[x86] Adding support for some missing intrinsics: _castf32_u32, _castf64_u64, _castu32_f32, _castu64_f64

[yubing]

Summary

Adding support for some missing intrinsics:
_castf32_u32, _castf64_u64, _castu32_f32, _castu64_f64

[RKSimon]

clang/lib/Headers/ia32intrin.h

198 ↗

(On Diff #218845)

There's an unofficial agreement that new x86 intrinsics should have doxygen comments - someday we'll get around to adding them to the existing ones as well.....

207 ↗

(On Diff #218845)

Shouldn't the unsigned long long cases be hidden by x86_64 wrappers?

clang/test/CodeGen/miscellaneous-builtins.c

1 ↗

(On Diff #218845)

32-bit target tests?

32 ↗

(On Diff #218845)

newline please

[RKSimon]

Also, as its x86 specific miscellaneous-builtins.c should be called x86-builtins.c (or similar).

[craig.topper]

clang/lib/Headers/ia32intrin.h

207 ↗

(On Diff #218845)

icc doesn't restrict this to 64-bit targets. On 64-bit targets it corresponds to movq, but on other targets we can just lower to a 64-bit store and two 32-bit loads in the worst case.

[RKSimon]

clang/lib/Headers/ia32intrin.h

207 ↗

(On Diff #218845)

Thanks - in which case we should definitely add 32-bit triple tests

[RKSimon]

Please rename miscellaneous-builtins.c to x86-builtins.c

[RKSimon]

LGTM

cfe/trunk/lib/Headers/ia32intrin.h

 
 static __inline__ void __attribute__((__always_inline__, __nodebug__))
 __writeeflags(unsigned int __f)
 {
   __builtin_ia32_writeeflags_u32(__f);
 }
 #endif /* !__x86_64__ */
 
+/** Cast a 32-bit float value to a 32-bit unsigned integer value
+ *
+ *  \headerfile <x86intrin.h>
+ *  This intrinsic corresponds to the <c> VMOVD / MOVD </c> instruction in x86_64,
+ *  and corresponds to the <c> VMOVL / MOVL </c> instruction in ia32.
+ *
+ *  \param __A
+ *     A 32-bit float value.
+ *  \returns a 32-bit unsigned integer containing the converted value.
+ */
+static __inline__ unsigned int __attribute__((__always_inline__))
+_castf32_u32(float __A) {
+  unsigned int D;
+  __builtin_memcpy(&D, &__A, sizeof(__A));
+  return D;
+}
+
+/** Cast a 64-bit float value to a 64-bit unsigned integer value
+ *
+ *  \headerfile <x86intrin.h>
+ *  This intrinsic corresponds to the <c> VMOVQ / MOVQ </c> instruction in x86_64,
+ *  and corresponds to the <c> VMOVL / MOVL </c> instruction in ia32.
+ *
+ *  \param __A
+ *     A 64-bit float value.
+ *  \returns a 64-bit unsigned integer containing the converted value.
+ */
+static __inline__ unsigned long long __attribute__((__always_inline__))
+_castf64_u64(double __A) {
+  unsigned long long D;
+  __builtin_memcpy(&D, &__A, sizeof(__A));
+  return D;
+}
+
+/** Cast a 32-bit unsigned integer value to a 32-bit float value
+ *
+ *  \headerfile <x86intrin.h>
+ *  This intrinsic corresponds to the <c> VMOVQ / MOVQ </c> instruction in x86_64,
+ *  and corresponds to the <c> FLDS </c> instruction in ia32.
+ *
+ *  \param __A
+ *     A 32-bit unsigned integer value.
+ *  \returns a 32-bit float value containing the converted value.
+ */
+static __inline__ float __attribute__((__always_inline__))
+_castu32_f32(unsigned int __A) {
+  float D;
+  __builtin_memcpy(&D, &__A, sizeof(__A));
+  return D;
+}
+
+/** Cast a 64-bit unsigned integer value to a 64-bit float value
+ *
+ *  \headerfile <x86intrin.h>
+ *  This intrinsic corresponds to the <c> VMOVQ / MOVQ </c> instruction in x86_64,
+ *  and corresponds to the <c> FLDL </c> instruction in ia32.
+ *
+ *  \param __A
+ *     A 64-bit unsigned integer value.
+ *  \returns a 64-bit float value containing the converted value.
+ */
+static __inline__ double __attribute__((__always_inline__))
+_castu64_f64(unsigned long long __A) {
+  double D;
+  __builtin_memcpy(&D, &__A, sizeof(__A));
+  return D;
+}
+
 /** Adds the unsigned integer operand to the CRC-32C checksum of the
  *     unsigned char operand.
  *
  *  \headerfile <x86intrin.h>
  *
  *  This intrinsic corresponds to the <c> CRC32B </c> instruction.
  *
  *  \param __C

cfe/trunk/test/CodeGen/x86-builtins.c

+// RUN: %clang_cc1 -ffreestanding %s -triple=x86_64-unknown-unknown -emit-llvm -o - -Wall -Werror | FileCheck %s -check-prefix=CHECK-64
+// RUN: %clang_cc1 -ffreestanding %s -triple=i386-unknown-unknown -emit-llvm -o - -Wall -Werror | FileCheck %s -check-prefix=CHECK-32
+
+#include <x86intrin.h>
+
+unsigned int test_castf32_u32 (float __A){
+  // CHECK-64-LABEL: @test_castf32_u32
+  // CHECK-64: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.*}}, i8* align 4 %{{.*}}, i64 4, i1 false)
+  // CHECK-64: %{{.*}} = load i32, i32* %{{.*}}, align 4
+  // CHECK-32-LABEL: @test_castf32_u32
+  // CHECK-32: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %{{.*}}, i8* align 4 %{{.*}}, i32 4, i1 false)
+  // CHECK-32: %{{.*}} = load i32, i32* %{{.*}}, align 4
+  return _castf32_u32(__A);
+}
+
+unsigned long long test_castf64_u64 (double __A){
+  // CHECK-64-LABEL: @test_castf64_u64
+  // CHECK-64: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %{{.*}}, i8* align 8 %{{.*}}, i64 8, i1 false)
+  // CHECK-64: %{{.*}} = load i64, i64* %{{.*}}, align 8
+  // CHECK-32-LABEL: @test_castf64_u64
+  // CHECK-32: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 8 %{{.*}}, i8* align 8 %{{.*}}, i32 8, i1 false)
+  // CHECK-32: %{{.*}} = load i64, i64* %{{.*}}, align 8
+  return _castf64_u64(__A);
+}
+
+float test_castu32_f32 (unsigned int __A){
+  // CHECK-64-LABEL: @test_castu32_f32
+  // CHECK-64: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 %{{.*}}, i8* align 4 %{{.*}}, i64 4, i1 false)
+  // CHECK-64: %{{.*}} = load float, float* %{{.*}}, align 4
+  // CHECK-32-LABEL: @test_castu32_f32
+  // CHECK-32: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %{{.*}}, i8* align 4 %{{.*}}, i32 4, i1 false)
+  // CHECK-32: %{{.*}} = load float, float* %{{.*}}, align 4
+  return _castu32_f32(__A);
+}
+
+double test_castu64_f64 (unsigned long long __A){
+  // CHECK-64-LABEL: @test_castu64_f64
+  // CHECK-64: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 %{{.*}}, i8* align 8 %{{.*}}, i64 8, i1 false)
+  // CHECK-64: %{{.*}} = load double, double* %{{.*}}, align 8
+  // CHECK-32-LABEL: @test_castu64_f64
+  // CHECK-32: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 8 %{{.*}}, i8* align 8 %{{.*}}, i32 8, i1 false)
+  // CHECK-32: %{{.*}} = load double, double* %{{.*}}, align 8
+  return _castu64_f64(__A);
+}
+