[libclc] Implement erf/erfc vector function with loop since scalar function is large (#157055)

This PR reduces amdgcn--amdhsa.bc size by 1.8% and nvptx64--nvidiacl.bc
size by 4%.
Loop trip count is constant and backend can decide whether to unroll.

---------

Co-authored-by: Copilot <[email protected]>

Author: wenju-he

libclc/clc/include/clc/shared/unary_def_scalarize_loop.inc

@@ -0,0 +1,28 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include <clc/utils.h>
+
+#if __CLC_VECSIZE_OR_1 >= 2
+
+#ifndef __CLC_IMPL_FUNCTION
+#define __CLC_IMPL_FUNCTION __CLC_FUNCTION
+#endif
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION(__CLC_GENTYPE x) {
+  union {
+    __CLC_GENTYPE vec;
+    __CLC_SCALAR_GENTYPE arr[__CLC_VECSIZE_OR_1];
+  } u_x, u_result;
+  u_x.vec = x;
+  for (int i = 0; i < __CLC_VECSIZE_OR_1; ++i)
+    u_result.arr[i] = __CLC_IMPL_FUNCTION(u_x.arr[i]);
+  return u_result.vec;
+}
+
+#endif // __CLC_VECSIZE_OR_1 >= 2

libclc/clc/lib/generic/math/clc_erf.cl

@@ -507,5 +507,5 @@ _CLC_OVERLOAD _CLC_DEF half __clc_erf(half x) {
 #endif
 
 #define __CLC_FUNCTION __clc_erf
-#define __CLC_BODY <clc/shared/unary_def_scalarize.inc>
+#define __CLC_BODY <clc/shared/unary_def_scalarize_loop.inc>
 #include <clc/math/gentype.inc>

libclc/clc/lib/generic/math/clc_erfc.cl

@@ -518,5 +518,5 @@ _CLC_OVERLOAD _CLC_DEF half __clc_erfc(half x) {
 #endif
 
 #define __CLC_FUNCTION __clc_erfc
-#define __CLC_BODY <clc/shared/unary_def_scalarize.inc>
+#define __CLC_BODY <clc/shared/unary_def_scalarize_loop.inc>
 #include <clc/math/gentype.inc>