diff options
| author | Matt Arsenault <Matthew.Arsenault@amd.com> | 2016-12-08 20:14:46 +0000 |
|---|---|---|
| committer | Matt Arsenault <Matthew.Arsenault@amd.com> | 2016-12-08 20:14:46 +0000 |
| commit | e96d03745d55c18b9afa7f581de06297bf1153b0 (patch) | |
| tree | 5e8e8c91da3915e73ab09bbcbce9bab71c66b4e1 /llvm/lib/Target/AMDGPU/SIISelLowering.h | |
| parent | 73b54f413430ec69547b7bf25e2bce7a9ab1d0c9 (diff) | |
| download | bcm5719-llvm-e96d03745d55c18b9afa7f581de06297bf1153b0.tar.gz bcm5719-llvm-e96d03745d55c18b9afa7f581de06297bf1153b0.zip | |
AMDGPU: Make f16 ConstantFP legal
Not having this legal led to combine failures, resulting
in dumb things like bitcasts of constants not being folded
away.
The only reason I'm leaving the v_mov_b32 hack that f32
already uses is to avoid madak formation test regressions.
PeepholeOptimizer has an ordering issue where the immediate
fold attempt is into the sgpr->vgpr copy instead of the actual
use. Running it twice avoids that problem.
llvm-svn: 289096
Diffstat (limited to 'llvm/lib/Target/AMDGPU/SIISelLowering.h')
| -rw-r--r-- | llvm/lib/Target/AMDGPU/SIISelLowering.h | 3 |
1 files changed, 0 insertions, 3 deletions
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.h b/llvm/lib/Target/AMDGPU/SIISelLowering.h index 56d6ef2a0c1..cb6d5364793 100644 --- a/llvm/lib/Target/AMDGPU/SIISelLowering.h +++ b/llvm/lib/Target/AMDGPU/SIISelLowering.h @@ -53,9 +53,6 @@ class SITargetLowering final : public AMDGPUTargetLowering { const SDLoc &DL, EVT VT) const; - /// \brief Custom lowering for ISD::ConstantFP. - SDValue lowerConstantFP(SDValue Op, SelectionDAG &DAG) const; - /// \brief Custom lowering for ISD::FP_ROUND for MVT::f16. SDValue lowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const; |
