From: Chad Rosier <mcrosier@codeaurora.org>
Date: Tue, 8 Oct 2013 22:09:29 +0000 (+0000)
Subject: [AArch64] Add support for NEON scalar floating-point reciprocal estimate,
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=3cb333070d35c9a6843e4211c9a74322f6c126ec;p=clang

[AArch64] Add support for NEON scalar floating-point reciprocal estimate,
reciprocal exponent, and reciprocal square root estimate instructions.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@192243 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/include/clang/Basic/arm_neon.td b/include/clang/Basic/arm_neon.td
index 6228bdc043..01934ae780 100644
--- a/include/clang/Basic/arm_neon.td
+++ b/include/clang/Basic/arm_neon.td
@@ -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">;
+
 }
diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index a2b0bbc282..8378d99171 100644
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -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)
diff --git a/test/CodeGen/aarch64-neon-intrinsics.c b/test/CodeGen/aarch64-neon-intrinsics.c
index d672f09600..bc071bc490 100644
--- a/test/CodeGen/aarch64-neon-intrinsics.c
+++ b/test/CodeGen/aarch64-neon-intrinsics.c
@@ -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);
+}