. */
|
|
|
static void
|
static void
|
alloc_code_hoist_mem (int n_blocks, int n_exprs)
|
alloc_code_hoist_mem (int n_blocks, int n_exprs)
|
{
|
{
|
antloc = sbitmap_vector_alloc (n_blocks, n_exprs);
|
antloc = sbitmap_vector_alloc (n_blocks, n_exprs);
|
transp = sbitmap_vector_alloc (n_blocks, n_exprs);
|
transp = sbitmap_vector_alloc (n_blocks, n_exprs);
|
comp = sbitmap_vector_alloc (n_blocks, n_exprs);
|
comp = sbitmap_vector_alloc (n_blocks, n_exprs);
|
|
|
hoist_vbein = sbitmap_vector_alloc (n_blocks, n_exprs);
|
hoist_vbein = sbitmap_vector_alloc (n_blocks, n_exprs);
|
hoist_vbeout = sbitmap_vector_alloc (n_blocks, n_exprs);
|
hoist_vbeout = sbitmap_vector_alloc (n_blocks, n_exprs);
|
hoist_exprs = sbitmap_vector_alloc (n_blocks, n_exprs);
|
hoist_exprs = sbitmap_vector_alloc (n_blocks, n_exprs);
|
transpout = sbitmap_vector_alloc (n_blocks, n_exprs);
|
transpout = sbitmap_vector_alloc (n_blocks, n_exprs);
|
}
|
}
|
|
|
/* Free vars used for code hoisting analysis. */
|
/* Free vars used for code hoisting analysis. */
|
|
|
static void
|
static void
|
free_code_hoist_mem (void)
|
free_code_hoist_mem (void)
|
{
|
{
|
sbitmap_vector_free (antloc);
|
sbitmap_vector_free (antloc);
|
sbitmap_vector_free (transp);
|
sbitmap_vector_free (transp);
|
sbitmap_vector_free (comp);
|
sbitmap_vector_free (comp);
|
|
|
sbitmap_vector_free (hoist_vbein);
|
sbitmap_vector_free (hoist_vbein);
|
sbitmap_vector_free (hoist_vbeout);
|
sbitmap_vector_free (hoist_vbeout);
|
sbitmap_vector_free (hoist_exprs);
|
sbitmap_vector_free (hoist_exprs);
|
sbitmap_vector_free (transpout);
|
sbitmap_vector_free (transpout);
|
|
|
free_dominance_info (CDI_DOMINATORS);
|
free_dominance_info (CDI_DOMINATORS);
|
}
|
}
|
|
|
/* Compute the very busy expressions at entry/exit from each block.
|
/* Compute the very busy expressions at entry/exit from each block.
|
|
|
An expression is very busy if all paths from a given point
|
An expression is very busy if all paths from a given point
|
compute the expression. */
|
compute the expression. */
|
|
|
static void
|
static void
|
compute_code_hoist_vbeinout (void)
|
compute_code_hoist_vbeinout (void)
|
{
|
{
|
int changed, passes;
|
int changed, passes;
|
basic_block bb;
|
basic_block bb;
|
|
|
sbitmap_vector_zero (hoist_vbeout, last_basic_block);
|
sbitmap_vector_zero (hoist_vbeout, last_basic_block);
|
sbitmap_vector_zero (hoist_vbein, last_basic_block);
|
sbitmap_vector_zero (hoist_vbein, last_basic_block);
|
|
|
passes = 0;
|
passes = 0;
|
changed = 1;
|
changed = 1;
|
|
|
while (changed)
|
while (changed)
|
{
|
{
|
changed = 0;
|
changed = 0;
|
|
|
/* We scan the blocks in the reverse order to speed up
|
/* We scan the blocks in the reverse order to speed up
|
the convergence. */
|
the convergence. */
|
FOR_EACH_BB_REVERSE (bb)
|
FOR_EACH_BB_REVERSE (bb)
|
{
|
{
|
changed |= sbitmap_a_or_b_and_c_cg (hoist_vbein[bb->index], antloc[bb->index],
|
changed |= sbitmap_a_or_b_and_c_cg (hoist_vbein[bb->index], antloc[bb->index],
|
hoist_vbeout[bb->index], transp[bb->index]);
|
hoist_vbeout[bb->index], transp[bb->index]);
|
if (bb->next_bb != EXIT_BLOCK_PTR)
|
if (bb->next_bb != EXIT_BLOCK_PTR)
|
sbitmap_intersection_of_succs (hoist_vbeout[bb->index], hoist_vbein, bb->index);
|
sbitmap_intersection_of_succs (hoist_vbeout[bb->index], hoist_vbein, bb->index);
|
}
|
}
|
|
|
passes++;
|
passes++;
|
}
|
}
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "hoisting vbeinout computation: %d passes\n", passes);
|
fprintf (dump_file, "hoisting vbeinout computation: %d passes\n", passes);
|
}
|
}
|
|
|
/* Top level routine to do the dataflow analysis needed by code hoisting. */
|
/* Top level routine to do the dataflow analysis needed by code hoisting. */
|
|
|
static void
|
static void
|
compute_code_hoist_data (void)
|
compute_code_hoist_data (void)
|
{
|
{
|
compute_local_properties (transp, comp, antloc, &expr_hash_table);
|
compute_local_properties (transp, comp, antloc, &expr_hash_table);
|
compute_transpout ();
|
compute_transpout ();
|
compute_code_hoist_vbeinout ();
|
compute_code_hoist_vbeinout ();
|
calculate_dominance_info (CDI_DOMINATORS);
|
calculate_dominance_info (CDI_DOMINATORS);
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
/* Determine if the expression identified by EXPR_INDEX would
|
/* Determine if the expression identified by EXPR_INDEX would
|
reach BB unimpared if it was placed at the end of EXPR_BB.
|
reach BB unimpared if it was placed at the end of EXPR_BB.
|
|
|
It's unclear exactly what Muchnick meant by "unimpared". It seems
|
It's unclear exactly what Muchnick meant by "unimpared". It seems
|
to me that the expression must either be computed or transparent in
|
to me that the expression must either be computed or transparent in
|
*every* block in the path(s) from EXPR_BB to BB. Any other definition
|
*every* block in the path(s) from EXPR_BB to BB. Any other definition
|
would allow the expression to be hoisted out of loops, even if
|
would allow the expression to be hoisted out of loops, even if
|
the expression wasn't a loop invariant.
|
the expression wasn't a loop invariant.
|
|
|
Contrast this to reachability for PRE where an expression is
|
Contrast this to reachability for PRE where an expression is
|
considered reachable if *any* path reaches instead of *all*
|
considered reachable if *any* path reaches instead of *all*
|
paths. */
|
paths. */
|
|
|
static int
|
static int
|
hoist_expr_reaches_here_p (basic_block expr_bb, int expr_index, basic_block bb, char *visited)
|
hoist_expr_reaches_here_p (basic_block expr_bb, int expr_index, basic_block bb, char *visited)
|
{
|
{
|
edge pred;
|
edge pred;
|
edge_iterator ei;
|
edge_iterator ei;
|
int visited_allocated_locally = 0;
|
int visited_allocated_locally = 0;
|
|
|
|
|
if (visited == NULL)
|
if (visited == NULL)
|
{
|
{
|
visited_allocated_locally = 1;
|
visited_allocated_locally = 1;
|
visited = XCNEWVEC (char, last_basic_block);
|
visited = XCNEWVEC (char, last_basic_block);
|
}
|
}
|
|
|
FOR_EACH_EDGE (pred, ei, bb->preds)
|
FOR_EACH_EDGE (pred, ei, bb->preds)
|
{
|
{
|
basic_block pred_bb = pred->src;
|
basic_block pred_bb = pred->src;
|
|
|
if (pred->src == ENTRY_BLOCK_PTR)
|
if (pred->src == ENTRY_BLOCK_PTR)
|
break;
|
break;
|
else if (pred_bb == expr_bb)
|
else if (pred_bb == expr_bb)
|
continue;
|
continue;
|
else if (visited[pred_bb->index])
|
else if (visited[pred_bb->index])
|
continue;
|
continue;
|
|
|
/* Does this predecessor generate this expression? */
|
/* Does this predecessor generate this expression? */
|
else if (TEST_BIT (comp[pred_bb->index], expr_index))
|
else if (TEST_BIT (comp[pred_bb->index], expr_index))
|
break;
|
break;
|
else if (! TEST_BIT (transp[pred_bb->index], expr_index))
|
else if (! TEST_BIT (transp[pred_bb->index], expr_index))
|
break;
|
break;
|
|
|
/* Not killed. */
|
/* Not killed. */
|
else
|
else
|
{
|
{
|
visited[pred_bb->index] = 1;
|
visited[pred_bb->index] = 1;
|
if (! hoist_expr_reaches_here_p (expr_bb, expr_index,
|
if (! hoist_expr_reaches_here_p (expr_bb, expr_index,
|
pred_bb, visited))
|
pred_bb, visited))
|
break;
|
break;
|
}
|
}
|
}
|
}
|
if (visited_allocated_locally)
|
if (visited_allocated_locally)
|
free (visited);
|
free (visited);
|
|
|
return (pred == NULL);
|
return (pred == NULL);
|
}
|
}
|
|
|
/* Actually perform code hoisting. */
|
/* Actually perform code hoisting. */
|
|
|
static void
|
static void
|
hoist_code (void)
|
hoist_code (void)
|
{
|
{
|
basic_block bb, dominated;
|
basic_block bb, dominated;
|
basic_block *domby;
|
basic_block *domby;
|
unsigned int domby_len;
|
unsigned int domby_len;
|
unsigned int i,j;
|
unsigned int i,j;
|
struct expr **index_map;
|
struct expr **index_map;
|
struct expr *expr;
|
struct expr *expr;
|
|
|
sbitmap_vector_zero (hoist_exprs, last_basic_block);
|
sbitmap_vector_zero (hoist_exprs, last_basic_block);
|
|
|
/* Compute a mapping from expression number (`bitmap_index') to
|
/* Compute a mapping from expression number (`bitmap_index') to
|
hash table entry. */
|
hash table entry. */
|
|
|
index_map = XCNEWVEC (struct expr *, expr_hash_table.n_elems);
|
index_map = XCNEWVEC (struct expr *, expr_hash_table.n_elems);
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (i = 0; i < expr_hash_table.size; i++)
|
for (expr = expr_hash_table.table[i]; expr != NULL; expr = expr->next_same_hash)
|
for (expr = expr_hash_table.table[i]; expr != NULL; expr = expr->next_same_hash)
|
index_map[expr->bitmap_index] = expr;
|
index_map[expr->bitmap_index] = expr;
|
|
|
/* Walk over each basic block looking for potentially hoistable
|
/* Walk over each basic block looking for potentially hoistable
|
expressions, nothing gets hoisted from the entry block. */
|
expressions, nothing gets hoisted from the entry block. */
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
int found = 0;
|
int found = 0;
|
int insn_inserted_p;
|
int insn_inserted_p;
|
|
|
domby_len = get_dominated_by (CDI_DOMINATORS, bb, &domby);
|
domby_len = get_dominated_by (CDI_DOMINATORS, bb, &domby);
|
/* Examine each expression that is very busy at the exit of this
|
/* Examine each expression that is very busy at the exit of this
|
block. These are the potentially hoistable expressions. */
|
block. These are the potentially hoistable expressions. */
|
for (i = 0; i < hoist_vbeout[bb->index]->n_bits; i++)
|
for (i = 0; i < hoist_vbeout[bb->index]->n_bits; i++)
|
{
|
{
|
int hoistable = 0;
|
int hoistable = 0;
|
|
|
if (TEST_BIT (hoist_vbeout[bb->index], i)
|
if (TEST_BIT (hoist_vbeout[bb->index], i)
|
&& TEST_BIT (transpout[bb->index], i))
|
&& TEST_BIT (transpout[bb->index], i))
|
{
|
{
|
/* We've found a potentially hoistable expression, now
|
/* We've found a potentially hoistable expression, now
|
we look at every block BB dominates to see if it
|
we look at every block BB dominates to see if it
|
computes the expression. */
|
computes the expression. */
|
for (j = 0; j < domby_len; j++)
|
for (j = 0; j < domby_len; j++)
|
{
|
{
|
dominated = domby[j];
|
dominated = domby[j];
|
/* Ignore self dominance. */
|
/* Ignore self dominance. */
|
if (bb == dominated)
|
if (bb == dominated)
|
continue;
|
continue;
|
/* We've found a dominated block, now see if it computes
|
/* We've found a dominated block, now see if it computes
|
the busy expression and whether or not moving that
|
the busy expression and whether or not moving that
|
expression to the "beginning" of that block is safe. */
|
expression to the "beginning" of that block is safe. */
|
if (!TEST_BIT (antloc[dominated->index], i))
|
if (!TEST_BIT (antloc[dominated->index], i))
|
continue;
|
continue;
|
|
|
/* Note if the expression would reach the dominated block
|
/* Note if the expression would reach the dominated block
|
unimpared if it was placed at the end of BB.
|
unimpared if it was placed at the end of BB.
|
|
|
Keep track of how many times this expression is hoistable
|
Keep track of how many times this expression is hoistable
|
from a dominated block into BB. */
|
from a dominated block into BB. */
|
if (hoist_expr_reaches_here_p (bb, i, dominated, NULL))
|
if (hoist_expr_reaches_here_p (bb, i, dominated, NULL))
|
hoistable++;
|
hoistable++;
|
}
|
}
|
|
|
/* If we found more than one hoistable occurrence of this
|
/* If we found more than one hoistable occurrence of this
|
expression, then note it in the bitmap of expressions to
|
expression, then note it in the bitmap of expressions to
|
hoist. It makes no sense to hoist things which are computed
|
hoist. It makes no sense to hoist things which are computed
|
in only one BB, and doing so tends to pessimize register
|
in only one BB, and doing so tends to pessimize register
|
allocation. One could increase this value to try harder
|
allocation. One could increase this value to try harder
|
to avoid any possible code expansion due to register
|
to avoid any possible code expansion due to register
|
allocation issues; however experiments have shown that
|
allocation issues; however experiments have shown that
|
the vast majority of hoistable expressions are only movable
|
the vast majority of hoistable expressions are only movable
|
from two successors, so raising this threshold is likely
|
from two successors, so raising this threshold is likely
|
to nullify any benefit we get from code hoisting. */
|
to nullify any benefit we get from code hoisting. */
|
if (hoistable > 1)
|
if (hoistable > 1)
|
{
|
{
|
SET_BIT (hoist_exprs[bb->index], i);
|
SET_BIT (hoist_exprs[bb->index], i);
|
found = 1;
|
found = 1;
|
}
|
}
|
}
|
}
|
}
|
}
|
/* If we found nothing to hoist, then quit now. */
|
/* If we found nothing to hoist, then quit now. */
|
if (! found)
|
if (! found)
|
{
|
{
|
free (domby);
|
free (domby);
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Loop over all the hoistable expressions. */
|
/* Loop over all the hoistable expressions. */
|
for (i = 0; i < hoist_exprs[bb->index]->n_bits; i++)
|
for (i = 0; i < hoist_exprs[bb->index]->n_bits; i++)
|
{
|
{
|
/* We want to insert the expression into BB only once, so
|
/* We want to insert the expression into BB only once, so
|
note when we've inserted it. */
|
note when we've inserted it. */
|
insn_inserted_p = 0;
|
insn_inserted_p = 0;
|
|
|
/* These tests should be the same as the tests above. */
|
/* These tests should be the same as the tests above. */
|
if (TEST_BIT (hoist_exprs[bb->index], i))
|
if (TEST_BIT (hoist_exprs[bb->index], i))
|
{
|
{
|
/* We've found a potentially hoistable expression, now
|
/* We've found a potentially hoistable expression, now
|
we look at every block BB dominates to see if it
|
we look at every block BB dominates to see if it
|
computes the expression. */
|
computes the expression. */
|
for (j = 0; j < domby_len; j++)
|
for (j = 0; j < domby_len; j++)
|
{
|
{
|
dominated = domby[j];
|
dominated = domby[j];
|
/* Ignore self dominance. */
|
/* Ignore self dominance. */
|
if (bb == dominated)
|
if (bb == dominated)
|
continue;
|
continue;
|
|
|
/* We've found a dominated block, now see if it computes
|
/* We've found a dominated block, now see if it computes
|
the busy expression and whether or not moving that
|
the busy expression and whether or not moving that
|
expression to the "beginning" of that block is safe. */
|
expression to the "beginning" of that block is safe. */
|
if (!TEST_BIT (antloc[dominated->index], i))
|
if (!TEST_BIT (antloc[dominated->index], i))
|
continue;
|
continue;
|
|
|
/* The expression is computed in the dominated block and
|
/* The expression is computed in the dominated block and
|
it would be safe to compute it at the start of the
|
it would be safe to compute it at the start of the
|
dominated block. Now we have to determine if the
|
dominated block. Now we have to determine if the
|
expression would reach the dominated block if it was
|
expression would reach the dominated block if it was
|
placed at the end of BB. */
|
placed at the end of BB. */
|
if (hoist_expr_reaches_here_p (bb, i, dominated, NULL))
|
if (hoist_expr_reaches_here_p (bb, i, dominated, NULL))
|
{
|
{
|
struct expr *expr = index_map[i];
|
struct expr *expr = index_map[i];
|
struct occr *occr = expr->antic_occr;
|
struct occr *occr = expr->antic_occr;
|
rtx insn;
|
rtx insn;
|
rtx set;
|
rtx set;
|
|
|
/* Find the right occurrence of this expression. */
|
/* Find the right occurrence of this expression. */
|
while (BLOCK_FOR_INSN (occr->insn) != dominated && occr)
|
while (BLOCK_FOR_INSN (occr->insn) != dominated && occr)
|
occr = occr->next;
|
occr = occr->next;
|
|
|
gcc_assert (occr);
|
gcc_assert (occr);
|
insn = occr->insn;
|
insn = occr->insn;
|
set = single_set (insn);
|
set = single_set (insn);
|
gcc_assert (set);
|
gcc_assert (set);
|
|
|
/* Create a pseudo-reg to store the result of reaching
|
/* Create a pseudo-reg to store the result of reaching
|
expressions into. Get the mode for the new pseudo
|
expressions into. Get the mode for the new pseudo
|
from the mode of the original destination pseudo. */
|
from the mode of the original destination pseudo. */
|
if (expr->reaching_reg == NULL)
|
if (expr->reaching_reg == NULL)
|
expr->reaching_reg
|
expr->reaching_reg
|
= gen_reg_rtx (GET_MODE (SET_DEST (set)));
|
= gen_reg_rtx (GET_MODE (SET_DEST (set)));
|
|
|
gcse_emit_move_after (expr->reaching_reg, SET_DEST (set), insn);
|
gcse_emit_move_after (expr->reaching_reg, SET_DEST (set), insn);
|
delete_insn (insn);
|
delete_insn (insn);
|
occr->deleted_p = 1;
|
occr->deleted_p = 1;
|
if (!insn_inserted_p)
|
if (!insn_inserted_p)
|
{
|
{
|
insert_insn_end_bb (index_map[i], bb, 0);
|
insert_insn_end_bb (index_map[i], bb, 0);
|
insn_inserted_p = 1;
|
insn_inserted_p = 1;
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
free (domby);
|
free (domby);
|
}
|
}
|
|
|
free (index_map);
|
free (index_map);
|
}
|
}
|
|
|
/* Top level routine to perform one code hoisting (aka unification) pass
|
/* Top level routine to perform one code hoisting (aka unification) pass
|
|
|
Return nonzero if a change was made. */
|
Return nonzero if a change was made. */
|
|
|
static int
|
static int
|
one_code_hoisting_pass (void)
|
one_code_hoisting_pass (void)
|
{
|
{
|
int changed = 0;
|
int changed = 0;
|
|
|
alloc_hash_table (max_cuid, &expr_hash_table, 0);
|
alloc_hash_table (max_cuid, &expr_hash_table, 0);
|
compute_hash_table (&expr_hash_table);
|
compute_hash_table (&expr_hash_table);
|
if (dump_file)
|
if (dump_file)
|
dump_hash_table (dump_file, "Code Hosting Expressions", &expr_hash_table);
|
dump_hash_table (dump_file, "Code Hosting Expressions", &expr_hash_table);
|
|
|
if (expr_hash_table.n_elems > 0)
|
if (expr_hash_table.n_elems > 0)
|
{
|
{
|
alloc_code_hoist_mem (last_basic_block, expr_hash_table.n_elems);
|
alloc_code_hoist_mem (last_basic_block, expr_hash_table.n_elems);
|
compute_code_hoist_data ();
|
compute_code_hoist_data ();
|
hoist_code ();
|
hoist_code ();
|
free_code_hoist_mem ();
|
free_code_hoist_mem ();
|
}
|
}
|
|
|
free_hash_table (&expr_hash_table);
|
free_hash_table (&expr_hash_table);
|
|
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* Here we provide the things required to do store motion towards
|
/* Here we provide the things required to do store motion towards
|
the exit. In order for this to be effective, gcse also needed to
|
the exit. In order for this to be effective, gcse also needed to
|
be taught how to move a load when it is kill only by a store to itself.
|
be taught how to move a load when it is kill only by a store to itself.
|
|
|
int i;
|
int i;
|
float a[10];
|
float a[10];
|
|
|
void foo(float scale)
|
void foo(float scale)
|
{
|
{
|
for (i=0; i<10; i++)
|
for (i=0; i<10; i++)
|
a[i] *= scale;
|
a[i] *= scale;
|
}
|
}
|
|
|
'i' is both loaded and stored to in the loop. Normally, gcse cannot move
|
'i' is both loaded and stored to in the loop. Normally, gcse cannot move
|
the load out since its live around the loop, and stored at the bottom
|
the load out since its live around the loop, and stored at the bottom
|
of the loop.
|
of the loop.
|
|
|
The 'Load Motion' referred to and implemented in this file is
|
The 'Load Motion' referred to and implemented in this file is
|
an enhancement to gcse which when using edge based lcm, recognizes
|
an enhancement to gcse which when using edge based lcm, recognizes
|
this situation and allows gcse to move the load out of the loop.
|
this situation and allows gcse to move the load out of the loop.
|
|
|
Once gcse has hoisted the load, store motion can then push this
|
Once gcse has hoisted the load, store motion can then push this
|
load towards the exit, and we end up with no loads or stores of 'i'
|
load towards the exit, and we end up with no loads or stores of 'i'
|
in the loop. */
|
in the loop. */
|
|
|
static hashval_t
|
static hashval_t
|
pre_ldst_expr_hash (const void *p)
|
pre_ldst_expr_hash (const void *p)
|
{
|
{
|
int do_not_record_p = 0;
|
int do_not_record_p = 0;
|
const struct ls_expr *x = p;
|
const struct ls_expr *x = p;
|
return hash_rtx (x->pattern, GET_MODE (x->pattern), &do_not_record_p, NULL, false);
|
return hash_rtx (x->pattern, GET_MODE (x->pattern), &do_not_record_p, NULL, false);
|
}
|
}
|
|
|
static int
|
static int
|
pre_ldst_expr_eq (const void *p1, const void *p2)
|
pre_ldst_expr_eq (const void *p1, const void *p2)
|
{
|
{
|
const struct ls_expr *ptr1 = p1, *ptr2 = p2;
|
const struct ls_expr *ptr1 = p1, *ptr2 = p2;
|
return expr_equiv_p (ptr1->pattern, ptr2->pattern);
|
return expr_equiv_p (ptr1->pattern, ptr2->pattern);
|
}
|
}
|
|
|
/* This will search the ldst list for a matching expression. If it
|
/* This will search the ldst list for a matching expression. If it
|
doesn't find one, we create one and initialize it. */
|
doesn't find one, we create one and initialize it. */
|
|
|
static struct ls_expr *
|
static struct ls_expr *
|
ldst_entry (rtx x)
|
ldst_entry (rtx x)
|
{
|
{
|
int do_not_record_p = 0;
|
int do_not_record_p = 0;
|
struct ls_expr * ptr;
|
struct ls_expr * ptr;
|
unsigned int hash;
|
unsigned int hash;
|
void **slot;
|
void **slot;
|
struct ls_expr e;
|
struct ls_expr e;
|
|
|
hash = hash_rtx (x, GET_MODE (x), &do_not_record_p,
|
hash = hash_rtx (x, GET_MODE (x), &do_not_record_p,
|
NULL, /*have_reg_qty=*/false);
|
NULL, /*have_reg_qty=*/false);
|
|
|
e.pattern = x;
|
e.pattern = x;
|
slot = htab_find_slot_with_hash (pre_ldst_table, &e, hash, INSERT);
|
slot = htab_find_slot_with_hash (pre_ldst_table, &e, hash, INSERT);
|
if (*slot)
|
if (*slot)
|
return (struct ls_expr *)*slot;
|
return (struct ls_expr *)*slot;
|
|
|
ptr = XNEW (struct ls_expr);
|
ptr = XNEW (struct ls_expr);
|
|
|
ptr->next = pre_ldst_mems;
|
ptr->next = pre_ldst_mems;
|
ptr->expr = NULL;
|
ptr->expr = NULL;
|
ptr->pattern = x;
|
ptr->pattern = x;
|
ptr->pattern_regs = NULL_RTX;
|
ptr->pattern_regs = NULL_RTX;
|
ptr->loads = NULL_RTX;
|
ptr->loads = NULL_RTX;
|
ptr->stores = NULL_RTX;
|
ptr->stores = NULL_RTX;
|
ptr->reaching_reg = NULL_RTX;
|
ptr->reaching_reg = NULL_RTX;
|
ptr->invalid = 0;
|
ptr->invalid = 0;
|
ptr->index = 0;
|
ptr->index = 0;
|
ptr->hash_index = hash;
|
ptr->hash_index = hash;
|
pre_ldst_mems = ptr;
|
pre_ldst_mems = ptr;
|
*slot = ptr;
|
*slot = ptr;
|
|
|
return ptr;
|
return ptr;
|
}
|
}
|
|
|
/* Free up an individual ldst entry. */
|
/* Free up an individual ldst entry. */
|
|
|
static void
|
static void
|
free_ldst_entry (struct ls_expr * ptr)
|
free_ldst_entry (struct ls_expr * ptr)
|
{
|
{
|
free_INSN_LIST_list (& ptr->loads);
|
free_INSN_LIST_list (& ptr->loads);
|
free_INSN_LIST_list (& ptr->stores);
|
free_INSN_LIST_list (& ptr->stores);
|
|
|
free (ptr);
|
free (ptr);
|
}
|
}
|
|
|
/* Free up all memory associated with the ldst list. */
|
/* Free up all memory associated with the ldst list. */
|
|
|
static void
|
static void
|
free_ldst_mems (void)
|
free_ldst_mems (void)
|
{
|
{
|
if (pre_ldst_table)
|
if (pre_ldst_table)
|
htab_delete (pre_ldst_table);
|
htab_delete (pre_ldst_table);
|
pre_ldst_table = NULL;
|
pre_ldst_table = NULL;
|
|
|
while (pre_ldst_mems)
|
while (pre_ldst_mems)
|
{
|
{
|
struct ls_expr * tmp = pre_ldst_mems;
|
struct ls_expr * tmp = pre_ldst_mems;
|
|
|
pre_ldst_mems = pre_ldst_mems->next;
|
pre_ldst_mems = pre_ldst_mems->next;
|
|
|
free_ldst_entry (tmp);
|
free_ldst_entry (tmp);
|
}
|
}
|
|
|
pre_ldst_mems = NULL;
|
pre_ldst_mems = NULL;
|
}
|
}
|
|
|
/* Dump debugging info about the ldst list. */
|
/* Dump debugging info about the ldst list. */
|
|
|
static void
|
static void
|
print_ldst_list (FILE * file)
|
print_ldst_list (FILE * file)
|
{
|
{
|
struct ls_expr * ptr;
|
struct ls_expr * ptr;
|
|
|
fprintf (file, "LDST list: \n");
|
fprintf (file, "LDST list: \n");
|
|
|
for (ptr = first_ls_expr(); ptr != NULL; ptr = next_ls_expr (ptr))
|
for (ptr = first_ls_expr(); ptr != NULL; ptr = next_ls_expr (ptr))
|
{
|
{
|
fprintf (file, " Pattern (%3d): ", ptr->index);
|
fprintf (file, " Pattern (%3d): ", ptr->index);
|
|
|
print_rtl (file, ptr->pattern);
|
print_rtl (file, ptr->pattern);
|
|
|
fprintf (file, "\n Loads : ");
|
fprintf (file, "\n Loads : ");
|
|
|
if (ptr->loads)
|
if (ptr->loads)
|
print_rtl (file, ptr->loads);
|
print_rtl (file, ptr->loads);
|
else
|
else
|
fprintf (file, "(nil)");
|
fprintf (file, "(nil)");
|
|
|
fprintf (file, "\n Stores : ");
|
fprintf (file, "\n Stores : ");
|
|
|
if (ptr->stores)
|
if (ptr->stores)
|
print_rtl (file, ptr->stores);
|
print_rtl (file, ptr->stores);
|
else
|
else
|
fprintf (file, "(nil)");
|
fprintf (file, "(nil)");
|
|
|
fprintf (file, "\n\n");
|
fprintf (file, "\n\n");
|
}
|
}
|
|
|
fprintf (file, "\n");
|
fprintf (file, "\n");
|
}
|
}
|
|
|
/* Returns 1 if X is in the list of ldst only expressions. */
|
/* Returns 1 if X is in the list of ldst only expressions. */
|
|
|
static struct ls_expr *
|
static struct ls_expr *
|
find_rtx_in_ldst (rtx x)
|
find_rtx_in_ldst (rtx x)
|
{
|
{
|
struct ls_expr e;
|
struct ls_expr e;
|
void **slot;
|
void **slot;
|
if (!pre_ldst_table)
|
if (!pre_ldst_table)
|
return NULL;
|
return NULL;
|
e.pattern = x;
|
e.pattern = x;
|
slot = htab_find_slot (pre_ldst_table, &e, NO_INSERT);
|
slot = htab_find_slot (pre_ldst_table, &e, NO_INSERT);
|
if (!slot || ((struct ls_expr *)*slot)->invalid)
|
if (!slot || ((struct ls_expr *)*slot)->invalid)
|
return NULL;
|
return NULL;
|
return *slot;
|
return *slot;
|
}
|
}
|
|
|
/* Assign each element of the list of mems a monotonically increasing value. */
|
/* Assign each element of the list of mems a monotonically increasing value. */
|
|
|
static int
|
static int
|
enumerate_ldsts (void)
|
enumerate_ldsts (void)
|
{
|
{
|
struct ls_expr * ptr;
|
struct ls_expr * ptr;
|
int n = 0;
|
int n = 0;
|
|
|
for (ptr = pre_ldst_mems; ptr != NULL; ptr = ptr->next)
|
for (ptr = pre_ldst_mems; ptr != NULL; ptr = ptr->next)
|
ptr->index = n++;
|
ptr->index = n++;
|
|
|
return n;
|
return n;
|
}
|
}
|
|
|
/* Return first item in the list. */
|
/* Return first item in the list. */
|
|
|
static inline struct ls_expr *
|
static inline struct ls_expr *
|
first_ls_expr (void)
|
first_ls_expr (void)
|
{
|
{
|
return pre_ldst_mems;
|
return pre_ldst_mems;
|
}
|
}
|
|
|
/* Return the next item in the list after the specified one. */
|
/* Return the next item in the list after the specified one. */
|
|
|
static inline struct ls_expr *
|
static inline struct ls_expr *
|
next_ls_expr (struct ls_expr * ptr)
|
next_ls_expr (struct ls_expr * ptr)
|
{
|
{
|
return ptr->next;
|
return ptr->next;
|
}
|
}
|
|
|
/* Load Motion for loads which only kill themselves. */
|
/* Load Motion for loads which only kill themselves. */
|
|
|
/* Return true if x is a simple MEM operation, with no registers or
|
/* Return true if x is a simple MEM operation, with no registers or
|
side effects. These are the types of loads we consider for the
|
side effects. These are the types of loads we consider for the
|
ld_motion list, otherwise we let the usual aliasing take care of it. */
|
ld_motion list, otherwise we let the usual aliasing take care of it. */
|
|
|
static int
|
static int
|
simple_mem (rtx x)
|
simple_mem (rtx x)
|
{
|
{
|
if (! MEM_P (x))
|
if (! MEM_P (x))
|
return 0;
|
return 0;
|
|
|
if (MEM_VOLATILE_P (x))
|
if (MEM_VOLATILE_P (x))
|
return 0;
|
return 0;
|
|
|
if (GET_MODE (x) == BLKmode)
|
if (GET_MODE (x) == BLKmode)
|
return 0;
|
return 0;
|
|
|
/* If we are handling exceptions, we must be careful with memory references
|
/* If we are handling exceptions, we must be careful with memory references
|
that may trap. If we are not, the behavior is undefined, so we may just
|
that may trap. If we are not, the behavior is undefined, so we may just
|
continue. */
|
continue. */
|
if (flag_non_call_exceptions && may_trap_p (x))
|
if (flag_non_call_exceptions && may_trap_p (x))
|
return 0;
|
return 0;
|
|
|
if (side_effects_p (x))
|
if (side_effects_p (x))
|
return 0;
|
return 0;
|
|
|
/* Do not consider function arguments passed on stack. */
|
/* Do not consider function arguments passed on stack. */
|
if (reg_mentioned_p (stack_pointer_rtx, x))
|
if (reg_mentioned_p (stack_pointer_rtx, x))
|
return 0;
|
return 0;
|
|
|
if (flag_float_store && FLOAT_MODE_P (GET_MODE (x)))
|
if (flag_float_store && FLOAT_MODE_P (GET_MODE (x)))
|
return 0;
|
return 0;
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Make sure there isn't a buried reference in this pattern anywhere.
|
/* Make sure there isn't a buried reference in this pattern anywhere.
|
If there is, invalidate the entry for it since we're not capable
|
If there is, invalidate the entry for it since we're not capable
|
of fixing it up just yet.. We have to be sure we know about ALL
|
of fixing it up just yet.. We have to be sure we know about ALL
|
loads since the aliasing code will allow all entries in the
|
loads since the aliasing code will allow all entries in the
|
ld_motion list to not-alias itself. If we miss a load, we will get
|
ld_motion list to not-alias itself. If we miss a load, we will get
|
the wrong value since gcse might common it and we won't know to
|
the wrong value since gcse might common it and we won't know to
|
fix it up. */
|
fix it up. */
|
|
|
static void
|
static void
|
invalidate_any_buried_refs (rtx x)
|
invalidate_any_buried_refs (rtx x)
|
{
|
{
|
const char * fmt;
|
const char * fmt;
|
int i, j;
|
int i, j;
|
struct ls_expr * ptr;
|
struct ls_expr * ptr;
|
|
|
/* Invalidate it in the list. */
|
/* Invalidate it in the list. */
|
if (MEM_P (x) && simple_mem (x))
|
if (MEM_P (x) && simple_mem (x))
|
{
|
{
|
ptr = ldst_entry (x);
|
ptr = ldst_entry (x);
|
ptr->invalid = 1;
|
ptr->invalid = 1;
|
}
|
}
|
|
|
/* Recursively process the insn. */
|
/* Recursively process the insn. */
|
fmt = GET_RTX_FORMAT (GET_CODE (x));
|
fmt = GET_RTX_FORMAT (GET_CODE (x));
|
|
|
for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0; i--)
|
for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0; i--)
|
{
|
{
|
if (fmt[i] == 'e')
|
if (fmt[i] == 'e')
|
invalidate_any_buried_refs (XEXP (x, i));
|
invalidate_any_buried_refs (XEXP (x, i));
|
else if (fmt[i] == 'E')
|
else if (fmt[i] == 'E')
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
invalidate_any_buried_refs (XVECEXP (x, i, j));
|
invalidate_any_buried_refs (XVECEXP (x, i, j));
|
}
|
}
|
}
|
}
|
|
|
/* Find all the 'simple' MEMs which are used in LOADs and STORES. Simple
|
/* Find all the 'simple' MEMs which are used in LOADs and STORES. Simple
|
being defined as MEM loads and stores to symbols, with no side effects
|
being defined as MEM loads and stores to symbols, with no side effects
|
and no registers in the expression. For a MEM destination, we also
|
and no registers in the expression. For a MEM destination, we also
|
check that the insn is still valid if we replace the destination with a
|
check that the insn is still valid if we replace the destination with a
|
REG, as is done in update_ld_motion_stores. If there are any uses/defs
|
REG, as is done in update_ld_motion_stores. If there are any uses/defs
|
which don't match this criteria, they are invalidated and trimmed out
|
which don't match this criteria, they are invalidated and trimmed out
|
later. */
|
later. */
|
|
|
static void
|
static void
|
compute_ld_motion_mems (void)
|
compute_ld_motion_mems (void)
|
{
|
{
|
struct ls_expr * ptr;
|
struct ls_expr * ptr;
|
basic_block bb;
|
basic_block bb;
|
rtx insn;
|
rtx insn;
|
|
|
pre_ldst_mems = NULL;
|
pre_ldst_mems = NULL;
|
pre_ldst_table = htab_create (13, pre_ldst_expr_hash,
|
pre_ldst_table = htab_create (13, pre_ldst_expr_hash,
|
pre_ldst_expr_eq, NULL);
|
pre_ldst_expr_eq, NULL);
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
{
|
{
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
{
|
{
|
if (GET_CODE (PATTERN (insn)) == SET)
|
if (GET_CODE (PATTERN (insn)) == SET)
|
{
|
{
|
rtx src = SET_SRC (PATTERN (insn));
|
rtx src = SET_SRC (PATTERN (insn));
|
rtx dest = SET_DEST (PATTERN (insn));
|
rtx dest = SET_DEST (PATTERN (insn));
|
|
|
/* Check for a simple LOAD... */
|
/* Check for a simple LOAD... */
|
if (MEM_P (src) && simple_mem (src))
|
if (MEM_P (src) && simple_mem (src))
|
{
|
{
|
ptr = ldst_entry (src);
|
ptr = ldst_entry (src);
|
if (REG_P (dest))
|
if (REG_P (dest))
|
ptr->loads = alloc_INSN_LIST (insn, ptr->loads);
|
ptr->loads = alloc_INSN_LIST (insn, ptr->loads);
|
else
|
else
|
ptr->invalid = 1;
|
ptr->invalid = 1;
|
}
|
}
|
else
|
else
|
{
|
{
|
/* Make sure there isn't a buried load somewhere. */
|
/* Make sure there isn't a buried load somewhere. */
|
invalidate_any_buried_refs (src);
|
invalidate_any_buried_refs (src);
|
}
|
}
|
|
|
/* Check for stores. Don't worry about aliased ones, they
|
/* Check for stores. Don't worry about aliased ones, they
|
will block any movement we might do later. We only care
|
will block any movement we might do later. We only care
|
about this exact pattern since those are the only
|
about this exact pattern since those are the only
|
circumstance that we will ignore the aliasing info. */
|
circumstance that we will ignore the aliasing info. */
|
if (MEM_P (dest) && simple_mem (dest))
|
if (MEM_P (dest) && simple_mem (dest))
|
{
|
{
|
ptr = ldst_entry (dest);
|
ptr = ldst_entry (dest);
|
|
|
if (! MEM_P (src)
|
if (! MEM_P (src)
|
&& GET_CODE (src) != ASM_OPERANDS
|
&& GET_CODE (src) != ASM_OPERANDS
|
/* Check for REG manually since want_to_gcse_p
|
/* Check for REG manually since want_to_gcse_p
|
returns 0 for all REGs. */
|
returns 0 for all REGs. */
|
&& can_assign_to_reg_p (src))
|
&& can_assign_to_reg_p (src))
|
ptr->stores = alloc_INSN_LIST (insn, ptr->stores);
|
ptr->stores = alloc_INSN_LIST (insn, ptr->stores);
|
else
|
else
|
ptr->invalid = 1;
|
ptr->invalid = 1;
|
}
|
}
|
}
|
}
|
else
|
else
|
invalidate_any_buried_refs (PATTERN (insn));
|
invalidate_any_buried_refs (PATTERN (insn));
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Remove any references that have been either invalidated or are not in the
|
/* Remove any references that have been either invalidated or are not in the
|
expression list for pre gcse. */
|
expression list for pre gcse. */
|
|
|
static void
|
static void
|
trim_ld_motion_mems (void)
|
trim_ld_motion_mems (void)
|
{
|
{
|
struct ls_expr * * last = & pre_ldst_mems;
|
struct ls_expr * * last = & pre_ldst_mems;
|
struct ls_expr * ptr = pre_ldst_mems;
|
struct ls_expr * ptr = pre_ldst_mems;
|
|
|
while (ptr != NULL)
|
while (ptr != NULL)
|
{
|
{
|
struct expr * expr;
|
struct expr * expr;
|
|
|
/* Delete if entry has been made invalid. */
|
/* Delete if entry has been made invalid. */
|
if (! ptr->invalid)
|
if (! ptr->invalid)
|
{
|
{
|
/* Delete if we cannot find this mem in the expression list. */
|
/* Delete if we cannot find this mem in the expression list. */
|
unsigned int hash = ptr->hash_index % expr_hash_table.size;
|
unsigned int hash = ptr->hash_index % expr_hash_table.size;
|
|
|
for (expr = expr_hash_table.table[hash];
|
for (expr = expr_hash_table.table[hash];
|
expr != NULL;
|
expr != NULL;
|
expr = expr->next_same_hash)
|
expr = expr->next_same_hash)
|
if (expr_equiv_p (expr->expr, ptr->pattern))
|
if (expr_equiv_p (expr->expr, ptr->pattern))
|
break;
|
break;
|
}
|
}
|
else
|
else
|
expr = (struct expr *) 0;
|
expr = (struct expr *) 0;
|
|
|
if (expr)
|
if (expr)
|
{
|
{
|
/* Set the expression field if we are keeping it. */
|
/* Set the expression field if we are keeping it. */
|
ptr->expr = expr;
|
ptr->expr = expr;
|
last = & ptr->next;
|
last = & ptr->next;
|
ptr = ptr->next;
|
ptr = ptr->next;
|
}
|
}
|
else
|
else
|
{
|
{
|
*last = ptr->next;
|
*last = ptr->next;
|
htab_remove_elt_with_hash (pre_ldst_table, ptr, ptr->hash_index);
|
htab_remove_elt_with_hash (pre_ldst_table, ptr, ptr->hash_index);
|
free_ldst_entry (ptr);
|
free_ldst_entry (ptr);
|
ptr = * last;
|
ptr = * last;
|
}
|
}
|
}
|
}
|
|
|
/* Show the world what we've found. */
|
/* Show the world what we've found. */
|
if (dump_file && pre_ldst_mems != NULL)
|
if (dump_file && pre_ldst_mems != NULL)
|
print_ldst_list (dump_file);
|
print_ldst_list (dump_file);
|
}
|
}
|
|
|
/* This routine will take an expression which we are replacing with
|
/* This routine will take an expression which we are replacing with
|
a reaching register, and update any stores that are needed if
|
a reaching register, and update any stores that are needed if
|
that expression is in the ld_motion list. Stores are updated by
|
that expression is in the ld_motion list. Stores are updated by
|
copying their SRC to the reaching register, and then storing
|
copying their SRC to the reaching register, and then storing
|
the reaching register into the store location. These keeps the
|
the reaching register into the store location. These keeps the
|
correct value in the reaching register for the loads. */
|
correct value in the reaching register for the loads. */
|
|
|
static void
|
static void
|
update_ld_motion_stores (struct expr * expr)
|
update_ld_motion_stores (struct expr * expr)
|
{
|
{
|
struct ls_expr * mem_ptr;
|
struct ls_expr * mem_ptr;
|
|
|
if ((mem_ptr = find_rtx_in_ldst (expr->expr)))
|
if ((mem_ptr = find_rtx_in_ldst (expr->expr)))
|
{
|
{
|
/* We can try to find just the REACHED stores, but is shouldn't
|
/* We can try to find just the REACHED stores, but is shouldn't
|
matter to set the reaching reg everywhere... some might be
|
matter to set the reaching reg everywhere... some might be
|
dead and should be eliminated later. */
|
dead and should be eliminated later. */
|
|
|
/* We replace (set mem expr) with (set reg expr) (set mem reg)
|
/* We replace (set mem expr) with (set reg expr) (set mem reg)
|
where reg is the reaching reg used in the load. We checked in
|
where reg is the reaching reg used in the load. We checked in
|
compute_ld_motion_mems that we can replace (set mem expr) with
|
compute_ld_motion_mems that we can replace (set mem expr) with
|
(set reg expr) in that insn. */
|
(set reg expr) in that insn. */
|
rtx list = mem_ptr->stores;
|
rtx list = mem_ptr->stores;
|
|
|
for ( ; list != NULL_RTX; list = XEXP (list, 1))
|
for ( ; list != NULL_RTX; list = XEXP (list, 1))
|
{
|
{
|
rtx insn = XEXP (list, 0);
|
rtx insn = XEXP (list, 0);
|
rtx pat = PATTERN (insn);
|
rtx pat = PATTERN (insn);
|
rtx src = SET_SRC (pat);
|
rtx src = SET_SRC (pat);
|
rtx reg = expr->reaching_reg;
|
rtx reg = expr->reaching_reg;
|
rtx copy, new;
|
rtx copy, new;
|
|
|
/* If we've already copied it, continue. */
|
/* If we've already copied it, continue. */
|
if (expr->reaching_reg == src)
|
if (expr->reaching_reg == src)
|
continue;
|
continue;
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "PRE: store updated with reaching reg ");
|
fprintf (dump_file, "PRE: store updated with reaching reg ");
|
print_rtl (dump_file, expr->reaching_reg);
|
print_rtl (dump_file, expr->reaching_reg);
|
fprintf (dump_file, ":\n ");
|
fprintf (dump_file, ":\n ");
|
print_inline_rtx (dump_file, insn, 8);
|
print_inline_rtx (dump_file, insn, 8);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
copy = gen_move_insn ( reg, copy_rtx (SET_SRC (pat)));
|
copy = gen_move_insn ( reg, copy_rtx (SET_SRC (pat)));
|
new = emit_insn_before (copy, insn);
|
new = emit_insn_before (copy, insn);
|
record_one_set (REGNO (reg), new);
|
record_one_set (REGNO (reg), new);
|
SET_SRC (pat) = reg;
|
SET_SRC (pat) = reg;
|
|
|
/* un-recognize this pattern since it's probably different now. */
|
/* un-recognize this pattern since it's probably different now. */
|
INSN_CODE (insn) = -1;
|
INSN_CODE (insn) = -1;
|
gcse_create_count++;
|
gcse_create_count++;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Store motion code. */
|
/* Store motion code. */
|
|
|
#define ANTIC_STORE_LIST(x) ((x)->loads)
|
#define ANTIC_STORE_LIST(x) ((x)->loads)
|
#define AVAIL_STORE_LIST(x) ((x)->stores)
|
#define AVAIL_STORE_LIST(x) ((x)->stores)
|
#define LAST_AVAIL_CHECK_FAILURE(x) ((x)->reaching_reg)
|
#define LAST_AVAIL_CHECK_FAILURE(x) ((x)->reaching_reg)
|
|
|
/* This is used to communicate the target bitvector we want to use in the
|
/* This is used to communicate the target bitvector we want to use in the
|
reg_set_info routine when called via the note_stores mechanism. */
|
reg_set_info routine when called via the note_stores mechanism. */
|
static int * regvec;
|
static int * regvec;
|
|
|
/* And current insn, for the same routine. */
|
/* And current insn, for the same routine. */
|
static rtx compute_store_table_current_insn;
|
static rtx compute_store_table_current_insn;
|
|
|
/* Used in computing the reverse edge graph bit vectors. */
|
/* Used in computing the reverse edge graph bit vectors. */
|
static sbitmap * st_antloc;
|
static sbitmap * st_antloc;
|
|
|
/* Global holding the number of store expressions we are dealing with. */
|
/* Global holding the number of store expressions we are dealing with. */
|
static int num_stores;
|
static int num_stores;
|
|
|
/* Checks to set if we need to mark a register set. Called from
|
/* Checks to set if we need to mark a register set. Called from
|
note_stores. */
|
note_stores. */
|
|
|
static void
|
static void
|
reg_set_info (rtx dest, rtx setter ATTRIBUTE_UNUSED,
|
reg_set_info (rtx dest, rtx setter ATTRIBUTE_UNUSED,
|
void *data)
|
void *data)
|
{
|
{
|
sbitmap bb_reg = data;
|
sbitmap bb_reg = data;
|
|
|
if (GET_CODE (dest) == SUBREG)
|
if (GET_CODE (dest) == SUBREG)
|
dest = SUBREG_REG (dest);
|
dest = SUBREG_REG (dest);
|
|
|
if (REG_P (dest))
|
if (REG_P (dest))
|
{
|
{
|
regvec[REGNO (dest)] = INSN_UID (compute_store_table_current_insn);
|
regvec[REGNO (dest)] = INSN_UID (compute_store_table_current_insn);
|
if (bb_reg)
|
if (bb_reg)
|
SET_BIT (bb_reg, REGNO (dest));
|
SET_BIT (bb_reg, REGNO (dest));
|
}
|
}
|
}
|
}
|
|
|
/* Clear any mark that says that this insn sets dest. Called from
|
/* Clear any mark that says that this insn sets dest. Called from
|
note_stores. */
|
note_stores. */
|
|
|
static void
|
static void
|
reg_clear_last_set (rtx dest, rtx setter ATTRIBUTE_UNUSED,
|
reg_clear_last_set (rtx dest, rtx setter ATTRIBUTE_UNUSED,
|
void *data)
|
void *data)
|
{
|
{
|
int *dead_vec = data;
|
int *dead_vec = data;
|
|
|
if (GET_CODE (dest) == SUBREG)
|
if (GET_CODE (dest) == SUBREG)
|
dest = SUBREG_REG (dest);
|
dest = SUBREG_REG (dest);
|
|
|
if (REG_P (dest) &&
|
if (REG_P (dest) &&
|
dead_vec[REGNO (dest)] == INSN_UID (compute_store_table_current_insn))
|
dead_vec[REGNO (dest)] == INSN_UID (compute_store_table_current_insn))
|
dead_vec[REGNO (dest)] = 0;
|
dead_vec[REGNO (dest)] = 0;
|
}
|
}
|
|
|
/* Return zero if some of the registers in list X are killed
|
/* Return zero if some of the registers in list X are killed
|
due to set of registers in bitmap REGS_SET. */
|
due to set of registers in bitmap REGS_SET. */
|
|
|
static bool
|
static bool
|
store_ops_ok (rtx x, int *regs_set)
|
store_ops_ok (rtx x, int *regs_set)
|
{
|
{
|
rtx reg;
|
rtx reg;
|
|
|
for (; x; x = XEXP (x, 1))
|
for (; x; x = XEXP (x, 1))
|
{
|
{
|
reg = XEXP (x, 0);
|
reg = XEXP (x, 0);
|
if (regs_set[REGNO(reg)])
|
if (regs_set[REGNO(reg)])
|
return false;
|
return false;
|
}
|
}
|
|
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Returns a list of registers mentioned in X. */
|
/* Returns a list of registers mentioned in X. */
|
static rtx
|
static rtx
|
extract_mentioned_regs (rtx x)
|
extract_mentioned_regs (rtx x)
|
{
|
{
|
return extract_mentioned_regs_helper (x, NULL_RTX);
|
return extract_mentioned_regs_helper (x, NULL_RTX);
|
}
|
}
|
|
|
/* Helper for extract_mentioned_regs; ACCUM is used to accumulate used
|
/* Helper for extract_mentioned_regs; ACCUM is used to accumulate used
|
registers. */
|
registers. */
|
static rtx
|
static rtx
|
extract_mentioned_regs_helper (rtx x, rtx accum)
|
extract_mentioned_regs_helper (rtx x, rtx accum)
|
{
|
{
|
int i;
|
int i;
|
enum rtx_code code;
|
enum rtx_code code;
|
const char * fmt;
|
const char * fmt;
|
|
|
/* Repeat is used to turn tail-recursion into iteration. */
|
/* Repeat is used to turn tail-recursion into iteration. */
|
repeat:
|
repeat:
|
|
|
if (x == 0)
|
if (x == 0)
|
return accum;
|
return accum;
|
|
|
code = GET_CODE (x);
|
code = GET_CODE (x);
|
switch (code)
|
switch (code)
|
{
|
{
|
case REG:
|
case REG:
|
return alloc_EXPR_LIST (0, x, accum);
|
return alloc_EXPR_LIST (0, x, accum);
|
|
|
case MEM:
|
case MEM:
|
x = XEXP (x, 0);
|
x = XEXP (x, 0);
|
goto repeat;
|
goto repeat;
|
|
|
case PRE_DEC:
|
case PRE_DEC:
|
case PRE_INC:
|
case PRE_INC:
|
case POST_DEC:
|
case POST_DEC:
|
case POST_INC:
|
case POST_INC:
|
/* We do not run this function with arguments having side effects. */
|
/* We do not run this function with arguments having side effects. */
|
gcc_unreachable ();
|
gcc_unreachable ();
|
|
|
case PC:
|
case PC:
|
case CC0: /*FIXME*/
|
case CC0: /*FIXME*/
|
case CONST:
|
case CONST:
|
case CONST_INT:
|
case CONST_INT:
|
case CONST_DOUBLE:
|
case CONST_DOUBLE:
|
case CONST_VECTOR:
|
case CONST_VECTOR:
|
case SYMBOL_REF:
|
case SYMBOL_REF:
|
case LABEL_REF:
|
case LABEL_REF:
|
case ADDR_VEC:
|
case ADDR_VEC:
|
case ADDR_DIFF_VEC:
|
case ADDR_DIFF_VEC:
|
return accum;
|
return accum;
|
|
|
default:
|
default:
|
break;
|
break;
|
}
|
}
|
|
|
i = GET_RTX_LENGTH (code) - 1;
|
i = GET_RTX_LENGTH (code) - 1;
|
fmt = GET_RTX_FORMAT (code);
|
fmt = GET_RTX_FORMAT (code);
|
|
|
for (; i >= 0; i--)
|
for (; i >= 0; i--)
|
{
|
{
|
if (fmt[i] == 'e')
|
if (fmt[i] == 'e')
|
{
|
{
|
rtx tem = XEXP (x, i);
|
rtx tem = XEXP (x, i);
|
|
|
/* If we are about to do the last recursive call
|
/* If we are about to do the last recursive call
|
needed at this level, change it into iteration. */
|
needed at this level, change it into iteration. */
|
if (i == 0)
|
if (i == 0)
|
{
|
{
|
x = tem;
|
x = tem;
|
goto repeat;
|
goto repeat;
|
}
|
}
|
|
|
accum = extract_mentioned_regs_helper (tem, accum);
|
accum = extract_mentioned_regs_helper (tem, accum);
|
}
|
}
|
else if (fmt[i] == 'E')
|
else if (fmt[i] == 'E')
|
{
|
{
|
int j;
|
int j;
|
|
|
for (j = 0; j < XVECLEN (x, i); j++)
|
for (j = 0; j < XVECLEN (x, i); j++)
|
accum = extract_mentioned_regs_helper (XVECEXP (x, i, j), accum);
|
accum = extract_mentioned_regs_helper (XVECEXP (x, i, j), accum);
|
}
|
}
|
}
|
}
|
|
|
return accum;
|
return accum;
|
}
|
}
|
|
|
/* Determine whether INSN is MEM store pattern that we will consider moving.
|
/* Determine whether INSN is MEM store pattern that we will consider moving.
|
REGS_SET_BEFORE is bitmap of registers set before (and including) the
|
REGS_SET_BEFORE is bitmap of registers set before (and including) the
|
current insn, REGS_SET_AFTER is bitmap of registers set after (and
|
current insn, REGS_SET_AFTER is bitmap of registers set after (and
|
including) the insn in this basic block. We must be passing through BB from
|
including) the insn in this basic block. We must be passing through BB from
|
head to end, as we are using this fact to speed things up.
|
head to end, as we are using this fact to speed things up.
|
|
|
The results are stored this way:
|
The results are stored this way:
|
|
|
-- the first anticipatable expression is added into ANTIC_STORE_LIST
|
-- the first anticipatable expression is added into ANTIC_STORE_LIST
|
-- if the processed expression is not anticipatable, NULL_RTX is added
|
-- if the processed expression is not anticipatable, NULL_RTX is added
|
there instead, so that we can use it as indicator that no further
|
there instead, so that we can use it as indicator that no further
|
expression of this type may be anticipatable
|
expression of this type may be anticipatable
|
-- if the expression is available, it is added as head of AVAIL_STORE_LIST;
|
-- if the expression is available, it is added as head of AVAIL_STORE_LIST;
|
consequently, all of them but this head are dead and may be deleted.
|
consequently, all of them but this head are dead and may be deleted.
|
-- if the expression is not available, the insn due to that it fails to be
|
-- if the expression is not available, the insn due to that it fails to be
|
available is stored in reaching_reg.
|
available is stored in reaching_reg.
|
|
|
The things are complicated a bit by fact that there already may be stores
|
The things are complicated a bit by fact that there already may be stores
|
to the same MEM from other blocks; also caller must take care of the
|
to the same MEM from other blocks; also caller must take care of the
|
necessary cleanup of the temporary markers after end of the basic block.
|
necessary cleanup of the temporary markers after end of the basic block.
|
*/
|
*/
|
|
|
static void
|
static void
|
find_moveable_store (rtx insn, int *regs_set_before, int *regs_set_after)
|
find_moveable_store (rtx insn, int *regs_set_before, int *regs_set_after)
|
{
|
{
|
struct ls_expr * ptr;
|
struct ls_expr * ptr;
|
rtx dest, set, tmp;
|
rtx dest, set, tmp;
|
int check_anticipatable, check_available;
|
int check_anticipatable, check_available;
|
basic_block bb = BLOCK_FOR_INSN (insn);
|
basic_block bb = BLOCK_FOR_INSN (insn);
|
|
|
set = single_set (insn);
|
set = single_set (insn);
|
if (!set)
|
if (!set)
|
return;
|
return;
|
|
|
dest = SET_DEST (set);
|
dest = SET_DEST (set);
|
|
|
if (! MEM_P (dest) || MEM_VOLATILE_P (dest)
|
if (! MEM_P (dest) || MEM_VOLATILE_P (dest)
|
|| GET_MODE (dest) == BLKmode)
|
|| GET_MODE (dest) == BLKmode)
|
return;
|
return;
|
|
|
if (side_effects_p (dest))
|
if (side_effects_p (dest))
|
return;
|
return;
|
|
|
/* If we are handling exceptions, we must be careful with memory references
|
/* If we are handling exceptions, we must be careful with memory references
|
that may trap. If we are not, the behavior is undefined, so we may just
|
that may trap. If we are not, the behavior is undefined, so we may just
|
continue. */
|
continue. */
|
if (flag_non_call_exceptions && may_trap_p (dest))
|
if (flag_non_call_exceptions && may_trap_p (dest))
|
return;
|
return;
|
|
|
/* Even if the destination cannot trap, the source may. In this case we'd
|
/* Even if the destination cannot trap, the source may. In this case we'd
|
need to handle updating the REG_EH_REGION note. */
|
need to handle updating the REG_EH_REGION note. */
|
if (find_reg_note (insn, REG_EH_REGION, NULL_RTX))
|
if (find_reg_note (insn, REG_EH_REGION, NULL_RTX))
|
return;
|
return;
|
|
|
/* Make sure that the SET_SRC of this store insns can be assigned to
|
/* Make sure that the SET_SRC of this store insns can be assigned to
|
a register, or we will fail later on in replace_store_insn, which
|
a register, or we will fail later on in replace_store_insn, which
|
assumes that we can do this. But sometimes the target machine has
|
assumes that we can do this. But sometimes the target machine has
|
oddities like MEM read-modify-write instruction. See for example
|
oddities like MEM read-modify-write instruction. See for example
|
PR24257. */
|
PR24257. */
|
if (!can_assign_to_reg_p (SET_SRC (set)))
|
if (!can_assign_to_reg_p (SET_SRC (set)))
|
return;
|
return;
|
|
|
ptr = ldst_entry (dest);
|
ptr = ldst_entry (dest);
|
if (!ptr->pattern_regs)
|
if (!ptr->pattern_regs)
|
ptr->pattern_regs = extract_mentioned_regs (dest);
|
ptr->pattern_regs = extract_mentioned_regs (dest);
|
|
|
/* Do not check for anticipatability if we either found one anticipatable
|
/* Do not check for anticipatability if we either found one anticipatable
|
store already, or tested for one and found out that it was killed. */
|
store already, or tested for one and found out that it was killed. */
|
check_anticipatable = 0;
|
check_anticipatable = 0;
|
if (!ANTIC_STORE_LIST (ptr))
|
if (!ANTIC_STORE_LIST (ptr))
|
check_anticipatable = 1;
|
check_anticipatable = 1;
|
else
|
else
|
{
|
{
|
tmp = XEXP (ANTIC_STORE_LIST (ptr), 0);
|
tmp = XEXP (ANTIC_STORE_LIST (ptr), 0);
|
if (tmp != NULL_RTX
|
if (tmp != NULL_RTX
|
&& BLOCK_FOR_INSN (tmp) != bb)
|
&& BLOCK_FOR_INSN (tmp) != bb)
|
check_anticipatable = 1;
|
check_anticipatable = 1;
|
}
|
}
|
if (check_anticipatable)
|
if (check_anticipatable)
|
{
|
{
|
if (store_killed_before (dest, ptr->pattern_regs, insn, bb, regs_set_before))
|
if (store_killed_before (dest, ptr->pattern_regs, insn, bb, regs_set_before))
|
tmp = NULL_RTX;
|
tmp = NULL_RTX;
|
else
|
else
|
tmp = insn;
|
tmp = insn;
|
ANTIC_STORE_LIST (ptr) = alloc_INSN_LIST (tmp,
|
ANTIC_STORE_LIST (ptr) = alloc_INSN_LIST (tmp,
|
ANTIC_STORE_LIST (ptr));
|
ANTIC_STORE_LIST (ptr));
|
}
|
}
|
|
|
/* It is not necessary to check whether store is available if we did
|
/* It is not necessary to check whether store is available if we did
|
it successfully before; if we failed before, do not bother to check
|
it successfully before; if we failed before, do not bother to check
|
until we reach the insn that caused us to fail. */
|
until we reach the insn that caused us to fail. */
|
check_available = 0;
|
check_available = 0;
|
if (!AVAIL_STORE_LIST (ptr))
|
if (!AVAIL_STORE_LIST (ptr))
|
check_available = 1;
|
check_available = 1;
|
else
|
else
|
{
|
{
|
tmp = XEXP (AVAIL_STORE_LIST (ptr), 0);
|
tmp = XEXP (AVAIL_STORE_LIST (ptr), 0);
|
if (BLOCK_FOR_INSN (tmp) != bb)
|
if (BLOCK_FOR_INSN (tmp) != bb)
|
check_available = 1;
|
check_available = 1;
|
}
|
}
|
if (check_available)
|
if (check_available)
|
{
|
{
|
/* Check that we have already reached the insn at that the check
|
/* Check that we have already reached the insn at that the check
|
failed last time. */
|
failed last time. */
|
if (LAST_AVAIL_CHECK_FAILURE (ptr))
|
if (LAST_AVAIL_CHECK_FAILURE (ptr))
|
{
|
{
|
for (tmp = BB_END (bb);
|
for (tmp = BB_END (bb);
|
tmp != insn && tmp != LAST_AVAIL_CHECK_FAILURE (ptr);
|
tmp != insn && tmp != LAST_AVAIL_CHECK_FAILURE (ptr);
|
tmp = PREV_INSN (tmp))
|
tmp = PREV_INSN (tmp))
|
continue;
|
continue;
|
if (tmp == insn)
|
if (tmp == insn)
|
check_available = 0;
|
check_available = 0;
|
}
|
}
|
else
|
else
|
check_available = store_killed_after (dest, ptr->pattern_regs, insn,
|
check_available = store_killed_after (dest, ptr->pattern_regs, insn,
|
bb, regs_set_after,
|
bb, regs_set_after,
|
&LAST_AVAIL_CHECK_FAILURE (ptr));
|
&LAST_AVAIL_CHECK_FAILURE (ptr));
|
}
|
}
|
if (!check_available)
|
if (!check_available)
|
AVAIL_STORE_LIST (ptr) = alloc_INSN_LIST (insn, AVAIL_STORE_LIST (ptr));
|
AVAIL_STORE_LIST (ptr) = alloc_INSN_LIST (insn, AVAIL_STORE_LIST (ptr));
|
}
|
}
|
|
|
/* Find available and anticipatable stores. */
|
/* Find available and anticipatable stores. */
|
|
|
static int
|
static int
|
compute_store_table (void)
|
compute_store_table (void)
|
{
|
{
|
int ret;
|
int ret;
|
basic_block bb;
|
basic_block bb;
|
unsigned regno;
|
unsigned regno;
|
rtx insn, pat, tmp;
|
rtx insn, pat, tmp;
|
int *last_set_in, *already_set;
|
int *last_set_in, *already_set;
|
struct ls_expr * ptr, **prev_next_ptr_ptr;
|
struct ls_expr * ptr, **prev_next_ptr_ptr;
|
|
|
max_gcse_regno = max_reg_num ();
|
max_gcse_regno = max_reg_num ();
|
|
|
reg_set_in_block = sbitmap_vector_alloc (last_basic_block,
|
reg_set_in_block = sbitmap_vector_alloc (last_basic_block,
|
max_gcse_regno);
|
max_gcse_regno);
|
sbitmap_vector_zero (reg_set_in_block, last_basic_block);
|
sbitmap_vector_zero (reg_set_in_block, last_basic_block);
|
pre_ldst_mems = 0;
|
pre_ldst_mems = 0;
|
pre_ldst_table = htab_create (13, pre_ldst_expr_hash,
|
pre_ldst_table = htab_create (13, pre_ldst_expr_hash,
|
pre_ldst_expr_eq, NULL);
|
pre_ldst_expr_eq, NULL);
|
last_set_in = XCNEWVEC (int, max_gcse_regno);
|
last_set_in = XCNEWVEC (int, max_gcse_regno);
|
already_set = XNEWVEC (int, max_gcse_regno);
|
already_set = XNEWVEC (int, max_gcse_regno);
|
|
|
/* Find all the stores we care about. */
|
/* Find all the stores we care about. */
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
/* First compute the registers set in this block. */
|
/* First compute the registers set in this block. */
|
regvec = last_set_in;
|
regvec = last_set_in;
|
|
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
{
|
{
|
if (! INSN_P (insn))
|
if (! INSN_P (insn))
|
continue;
|
continue;
|
|
|
if (CALL_P (insn))
|
if (CALL_P (insn))
|
{
|
{
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
|
{
|
{
|
last_set_in[regno] = INSN_UID (insn);
|
last_set_in[regno] = INSN_UID (insn);
|
SET_BIT (reg_set_in_block[bb->index], regno);
|
SET_BIT (reg_set_in_block[bb->index], regno);
|
}
|
}
|
}
|
}
|
|
|
pat = PATTERN (insn);
|
pat = PATTERN (insn);
|
compute_store_table_current_insn = insn;
|
compute_store_table_current_insn = insn;
|
note_stores (pat, reg_set_info, reg_set_in_block[bb->index]);
|
note_stores (pat, reg_set_info, reg_set_in_block[bb->index]);
|
}
|
}
|
|
|
/* Now find the stores. */
|
/* Now find the stores. */
|
memset (already_set, 0, sizeof (int) * max_gcse_regno);
|
memset (already_set, 0, sizeof (int) * max_gcse_regno);
|
regvec = already_set;
|
regvec = already_set;
|
FOR_BB_INSNS (bb, insn)
|
FOR_BB_INSNS (bb, insn)
|
{
|
{
|
if (! INSN_P (insn))
|
if (! INSN_P (insn))
|
continue;
|
continue;
|
|
|
if (CALL_P (insn))
|
if (CALL_P (insn))
|
{
|
{
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno))
|
already_set[regno] = 1;
|
already_set[regno] = 1;
|
}
|
}
|
|
|
pat = PATTERN (insn);
|
pat = PATTERN (insn);
|
note_stores (pat, reg_set_info, NULL);
|
note_stores (pat, reg_set_info, NULL);
|
|
|
/* Now that we've marked regs, look for stores. */
|
/* Now that we've marked regs, look for stores. */
|
find_moveable_store (insn, already_set, last_set_in);
|
find_moveable_store (insn, already_set, last_set_in);
|
|
|
/* Unmark regs that are no longer set. */
|
/* Unmark regs that are no longer set. */
|
compute_store_table_current_insn = insn;
|
compute_store_table_current_insn = insn;
|
note_stores (pat, reg_clear_last_set, last_set_in);
|
note_stores (pat, reg_clear_last_set, last_set_in);
|
if (CALL_P (insn))
|
if (CALL_P (insn))
|
{
|
{
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
for (regno = 0; regno < FIRST_PSEUDO_REGISTER; regno++)
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno)
|
if (TEST_HARD_REG_BIT (regs_invalidated_by_call, regno)
|
&& last_set_in[regno] == INSN_UID (insn))
|
&& last_set_in[regno] == INSN_UID (insn))
|
last_set_in[regno] = 0;
|
last_set_in[regno] = 0;
|
}
|
}
|
}
|
}
|
|
|
#ifdef ENABLE_CHECKING
|
#ifdef ENABLE_CHECKING
|
/* last_set_in should now be all-zero. */
|
/* last_set_in should now be all-zero. */
|
for (regno = 0; regno < max_gcse_regno; regno++)
|
for (regno = 0; regno < max_gcse_regno; regno++)
|
gcc_assert (!last_set_in[regno]);
|
gcc_assert (!last_set_in[regno]);
|
#endif
|
#endif
|
|
|
/* Clear temporary marks. */
|
/* Clear temporary marks. */
|
for (ptr = first_ls_expr (); ptr != NULL; ptr = next_ls_expr (ptr))
|
for (ptr = first_ls_expr (); ptr != NULL; ptr = next_ls_expr (ptr))
|
{
|
{
|
LAST_AVAIL_CHECK_FAILURE(ptr) = NULL_RTX;
|
LAST_AVAIL_CHECK_FAILURE(ptr) = NULL_RTX;
|
if (ANTIC_STORE_LIST (ptr)
|
if (ANTIC_STORE_LIST (ptr)
|
&& (tmp = XEXP (ANTIC_STORE_LIST (ptr), 0)) == NULL_RTX)
|
&& (tmp = XEXP (ANTIC_STORE_LIST (ptr), 0)) == NULL_RTX)
|
ANTIC_STORE_LIST (ptr) = XEXP (ANTIC_STORE_LIST (ptr), 1);
|
ANTIC_STORE_LIST (ptr) = XEXP (ANTIC_STORE_LIST (ptr), 1);
|
}
|
}
|
}
|
}
|
|
|
/* Remove the stores that are not available anywhere, as there will
|
/* Remove the stores that are not available anywhere, as there will
|
be no opportunity to optimize them. */
|
be no opportunity to optimize them. */
|
for (ptr = pre_ldst_mems, prev_next_ptr_ptr = &pre_ldst_mems;
|
for (ptr = pre_ldst_mems, prev_next_ptr_ptr = &pre_ldst_mems;
|
ptr != NULL;
|
ptr != NULL;
|
ptr = *prev_next_ptr_ptr)
|
ptr = *prev_next_ptr_ptr)
|
{
|
{
|
if (!AVAIL_STORE_LIST (ptr))
|
if (!AVAIL_STORE_LIST (ptr))
|
{
|
{
|
*prev_next_ptr_ptr = ptr->next;
|
*prev_next_ptr_ptr = ptr->next;
|
htab_remove_elt_with_hash (pre_ldst_table, ptr, ptr->hash_index);
|
htab_remove_elt_with_hash (pre_ldst_table, ptr, ptr->hash_index);
|
free_ldst_entry (ptr);
|
free_ldst_entry (ptr);
|
}
|
}
|
else
|
else
|
prev_next_ptr_ptr = &ptr->next;
|
prev_next_ptr_ptr = &ptr->next;
|
}
|
}
|
|
|
ret = enumerate_ldsts ();
|
ret = enumerate_ldsts ();
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "ST_avail and ST_antic (shown under loads..)\n");
|
fprintf (dump_file, "ST_avail and ST_antic (shown under loads..)\n");
|
print_ldst_list (dump_file);
|
print_ldst_list (dump_file);
|
}
|
}
|
|
|
free (last_set_in);
|
free (last_set_in);
|
free (already_set);
|
free (already_set);
|
return ret;
|
return ret;
|
}
|
}
|
|
|
/* Check to see if the load X is aliased with STORE_PATTERN.
|
/* Check to see if the load X is aliased with STORE_PATTERN.
|
AFTER is true if we are checking the case when STORE_PATTERN occurs
|
AFTER is true if we are checking the case when STORE_PATTERN occurs
|
after the X. */
|
after the X. */
|
|
|
static bool
|
static bool
|
load_kills_store (rtx x, rtx store_pattern, int after)
|
load_kills_store (rtx x, rtx store_pattern, int after)
|
{
|
{
|
if (after)
|
if (after)
|
return anti_dependence (x, store_pattern);
|
return anti_dependence (x, store_pattern);
|
else
|
else
|
return true_dependence (store_pattern, GET_MODE (store_pattern), x,
|
return true_dependence (store_pattern, GET_MODE (store_pattern), x,
|
rtx_addr_varies_p);
|
rtx_addr_varies_p);
|
}
|
}
|
|
|
/* Go through the entire insn X, looking for any loads which might alias
|
/* Go through the entire insn X, looking for any loads which might alias
|
STORE_PATTERN. Return true if found.
|
STORE_PATTERN. Return true if found.
|
AFTER is true if we are checking the case when STORE_PATTERN occurs
|
AFTER is true if we are checking the case when STORE_PATTERN occurs
|
after the insn X. */
|
after the insn X. */
|
|
|
static bool
|
static bool
|
find_loads (rtx x, rtx store_pattern, int after)
|
find_loads (rtx x, rtx store_pattern, int after)
|
{
|
{
|
const char * fmt;
|
const char * fmt;
|
int i, j;
|
int i, j;
|
int ret = false;
|
int ret = false;
|
|
|
if (!x)
|
if (!x)
|
return false;
|
return false;
|
|
|
if (GET_CODE (x) == SET)
|
if (GET_CODE (x) == SET)
|
x = SET_SRC (x);
|
x = SET_SRC (x);
|
|
|
if (MEM_P (x))
|
if (MEM_P (x))
|
{
|
{
|
if (load_kills_store (x, store_pattern, after))
|
if (load_kills_store (x, store_pattern, after))
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Recursively process the insn. */
|
/* Recursively process the insn. */
|
fmt = GET_RTX_FORMAT (GET_CODE (x));
|
fmt = GET_RTX_FORMAT (GET_CODE (x));
|
|
|
for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0 && !ret; i--)
|
for (i = GET_RTX_LENGTH (GET_CODE (x)) - 1; i >= 0 && !ret; i--)
|
{
|
{
|
if (fmt[i] == 'e')
|
if (fmt[i] == 'e')
|
ret |= find_loads (XEXP (x, i), store_pattern, after);
|
ret |= find_loads (XEXP (x, i), store_pattern, after);
|
else if (fmt[i] == 'E')
|
else if (fmt[i] == 'E')
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
for (j = XVECLEN (x, i) - 1; j >= 0; j--)
|
ret |= find_loads (XVECEXP (x, i, j), store_pattern, after);
|
ret |= find_loads (XVECEXP (x, i, j), store_pattern, after);
|
}
|
}
|
return ret;
|
return ret;
|
}
|
}
|
|
|
/* Check if INSN kills the store pattern X (is aliased with it).
|
/* Check if INSN kills the store pattern X (is aliased with it).
|
AFTER is true if we are checking the case when store X occurs
|
AFTER is true if we are checking the case when store X occurs
|
after the insn. Return true if it does. */
|
after the insn. Return true if it does. */
|
|
|
static bool
|
static bool
|
store_killed_in_insn (rtx x, rtx x_regs, rtx insn, int after)
|
store_killed_in_insn (rtx x, rtx x_regs, rtx insn, int after)
|
{
|
{
|
rtx reg, base, note;
|
rtx reg, base, note;
|
|
|
if (!INSN_P (insn))
|
if (!INSN_P (insn))
|
return false;
|
return false;
|
|
|
if (CALL_P (insn))
|
if (CALL_P (insn))
|
{
|
{
|
/* A normal or pure call might read from pattern,
|
/* A normal or pure call might read from pattern,
|
but a const call will not. */
|
but a const call will not. */
|
if (! CONST_OR_PURE_CALL_P (insn) || pure_call_p (insn))
|
if (! CONST_OR_PURE_CALL_P (insn) || pure_call_p (insn))
|
return true;
|
return true;
|
|
|
/* But even a const call reads its parameters. Check whether the
|
/* But even a const call reads its parameters. Check whether the
|
base of some of registers used in mem is stack pointer. */
|
base of some of registers used in mem is stack pointer. */
|
for (reg = x_regs; reg; reg = XEXP (reg, 1))
|
for (reg = x_regs; reg; reg = XEXP (reg, 1))
|
{
|
{
|
base = find_base_term (XEXP (reg, 0));
|
base = find_base_term (XEXP (reg, 0));
|
if (!base
|
if (!base
|
|| (GET_CODE (base) == ADDRESS
|
|| (GET_CODE (base) == ADDRESS
|
&& GET_MODE (base) == Pmode
|
&& GET_MODE (base) == Pmode
|
&& XEXP (base, 0) == stack_pointer_rtx))
|
&& XEXP (base, 0) == stack_pointer_rtx))
|
return true;
|
return true;
|
}
|
}
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
if (GET_CODE (PATTERN (insn)) == SET)
|
if (GET_CODE (PATTERN (insn)) == SET)
|
{
|
{
|
rtx pat = PATTERN (insn);
|
rtx pat = PATTERN (insn);
|
rtx dest = SET_DEST (pat);
|
rtx dest = SET_DEST (pat);
|
|
|
if (GET_CODE (dest) == ZERO_EXTRACT)
|
if (GET_CODE (dest) == ZERO_EXTRACT)
|
dest = XEXP (dest, 0);
|
dest = XEXP (dest, 0);
|
|
|
/* Check for memory stores to aliased objects. */
|
/* Check for memory stores to aliased objects. */
|
if (MEM_P (dest)
|
if (MEM_P (dest)
|
&& !expr_equiv_p (dest, x))
|
&& !expr_equiv_p (dest, x))
|
{
|
{
|
if (after)
|
if (after)
|
{
|
{
|
if (output_dependence (dest, x))
|
if (output_dependence (dest, x))
|
return true;
|
return true;
|
}
|
}
|
else
|
else
|
{
|
{
|
if (output_dependence (x, dest))
|
if (output_dependence (x, dest))
|
return true;
|
return true;
|
}
|
}
|
}
|
}
|
if (find_loads (SET_SRC (pat), x, after))
|
if (find_loads (SET_SRC (pat), x, after))
|
return true;
|
return true;
|
}
|
}
|
else if (find_loads (PATTERN (insn), x, after))
|
else if (find_loads (PATTERN (insn), x, after))
|
return true;
|
return true;
|
|
|
/* If this insn has a REG_EQUAL or REG_EQUIV note referencing a memory
|
/* If this insn has a REG_EQUAL or REG_EQUIV note referencing a memory
|
location aliased with X, then this insn kills X. */
|
location aliased with X, then this insn kills X. */
|
note = find_reg_equal_equiv_note (insn);
|
note = find_reg_equal_equiv_note (insn);
|
if (! note)
|
if (! note)
|
return false;
|
return false;
|
note = XEXP (note, 0);
|
note = XEXP (note, 0);
|
|
|
/* However, if the note represents a must alias rather than a may
|
/* However, if the note represents a must alias rather than a may
|
alias relationship, then it does not kill X. */
|
alias relationship, then it does not kill X. */
|
if (expr_equiv_p (note, x))
|
if (expr_equiv_p (note, x))
|
return false;
|
return false;
|
|
|
/* See if there are any aliased loads in the note. */
|
/* See if there are any aliased loads in the note. */
|
return find_loads (note, x, after);
|
return find_loads (note, x, after);
|
}
|
}
|
|
|
/* Returns true if the expression X is loaded or clobbered on or after INSN
|
/* Returns true if the expression X is loaded or clobbered on or after INSN
|
within basic block BB. REGS_SET_AFTER is bitmap of registers set in
|
within basic block BB. REGS_SET_AFTER is bitmap of registers set in
|
or after the insn. X_REGS is list of registers mentioned in X. If the store
|
or after the insn. X_REGS is list of registers mentioned in X. If the store
|
is killed, return the last insn in that it occurs in FAIL_INSN. */
|
is killed, return the last insn in that it occurs in FAIL_INSN. */
|
|
|
static bool
|
static bool
|
store_killed_after (rtx x, rtx x_regs, rtx insn, basic_block bb,
|
store_killed_after (rtx x, rtx x_regs, rtx insn, basic_block bb,
|
int *regs_set_after, rtx *fail_insn)
|
int *regs_set_after, rtx *fail_insn)
|
{
|
{
|
rtx last = BB_END (bb), act;
|
rtx last = BB_END (bb), act;
|
|
|
if (!store_ops_ok (x_regs, regs_set_after))
|
if (!store_ops_ok (x_regs, regs_set_after))
|
{
|
{
|
/* We do not know where it will happen. */
|
/* We do not know where it will happen. */
|
if (fail_insn)
|
if (fail_insn)
|
*fail_insn = NULL_RTX;
|
*fail_insn = NULL_RTX;
|
return true;
|
return true;
|
}
|
}
|
|
|
/* Scan from the end, so that fail_insn is determined correctly. */
|
/* Scan from the end, so that fail_insn is determined correctly. */
|
for (act = last; act != PREV_INSN (insn); act = PREV_INSN (act))
|
for (act = last; act != PREV_INSN (insn); act = PREV_INSN (act))
|
if (store_killed_in_insn (x, x_regs, act, false))
|
if (store_killed_in_insn (x, x_regs, act, false))
|
{
|
{
|
if (fail_insn)
|
if (fail_insn)
|
*fail_insn = act;
|
*fail_insn = act;
|
return true;
|
return true;
|
}
|
}
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Returns true if the expression X is loaded or clobbered on or before INSN
|
/* Returns true if the expression X is loaded or clobbered on or before INSN
|
within basic block BB. X_REGS is list of registers mentioned in X.
|
within basic block BB. X_REGS is list of registers mentioned in X.
|
REGS_SET_BEFORE is bitmap of registers set before or in this insn. */
|
REGS_SET_BEFORE is bitmap of registers set before or in this insn. */
|
static bool
|
static bool
|
store_killed_before (rtx x, rtx x_regs, rtx insn, basic_block bb,
|
store_killed_before (rtx x, rtx x_regs, rtx insn, basic_block bb,
|
int *regs_set_before)
|
int *regs_set_before)
|
{
|
{
|
rtx first = BB_HEAD (bb);
|
rtx first = BB_HEAD (bb);
|
|
|
if (!store_ops_ok (x_regs, regs_set_before))
|
if (!store_ops_ok (x_regs, regs_set_before))
|
return true;
|
return true;
|
|
|
for ( ; insn != PREV_INSN (first); insn = PREV_INSN (insn))
|
for ( ; insn != PREV_INSN (first); insn = PREV_INSN (insn))
|
if (store_killed_in_insn (x, x_regs, insn, true))
|
if (store_killed_in_insn (x, x_regs, insn, true))
|
return true;
|
return true;
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
/* Fill in available, anticipatable, transparent and kill vectors in
|
/* Fill in available, anticipatable, transparent and kill vectors in
|
STORE_DATA, based on lists of available and anticipatable stores. */
|
STORE_DATA, based on lists of available and anticipatable stores. */
|
static void
|
static void
|
build_store_vectors (void)
|
build_store_vectors (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
int *regs_set_in_block;
|
int *regs_set_in_block;
|
rtx insn, st;
|
rtx insn, st;
|
struct ls_expr * ptr;
|
struct ls_expr * ptr;
|
unsigned regno;
|
unsigned regno;
|
|
|
/* Build the gen_vector. This is any store in the table which is not killed
|
/* Build the gen_vector. This is any store in the table which is not killed
|
by aliasing later in its block. */
|
by aliasing later in its block. */
|
ae_gen = sbitmap_vector_alloc (last_basic_block, num_stores);
|
ae_gen = sbitmap_vector_alloc (last_basic_block, num_stores);
|
sbitmap_vector_zero (ae_gen, last_basic_block);
|
sbitmap_vector_zero (ae_gen, last_basic_block);
|
|
|
st_antloc = sbitmap_vector_alloc (last_basic_block, num_stores);
|
st_antloc = sbitmap_vector_alloc (last_basic_block, num_stores);
|
sbitmap_vector_zero (st_antloc, last_basic_block);
|
sbitmap_vector_zero (st_antloc, last_basic_block);
|
|
|
for (ptr = first_ls_expr (); ptr != NULL; ptr = next_ls_expr (ptr))
|
for (ptr = first_ls_expr (); ptr != NULL; ptr = next_ls_expr (ptr))
|
{
|
{
|
for (st = AVAIL_STORE_LIST (ptr); st != NULL; st = XEXP (st, 1))
|
for (st = AVAIL_STORE_LIST (ptr); st != NULL; st = XEXP (st, 1))
|
{
|
{
|
insn = XEXP (st, 0);
|
insn = XEXP (st, 0);
|
bb = BLOCK_FOR_INSN (insn);
|
bb = BLOCK_FOR_INSN (insn);
|
|
|
/* If we've already seen an available expression in this block,
|
/* If we've already seen an available expression in this block,
|
we can delete this one (It occurs earlier in the block). We'll
|
we can delete this one (It occurs earlier in the block). We'll
|
copy the SRC expression to an unused register in case there
|
copy the SRC expression to an unused register in case there
|
are any side effects. */
|
are any side effects. */
|
if (TEST_BIT (ae_gen[bb->index], ptr->index))
|
if (TEST_BIT (ae_gen[bb->index], ptr->index))
|
{
|
{
|
rtx r = gen_reg_rtx (GET_MODE (ptr->pattern));
|
rtx r = gen_reg_rtx (GET_MODE (ptr->pattern));
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "Removing redundant store:\n");
|
fprintf (dump_file, "Removing redundant store:\n");
|
replace_store_insn (r, XEXP (st, 0), bb, ptr);
|
replace_store_insn (r, XEXP (st, 0), bb, ptr);
|
continue;
|
continue;
|
}
|
}
|
SET_BIT (ae_gen[bb->index], ptr->index);
|
SET_BIT (ae_gen[bb->index], ptr->index);
|
}
|
}
|
|
|
for (st = ANTIC_STORE_LIST (ptr); st != NULL; st = XEXP (st, 1))
|
for (st = ANTIC_STORE_LIST (ptr); st != NULL; st = XEXP (st, 1))
|
{
|
{
|
insn = XEXP (st, 0);
|
insn = XEXP (st, 0);
|
bb = BLOCK_FOR_INSN (insn);
|
bb = BLOCK_FOR_INSN (insn);
|
SET_BIT (st_antloc[bb->index], ptr->index);
|
SET_BIT (st_antloc[bb->index], ptr->index);
|
}
|
}
|
}
|
}
|
|
|
ae_kill = sbitmap_vector_alloc (last_basic_block, num_stores);
|
ae_kill = sbitmap_vector_alloc (last_basic_block, num_stores);
|
sbitmap_vector_zero (ae_kill, last_basic_block);
|
sbitmap_vector_zero (ae_kill, last_basic_block);
|
|
|
transp = sbitmap_vector_alloc (last_basic_block, num_stores);
|
transp = sbitmap_vector_alloc (last_basic_block, num_stores);
|
sbitmap_vector_zero (transp, last_basic_block);
|
sbitmap_vector_zero (transp, last_basic_block);
|
regs_set_in_block = XNEWVEC (int, max_gcse_regno);
|
regs_set_in_block = XNEWVEC (int, max_gcse_regno);
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
{
|
{
|
for (regno = 0; regno < max_gcse_regno; regno++)
|
for (regno = 0; regno < max_gcse_regno; regno++)
|
regs_set_in_block[regno] = TEST_BIT (reg_set_in_block[bb->index], regno);
|
regs_set_in_block[regno] = TEST_BIT (reg_set_in_block[bb->index], regno);
|
|
|
for (ptr = first_ls_expr (); ptr != NULL; ptr = next_ls_expr (ptr))
|
for (ptr = first_ls_expr (); ptr != NULL; ptr = next_ls_expr (ptr))
|
{
|
{
|
if (store_killed_after (ptr->pattern, ptr->pattern_regs, BB_HEAD (bb),
|
if (store_killed_after (ptr->pattern, ptr->pattern_regs, BB_HEAD (bb),
|
bb, regs_set_in_block, NULL))
|
bb, regs_set_in_block, NULL))
|
{
|
{
|
/* It should not be necessary to consider the expression
|
/* It should not be necessary to consider the expression
|
killed if it is both anticipatable and available. */
|
killed if it is both anticipatable and available. */
|
if (!TEST_BIT (st_antloc[bb->index], ptr->index)
|
if (!TEST_BIT (st_antloc[bb->index], ptr->index)
|
|| !TEST_BIT (ae_gen[bb->index], ptr->index))
|
|| !TEST_BIT (ae_gen[bb->index], ptr->index))
|
SET_BIT (ae_kill[bb->index], ptr->index);
|
SET_BIT (ae_kill[bb->index], ptr->index);
|
}
|
}
|
else
|
else
|
SET_BIT (transp[bb->index], ptr->index);
|
SET_BIT (transp[bb->index], ptr->index);
|
}
|
}
|
}
|
}
|
|
|
free (regs_set_in_block);
|
free (regs_set_in_block);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
dump_sbitmap_vector (dump_file, "st_antloc", "", st_antloc, last_basic_block);
|
dump_sbitmap_vector (dump_file, "st_antloc", "", st_antloc, last_basic_block);
|
dump_sbitmap_vector (dump_file, "st_kill", "", ae_kill, last_basic_block);
|
dump_sbitmap_vector (dump_file, "st_kill", "", ae_kill, last_basic_block);
|
dump_sbitmap_vector (dump_file, "Transpt", "", transp, last_basic_block);
|
dump_sbitmap_vector (dump_file, "Transpt", "", transp, last_basic_block);
|
dump_sbitmap_vector (dump_file, "st_avloc", "", ae_gen, last_basic_block);
|
dump_sbitmap_vector (dump_file, "st_avloc", "", ae_gen, last_basic_block);
|
}
|
}
|
}
|
}
|
|
|
/* Insert an instruction at the beginning of a basic block, and update
|
/* Insert an instruction at the beginning of a basic block, and update
|
the BB_HEAD if needed. */
|
the BB_HEAD if needed. */
|
|
|
static void
|
static void
|
insert_insn_start_bb (rtx insn, basic_block bb)
|
insert_insn_start_bb (rtx insn, basic_block bb)
|
{
|
{
|
/* Insert at start of successor block. */
|
/* Insert at start of successor block. */
|
rtx prev = PREV_INSN (BB_HEAD (bb));
|
rtx prev = PREV_INSN (BB_HEAD (bb));
|
rtx before = BB_HEAD (bb);
|
rtx before = BB_HEAD (bb);
|
while (before != 0)
|
while (before != 0)
|
{
|
{
|
if (! LABEL_P (before)
|
if (! LABEL_P (before)
|
&& (! NOTE_P (before)
|
&& (! NOTE_P (before)
|
|| NOTE_LINE_NUMBER (before) != NOTE_INSN_BASIC_BLOCK))
|
|| NOTE_LINE_NUMBER (before) != NOTE_INSN_BASIC_BLOCK))
|
break;
|
break;
|
prev = before;
|
prev = before;
|
if (prev == BB_END (bb))
|
if (prev == BB_END (bb))
|
break;
|
break;
|
before = NEXT_INSN (before);
|
before = NEXT_INSN (before);
|
}
|
}
|
|
|
insn = emit_insn_after_noloc (insn, prev);
|
insn = emit_insn_after_noloc (insn, prev);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "STORE_MOTION insert store at start of BB %d:\n",
|
fprintf (dump_file, "STORE_MOTION insert store at start of BB %d:\n",
|
bb->index);
|
bb->index);
|
print_inline_rtx (dump_file, insn, 6);
|
print_inline_rtx (dump_file, insn, 6);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
}
|
}
|
|
|
/* This routine will insert a store on an edge. EXPR is the ldst entry for
|
/* This routine will insert a store on an edge. EXPR is the ldst entry for
|
the memory reference, and E is the edge to insert it on. Returns nonzero
|
the memory reference, and E is the edge to insert it on. Returns nonzero
|
if an edge insertion was performed. */
|
if an edge insertion was performed. */
|
|
|
static int
|
static int
|
insert_store (struct ls_expr * expr, edge e)
|
insert_store (struct ls_expr * expr, edge e)
|
{
|
{
|
rtx reg, insn;
|
rtx reg, insn;
|
basic_block bb;
|
basic_block bb;
|
edge tmp;
|
edge tmp;
|
edge_iterator ei;
|
edge_iterator ei;
|
|
|
/* We did all the deleted before this insert, so if we didn't delete a
|
/* We did all the deleted before this insert, so if we didn't delete a
|
store, then we haven't set the reaching reg yet either. */
|
store, then we haven't set the reaching reg yet either. */
|
if (expr->reaching_reg == NULL_RTX)
|
if (expr->reaching_reg == NULL_RTX)
|
return 0;
|
return 0;
|
|
|
if (e->flags & EDGE_FAKE)
|
if (e->flags & EDGE_FAKE)
|
return 0;
|
return 0;
|
|
|
reg = expr->reaching_reg;
|
reg = expr->reaching_reg;
|
insn = gen_move_insn (copy_rtx (expr->pattern), reg);
|
insn = gen_move_insn (copy_rtx (expr->pattern), reg);
|
|
|
/* If we are inserting this expression on ALL predecessor edges of a BB,
|
/* If we are inserting this expression on ALL predecessor edges of a BB,
|
insert it at the start of the BB, and reset the insert bits on the other
|
insert it at the start of the BB, and reset the insert bits on the other
|
edges so we don't try to insert it on the other edges. */
|
edges so we don't try to insert it on the other edges. */
|
bb = e->dest;
|
bb = e->dest;
|
FOR_EACH_EDGE (tmp, ei, e->dest->preds)
|
FOR_EACH_EDGE (tmp, ei, e->dest->preds)
|
if (!(tmp->flags & EDGE_FAKE))
|
if (!(tmp->flags & EDGE_FAKE))
|
{
|
{
|
int index = EDGE_INDEX (edge_list, tmp->src, tmp->dest);
|
int index = EDGE_INDEX (edge_list, tmp->src, tmp->dest);
|
|
|
gcc_assert (index != EDGE_INDEX_NO_EDGE);
|
gcc_assert (index != EDGE_INDEX_NO_EDGE);
|
if (! TEST_BIT (pre_insert_map[index], expr->index))
|
if (! TEST_BIT (pre_insert_map[index], expr->index))
|
break;
|
break;
|
}
|
}
|
|
|
/* If tmp is NULL, we found an insertion on every edge, blank the
|
/* If tmp is NULL, we found an insertion on every edge, blank the
|
insertion vector for these edges, and insert at the start of the BB. */
|
insertion vector for these edges, and insert at the start of the BB. */
|
if (!tmp && bb != EXIT_BLOCK_PTR)
|
if (!tmp && bb != EXIT_BLOCK_PTR)
|
{
|
{
|
FOR_EACH_EDGE (tmp, ei, e->dest->preds)
|
FOR_EACH_EDGE (tmp, ei, e->dest->preds)
|
{
|
{
|
int index = EDGE_INDEX (edge_list, tmp->src, tmp->dest);
|
int index = EDGE_INDEX (edge_list, tmp->src, tmp->dest);
|
RESET_BIT (pre_insert_map[index], expr->index);
|
RESET_BIT (pre_insert_map[index], expr->index);
|
}
|
}
|
insert_insn_start_bb (insn, bb);
|
insert_insn_start_bb (insn, bb);
|
return 0;
|
return 0;
|
}
|
}
|
|
|
/* We can't put stores in the front of blocks pointed to by abnormal
|
/* We can't put stores in the front of blocks pointed to by abnormal
|
edges since that may put a store where one didn't used to be. */
|
edges since that may put a store where one didn't used to be. */
|
gcc_assert (!(e->flags & EDGE_ABNORMAL));
|
gcc_assert (!(e->flags & EDGE_ABNORMAL));
|
|
|
insert_insn_on_edge (insn, e);
|
insert_insn_on_edge (insn, e);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "STORE_MOTION insert insn on edge (%d, %d):\n",
|
fprintf (dump_file, "STORE_MOTION insert insn on edge (%d, %d):\n",
|
e->src->index, e->dest->index);
|
e->src->index, e->dest->index);
|
print_inline_rtx (dump_file, insn, 6);
|
print_inline_rtx (dump_file, insn, 6);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
return 1;
|
return 1;
|
}
|
}
|
|
|
/* Remove any REG_EQUAL or REG_EQUIV notes containing a reference to the
|
/* Remove any REG_EQUAL or REG_EQUIV notes containing a reference to the
|
memory location in SMEXPR set in basic block BB.
|
memory location in SMEXPR set in basic block BB.
|
|
|
This could be rather expensive. */
|
This could be rather expensive. */
|
|
|
static void
|
static void
|
remove_reachable_equiv_notes (basic_block bb, struct ls_expr *smexpr)
|
remove_reachable_equiv_notes (basic_block bb, struct ls_expr *smexpr)
|
{
|
{
|
edge_iterator *stack, ei;
|
edge_iterator *stack, ei;
|
int sp;
|
int sp;
|
edge act;
|
edge act;
|
sbitmap visited = sbitmap_alloc (last_basic_block);
|
sbitmap visited = sbitmap_alloc (last_basic_block);
|
rtx last, insn, note;
|
rtx last, insn, note;
|
rtx mem = smexpr->pattern;
|
rtx mem = smexpr->pattern;
|
|
|
stack = XNEWVEC (edge_iterator, n_basic_blocks);
|
stack = XNEWVEC (edge_iterator, n_basic_blocks);
|
sp = 0;
|
sp = 0;
|
ei = ei_start (bb->succs);
|
ei = ei_start (bb->succs);
|
|
|
sbitmap_zero (visited);
|
sbitmap_zero (visited);
|
|
|
act = (EDGE_COUNT (ei_container (ei)) > 0 ? EDGE_I (ei_container (ei), 0) : NULL);
|
act = (EDGE_COUNT (ei_container (ei)) > 0 ? EDGE_I (ei_container (ei), 0) : NULL);
|
while (1)
|
while (1)
|
{
|
{
|
if (!act)
|
if (!act)
|
{
|
{
|
if (!sp)
|
if (!sp)
|
{
|
{
|
free (stack);
|
free (stack);
|
sbitmap_free (visited);
|
sbitmap_free (visited);
|
return;
|
return;
|
}
|
}
|
act = ei_edge (stack[--sp]);
|
act = ei_edge (stack[--sp]);
|
}
|
}
|
bb = act->dest;
|
bb = act->dest;
|
|
|
if (bb == EXIT_BLOCK_PTR
|
if (bb == EXIT_BLOCK_PTR
|
|| TEST_BIT (visited, bb->index))
|
|| TEST_BIT (visited, bb->index))
|
{
|
{
|
if (!ei_end_p (ei))
|
if (!ei_end_p (ei))
|
ei_next (&ei);
|
ei_next (&ei);
|
act = (! ei_end_p (ei)) ? ei_edge (ei) : NULL;
|
act = (! ei_end_p (ei)) ? ei_edge (ei) : NULL;
|
continue;
|
continue;
|
}
|
}
|
SET_BIT (visited, bb->index);
|
SET_BIT (visited, bb->index);
|
|
|
if (TEST_BIT (st_antloc[bb->index], smexpr->index))
|
if (TEST_BIT (st_antloc[bb->index], smexpr->index))
|
{
|
{
|
for (last = ANTIC_STORE_LIST (smexpr);
|
for (last = ANTIC_STORE_LIST (smexpr);
|
BLOCK_FOR_INSN (XEXP (last, 0)) != bb;
|
BLOCK_FOR_INSN (XEXP (last, 0)) != bb;
|
last = XEXP (last, 1))
|
last = XEXP (last, 1))
|
continue;
|
continue;
|
last = XEXP (last, 0);
|
last = XEXP (last, 0);
|
}
|
}
|
else
|
else
|
last = NEXT_INSN (BB_END (bb));
|
last = NEXT_INSN (BB_END (bb));
|
|
|
for (insn = BB_HEAD (bb); insn != last; insn = NEXT_INSN (insn))
|
for (insn = BB_HEAD (bb); insn != last; insn = NEXT_INSN (insn))
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
{
|
{
|
note = find_reg_equal_equiv_note (insn);
|
note = find_reg_equal_equiv_note (insn);
|
if (!note || !expr_equiv_p (XEXP (note, 0), mem))
|
if (!note || !expr_equiv_p (XEXP (note, 0), mem))
|
continue;
|
continue;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "STORE_MOTION drop REG_EQUAL note at insn %d:\n",
|
fprintf (dump_file, "STORE_MOTION drop REG_EQUAL note at insn %d:\n",
|
INSN_UID (insn));
|
INSN_UID (insn));
|
remove_note (insn, note);
|
remove_note (insn, note);
|
}
|
}
|
|
|
if (!ei_end_p (ei))
|
if (!ei_end_p (ei))
|
ei_next (&ei);
|
ei_next (&ei);
|
act = (! ei_end_p (ei)) ? ei_edge (ei) : NULL;
|
act = (! ei_end_p (ei)) ? ei_edge (ei) : NULL;
|
|
|
if (EDGE_COUNT (bb->succs) > 0)
|
if (EDGE_COUNT (bb->succs) > 0)
|
{
|
{
|
if (act)
|
if (act)
|
stack[sp++] = ei;
|
stack[sp++] = ei;
|
ei = ei_start (bb->succs);
|
ei = ei_start (bb->succs);
|
act = (EDGE_COUNT (ei_container (ei)) > 0 ? EDGE_I (ei_container (ei), 0) : NULL);
|
act = (EDGE_COUNT (ei_container (ei)) > 0 ? EDGE_I (ei_container (ei), 0) : NULL);
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* This routine will replace a store with a SET to a specified register. */
|
/* This routine will replace a store with a SET to a specified register. */
|
|
|
static void
|
static void
|
replace_store_insn (rtx reg, rtx del, basic_block bb, struct ls_expr *smexpr)
|
replace_store_insn (rtx reg, rtx del, basic_block bb, struct ls_expr *smexpr)
|
{
|
{
|
rtx insn, mem, note, set, ptr, pair;
|
rtx insn, mem, note, set, ptr, pair;
|
|
|
mem = smexpr->pattern;
|
mem = smexpr->pattern;
|
insn = gen_move_insn (reg, SET_SRC (single_set (del)));
|
insn = gen_move_insn (reg, SET_SRC (single_set (del)));
|
insn = emit_insn_after (insn, del);
|
insn = emit_insn_after (insn, del);
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"STORE_MOTION delete insn in BB %d:\n ", bb->index);
|
"STORE_MOTION delete insn in BB %d:\n ", bb->index);
|
print_inline_rtx (dump_file, del, 6);
|
print_inline_rtx (dump_file, del, 6);
|
fprintf (dump_file, "\nSTORE MOTION replaced with insn:\n ");
|
fprintf (dump_file, "\nSTORE MOTION replaced with insn:\n ");
|
print_inline_rtx (dump_file, insn, 6);
|
print_inline_rtx (dump_file, insn, 6);
|
fprintf (dump_file, "\n");
|
fprintf (dump_file, "\n");
|
}
|
}
|
|
|
for (ptr = ANTIC_STORE_LIST (smexpr); ptr; ptr = XEXP (ptr, 1))
|
for (ptr = ANTIC_STORE_LIST (smexpr); ptr; ptr = XEXP (ptr, 1))
|
if (XEXP (ptr, 0) == del)
|
if (XEXP (ptr, 0) == del)
|
{
|
{
|
XEXP (ptr, 0) = insn;
|
XEXP (ptr, 0) = insn;
|
break;
|
break;
|
}
|
}
|
|
|
/* Move the notes from the deleted insn to its replacement, and patch
|
/* Move the notes from the deleted insn to its replacement, and patch
|
up the LIBCALL notes. */
|
up the LIBCALL notes. */
|
REG_NOTES (insn) = REG_NOTES (del);
|
REG_NOTES (insn) = REG_NOTES (del);
|
|
|
note = find_reg_note (insn, REG_RETVAL, NULL_RTX);
|
note = find_reg_note (insn, REG_RETVAL, NULL_RTX);
|
if (note)
|
if (note)
|
{
|
{
|
pair = XEXP (note, 0);
|
pair = XEXP (note, 0);
|
note = find_reg_note (pair, REG_LIBCALL, NULL_RTX);
|
note = find_reg_note (pair, REG_LIBCALL, NULL_RTX);
|
XEXP (note, 0) = insn;
|
XEXP (note, 0) = insn;
|
}
|
}
|
note = find_reg_note (insn, REG_LIBCALL, NULL_RTX);
|
note = find_reg_note (insn, REG_LIBCALL, NULL_RTX);
|
if (note)
|
if (note)
|
{
|
{
|
pair = XEXP (note, 0);
|
pair = XEXP (note, 0);
|
note = find_reg_note (pair, REG_RETVAL, NULL_RTX);
|
note = find_reg_note (pair, REG_RETVAL, NULL_RTX);
|
XEXP (note, 0) = insn;
|
XEXP (note, 0) = insn;
|
}
|
}
|
|
|
delete_insn (del);
|
delete_insn (del);
|
|
|
/* Now we must handle REG_EQUAL notes whose contents is equal to the mem;
|
/* Now we must handle REG_EQUAL notes whose contents is equal to the mem;
|
they are no longer accurate provided that they are reached by this
|
they are no longer accurate provided that they are reached by this
|
definition, so drop them. */
|
definition, so drop them. */
|
for (; insn != NEXT_INSN (BB_END (bb)); insn = NEXT_INSN (insn))
|
for (; insn != NEXT_INSN (BB_END (bb)); insn = NEXT_INSN (insn))
|
if (INSN_P (insn))
|
if (INSN_P (insn))
|
{
|
{
|
set = single_set (insn);
|
set = single_set (insn);
|
if (!set)
|
if (!set)
|
continue;
|
continue;
|
if (expr_equiv_p (SET_DEST (set), mem))
|
if (expr_equiv_p (SET_DEST (set), mem))
|
return;
|
return;
|
note = find_reg_equal_equiv_note (insn);
|
note = find_reg_equal_equiv_note (insn);
|
if (!note || !expr_equiv_p (XEXP (note, 0), mem))
|
if (!note || !expr_equiv_p (XEXP (note, 0), mem))
|
continue;
|
continue;
|
|
|
if (dump_file)
|
if (dump_file)
|
fprintf (dump_file, "STORE_MOTION drop REG_EQUAL note at insn %d:\n",
|
fprintf (dump_file, "STORE_MOTION drop REG_EQUAL note at insn %d:\n",
|
INSN_UID (insn));
|
INSN_UID (insn));
|
remove_note (insn, note);
|
remove_note (insn, note);
|
}
|
}
|
remove_reachable_equiv_notes (bb, smexpr);
|
remove_reachable_equiv_notes (bb, smexpr);
|
}
|
}
|
|
|
|
|
/* Delete a store, but copy the value that would have been stored into
|
/* Delete a store, but copy the value that would have been stored into
|
the reaching_reg for later storing. */
|
the reaching_reg for later storing. */
|
|
|
static void
|
static void
|
delete_store (struct ls_expr * expr, basic_block bb)
|
delete_store (struct ls_expr * expr, basic_block bb)
|
{
|
{
|
rtx reg, i, del;
|
rtx reg, i, del;
|
|
|
if (expr->reaching_reg == NULL_RTX)
|
if (expr->reaching_reg == NULL_RTX)
|
expr->reaching_reg = gen_reg_rtx (GET_MODE (expr->pattern));
|
expr->reaching_reg = gen_reg_rtx (GET_MODE (expr->pattern));
|
|
|
reg = expr->reaching_reg;
|
reg = expr->reaching_reg;
|
|
|
for (i = AVAIL_STORE_LIST (expr); i; i = XEXP (i, 1))
|
for (i = AVAIL_STORE_LIST (expr); i; i = XEXP (i, 1))
|
{
|
{
|
del = XEXP (i, 0);
|
del = XEXP (i, 0);
|
if (BLOCK_FOR_INSN (del) == bb)
|
if (BLOCK_FOR_INSN (del) == bb)
|
{
|
{
|
/* We know there is only one since we deleted redundant
|
/* We know there is only one since we deleted redundant
|
ones during the available computation. */
|
ones during the available computation. */
|
replace_store_insn (reg, del, bb, expr);
|
replace_store_insn (reg, del, bb, expr);
|
break;
|
break;
|
}
|
}
|
}
|
}
|
}
|
}
|
|
|
/* Free memory used by store motion. */
|
/* Free memory used by store motion. */
|
|
|
static void
|
static void
|
free_store_memory (void)
|
free_store_memory (void)
|
{
|
{
|
free_ldst_mems ();
|
free_ldst_mems ();
|
|
|
if (ae_gen)
|
if (ae_gen)
|
sbitmap_vector_free (ae_gen);
|
sbitmap_vector_free (ae_gen);
|
if (ae_kill)
|
if (ae_kill)
|
sbitmap_vector_free (ae_kill);
|
sbitmap_vector_free (ae_kill);
|
if (transp)
|
if (transp)
|
sbitmap_vector_free (transp);
|
sbitmap_vector_free (transp);
|
if (st_antloc)
|
if (st_antloc)
|
sbitmap_vector_free (st_antloc);
|
sbitmap_vector_free (st_antloc);
|
if (pre_insert_map)
|
if (pre_insert_map)
|
sbitmap_vector_free (pre_insert_map);
|
sbitmap_vector_free (pre_insert_map);
|
if (pre_delete_map)
|
if (pre_delete_map)
|
sbitmap_vector_free (pre_delete_map);
|
sbitmap_vector_free (pre_delete_map);
|
if (reg_set_in_block)
|
if (reg_set_in_block)
|
sbitmap_vector_free (reg_set_in_block);
|
sbitmap_vector_free (reg_set_in_block);
|
|
|
ae_gen = ae_kill = transp = st_antloc = NULL;
|
ae_gen = ae_kill = transp = st_antloc = NULL;
|
pre_insert_map = pre_delete_map = reg_set_in_block = NULL;
|
pre_insert_map = pre_delete_map = reg_set_in_block = NULL;
|
}
|
}
|
|
|
/* Perform store motion. Much like gcse, except we move expressions the
|
/* Perform store motion. Much like gcse, except we move expressions the
|
other way by looking at the flowgraph in reverse. */
|
other way by looking at the flowgraph in reverse. */
|
|
|
static void
|
static void
|
store_motion (void)
|
store_motion (void)
|
{
|
{
|
basic_block bb;
|
basic_block bb;
|
int x;
|
int x;
|
struct ls_expr * ptr;
|
struct ls_expr * ptr;
|
int update_flow = 0;
|
int update_flow = 0;
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "before store motion\n");
|
fprintf (dump_file, "before store motion\n");
|
print_rtl (dump_file, get_insns ());
|
print_rtl (dump_file, get_insns ());
|
}
|
}
|
|
|
init_alias_analysis ();
|
init_alias_analysis ();
|
|
|
/* Find all the available and anticipatable stores. */
|
/* Find all the available and anticipatable stores. */
|
num_stores = compute_store_table ();
|
num_stores = compute_store_table ();
|
if (num_stores == 0)
|
if (num_stores == 0)
|
{
|
{
|
htab_delete (pre_ldst_table);
|
htab_delete (pre_ldst_table);
|
pre_ldst_table = NULL;
|
pre_ldst_table = NULL;
|
sbitmap_vector_free (reg_set_in_block);
|
sbitmap_vector_free (reg_set_in_block);
|
end_alias_analysis ();
|
end_alias_analysis ();
|
return;
|
return;
|
}
|
}
|
|
|
/* Now compute kill & transp vectors. */
|
/* Now compute kill & transp vectors. */
|
build_store_vectors ();
|
build_store_vectors ();
|
add_noreturn_fake_exit_edges ();
|
add_noreturn_fake_exit_edges ();
|
connect_infinite_loops_to_exit ();
|
connect_infinite_loops_to_exit ();
|
|
|
edge_list = pre_edge_rev_lcm (num_stores, transp, ae_gen,
|
edge_list = pre_edge_rev_lcm (num_stores, transp, ae_gen,
|
st_antloc, ae_kill, &pre_insert_map,
|
st_antloc, ae_kill, &pre_insert_map,
|
&pre_delete_map);
|
&pre_delete_map);
|
|
|
/* Now we want to insert the new stores which are going to be needed. */
|
/* Now we want to insert the new stores which are going to be needed. */
|
for (ptr = first_ls_expr (); ptr != NULL; ptr = next_ls_expr (ptr))
|
for (ptr = first_ls_expr (); ptr != NULL; ptr = next_ls_expr (ptr))
|
{
|
{
|
/* If any of the edges we have above are abnormal, we can't move this
|
/* If any of the edges we have above are abnormal, we can't move this
|
store. */
|
store. */
|
for (x = NUM_EDGES (edge_list) - 1; x >= 0; x--)
|
for (x = NUM_EDGES (edge_list) - 1; x >= 0; x--)
|
if (TEST_BIT (pre_insert_map[x], ptr->index)
|
if (TEST_BIT (pre_insert_map[x], ptr->index)
|
&& (INDEX_EDGE (edge_list, x)->flags & EDGE_ABNORMAL))
|
&& (INDEX_EDGE (edge_list, x)->flags & EDGE_ABNORMAL))
|
break;
|
break;
|
|
|
if (x >= 0)
|
if (x >= 0)
|
{
|
{
|
if (dump_file != NULL)
|
if (dump_file != NULL)
|
fprintf (dump_file,
|
fprintf (dump_file,
|
"Can't replace store %d: abnormal edge from %d to %d\n",
|
"Can't replace store %d: abnormal edge from %d to %d\n",
|
ptr->index, INDEX_EDGE (edge_list, x)->src->index,
|
ptr->index, INDEX_EDGE (edge_list, x)->src->index,
|
INDEX_EDGE (edge_list, x)->dest->index);
|
INDEX_EDGE (edge_list, x)->dest->index);
|
continue;
|
continue;
|
}
|
}
|
|
|
/* Now we want to insert the new stores which are going to be needed. */
|
/* Now we want to insert the new stores which are going to be needed. */
|
|
|
FOR_EACH_BB (bb)
|
FOR_EACH_BB (bb)
|
if (TEST_BIT (pre_delete_map[bb->index], ptr->index))
|
if (TEST_BIT (pre_delete_map[bb->index], ptr->index))
|
delete_store (ptr, bb);
|
delete_store (ptr, bb);
|
|
|
for (x = 0; x < NUM_EDGES (edge_list); x++)
|
for (x = 0; x < NUM_EDGES (edge_list); x++)
|
if (TEST_BIT (pre_insert_map[x], ptr->index))
|
if (TEST_BIT (pre_insert_map[x], ptr->index))
|
update_flow |= insert_store (ptr, INDEX_EDGE (edge_list, x));
|
update_flow |= insert_store (ptr, INDEX_EDGE (edge_list, x));
|
}
|
}
|
|
|
if (update_flow)
|
if (update_flow)
|
commit_edge_insertions ();
|
commit_edge_insertions ();
|
|
|
free_store_memory ();
|
free_store_memory ();
|
free_edge_list (edge_list);
|
free_edge_list (edge_list);
|
remove_fake_exit_edges ();
|
remove_fake_exit_edges ();
|
end_alias_analysis ();
|
end_alias_analysis ();
|
}
|
}
|
|
|
|
|
/* Entry point for jump bypassing optimization pass. */
|
/* Entry point for jump bypassing optimization pass. */
|
|
|
static int
|
static int
|
bypass_jumps (void)
|
bypass_jumps (void)
|
{
|
{
|
int changed;
|
int changed;
|
|
|
/* We do not construct an accurate cfg in functions which call
|
/* We do not construct an accurate cfg in functions which call
|
setjmp, so just punt to be safe. */
|
setjmp, so just punt to be safe. */
|
if (current_function_calls_setjmp)
|
if (current_function_calls_setjmp)
|
return 0;
|
return 0;
|
|
|
/* Identify the basic block information for this function, including
|
/* Identify the basic block information for this function, including
|
successors and predecessors. */
|
successors and predecessors. */
|
max_gcse_regno = max_reg_num ();
|
max_gcse_regno = max_reg_num ();
|
|
|
if (dump_file)
|
if (dump_file)
|
dump_flow_info (dump_file, dump_flags);
|
dump_flow_info (dump_file, dump_flags);
|
|
|
/* Return if there's nothing to do, or it is too expensive. */
|
/* Return if there's nothing to do, or it is too expensive. */
|
if (n_basic_blocks <= NUM_FIXED_BLOCKS + 1
|
if (n_basic_blocks <= NUM_FIXED_BLOCKS + 1
|
|| is_too_expensive (_ ("jump bypassing disabled")))
|
|| is_too_expensive (_ ("jump bypassing disabled")))
|
return 0;
|
return 0;
|
|
|
gcc_obstack_init (&gcse_obstack);
|
gcc_obstack_init (&gcse_obstack);
|
bytes_used = 0;
|
bytes_used = 0;
|
|
|
/* We need alias. */
|
/* We need alias. */
|
init_alias_analysis ();
|
init_alias_analysis ();
|
|
|
/* Record where pseudo-registers are set. This data is kept accurate
|
/* Record where pseudo-registers are set. This data is kept accurate
|
during each pass. ??? We could also record hard-reg information here
|
during each pass. ??? We could also record hard-reg information here
|
[since it's unchanging], however it is currently done during hash table
|
[since it's unchanging], however it is currently done during hash table
|
computation.
|
computation.
|
|
|
It may be tempting to compute MEM set information here too, but MEM sets
|
It may be tempting to compute MEM set information here too, but MEM sets
|
will be subject to code motion one day and thus we need to compute
|
will be subject to code motion one day and thus we need to compute
|
information about memory sets when we build the hash tables. */
|
information about memory sets when we build the hash tables. */
|
|
|
alloc_reg_set_mem (max_gcse_regno);
|
alloc_reg_set_mem (max_gcse_regno);
|
compute_sets ();
|
compute_sets ();
|
|
|
max_gcse_regno = max_reg_num ();
|
max_gcse_regno = max_reg_num ();
|
alloc_gcse_mem ();
|
alloc_gcse_mem ();
|
changed = one_cprop_pass (MAX_GCSE_PASSES + 2, true, true);
|
changed = one_cprop_pass (MAX_GCSE_PASSES + 2, true, true);
|
free_gcse_mem ();
|
free_gcse_mem ();
|
|
|
if (dump_file)
|
if (dump_file)
|
{
|
{
|
fprintf (dump_file, "BYPASS of %s: %d basic blocks, ",
|
fprintf (dump_file, "BYPASS of %s: %d basic blocks, ",
|
current_function_name (), n_basic_blocks);
|
current_function_name (), n_basic_blocks);
|
fprintf (dump_file, "%d bytes\n\n", bytes_used);
|
fprintf (dump_file, "%d bytes\n\n", bytes_used);
|
}
|
}
|
|
|
obstack_free (&gcse_obstack, NULL);
|
obstack_free (&gcse_obstack, NULL);
|
free_reg_set_mem ();
|
free_reg_set_mem ();
|
|
|
/* We are finished with alias. */
|
/* We are finished with alias. */
|
end_alias_analysis ();
|
end_alias_analysis ();
|
allocate_reg_info (max_reg_num (), FALSE, FALSE);
|
allocate_reg_info (max_reg_num (), FALSE, FALSE);
|
|
|
return changed;
|
return changed;
|
}
|
}
|
|
|
/* Return true if the graph is too expensive to optimize. PASS is the
|
/* Return true if the graph is too expensive to optimize. PASS is the
|
optimization about to be performed. */
|
optimization about to be performed. */
|
|
|
static bool
|
static bool
|
is_too_expensive (const char *pass)
|
is_too_expensive (const char *pass)
|
{
|
{
|
/* Trying to perform global optimizations on flow graphs which have
|
/* Trying to perform global optimizations on flow graphs which have
|
a high connectivity will take a long time and is unlikely to be
|
a high connectivity will take a long time and is unlikely to be
|
particularly useful.
|
particularly useful.
|
|
|
In normal circumstances a cfg should have about twice as many
|
In normal circumstances a cfg should have about twice as many
|
edges as blocks. But we do not want to punish small functions
|
edges as blocks. But we do not want to punish small functions
|
which have a couple switch statements. Rather than simply
|
which have a couple switch statements. Rather than simply
|
threshold the number of blocks, uses something with a more
|
threshold the number of blocks, uses something with a more
|
graceful degradation. */
|
graceful degradation. */
|
if (n_edges > 20000 + n_basic_blocks * 4)
|
if (n_edges > 20000 + n_basic_blocks * 4)
|
{
|
{
|
warning (OPT_Wdisabled_optimization,
|
warning (OPT_Wdisabled_optimization,
|
"%s: %d basic blocks and %d edges/basic block",
|
"%s: %d basic blocks and %d edges/basic block",
|
pass, n_basic_blocks, n_edges / n_basic_blocks);
|
pass, n_basic_blocks, n_edges / n_basic_blocks);
|
|
|
return true;
|
return true;
|
}
|
}
|
|
|
/* If allocating memory for the cprop bitmap would take up too much
|
/* If allocating memory for the cprop bitmap would take up too much
|
storage it's better just to disable the optimization. */
|
storage it's better just to disable the optimization. */
|
if ((n_basic_blocks
|
if ((n_basic_blocks
|
* SBITMAP_SET_SIZE (max_reg_num ())
|
* SBITMAP_SET_SIZE (max_reg_num ())
|
* sizeof (SBITMAP_ELT_TYPE)) > MAX_GCSE_MEMORY)
|
* sizeof (SBITMAP_ELT_TYPE)) > MAX_GCSE_MEMORY)
|
{
|
{
|
warning (OPT_Wdisabled_optimization,
|
warning (OPT_Wdisabled_optimization,
|
"%s: %d basic blocks and %d registers",
|
"%s: %d basic blocks and %d registers",
|
pass, n_basic_blocks, max_reg_num ());
|
pass, n_basic_blocks, max_reg_num ());
|
|
|
return true;
|
return true;
|
}
|
}
|
|
|
return false;
|
return false;
|
}
|
}
|
|
|
static bool
|
static bool
|
gate_handle_jump_bypass (void)
|
gate_handle_jump_bypass (void)
|
{
|
{
|
return optimize > 0 && flag_gcse;
|
return optimize > 0 && flag_gcse;
|
}
|
}
|
|
|
/* Perform jump bypassing and control flow optimizations. */
|
/* Perform jump bypassing and control flow optimizations. */
|
static unsigned int
|
static unsigned int
|
rest_of_handle_jump_bypass (void)
|
rest_of_handle_jump_bypass (void)
|
{
|
{
|
cleanup_cfg (CLEANUP_EXPENSIVE);
|
cleanup_cfg (CLEANUP_EXPENSIVE);
|
reg_scan (get_insns (), max_reg_num ());
|
reg_scan (get_insns (), max_reg_num ());
|
|
|
if (bypass_jumps ())
|
if (bypass_jumps ())
|
{
|
{
|
rebuild_jump_labels (get_insns ());
|
rebuild_jump_labels (get_insns ());
|
cleanup_cfg (CLEANUP_EXPENSIVE);
|
cleanup_cfg (CLEANUP_EXPENSIVE);
|
delete_trivially_dead_insns (get_insns (), max_reg_num ());
|
delete_trivially_dead_insns (get_insns (), max_reg_num ());
|
}
|
}
|
return 0;
|
return 0;
|
}
|
}
|
|
|
struct tree_opt_pass pass_jump_bypass =
|
struct tree_opt_pass pass_jump_bypass =
|
{
|
{
|
"bypass", /* name */
|
"bypass", /* name */
|
gate_handle_jump_bypass, /* gate */
|
gate_handle_jump_bypass, /* gate */
|
rest_of_handle_jump_bypass, /* execute */
|
rest_of_handle_jump_bypass, /* execute */
|
NULL, /* sub */
|
NULL, /* sub */
|
NULL, /* next */
|
NULL, /* next */
|
0, /* static_pass_number */
|
0, /* static_pass_number */
|
TV_BYPASS, /* tv_id */
|
TV_BYPASS, /* tv_id */
|
0, /* properties_required */
|
0, /* properties_required */
|
0, /* properties_provided */
|
0, /* properties_provided */
|
0, /* properties_destroyed */
|
0, /* properties_destroyed */
|
0, /* todo_flags_start */
|
0, /* todo_flags_start */
|
TODO_dump_func |
|
TODO_dump_func |
|
TODO_ggc_collect | TODO_verify_flow, /* todo_flags_finish */
|
TODO_ggc_collect | TODO_verify_flow, /* todo_flags_finish */
|
'G' /* letter */
|
'G' /* letter */
|
};
|
};
|
|
|
|
|
static bool
|
static bool
|
gate_handle_gcse (void)
|
gate_handle_gcse (void)
|
{
|
{
|
return optimize > 0 && flag_gcse;
|
return optimize > 0 && flag_gcse;
|
}
|
}
|
|
|
|
|
static unsigned int
|
static unsigned int
|
rest_of_handle_gcse (void)
|
rest_of_handle_gcse (void)
|
{
|
{
|
int save_csb, save_cfj;
|
int save_csb, save_cfj;
|
int tem2 = 0, tem;
|
int tem2 = 0, tem;
|
|
|
tem = gcse_main (get_insns ());
|
tem = gcse_main (get_insns ());
|
rebuild_jump_labels (get_insns ());
|
rebuild_jump_labels (get_insns ());
|
delete_trivially_dead_insns (get_insns (), max_reg_num ());
|
delete_trivially_dead_insns (get_insns (), max_reg_num ());
|
|
|
save_csb = flag_cse_skip_blocks;
|
save_csb = flag_cse_skip_blocks;
|
save_cfj = flag_cse_follow_jumps;
|
save_cfj = flag_cse_follow_jumps;
|
flag_cse_skip_blocks = flag_cse_follow_jumps = 0;
|
flag_cse_skip_blocks = flag_cse_follow_jumps = 0;
|
|
|
/* If -fexpensive-optimizations, re-run CSE to clean up things done
|
/* If -fexpensive-optimizations, re-run CSE to clean up things done
|
by gcse. */
|
by gcse. */
|
if (flag_expensive_optimizations)
|
if (flag_expensive_optimizations)
|
{
|
{
|
timevar_push (TV_CSE);
|
timevar_push (TV_CSE);
|
reg_scan (get_insns (), max_reg_num ());
|
reg_scan (get_insns (), max_reg_num ());
|
tem2 = cse_main (get_insns (), max_reg_num ());
|
tem2 = cse_main (get_insns (), max_reg_num ());
|
purge_all_dead_edges ();
|
purge_all_dead_edges ();
|
delete_trivially_dead_insns (get_insns (), max_reg_num ());
|
delete_trivially_dead_insns (get_insns (), max_reg_num ());
|
timevar_pop (TV_CSE);
|
timevar_pop (TV_CSE);
|
cse_not_expected = !flag_rerun_cse_after_loop;
|
cse_not_expected = !flag_rerun_cse_after_loop;
|
}
|
}
|
|
|
/* If gcse or cse altered any jumps, rerun jump optimizations to clean
|
/* If gcse or cse altered any jumps, rerun jump optimizations to clean
|
things up. */
|
things up. */
|
if (tem || tem2)
|
if (tem || tem2)
|
{
|
{
|
timevar_push (TV_JUMP);
|
timevar_push (TV_JUMP);
|
rebuild_jump_labels (get_insns ());
|
rebuild_jump_labels (get_insns ());
|
delete_dead_jumptables ();
|
delete_dead_jumptables ();
|
cleanup_cfg (CLEANUP_EXPENSIVE);
|
cleanup_cfg (CLEANUP_EXPENSIVE);
|
timevar_pop (TV_JUMP);
|
timevar_pop (TV_JUMP);
|
}
|
}
|
|
|
flag_cse_skip_blocks = save_csb;
|
flag_cse_skip_blocks = save_csb;
|
flag_cse_follow_jumps = save_cfj;
|
flag_cse_follow_jumps = save_cfj;
|
return 0;
|
return 0;
|
}
|
}
|
|
|
struct tree_opt_pass pass_gcse =
|
struct tree_opt_pass pass_gcse =
|
{
|
{
|
"gcse1", /* name */
|
"gcse1", /* name */
|
gate_handle_gcse, /* gate */
|
gate_handle_gcse, /* gate */
|
rest_of_handle_gcse, /* execute */
|
rest_of_handle_gcse, /* execute */
|
NULL, /* sub */
|
NULL, /* sub */
|
NULL, /* next */
|
NULL, /* next */
|
0, /* static_pass_number */
|
0, /* static_pass_number */
|
TV_GCSE, /* tv_id */
|
TV_GCSE, /* tv_id */
|
0, /* properties_required */
|
0, /* properties_required */
|
0, /* properties_provided */
|
0, /* properties_provided */
|
0, /* properties_destroyed */
|
0, /* properties_destroyed */
|
0, /* todo_flags_start */
|
0, /* todo_flags_start */
|
TODO_dump_func |
|
TODO_dump_func |
|
TODO_verify_flow | TODO_ggc_collect, /* todo_flags_finish */
|
TODO_verify_flow | TODO_ggc_collect, /* todo_flags_finish */
|
'G' /* letter */
|
'G' /* letter */
|
};
|
};
|
|
|
|
|
#include "gt-gcse.h"
|
#include "gt-gcse.h"
|
|
|
