Re: [PATCH v2] A new copy propagation and PHI elimination pass

[Richard Biener <rguenther@suse.de>]

On Thu, 2 Nov 2023, Filip Kastl wrote:

> > Hi,
> > 
> > this is a patch that I submitted two months ago as an RFC. I added some polish
> > since.
> > 
> > It is a new lightweight pass that removes redundant PHI functions and as a
> > bonus does basic copy propagation. With Jan Hubi?ka we measured that it is able
> > to remove usually more than 5% of all PHI functions when run among early passes
> > (sometimes even 13% or more). Those are mostly PHI functions that would be
> > later optimized away but with this pass it is possible to remove them early
> > enough so that they don't get streamed when runing LTO (and also potentially
> > inlined at multiple places). It is also able to remove some redundant PHIs
> > that otherwise would still be present during RTL expansion.
> > 
> > Jakub Jel?nek was concerned about debug info coverage so I compiled cc1plus
> > with and without this patch. These are the sizes of .debug_info and
> > .debug_loclists
> > 
> > .debug_info without patch 181694311
> > .debug_info    with patch 181692320
> > +0.0011% change
> > 
> > .debug_loclists without patch 47934753
> > .debug_loclists    with patch 47934966
> > -0.0004% change
> > 
> > I wanted to use dwlocstat to compare debug coverages but didn't manage to get
> > the program working on my machine sadly. Hope this suffices. Seems to me that
> > my patch doesn't have a significant impact on debug info.
> > 
> > Bootstraped and tested* on x86_64-pc-linux-gnu.
> > 
> > * One testcase (pr79691.c) did regress. However that is because the test is
> > dependent on a certain variable not being copy propagated. I will go into more
> > detail about this in a reply to this mail.
> > 
> > Ok to commit?
> 
> This is a second version of the patch.  In this version, I modified the
> pr79691.c testcase so that it works as intended with other changes from the
> patch.
> 
> The pr79691.c testcase checks that we get constants from snprintf calls and
> that they simplify into a single constant.  The testcase doesn't account for
> the fact that this constant may be further copy propagated which is exactly
> what happens with this patch applied.
> 
> Bootstrapped and tested on x86_64-pc-linux-gnu.
> 
> Ok to commit?
> 
> Filip Kastl
> 
> -- >8 --
> 
> This patch adds the strongly-connected copy propagation (SCCOPY) pass.
> It is a lightweight GIMPLE copy propagation pass that also removes some
> redundant PHI statements. It handles degenerate PHIs, e.g.:
> 
> _5 = PHI <_1>;
> _6 = PHI <_6, _6, _1, _1>;
> _7 = PHI <16, _7>;
> // Replaces occurences of _5 and _6 by _1 and _7 by 16
> 
> It also handles more complicated situations, e.g.:
> 
> _8 = PHI <_9, _10>;
> _9 = PHI <_8, _10>;
> _10 = PHI <_8, _9, _1>;
> // Replaces occurences of _8, _9 and _10 by _1
> 
> gcc/ChangeLog:
> 
> 	* Makefile.in: Added sccopy pass.
> 	* passes.def: Added sccopy pass before LTO streaming and before
> 	  RTL expansion.
> 	* tree-pass.h (make_pass_sccopy): Added sccopy pass.
> 	* tree-ssa-sccopy.cc: New file.

Can you name the new file gimple-ssa-sccopy.cc please?

> 
> gcc/testsuite/ChangeLog:
> 
> 	* gcc.dg/tree-ssa/pr79691.c: Updated scan-tree-dump to account
> 	  for additional copy propagation this patch adds.
> 	* gcc.dg/sccopy-1.c: New test.
> 
> Signed-off-by: Filip Kastl <fkastl@suse.cz>
> ---
>  gcc/Makefile.in                         |   1 +
>  gcc/passes.def                          |   3 +
>  gcc/testsuite/gcc.dg/sccopy-1.c         |  78 +++
>  gcc/testsuite/gcc.dg/tree-ssa/pr79691.c |   2 +-
>  gcc/tree-pass.h                         |   1 +
>  gcc/tree-ssa-sccopy.cc                  | 867 ++++++++++++++++++++++++
>  6 files changed, 951 insertions(+), 1 deletion(-)
>  create mode 100644 gcc/testsuite/gcc.dg/sccopy-1.c
>  create mode 100644 gcc/tree-ssa-sccopy.cc
> 
> diff --git a/gcc/Makefile.in b/gcc/Makefile.in
> index a25a1e32fbc..2bd5a015676 100644
> --- a/gcc/Makefile.in
> +++ b/gcc/Makefile.in
> @@ -1736,6 +1736,7 @@ OBJS = \
>  	tree-ssa-pre.o \
>  	tree-ssa-propagate.o \
>  	tree-ssa-reassoc.o \
> +	tree-ssa-sccopy.o \
>  	tree-ssa-sccvn.o \
>  	tree-ssa-scopedtables.o \
>  	tree-ssa-sink.o \
> diff --git a/gcc/passes.def b/gcc/passes.def
> index 1e1950bdb39..fa6c5a2c9fa 100644
> --- a/gcc/passes.def
> +++ b/gcc/passes.def
> @@ -100,6 +100,7 @@ along with GCC; see the file COPYING3.  If not see
>  	  NEXT_PASS (pass_if_to_switch);
>  	  NEXT_PASS (pass_convert_switch);
>  	  NEXT_PASS (pass_cleanup_eh);
> +	  NEXT_PASS (pass_sccopy);
>  	  NEXT_PASS (pass_profile);
>  	  NEXT_PASS (pass_local_pure_const);
>  	  NEXT_PASS (pass_modref);
> @@ -368,6 +369,7 @@ along with GCC; see the file COPYING3.  If not see
>  	 However, this also causes us to misdiagnose cases that should be
>  	 real warnings (e.g., testsuite/gcc.dg/pr18501.c).  */
>        NEXT_PASS (pass_cd_dce, false /* update_address_taken_p */);
> +      NEXT_PASS (pass_sccopy);
>        NEXT_PASS (pass_tail_calls);
>        /* Split critical edges before late uninit warning to reduce the
>           number of false positives from it.  */
> @@ -409,6 +411,7 @@ along with GCC; see the file COPYING3.  If not see
>        NEXT_PASS (pass_sancov);
>        NEXT_PASS (pass_asan);
>        NEXT_PASS (pass_tsan);
> +      NEXT_PASS (pass_sccopy);
>        /* ???  We do want some kind of loop invariant motion, but we possibly
>           need to adjust LIM to be more friendly towards preserving accurate
>  	 debug information here.  */
> diff --git a/gcc/testsuite/gcc.dg/sccopy-1.c b/gcc/testsuite/gcc.dg/sccopy-1.c
> new file mode 100644
> index 00000000000..1e61a6b320e
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/sccopy-1.c
> @@ -0,0 +1,78 @@
> +/* { dg-do compile } */
> +/* { dg-options "-fgimple -fdump-tree-sccopy -O2" } */
> +/* { dg-final { scan-tree-dump "Replacing SCC of size 2" "sccopy1" } } */
> +
> +int __GIMPLE (ssa, startwith ("sccopy"))
> +main ()
> +{
> +  int a;
> +  int y;
> +  int x;
> +  int _1;
> +  int _2;
> +  int _13;
> +
> +  __BB(2):
> +  if (x_7(D) == 5)
> +    goto __BB3;
> +  else
> +    goto __BB4;
> +
> +  __BB(3):
> +  a_10 = x_7(D);
> +  goto __BB5;
> +
> +  __BB(4):
> +  a_9 = y_8(D);
> +  goto __BB5;
> +
> +  __BB(5):
> +  a_3 = __PHI (__BB3: a_10, __BB4: a_9);
> +  if (x_7(D) == y_8(D))
> +    goto __BB6;
> +  else
> +    goto __BB11;
> +
> +  __BB(6):
> +  a_11 = a_3 + 1;
> +  goto __BB7;
> +
> +  __BB(7):
> +  a_4 = __PHI (__BB6: a_11, __BB11: a_6);
> +label1:
> +  if (x_7(D) != y_8(D))
> +    goto __BB8;
> +  else
> +    goto __BB10;
> +
> +  __BB(8):
> +  goto __BB9;
> +
> +  __BB(9):
> +  a_12 = __PHI (__BB8: a_4, __BB10: a_5);
> +  goto __BB10;
> +
> +  __BB(10,loop_header(1)):
> +  a_5 = __PHI (__BB7: a_4, __BB9: a_12);
> +label2:
> +  _1 = y_8(D) * 2;
> +  if (x_7(D) == _1)
> +    goto __BB9;
> +  else
> +    goto __BB11;
> +
> +  __BB(11):
> +  a_6 = __PHI (__BB5: a_3, __BB10: a_5);
> +  _2 = x_7(D) * 3;
> +  if (y_8(D) == _2)
> +    goto __BB7;
> +  else
> +    goto __BB12;
> +
> +  __BB(12):
> +  _13 = 0;
> +  return _13;
> +
> +}
> +
> +
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr79691.c b/gcc/testsuite/gcc.dg/tree-ssa/pr79691.c
> index bf889318c06..43770c95bca 100644
> --- a/gcc/testsuite/gcc.dg/tree-ssa/pr79691.c
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr79691.c
> @@ -34,4 +34,4 @@ int f4 (int i)
>  
>  /* { dg-final { scan-tree-dump-times "sprintf" 1 "optimized" } }
>     { dg-final { scan-tree-dump-times "snprintf" 1 "optimized" } }
> -   { dg-final { scan-tree-dump " = 9;" "optimized" } } */
> +   { dg-final { scan-tree-dump "return 9;" "optimized" } } */
> diff --git a/gcc/tree-pass.h b/gcc/tree-pass.h
> index 09e6ada5b2f..94a48461590 100644
> --- a/gcc/tree-pass.h
> +++ b/gcc/tree-pass.h
> @@ -399,6 +399,7 @@ extern gimple_opt_pass *make_pass_iv_optimize (gcc::context *ctxt);
>  extern gimple_opt_pass *make_pass_tree_loop_done (gcc::context *ctxt);
>  extern gimple_opt_pass *make_pass_ch (gcc::context *ctxt);
>  extern gimple_opt_pass *make_pass_ch_vect (gcc::context *ctxt);
> +extern gimple_opt_pass *make_pass_sccopy (gcc::context *ctxt);
>  extern gimple_opt_pass *make_pass_ccp (gcc::context *ctxt);
>  extern gimple_opt_pass *make_pass_split_paths (gcc::context *ctxt);
>  extern gimple_opt_pass *make_pass_build_ssa (gcc::context *ctxt);
> diff --git a/gcc/tree-ssa-sccopy.cc b/gcc/tree-ssa-sccopy.cc
> new file mode 100644
> index 00000000000..f6e462f4c71
> --- /dev/null
> +++ b/gcc/tree-ssa-sccopy.cc
> @@ -0,0 +1,867 @@
> +/* Strongly-connected copy propagation pass for the GNU compiler.
> +   Copyright (C) 2023 Free Software Foundation, Inc.
> +   Contributed by Filip Kastl <fkastl@suse.cz>
> +
> +This file is part of GCC.
> +
> +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
> +Software Foundation; either version 3, or (at your option) any later
> +version.
> +
> +GCC is distributed in the hope that it will be useful, but WITHOUT ANY
> +WARRANTY; without even the implied warranty of MERCHANTABILITY or
> +FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
> +for more details.
> +
> +You should have received a copy of the GNU General Public License
> +along with GCC; see the file COPYING3.  If not see
> +<http://www.gnu.org/licenses/>.  */
> +
> +#include "config.h"
> +#include "system.h"
> +#include "coretypes.h"
> +#include "backend.h"
> +#include "tree.h"
> +#include "gimple.h"
> +#include "tree-pass.h"
> +#include "ssa.h"
> +#include "gimple-iterator.h"
> +#include "vec.h"
> +#include "hash-set.h"
> +#include <algorithm>
> +#include "ssa-iterators.h"
> +#include "gimple-fold.h"
> +#include "gimplify.h"
> +#include "tree-cfg.h"
> +#include "tree-eh.h"
> +#include "tree-cfgcleanup.h"
> +#include "builtins.h"
> +
> +/* Strongly connected copy propagation pass.
> +
> +   This is a lightweight copy propagation pass that is also able to eliminate
> +   redundant PHI statements.  The pass considers the following types of copy
> +   statements:
> +
> +   1 An assignment statement with a single argument.
> +
> +   _3 = _2;
> +   _4 = 5;
> +
> +   2 A degenerate PHI statement.  A degenerate PHI is a PHI that only refers to
> +     itself or one other value.
> +
> +   _5 = PHI <_1>;
> +   _6 = PHI <_6, _6, _1, _1>;
> +   _7 = PHI <16, _7>;
> +
> +   3 A set of PHI statements that only refer to each other or to one other
> +     value.
> +
> +   _8 = PHI <_9, _10>;
> +   _9 = PHI <_8, _10>;
> +   _10 = PHI <_8, _9, _1>;

this case necessarily involves a cyclic CFG, so maybe say

"This is a lightweight SSA copy propagation pass that is able to handle
cycles optimistically, eliminating PHIs within those."

?  Or is this a mis-characterization?

> +   All of these statements produce copies and can be eliminated from the
> +   program.  For a copy statement we identify the value it creates a copy of
> +   and replace references to the statement with the value -- we propagate the
> +   copy.
> +
> +   _3 = _2; // Replace all occurences of _3 by _2
> +
> +   _8 = PHI <_9, _10>;
> +   _9 = PHI <_8, _10>;
> +   _10 = PHI <_8, _9, _1>; // Replace all occurences of _8, _9 and _10 by _1
> +
> +   To find all three types of copy statements we use an algorithm based on
> +   strongly-connected components (SCCs) in dataflow graph.  The algorithm was
> +   introduced in an article from 2013[1]. We describe the algorithm bellow.
> +
> +   To identify SCCs we implement the Robert Tarjan's SCC algorithm.  For the
> +   SCC computation we wrap potential copy statements in the 'vertex' struct.
> +   To each of these statements we also assign a vertex number ('vxnum'). Since
> +   the main algorithm has to be able to compute SCCs of subgraphs of the whole
> +   dataflow graph we use GIMPLE stmt flags to prevent Tarjan's algorithm from
> +   leaving the subgraph.
> +
> +   References:
> +
> +     [1] Simple and Efficient Construction of Static Single Assignmemnt Form,
> +     Braun, Buchwald, Hack, Leissa, Mallon, Zwinkau, 2013, LNCS vol. 7791,
> +     Section 3.2.  */
> +
> +/* State of vertex during Tarjan computation.
> +
> +   unvisited  Vertex hasn't yet been popped from worklist.
> +   vopen      DFS has visited vertex for the first time.  Vertex has been put
> +	      on Tarjan stack.
> +   closed     DFS has backtracked through vertex.  At this point, vertex
> +	      doesn't have any unvisited neighbors.
> +   in_scc     Vertex has been popped from Tarjan stack.  */
> +
> +enum vstate
> +{
> +  unvisited,
> +  vopen,
> +  closed,
> +  in_scc
> +};
> +
> +/* Information about a vertex.  Used by Tarjan.  */
> +
> +struct vertex
> +{
> +  vstate state;
> +  unsigned index;
> +  unsigned lowlink;
> +  gimple* stmt;
> +};
> +
> +/* Set 'dead' flag of gimple statement to true.
> +   We remove these statements at the end of the pass.  */
> +
> +static void
> +set_stmt_dead (gimple* stmt)
> +{
> +  gimple_set_plf (stmt, GF_PLF_1, true);
> +}
> +
> +/* For each statement from given SCC, mark it 'dead'.  */
> +
> +static void
> +set_scc_dead (vec<gimple *> scc)
> +{
> +  for (gimple *stmt : scc)
> +    set_stmt_dead (stmt);
> +}
> +
> +/* Clear 'dead' flag of gimple statement to false.  */
> +
> +static void
> +clear_stmt_dead (gimple* stmt)
> +{
> +  gimple_set_plf (stmt, GF_PLF_1, false);
> +}
> +
> +/* Return value of 'dead' flag of gimple statement.  */
> +
> +static bool
> +is_stmt_dead (gimple* stmt)
> +{
> +  return gimple_plf (stmt, GF_PLF_1);
> +}
> +
> +/* Set 'using' flag of gimple statement to true.
> +   If the flag isn't set, Tarjan will ignore the statement.  */
> +
> +static void
> +tarjan_set_using (gimple* stmt)
> +{
> +  gimple_set_plf (stmt, GF_PLF_2, true);
> +}
> +
> +/* Clear 'using' flag of gimple statement to false.  */
> +
> +static void
> +tarjan_clear_using (gimple* stmt)
> +{
> +  gimple_set_plf (stmt, GF_PLF_2, false);
> +}
> +
> +/* Return value of 'using' flag of gimple statement.  */
> +
> +static bool
> +tarjan_is_using (gimple* stmt)
> +{
> +  return gimple_plf (stmt, GF_PLF_2);
> +}
> +
> +/* Set 'vxnum' (vertex number) of statement.  Used by Tarjan.  */
> +
> +static void
> +tarjan_set_vxnum (gimple* stmt, unsigned num)
> +{
> +  gimple_set_uid (stmt, num);
> +}
> +
> +/* Return 'vxnum' (vertex number) of statement.  Used by Tarjan.  */
> +
> +static unsigned
> +tarjan_vxnum (gimple* stmt)
> +{
> +  return gimple_uid (stmt);
> +}
> +
> +/* Create and initialize vertex struct for each given statement.  */
> +
> +static auto_vec<vertex>
> +tarjan_stmts_to_vertices (auto_vec<gimple *> &stmts)
> +{
> +  auto_vec<vertex> result;
> +  for (gimple *stmt : stmts)
> +    {
> +      vertex v;
> +      v.state = unvisited;
> +      v.index = 0;
> +      v.lowlink = 0;
> +      v.stmt = stmt;
> +
> +      result.safe_push (v);
> +    }
> +  return result;
> +}
> +
> +/* Part of 'tarjan_compute_sccs ()'.  */
> +
> +static void
> +tarjan_visit_neighbor (tree neigh_tree, unsigned parent_vxnum,
> +		       auto_vec<vertex> &vs, auto_vec<unsigned> &worklist)
> +{
> +  if (TREE_CODE (neigh_tree) != SSA_NAME)
> +    return; /* Skip any neighbor that isn't an SSA name.  */
> +
> +  gimple *neigh_stmt = SSA_NAME_DEF_STMT (neigh_tree);
> +
> +  /* Skip neighbors outside the induced subgraph that Tarjan currently works
> +     with.  */
> +  if (!tarjan_is_using (neigh_stmt))
> +    return;
> +  unsigned neigh_vxnum = tarjan_vxnum (neigh_stmt);
> +
> +  vstate neigh_state = vs[neigh_vxnum].state;
> +  vstate parent_state = vs[parent_vxnum].state;
> +  if (parent_state == vopen) /* We're currently opening parent.  */
> +    {
> +      /* Put unvisited neighbors on worklist.  Update lowlink of parent
> +	 vertex according to indices of neighbors present on stack.  */
> +      switch (neigh_state)
> +	{
> +	case unvisited:
> +	  worklist.safe_push (neigh_vxnum);
> +	  break;
> +	case vopen:
> +	case closed:
> +	  vs[parent_vxnum].lowlink = std::min (vs[parent_vxnum].lowlink,
> +					       vs[neigh_vxnum].index);
> +	  break;
> +	case in_scc:
> +	  /* Ignore these edges.  */
> +	  break;
> +	}
> +    }
> +  else if (parent_state == closed) /* We're currently closing parent.  */
> +    {
> +      /* Update lowlink of parent vertex according to lowlinks of
> +	 children of parent (in terms of DFS tree).  */
> +      if (neigh_state == closed)
> +	{
> +	  vs[parent_vxnum].lowlink = std::min (vs[parent_vxnum].lowlink,
> +					       vs[neigh_vxnum].lowlink);
> +	}
> +    }
> +}
> +
> +/* Implementation of Tarjan's SCC algorithm.
> +
> +   Compute SCCs in dataflow graph on given statements.  Return the
> +   SCCs in a reverse topological order.
> +
> +   Given statements should be only those for which stmt_may_generate_copy
> +   returns 'true'.  */
> +
> +static auto_vec<vec<gimple *>>
> +tarjan_compute_sccs (auto_vec<gimple *> &copy_stmts)
> +{
> +  auto_vec<vec<gimple *>> sccs;
> +  auto_vec<unsigned> worklist; /* DFS stack.  */
> +  auto_vec<unsigned> stack; /* Tarjan stack.  */
> +  unsigned curr_index = 0;
> +
> +  auto_vec<vertex> vs = tarjan_stmts_to_vertices (copy_stmts);
> +
> +  /* Mark the subgraph induced by 'copy_stmts' and index it by vxnums.  */
> +  unsigned i;
> +  for (i = 0; i < vs.length (); i++)
> +    {
> +      gimple *stmt = vs[i].stmt;
> +      tarjan_set_using (stmt);
> +      tarjan_set_vxnum (stmt, i);
> +    }
> +
> +  /* Put all vertices on worklist.  */
> +  for (i = 0; i < vs.length (); i++)
> +    {
> +      worklist.safe_push (i);
> +    }
> +
> +  /* Worklist loop.  */
> +  while (!worklist.is_empty ())
> +    {
> +      unsigned i = worklist.pop ();
> +      gimple *stmt = vs[i].stmt;
> +      vstate state = vs[i].state;
> +
> +      if (state == unvisited)
> +	{
> +	  vs[i].state = vopen;
> +
> +	  /* Assign index to this vertex.  */
> +	  vs[i].index = curr_index;
> +	  vs[i].lowlink = curr_index;
> +	  curr_index++;
> +
> +	  /* Put vertex on stack and also on worklist to be closed later.  */
> +	  stack.safe_push (i);
> +	  worklist.safe_push (i);
> +	}
> +      else if (state == vopen)
> +	vs[i].state = closed;
> +
> +      /* Visit neighbors of this vertex.  */
> +      tree op;
> +      gphi *phi;
> +      switch (gimple_code (stmt))
> +	{
> +	  case GIMPLE_PHI:
> +	    phi = as_a <gphi *> (stmt);
> +	    unsigned j;
> +	    for (j = 0; j < gimple_phi_num_args (phi); j++)
> +	      {
> +		op = gimple_phi_arg_def (phi, j);
> +		tarjan_visit_neighbor (op, i, vs, worklist);
> +	      }
> +	    break;
> +	  case GIMPLE_ASSIGN:
> +	    op = gimple_assign_rhs1 (stmt);
> +	    tarjan_visit_neighbor (op, i, vs, worklist);
> +	    break;
> +	  default:
> +	    gcc_unreachable ();
> +	}
> +
> +      /* If we've just closed a root vertex of an scc, pop scc from stack.  */
> +      if (state == vopen && vs[i].lowlink == vs[i].index)
> +	{
> +	  vec<gimple *> scc = vNULL;
> +
> +	  unsigned j;
> +	  do
> +	    {
> +	      j = stack.pop ();
> +	      scc.safe_push (vs[j].stmt);
> +	      vs[j].state = in_scc;
> +	    }
> +	  while (j != i);
> +
> +	  sccs.safe_push (scc);

It might be nice to optimize SCCs of size 1 somehow, not sure how
many times these appear - possibly prevent them from even entering
the SCC discovery?

I'll note that while you are working with stmts everywhere that
you are really bound to using SSA defs and those would already
nicely have numbers (the SSA_NAME_VERSION).  In principle the
SCC lattice could be pre-allocated once, indexed by
SSA_NAME_VERSION and you could keep a "generation" number
indicating what SCC discovery round it belongs to (aka the
set_using).

There's a old SCC finding algorithm working on the SSA graph
in the old SCC based value-numbering, for example on the
gcc 7 branch in tree-ssa-sccvn.c:DFS

For reading it would be nice to put the SCC finding in its
own class.

> +	}
> +    }
> +
> +  if (!stack.is_empty ())
> +    gcc_unreachable ();
> +
> +  /* Clear copy stmts' 'using' flags.  */
> +  for (vertex v : vs)
> +    {
> +      gimple *s = v.stmt;
> +      tarjan_clear_using (s);
> +    }
> +
> +  return sccs;
> +}
> +
> +/* Could this statement potentially be a copy statement?
> +
> +   This pass only considers statements for which this function returns 'true'.
> +   Those are basically PHI functions and assignment statements similar to
> +
> +   _2 = _1;
> +   or
> +   _2 = 5;  */
> +
> +static bool
> +stmt_may_generate_copy (gimple *stmt)
> +{
> +  if (gimple_code (stmt) == GIMPLE_PHI)
> +    {
> +      gphi *phi = as_a <gphi *> (stmt);
> +
> +      /* No OCCURS_IN_ABNORMAL_PHI SSA names in lhs nor rhs.  */
> +      if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (gimple_phi_result (phi)))
> +	return false;
> +
> +      unsigned i;
> +      for (i = 0; i < gimple_phi_num_args (phi); i++)
> +	{
> +	  tree op = gimple_phi_arg_def (phi, i);
> +	  if (TREE_CODE (op) == SSA_NAME
> +	      && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op))
> +	    return false;
> +	}

When there's more than one non-SSA PHI argument and they are not
the same then the stmt also cannot be a copy, right?

> +      return true;
> +    }
> +
> +  if (gimple_code (stmt) != GIMPLE_ASSIGN)
> +    return false;
> +
> +  /* If the statement has volatile operands, it won't generate a
> +     useful copy.  */
> +  if (gimple_has_volatile_ops (stmt))
> +    return false;
> +
> +  /* Statements with loads and/or stores will never generate a useful copy.  */
> +  if (gimple_store_p (stmt) || gimple_assign_load_p (stmt))
> +    return false;
> +
> +  if (!gimple_assign_single_p (stmt))
> +    return false;
> +
> +  tree lhs = gimple_assign_lhs (stmt);
> +  if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (lhs))
> +    return false;
> +
> +  /* If the assignment is from a constant it generates a useful copy.  */
> +  if (is_gimple_min_invariant (gimple_assign_rhs1 (stmt)))
> +    return true;
> +
> +  tree rhs = single_ssa_tree_operand (stmt, SSA_OP_USE);
> +
> +  if (!is_gimple_val (gimple_assign_rhs1 (stmt)))
> +    return false;
> +
> +  /* rhs shouldn't flow through any abnormal edges.  */
> +  if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (rhs))
> +    return false;
> +
> +  if (!rhs)
> +    return false;

so this means

 _1 = 0;

doesn't generate a copy?  But you allowed those in PHIs?

Generally I wanted to mention there's gimple_assign_ssa_name_copy_p
which exactly matches _1 = _2 (and nothing more).  There cannot
be volatiles or side-effects in those, no stores or loads
(but SSA_NAME_OCCURS_IN_ABNORMAL_PHI can appear).

Can you use that and remove the unneeded checks above?

> +
> +  /* If rhs and lhs are pointers, alignment of lhs and rhs should be the same.
> +     Usage of __builtin_assume_aligned can cause alignment of lhs to be greater
> +     than alignment of rhs.  In that case we don't want to propagate rhs since
> +     we would lose the alignment information.  */

The same might be true for value-ranges.  I think it would be better to
do it like copy-prop does and in cycles propagate the info to the
copy-of variable we substitute with.  For non-cyclic cases we cannot
generally copy flow-sensitive info.

A more conservative test would be to compare
SSA_NAME_PTR_INFO for pointers and SSA_NAME_RANGE_INFO for non-pointers.

> +  if (POINTER_TYPE_P (TREE_TYPE (lhs))
> +      && POINTER_TYPE_P (TREE_TYPE (rhs))
> +      && get_pointer_alignment (lhs) != get_pointer_alignment (rhs))
> +    return false;
> +
> +  return true;
> +}
> +
> +/* Return all statements in cfun that could generate copies.  All statements
> +   for which stmt_may_generate_copy returns 'true'.  */
> +
> +static auto_vec<gimple *>
> +get_all_stmt_may_generate_copy (void)
> +{
> +  auto_vec<gimple *> result;
> +
> +  basic_block bb;
> +  FOR_EACH_BB_FN (bb, cfun)
> +    {
> +      gimple_stmt_iterator gsi;
> +      for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
> +	{
> +	  gimple *s = gsi_stmt (gsi);
> +	  if (stmt_may_generate_copy (s))
> +	    result.safe_push (s);
> +	}
> +
> +      gphi_iterator pi;
> +      for (pi = gsi_start_phis (bb); !gsi_end_p (pi); gsi_next (&pi))
> +	{
> +	  gimple *s = pi.phi ();
> +	  if (stmt_may_generate_copy (s))
> +	    result.safe_push (s);
> +	}
> +    }

I wonder if this can do stmt-to-vertices on-the-fly, thus not
build the vector of gimple *.  That is, you build very many
vec<>s (probably not many in practice, but ...)

> +  return result;
> +}
> +
> +/* Cleanup to be performed after every call of 'replace_use_by ()'.  */
> +
> +void
> +cleanup_after_replace (gimple *old_stmt, gimple *stmt, bitmap need_eh_cleanup,
> +		       bitmap need_ab_cleanup, vec<gimple *> stmts_to_fixup,
> +		       bool can_make_abnormal_goto, bool was_noreturn)
> +{
> +  basic_block bb = stmt->bb;
> +
> +  /* If we cleaned up EH information from the statement,
> +     remove EH edges.  */
> +  if (maybe_clean_or_replace_eh_stmt (old_stmt, stmt))
> +    bitmap_set_bit (need_eh_cleanup, bb->index);
> +
> +  /* If we turned a call with possible abnormal control transfer
> +     into one that doesn't, remove abnormal edges.  */
> +  if (can_make_abnormal_goto
> +      && !stmt_can_make_abnormal_goto (stmt))
> +    bitmap_set_bit (need_ab_cleanup, bb->index);
> +
> +  /* If we turned a not noreturn call into a noreturn one
> +     schedule it for fixup.  */
> +  if (!was_noreturn
> +      && is_gimple_call (stmt)
> +      && gimple_call_noreturn_p (stmt))
> +    stmts_to_fixup.safe_push (stmt);
> +
> +  if (gimple_assign_single_p (stmt))
> +    {
> +      tree rhs = gimple_assign_rhs1 (stmt);
> +
> +      if (TREE_CODE (rhs) == ADDR_EXPR)
> +	recompute_tree_invariant_for_addr_expr (rhs);
> +    }
> +
> +  update_stmt_if_modified (stmt);
> +}
> +
> +/* Cleanup related to 'replace_use_by ()'. In contrast to
> +   'cleanup_after_replace ()', this function needs to be called only at the
> +   end of the pass.  */
> +
> +void
> +cleanup_after_all_replaces_done (bitmap need_eh_cleanup, bitmap
> +				 need_ab_cleanup, vec<gimple *> stmts_to_fixup)
> +{
> +  if (!bitmap_empty_p (need_eh_cleanup))
> +    gimple_purge_all_dead_eh_edges (need_eh_cleanup);
> +  if (!bitmap_empty_p (need_ab_cleanup))
> +    gimple_purge_all_dead_abnormal_call_edges (need_ab_cleanup);
> +
> +  /* Fixup stmts that became noreturn calls.  This may require splitting
> +     blocks and thus isn't possible during the dominator walk.  Do this
> +     in reverse order so we don't inadvertedly remove a stmt we want to
> +     fixup by visiting a dominating now noreturn call first.  */
> +  while (!stmts_to_fixup.is_empty ())
> +    {
> +      gimple *stmt = stmts_to_fixup.pop ();
> +      fixup_noreturn_call (stmt);
> +    }
> +}
> +
> +/* Replace each use of stmt 'get_replaced' by a use of stmt 'replace_by'.  */
> +
> +static void
> +replace_use_by (tree get_replaced, tree replace_by, bitmap need_eh_cleanup,
> +		bitmap need_ab_cleanup, vec<gimple *> stmts_to_fixup)
> +{

If I understand correctly sofar you'll never replace a SSA name by
a constant, so REPLACE_BY is always an SSA_NAME.

I wonder if you ever hit the need to do any of the fancy cleanups.

In fact I suggest you use the existing replace_uses_by API instead
of re-inventing it.

> +  use_operand_p use_p;
> +  imm_use_iterator iter;
> +  gimple *use_stmt;
> +  FOR_EACH_IMM_USE_STMT (use_stmt, iter, get_replaced)
> +    {
> +      bool was_noreturn = false;
> +      bool can_make_abnormal_goto = false;
> +      if (is_gimple_call (use_stmt))
> +	{
> +	  was_noreturn = gimple_call_noreturn_p (use_stmt);
> +	  can_make_abnormal_goto = stmt_can_make_abnormal_goto (use_stmt);
> +	}
> +
> +      FOR_EACH_IMM_USE_ON_STMT (use_p, iter)
> +	SET_USE (use_p, unshare_expr (replace_by));
> +
> +      /* Recompute tree invariant.  */
> +      if (gimple_assign_single_p (use_stmt))
> +	{
> +	  tree rhs = gimple_assign_rhs1 (use_stmt);
> +
> +	  if (TREE_CODE (rhs) == ADDR_EXPR)
> +	    recompute_tree_invariant_for_addr_expr (rhs);
> +	}
> +
> +      /* Cleanup.  */
> +      gimple_stmt_iterator gsi = gsi_for_stmt (use_stmt);
> +      fold_stmt (&gsi);
> +      gimple_set_modified (gsi_stmt (gsi), true);
> +      cleanup_after_replace (use_stmt, gsi_stmt (gsi), need_eh_cleanup,
> +			     need_ab_cleanup, stmts_to_fixup,
> +			     can_make_abnormal_goto, was_noreturn);
> +    }
> +}
> +
> +/* For each statement from given SCC, replace its usages by value
> +   'replace_by'.  */
> +
> +static void
> +replace_scc_by_value (vec<gimple *> scc, tree replace_by, bitmap
> +		      need_eh_cleanup, bitmap need_ab_cleanup, vec<gimple *>
> +		      stmts_to_fixup)
> +{
> +  for (gimple *stmt : scc)
> +    {
> +      tree get_replaced = gimple_get_lhs (stmt);
> +      replace_use_by (get_replaced, replace_by, need_eh_cleanup,
> +		      need_ab_cleanup, stmts_to_fixup);
> +    }
> +
> +  if (dump_file)
> +    fprintf (dump_file, "Replacing SCC of size %d\n", scc.length ());
> +}
> +
> +/* Part of 'sccopy_propagate ()'.  */
> +
> +static void
> +sccopy_visit_op (tree op, hash_set<tree> &outer_ops,
> +		 hash_set<gimple *> &scc_set, bool &is_inner,
> +		 tree &last_outer_op)
> +{
> +  bool op_in_scc = false;
> +
> +  if (TREE_CODE (op) == SSA_NAME)
> +    {
> +      gimple *op_stmt = SSA_NAME_DEF_STMT (op);
> +      if (scc_set.contains (op_stmt))
> +	op_in_scc = true;
> +    }
> +
> +  if (!op_in_scc)
> +    {
> +      outer_ops.add (op);
> +      last_outer_op = op;
> +      is_inner = false;
> +    }
> +}
> +
> +/* Main function of this pass.  Find and propagate all three types of copy
> +   statements (see pass description above).
> +
> +   This is an implementation of an algorithm from the paper Simple and
> +   Efficient Construction of Static Single Assignmemnt Form[1].  It is based
> +   on strongly-connected components (SCCs) in dataflow graph.  The original
> +   algorithm only considers PHI statements.  We extend it to also consider
> +   assignment statements of type _2 = _1;.
> +
> +   The algorithm is based on this definition of a set of redundant PHIs[1]:
> +
> +     A non-empty set P of PHI functions is redundant iff the PHI functions just
> +     reference each other or one other value
> +
> +   It uses this lemma[1]:
> +
> +     Let P be a redundant set of PHI functions.  Then there is a
> +     strongly-connected component S subset of P that is also redundant.
> +
> +   The algorithm works in this way:
> +
> +     1 Find SCCs
> +     2 For each SCC S in topological order:
> +     3   Construct set 'inner' of statements that only have other statements
> +	 from S on their right hand side
> +     4   Construct set 'outer' of values that originate outside S and appear on
> +	 right hand side of some statement from S
> +     5   If |outer| = 1, outer only contains a value v.  Statements in S only
> +	 refer to each other or to v -- they are redundant.  Propagate v.
> +	 Else, recurse on statements in inner.
> +
> +   The implementation is non-recursive.
> +
> +   References:
> +
> +     [1] Simple and Efficient Construction of Static Single Assignmemnt Form,
> +     Braun, Buchwald, Hack, Leissa, Mallon, Zwinkau, 2013, LNCS vol. 7791,
> +     Section 3.2.  */
> +
> +static void
> +sccopy_propagate ()
> +{
> +  auto_vec<gimple *> useful_stmts = get_all_stmt_may_generate_copy ();
> +
> +  auto_vec<vec<gimple *>> worklist;
> +  worklist = tarjan_compute_sccs (useful_stmts);
> +
> +  /* Prepare data structs for cleanup after stmt modification.  */
> +  bitmap need_eh_cleanup = BITMAP_ALLOC (NULL);
> +  bitmap need_ab_cleanup = BITMAP_ALLOC (NULL);
> +  vec<gimple *> stmts_to_fixup = vNULL;
> +
> +  while (!worklist.is_empty ())
> +    {
> +      vec<gimple *> scc = worklist.pop ();
> +
> +      auto_vec<gimple *> inner;
> +      hash_set<tree> outer_ops;
> +      tree last_outer_op = NULL_TREE;
> +
> +      /* Prepare hash set of PHIs in scc to query later.  */
> +      hash_set<gimple *> scc_set;
> +      for (gimple *stmt : scc)
> +	scc_set.add (stmt);
> +
> +      for (gimple *stmt : scc)
> +	{
> +	  bool is_inner = true;
> +
> +	  gphi *phi;
> +	  tree op;
> +
> +	  switch (gimple_code (stmt))
> +	    {
> +	      case GIMPLE_PHI:
> +		phi = as_a <gphi *> (stmt);
> +		unsigned j;
> +		for (j = 0; j < gimple_phi_num_args (phi); j++)
> +		  {
> +		    op = gimple_phi_arg_def (phi, j);
> +		    sccopy_visit_op (op, outer_ops, scc_set, is_inner,
> +				   last_outer_op);
> +		  }
> +		break;
> +	      case GIMPLE_ASSIGN:
> +		op = gimple_assign_rhs1 (stmt);
> +		sccopy_visit_op (op, outer_ops, scc_set, is_inner,
> +			       last_outer_op);
> +		break;
> +	      default:
> +		gcc_unreachable ();
> +	    }
> +
> +	  if (is_inner)
> +	    {
> +	      inner.safe_push (stmt);
> +	    }
> +	}
> +
> +      if (outer_ops.elements () == 1)
> +	{
> +	  /* The only operand in outer_ops.  */
> +	  tree outer_op = last_outer_op;
> +
> +	  replace_scc_by_value (scc, outer_op, need_eh_cleanup,
> +				need_ab_cleanup, stmts_to_fixup);
> +	  set_scc_dead (scc);
> +	}
> +      else if (outer_ops.elements () > 1)
> +	{
> +	  /* Add inner sccs to worklist.  */
> +	  auto_vec<vec<gimple *>> inner_sccs = tarjan_compute_sccs (inner);
> +	  for (vec<gimple *> inner_scc : inner_sccs)
> +	    worklist.safe_push (inner_scc);
> +	}
> +      else
> +	gcc_unreachable ();
> +
> +      scc.release ();
> +    }
> +
> +  cleanup_after_all_replaces_done (need_eh_cleanup, need_ab_cleanup,
> +				   stmts_to_fixup);
> +
> +  /* Remove data structs for cleanup after stmt modification.  */
> +  BITMAP_FREE (need_eh_cleanup);
> +  BITMAP_FREE (need_ab_cleanup);
> +  stmts_to_fixup.release ();
> +}
> +
> +/* Called when pass execution starts.  */
> +
> +static void
> +init_sccopy (void)
> +{
> +  /* Clear statement flags.  */
> +  basic_block bb;
> +  FOR_EACH_BB_FN (bb, cfun)
> +    {
> +      gimple_stmt_iterator gsi;
> +      for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
> +	{
> +	  gimple* stmt = gsi_stmt (gsi);
> +	  clear_stmt_dead (stmt);
> +	  tarjan_clear_using (stmt);
> +	}
> +
> +      gphi_iterator pi;
> +      for (pi = gsi_start_phis (bb); !gsi_end_p (pi); gsi_next (&pi))
> +	{
> +	  gimple *stmt = pi.phi ();
> +	  clear_stmt_dead (stmt);
> +	  tarjan_clear_using (stmt);
> +	}
> +    }
> +}
> +
> +/* Called before pass execution ends.  */
> +
> +static void
> +finalize_sccopy (void)
> +{
> +  basic_block bb;
> +
> +  /* Remove all propagated statements.  */
> +  FOR_EACH_BB_FN (bb, cfun)
> +    {
> +      gimple_stmt_iterator gsi;
> +      for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi);)

You generally should remove stmts back-to-front to make
debug stmt generation work.  To not re-invent the wheel
it might be simplest to use simple_dce_from_worklist with
the dead stmts defs as seeds (just remove set/is_stmt_dead
and track dead SSA defs in a bitmap).

> +	{
> +	  gimple *stmt = gsi_stmt (gsi);
> +	  if (is_stmt_dead (stmt))
> +	    gsi_remove (&gsi, true);
> +	  else
> +	    gsi_next (&gsi);

this also fails to release the SSA def in the stmt.

  release_defs (stmt);

would do that (or simple_dce_from_worklist).

> +	}
> +
> +      gphi_iterator pi;
> +      for (pi = gsi_start_phis (bb); !gsi_end_p (pi);)
> +	{
> +	  gphi *stmt = pi.phi ();
> +
> +	  if (is_stmt_dead (stmt))
> +	    remove_phi_node (&pi, true);
> +	  else
> +	    gsi_next (&pi);
> +	}
> +    }
> +
> +  /* More cleanup.  */
> +  FOR_EACH_BB_FN (bb, cfun)
> +    gimple_purge_dead_eh_edges (bb);

You only removed copies, I'm sure this should neve do anything?

As a general comment I wonder if we can simplify most of the
pass by starting SSA based SCC discovery from a copy stmt and
whether we can reduce the number of vec<>s that are allocated
and possibly re-allocated many times?

Thanks,
Richard.

> +}
> +
> +namespace {
> +
> +const pass_data pass_data_sccopy =
> +{
> +  GIMPLE_PASS, /* type */
> +  "sccopy", /* name */
> +  OPTGROUP_NONE, /* optinfo_flags */
> +  TV_NONE, /* tv_id */
> +  ( PROP_cfg | PROP_ssa ), /* properties_required */
> +  0, /* properties_provided */
> +  0, /* properties_destroyed */
> +  0, /* todo_flags_start */
> +  TODO_update_ssa | TODO_cleanup_cfg, /* todo_flags_finish */
> +};
> +
> +class pass_sccopy : public gimple_opt_pass
> +{
> +public:
> +  pass_sccopy (gcc::context *ctxt)
> +    : gimple_opt_pass (pass_data_sccopy, ctxt)
> +  {}
> +
> +  /* opt_pass methods: */
> +  virtual bool gate (function *) { return true; }
> +  virtual unsigned int execute (function *);
> +  opt_pass * clone () final override { return new pass_sccopy (m_ctxt); }
> +}; // class pass_sccopy
> +
> +unsigned
> +pass_sccopy::execute (function *)
> +{
> +  init_sccopy ();
> +  sccopy_propagate ();
> +  finalize_sccopy ();
> +  return 0;
> +}
> +
> +} // anon namespace
> +
> +gimple_opt_pass *
> +make_pass_sccopy (gcc::context *ctxt)
> +{
> +  return new pass_sccopy (ctxt);
> +}
> 

-- 
Richard Biener <rguenther@suse.de>
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)