[PATCH][GCC 13] middle-end: fix missing converts in vec_cond_expr simplification [PR112469]

[Alex Coplan]

From: Richard Biener <rguent...@suse.de>

The following avoids type inconsistencies in .COND_op generated by
simplifications of VEC_COND_EXPRs.

This is a backport of r14-5317-g20aa06490ab57da7729a24bae7c4ec2f5918ec91
but with a testcase that triggered an ICE on the 13 branch (with the
same root cause as the original PR).  It isn't an exact cherry-pick
because some of the patterns that were patched in the original fix don't
exist in GCC 13.

Bootstrapped/regtested on aarch64-linux-gnu, arm-linux-gnueabihf, and
x86_64-linux-gnu.  OK for the 13 branch?

Thanks,
Alex

gcc/ChangeLog:

        PR middle-end/112469
        * match.pd (cond ? op a : b -> .COND_op (cond, a, b)): Add
        missing view_converts.

gcc/testsuite/ChangeLog:

        PR middle-end/112469
        * gcc.target/aarch64/sve/pr112469.c: New test.

Co-Authored-By: Alex Coplan <alex.cop...@arm.com>

diff --git a/gcc/match.pd b/gcc/match.pd
index d3af36f8de3..bbcc6587569 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -7934,13 +7934,13 @@ and,
    (with { tree op_type = TREE_TYPE (@3); }
     (if (vectorized_internal_fn_supported_p (as_internal_fn (cond_op), op_type)
         && is_truth_type_for (op_type, TREE_TYPE (@0)))
-     (cond_op @0 @1 @2))))
+     (cond_op @0 (view_convert @1) @2))))
  (simplify
   (vec_cond @0 @1 (view_convert? (uncond_op@3 @2)))
    (with { tree op_type = TREE_TYPE (@3); }
     (if (vectorized_internal_fn_supported_p (as_internal_fn (cond_op), op_type)
         && is_truth_type_for (op_type, TREE_TYPE (@0)))
-     (cond_op (bit_not @0) @2 @1)))))
+     (cond_op (bit_not @0) (view_convert @2) @1)))))
 
 /* Simplify:
 
diff --git a/gcc/testsuite/gcc.target/aarch64/sve/pr112469.c 
b/gcc/testsuite/gcc.target/aarch64/sve/pr112469.c
new file mode 100644
index 00000000000..3c347b2e986
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve/pr112469.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-options "-O2" } */
+_Float16 in[128];
+short out[128];
+void foo(void) {
+  for (int i = 0; i < 128; i++) {
+    _Float16 x = in[i];
+    _Float16 y = x ? -x : 0.0;
+    short dst;
+    __builtin_memcpy (&dst, &y, sizeof(dst));
+    out[i] = dst;
+  }
+}