The attached patches implement most of the ARM AAPCS-VFP hard float
ABI. The missing piece is support for putting "homogeneous aggregates"
into registers.
Patch by Sandeep Patel!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73095 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/ARM/ARMCallingConv.td b/lib/Target/ARM/ARMCallingConv.td
index 6cd786e..1c5ec5f 100644
--- a/lib/Target/ARM/ARMCallingConv.td
+++ b/lib/Target/ARM/ARMCallingConv.td
@@ -17,6 +17,11 @@
class CCIfAlign<string Align, CCAction A>:
CCIf<!strconcat("ArgFlags.getOrigAlign() == ", Align), A>;
+/// CCIfFloatABI - Match of the float ABI and the arg. ABIType may be "Hard" or
+/// "Soft".
+class CCIfFloatABI<string ABIType, CCAction A>:
+ CCIf<!strconcat("llvm::FloatABIType == llvm::FloatABI::", ABIType), A>;
+
//===----------------------------------------------------------------------===//
// ARM APCS Calling Convention
//===----------------------------------------------------------------------===//
@@ -43,9 +48,10 @@
]>;
//===----------------------------------------------------------------------===//
-// ARM AAPCS (EABI) Calling Convention
+// ARM AAPCS (EABI) Calling Convention, common parts
//===----------------------------------------------------------------------===//
-def CC_ARM_AAPCS : CallingConv<[
+
+def CC_ARM_AAPCS_Common : CallingConv<[
CCIfType<[i8, i16], CCPromoteToType<i32>>,
@@ -53,35 +59,71 @@
// i64 is 8-aligned i32 here, so we may need to eat R1 as a pad register
// (and the same is true for f64 if VFP is not enabled)
CCIfType<[i32], CCIfAlign<"8", CCAssignToRegWithShadow<[R0, R2], [R0, R1]>>>,
- CCIfType<[f64], CCCustom<"CC_ARM_AAPCS_Custom_f64">>,
-
- CCIfType<[f32], CCBitConvertToType<i32>>,
CCIfType<[i32], CCIf<"State.getNextStackOffset() == 0 &&"
"ArgFlags.getOrigAlign() != 8",
CCAssignToReg<[R0, R1, R2, R3]>>>,
- CCIfType<[i32], CCAssignToStack<4, 4>>,
+ CCIfType<[i32, f32], CCAssignToStack<4, 4>>,
CCIfType<[f64], CCAssignToStack<8, 8>>
]>;
-def RetCC_ARM_AAPCS : CallingConv<[
- CCIfType<[f32], CCBitConvertToType<i32>>,
- CCIfType<[f64], CCCustom<"RetCC_ARM_AAPCS_Custom_f64">>,
-
- CCIfType<[i32], CCAssignToReg<[R0, R1, R2, R3]>>,
+def RetCC_ARM_AAPCS_Common : CallingConv<[
+ CCIfType<[i32], CCAssignToReg<[R0, R1, R2, R3]>>
CCIfType<[i64], CCAssignToRegWithShadow<[R0, R2], [R1, R3]>>
]>;
//===----------------------------------------------------------------------===//
+// ARM AAPCS (EABI) Calling Convention
+//===----------------------------------------------------------------------===//
+
+def CC_ARM_AAPCS : CallingConv<[
+ CCIfType<[f64], CCCustom<"CC_ARM_AAPCS_Custom_f64">>,
+ CCIfType<[f32], CCBitConvertToType<i32>>,
+ CCDelegateTo<CC_ARM_AAPCS_Common>
+]>;
+
+def RetCC_ARM_AAPCS : CallingConv<[
+ CCIfType<[f64], CCCustom<"RetCC_ARM_AAPCS_Custom_f64">>,
+ CCIfType<[f32], CCBitConvertToType<i32>>,
+ CCDelegateTo<RetCC_ARM_AAPCS_Common>
+]>;
+
+//===----------------------------------------------------------------------===//
+// ARM AAPCS-VFP (EABI) Calling Convention
+//===----------------------------------------------------------------------===//
+
+def CC_ARM_AAPCS_VFP : CallingConv<[
+ CCIfType<[f64], CCAssignToReg<[D0, D1, D2, D3, D4, D5, D6, D7]>>,
+ CCIfType<[f32], CCAssignToReg<[S0, S1, S2, S3, S4, S5, S6, S7, S8,
+ S9, S10, S11, S12, S13, S14, S15]>>,
+ CCDelegateTo<CC_ARM_AAPCS_Common>
+]>;
+
+def RetCC_ARM_AAPCS_VFP : CallingConv<[
+ CCIfType<[f64], CCAssignToReg<[D0, D1, D2, D3, D4, D5, D6, D7]>>,
+ CCIfType<[f32], CCAssignToReg<[S0, S1, S2, S3, S4, S5, S6, S7, S8,
+ S9, S10, S11, S12, S13, S14, S15]>>,
+ CCDelegateTo<RetCC_ARM_AAPCS_Common>
+]>;
+
+//===----------------------------------------------------------------------===//
// ARM Calling Convention Dispatch
//===----------------------------------------------------------------------===//
def CC_ARM : CallingConv<[
+ CCIfSubtarget<"isAAPCS_ABI()",
+ CCIfSubtarget<"hasVFP2()",
+ CCIfFloatABI<"Hard",
+ CCDelegateTo<CC_ARM_AAPCS_VFP>>>>,
CCIfSubtarget<"isAAPCS_ABI()", CCDelegateTo<CC_ARM_AAPCS>>,
CCDelegateTo<CC_ARM_APCS>
]>;
def RetCC_ARM : CallingConv<[
+ CCIfSubtarget<"isAAPCS_ABI()",
+ CCIfSubtarget<"hasVFP2()",
+ CCIfFloatABI<"Hard",
+ CCDelegateTo<RetCC_ARM_AAPCS_VFP>>>>,
CCIfSubtarget<"isAAPCS_ABI()", CCDelegateTo<RetCC_ARM_AAPCS>>,
CCDelegateTo<RetCC_ARM_APCS>
]>;
diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp
index ca3a9cb..a413ee6 100644
--- a/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -549,6 +549,10 @@
switch (N->getOpcode()) {
default: break;
case ISD::Constant: {
+ // ARMv6T2 and later should materialize imms via MOV / MOVT pair.
+ if (Subtarget->hasV6T2Ops())
+ break;
+
unsigned Val = cast<ConstantSDNode>(N)->getZExtValue();
bool UseCP = true;
if (Subtarget->isThumb())
@@ -559,6 +563,7 @@
UseCP = (ARM_AM::getSOImmVal(Val) == -1 && // MOV
ARM_AM::getSOImmVal(~Val) == -1 && // MVN
!ARM_AM::isSOImmTwoPartVal(Val)); // two instrs.
+
if (UseCP) {
SDValue CPIdx =
CurDAG->getTargetConstantPool(ConstantInt::get(Type::Int32Ty, Val),
@@ -570,7 +575,7 @@
CPIdx, CurDAG->getEntryNode());
else {
SDValue Ops[] = {
- CPIdx,
+ CPIdx,
CurDAG->getRegister(0, MVT::i32),
CurDAG->getTargetConstant(0, MVT::i32),
getAL(CurDAG),
@@ -583,7 +588,7 @@
ReplaceUses(Op, SDValue(ResNode, 0));
return NULL;
}
-
+
// Other cases are autogenerated.
break;
}
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index c0fd9dc..ec8bd1f 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -1101,7 +1101,12 @@
else
RC = ARM::GPRRegisterClass;
- if (RegVT == MVT::f64) {
+ if (FloatABIType == FloatABI::Hard) {
+ if (RegVT == MVT::f32)
+ RC = ARM::SPRRegisterClass;
+ else if (RegVT == MVT::f64)
+ RC = ARM::DPRRegisterClass;
+ } else if (RegVT == MVT::f64) {
// f64 is passed in pairs of GPRs and must be combined.
RegVT = MVT::i32;
} else if (!((RegVT == MVT::i32) || (RegVT == MVT::f32)))
diff --git a/lib/Target/ARM/ARMRegisterInfo.td b/lib/Target/ARM/ARMRegisterInfo.td
index b95d1f9..ebe7d58 100644
--- a/lib/Target/ARM/ARMRegisterInfo.td
+++ b/lib/Target/ARM/ARMRegisterInfo.td
@@ -219,3 +219,18 @@
// Condition code registers.
def CCR : RegisterClass<"ARM", [i32], 32, [CPSR]>;
+
+//===----------------------------------------------------------------------===//
+// Subregister Set Definitions... now that we have all of the pieces, define the
+// sub registers for each register.
+//
+
+def : SubRegSet<1, [D0, D1, D2, D3, D4, D5, D6, D7,
+ D8, D9, D10, D11, D12, D13, D14, D15],
+ [S0, S2, S4, S6, S8, S10, S12, S14,
+ S16, S18, S20, S22, S24, S26, S28, S30]>;
+
+def : SubRegSet<2, [D0, D1, D2, D3, D4, D5, D6, D7,
+ D8, D9, D10, D11, D12, D13, D14, D15],
+ [S1, S3, S5, S7, S9, S11, S13, S15,
+ S17, S19, S21, S23, S25, S27, S29, S31]>;
diff --git a/lib/Target/ARM/ARMSubtarget.cpp b/lib/Target/ARM/ARMSubtarget.cpp
index ef78cd5..a978380 100644
--- a/lib/Target/ARM/ARMSubtarget.cpp
+++ b/lib/Target/ARM/ARMSubtarget.cpp
@@ -14,6 +14,8 @@
#include "ARMSubtarget.h"
#include "ARMGenSubtarget.inc"
#include "llvm/Module.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
using namespace llvm;
ARMSubtarget::ARMSubtarget(const Module &M, const std::string &FS,
@@ -28,6 +30,10 @@
, CPUString("generic")
, TargetType(isELF) // Default to ELF unless otherwise specified.
, TargetABI(ARM_ABI_APCS) {
+ // default to soft float ABI
+ if (FloatABIType == FloatABI::Default)
+ FloatABIType = FloatABI::Soft;
+
// Determine default and user specified characteristics
// Parse features string.
diff --git a/lib/Target/TargetMachine.cpp b/lib/Target/TargetMachine.cpp
index dea293b..c487cb8 100644
--- a/lib/Target/TargetMachine.cpp
+++ b/lib/Target/TargetMachine.cpp
@@ -30,6 +30,7 @@
bool FiniteOnlyFPMathOption;
bool HonorSignDependentRoundingFPMathOption;
bool UseSoftFloat;
+ FloatABI::ABIType FloatABIType;
bool NoImplicitFloat;
bool NoZerosInBSS;
bool ExceptionHandling;
@@ -84,6 +85,19 @@
cl::desc("Generate software floating point library calls"),
cl::location(UseSoftFloat),
cl::init(false));
+static cl::opt<llvm::FloatABI::ABIType, true>
+FloatABIForCalls("float-abi",
+ cl::desc("Choose float ABI type"),
+ cl::location(FloatABIType),
+ cl::init(FloatABI::Default),
+ cl::values(
+ clEnumValN(FloatABI::Default, "default",
+ "Target default float ABI type"),
+ clEnumValN(FloatABI::Soft, "soft",
+ "Soft float ABI (implied by -soft-float)"),
+ clEnumValN(FloatABI::Hard, "hard",
+ "Hard float ABI (uses FP registers)"),
+ clEnumValEnd));
static cl::opt<bool, true>
DontPlaceZerosInBSS("nozero-initialized-in-bss",
cl::desc("Don't place zero-initialized symbols into bss section"),
@@ -162,6 +176,14 @@
// TargetMachine Class
//
+TargetMachine::TargetMachine()
+ : AsmInfo(0) {
+ // Typically it will be subtargets that will adjust FloatABIType from Default
+ // to Soft or Hard.
+ if (UseSoftFloat)
+ FloatABIType = FloatABI::Soft;
+}
+
TargetMachine::~TargetMachine() {
delete AsmInfo;
}
diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp
index 03ce1ae..56983ce 100644
--- a/lib/Target/X86/X86Subtarget.cpp
+++ b/lib/Target/X86/X86Subtarget.cpp
@@ -350,6 +350,10 @@
, MaxInlineSizeThreshold(128)
, Is64Bit(is64Bit)
, TargetType(isELF) { // Default to ELF unless otherwise specified.
+
+ // default to hard float ABI
+ if (FloatABIType == FloatABI::Default)
+ FloatABIType = FloatABI::Hard;
// Determine default and user specified characteristics
if (!FS.empty()) {