r189462 - ARM: use vqdmull and vqadds/vqsubs to implement vqdmlal/vqdmlsl

[Tim Northover]

Author: tnorthover
Date: Wed Aug 28 04:46:34 2013
New Revision: 189462

URL: http://llvm.org/viewvc/llvm-project?rev=189462&view=rev
Log:
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.

Modified:
    cfe/trunk/lib/CodeGen/CGBuiltin.cpp

Modified: cfe/trunk/lib/CodeGen/CGBuiltin.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGBuiltin.cpp?rev=189462&r1=189461&r2=189462&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGBuiltin.cpp (original)
+++ cfe/trunk/lib/CodeGen/CGBuiltin.cpp Wed Aug 28 04:46:34 2013
@@ -2561,12 +2561,28 @@ Value *CodeGenFunction::EmitARMBuiltinEx
   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),