From: Richard Sandiford <richard.sandiford@linaro.org>
Date: Wed, 10 Jan 2018 13:07:54 +0000 (+0000)
Subject: Don't use permutes for single-element accesses (PR83753)
X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=6737facbb3c53a1f0158b05e4116c161ed9bc319;p=gcc.git

Don't use permutes for single-element accesses (PR83753)

After cunrolling the inner loop, the remaining loop in the testcase
has a single 32-bit access and a group of 64-bit accesses.  We first
try to vectorise at 128 bits (VF 4), but decide not to for cost reasons.
We then try with 64 bits (VF 2) instead.  This means that the group
of 64-bit accesses uses a single-element vector, which is deliberately
supported as of r251538.  We then try to create "permutes" for these
single-element vectors and fall foul of:

	      for (i = 0; i < 6; i++)
		sel[i] += exact_div (nelt, 2);

in vect_grouped_store_supported, since nelt==1.

Maybe we shouldn't even be trying to vectorise statements in the
single-element case, and instead just copy the scalar statement
for each member of the group.  But until then, this patch treats
non-strided grouped accesses as VMAT_CONTIGUOUS if no permutation
is necessary.

2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
	PR tree-optimization/83753
	* tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
	for non-strided grouped accesses if the number of elements is 1.

gcc/testsuite/
	PR tree-optimization/83753
	* gcc.dg/torture/pr83753.c: New test.

From-SVN: r256427
---

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index a30e2854555..f0e9e5948a2 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,9 @@
+2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
+
+	PR tree-optimization/83753
+	* tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
+	for non-strided grouped accesses if the number of elements is 1.
+
 2018-01-10  Jan Hubicka  <hubicka@ucw.cz>
 
 	PR target/81616
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 32a1460a35a..2fda0178a97 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2018-01-10  Richard Sandiford  <richard.sandiford@linaro.org>
+
+	PR tree-optimization/83753
+	* gcc.dg/torture/pr83753.c: New test.
+
 2018-01-09  Jan Hubicka  <hubicka@ucw.cz>
 
 	* gcc.target/i386/avx2-gather-1.c: Add -march.
diff --git a/gcc/testsuite/gcc.dg/torture/pr83753.c b/gcc/testsuite/gcc.dg/torture/pr83753.c
new file mode 100644
index 00000000000..54aafc0b321
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/torture/pr83753.c
@@ -0,0 +1,19 @@
+/* { dg-do compile } */
+/* { dg-options "-mcpu=xgene1" { target aarch64*-*-* } } */
+
+typedef struct {
+  int m1[10];
+  double m2[10][8];
+} blah;
+
+void
+foo (blah *info) {
+  int i, d;
+
+  for (d=0; d<10; d++) {
+    info->m1[d] = 0;
+    info->m2[d][0] = 1;
+    for (i=1; i<8; i++)
+      info->m2[d][i] = 2;
+  }
+}
diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
index a4f2d713959..819a981d57a 100644
--- a/gcc/tree-vect-stmts.c
+++ b/gcc/tree-vect-stmts.c
@@ -1849,10 +1849,16 @@ get_group_load_store_type (gimple *stmt, tree vectype, bool slp,
 	  && (can_overrun_p || !would_overrun_p)
 	  && compare_step_with_zero (stmt) > 0)
 	{
-	  /* First try using LOAD/STORE_LANES.  */
-	  if (vls_type == VLS_LOAD
-	      ? vect_load_lanes_supported (vectype, group_size)
-	      : vect_store_lanes_supported (vectype, group_size))
+	  /* First cope with the degenerate case of a single-element
+	     vector.  */
+	  if (known_eq (TYPE_VECTOR_SUBPARTS (vectype), 1U))
+	    *memory_access_type = VMAT_CONTIGUOUS;
+
+	  /* Otherwise try using LOAD/STORE_LANES.  */
+	  if (*memory_access_type == VMAT_ELEMENTWISE
+	      && (vls_type == VLS_LOAD
+		  ? vect_load_lanes_supported (vectype, group_size)
+		  : vect_store_lanes_supported (vectype, group_size)))
 	    {
 	      *memory_access_type = VMAT_LOAD_STORE_LANES;
 	      overrun_p = would_overrun_p;