Two fixes for live-out SLP inductions (PR 83857)

vect_analyze_loop_operations was calling vectorizable_live_operation
for all live-out phis, which led to a bogus ncopies calculation in
the pure SLP case.  I think v_a_l_o should only be passing phis
that are vectorised using normal loop vectorisation, since
vect_slp_analyze_node_operations handles the SLP side (and knows
the correct slp_index and slp_node arguments to pass in, via
vect_analyze_stmt).

With that fixed we hit an older bug that vectorizable_live_operation
didn't handle live-out SLP inductions.  Fixed by using gimple_phi_result
rather than gimple_get_lhs for phis.

2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>

gcc/
	PR tree-optimization/83857
	* tree-vect-loop.c (vect_analyze_loop_operations): Don't call
	vectorizable_live_operation for pure SLP statements.
	(vectorizable_live_operation): Handle PHIs.

gcc/testsuite/
	PR tree-optimization/83857
	* gcc.dg/vect/pr83857.c: New test.

From-SVN: r256747

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 055e924a8b0bf6a5827c6afeb9f71182c1ff2ff7..979f5ae487144845c7f19b6ef8b619d825fa330b 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,10 @@
+2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
+
+       PR tree-optimization/83857
+       * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
+       vectorizable_live_operation for pure SLP statements.
+       (vectorizable_live_operation): Handle PHIs.
+
 2018-01-16  Richard Biener  <rguenther@suse.de>
 
        PR tree-optimization/83867

diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index ae578285db08874236b6e1670a1a8fbbd040502c..0e24ae440bd8d8627c428edd5b212931a33077b0 100644 (file)
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,8 @@
+2018-01-16  Richard Sandiford  <richard.sandiford@linaro.org>
+
+       PR tree-optimization/83857
+       * gcc.dg/vect/pr83857.c: New test.
+
 2018-01-16  Richard Biener  <rguenther@suse.de>
 
        PR tree-optimization/83867

diff --git a/gcc/testsuite/gcc.dg/vect/pr83857.c b/gcc/testsuite/gcc.dg/vect/pr83857.c
new file mode 100644 (file)
index 0000000..92e648b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/pr83857.c
@@ -0,0 +1,30 @@
+/* { dg-do run } */
+/* { dg-additional-options "-ffast-math" } */
+
+#define N 100
+
+double __attribute__ ((noinline, noclone))
+f (double *x, double y)
+{
+  double a = 0;
+  for (int i = 0; i < N; ++i)
+    {
+      a += y;
+      x[i * 2] += a;
+      x[i * 2 + 1] += a;
+    }
+  return a - y;
+}
+
+double x[N * 2];
+
+int
+main (void)
+{
+  if (f (x, 5) != (N - 1) * 5)
+    __builtin_abort ();
+  return 0;
+}
+
+/* { dg-final { scan-tree-dump "Loop contains only SLP stmts" "vect" { target vect_double } } } */
+/* { dg-final { scan-tree-dump "vectorized 1 loops" "vect" { target vect_double } } } */

diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index 64b9ce3fd2306f707a1989dcee8bce63b415c859..79b818608dff1633b9feec654174b30e163b270c 100644 (file)
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -1851,7 +1851,10 @@ vect_analyze_loop_operations (loop_vec_info loop_vinfo)
                ok = vectorizable_reduction (phi, NULL, NULL, NULL, NULL);
             }
 
-         if (ok && STMT_VINFO_LIVE_P (stmt_info))
+         /* SLP PHIs are tested by vect_slp_analyze_node_operations.  */
+         if (ok
+             && STMT_VINFO_LIVE_P (stmt_info)
+             && !PURE_SLP_STMT (stmt_info))
            ok = vectorizable_live_operation (phi, NULL, NULL, -1, NULL);
 
           if (!ok)
@@ -8217,7 +8220,11 @@ vectorizable_live_operation (gimple *stmt,
       gcc_assert (!LOOP_VINFO_FULLY_MASKED_P (loop_vinfo));
 
       /* Get the correct slp vectorized stmt.  */
-      vec_lhs = gimple_get_lhs (SLP_TREE_VEC_STMTS (slp_node)[vec_entry]);
+      gimple *vec_stmt = SLP_TREE_VEC_STMTS (slp_node)[vec_entry];
+      if (gphi *phi = dyn_cast <gphi *> (vec_stmt))
+       vec_lhs = gimple_phi_result (phi);
+      else
+       vec_lhs = gimple_get_lhs (vec_stmt);
 
       /* Get entry to use.  */
       bitstart = bitsize_int (vec_index);