[AArch64][Clang] Implement ACLE rintn intrinsics (llvm#66112)

This patch adds support for two missing ACLE intrinsics for floating
point round with ties to even:

- rintn
- rintnf

These are specified in ACLE section 8.6:
[https://arm-software.github.io/acle/main/acle.html#floating-point-data-processing-intrinsics]

clang/lib/Headers/arm_acle.h

@@ -592,6 +592,21 @@ __smusdx(int16x2_t __a, int16x2_t __b) {
 }
 #endif
 
+/* 8.6 Floating-point data-processing intrinsics */
+#if (defined(__ARM_FEATURE_DIRECTED_ROUNDING)    &&                         \
+  (__ARM_FEATURE_DIRECTED_ROUNDING))             &&                         \
+  (defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE)
+static __inline__ double __attribute__((__always_inline__, __nodebug__))
+__rintn(double __a) {
+  return __builtin_roundeven(__a);
+}
+
+static __inline__ float __attribute__((__always_inline__, __nodebug__))
+__rintnf(float __a) {
+  return __builtin_roundevenf(__a);
+}
+#endif
+
 /* 9.7 CRC32 intrinsics */
 #if (defined(__ARM_FEATURE_CRC32) && __ARM_FEATURE_CRC32) ||                   \
     (defined(__ARM_64BIT_STATE) && __ARM_64BIT_STATE)

clang/test/CodeGen/arm_acle.c

@@ -1690,6 +1690,22 @@ int32_t test_jcvt(double v) {
 }
 #endif
 
+#if defined(__ARM_FEATURE_DIRECTED_ROUNDING) && defined(__ARM_64BIT_STATE)
+
+// AArch64-LABEL: @test_rintn(
+// AArch64-NEXT:  entry:
+// AArch64-NEXT:    call double @llvm.roundeven.f64(double [[TMP0:%.*]])
+double test_rintn(double a) {
+  return __rintn(a);
+}
+
+// AArch64-LABEL: @test_rintnf(
+// AArch64-NEXT: entry:
+// AArch64-NEXT:      call float @llvm.roundeven.f32(float [[TMP0:%.*]])
+float test_rintnf(float b) {
+  return __rintnf(b);
+}
+#endif
 
 #if defined(__ARM_64BIT_STATE) && defined(__ARM_FEATURE_RNG)