From: Tim Northover <tnorthover@apple.com>
Date: Wed, 28 Aug 2013 09:46:34 +0000 (+0000)
Subject: ARM: use vqdmull and vqadds/vqsubs to implement vqdmlal/vqdmlsl
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=b1c09b913e331824da29f8017452b6171f7eccbc;p=clang

ARM: use vqdmull and vqadds/vqsubs to implement vqdmlal/vqdmlsl

The NEON intrinsics vqdmlal and vqdmlsl are really just combinations of a
saturating-doubling-multiply (vqdmull) and a saturating add/sub, so now that
LLVM can spot those patterns Clang should emit them instead of specialised
intrinsics.

Feature already tested by existing ARM NEON intrinsics tests.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@189462 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index fc8d284839..c9a2fcbf46 100644
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -2561,12 +2561,28 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
   case ARM::BI__builtin_neon_vqaddq_v:
     Int = usgn ? Intrinsic::arm_neon_vqaddu : Intrinsic::arm_neon_vqadds;
     return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqadd");
-  case ARM::BI__builtin_neon_vqdmlal_v:
-    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqdmlal, Ty),
-                        Ops, "vqdmlal");
-  case ARM::BI__builtin_neon_vqdmlsl_v:
-    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqdmlsl, Ty),
-                        Ops, "vqdmlsl");
+  case ARM::BI__builtin_neon_vqdmlal_v: {
+    SmallVector<Value *, 2> MulOps(Ops.begin() + 1, Ops.end());
+    Value *Mul = EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqdmull, Ty),
+                              MulOps, "vqdmlal");
+
+    SmallVector<Value *, 2> AddOps;
+    AddOps.push_back(Ops[0]);
+    AddOps.push_back(Mul);
+    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqadds, Ty),
+                        AddOps, "vqdmlal");
+  }
+  case ARM::BI__builtin_neon_vqdmlsl_v: {
+    SmallVector<Value *, 2> MulOps(Ops.begin() + 1, Ops.end());
+    Value *Mul = EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqdmull, Ty),
+                              MulOps, "vqdmlsl");
+
+    SmallVector<Value *, 2> SubOps;
+    SubOps.push_back(Ops[0]);
+    SubOps.push_back(Mul);
+    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqsubs, Ty),
+                        SubOps, "vqdmlsl");
+  }
   case ARM::BI__builtin_neon_vqdmulh_v:
   case ARM::BI__builtin_neon_vqdmulhq_v:
     return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqdmulh, Ty),