On August 1, 2018 4:58:09 PM GMT+02:00, Richard Sandiford
<richard.sandif...@arm.com> wrote:
>Richard Biener <rguent...@suse.de> writes:
>> This should be 4/4 but I have the main patch on top, so...
>>
>> This uses the region-based VN from GIMPLE unrolling which means
>> we better approximate the effects optimizations on unrolled inner
>> loops when evaluating whether to unroll outer ones.
>
>Great! Sounds like it should also fix cases where missed
>value-numbering
>opportunities after unrolling prevented SLP vectorisation. (Hit a few
>of
>those, but don't have the testcases to hand unfortunately.)
Yes though as before we do not value number the blocks if they are not
contained inside a loop. That's of course easy to change.
Richard.
>Thanks,
>Richard
>
>>
>> * tree-ssa-loop-ivcanon.c: Include tree-ssa-sccvn.h.
>> (propagate_constants_for_unrolling): Remove.
>> (tree_unroll_loops_completely): Perform value-numbering
>> on the unrolled bodies loop parent.
>>
>> * gfortran.dg/reassoc_4.f: Change max-completely-peeled-insns
>> param to current default.
>> ---
>> gcc/testsuite/gfortran.dg/reassoc_4.f | 2 +-
>> gcc/tree-ssa-loop-ivcanon.c | 57
>++++++-----------------------------
>> 2 files changed, 10 insertions(+), 49 deletions(-)
>>
>> diff --git a/gcc/testsuite/gfortran.dg/reassoc_4.f
>b/gcc/testsuite/gfortran.dg/reassoc_4.f
>> index b155cba768c..07b4affb2a4 100644
>> --- a/gcc/testsuite/gfortran.dg/reassoc_4.f
>> +++ b/gcc/testsuite/gfortran.dg/reassoc_4.f
>> @@ -1,5 +1,5 @@
>> ! { dg-do compile }
>> -! { dg-options "-O3 -ffast-math -fdump-tree-reassoc1 --param
>max-completely-peeled-insns=400" }
>> +! { dg-options "-O3 -ffast-math -fdump-tree-reassoc1 --param
>max-completely-peeled-insns=200" }
>> ! { dg-additional-options "--param max-completely-peel-times=16" {
>target spu-*-* } }
>> subroutine anisonl(w,vo,anisox,s,ii1,jj1,weight)
>> integer ii1,jj1,i1,iii1,j1,jjj1,k1,l1,m1,n1
>> diff --git a/gcc/tree-ssa-loop-ivcanon.c
>b/gcc/tree-ssa-loop-ivcanon.c
>> index 326589f63c3..97c2ad94985 100644
>> --- a/gcc/tree-ssa-loop-ivcanon.c
>> +++ b/gcc/tree-ssa-loop-ivcanon.c
>> @@ -63,6 +63,7 @@ along with GCC; see the file COPYING3. If not see
>> #include "tree-inline.h"
>> #include "tree-cfgcleanup.h"
>> #include "builtins.h"
>> +#include "tree-ssa-sccvn.h"
>>
>> /* Specifies types of loops that may be unrolled. */
>>
>> @@ -1318,50 +1319,6 @@ canonicalize_induction_variables (void)
>> return 0;
>> }
>>
>> -/* Propagate constant SSA_NAMEs defined in basic block BB. */
>> -
>> -static void
>> -propagate_constants_for_unrolling (basic_block bb)
>> -{
>> - /* Look for degenerate PHI nodes with constant argument. */
>> - for (gphi_iterator gsi = gsi_start_phis (bb); !gsi_end_p (gsi); )
>> - {
>> - gphi *phi = gsi.phi ();
>> - tree result = gimple_phi_result (phi);
>> - tree arg = gimple_phi_arg_def (phi, 0);
>> -
>> - if (! SSA_NAME_OCCURS_IN_ABNORMAL_PHI (result)
>> - && gimple_phi_num_args (phi) == 1
>> - && CONSTANT_CLASS_P (arg))
>> - {
>> - replace_uses_by (result, arg);
>> - gsi_remove (&gsi, true);
>> - release_ssa_name (result);
>> - }
>> - else
>> - gsi_next (&gsi);
>> - }
>> -
>> - /* Look for assignments to SSA names with constant RHS. */
>> - for (gimple_stmt_iterator gsi = gsi_start_bb (bb); !gsi_end_p
>(gsi); )
>> - {
>> - gimple *stmt = gsi_stmt (gsi);
>> - tree lhs;
>> -
>> - if (is_gimple_assign (stmt)
>> - && TREE_CODE_CLASS (gimple_assign_rhs_code (stmt)) ==
>tcc_constant
>> - && (lhs = gimple_assign_lhs (stmt), TREE_CODE (lhs) == SSA_NAME)
>> - && !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (lhs))
>> - {
>> - replace_uses_by (lhs, gimple_assign_rhs1 (stmt));
>> - gsi_remove (&gsi, true);
>> - release_ssa_name (lhs);
>> - }
>> - else
>> - gsi_next (&gsi);
>> - }
>> -}
>> -
>> /* Process loops from innermost to outer, stopping at the innermost
>> loop we unrolled. */
>>
>> @@ -1512,10 +1469,14 @@ tree_unroll_loops_completely (bool
>may_increase_size, bool unroll_outer)
>> EXECUTE_IF_SET_IN_BITMAP (fathers, 0, i, bi)
>> {
>> loop_p father = get_loop (cfun, i);
>> - basic_block *body = get_loop_body_in_dom_order (father);
>> - for (unsigned j = 0; j < father->num_nodes; j++)
>> - propagate_constants_for_unrolling (body[j]);
>> - free (body);
>> + bitmap exit_bbs = BITMAP_ALLOC (NULL);
>> + loop_exit *exit = father->exits->next;
>> + while (exit->e)
>> + {
>> + bitmap_set_bit (exit_bbs, exit->e->dest->index);
>> + exit = exit->next;
>> + }
>> + do_rpo_vn (cfun, loop_preheader_edge (father), exit_bbs);
>> }
>> BITMAP_FREE (fathers);
