AArch64: Add missing scalar pair intrinsics.

E.g. "float32_t vaddv_f32(float32x2_t a)" to be matched into "faddp s0, v1.2s".

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@196199 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/arm_neon.td b/include/clang/Basic/arm_neon.td
index 3706ff6d0553051641d62820b191844b92213aba..bc6e23362d1047066d5ae2ebd56e5ca46f56c309 100644 (file)
--- a/include/clang/Basic/arm_neon.td
+++ b/include/clang/Basic/arm_neon.td
@@ -779,7 +779,7 @@ def FMINNMP : SInst<"vpminnm", "ddd", "fQfQd">;
 ////////////////////////////////////////////////////////////////////////////////
 // Pairwise Addition
 // With additional Qc Qs Qi QUc QUs QUi Qf Qd types.
-def ADDP  : IInst<"vpadd", "ddd", "csiUcUsUifQcQsQiQUcQUsQUiQfQd">;
+def ADDP  : IInst<"vpadd", "ddd", "csiUcUsUifQcQsQiQlQUcQUsQUiQUlQfQd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Shifts by constant
@@ -951,11 +951,11 @@ def VMULX_LANEQ : IOpInst<"vmulx_laneq", "ddji", "fQfQd", OP_MULX_LN>;
 ////////////////////////////////////////////////////////////////////////////////
 // Across vectors class
 def VADDLV  : SInst<"vaddlv", "rd", "csiUcUsUiQcQsQiQUcQUsQUi">;
-def VMAXV   : SInst<"vmaxv", "sd", "csiUcUsUiQcQsQiQUcQUsQUiQf">;
-def VMINV   : SInst<"vminv", "sd", "csiUcUsUiQcQsQiQUcQUsQUiQf">;
-def VADDV   : SInst<"vaddv", "sd", "csiUcUsUiQcQsQiQUcQUsQUi">;
-def FMAXNMV : SInst<"vmaxnmv", "sd", "Qf">;
-def FMINNMV : SInst<"vminnmv", "sd", "Qf">;
+def VMAXV   : SInst<"vmaxv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
+def VMINV   : SInst<"vminv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
+def VADDV   : SInst<"vaddv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQdQlQUl">;
+def FMAXNMV : SInst<"vmaxnmv", "sd", "fQfQd">;
+def FMINNMV : SInst<"vminnmv", "sd", "fQfQd">;
  
 ////////////////////////////////////////////////////////////////////////////////
 // Newly added Vector Extract for f64
@@ -1104,7 +1104,7 @@ def SCALAR_FCVTZU_N_U64 : SInst<"vcvt_n_u64", "bsi", "Sd">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Reduce Pairwise Addition (Scalar and Floating Point)
-def SCALAR_ADDP  : SInst<"vpadd", "sd", "SfSHlSHd">;
+def SCALAR_ADDP  : SInst<"vpadd", "sd", "SfSHlSHdSHUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Reduce Floating Point Pairwise Max/Min

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 3031dcb395861c63369bf1d6eaa601d15ba22769..f532545407ab8fed09a9ad2a0fb621176f9d8e9b 100644 (file)
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -1985,6 +1985,7 @@ static Value *EmitAArch64ScalarBuiltinExpr(CodeGenFunction &CGF,
     s = "vqrshlu"; OverloadInt = true; break;
   // Scalar Reduce Pairwise Add
   case AArch64::BI__builtin_neon_vpaddd_s64:
+  case AArch64::BI__builtin_neon_vpaddd_u64:
     Int = Intrinsic::aarch64_neon_vpadd; s = "vpadd";
     break;
   case AArch64::BI__builtin_neon_vpadds_f32:
@@ -2069,23 +2070,36 @@ static Value *EmitAArch64ScalarBuiltinExpr(CodeGenFunction &CGF,
   case AArch64::BI__builtin_neon_vaddvq_s8:
   case AArch64::BI__builtin_neon_vaddvq_s16:
   case AArch64::BI__builtin_neon_vaddvq_s32:
+  case AArch64::BI__builtin_neon_vaddvq_s64:
   case AArch64::BI__builtin_neon_vaddv_u8:
   case AArch64::BI__builtin_neon_vaddv_u16:
   case AArch64::BI__builtin_neon_vaddvq_u8:
   case AArch64::BI__builtin_neon_vaddvq_u16:
   case AArch64::BI__builtin_neon_vaddvq_u32:
+  case AArch64::BI__builtin_neon_vaddvq_u64:
+  case AArch64::BI__builtin_neon_vaddv_f32:
+  case AArch64::BI__builtin_neon_vaddvq_f32:
+  case AArch64::BI__builtin_neon_vaddvq_f64:
     Int = Intrinsic::aarch64_neon_vaddv;
     AcrossVec = true; ExtendEle = false; s = "vaddv"; break;      
+  case AArch64::BI__builtin_neon_vmaxv_f32:
   case AArch64::BI__builtin_neon_vmaxvq_f32:
+  case AArch64::BI__builtin_neon_vmaxvq_f64:
     Int = Intrinsic::aarch64_neon_vmaxv;
     AcrossVec = true; ExtendEle = false; s = "vmaxv"; break;
+  case AArch64::BI__builtin_neon_vminv_f32:
   case AArch64::BI__builtin_neon_vminvq_f32:
+  case AArch64::BI__builtin_neon_vminvq_f64:
     Int = Intrinsic::aarch64_neon_vminv;
     AcrossVec = true; ExtendEle = false; s = "vminv"; break;
+  case AArch64::BI__builtin_neon_vmaxnmv_f32:
   case AArch64::BI__builtin_neon_vmaxnmvq_f32:
+  case AArch64::BI__builtin_neon_vmaxnmvq_f64:
     Int = Intrinsic::aarch64_neon_vmaxnmv;
     AcrossVec = true; ExtendEle = false; s = "vmaxnmv"; break;
+  case AArch64::BI__builtin_neon_vminnmv_f32:
   case AArch64::BI__builtin_neon_vminnmvq_f32:
+  case AArch64::BI__builtin_neon_vminnmvq_f64:
     Int = Intrinsic::aarch64_neon_vminnmv;
     AcrossVec = true; ExtendEle = false; s = "vminnmv"; break;
   // Scalar Integer Saturating Doubling Multiply Half High

diff --git a/test/CodeGen/aarch64-neon-intrinsics.c b/test/CodeGen/aarch64-neon-intrinsics.c
index d9c1b8f96c40479f934cfda81d0bdb26f6bed5ed..f0733c96266cef6070378488c0887561dc00cf10 100644 (file)
--- a/test/CodeGen/aarch64-neon-intrinsics.c
+++ b/test/CodeGen/aarch64-neon-intrinsics.c
@@ -11272,3 +11272,100 @@ int64x1_t test_vneg_s64(int64x1_t a) {
   return vneg_s64(a);
   // CHECK: neg d{{[0-9]+}}, d{{[0-9]+}}
 }
+
+float32_t test_vaddv_f32(float32x2_t a) {
+  // CHECK-LABEL: test_vaddv_f32
+  return vaddv_f32(a);
+  // CHECK: faddp {{s[0-9]+}}, {{v[0-9]+}}.2s
+}
+
+float32_t test_vaddvq_f32(float32x4_t a) {
+  // CHECK-LABEL: test_vaddvq_f32
+  return vaddvq_f32(a);
+  // CHECK: faddp {{v[0-9]+}}.4s, {{v[0-9]+}}.4s, {{v[0-9]+}}.4s
+  // CHECK: faddp {{s[0-9]+}}, {{v[0-9]+}}.2s
+}
+
+float64_t test_vaddvq_f64(float64x2_t a) {
+  // CHECK-LABEL: test_vaddvq_f64
+  return vaddvq_f64(a);
+  // CHECK: faddp {{d[0-9]+}}, {{v[0-9]+}}.2d
+}
+
+float32_t test_vmaxv_f32(float32x2_t a) {
+  // CHECK-LABEL: test_vmaxv_f32
+  return vmaxv_f32(a);
+  // CHECK: fmaxp {{s[0-9]+}}, {{v[0-9]+}}.2s
+}
+
+float64_t test_vmaxvq_f64(float64x2_t a) {
+  // CHECK-LABEL: test_vmaxvq_f64
+  return vmaxvq_f64(a);
+  // CHECK: fmaxp {{d[0-9]+}}, {{v[0-9]+}}.2d
+}
+
+float32_t test_vminv_f32(float32x2_t a) {
+  // CHECK-LABEL: test_vminv_f32
+  return vminv_f32(a);
+  // CHECK: fminp {{s[0-9]+}}, {{v[0-9]+}}.2s
+}
+
+float64_t test_vminvq_f64(float64x2_t a) {
+  // CHECK-LABEL: test_vminvq_f64
+  return vminvq_f64(a);
+  // CHECK: fminp {{d[0-9]+}}, {{v[0-9]+}}.2d
+}
+
+float64_t test_vmaxnmvq_f64(float64x2_t a) {
+  // CHECK-LABEL: test_vmaxnmvq_f64
+  return vmaxnmvq_f64(a);
+  // CHECK: fmaxnmp {{d[0-9]+}}, {{v[0-9]+}}.2d
+}
+
+float32_t test_vmaxnmv_f32(float32x2_t a) {
+  // CHECK-LABEL: test_vmaxnmv_f32
+  return vmaxnmv_f32(a);
+  // CHECK: fmaxnmp {{s[0-9]+}}, {{v[0-9]+}}.2s
+}
+
+float64_t test_vminnmvq_f64(float64x2_t a) {
+  // CHECK-LABEL: test_vminnmvq_f64
+  return vminnmvq_f64(a);
+  // CHECK: fminnmp {{d[0-9]+}}, {{v[0-9]+}}.2d
+}
+
+float32_t test_vminnmv_f32(float32x2_t a) {
+  // CHECK-LABEL: test_vminnmv_f32
+  return vminnmv_f32(a);
+  // CHECK: fminnmp {{s[0-9]+}}, {{v[0-9]+}}.2s
+}
+
+int64x2_t test_vpaddq_s64(int64x2_t a, int64x2_t b) {
+  // CHECK-LABEL: test_vpaddq_s64
+  return vpaddq_s64(a, b);
+  // CHECK: addp {{v[0-9]+}}.2d, {{v[0-9]+}}.2d, {{v[0-9]+}}.2d
+}
+
+uint64x2_t test_vpaddq_u64(uint64x2_t a, uint64x2_t b) {
+  // CHECK-LABEL: test_vpaddq_u64
+  return vpaddq_u64(a, b);
+  // CHECK: addp {{v[0-9]+}}.2d, {{v[0-9]+}}.2d, {{v[0-9]+}}.2d
+}
+
+uint64_t test_vpaddd_u64(uint64x2_t a) {
+  // CHECK-LABEL: test_vpaddd_u64
+  return vpaddd_u64(a);
+  // CHECK: addp {{d[0-9]+}}, {{v[0-9]+}}.2d
+}
+
+int64_t test_vaddvq_s64(int64x2_t a) {
+  // CHECK-LABEL: test_vaddvq_s64
+  return vaddvq_s64(a);
+  // CHECK: addp {{d[0-9]+}}, {{v[0-9]+}}.2d
+}
+
+uint64_t test_vaddvq_u64(uint64x2_t a) {
+  // CHECK-LABEL: test_vaddvq_u64
+  return vaddvq_u64(a);
+  // CHECK: addp {{d[0-9]+}}, {{v[0-9]+}}.2d
+}