We currently use subst definitions to handle the different variants of shift
count operands. Unfortunately, in the vector shift pattern the shift count
operand is used directly. Without it being adjusted for the 'subst' variants the
displacement value is omitted resulting in a wrong shift count being applied.

This patch needs to be applied to older branches as well.

Bootstrapped and regression tested on s390x (IBM z14).
Committed to mainline

gcc/ChangeLog:

2019-07-01  Andreas Krebbel  <kreb...@linux.ibm.com>

        * config/s390/vector.md:

gcc/testsuite/ChangeLog:

2019-07-01  Andreas Krebbel  <kreb...@linux.ibm.com>

        * gcc.target/s390/vector/vec-shift-2.c: New test.
---
 gcc/config/s390/vector.md                          |  2 +-
 gcc/testsuite/gcc.target/s390/vector/vec-shift-2.c | 24 ++++++++++++++++++++++
 2 files changed, 25 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/gcc.target/s390/vector/vec-shift-2.c

diff --git a/gcc/config/s390/vector.md b/gcc/config/s390/vector.md
index a2c1012..140ef47 100644
--- a/gcc/config/s390/vector.md
+++ b/gcc/config/s390/vector.md
@@ -981,7 +981,7 @@
        (VEC_SHIFTS:VI (match_operand:VI 1 "register_operand"   "v")
                       (match_operand:SI 2 "nonmemory_operand" "an")))]
   "TARGET_VX"
-  "<vec_shifts_mnem><bhfgq>\t%v0,%v1,%Y2"
+  "<vec_shifts_mnem><bhfgq>\t%v0,%v1,<addr_style_op_ops>"
   [(set_attr "op_type" "VRS")])
 
 ; Shift each element by corresponding vector element
diff --git a/gcc/testsuite/gcc.target/s390/vector/vec-shift-2.c 
b/gcc/testsuite/gcc.target/s390/vector/vec-shift-2.c
new file mode 100644
index 0000000..c7a1d93
--- /dev/null
+++ b/gcc/testsuite/gcc.target/s390/vector/vec-shift-2.c
@@ -0,0 +1,24 @@
+/* { dg-do run } */
+/* { dg-options "-O3 -mzarch -march=z13 --save-temps" } */
+
+/* { dg-final { scan-assembler-times "veslf" 1 } } */
+
+typedef __attribute__((vector_size(16))) signed int v4si;
+
+v4si __attribute__((noinline,noclone))
+shift_left_by_scalar (v4si in, int shift_count)
+{
+  return in << (3 + shift_count);
+}
+
+int
+main ()
+{
+  v4si a = { 1, 2, 3, 4 };
+  v4si result = shift_left_by_scalar (a, 1);
+
+  if (result[1] != 32)
+    __builtin_abort ();
+
+  return 0;
+}
-- 
2.7.4

Reply via email to