[PATCHv2,2/2] PHIOPT: Improve replace_phi_edge_with_variable for diamond shapped bb
Checks
Commit Message
While looking at differences between what minmax_replacement
and match_simplify_replacement does. I noticed that they sometimes
chose different edges to remove. I decided we should be able to do
better and be able to remove both empty basic blocks in the
case of match_simplify_replacement as that moves the statements.
This also updates the testcases as now match_simplify_replacement
will remove the unused MIN/MAX_EXPR and they were checking for
those.
OK? Bootstrapped and tested on x86_64-linux-gnu with no regressions.
gcc/ChangeLog:
* tree-ssa-phiopt.cc (replace_phi_edge_with_variable): Handle
diamond form bb with forwarder only empty blocks better.
gcc/testsuite/ChangeLog:
* gcc.dg/tree-ssa/minmax-15.c: Update test.
* gcc.dg/tree-ssa/minmax-16.c: Update test.
* gcc.dg/tree-ssa/minmax-3.c: Update test.
* gcc.dg/tree-ssa/minmax-4.c: Update test.
* gcc.dg/tree-ssa/minmax-5.c: Update test.
* gcc.dg/tree-ssa/minmax-8.c: Update test.
---
gcc/testsuite/gcc.dg/tree-ssa/minmax-15.c | 3 +-
gcc/testsuite/gcc.dg/tree-ssa/minmax-16.c | 9 ++----
gcc/testsuite/gcc.dg/tree-ssa/minmax-3.c | 2 +-
gcc/testsuite/gcc.dg/tree-ssa/minmax-4.c | 2 +-
gcc/testsuite/gcc.dg/tree-ssa/minmax-5.c | 2 +-
gcc/testsuite/gcc.dg/tree-ssa/minmax-8.c | 2 +-
gcc/tree-ssa-phiopt.cc | 35 ++++++++++++++++++++++-
7 files changed, 43 insertions(+), 12 deletions(-)
Comments
On Thu, May 4, 2023 at 1:13 AM Andrew Pinski via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> While looking at differences between what minmax_replacement
> and match_simplify_replacement does. I noticed that they sometimes
> chose different edges to remove. I decided we should be able to do
> better and be able to remove both empty basic blocks in the
> case of match_simplify_replacement as that moves the statements.
>
> This also updates the testcases as now match_simplify_replacement
> will remove the unused MIN/MAX_EXPR and they were checking for
> those.
>
> OK? Bootstrapped and tested on x86_64-linux-gnu with no regressions.
OK.
> gcc/ChangeLog:
>
> * tree-ssa-phiopt.cc (replace_phi_edge_with_variable): Handle
> diamond form bb with forwarder only empty blocks better.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.dg/tree-ssa/minmax-15.c: Update test.
> * gcc.dg/tree-ssa/minmax-16.c: Update test.
> * gcc.dg/tree-ssa/minmax-3.c: Update test.
> * gcc.dg/tree-ssa/minmax-4.c: Update test.
> * gcc.dg/tree-ssa/minmax-5.c: Update test.
> * gcc.dg/tree-ssa/minmax-8.c: Update test.
> ---
> gcc/testsuite/gcc.dg/tree-ssa/minmax-15.c | 3 +-
> gcc/testsuite/gcc.dg/tree-ssa/minmax-16.c | 9 ++----
> gcc/testsuite/gcc.dg/tree-ssa/minmax-3.c | 2 +-
> gcc/testsuite/gcc.dg/tree-ssa/minmax-4.c | 2 +-
> gcc/testsuite/gcc.dg/tree-ssa/minmax-5.c | 2 +-
> gcc/testsuite/gcc.dg/tree-ssa/minmax-8.c | 2 +-
> gcc/tree-ssa-phiopt.cc | 35 ++++++++++++++++++++++-
> 7 files changed, 43 insertions(+), 12 deletions(-)
>
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/minmax-15.c b/gcc/testsuite/gcc.dg/tree-ssa/minmax-15.c
> index 8a39871c938..6731f91e6c3 100644
> --- a/gcc/testsuite/gcc.dg/tree-ssa/minmax-15.c
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/minmax-15.c
> @@ -30,5 +30,6 @@ main (void)
> return 0;
> }
>
> -/* { dg-final { scan-tree-dump-times "MIN_EXPR" 3 "phiopt1" } } */
> +/* There should only be two MIN_EXPR left, the 3rd one was removed. */
> +/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "phiopt1" } } */
> /* { dg-final { scan-tree-dump-times "MAX_EXPR" 0 "phiopt1" } } */
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/minmax-16.c b/gcc/testsuite/gcc.dg/tree-ssa/minmax-16.c
> index 623b12b3f74..094364e6424 100644
> --- a/gcc/testsuite/gcc.dg/tree-ssa/minmax-16.c
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/minmax-16.c
> @@ -25,11 +25,8 @@ main (void)
> return 0;
> }
>
> -/* After phiopt1, there really should be only 3 MIN_EXPR in the IR (including debug statements).
> - But the way phiopt does not cleanup the CFG all the time, the PHI might still reference the
> - alternative bb's moved statement.
> - Note in the end, we do dce the statement and other debug statements to end up with only 2 MIN_EXPR.
> - So check that too. */
> -/* { dg-final { scan-tree-dump-times "MIN_EXPR" 4 "phiopt1" } } */
> +/* After phiopt1, will be only 2 MIN_EXPR in the IR (including debug statements). */
> +/* xk will only have the final result so the extra debug info does not change anything. */
> +/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "phiopt1" } } */
> /* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "optimized" } } */
> /* { dg-final { scan-tree-dump-times "MAX_EXPR" 0 "phiopt1" } } */
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/minmax-3.c b/gcc/testsuite/gcc.dg/tree-ssa/minmax-3.c
> index 2af10776346..521afe3e4d9 100644
> --- a/gcc/testsuite/gcc.dg/tree-ssa/minmax-3.c
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/minmax-3.c
> @@ -25,5 +25,5 @@ main (void)
> return 0;
> }
>
> -/* { dg-final { scan-tree-dump-times "MIN_EXPR" 3 "phiopt1" } } */
> +/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "phiopt1" } } */
> /* { dg-final { scan-tree-dump-times "MAX_EXPR" 0 "phiopt1" } } */
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/minmax-4.c b/gcc/testsuite/gcc.dg/tree-ssa/minmax-4.c
> index 973f39bfed3..49e27185b5e 100644
> --- a/gcc/testsuite/gcc.dg/tree-ssa/minmax-4.c
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/minmax-4.c
> @@ -26,4 +26,4 @@ main (void)
> }
>
> /* { dg-final { scan-tree-dump-times "MIN_EXPR" 0 "phiopt1" } } */
> -/* { dg-final { scan-tree-dump-times "MAX_EXPR" 3 "phiopt1" } } */
> +/* { dg-final { scan-tree-dump-times "MAX_EXPR" 2 "phiopt1" } } */
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/minmax-5.c b/gcc/testsuite/gcc.dg/tree-ssa/minmax-5.c
> index 34e4e720511..194c881cc98 100644
> --- a/gcc/testsuite/gcc.dg/tree-ssa/minmax-5.c
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/minmax-5.c
> @@ -25,5 +25,5 @@ main (void)
> return 0;
> }
>
> -/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "phiopt1" } } */
> +/* { dg-final { scan-tree-dump-times "MIN_EXPR" 1 "phiopt1" } } */
> /* { dg-final { scan-tree-dump-times "MAX_EXPR" 1 "phiopt1" } } */
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/minmax-8.c b/gcc/testsuite/gcc.dg/tree-ssa/minmax-8.c
> index 0160e573fef..d5cb53145ea 100644
> --- a/gcc/testsuite/gcc.dg/tree-ssa/minmax-8.c
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/minmax-8.c
> @@ -26,4 +26,4 @@ main (void)
> }
>
> /* { dg-final { scan-tree-dump-times "MIN_EXPR" 1 "phiopt1" } } */
> -/* { dg-final { scan-tree-dump-times "MAX_EXPR" 2 "phiopt1" } } */
> +/* { dg-final { scan-tree-dump-times "MAX_EXPR" 1 "phiopt1" } } */
> diff --git a/gcc/tree-ssa-phiopt.cc b/gcc/tree-ssa-phiopt.cc
> index fbc1fd66607..7ba24211025 100644
> --- a/gcc/tree-ssa-phiopt.cc
> +++ b/gcc/tree-ssa-phiopt.cc
> @@ -94,6 +94,7 @@ replace_phi_edge_with_variable (basic_block cond_block,
> basic_block bb = gimple_bb (phi);
> gimple_stmt_iterator gsi;
> tree phi_result = PHI_RESULT (phi);
> + bool deleteboth = false;
>
> /* Duplicate range info if they are the only things setting the target PHI.
> This is needed as later on, the new_tree will be replacing
> @@ -137,7 +138,14 @@ replace_phi_edge_with_variable (basic_block cond_block,
> keep_edge = EDGE_SUCC (cond_block, 1);
> }
> else if ((keep_edge = find_edge (cond_block, e->src)))
> - ;
> + {
> + basic_block bb1 = EDGE_SUCC (cond_block, 0)->dest;
> + basic_block bb2 = EDGE_SUCC (cond_block, 1)->dest;
> + if (single_pred_p (bb1) && single_pred_p (bb2)
> + && single_succ_p (bb1) && single_succ_p (bb2)
> + && empty_block_p (bb1) && empty_block_p (bb2))
> + deleteboth = true;
> + }
> else
> gcc_unreachable ();
>
> @@ -148,6 +156,31 @@ replace_phi_edge_with_variable (basic_block cond_block,
> e->probability = profile_probability::always ();
> delete_basic_block (edge_to_remove->dest);
>
> + /* Eliminate the COND_EXPR at the end of COND_BLOCK. */
> + gsi = gsi_last_bb (cond_block);
> + gsi_remove (&gsi, true);
> + }
> + else if (deleteboth)
> + {
> + basic_block bb1 = EDGE_SUCC (cond_block, 0)->dest;
> + basic_block bb2 = EDGE_SUCC (cond_block, 1)->dest;
> +
> + edge newedge = redirect_edge_and_branch (keep_edge, bb);
> +
> + /* The new edge should be the same. */
> + gcc_assert (newedge == keep_edge);
> +
> + keep_edge->flags |= EDGE_FALLTHRU;
> + keep_edge->flags &= ~(EDGE_TRUE_VALUE | EDGE_FALSE_VALUE);
> + keep_edge->probability = profile_probability::always ();
> +
> + /* Copy the edge's phi entry from the old one. */
> + copy_phi_arg_into_existing_phi (e, keep_edge);
> +
> + /* Delete the old 2 empty basic blocks */
> + delete_basic_block (bb1);
> + delete_basic_block (bb2);
> +
> /* Eliminate the COND_EXPR at the end of COND_BLOCK. */
> gsi = gsi_last_bb (cond_block);
> gsi_remove (&gsi, true);
> --
> 2.31.1
>
@@ -30,5 +30,6 @@ main (void)
return 0;
}
-/* { dg-final { scan-tree-dump-times "MIN_EXPR" 3 "phiopt1" } } */
+/* There should only be two MIN_EXPR left, the 3rd one was removed. */
+/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "phiopt1" } } */
/* { dg-final { scan-tree-dump-times "MAX_EXPR" 0 "phiopt1" } } */
@@ -25,11 +25,8 @@ main (void)
return 0;
}
-/* After phiopt1, there really should be only 3 MIN_EXPR in the IR (including debug statements).
- But the way phiopt does not cleanup the CFG all the time, the PHI might still reference the
- alternative bb's moved statement.
- Note in the end, we do dce the statement and other debug statements to end up with only 2 MIN_EXPR.
- So check that too. */
-/* { dg-final { scan-tree-dump-times "MIN_EXPR" 4 "phiopt1" } } */
+/* After phiopt1, will be only 2 MIN_EXPR in the IR (including debug statements). */
+/* xk will only have the final result so the extra debug info does not change anything. */
+/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "phiopt1" } } */
/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "optimized" } } */
/* { dg-final { scan-tree-dump-times "MAX_EXPR" 0 "phiopt1" } } */
@@ -25,5 +25,5 @@ main (void)
return 0;
}
-/* { dg-final { scan-tree-dump-times "MIN_EXPR" 3 "phiopt1" } } */
+/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "phiopt1" } } */
/* { dg-final { scan-tree-dump-times "MAX_EXPR" 0 "phiopt1" } } */
@@ -26,4 +26,4 @@ main (void)
}
/* { dg-final { scan-tree-dump-times "MIN_EXPR" 0 "phiopt1" } } */
-/* { dg-final { scan-tree-dump-times "MAX_EXPR" 3 "phiopt1" } } */
+/* { dg-final { scan-tree-dump-times "MAX_EXPR" 2 "phiopt1" } } */
@@ -25,5 +25,5 @@ main (void)
return 0;
}
-/* { dg-final { scan-tree-dump-times "MIN_EXPR" 2 "phiopt1" } } */
+/* { dg-final { scan-tree-dump-times "MIN_EXPR" 1 "phiopt1" } } */
/* { dg-final { scan-tree-dump-times "MAX_EXPR" 1 "phiopt1" } } */
@@ -26,4 +26,4 @@ main (void)
}
/* { dg-final { scan-tree-dump-times "MIN_EXPR" 1 "phiopt1" } } */
-/* { dg-final { scan-tree-dump-times "MAX_EXPR" 2 "phiopt1" } } */
+/* { dg-final { scan-tree-dump-times "MAX_EXPR" 1 "phiopt1" } } */
@@ -94,6 +94,7 @@ replace_phi_edge_with_variable (basic_block cond_block,
basic_block bb = gimple_bb (phi);
gimple_stmt_iterator gsi;
tree phi_result = PHI_RESULT (phi);
+ bool deleteboth = false;
/* Duplicate range info if they are the only things setting the target PHI.
This is needed as later on, the new_tree will be replacing
@@ -137,7 +138,14 @@ replace_phi_edge_with_variable (basic_block cond_block,
keep_edge = EDGE_SUCC (cond_block, 1);
}
else if ((keep_edge = find_edge (cond_block, e->src)))
- ;
+ {
+ basic_block bb1 = EDGE_SUCC (cond_block, 0)->dest;
+ basic_block bb2 = EDGE_SUCC (cond_block, 1)->dest;
+ if (single_pred_p (bb1) && single_pred_p (bb2)
+ && single_succ_p (bb1) && single_succ_p (bb2)
+ && empty_block_p (bb1) && empty_block_p (bb2))
+ deleteboth = true;
+ }
else
gcc_unreachable ();
@@ -148,6 +156,31 @@ replace_phi_edge_with_variable (basic_block cond_block,
e->probability = profile_probability::always ();
delete_basic_block (edge_to_remove->dest);
+ /* Eliminate the COND_EXPR at the end of COND_BLOCK. */
+ gsi = gsi_last_bb (cond_block);
+ gsi_remove (&gsi, true);
+ }
+ else if (deleteboth)
+ {
+ basic_block bb1 = EDGE_SUCC (cond_block, 0)->dest;
+ basic_block bb2 = EDGE_SUCC (cond_block, 1)->dest;
+
+ edge newedge = redirect_edge_and_branch (keep_edge, bb);
+
+ /* The new edge should be the same. */
+ gcc_assert (newedge == keep_edge);
+
+ keep_edge->flags |= EDGE_FALLTHRU;
+ keep_edge->flags &= ~(EDGE_TRUE_VALUE | EDGE_FALSE_VALUE);
+ keep_edge->probability = profile_probability::always ();
+
+ /* Copy the edge's phi entry from the old one. */
+ copy_phi_arg_into_existing_phi (e, keep_edge);
+
+ /* Delete the old 2 empty basic blocks */
+ delete_basic_block (bb1);
+ delete_basic_block (bb2);
+
/* Eliminate the COND_EXPR at the end of COND_BLOCK. */
gsi = gsi_last_bb (cond_block);
gsi_remove (&gsi, true);