Files
FEX-Emu--FEX/FEXCore/Source/Interface/IR/IRDumper.cpp
T
Alyssa Rosenzweig e6db2d0b96 IR: Remove phi nodes
It turns out that pure SSA isn't a great choice for the sort of emulation we do.
On one hand, it discards information from the guest binary's register allocation
that would let us skip stuff. On the other hand, it doesn't have nearly as many
benefits in this setting as in a traditional compiler... We really *don't* want
to do global RA or really any global optimization. We assume the guest optimizer
did its job for x86, we just need to clean up the mess left from going x86 ->
arm. So we just need enough SSA to peephole optimize.

My concrete IR proposals are that:

  * SSA values must be killed in the same block that they are defined.
  * Explicit LoadGPR/StoreGPR instructions can be used for global persistence.
  * LoadGPR/StoreGPR are eliminated in favour of SSA within a block.

This has a lot of nice properties for our setting:

  * Except for some internal REP instruction emulation (etc), we already have
    registers for everything that escapes block boundaries, so this form is very
    easy to go into -- straightforward local value numbering, not a full into
    SSA pass.

  * Spilling is entirely local (if it happens at all), since everything is in
    registers at block boundaries. This is excellent, because Belady's algorithm
    lets us spill nearly optimally in linear-time for individual blocks. (And
    the global version of Belady's algorithm is massively more complicated...)
    A nice fit for a JIT.

    Relatedly, it turns out allowing spilling is probably a decent decision,
    since the same spiller code can be used to rematerialize constants in a
    straightforward way. This is an issue with the current RA.

  * Register assignment is entirely local. For the same reason, we can assign
    registers "optimally" in linear time & memory (e.g. with linear scan). And
    the impl is massively simpler than a full blown SSA-based tree scan RA. For
    example, we don't have to worry about parallel copies or coalescing phis or
    anything. Massively nicer algorithm to deal with.

  * SSA value names can be block local which makes the validation implicit :~)

It also has remarkably few drawbacks, because we didn't want to do CFG global
optimization anyway given our time budget and the diminishng returns. The few
global optimizations we might want (flag escape analysis?) don't necessarily
benefit from pure SSA anyway.

Anyway, we explicitly don't want phi nodes in any of this. They're currently
unused. Let's just remove them so nobody gets the bright idea of changing that.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
2023-09-05 16:35:12 -04:00

360 lines
11 KiB
C++

/*
$info$
meta: ir|dumper ~ IR -> Text
tags: ir|dumper
$end_info$
*/
#include <FEXCore/IR/IR.h>
#include <FEXCore/IR/IntrusiveIRList.h>
#include <FEXCore/IR/RegisterAllocationData.h>
#include <FEXCore/fextl/sstream.h>
#include <algorithm>
#include <array>
#include <ostream>
#include <stdint.h>
#include <string_view>
#include <iomanip>
namespace FEXCore::IR {
#define IROP_GETNAME_IMPL
#define IROP_GETRAARGS_IMPL
#define IROP_REG_CLASSES_IMPL
#define IROP_HASSIDEEFFECTS_IMPL
#define IROP_SIZES_IMPL
#define IROP_GETHASDEST_IMPL
#include <FEXCore/IR/IRDefines.inc>
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, const SHA256Sum &Arg) {
*out << "sha256:";
for(auto byte: Arg.data)
*out << std::hex << std::setfill('0') << std::setw(2) << (unsigned int)byte;
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, uint64_t Arg) {
*out << "#0x" << std::hex << Arg;
}
[[maybe_unused]]
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, const char* Arg) {
*out << Arg;
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, CondClassType Arg) {
static constexpr std::array<std::string_view, 22> CondNames = {
"EQ",
"NEQ",
"UGE",
"ULT",
"MI",
"PL",
"VS",
"VC",
"UGT",
"ULE",
"SGE",
"SLT",
"SGT",
"SLE",
"Invalid Cond",
"Invalid Cond",
"FLU",
"FGE",
"FLEU",
"FGT",
"FU",
"FNU"
};
*out << CondNames[Arg];
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, MemOffsetType Arg) {
static constexpr std::array<std::string_view, 3> Names = {
"SXTX",
"UXTW",
"SXTW",
};
*out << Names[Arg];
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, RegisterClassType Arg) {
if (Arg == GPRClass.Val)
*out << "GPR";
else if (Arg == GPRFixedClass.Val)
*out << "GPRFixed";
else if (Arg == FPRClass.Val)
*out << "FPR";
else if (Arg == FPRFixedClass.Val)
*out << "FPRFixed";
else if (Arg == GPRPairClass.Val)
*out << "GPRPair";
else
*out << "Unknown Registerclass " << Arg;
}
static void PrintArg(fextl::stringstream *out, IRListView const* IR, OrderedNodeWrapper Arg, IR::RegisterAllocationData *RAData) {
auto [CodeNode, IROp] = IR->at(Arg)();
const auto ArgID = Arg.ID();
if (ArgID.IsInvalid()) {
*out << "%Invalid";
} else {
*out << "%" << std::dec << ArgID;
if (RAData) {
auto PhyReg = RAData->GetNodeRegister(ArgID);
switch (PhyReg.Class) {
case FEXCore::IR::GPRClass.Val: *out << "(GPR"; break;
case FEXCore::IR::GPRFixedClass.Val: *out << "(GPRFixed"; break;
case FEXCore::IR::FPRClass.Val: *out << "(FPR"; break;
case FEXCore::IR::FPRFixedClass.Val: *out << "(FPRFixed"; break;
case FEXCore::IR::GPRPairClass.Val: *out << "(GPRPair"; break;
case FEXCore::IR::ComplexClass.Val: *out << "(Complex"; break;
case FEXCore::IR::InvalidClass.Val: *out << "(Invalid"; break;
default: *out << "(Unknown"; break;
}
if (PhyReg.Class != FEXCore::IR::InvalidClass.Val) {
*out << std::dec << (uint32_t)PhyReg.Reg << ")";
} else {
*out << ")";
}
}
}
if (GetHasDest(IROp->Op)) {
uint32_t ElementSize = IROp->ElementSize;
uint32_t NumElements = IROp->Size;
if (!IROp->ElementSize) {
ElementSize = IROp->Size;
}
if (ElementSize) {
NumElements /= ElementSize;
}
*out << " i" << std::dec << (ElementSize * 8);
if (NumElements > 1) {
*out << "v" << std::dec << NumElements;
}
}
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, FEXCore::IR::FenceType Arg) {
if (Arg == IR::Fence_Load) {
*out << "Loads";
}
else if (Arg == IR::Fence_Store) {
*out << "Stores";
}
else if (Arg == IR::Fence_LoadStore) {
*out << "LoadStores";
}
else {
*out << "<Unknown Fence Type>";
}
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, FEXCore::IR::RoundType Arg) {
switch (Arg) {
case FEXCore::IR::Round_Nearest: *out << "Nearest"; break;
case FEXCore::IR::Round_Negative_Infinity: *out << "-Inf"; break;
case FEXCore::IR::Round_Positive_Infinity: *out << "+Inf"; break;
case FEXCore::IR::Round_Towards_Zero: *out << "Towards Zero"; break;
case FEXCore::IR::Round_Host: *out << "Host"; break;
default: *out << "<Unknown Round Type>"; break;
}
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, FEXCore::IR::SyscallFlags Arg) {
switch (Arg) {
case FEXCore::IR::SyscallFlags::DEFAULT: *out << "Default"; break;
case FEXCore::IR::SyscallFlags::OPTIMIZETHROUGH: *out << "Optimize Through"; break;
case FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY: *out << "No Sync State on Entry"; break;
case FEXCore::IR::SyscallFlags::NORETURN: *out << "No Return"; break;
case FEXCore::IR::SyscallFlags::NOSIDEEFFECTS: *out << "No Side Effects"; break;
default: *out << "<Unknown Round Type>"; break;
}
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, FEXCore::IR::NamedVectorConstant Arg) {
switch (Arg) {
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_INCREMENTAL_U16_INDEX: {
*out << "u16_incremental_index";
break;
}
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_INCREMENTAL_U16_INDEX_UPPER: {
*out << "u16_incremental_index_upper";
break;
}
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_PADDSUBPS_INVERT: {
*out << "addsubps_invert";
break;
}
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_PADDSUBPS_INVERT_UPPER: {
*out << "addsubps_invert_upper";
break;
}
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_PADDSUBPD_INVERT: {
*out << "addsubpd_invert";
break;
}
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_PADDSUBPD_INVERT_UPPER: {
*out << "addsubpd_invert_upper";
break;
}
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_MOVMSKPS_SHIFT: {
*out << "movmskps_shift";
break;
}
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_AESKEYGENASSIST_SWIZZLE: {
*out << "aeskeygenassist_swizzle";
break;
}
case FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_ZERO: {
*out << "vectorzero";
break;
}
default: *out << "<Unknown Named Vector Constant>"; break;
}
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, FEXCore::IR::OpSize Arg) {
switch (Arg) {
case OpSize::i8Bit: *out << "i8"; break;
case OpSize::i16Bit: *out << "i16"; break;
case OpSize::i32Bit: *out << "i32"; break;
case OpSize::i64Bit: *out << "i64"; break;
case OpSize::i128Bit: *out << "i128"; break;
case OpSize::i256Bit: *out << "i256"; break;
default: *out << "<Unknown OpSize Type>"; break;
}
}
static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, FEXCore::IR::BreakDefinition Arg) {
*out << "{" << Arg.ErrorRegister << ".";
*out << static_cast<uint32_t>(Arg.Signal) << ".";
*out << static_cast<uint32_t>(Arg.TrapNumber) << ".";
*out << static_cast<uint32_t>(Arg.si_code) << "}";
}
void Dump(fextl::stringstream *out, IRListView const* IR, IR::RegisterAllocationData *RAData) {
auto HeaderOp = IR->GetHeader();
int8_t CurrentIndent = 0;
auto AddIndent = [&out, &CurrentIndent]() {
for (uint8_t i = 0; i < CurrentIndent; ++i) {
*out << "\t";
}
};
++CurrentIndent;
AddIndent();
*out << "(%0) " << "IRHeader ";
*out << "%" << HeaderOp->Blocks.ID() << ", ";
*out << "#" << std::dec << HeaderOp->OriginalRIP << ", ";
*out << "#" << std::dec << HeaderOp->BlockCount << ", ";
*out << "#" << std::dec << HeaderOp->NumHostInstructions << std::endl;
for (auto [BlockNode, BlockHeader] : IR->GetBlocks()) {
{
auto BlockIROp = BlockHeader->C<FEXCore::IR::IROp_CodeBlock>();
AddIndent();
*out << "(%" << IR->GetID(BlockNode) << ") " << "CodeBlock ";
*out << "%" << BlockIROp->Begin.ID() << ", ";
*out << "%" << BlockIROp->Last.ID() << std::endl;
}
++CurrentIndent;
for (auto [CodeNode, IROp] : IR->GetCode(BlockNode)) {
const auto ID = IR->GetID(CodeNode);
const auto Name = FEXCore::IR::GetName(IROp->Op);
{
AddIndent();
if (GetHasDest(IROp->Op)) {
uint32_t ElementSize = IROp->ElementSize;
uint32_t NumElements = IROp->Size;
if (!IROp->ElementSize) {
ElementSize = IROp->Size;
}
if (ElementSize) {
NumElements /= ElementSize;
}
*out << "%" << std::dec << ID;
if (RAData) {
auto PhyReg = RAData->GetNodeRegister(ID);
switch (PhyReg.Class) {
case FEXCore::IR::GPRClass.Val: *out << "(GPR"; break;
case FEXCore::IR::GPRFixedClass.Val: *out << "(GPRFixed"; break;
case FEXCore::IR::FPRClass.Val: *out << "(FPR"; break;
case FEXCore::IR::FPRFixedClass.Val: *out << "(FPRFixed"; break;
case FEXCore::IR::GPRPairClass.Val: *out << "(GPRPair"; break;
case FEXCore::IR::ComplexClass.Val: *out << "(Complex"; break;
case FEXCore::IR::InvalidClass.Val: *out << "(Invalid"; break;
default: *out << "(Unknown"; break;
}
if (PhyReg.Class != FEXCore::IR::InvalidClass.Val) {
*out << std::dec << (uint32_t)PhyReg.Reg << ")";
} else {
*out << ")";
}
}
*out << " i" << std::dec << (ElementSize * 8);
if (NumElements > 1) {
*out << "v" << std::dec << NumElements;
}
*out << " = ";
}
else {
uint32_t ElementSize = IROp->ElementSize;
if (!IROp->ElementSize) {
ElementSize = IROp->Size;
}
uint32_t NumElements = 0;
if (ElementSize) {
NumElements = IROp->Size / ElementSize;
}
*out << "(%" << std::dec << ID << ' ';
*out << 'i' << std::dec << (ElementSize * 8);
if (NumElements > 1) {
*out << 'v' << std::dec << NumElements;
}
*out << ") ";
}
*out << Name;
#define IROP_ARGPRINTER_HELPER
#include <FEXCore/IR/IRDefines.inc>
default: *out << "<Unknown Args>"; break;
}
//*out << " (" << std::dec << CodeNode->GetUses() << ")";
*out << "\n";
}
}
CurrentIndent = std::max(0, CurrentIndent - 1);
}
}
}