maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[AArch64] Add support for NEON scalar floating-point reciprocal estimate, 

reciprocal exponent, and reciprocal square root estimate instructions.

llvm-svn: 192243
This commit is contained in:
Chad Rosier 2013-10-08 22:09:29 +00:00
parent 9849cc6696
commit 0a903478c6
3 changed files with 63 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
clang/include/clang/Basic/arm_neon.td
View File

@ -777,4 +777,16 @@ def SCALAR_SCVTFD : SInst<"vcvt_f64", "os", "Sl">;
def SCALAR_UCVTFS : SInst<"vcvt_f32", "ys", "SUi">;
def SCALAR_UCVTFD : SInst<"vcvt_f64", "os", "SUl">;
////////////////////////////////////////////////////////////////////////////////
// Scalar Floating-point Reciprocal Estimate
def SCALAR_FRECPE : IInst<"vrecpe", "ss", "SfSd">;
////////////////////////////////////////////////////////////////////////////////
// Scalar Floating-point Reciprocal Exponent
def SCALAR_FRECPX : IInst<"vrecpx", "ss", "SfSd">;
////////////////////////////////////////////////////////////////////////////////
// Scalar Floating-point Reciprocal Square Root Estimate
def SCALAR_FRSQRTE : IInst<"vrsqrte", "ss", "SfSd">;
}

15
clang/lib/CodeGen/CGBuiltin.cpp
View File

@ -1985,6 +1985,21 @@ static Value *EmitAArch64ScalarBuiltinExpr(CodeGenFunction &CGF,
case AArch64::BI__builtin_neon_vcvtd_f64_u64:
Int = Intrinsic::aarch64_neon_vcvtf64_u64,
s = "vcvtf"; OverloadInt = false; break;
// Scalar Floating-point Reciprocal Estimate
case AArch64::BI__builtin_neon_vrecpes_f32:
case AArch64::BI__builtin_neon_vrecped_f64:
Int = Intrinsic::arm_neon_vrecpe;
s = "vrecpe"; OverloadInt = true; break;
// Scalar Floating-point Reciprocal Exponent
case AArch64::BI__builtin_neon_vrecpxs_f32:
case AArch64::BI__builtin_neon_vrecpxd_f64:
Int = Intrinsic::aarch64_neon_vrecpx;
s = "vrecpx"; OverloadInt = true; break;
// Scalar Floating-point Reciprocal Square Root Estimate
case AArch64::BI__builtin_neon_vrsqrtes_f32:
case AArch64::BI__builtin_neon_vrsqrted_f64:
Int = Intrinsic::arm_neon_vrsqrte;
s = "vrsqrte"; OverloadInt = true; break;
}
if (!Int)

36
clang/test/CodeGen/aarch64-neon-intrinsics.c
View File

@ -5621,3 +5621,39 @@ float64_t test_vcvtd_f64_u64(uint64_t a) {
// CHECK: ucvtf {{d[0-9]+}}, {{d[0-9]+}}
return (float64_t)vcvtd_f64_u64(a);
}
float32_t test_vrecpes_f32(float32_t a) {
// CHECK: test_vrecpes_f32
// CHECK: frecpe {{s[0-9]+}}, {{s[0-9]+}}
return vrecpes_f32(a);
}
float64_t test_vrecped_f64(float64_t a) {
// CHECK: test_vrecped_f64
// CHECK: frecpe {{d[0-9]+}}, {{d[0-9]+}}
return vrecped_f64(a);
}
float32_t test_vrecpxs_f32(float32_t a) {
// CHECK: test_vrecpxs_f32
// CHECK: frecpx {{s[0-9]+}}, {{s[0-9]+}}
return vrecpxs_f32(a);
}
float64_t test_vrecpxd_f64(float64_t a) {
// CHECK: test_vrecpxd_f64
// CHECK: frecpx {{d[0-9]+}}, {{d[0-9]+}}
return vrecpxd_f64(a);
}
float32_t test_vrsqrtes_f32(float32_t a) {
// CHECK: vrsqrtes_f32
// CHECK: frsqrte {{s[0-9]+}}, {{s[0-9]+}}
return vrsqrtes_f32(a);
}
float64_t test_vrsqrted_f64(float64_t a) {
// CHECK: vrsqrted_f64
// CHECK: frsqrte {{d[0-9]+}}, {{d[0-9]+}}
return vrsqrted_f64(a);
}