[llvm] [RISCV] Use separate CCValAssign for both parts of f64 with ilp32. (PR #69129)
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Mon Oct 16 14:33:41 PDT 2023
https://github.com/topperc updated https://github.com/llvm/llvm-project/pull/69129
>From 90484af6eb66867140e00ca360d31f5e6347bcc4 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper at sifive.com>
Date: Sun, 15 Oct 2023 13:32:19 -0700
Subject: [PATCH 1/3] [RISCV] Use separate CCValAssign for both parts of f64
with ilp32.
Mark any registers as CustomReg.
This allows us to more directly emit the register or memory
access for the high part. Previously we assume the offset was
0 and we needed a memory access if the low register was X17.
This is another part of supporting FP arguments with GISel.
---
llvm/lib/Target/RISCV/RISCVISelLowering.cpp | 94 +++++++++++++--------
1 file changed, 57 insertions(+), 37 deletions(-)
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 6eb253cc5146635..6eb2e7697e5cd91 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -16435,9 +16435,16 @@ bool RISCV::CC_RISCV(const DataLayout &DL, RISCVABI::ABI ABI, unsigned ValNo,
return false;
}
LocVT = MVT::i32;
- if (!State.AllocateReg(ArgGPRs))
- State.AllocateStack(4, Align(4));
- State.addLoc(CCValAssign::getReg(ValNo, ValVT, Reg, LocVT, LocInfo));
+ State.addLoc(CCValAssign::getCustomReg(ValNo, ValVT, Reg, LocVT, LocInfo));
+ Register HiReg = State.AllocateReg(ArgGPRs);
+ if (HiReg) {
+ State.addLoc(
+ CCValAssign::getCustomReg(ValNo, ValVT, HiReg, LocVT, LocInfo));
+ } else {
+ unsigned StackOffset = State.AllocateStack(4, Align(4));
+ State.addLoc(
+ CCValAssign::getMem(ValNo, ValVT, StackOffset, LocVT, LocInfo));
+ }
return false;
}
@@ -16746,7 +16753,9 @@ static SDValue unpackFromMemLoc(SelectionDAG &DAG, SDValue Chain,
}
static SDValue unpackF64OnRV32DSoftABI(SelectionDAG &DAG, SDValue Chain,
- const CCValAssign &VA, const SDLoc &DL) {
+ const CCValAssign &VA,
+ const CCValAssign &HiVA,
+ const SDLoc &DL) {
assert(VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64 &&
"Unexpected VA");
MachineFunction &MF = DAG.getMachineFunction();
@@ -16759,16 +16768,17 @@ static SDValue unpackF64OnRV32DSoftABI(SelectionDAG &DAG, SDValue Chain,
RegInfo.addLiveIn(VA.getLocReg(), LoVReg);
SDValue Lo = DAG.getCopyFromReg(Chain, DL, LoVReg, MVT::i32);
SDValue Hi;
- if (VA.getLocReg() == RISCV::X17) {
+ if (HiVA.isMemLoc()) {
// Second half of f64 is passed on the stack.
- int FI = MFI.CreateFixedObject(4, 0, /*IsImmutable=*/true);
+ int FI = MFI.CreateFixedObject(4, HiVA.getLocMemOffset(),
+ /*IsImmutable=*/true);
SDValue FIN = DAG.getFrameIndex(FI, MVT::i32);
Hi = DAG.getLoad(MVT::i32, DL, Chain, FIN,
MachinePointerInfo::getFixedStack(MF, FI));
} else {
// Second half of f64 is passed in another GPR.
Register HiVReg = RegInfo.createVirtualRegister(&RISCV::GPRRegClass);
- RegInfo.addLiveIn(VA.getLocReg() + 1, HiVReg);
+ RegInfo.addLiveIn(HiVA.getLocReg(), HiVReg);
Hi = DAG.getCopyFromReg(Chain, DL, HiVReg, MVT::i32);
}
return DAG.getNode(RISCVISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
@@ -17011,15 +17021,16 @@ SDValue RISCVTargetLowering::LowerFormalArguments(
CallConv == CallingConv::Fast ? RISCV::CC_RISCV_FastCC
: RISCV::CC_RISCV);
- for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
+ for (unsigned i = 0, e = ArgLocs.size(), InsIdx = 0; i != e; ++i, ++InsIdx) {
CCValAssign &VA = ArgLocs[i];
SDValue ArgValue;
// Passing f64 on RV32D with a soft float ABI must be handled as a special
// case.
- if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64)
- ArgValue = unpackF64OnRV32DSoftABI(DAG, Chain, VA, DL);
- else if (VA.isRegLoc())
- ArgValue = unpackFromRegLoc(DAG, Chain, VA, DL, Ins[i], *this);
+ if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64) {
+ assert(VA.needsCustom());
+ ArgValue = unpackF64OnRV32DSoftABI(DAG, Chain, VA, ArgLocs[++i], DL);
+ } else if (VA.isRegLoc())
+ ArgValue = unpackFromRegLoc(DAG, Chain, VA, DL, Ins[InsIdx], *this);
else
ArgValue = unpackFromMemLoc(DAG, Chain, VA, DL);
@@ -17031,12 +17042,12 @@ SDValue RISCVTargetLowering::LowerFormalArguments(
// stores are relative to that.
InVals.push_back(DAG.getLoad(VA.getValVT(), DL, Chain, ArgValue,
MachinePointerInfo()));
- unsigned ArgIndex = Ins[i].OrigArgIndex;
- unsigned ArgPartOffset = Ins[i].PartOffset;
+ unsigned ArgIndex = Ins[InsIdx].OrigArgIndex;
+ unsigned ArgPartOffset = Ins[InsIdx].PartOffset;
assert(VA.getValVT().isVector() || ArgPartOffset == 0);
- while (i + 1 != e && Ins[i + 1].OrigArgIndex == ArgIndex) {
+ while (i + 1 != e && Ins[InsIdx + 1].OrigArgIndex == ArgIndex) {
CCValAssign &PartVA = ArgLocs[i + 1];
- unsigned PartOffset = Ins[i + 1].PartOffset - ArgPartOffset;
+ unsigned PartOffset = Ins[InsIdx + 1].PartOffset - ArgPartOffset;
SDValue Offset = DAG.getIntPtrConstant(PartOffset, DL);
if (PartVA.getValVT().isScalableVector())
Offset = DAG.getNode(ISD::VSCALE, DL, XLenVT, Offset);
@@ -17044,6 +17055,7 @@ SDValue RISCVTargetLowering::LowerFormalArguments(
InVals.push_back(DAG.getLoad(PartVA.getValVT(), DL, Chain, Address,
MachinePointerInfo()));
++i;
+ ++InsIdx;
}
continue;
}
@@ -17259,14 +17271,16 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
SmallVector<std::pair<Register, SDValue>, 8> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
SDValue StackPtr;
- for (unsigned i = 0, j = 0, e = ArgLocs.size(); i != e; ++i) {
+ for (unsigned i = 0, j = 0, e = ArgLocs.size(), OutIdx = 0; i != e;
+ ++i, ++OutIdx) {
CCValAssign &VA = ArgLocs[i];
- SDValue ArgValue = OutVals[i];
- ISD::ArgFlagsTy Flags = Outs[i].Flags;
+ SDValue ArgValue = OutVals[OutIdx];
+ ISD::ArgFlagsTy Flags = Outs[OutIdx].Flags;
// Handle passing f64 on RV32D with a soft float ABI as a special case.
if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64) {
assert(VA.isRegLoc() && "Expected register VA assignment");
+ assert(VA.needsCustom());
SDValue SplitF64 = DAG.getNode(
RISCVISD::SplitF64, DL, DAG.getVTList(MVT::i32, MVT::i32), ArgValue);
SDValue Lo = SplitF64.getValue(0);
@@ -17275,18 +17289,22 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
Register RegLo = VA.getLocReg();
RegsToPass.push_back(std::make_pair(RegLo, Lo));
- if (RegLo == RISCV::X17) {
+ // Get the CCValAssign for the Hi part.
+ CCValAssign &HiVA = ArgLocs[++i];
+
+ if (HiVA.isMemLoc()) {
// Second half of f64 is passed on the stack.
- // Work out the address of the stack slot.
if (!StackPtr.getNode())
StackPtr = DAG.getCopyFromReg(Chain, DL, RISCV::X2, PtrVT);
+ SDValue Address =
+ DAG.getNode(ISD::ADD, DL, PtrVT, StackPtr,
+ DAG.getIntPtrConstant(HiVA.getLocMemOffset(), DL));
// Emit the store.
MemOpChains.push_back(
- DAG.getStore(Chain, DL, Hi, StackPtr, MachinePointerInfo()));
+ DAG.getStore(Chain, DL, Hi, Address, MachinePointerInfo()));
} else {
// Second half of f64 is passed in another GPR.
- assert(RegLo < RISCV::X31 && "Invalid register pair");
- Register RegHigh = RegLo + 1;
+ Register RegHigh = HiVA.getLocReg();
RegsToPass.push_back(std::make_pair(RegHigh, Hi));
}
continue;
@@ -17297,7 +17315,7 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
if (VA.getLocInfo() == CCValAssign::Indirect) {
// Store the argument in a stack slot and pass its address.
Align StackAlign =
- std::max(getPrefTypeAlign(Outs[i].ArgVT, DAG),
+ std::max(getPrefTypeAlign(Outs[OutIdx].ArgVT, DAG),
getPrefTypeAlign(ArgValue.getValueType(), DAG));
TypeSize StoredSize = ArgValue.getValueType().getStoreSize();
// If the original argument was split (e.g. i128), we need
@@ -17305,16 +17323,16 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
// Vectors may be partly split to registers and partly to the stack, in
// which case the base address is partly offset and subsequent stores are
// relative to that.
- unsigned ArgIndex = Outs[i].OrigArgIndex;
- unsigned ArgPartOffset = Outs[i].PartOffset;
+ unsigned ArgIndex = Outs[OutIdx].OrigArgIndex;
+ unsigned ArgPartOffset = Outs[OutIdx].PartOffset;
assert(VA.getValVT().isVector() || ArgPartOffset == 0);
// Calculate the total size to store. We don't have access to what we're
// actually storing other than performing the loop and collecting the
// info.
SmallVector<std::pair<SDValue, SDValue>> Parts;
- while (i + 1 != e && Outs[i + 1].OrigArgIndex == ArgIndex) {
+ while (i + 1 != e && Outs[OutIdx + 1].OrigArgIndex == ArgIndex) {
SDValue PartValue = OutVals[i + 1];
- unsigned PartOffset = Outs[i + 1].PartOffset - ArgPartOffset;
+ unsigned PartOffset = Outs[OutIdx + 1].PartOffset - ArgPartOffset;
SDValue Offset = DAG.getIntPtrConstant(PartOffset, DL);
EVT PartVT = PartValue.getValueType();
if (PartVT.isScalableVector())
@@ -17323,6 +17341,7 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
StackAlign = std::max(StackAlign, getPrefTypeAlign(PartVT, DAG));
Parts.push_back(std::make_pair(PartValue, Offset));
++i;
+ ++OutIdx;
}
SDValue SpillSlot = DAG.CreateStackTemporary(StoredSize, StackAlign);
int FI = cast<FrameIndexSDNode>(SpillSlot)->getIndex();
@@ -17464,7 +17483,8 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
analyzeInputArgs(MF, RetCCInfo, Ins, /*IsRet=*/true, RISCV::CC_RISCV);
// Copy all of the result registers out of their specified physreg.
- for (auto &VA : RVLocs) {
+ for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
+ auto &VA = RVLocs[i];
// Copy the value out
SDValue RetValue =
DAG.getCopyFromReg(Chain, DL, VA.getLocReg(), VA.getLocVT(), Glue);
@@ -17473,9 +17493,9 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
Glue = RetValue.getValue(2);
if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64) {
- assert(VA.getLocReg() == ArgGPRs[0] && "Unexpected reg assignment");
- SDValue RetValue2 =
- DAG.getCopyFromReg(Chain, DL, ArgGPRs[1], MVT::i32, Glue);
+ assert(VA.needsCustom());
+ SDValue RetValue2 = DAG.getCopyFromReg(Chain, DL, RVLocs[++i].getLocReg(),
+ MVT::i32, Glue);
Chain = RetValue2.getValue(1);
Glue = RetValue2.getValue(2);
RetValue = DAG.getNode(RISCVISD::BuildPairF64, DL, MVT::f64, RetValue,
@@ -17538,21 +17558,21 @@ RISCVTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
SmallVector<SDValue, 4> RetOps(1, Chain);
// Copy the result values into the output registers.
- for (unsigned i = 0, e = RVLocs.size(); i < e; ++i) {
- SDValue Val = OutVals[i];
+ for (unsigned i = 0, e = RVLocs.size(), OutIdx = 0; i < e; ++i, ++OutIdx) {
+ SDValue Val = OutVals[OutIdx];
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64) {
// Handle returning f64 on RV32D with a soft float ABI.
assert(VA.isRegLoc() && "Expected return via registers");
+ assert(VA.needsCustom());
SDValue SplitF64 = DAG.getNode(RISCVISD::SplitF64, DL,
DAG.getVTList(MVT::i32, MVT::i32), Val);
SDValue Lo = SplitF64.getValue(0);
SDValue Hi = SplitF64.getValue(1);
Register RegLo = VA.getLocReg();
- assert(RegLo < RISCV::X31 && "Invalid register pair");
- Register RegHi = RegLo + 1;
+ Register RegHi = RVLocs[++i].getLocReg();
if (STI.isRegisterReservedByUser(RegLo) ||
STI.isRegisterReservedByUser(RegHi))
>From 50c0305483b13472463c32e69d44c6aa330ecd02 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper at sifive.com>
Date: Sun, 15 Oct 2023 14:59:35 -0700
Subject: [PATCH 2/3] !fixup Use getCustomMem for the high memory location.
---
llvm/lib/Target/RISCV/RISCVISelLowering.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 6eb2e7697e5cd91..71dcde018b6dd70 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -16443,7 +16443,7 @@ bool RISCV::CC_RISCV(const DataLayout &DL, RISCVABI::ABI ABI, unsigned ValNo,
} else {
unsigned StackOffset = State.AllocateStack(4, Align(4));
State.addLoc(
- CCValAssign::getMem(ValNo, ValVT, StackOffset, LocVT, LocInfo));
+ CCValAssign::getCustomMem(ValNo, ValVT, StackOffset, LocVT, LocInfo));
}
return false;
}
>From 3e90ffa7c3fdd5d1d929748b1ff78429c4529386 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper at sifive.com>
Date: Mon, 16 Oct 2023 14:33:18 -0700
Subject: [PATCH 3/3] !fixup Change use of i to OutIdx in LowerCall.
---
llvm/lib/Target/RISCV/RISCVISelLowering.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 71dcde018b6dd70..b10b019530906ed 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -17331,7 +17331,7 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
// info.
SmallVector<std::pair<SDValue, SDValue>> Parts;
while (i + 1 != e && Outs[OutIdx + 1].OrigArgIndex == ArgIndex) {
- SDValue PartValue = OutVals[i + 1];
+ SDValue PartValue = OutVals[OutIdx + 1];
unsigned PartOffset = Outs[OutIdx + 1].PartOffset - ArgPartOffset;
SDValue Offset = DAG.getIntPtrConstant(PartOffset, DL);
EVT PartVT = PartValue.getValueType();
More information about the llvm-commits
mailing list