From 4e423f724ac2b9aad30dc7a01f493fe0395f7d06 Mon Sep 17 00:00:00 2001 From: Tim Northover Date: Wed, 28 Aug 2013 09:46:34 +0000 Subject: 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. llvm-svn: 189462 --- clang/lib/CodeGen/CGBuiltin.cpp | 28 ++++++++++++++++++++++------ 1 file changed, 22 insertions(+), 6 deletions(-) (limited to 'clang/lib/CodeGen') diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp index fc8d2848390..c9a2fcbf460 100644 --- a/clang/lib/CodeGen/CGBuiltin.cpp +++ b/clang/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 MulOps(Ops.begin() + 1, Ops.end()); + Value *Mul = EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqdmull, Ty), + MulOps, "vqdmlal"); + + SmallVector 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 MulOps(Ops.begin() + 1, Ops.end()); + Value *Mul = EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqdmull, Ty), + MulOps, "vqdmlsl"); + + SmallVector 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), -- cgit v1.2.3