1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
llvm-mirror/lib/Target/Mips/Mips16HardFloat.cpp
Bjorn Pettersson 29ffba4b56 Update @llvm.powi to handle different int sizes for the exponent
This can be seen as a follow up to commit 0ee439b705e82a4fe20e2,
that changed the second argument of __powidf2, __powisf2 and
__powitf2 in compiler-rt from si_int to int. That was to align with
how those runtimes are defined in libgcc.
One thing that seem to have been missing in that patch was to make
sure that the rest of LLVM also handle that the argument now depends
on the size of int (not using the si_int machine mode for 32-bit).
When using __builtin_powi for a target with 16-bit int clang crashed.
And when emitting libcalls to those rtlib functions, typically when
lowering @llvm.powi), the backend would always prepare the exponent
argument as an i32 which caused miscompiles when the rtlib was
compiled with 16-bit int.

The solution used here is to use an overloaded type for the second
argument in @llvm.powi. This way clang can use the "correct" type
when lowering __builtin_powi, and then later when emitting the libcall
it is assumed that the type used in @llvm.powi matches the rtlib
function.

One thing that needed some extra attention was that when vectorizing
calls several passes did not support that several arguments could
be overloaded in the intrinsics. This patch allows overload of a
scalar operand by adding hasVectorInstrinsicOverloadedScalarOpd, with
an entry for powi.

Differential Revision: https://reviews.llvm.org/D99439
2021-06-17 09:38:28 +02:00

537 lines
16 KiB
C++

//===- Mips16HardFloat.cpp for Mips16 Hard Float --------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file defines a pass needed for Mips16 Hard Float
//
//===----------------------------------------------------------------------===//
#include "MipsTargetMachine.h"
#include "llvm/CodeGen/TargetPassConfig.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Value.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
#include <string>
using namespace llvm;
#define DEBUG_TYPE "mips16-hard-float"
namespace {
class Mips16HardFloat : public ModulePass {
public:
static char ID;
Mips16HardFloat() : ModulePass(ID) {}
StringRef getPassName() const override { return "MIPS16 Hard Float Pass"; }
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.addRequired<TargetPassConfig>();
ModulePass::getAnalysisUsage(AU);
}
bool runOnModule(Module &M) override;
};
} // end anonymous namespace
static void emitInlineAsm(LLVMContext &C, BasicBlock *BB, StringRef AsmText) {
std::vector<Type *> AsmArgTypes;
std::vector<Value *> AsmArgs;
FunctionType *AsmFTy =
FunctionType::get(Type::getVoidTy(C), AsmArgTypes, false);
InlineAsm *IA = InlineAsm::get(AsmFTy, AsmText, "", true,
/* IsAlignStack */ false, InlineAsm::AD_ATT);
CallInst::Create(IA, AsmArgs, "", BB);
}
char Mips16HardFloat::ID = 0;
//
// Return types that matter for hard float are:
// float, double, complex float, and complex double
//
enum FPReturnVariant {
FRet, DRet, CFRet, CDRet, NoFPRet
};
//
// Determine which FP return type this function has
//
static FPReturnVariant whichFPReturnVariant(Type *T) {
switch (T->getTypeID()) {
case Type::FloatTyID:
return FRet;
case Type::DoubleTyID:
return DRet;
case Type::StructTyID: {
StructType *ST = cast<StructType>(T);
if (ST->getNumElements() != 2)
break;
if ((ST->getElementType(0)->isFloatTy()) &&
(ST->getElementType(1)->isFloatTy()))
return CFRet;
if ((ST->getElementType(0)->isDoubleTy()) &&
(ST->getElementType(1)->isDoubleTy()))
return CDRet;
break;
}
default:
break;
}
return NoFPRet;
}
// Parameter type that matter are float, (float, float), (float, double),
// double, (double, double), (double, float)
enum FPParamVariant {
FSig, FFSig, FDSig,
DSig, DDSig, DFSig, NoSig
};
// which floating point parameter signature variant we are dealing with
using TypeID = Type::TypeID;
const Type::TypeID FloatTyID = Type::FloatTyID;
const Type::TypeID DoubleTyID = Type::DoubleTyID;
static FPParamVariant whichFPParamVariantNeeded(Function &F) {
switch (F.arg_size()) {
case 0:
return NoSig;
case 1:{
TypeID ArgTypeID = F.getFunctionType()->getParamType(0)->getTypeID();
switch (ArgTypeID) {
case FloatTyID:
return FSig;
case DoubleTyID:
return DSig;
default:
return NoSig;
}
}
default: {
TypeID ArgTypeID0 = F.getFunctionType()->getParamType(0)->getTypeID();
TypeID ArgTypeID1 = F.getFunctionType()->getParamType(1)->getTypeID();
switch(ArgTypeID0) {
case FloatTyID: {
switch (ArgTypeID1) {
case FloatTyID:
return FFSig;
case DoubleTyID:
return FDSig;
default:
return FSig;
}
}
case DoubleTyID: {
switch (ArgTypeID1) {
case FloatTyID:
return DFSig;
case DoubleTyID:
return DDSig;
default:
return DSig;
}
}
default:
return NoSig;
}
}
}
llvm_unreachable("can't get here");
}
// Figure out if we need float point based on the function parameters.
// We need to move variables in and/or out of floating point
// registers because of the ABI
static bool needsFPStubFromParams(Function &F) {
if (F.arg_size() >=1) {
Type *ArgType = F.getFunctionType()->getParamType(0);
switch (ArgType->getTypeID()) {
case Type::FloatTyID:
case Type::DoubleTyID:
return true;
default:
break;
}
}
return false;
}
static bool needsFPReturnHelper(Function &F) {
Type* RetType = F.getReturnType();
return whichFPReturnVariant(RetType) != NoFPRet;
}
static bool needsFPReturnHelper(FunctionType &FT) {
Type* RetType = FT.getReturnType();
return whichFPReturnVariant(RetType) != NoFPRet;
}
static bool needsFPHelperFromSig(Function &F) {
return needsFPStubFromParams(F) || needsFPReturnHelper(F);
}
// We swap between FP and Integer registers to allow Mips16 and Mips32 to
// interoperate
static std::string swapFPIntParams(FPParamVariant PV, Module *M, bool LE,
bool ToFP) {
std::string MI = ToFP ? "mtc1 ": "mfc1 ";
std::string AsmText;
switch (PV) {
case FSig:
AsmText += MI + "$$4, $$f12\n";
break;
case FFSig:
AsmText += MI + "$$4, $$f12\n";
AsmText += MI + "$$5, $$f14\n";
break;
case FDSig:
AsmText += MI + "$$4, $$f12\n";
if (LE) {
AsmText += MI + "$$6, $$f14\n";
AsmText += MI + "$$7, $$f15\n";
} else {
AsmText += MI + "$$7, $$f14\n";
AsmText += MI + "$$6, $$f15\n";
}
break;
case DSig:
if (LE) {
AsmText += MI + "$$4, $$f12\n";
AsmText += MI + "$$5, $$f13\n";
} else {
AsmText += MI + "$$5, $$f12\n";
AsmText += MI + "$$4, $$f13\n";
}
break;
case DDSig:
if (LE) {
AsmText += MI + "$$4, $$f12\n";
AsmText += MI + "$$5, $$f13\n";
AsmText += MI + "$$6, $$f14\n";
AsmText += MI + "$$7, $$f15\n";
} else {
AsmText += MI + "$$5, $$f12\n";
AsmText += MI + "$$4, $$f13\n";
AsmText += MI + "$$7, $$f14\n";
AsmText += MI + "$$6, $$f15\n";
}
break;
case DFSig:
if (LE) {
AsmText += MI + "$$4, $$f12\n";
AsmText += MI + "$$5, $$f13\n";
} else {
AsmText += MI + "$$5, $$f12\n";
AsmText += MI + "$$4, $$f13\n";
}
AsmText += MI + "$$6, $$f14\n";
break;
case NoSig:
break;
}
return AsmText;
}
// Make sure that we know we already need a stub for this function.
// Having called needsFPHelperFromSig
static void assureFPCallStub(Function &F, Module *M,
const MipsTargetMachine &TM) {
// for now we only need them for static relocation
if (TM.isPositionIndependent())
return;
LLVMContext &Context = M->getContext();
bool LE = TM.isLittleEndian();
std::string Name(F.getName());
std::string SectionName = ".mips16.call.fp." + Name;
std::string StubName = "__call_stub_fp_" + Name;
//
// see if we already have the stub
//
Function *FStub = M->getFunction(StubName);
if (FStub && !FStub->isDeclaration()) return;
FStub = Function::Create(F.getFunctionType(),
Function::InternalLinkage, StubName, M);
FStub->addFnAttr("mips16_fp_stub");
FStub->addFnAttr(Attribute::Naked);
FStub->addFnAttr(Attribute::NoInline);
FStub->addFnAttr(Attribute::NoUnwind);
FStub->addFnAttr("nomips16");
FStub->setSection(SectionName);
BasicBlock *BB = BasicBlock::Create(Context, "entry", FStub);
FPReturnVariant RV = whichFPReturnVariant(FStub->getReturnType());
FPParamVariant PV = whichFPParamVariantNeeded(F);
std::string AsmText;
AsmText += ".set reorder\n";
AsmText += swapFPIntParams(PV, M, LE, true);
if (RV != NoFPRet) {
AsmText += "move $$18, $$31\n";
AsmText += "jal " + Name + "\n";
} else {
AsmText += "lui $$25, %hi(" + Name + ")\n";
AsmText += "addiu $$25, $$25, %lo(" + Name + ")\n";
}
switch (RV) {
case FRet:
AsmText += "mfc1 $$2, $$f0\n";
break;
case DRet:
if (LE) {
AsmText += "mfc1 $$2, $$f0\n";
AsmText += "mfc1 $$3, $$f1\n";
} else {
AsmText += "mfc1 $$3, $$f0\n";
AsmText += "mfc1 $$2, $$f1\n";
}
break;
case CFRet:
if (LE) {
AsmText += "mfc1 $$2, $$f0\n";
AsmText += "mfc1 $$3, $$f2\n";
} else {
AsmText += "mfc1 $$3, $$f0\n";
AsmText += "mfc1 $$3, $$f2\n";
}
break;
case CDRet:
if (LE) {
AsmText += "mfc1 $$4, $$f2\n";
AsmText += "mfc1 $$5, $$f3\n";
AsmText += "mfc1 $$2, $$f0\n";
AsmText += "mfc1 $$3, $$f1\n";
} else {
AsmText += "mfc1 $$5, $$f2\n";
AsmText += "mfc1 $$4, $$f3\n";
AsmText += "mfc1 $$3, $$f0\n";
AsmText += "mfc1 $$2, $$f1\n";
}
break;
case NoFPRet:
break;
}
if (RV != NoFPRet)
AsmText += "jr $$18\n";
else
AsmText += "jr $$25\n";
emitInlineAsm(Context, BB, AsmText);
new UnreachableInst(Context, BB);
}
// Functions that are llvm intrinsics and don't need helpers.
static const char *const IntrinsicInline[] = {
"fabs", "fabsf",
"llvm.ceil.f32", "llvm.ceil.f64",
"llvm.copysign.f32", "llvm.copysign.f64",
"llvm.cos.f32", "llvm.cos.f64",
"llvm.exp.f32", "llvm.exp.f64",
"llvm.exp2.f32", "llvm.exp2.f64",
"llvm.fabs.f32", "llvm.fabs.f64",
"llvm.floor.f32", "llvm.floor.f64",
"llvm.fma.f32", "llvm.fma.f64",
"llvm.log.f32", "llvm.log.f64",
"llvm.log10.f32", "llvm.log10.f64",
"llvm.nearbyint.f32", "llvm.nearbyint.f64",
"llvm.pow.f32", "llvm.pow.f64",
"llvm.powi.f32.i32", "llvm.powi.f64.i32",
"llvm.rint.f32", "llvm.rint.f64",
"llvm.round.f32", "llvm.round.f64",
"llvm.sin.f32", "llvm.sin.f64",
"llvm.sqrt.f32", "llvm.sqrt.f64",
"llvm.trunc.f32", "llvm.trunc.f64",
};
static bool isIntrinsicInline(Function *F) {
return std::binary_search(std::begin(IntrinsicInline),
std::end(IntrinsicInline), F->getName());
}
// Returns of float, double and complex need to be handled with a helper
// function.
static bool fixupFPReturnAndCall(Function &F, Module *M,
const MipsTargetMachine &TM) {
bool Modified = false;
LLVMContext &C = M->getContext();
Type *MyVoid = Type::getVoidTy(C);
for (auto &BB: F)
for (auto &I: BB) {
if (const ReturnInst *RI = dyn_cast<ReturnInst>(&I)) {
Value *RVal = RI->getReturnValue();
if (!RVal) continue;
//
// If there is a return value and it needs a helper function,
// figure out which one and add a call before the actual
// return to this helper. The purpose of the helper is to move
// floating point values from their soft float return mapping to
// where they would have been mapped to in floating point registers.
//
Type *T = RVal->getType();
FPReturnVariant RV = whichFPReturnVariant(T);
if (RV == NoFPRet) continue;
static const char *const Helper[NoFPRet] = {
"__mips16_ret_sf", "__mips16_ret_df", "__mips16_ret_sc",
"__mips16_ret_dc"
};
const char *Name = Helper[RV];
AttributeList A;
Value *Params[] = {RVal};
Modified = true;
//
// These helper functions have a different calling ABI so
// this __Mips16RetHelper indicates that so that later
// during call setup, the proper call lowering to the helper
// functions will take place.
//
A = A.addAttribute(C, AttributeList::FunctionIndex,
"__Mips16RetHelper");
A = A.addAttribute(C, AttributeList::FunctionIndex,
Attribute::ReadNone);
A = A.addAttribute(C, AttributeList::FunctionIndex,
Attribute::NoInline);
FunctionCallee F = (M->getOrInsertFunction(Name, A, MyVoid, T));
CallInst::Create(F, Params, "", &I);
} else if (const CallInst *CI = dyn_cast<CallInst>(&I)) {
FunctionType *FT = CI->getFunctionType();
Function *F_ = CI->getCalledFunction();
if (needsFPReturnHelper(*FT) &&
!(F_ && isIntrinsicInline(F_))) {
Modified=true;
F.addFnAttr("saveS2");
}
if (F_ && !isIntrinsicInline(F_)) {
// pic mode calls are handled by already defined
// helper functions
if (needsFPReturnHelper(*F_)) {
Modified=true;
F.addFnAttr("saveS2");
}
if (!TM.isPositionIndependent()) {
if (needsFPHelperFromSig(*F_)) {
assureFPCallStub(*F_, M, TM);
Modified=true;
}
}
}
}
}
return Modified;
}
static void createFPFnStub(Function *F, Module *M, FPParamVariant PV,
const MipsTargetMachine &TM) {
bool PicMode = TM.isPositionIndependent();
bool LE = TM.isLittleEndian();
LLVMContext &Context = M->getContext();
std::string Name(F->getName());
std::string SectionName = ".mips16.fn." + Name;
std::string StubName = "__fn_stub_" + Name;
std::string LocalName = "$$__fn_local_" + Name;
Function *FStub = Function::Create
(F->getFunctionType(),
Function::InternalLinkage, StubName, M);
FStub->addFnAttr("mips16_fp_stub");
FStub->addFnAttr(Attribute::Naked);
FStub->addFnAttr(Attribute::NoUnwind);
FStub->addFnAttr(Attribute::NoInline);
FStub->addFnAttr("nomips16");
FStub->setSection(SectionName);
BasicBlock *BB = BasicBlock::Create(Context, "entry", FStub);
std::string AsmText;
if (PicMode) {
AsmText += ".set noreorder\n";
AsmText += ".cpload $$25\n";
AsmText += ".set reorder\n";
AsmText += ".reloc 0, R_MIPS_NONE, " + Name + "\n";
AsmText += "la $$25, " + LocalName + "\n";
} else
AsmText += "la $$25, " + Name + "\n";
AsmText += swapFPIntParams(PV, M, LE, false);
AsmText += "jr $$25\n";
AsmText += LocalName + " = " + Name + "\n";
emitInlineAsm(Context, BB, AsmText);
new UnreachableInst(FStub->getContext(), BB);
}
// remove the use-soft-float attribute
static void removeUseSoftFloat(Function &F) {
AttrBuilder B;
LLVM_DEBUG(errs() << "removing -use-soft-float\n");
B.addAttribute("use-soft-float", "false");
F.removeAttributes(AttributeList::FunctionIndex, B);
if (F.hasFnAttribute("use-soft-float")) {
LLVM_DEBUG(errs() << "still has -use-soft-float\n");
}
F.addAttributes(AttributeList::FunctionIndex, B);
}
// This pass only makes sense when the underlying chip has floating point but
// we are compiling as mips16.
// For all mips16 functions (that are not stubs we have already generated), or
// declared via attributes as nomips16, we must:
// 1) fixup all returns of float, double, single and double complex
// by calling a helper function before the actual return.
// 2) generate helper functions (stubs) that can be called by mips32
// functions that will move parameters passed normally passed in
// floating point
// registers the soft float equivalents.
// 3) in the case of static relocation, generate helper functions so that
// mips16 functions can call extern functions of unknown type (mips16 or
// mips32).
// 4) TBD. For pic, calls to extern functions of unknown type are handled by
// predefined helper functions in libc but this work is currently done
// during call lowering but it should be moved here in the future.
bool Mips16HardFloat::runOnModule(Module &M) {
auto &TM = static_cast<const MipsTargetMachine &>(
getAnalysis<TargetPassConfig>().getTM<TargetMachine>());
LLVM_DEBUG(errs() << "Run on Module Mips16HardFloat\n");
bool Modified = false;
for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
if (F->hasFnAttribute("nomips16") &&
F->hasFnAttribute("use-soft-float")) {
removeUseSoftFloat(*F);
continue;
}
if (F->isDeclaration() || F->hasFnAttribute("mips16_fp_stub") ||
F->hasFnAttribute("nomips16")) continue;
Modified |= fixupFPReturnAndCall(*F, &M, TM);
FPParamVariant V = whichFPParamVariantNeeded(*F);
if (V != NoSig) {
Modified = true;
createFPFnStub(&*F, &M, V, TM);
}
}
return Modified;
}
ModulePass *llvm::createMips16HardFloatPass() {
return new Mips16HardFloat();
}