HI,
This patch backports r254778 and test case in r244815 to GCC6. Bootstrap and
test on x86_64. Is it OK?
Thanks,
bin
2017-12-18 Bin Cheng <bin.ch...@arm.com>
Backport from mainline
2017-11-15 Bin Cheng <bin.ch...@arm.com>
PR tree-optimization/82726
PR tree-optimization/70754
* tree-predcom.c (order_drefs_by_pos): New function.
(combine_chains): Move code setting has_max_use_after to...
(try_combine_chains): ...here. New parameter. Sort combined chains
according to position information.
(tree_predictive_commoning_loop): Update call to above function.
(update_pos_for_combined_chains, pcom_stmt_dominates_stmt_p): New.
gcc/testsuite
2017-12-18 Bin Cheng <bin.ch...@arm.com>
Backport from mainline
2017-11-15 Bin Cheng <bin.ch...@arm.com>
PR tree-optimization/82726
* gcc.dg/tree-ssa/pr82726.c: New test.
Backport from mainline
2017-01-23 Bin Cheng <bin.ch...@arm.com>
PR tree-optimization/70754
* gfortran.dg/pr70754.f90: New test.Index: gcc/testsuite/gcc.dg/tree-ssa/pr82726.c
===================================================================
--- gcc/testsuite/gcc.dg/tree-ssa/pr82726.c (revision 0)
+++ gcc/testsuite/gcc.dg/tree-ssa/pr82726.c (working copy)
@@ -0,0 +1,26 @@
+/* { dg-do compile } */
+/* { dg-options "-O3 --param tree-reassoc-width=4" } */
+/* { dg-additional-options "-mavx2" { target { x86_64-*-* i?86-*-* } } } */
+
+#define N 40
+#define M 128
+unsigned int in[N+M];
+unsigned short out[N];
+
+/* Outer-loop vectorization. */
+
+void
+foo (){
+ int i,j;
+ unsigned int diff;
+
+ for (i = 0; i < N; i++) {
+ diff = 0;
+ for (j = 0; j < M; j+=8) {
+ diff += in[j+i];
+ }
+ out[i]=(unsigned short)diff;
+ }
+
+ return;
+}
Index: gcc/testsuite/gfortran.dg/pr70754.f90
===================================================================
--- gcc/testsuite/gfortran.dg/pr70754.f90 (revision 0)
+++ gcc/testsuite/gfortran.dg/pr70754.f90 (working copy)
@@ -0,0 +1,35 @@
+! { dg-do compile }
+! { dg-options "-Ofast" }
+module m
+ implicit none
+ private
+ save
+
+ integer, parameter, public :: &
+ ii4 = selected_int_kind(6), &
+ rr8 = selected_real_kind(13)
+
+ integer (ii4), dimension(40,40,199), public :: xyz
+ public :: foo
+contains
+ subroutine foo(a)
+ real (rr8), dimension(40,40), intent(out) :: a
+ real (rr8), dimension(40,40) :: b
+ integer (ii4), dimension(40,40) :: c
+ integer i, j
+
+ do i=1,20
+ b(i,j) = 123 * a(i,j) + 34 * a(i,j+1) &
+ + 34 * a(i,j-1) + a(i+1,j+1) &
+ + a(i+1,j-1) + a(i-1,j+1) &
+ + a(i-1,j-1)
+ c(i,j) = 123
+ end do
+
+ where ((xyz(:,:,2) /= 0) .and. (c /= 0))
+ a = b/real(c)
+ elsewhere
+ a = 456
+ endwhere
+ end subroutine foo
+end module m
Index: gcc/tree-predcom.c
===================================================================
--- gcc/tree-predcom.c (revision 255817)
+++ gcc/tree-predcom.c (working copy)
@@ -943,6 +943,17 @@
return (*da)->pos - (*db)->pos;
}
+/* Compares two drefs A and B by their position. Callback for qsort. */
+
+static int
+order_drefs_by_pos (const void *a, const void *b)
+{
+ const dref *const da = (const dref *) a;
+ const dref *const db = (const dref *) b;
+
+ return (*da)->pos - (*db)->pos;
+}
+
/* Returns root of the CHAIN. */
static inline dref
@@ -2250,7 +2261,6 @@
bool swap = false;
chain_p new_chain;
unsigned i;
- gimple *root_stmt;
tree rslt_type = NULL_TREE;
if (ch1 == ch2)
@@ -2292,31 +2302,55 @@
new_chain->refs.safe_push (nw);
}
- new_chain->has_max_use_after = false;
- root_stmt = get_chain_root (new_chain)->stmt;
- for (i = 1; new_chain->refs.iterate (i, &nw); i++)
- {
- if (nw->distance == new_chain->length
- && !stmt_dominates_stmt_p (nw->stmt, root_stmt))
- {
- new_chain->has_max_use_after = true;
- break;
- }
- }
-
ch1->combined = true;
ch2->combined = true;
return new_chain;
}
-/* Try to combine the CHAINS. */
+/* Recursively update position information of all offspring chains to ROOT
+ chain's position information. */
static void
-try_combine_chains (vec<chain_p> *chains)
+update_pos_for_combined_chains (chain_p root)
{
+ chain_p ch1 = root->ch1, ch2 = root->ch2;
+ dref ref, ref1, ref2;
+ for (unsigned j = 0; (root->refs.iterate (j, &ref)
+ && ch1->refs.iterate (j, &ref1)
+ && ch2->refs.iterate (j, &ref2)); ++j)
+ ref1->pos = ref2->pos = ref->pos;
+
+ if (ch1->type == CT_COMBINATION)
+ update_pos_for_combined_chains (ch1);
+ if (ch2->type == CT_COMBINATION)
+ update_pos_for_combined_chains (ch2);
+}
+
+/* Returns true if statement S1 dominates statement S2. */
+
+static bool
+pcom_stmt_dominates_stmt_p (gimple *s1, gimple *s2)
+{
+ basic_block bb1 = gimple_bb (s1), bb2 = gimple_bb (s2);
+
+ if (!bb1 || s1 == s2)
+ return true;
+
+ if (bb1 == bb2)
+ return gimple_uid (s1) < gimple_uid (s2);
+
+ return dominated_by_p (CDI_DOMINATORS, bb2, bb1);
+}
+
+/* Try to combine the CHAINS in LOOP. */
+
+static void
+try_combine_chains (struct loop *loop, vec<chain_p> *chains)
+{
unsigned i, j;
chain_p ch1, ch2, cch;
auto_vec<chain_p> worklist;
+ bool combined_p = false;
FOR_EACH_VEC_ELT (*chains, i, ch1)
if (chain_can_be_combined_p (ch1))
@@ -2338,10 +2372,82 @@
{
worklist.safe_push (cch);
chains->safe_push (cch);
+ combined_p = true;
break;
}
}
}
+ if (!combined_p)
+ return;
+
+ /* Setup UID for all statements in dominance order. */
+ basic_block *bbs = get_loop_body (loop);
+ renumber_gimple_stmt_uids_in_blocks (bbs, loop->num_nodes);
+ free (bbs);
+
+ /* Re-association in combined chains may generate statements different to
+ order of references of the original chain. We need to keep references
+ of combined chain in dominance order so that all uses will be inserted
+ after definitions. Note:
+ A) This is necessary for all combined chains.
+ B) This is only necessary for ZERO distance references because other
+ references inherit value from loop carried PHIs.
+
+ We first update position information for all combined chains. */
+ dref ref;
+ for (i = 0; chains->iterate (i, &ch1); ++i)
+ {
+ if (ch1->type != CT_COMBINATION || ch1->combined)
+ continue;
+
+ for (j = 0; ch1->refs.iterate (j, &ref); ++j)
+ ref->pos = gimple_uid (ref->stmt);
+
+ update_pos_for_combined_chains (ch1);
+ }
+ /* Then sort references according to newly updated position information. */
+ for (i = 0; chains->iterate (i, &ch1); ++i)
+ {
+ if (ch1->type != CT_COMBINATION && !ch1->combined)
+ continue;
+
+ /* Find the first reference with non-ZERO distance. */
+ if (ch1->length == 0)
+ j = ch1->refs.length();
+ else
+ {
+ for (j = 0; ch1->refs.iterate (j, &ref); ++j)
+ if (ref->distance != 0)
+ break;
+ }
+
+ /* Sort all ZERO distance references by position. */
+ qsort (&ch1->refs[0], j, sizeof (ch1->refs[0]), order_drefs_by_pos);
+
+ if (ch1->combined)
+ continue;
+
+ /* For ZERO length chain, has_max_use_after must be true since root
+ combined stmt must dominates others. */
+ if (ch1->length == 0)
+ {
+ ch1->has_max_use_after = true;
+ continue;
+ }
+ /* Check if there is use at max distance after root for combined chains
+ and set flag accordingly. */
+ ch1->has_max_use_after = false;
+ gimple *root_stmt = get_chain_root (ch1)->stmt;
+ for (j = 1; ch1->refs.iterate (j, &ref); ++j)
+ {
+ if (ref->distance == ch1->length
+ && !pcom_stmt_dominates_stmt_p (ref->stmt, root_stmt))
+ {
+ ch1->has_max_use_after = true;
+ break;
+ }
+ }
+ }
}
/* Prepare initializers for CHAIN in LOOP. Returns false if this is
@@ -2490,7 +2596,7 @@
prepare_initializers (loop, chains);
/* Try to combine the chains that are always worked with together. */
- try_combine_chains (&chains);
+ try_combine_chains (loop, &chains);
if (dump_file && (dump_flags & TDF_DETAILS))
{
