[PR64164] drop copyrename, integrate into expand

On Mar 27, 2015, Alexandre Oliva <aoliva@redhat.com> wrote:

> This patch reworks the out-of-ssa expander to enable coalescing of SSA
> partitions that don't share the same base name.  This is done only when
> optimizing.

> The test we use to tell whether two partitions can be merged no longer
> demands them to have the same base variable when optimizing, so they
> become eligible for coalescing, as they would after copyrename.  We then
> compute the partitioning we'd get if all coalescible partitions were
> coalesced, using this partition assignment to assign base vars numbers.
> These base var numbers are then used to identify conflicts, which used
> to be based on shared base vars or base types.

> We now propagate base var names during coalescing proper, only towards
> the leader variable.  I'm no longer sure this is still needed, but
> something about handling variables and results led me this way and I
> didn't revisit it.  I might rework that with a later patch, or a later
> revision of this patch; it would require other means to identify
> partitions holding result_decls during merging, or allow that and deal
> with param and result decls in a different way during expand proper.

> I had to fix two lingering bugs in order for the whole thing to work: we
> perform conflict detection after abnormal coalescing, but we computed
> live ranges involving only the partition leaders, so conflicts with
> other names already coalesced wouldn't be detected.

This early abnormal coalescing was only present for a few days in the
trunk, and I was lucky enough to start working on a tree that had it.
It turns out that the fix for it was thus rendered unnecessary, so I
dropped it.  It was the fix for it, that didn't cover the live range
check, that caused the two ICEs I saw in the regressions tests.  Since
the ultimate cause of the problem is gone, and the change that
introduced the check failures, both problems went *poof* after I updated
the tree, resolved the conflicts and dropped the redundant code.

> The other problem was that we didn't track default defs for parms as
> live at entry, so they might end up coalesced.

I improved this a little bit, using the bitmap of partitions containing
default params to check that we only process function-entry defs for
them, rather than for all param decls in case they end up in other
partitions.

> I guess none of these problems would have been exercised in practice,
> because we wouldn't even consider merging ssa names associated with
> different variables.

> In the end, I verified that this fixed the codegen regression in the
> PR64164 testcase, that failed to merge two partitions that could in
> theory be merged, but that wasn't even considered due to differences in
> the SSA var names.

> I'd agree that disregarding the var names and dropping 4 passes is too
> much of a change to fix this one problem, but...  it's something we
> should have long tackled, and it gets this and other jobs done, so...

Regstrapped on x86_64-linux-gnu native and on i686-pc-linux-gnu native
on x86_64, so without lto plugin.  The only regression is in
gcc.dg/guality/pr54200.c, that explicitly disables VTA.  When
optimization is enabled, the different coalescing we perform now causes
VTA-less variable tracking to lose track of variable "z".  This
regression in non-VTA var-tracking is expected and, as richi put it in
PR 64164, I guess we don't care about that, do we? :-)

The other guality regressions I mentioned in my other email turned out
not to be regressions, but preexisting failures that somehow did not
make to the test_summary of my earlier pristine build.

Is this ok to install?

for  gcc/ChangeLog

	PR rtl-optimization/64164
	* Makefile.in (OBJS): Drop tree-ssa-copyrename.o.
	* tree-ssa-copyrename.c: Removed.
	* opts.c (default_options_table): Drop -ftree-copyrename.
	* passes.def: Drop all occurrences of pass_rename_ssa_copies.
	* common.opt (ftree-copyrename): Ignore.
	(ftree-coalesce-vars, ftree-coalesce-inlined-vars): Likewise.
	* doc/invoke.texi: Remove the ignored options above.
	* gimple-expr.c (gimple_can_coalesce_p): Allow coalescing
	across base variables when optimizing.
	* tree-ssa-coalesce.c (build_ssa_conflict_graph): Add
	param_defaults argument.  Process PARM_DECLs's default defs at
	the entry point.
	(attempt_coalesce): Add param_defaults argument, and
	track the presence of default defs for params in each
	partition.  Propagate base var to leader on merge, preferring
	parms and results, named vars, ignored vars, and then anon
	vars.  Refuse to merge a RESULT_DECL partition with a default
	PARM_DECL one.
	(coalesce_partitions): Add param_defaults argument,
	and pass it to attempt_coalesce.
	(dump_part_var_map): New.
	(compute_optimized_partition_bases): New, called by...
	(coalesce_ssa_name): ... when optimizing, disabling
	partition_view_bitmap's base assignment.  Pass local
	param_defaults to coalescer functions.
	* tree-ssa-live.c (var_map_base_init): Note use only when not
	optimizing.
---
 gcc/Makefile.in           |    1 
 gcc/common.opt            |   12 +
 gcc/doc/invoke.texi       |   29 ---
 gcc/gimple-expr.c         |    7 -
 gcc/opts.c                |    1 
 gcc/passes.def            |    5 
 gcc/tree-ssa-coalesce.c   |  342 ++++++++++++++++++++++++++++++-
 gcc/tree-ssa-copyrename.c |  499 ---------------------------------------------
 gcc/tree-ssa-live.c       |    5 
 9 files changed, 347 insertions(+), 554 deletions(-)
 delete mode 100644 gcc/tree-ssa-copyrename.c

[PR64164] drop copyrename, integrate into expand

Commit Message

Comments

Patch