i386: Fix vec_unpacks_float_lo_v4si operand constraint [PR104469]
Commit Message
2022-02-10 Uroš Bizjak <ubizjak@gmail.com>
gcc/ChangeLog:
PR target/104469
* config/i386/sse.md (vec_unpacks_float_lo_v4si):
Change operand 1 constraint to register_operand.
gcc/testsuite/ChangeLog:
PR target/104469
* gcc.target/i386/pr104469.c: New test.
Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.
Pushed to master and release branches.
Uros.
@@ -9223,7 +9223,7 @@ (define_expand "vec_unpacks_float_lo_v4si"
(define_expand "vec_unpacks_float_hi_v8si"
[(set (match_dup 2)
(vec_select:V4SI
- (match_operand:V8SI 1 "vector_operand")
+ (match_operand:V8SI 1 "register_operand")
(parallel [(const_int 4) (const_int 5)
(const_int 6) (const_int 7)])))
(set (match_operand:V4DF 0 "register_operand")
new file mode 100644
@@ -0,0 +1,12 @@
+/* PR target/104469 */
+/* { dg-do compile } */
+/* { dg-options "-mavx512f" } */
+
+typedef double __attribute__((__vector_size__ (64))) F;
+typedef int __attribute__((__vector_size__ (32))) V;
+
+F
+foo (V v)
+{
+ return __builtin_convertvector (v, F);
+}