[ARM] Guard the declarations of f16 to f32 vcvt intrinsics in arm_neon.h by testing __ARM_FP

[sbaranga]

Summary

Conversions between float and half are only available when the
taraget has the half-precision extension. Guard these intrinsics
so that they don't cause crashes in the backend.

Fixes PR27550.

[t.p.northover]

Couple of issues, as far as I can see:

include/clang/Basic/arm_neon.td

710–711

I think these _high variants are supposed to be AArch64-only.

test/CodeGen/arm-negative-fp16.c

1 ↗

(On Diff #55423)

This is a Sema test.

[sbaranga]

Don't change the AArch64 intrinsics and move the test to Sema.

include/clang/Basic/arm_neon.td

710–711

Thanks for catching this!

[t.p.northover]

Looks good. Thanks for updating the patch.

Tim.

[sbaranga]

Thanks, r268047!

Cheers,
Silviu

include/clang/Basic/arm_neon.td

 let InstName = "vmov" in {
 def VGET_HIGH : NoTestOpInst<"vget_high", "dk", "csilhfUcUsUiUlPcPs", OP_HI>;
 def VGET_LOW  : NoTestOpInst<"vget_low", "dk", "csilhfUcUsUiUlPcPs", OP_LO>;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.22 Converting vectors
 
+let ArchGuard = "(__ARM_FP & 2)" in {
   def VCVT_F16_F32 : SInst<"vcvt_f16_f32", "md", "Hf">;
   def VCVT_F32_F16 : SInst<"vcvt_f32_f16", "wd", "h">;
+}
 
 def VCVT_S32     : SInst<"vcvt_s32", "xd",  "fQf">;
 def VCVT_U32     : SInst<"vcvt_u32", "ud",  "fQf">;
 def VCVT_F32     : SInst<"vcvt_f32", "fd",  "iUiQiQUi">;
 let isVCVT_N = 1 in {
 def VCVT_N_S32   : SInst<"vcvt_n_s32", "xdi", "fQf">;
 def VCVT_N_U32   : SInst<"vcvt_n_u32", "udi", "fQf">;
 def VCVT_N_F32   : SInst<"vcvt_n_f32", "fdi", "iUiQiQUi">;
 }

test/Sema/arm-no-fp16.c

+// RUN: %clang_cc1 -triple thumbv7-none-eabi %s -target-feature +neon -target-feature -fp16 -fsyntax-only -verify
+
+#include <arm_neon.h>
+
+float16x4_t test_vcvt_f16_f32(float32x4_t a) {
+  return vcvt_f16_f32(a); // expected-warning{{implicit declaration of function 'vcvt_f16_f32'}}  expected-error{{returning 'int' from a function with incompatible result type 'float16x4_t'}}
+}
+
+float32x4_t test_vcvt_f32_f16(float16x4_t a) {
+  return vcvt_f32_f16(a); // expected-warning{{implicit declaration of function 'vcvt_f32_f16'}} expected-error{{returning 'int' from a function with incompatible result type 'float32x4_t'}}
+}