From a48c4a87034a9478b73c25dada9921d1aa16d318 Mon Sep 17 00:00:00 2001
From: Chad Rosier <mcrosier@codeaurora.org>
Date: Wed, 16 Oct 2013 16:09:16 +0000
Subject: [PATCH] [AArch64] Add support for NEON scalar signed saturating
 accumulated of unsigned value and unsigned saturating accumulate of signed
 value instructions.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@192801 91177308-0d34-0410-b5e6-96231b3b80d8
---
 include/clang/Basic/arm_neon.td        |  8 +++++
 lib/CodeGen/CGBuiltin.cpp              | 14 ++++++++
 test/CodeGen/aarch64-neon-intrinsics.c | 48 ++++++++++++++++++++++++++
 3 files changed, 70 insertions(+)

diff --git a/include/clang/Basic/arm_neon.td b/include/clang/Basic/arm_neon.td
index 2cca85cc95..06fb9a3c37 100644
--- a/include/clang/Basic/arm_neon.td
+++ b/include/clang/Basic/arm_neon.td
@@ -862,4 +862,12 @@ def SCALAR_SQABS : SInst<"vqabs", "ss", "ScSsSiSl">;
 ////////////////////////////////////////////////////////////////////////////////
 // Scalar Signed Saturating Negate
 def SCALAR_SQNEG : SInst<"vqneg", "ss", "ScSsSiSl">;
+
+////////////////////////////////////////////////////////////////////////////////
+// Scalar Signed Saturating Accumulated of Unsigned Value
+def SCALAR_SUQADD : SInst<"vuqadd", "sss", "ScSsSiSl">;
+
+////////////////////////////////////////////////////////////////////////////////
+// Scalar Unsigned Saturating Accumulated of Unsigned Value
+def SCALAR_USQADD : SInst<"vsqadd", "sss", "SUcSUsSUiSUl">;
 }
diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 361758c51b..41ca24ed83 100644
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -2088,6 +2088,20 @@ static Value *EmitAArch64ScalarBuiltinExpr(CodeGenFunction &CGF,
   case AArch64::BI__builtin_neon_vqnegd_s64:
     Int = Intrinsic::arm_neon_vqneg;
     s = "vqneg"; OverloadInt = true; break;
+  // Scalar Signed Saturating Accumulated of Unsigned Value
+  case AArch64::BI__builtin_neon_vuqaddb_s8:
+  case AArch64::BI__builtin_neon_vuqaddh_s16:
+  case AArch64::BI__builtin_neon_vuqadds_s32:
+  case AArch64::BI__builtin_neon_vuqaddd_s64:
+    Int = Intrinsic::aarch64_neon_vuqadd;
+    s = "vuqadd"; OverloadInt = true; break;
+  // Scalar Unsigned Saturating Accumulated of Unsigned Value
+  case AArch64::BI__builtin_neon_vsqaddb_u8:
+  case AArch64::BI__builtin_neon_vsqaddh_u16:
+  case AArch64::BI__builtin_neon_vsqadds_u32:
+  case AArch64::BI__builtin_neon_vsqaddd_u64:
+    Int = Intrinsic::aarch64_neon_vsqadd;
+    s = "vsqadd"; OverloadInt = true; break;
   }
 
   if (!Int)
diff --git a/test/CodeGen/aarch64-neon-intrinsics.c b/test/CodeGen/aarch64-neon-intrinsics.c
index a77b56eb71..54b4e7536a 100644
--- a/test/CodeGen/aarch64-neon-intrinsics.c
+++ b/test/CodeGen/aarch64-neon-intrinsics.c
@@ -7165,3 +7165,51 @@ int64_t test_vqnegd_s64(int64_t a) {
 // CHECK: sqneg {{d[0-9]+}}, {{d[0-9]+}}
   return (int64_t)vqnegd_s64(a);
 }
+
+int8_t test_vuqaddb_s8(int8_t a, int8_t b) {
+// CHECK: test_vuqaddb_s8
+// CHECK: suqadd {{b[0-9]+}}, {{b[0-9]+}}
+  return (int8_t)vuqaddb_s8(a, b);
+}
+
+int16_t test_vuqaddh_s16(int16_t a, int16_t b) {
+// CHECK: test_vuqaddh_s16
+// CHECK: suqadd {{h[0-9]+}}, {{h[0-9]+}}
+  return (int16_t)vuqaddh_s16(a, b);
+}
+
+int32_t test_vuqadds_s32(int32_t a, int32_t b) {
+// CHECK: test_vuqadds_s32
+// CHECK: suqadd {{s[0-9]+}}, {{s[0-9]+}}
+  return (int32_t)vuqadds_s32(a, b);
+}
+
+int64_t test_vuqaddd_s64(int64_t a, int64_t b) {
+// CHECK: test_vuqaddd_s64
+// CHECK: suqadd {{d[0-9]+}}, {{d[0-9]+}}
+  return (int64_t)vuqaddd_s64(a, b);
+}
+
+uint8_t test_vsqaddb_u8(uint8_t a, uint8_t b) {
+// CHECK: test_vsqaddb_u8
+// CHECK: usqadd {{b[0-9]+}}, {{b[0-9]+}}
+  return (uint8_t)vsqaddb_u8(a, b);
+}
+
+uint16_t test_vsqaddh_u16(uint16_t a, uint16_t b) {
+// CHECK: test_vsqaddh_u16
+// CHECK: usqadd {{h[0-9]+}}, {{h[0-9]+}}
+  return (uint16_t)vsqaddh_u16(a, b);
+}
+
+uint32_t test_vsqadds_u32(uint32_t a, uint32_t b) {
+// CHECK: test_vsqadds_u32
+// CHECK: usqadd {{s[0-9]+}}, {{s[0-9]+}}
+  return (uint32_t)vsqadds_u32(a, b);
+}
+
+uint64_t test_vsqaddd_u64(uint64_t a, uint64_t b) {
+// CHECK: test_vsqaddd_u64
+// CHECK: usqadd {{d[0-9]+}}, {{d[0-9]+}}
+  return (uint64_t)vsqaddd_u64(a, b);
+}
-- 
2.40.0