diff options
| author | Craig Topper <craig.topper@intel.com> | 2019-01-07 21:00:41 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@intel.com> | 2019-01-07 21:00:41 +0000 |
| commit | cd9e232a4d35f5983e48e626cf97e0ac5151f722 (patch) | |
| tree | 48367973935ccb0f57a8b67864b6ba432a53f9a2 /clang/lib/CodeGen/CGBuiltin.cpp | |
| parent | 486313b5f7361d4f61c8f9ef6b79f9aa9678c5d2 (diff) | |
| download | bcm5719-llvm-cd9e232a4d35f5983e48e626cf97e0ac5151f722.tar.gz bcm5719-llvm-cd9e232a4d35f5983e48e626cf97e0ac5151f722.zip | |
Recommit r350555 "[X86] Use funnel shift intrinsics for the VBMI2 vshld/vshrd builtins."
The MSVC limit hit in AutoUpgrade.cpp has been worked around for now.
llvm-svn: 350568
Diffstat (limited to 'clang/lib/CodeGen/CGBuiltin.cpp')
| -rw-r--r-- | clang/lib/CodeGen/CGBuiltin.cpp | 46 |
1 files changed, 46 insertions, 0 deletions
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp index 93484f82c30..ca7b4691ff5 100644 --- a/clang/lib/CodeGen/CGBuiltin.cpp +++ b/clang/lib/CodeGen/CGBuiltin.cpp @@ -10999,6 +10999,52 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID, case X86::BI__builtin_ia32_pternlogq256_maskz: return EmitX86Ternlog(*this, /*ZeroMask*/true, Ops); + case X86::BI__builtin_ia32_vpshldd128: + case X86::BI__builtin_ia32_vpshldd256: + case X86::BI__builtin_ia32_vpshldd512: + case X86::BI__builtin_ia32_vpshldq128: + case X86::BI__builtin_ia32_vpshldq256: + case X86::BI__builtin_ia32_vpshldq512: + case X86::BI__builtin_ia32_vpshldw128: + case X86::BI__builtin_ia32_vpshldw256: + case X86::BI__builtin_ia32_vpshldw512: + return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false); + + case X86::BI__builtin_ia32_vpshrdd128: + case X86::BI__builtin_ia32_vpshrdd256: + case X86::BI__builtin_ia32_vpshrdd512: + case X86::BI__builtin_ia32_vpshrdq128: + case X86::BI__builtin_ia32_vpshrdq256: + case X86::BI__builtin_ia32_vpshrdq512: + case X86::BI__builtin_ia32_vpshrdw128: + case X86::BI__builtin_ia32_vpshrdw256: + case X86::BI__builtin_ia32_vpshrdw512: + // Ops 0 and 1 are swapped. + return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true); + + case X86::BI__builtin_ia32_vpshldvd128: + case X86::BI__builtin_ia32_vpshldvd256: + case X86::BI__builtin_ia32_vpshldvd512: + case X86::BI__builtin_ia32_vpshldvq128: + case X86::BI__builtin_ia32_vpshldvq256: + case X86::BI__builtin_ia32_vpshldvq512: + case X86::BI__builtin_ia32_vpshldvw128: + case X86::BI__builtin_ia32_vpshldvw256: + case X86::BI__builtin_ia32_vpshldvw512: + return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false); + + case X86::BI__builtin_ia32_vpshrdvd128: + case X86::BI__builtin_ia32_vpshrdvd256: + case X86::BI__builtin_ia32_vpshrdvd512: + case X86::BI__builtin_ia32_vpshrdvq128: + case X86::BI__builtin_ia32_vpshrdvq256: + case X86::BI__builtin_ia32_vpshrdvq512: + case X86::BI__builtin_ia32_vpshrdvw128: + case X86::BI__builtin_ia32_vpshrdvw256: + case X86::BI__builtin_ia32_vpshrdvw512: + // Ops 0 and 1 are swapped. + return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true); + // 3DNow! case X86::BI__builtin_ia32_pswapdsf: case X86::BI__builtin_ia32_pswapdsi: { |
