1 files changed, 15 insertions, 3 deletions
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index 6118933..d059480 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -367,6 +367,18 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(const TargetMachine &TM,
   setTruncStoreAction(MVT::v4f64, MVT::v4bf16, Expand);
   setTruncStoreAction(MVT::v4f64, MVT::v4f16, Expand);
 
+  setTruncStoreAction(MVT::v5i32, MVT::v5i1, Expand);
+  setTruncStoreAction(MVT::v5i32, MVT::v5i8, Expand);
+  setTruncStoreAction(MVT::v5i32, MVT::v5i16, Expand);
+
+  setTruncStoreAction(MVT::v6i32, MVT::v6i1, Expand);
+  setTruncStoreAction(MVT::v6i32, MVT::v6i8, Expand);
+  setTruncStoreAction(MVT::v6i32, MVT::v6i16, Expand);
+
+  setTruncStoreAction(MVT::v7i32, MVT::v7i1, Expand);
+  setTruncStoreAction(MVT::v7i32, MVT::v7i8, Expand);
+  setTruncStoreAction(MVT::v7i32, MVT::v7i16, Expand);
+
   setTruncStoreAction(MVT::v8f64, MVT::v8f32, Expand);
   setTruncStoreAction(MVT::v8f64, MVT::v8bf16, Expand);
   setTruncStoreAction(MVT::v8f64, MVT::v8f16, Expand);
@@ -2634,7 +2646,7 @@ bool AMDGPUTargetLowering::allowApproxFunc(const SelectionDAG &DAG,
   if (Flags.hasApproximateFuncs())
     return true;
   auto &Options = DAG.getTarget().Options;
-  return Options.UnsafeFPMath || Options.ApproxFuncFPMath;
+  return Options.ApproxFuncFPMath;
 }
 
 bool AMDGPUTargetLowering::needsDenormHandlingF32(const SelectionDAG &DAG,
@@ -2757,7 +2769,7 @@ SDValue AMDGPUTargetLowering::LowerFLOGCommon(SDValue Op,
 
   const auto &Options = getTargetMachine().Options;
   if (VT == MVT::f16 || Flags.hasApproximateFuncs() ||
-      Options.ApproxFuncFPMath || Options.UnsafeFPMath) {
+      Options.ApproxFuncFPMath) {
 
     if (VT == MVT::f16 && !Subtarget->has16BitInsts()) {
       // Log and multiply in f32 is good enough for f16.
@@ -3585,7 +3597,7 @@ SDValue AMDGPUTargetLowering::LowerFP_TO_FP16(SDValue Op, SelectionDAG &DAG) con
   if (N0.getValueType() == MVT::f32)
     return DAG.getNode(AMDGPUISD::FP_TO_FP16, DL, Op.getValueType(), N0);
 
-  if (getTargetMachine().Options.UnsafeFPMath) {
+  if (Op->getFlags().hasApproximateFuncs()) {
     // There is a generic expand for FP_TO_FP16 with unsafe fast math.
     return SDValue();
   }