Make default expansion for FP16 <-> FP32 nodes into libcalls
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@98501 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/include/llvm/CodeGen/RuntimeLibcalls.h b/include/llvm/CodeGen/RuntimeLibcalls.h
index c404ab6..0ac945b 100644
--- a/include/llvm/CodeGen/RuntimeLibcalls.h
+++ b/include/llvm/CodeGen/RuntimeLibcalls.h
@@ -153,6 +153,8 @@
// CONVERSION
FPEXT_F32_F64,
+ FPEXT_F16_F32,
+ FPROUND_F32_F16,
FPROUND_F64_F32,
FPROUND_F80_F32,
FPROUND_PPCF128_F32,
diff --git a/include/llvm/Target/TargetSelectionDAG.td b/include/llvm/Target/TargetSelectionDAG.td
index 4365d33..c718c86 100644
--- a/include/llvm/Target/TargetSelectionDAG.td
+++ b/include/llvm/Target/TargetSelectionDAG.td
@@ -345,6 +345,8 @@
def uint_to_fp : SDNode<"ISD::UINT_TO_FP" , SDTIntToFPOp>;
def fp_to_sint : SDNode<"ISD::FP_TO_SINT" , SDTFPToIntOp>;
def fp_to_uint : SDNode<"ISD::FP_TO_UINT" , SDTFPToIntOp>;
+def f16_to_f32 : SDNode<"ISD::FP16_TO_FP32", SDTIntToFPOp>;
+def f32_to_f16 : SDNode<"ISD::FP32_TO_FP16", SDTFPToIntOp>;
def setcc : SDNode<"ISD::SETCC" , SDTSetCC>;
def select : SDNode<"ISD::SELECT" , SDTSelect>;
diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index a3175e8..b5af2c1 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -851,6 +851,8 @@
case ISD::MERGE_VALUES:
case ISD::EH_RETURN:
case ISD::FRAME_TO_ARGS_OFFSET:
+ case ISD::FP16_TO_FP32:
+ case ISD::FP32_TO_FP16:
// These operations lie about being legal: when they claim to be legal,
// they should actually be expanded.
Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
@@ -2636,6 +2638,12 @@
Results.push_back(ExpandFPLibCall(Node, RTLIB::REM_F32, RTLIB::REM_F64,
RTLIB::REM_F80, RTLIB::REM_PPCF128));
break;
+ case ISD::FP16_TO_FP32:
+ Results.push_back(ExpandLibCall(RTLIB::FPEXT_F16_F32, Node, false));
+ break;
+ case ISD::FP32_TO_FP16:
+ Results.push_back(ExpandLibCall(RTLIB::FPROUND_F32_F16, Node, false));
+ break;
case ISD::ConstantFP: {
ConstantFPSDNode *CFP = cast<ConstantFPSDNode>(Node);
// Check to see if this FP immediate is already legal.
diff --git a/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
index 35a7c7c..3a4b3de 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp
@@ -79,6 +79,7 @@
case ISD::FNEG: R = SoftenFloatRes_FNEG(N); break;
case ISD::FP_EXTEND: R = SoftenFloatRes_FP_EXTEND(N); break;
case ISD::FP_ROUND: R = SoftenFloatRes_FP_ROUND(N); break;
+ case ISD::FP16_TO_FP32:R = SoftenFloatRes_FP16_TO_FP32(N); break;
case ISD::FPOW: R = SoftenFloatRes_FPOW(N); break;
case ISD::FPOWI: R = SoftenFloatRes_FPOWI(N); break;
case ISD::FREM: R = SoftenFloatRes_FREM(N); break;
@@ -332,6 +333,14 @@
return MakeLibCall(LC, NVT, &Op, 1, false, N->getDebugLoc());
}
+// FIXME: Should we just use 'normal' FP_EXTEND / FP_TRUNC instead of special
+// nodes?
+SDValue DAGTypeLegalizer::SoftenFloatRes_FP16_TO_FP32(SDNode *N) {
+ EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+ SDValue Op = N->getOperand(0);
+ return MakeLibCall(RTLIB::FPEXT_F16_F32, NVT, &Op, 1, false, N->getDebugLoc());
+}
+
SDValue DAGTypeLegalizer::SoftenFloatRes_FP_ROUND(SDNode *N) {
EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
SDValue Op = N->getOperand(0);
@@ -548,6 +557,7 @@
case ISD::FP_ROUND: Res = SoftenFloatOp_FP_ROUND(N); break;
case ISD::FP_TO_SINT: Res = SoftenFloatOp_FP_TO_SINT(N); break;
case ISD::FP_TO_UINT: Res = SoftenFloatOp_FP_TO_UINT(N); break;
+ case ISD::FP32_TO_FP16:Res = SoftenFloatOp_FP32_TO_FP16(N); break;
case ISD::SELECT_CC: Res = SoftenFloatOp_SELECT_CC(N); break;
case ISD::SETCC: Res = SoftenFloatOp_SETCC(N); break;
case ISD::STORE: Res = SoftenFloatOp_STORE(N, OpNo); break;
@@ -704,6 +714,13 @@
return MakeLibCall(LC, RVT, &Op, 1, false, N->getDebugLoc());
}
+SDValue DAGTypeLegalizer::SoftenFloatOp_FP32_TO_FP16(SDNode *N) {
+ EVT RVT = N->getValueType(0);
+ RTLIB::Libcall LC = RTLIB::FPROUND_F32_F16;
+ SDValue Op = GetSoftenedFloat(N->getOperand(0));
+ return MakeLibCall(LC, RVT, &Op, 1, false, N->getDebugLoc());
+}
+
SDValue DAGTypeLegalizer::SoftenFloatOp_SELECT_CC(SDNode *N) {
SDValue NewLHS = N->getOperand(0), NewRHS = N->getOperand(1);
ISD::CondCode CCCode = cast<CondCodeSDNode>(N->getOperand(4))->get();
diff --git a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 81f28ad..48f64c3 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -80,6 +80,8 @@
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT: Res = PromoteIntRes_FP_TO_XINT(N); break;
+ case ISD::FP32_TO_FP16:Res = PromoteIntRes_FP32_TO_FP16(N); break;
+
case ISD::AND:
case ISD::OR:
case ISD::XOR:
@@ -324,6 +326,16 @@
NVT, Res, DAG.getValueType(N->getValueType(0)));
}
+SDValue DAGTypeLegalizer::PromoteIntRes_FP32_TO_FP16(SDNode *N) {
+ EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+ DebugLoc dl = N->getDebugLoc();
+
+ SDValue Res = DAG.getNode(N->getOpcode(), dl, NVT, N->getOperand(0));
+
+ return DAG.getNode(ISD::AssertZext, dl,
+ NVT, Res, DAG.getValueType(N->getValueType(0)));
+}
+
SDValue DAGTypeLegalizer::PromoteIntRes_INT_EXTEND(SDNode *N) {
EVT NVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
DebugLoc dl = N->getDebugLoc();
@@ -634,6 +646,7 @@
case ISD::STORE: Res = PromoteIntOp_STORE(cast<StoreSDNode>(N),
OpNo); break;
case ISD::TRUNCATE: Res = PromoteIntOp_TRUNCATE(N); break;
+ case ISD::FP16_TO_FP32:
case ISD::UINT_TO_FP: Res = PromoteIntOp_UINT_TO_FP(N); break;
case ISD::ZERO_EXTEND: Res = PromoteIntOp_ZERO_EXTEND(N); break;
diff --git a/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index b0af357..7c70663 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -257,6 +257,7 @@
SDValue PromoteIntRes_CTTZ(SDNode *N);
SDValue PromoteIntRes_EXTRACT_VECTOR_ELT(SDNode *N);
SDValue PromoteIntRes_FP_TO_XINT(SDNode *N);
+ SDValue PromoteIntRes_FP32_TO_FP16(SDNode *N);
SDValue PromoteIntRes_INT_EXTEND(SDNode *N);
SDValue PromoteIntRes_LOAD(LoadSDNode *N);
SDValue PromoteIntRes_Overflow(SDNode *N);
@@ -406,6 +407,7 @@
SDValue SoftenFloatRes_FNEARBYINT(SDNode *N);
SDValue SoftenFloatRes_FNEG(SDNode *N);
SDValue SoftenFloatRes_FP_EXTEND(SDNode *N);
+ SDValue SoftenFloatRes_FP16_TO_FP32(SDNode *N);
SDValue SoftenFloatRes_FP_ROUND(SDNode *N);
SDValue SoftenFloatRes_FPOW(SDNode *N);
SDValue SoftenFloatRes_FPOWI(SDNode *N);
@@ -429,6 +431,7 @@
SDValue SoftenFloatOp_FP_ROUND(SDNode *N);
SDValue SoftenFloatOp_FP_TO_SINT(SDNode *N);
SDValue SoftenFloatOp_FP_TO_UINT(SDNode *N);
+ SDValue SoftenFloatOp_FP32_TO_FP16(SDNode *N);
SDValue SoftenFloatOp_SELECT_CC(SDNode *N);
SDValue SoftenFloatOp_SETCC(SDNode *N);
SDValue SoftenFloatOp_STORE(SDNode *N, unsigned OpNo);
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 1911103..90c196c 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5665,6 +5665,8 @@
case ISD::FP_TO_SINT: return "fp_to_sint";
case ISD::FP_TO_UINT: return "fp_to_uint";
case ISD::BIT_CONVERT: return "bit_convert";
+ case ISD::FP16_TO_FP32: return "fp16_to_fp32";
+ case ISD::FP32_TO_FP16: return "fp32_to_fp16";
case ISD::CONVERT_RNDSAT: {
switch (cast<CvtRndSatSDNode>(this)->getCvtCode()) {
diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 2ae9d34..4df4d77 100644
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -175,6 +175,8 @@
Names[RTLIB::FLOOR_F80] = "floorl";
Names[RTLIB::FLOOR_PPCF128] = "floorl";
Names[RTLIB::FPEXT_F32_F64] = "__extendsfdf2";
+ Names[RTLIB::FPEXT_F16_F32] = "__gnu_h2f_ieee";
+ Names[RTLIB::FPROUND_F32_F16] = "__gnu_f2h_ieee";
Names[RTLIB::FPROUND_F64_F32] = "__truncdfsf2";
Names[RTLIB::FPROUND_F80_F32] = "__truncxfsf2";
Names[RTLIB::FPROUND_PPCF128_F32] = "__trunctfsf2";
@@ -269,6 +271,7 @@
if (RetVT == MVT::f64)
return FPEXT_F32_F64;
}
+
return UNKNOWN_LIBCALL;
}
@@ -288,6 +291,7 @@
if (OpVT == MVT::ppcf128)
return FPROUND_PPCF128_F64;
}
+
return UNKNOWN_LIBCALL;
}