fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Moving alloy/ into xenia/cpu/ to start simplifying things.

Browse Source
This commit is contained in:
Ben Vanik
2015-03-24 07:46:18 -07:00
parent 59395318f3
commit 29912f44c0
519 changed files with 2246 additions and 2296 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
src/xenia/cpu/backend/assembler.cc Normal file
View File

@@ -0,0 +1,26 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/assembler.h"
namespace xe {
namespace cpu {
namespace backend {
Assembler::Assembler(Backend* backend) : backend_(backend) {}
Assembler::~Assembler() { Reset(); }
int Assembler::Initialize() { return 0; }
void Assembler::Reset() {}
} // namespace backend
} // namespace cpu
} // namespace xe

58
src/xenia/cpu/backend/assembler.h Normal file
View File

@@ -0,0 +1,58 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_ASSEMBLER_H_
#define XENIA_BACKEND_ASSEMBLER_H_
#include <memory>
namespace xe {
namespace cpu {
namespace hir {
class HIRBuilder;
} // namespace hir
namespace runtime {
class DebugInfo;
class Function;
class FunctionInfo;
class Runtime;
} // namespace runtime
} // namespace cpu
} // namespace xe
namespace xe {
namespace cpu {
namespace backend {
class Backend;
class Assembler {
public:
Assembler(Backend* backend);
virtual ~Assembler();
virtual int Initialize();
virtual void Reset();
virtual int Assemble(runtime::FunctionInfo* symbol_info,
hir::HIRBuilder* builder, uint32_t debug_info_flags,
std::unique_ptr<runtime::DebugInfo> debug_info,
uint32_t trace_flags,
runtime::Function** out_function) = 0;
protected:
Backend* backend_;
};
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_ASSEMBLER_H_

32
src/xenia/cpu/backend/backend.cc Normal file
View File

@@ -0,0 +1,32 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/backend.h"
namespace xe {
namespace cpu {
namespace backend {
using xe::cpu::runtime::Runtime;
Backend::Backend(Runtime* runtime) : runtime_(runtime) {
memset(&machine_info_, 0, sizeof(machine_info_));
}
Backend::~Backend() = default;
int Backend::Initialize() { return 0; }
void* Backend::AllocThreadData() { return nullptr; }
void Backend::FreeThreadData(void* thread_data) {}
} // namespace backend
} // namespace cpu
} // namespace xe

55
src/xenia/cpu/backend/backend.h Normal file
View File

@@ -0,0 +1,55 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_BACKEND_H_
#define XENIA_BACKEND_BACKEND_H_
#include <memory>
#include "xenia/cpu/backend/machine_info.h"
namespace xe {
namespace cpu {
namespace runtime {
class Runtime;
} // namespace runtime
} // namespace cpu
} // namespace xe
namespace xe {
namespace cpu {
namespace backend {
class Assembler;
class Backend {
public:
Backend(runtime::Runtime* runtime);
virtual ~Backend();
runtime::Runtime* runtime() const { return runtime_; }
const MachineInfo* machine_info() const { return &machine_info_; }
virtual int Initialize();
virtual void* AllocThreadData();
virtual void FreeThreadData(void* thread_data);
virtual std::unique_ptr<Assembler> CreateAssembler() = 0;
protected:
runtime::Runtime* runtime_;
MachineInfo machine_info_;
};
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_BACKEND_H_

37
src/xenia/cpu/backend/machine_info.h Normal file
View File

@@ -0,0 +1,37 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2014 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_MACHINE_INFO_H_
#define XENIA_BACKEND_MACHINE_INFO_H_
#include <cstdint>
namespace xe {
namespace cpu {
namespace backend {
struct MachineInfo {
struct RegisterSet {
enum Types {
INT_TYPES = (1 << 1),
FLOAT_TYPES = (1 << 2),
VEC_TYPES = (1 << 3),
};
uint8_t id;
char name[4];
uint32_t types;
uint32_t count;
} register_sets[8];
};
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_MACHINE_INFO_H_

14
src/xenia/cpu/backend/sources.gypi Normal file
View File

@@ -0,0 +1,14 @@
# Copyright 2013 Ben Vanik. All Rights Reserved.
{
'sources': [
'assembler.cc',
'assembler.h',
'backend.cc',
'backend.h',
'machine_info.h',
],
'includes': [
'x64/sources.gypi',
],
}

34
src/xenia/cpu/backend/x64/sources.gypi Normal file
View File

@@ -0,0 +1,34 @@
# Copyright 2013 Ben Vanik. All Rights Reserved.
{
'sources': [
'x64_assembler.cc',
'x64_assembler.h',
'x64_backend.cc',
'x64_backend.h',
'x64_code_cache.h',
'x64_emitter.cc',
'x64_emitter.h',
'x64_function.cc',
'x64_function.h',
'x64_sequence.inl',
'x64_sequences.cc',
'x64_sequences.h',
'x64_thunk_emitter.cc',
'x64_thunk_emitter.h',
'x64_tracers.cc',
'x64_tracers.h',
],
'conditions': [
['OS == "mac" or OS == "linux"', {
'sources': [
'x64_code_cache_posix.cc',
],
}],
['OS == "win"', {
'sources': [
'x64_code_cache_win.cc',
],
}],
],
}

135
src/xenia/cpu/backend/x64/x64_assembler.cc Normal file
View File

@@ -0,0 +1,135 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/x64/x64_assembler.h"
#include "xenia/cpu/backend/x64/x64_backend.h"
#include "xenia/cpu/backend/x64/x64_emitter.h"
#include "xenia/cpu/backend/x64/x64_function.h"
#include "xenia/cpu/hir/hir_builder.h"
#include "xenia/cpu/hir/label.h"
#include "xenia/cpu/runtime/runtime.h"
#include "poly/reset_scope.h"
#include "xenia/profiling.h"
namespace BE {
#include <beaengine/BeaEngine.h>
} // namespace BE
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
// TODO(benvanik): remove when enums redefined.
using namespace xe::cpu::runtime;
using xe::cpu::hir::HIRBuilder;
using xe::cpu::runtime::DebugInfo;
using xe::cpu::runtime::Function;
using xe::cpu::runtime::FunctionInfo;
X64Assembler::X64Assembler(X64Backend* backend)
: Assembler(backend), x64_backend_(backend) {}
X64Assembler::~X64Assembler() {
// Emitter must be freed before the allocator.
emitter_.reset();
allocator_.reset();
}
int X64Assembler::Initialize() {
int result = Assembler::Initialize();
if (result) {
return result;
}
allocator_.reset(new XbyakAllocator());
emitter_.reset(new X64Emitter(x64_backend_, allocator_.get()));
return result;
}
void X64Assembler::Reset() {
string_buffer_.Reset();
Assembler::Reset();
}
int X64Assembler::Assemble(FunctionInfo* symbol_info, HIRBuilder* builder,
uint32_t debug_info_flags,
std::unique_ptr<DebugInfo> debug_info,
uint32_t trace_flags, Function** out_function) {
SCOPE_profile_cpu_f("cpu");
// Reset when we leave.
poly::make_reset_scope(this);
// Lower HIR -> x64.
void* machine_code = 0;
size_t code_size = 0;
int result = emitter_->Emit(builder, debug_info_flags, debug_info.get(),
trace_flags, machine_code, code_size);
if (result) {
return result;
}
// Stash generated machine code.
if (debug_info_flags & DebugInfoFlags::DEBUG_INFO_MACHINE_CODE_DISASM) {
DumpMachineCode(debug_info.get(), machine_code, code_size, &string_buffer_);
debug_info->set_machine_code_disasm(string_buffer_.ToString());
string_buffer_.Reset();
}
{
X64Function* fn = new X64Function(symbol_info);
fn->set_debug_info(std::move(debug_info));
fn->Setup(machine_code, code_size);
*out_function = fn;
}
return 0;
}
void X64Assembler::DumpMachineCode(DebugInfo* debug_info, void* machine_code,
size_t code_size, poly::StringBuffer* str) {
BE::DISASM disasm = {0};
disasm.Archi = 64;
disasm.Options = BE::Tabulation + BE::MasmSyntax + BE::PrefixedNumeral;
disasm.EIP = (BE::UIntPtr)machine_code;
BE::UIntPtr eip_end = disasm.EIP + code_size;
uint64_t prev_source_offset = 0;
while (disasm.EIP < eip_end) {
// Look up source offset.
auto map_entry =
debug_info->LookupCodeOffset(disasm.EIP - (BE::UIntPtr)machine_code);
if (map_entry) {
if (map_entry->source_offset == prev_source_offset) {
str->Append(" ");
} else {
str->Append("%.8X ", map_entry->source_offset);
prev_source_offset = map_entry->source_offset;
}
} else {
str->Append("? ");
}
size_t len = BE::Disasm(&disasm);
if (len == BE::UNKNOWN_OPCODE) {
break;
}
str->Append("%p %s\n", disasm.EIP, disasm.CompleteInstr);
disasm.EIP += len;
}
}
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe

58
src/xenia/cpu/backend/x64/x64_assembler.h Normal file
View File

@@ -0,0 +1,58 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_X64_X64_ASSEMBLER_H_
#define XENIA_BACKEND_X64_X64_ASSEMBLER_H_
#include <memory>
#include "xenia/cpu/backend/assembler.h"
#include "poly/string_buffer.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
class X64Backend;
class X64Emitter;
class XbyakAllocator;
class X64Assembler : public Assembler {
public:
X64Assembler(X64Backend* backend);
~X64Assembler() override;
int Initialize() override;
void Reset() override;
int Assemble(runtime::FunctionInfo* symbol_info, hir::HIRBuilder* builder,
uint32_t debug_info_flags,
std::unique_ptr<runtime::DebugInfo> debug_info,
uint32_t trace_flags, runtime::Function** out_function) override;
private:
void DumpMachineCode(runtime::DebugInfo* debug_info, void* machine_code,
size_t code_size, poly::StringBuffer* str);
private:
X64Backend* x64_backend_;
std::unique_ptr<X64Emitter> emitter_;
std::unique_ptr<XbyakAllocator> allocator_;
poly::StringBuffer string_buffer_;
};
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_X64_X64_ASSEMBLER_H_

67
src/xenia/cpu/backend/x64/x64_backend.cc Normal file
View File

@@ -0,0 +1,67 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/x64/x64_backend.h"
#include "xenia/cpu/backend/x64/x64_assembler.h"
#include "xenia/cpu/backend/x64/x64_code_cache.h"
#include "xenia/cpu/backend/x64/x64_sequences.h"
#include "xenia/cpu/backend/x64/x64_thunk_emitter.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
using xe::cpu::runtime::Runtime;
X64Backend::X64Backend(Runtime* runtime) : Backend(runtime), code_cache_(0) {}
X64Backend::~X64Backend() { delete code_cache_; }
int X64Backend::Initialize() {
int result = Backend::Initialize();
if (result) {
return result;
}
RegisterSequences();
machine_info_.register_sets[0] = {
0, "gpr", MachineInfo::RegisterSet::INT_TYPES, X64Emitter::GPR_COUNT,
};
machine_info_.register_sets[1] = {
1, "xmm", MachineInfo::RegisterSet::FLOAT_TYPES |
MachineInfo::RegisterSet::VEC_TYPES,
X64Emitter::XMM_COUNT,
};
code_cache_ = new X64CodeCache();
result = code_cache_->Initialize();
if (result) {
return result;
}
// Generate thunks used to transition between jitted code and host code.
auto allocator = std::make_unique<XbyakAllocator>();
auto thunk_emitter = std::make_unique<X64ThunkEmitter>(this, allocator.get());
host_to_guest_thunk_ = thunk_emitter->EmitHostToGuestThunk();
guest_to_host_thunk_ = thunk_emitter->EmitGuestToHostThunk();
return result;
}
std::unique_ptr<Assembler> X64Backend::CreateAssembler() {
return std::make_unique<X64Assembler>(this);
}
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe

51
src/xenia/cpu/backend/x64/x64_backend.h Normal file
View File

@@ -0,0 +1,51 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_X64_X64_BACKEND_H_
#define XENIA_BACKEND_X64_X64_BACKEND_H_
#include "xenia/cpu/backend/backend.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
class X64CodeCache;
#define XENIA_HAS_X64_BACKEND 1
typedef void* (*HostToGuestThunk)(void* target, void* arg0, void* arg1);
typedef void* (*GuestToHostThunk)(void* target, void* arg0, void* arg1);
class X64Backend : public Backend {
public:
X64Backend(runtime::Runtime* runtime);
~X64Backend() override;
X64CodeCache* code_cache() const { return code_cache_; }
HostToGuestThunk host_to_guest_thunk() const { return host_to_guest_thunk_; }
GuestToHostThunk guest_to_host_thunk() const { return guest_to_host_thunk_; }
int Initialize() override;
std::unique_ptr<Assembler> CreateAssembler() override;
private:
X64CodeCache* code_cache_;
HostToGuestThunk host_to_guest_thunk_;
GuestToHostThunk guest_to_host_thunk_;
};
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_X64_X64_BACKEND_H_

48
src/xenia/cpu/backend/x64/x64_code_cache.h Normal file
View File

@@ -0,0 +1,48 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_X64_X64_CODE_CACHE_H_
#define XENIA_BACKEND_X64_X64_CODE_CACHE_H_
#include <mutex>
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
class X64CodeChunk;
class X64CodeCache {
public:
X64CodeCache(size_t chunk_size = DEFAULT_CHUNK_SIZE);
virtual ~X64CodeCache();
int Initialize();
// TODO(benvanik): ELF serialization/etc
// TODO(benvanik): keep track of code blocks
// TODO(benvanik): padding/guards/etc
void* PlaceCode(void* machine_code, size_t code_size, size_t stack_size);
private:
const static size_t DEFAULT_CHUNK_SIZE = 4 * 1024 * 1024;
std::mutex lock_;
size_t chunk_size_;
X64CodeChunk* head_chunk_;
X64CodeChunk* active_chunk_;
};
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_X64_X64_CODE_CACHE_H_

98
src/xenia/cpu/backend/x64/x64_code_cache_posix.cc Normal file
View File

@@ -0,0 +1,98 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2014 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/x64/x64_code_cache.h"
#include <sys/mman.h>
#include "poly/assert.h"
#include "poly/math.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
class X64CodeChunk {
public:
X64CodeChunk(size_t chunk_size);
~X64CodeChunk();
public:
X64CodeChunk* next;
size_t capacity;
uint8_t* buffer;
size_t offset;
};
X64CodeCache::X64CodeCache(size_t chunk_size)
: chunk_size_(chunk_size), head_chunk_(NULL), active_chunk_(NULL) {}
X64CodeCache::~X64CodeCache() {
std::lock_guard<std::mutex> guard(lock_);
auto chunk = head_chunk_;
while (chunk) {
auto next = chunk->next;
delete chunk;
chunk = next;
}
head_chunk_ = NULL;
}
int X64CodeCache::Initialize() { return 0; }
void* X64CodeCache::PlaceCode(void* machine_code, size_t code_size,
size_t stack_size) {
// Always move the code to land on 16b alignment. We do this by rounding up
// to 16b so that all offsets are aligned.
code_size = poly::round_up(code_size, 16);
lock_.lock();
if (active_chunk_) {
if (active_chunk_->capacity - active_chunk_->offset < code_size) {
auto next = active_chunk_->next;
if (!next) {
assert_true(code_size < chunk_size_, "need to support larger chunks");
next = new X64CodeChunk(chunk_size_);
active_chunk_->next = next;
}
active_chunk_ = next;
}
} else {
head_chunk_ = active_chunk_ = new X64CodeChunk(chunk_size_);
}
uint8_t* final_address = active_chunk_->buffer + active_chunk_->offset;
active_chunk_->offset += code_size;
lock_.unlock();
// Copy code.
memcpy(final_address, machine_code, code_size);
return final_address;
}
X64CodeChunk::X64CodeChunk(size_t chunk_size)
: next(NULL), capacity(chunk_size), buffer(0), offset(0) {
buffer = (uint8_t*)mmap(nullptr, chunk_size, PROT_WRITE | PROT_EXEC,
MAP_ANON | MAP_PRIVATE, -1, 0);
}
X64CodeChunk::~X64CodeChunk() {
if (buffer) {
munmap(buffer, capacity);
}
}
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe

281
src/xenia/cpu/backend/x64/x64_code_cache_win.cc Normal file
View File

@@ -0,0 +1,281 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/x64/x64_code_cache.h"
#include "poly/poly.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
class X64CodeChunk {
public:
X64CodeChunk(size_t chunk_size);
~X64CodeChunk();
public:
X64CodeChunk* next;
size_t capacity;
uint8_t* buffer;
size_t offset;
// Estimate of function sized use to determine initial table capacity.
const static uint32_t ESTIMATED_FN_SIZE = 512;
// Size of unwind info per function.
// TODO(benvanik): move this to emitter.
const static uint32_t UNWIND_INFO_SIZE = 4 + (2 * 1 + 2 + 2);
void* fn_table_handle;
RUNTIME_FUNCTION* fn_table;
uint32_t fn_table_count;
uint32_t fn_table_capacity;
void AddTableEntry(uint8_t* code, size_t code_size, size_t stack_size);
};
X64CodeCache::X64CodeCache(size_t chunk_size)
: chunk_size_(chunk_size), head_chunk_(NULL), active_chunk_(NULL) {}
X64CodeCache::~X64CodeCache() {
std::lock_guard<std::mutex> guard(lock_);
auto chunk = head_chunk_;
while (chunk) {
auto next = chunk->next;
delete chunk;
chunk = next;
}
head_chunk_ = NULL;
}
int X64CodeCache::Initialize() { return 0; }
void* X64CodeCache::PlaceCode(void* machine_code, size_t code_size,
size_t stack_size) {
size_t alloc_size = code_size;
// Add unwind info into the allocation size. Keep things 16b aligned.
alloc_size += poly::round_up(X64CodeChunk::UNWIND_INFO_SIZE, 16);
// Always move the code to land on 16b alignment. We do this by rounding up
// to 16b so that all offsets are aligned.
alloc_size = poly::round_up(alloc_size, 16);
lock_.lock();
if (active_chunk_) {
if (active_chunk_->capacity - active_chunk_->offset < alloc_size) {
auto next = active_chunk_->next;
if (!next) {
assert_true(alloc_size < chunk_size_, "need to support larger chunks");
next = new X64CodeChunk(chunk_size_);
active_chunk_->next = next;
}
active_chunk_ = next;
}
} else {
head_chunk_ = active_chunk_ = new X64CodeChunk(chunk_size_);
}
uint8_t* final_address = active_chunk_->buffer + active_chunk_->offset;
active_chunk_->offset += alloc_size;
// Add entry to fn table.
active_chunk_->AddTableEntry(final_address, alloc_size, stack_size);
lock_.unlock();
// Copy code.
memcpy(final_address, machine_code, code_size);
// This isn't needed on x64 (probably), but is convention.
FlushInstructionCache(GetCurrentProcess(), final_address, alloc_size);
return final_address;
}
X64CodeChunk::X64CodeChunk(size_t chunk_size)
: next(NULL), capacity(chunk_size), buffer(0), offset(0) {
buffer = (uint8_t*)VirtualAlloc(NULL, capacity, MEM_RESERVE | MEM_COMMIT,
PAGE_EXECUTE_READWRITE);
fn_table_capacity =
static_cast<uint32_t>(poly::round_up(capacity / ESTIMATED_FN_SIZE, 16));
size_t table_size = fn_table_capacity * sizeof(RUNTIME_FUNCTION);
fn_table = (RUNTIME_FUNCTION*)malloc(table_size);
fn_table_count = 0;
fn_table_handle = 0;
RtlAddGrowableFunctionTable(&fn_table_handle, fn_table, fn_table_count,
fn_table_capacity, (ULONG_PTR)buffer,
(ULONG_PTR)buffer + capacity);
}
X64CodeChunk::~X64CodeChunk() {
if (fn_table_handle) {
RtlDeleteGrowableFunctionTable(fn_table_handle);
}
if (buffer) {
VirtualFree(buffer, 0, MEM_RELEASE);
}
}
// http://msdn.microsoft.com/en-us/library/ssa62fwe.aspx
namespace {
typedef enum _UNWIND_OP_CODES {
UWOP_PUSH_NONVOL = 0, /* info == register number */
UWOP_ALLOC_LARGE, /* no info, alloc size in next 2 slots */
UWOP_ALLOC_SMALL, /* info == size of allocation / 8 - 1 */
UWOP_SET_FPREG, /* no info, FP = RSP + UNWIND_INFO.FPRegOffset*16 */
UWOP_SAVE_NONVOL, /* info == register number, offset in next slot */
UWOP_SAVE_NONVOL_FAR, /* info == register number, offset in next 2 slots */
UWOP_SAVE_XMM128, /* info == XMM reg number, offset in next slot */
UWOP_SAVE_XMM128_FAR, /* info == XMM reg number, offset in next 2 slots */
UWOP_PUSH_MACHFRAME /* info == 0: no error-code, 1: error-code */
} UNWIND_CODE_OPS;
class UNWIND_REGISTER {
public:
enum _ {
RAX = 0,
RCX = 1,
RDX = 2,
RBX = 3,
RSP = 4,
RBP = 5,
RSI = 6,
RDI = 7,
R8 = 8,
R9 = 9,
R10 = 10,
R11 = 11,
R12 = 12,
R13 = 13,
R14 = 14,
R15 = 15,
};
};
typedef union _UNWIND_CODE {
struct {
uint8_t CodeOffset;
uint8_t UnwindOp : 4;
uint8_t OpInfo : 4;
};
USHORT FrameOffset;
} UNWIND_CODE, *PUNWIND_CODE;
typedef struct _UNWIND_INFO {
uint8_t Version : 3;
uint8_t Flags : 5;
uint8_t SizeOfProlog;
uint8_t CountOfCodes;
uint8_t FrameRegister : 4;
uint8_t FrameOffset : 4;
UNWIND_CODE UnwindCode[1];
/* UNWIND_CODE MoreUnwindCode[((CountOfCodes + 1) & ~1) - 1];
* union {
* OPTIONAL ULONG ExceptionHandler;
* OPTIONAL ULONG FunctionEntry;
* };
* OPTIONAL ULONG ExceptionData[]; */
} UNWIND_INFO, *PUNWIND_INFO;
} // namespace
void X64CodeChunk::AddTableEntry(uint8_t* code, size_t code_size,
size_t stack_size) {
// NOTE: we assume a chunk lock.
if (fn_table_count + 1 > fn_table_capacity) {
// Table exhausted, need to realloc. If this happens a lot we should tune
// the table size to prevent this.
PLOGW("X64CodeCache growing FunctionTable - adjust ESTIMATED_FN_SIZE");
RtlDeleteGrowableFunctionTable(fn_table_handle);
size_t old_size = fn_table_capacity * sizeof(RUNTIME_FUNCTION);
size_t new_size = old_size * 2;
auto new_table = (RUNTIME_FUNCTION*)realloc(fn_table, new_size);
assert_not_null(new_table);
if (!new_table) {
return;
}
fn_table = new_table;
fn_table_capacity *= 2;
RtlAddGrowableFunctionTable(&fn_table_handle, fn_table, fn_table_count,
fn_table_capacity, (ULONG_PTR)buffer,
(ULONG_PTR)buffer + capacity);
}
// Allocate unwind data. We know we have space because we overallocated.
// This should be the tailing 16b with 16b alignment.
size_t unwind_info_offset = offset - UNWIND_INFO_SIZE;
if (!stack_size) {
// http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
UNWIND_INFO* unwind_info = (UNWIND_INFO*)(buffer + unwind_info_offset);
unwind_info->Version = 1;
unwind_info->Flags = 0;
unwind_info->SizeOfProlog = 0;
unwind_info->CountOfCodes = 0;
unwind_info->FrameRegister = 0;
unwind_info->FrameOffset = 0;
} else if (stack_size <= 128) {
uint8_t prolog_size = 4;
// http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
UNWIND_INFO* unwind_info = (UNWIND_INFO*)(buffer + unwind_info_offset);
unwind_info->Version = 1;
unwind_info->Flags = 0;
unwind_info->SizeOfProlog = prolog_size;
unwind_info->CountOfCodes = 1;
unwind_info->FrameRegister = 0;
unwind_info->FrameOffset = 0;
// http://msdn.microsoft.com/en-us/library/ck9asaa9.aspx
size_t co = 0;
auto& unwind_code = unwind_info->UnwindCode[co++];
unwind_code.CodeOffset =
14; // end of instruction + 1 == offset of next instruction
unwind_code.UnwindOp = UWOP_ALLOC_SMALL;
unwind_code.OpInfo = stack_size / 8 - 1;
} else {
// TODO(benvanik): take as parameters?
uint8_t prolog_size = 7;
// http://msdn.microsoft.com/en-us/library/ddssxxy8.aspx
UNWIND_INFO* unwind_info = (UNWIND_INFO*)(buffer + unwind_info_offset);
unwind_info->Version = 1;
unwind_info->Flags = 0;
unwind_info->SizeOfProlog = prolog_size;
unwind_info->CountOfCodes = 3;
unwind_info->FrameRegister = 0;
unwind_info->FrameOffset = 0;
// http://msdn.microsoft.com/en-us/library/ck9asaa9.aspx
size_t co = 0;
auto& unwind_code = unwind_info->UnwindCode[co++];
unwind_code.CodeOffset =
7; // end of instruction + 1 == offset of next instruction
unwind_code.UnwindOp = UWOP_ALLOC_LARGE;
unwind_code.OpInfo = 0;
unwind_code = unwind_info->UnwindCode[co++];
unwind_code.FrameOffset = (USHORT)(stack_size) / 8;
}
// Add entry.
auto& fn_entry = fn_table[fn_table_count++];
fn_entry.BeginAddress = (DWORD)(code - buffer);
fn_entry.EndAddress = (DWORD)(fn_entry.BeginAddress + code_size);
fn_entry.UnwindData = (DWORD)unwind_info_offset;
// Notify the function table that it has new entries.
RtlGrowFunctionTable(fn_table_handle, fn_table_count);
}
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe

942
src/xenia/cpu/backend/x64/x64_emitter.cc Normal file
View File

@@ -0,0 +1,942 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/x64/x64_emitter.h"
#include "xenia/cpu/backend/x64/x64_backend.h"
#include "xenia/cpu/backend/x64/x64_code_cache.h"
#include "xenia/cpu/backend/x64/x64_function.h"
#include "xenia/cpu/backend/x64/x64_sequences.h"
#include "xenia/cpu/backend/x64/x64_thunk_emitter.h"
#include "xenia/cpu/cpu-private.h"
#include "xenia/cpu/hir/hir_builder.h"
#include "xenia/cpu/runtime/debug_info.h"
#include "xenia/cpu/runtime/runtime.h"
#include "xenia/cpu/runtime/symbol_info.h"
#include "xenia/cpu/runtime/thread_state.h"
#include "poly/vec128.h"
#include "xdb/protocol.h"
#include "xenia/profiling.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
// TODO(benvanik): remove when enums redefined.
using namespace xe::cpu::hir;
using namespace xe::cpu::runtime;
using poly::vec128b;
using poly::vec128f;
using poly::vec128i;
using namespace Xbyak;
using xe::cpu::hir::HIRBuilder;
using xe::cpu::hir::Instr;
using xe::cpu::runtime::Function;
using xe::cpu::runtime::FunctionInfo;
using xe::cpu::runtime::SourceMapEntry;
using xe::cpu::runtime::ThreadState;
static const size_t MAX_CODE_SIZE = 1 * 1024 * 1024;
static const size_t STASH_OFFSET = 32;
static const size_t STASH_OFFSET_HIGH = 32 + 32;
// If we are running with tracing on we have to store the EFLAGS in the stack,
// otherwise our calls out to C to print will clear it before DID_CARRY/etc
// can get the value.
#define STORE_EFLAGS 1
const uint32_t X64Emitter::gpr_reg_map_[X64Emitter::GPR_COUNT] = {
Operand::RBX, Operand::R12, Operand::R13, Operand::R14, Operand::R15,
};
const uint32_t X64Emitter::xmm_reg_map_[X64Emitter::XMM_COUNT] = {
6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
};
X64Emitter::X64Emitter(X64Backend* backend, XbyakAllocator* allocator)
: CodeGenerator(MAX_CODE_SIZE, AutoGrow, allocator),
runtime_(backend->runtime()),
backend_(backend),
code_cache_(backend->code_cache()),
allocator_(allocator),
current_instr_(0) {}
X64Emitter::~X64Emitter() {}
int X64Emitter::Initialize() { return 0; }
int X64Emitter::Emit(HIRBuilder* builder, uint32_t debug_info_flags,
runtime::DebugInfo* debug_info, uint32_t trace_flags,
void*& out_code_address, size_t& out_code_size) {
SCOPE_profile_cpu_f("cpu");
// Reset.
if (debug_info_flags & DEBUG_INFO_SOURCE_MAP) {
source_map_count_ = 0;
source_map_arena_.Reset();
}
trace_flags_ = trace_flags;
// Fill the generator with code.
size_t stack_size = 0;
int result = Emit(builder, stack_size);
if (result) {
return result;
}
// Copy the final code to the cache and relocate it.
out_code_size = getSize();
out_code_address = Emplace(stack_size);
// Stash source map.
if (debug_info_flags & DEBUG_INFO_SOURCE_MAP) {
debug_info->InitializeSourceMap(
source_map_count_, (SourceMapEntry*)source_map_arena_.CloneContents());
}
return 0;
}
void* X64Emitter::Emplace(size_t stack_size) {
// To avoid changing xbyak, we do a switcharoo here.
// top_ points to the Xbyak buffer, and since we are in AutoGrow mode
// it has pending relocations. We copy the top_ to our buffer, swap the
// pointer, relocate, then return the original scratch pointer for use.
uint8_t* old_address = top_;
void* new_address = code_cache_->PlaceCode(top_, size_, stack_size);
top_ = (uint8_t*)new_address;
ready();
top_ = old_address;
reset();
return new_address;
}
int X64Emitter::Emit(HIRBuilder* builder, size_t& out_stack_size) {
// Calculate stack size. We need to align things to their natural sizes.
// This could be much better (sort by type/etc).
auto locals = builder->locals();
size_t stack_offset = StackLayout::GUEST_STACK_SIZE;
for (auto it = locals.begin(); it != locals.end(); ++it) {
auto slot = *it;
size_t type_size = GetTypeSize(slot->type);
// Align to natural size.
stack_offset = poly::align(stack_offset, type_size);
slot->set_constant((uint32_t)stack_offset);
stack_offset += type_size;
}
// Ensure 16b alignment.
stack_offset -= StackLayout::GUEST_STACK_SIZE;
stack_offset = poly::align(stack_offset, static_cast<size_t>(16));
// Function prolog.
// Must be 16b aligned.
// Windows is very strict about the form of this and the epilog:
// http://msdn.microsoft.com/en-us/library/tawsa7cb.aspx
// TODO(benvanik): save off non-volatile registers so we can use them:
// RBX, RBP, RDI, RSI, RSP, R12, R13, R14, R15
// Only want to do this if we actually use them, though, otherwise
// it just adds overhead.
// IMPORTANT: any changes to the prolog must be kept in sync with
// X64CodeCache, which dynamically generates exception information.
// Adding or changing anything here must be matched!
const bool emit_prolog = true;
const size_t stack_size = StackLayout::GUEST_STACK_SIZE + stack_offset;
assert_true((stack_size + 8) % 16 == 0);
out_stack_size = stack_size;
stack_size_ = stack_size;
if (emit_prolog) {
sub(rsp, (uint32_t)stack_size);
mov(qword[rsp + StackLayout::GUEST_RCX_HOME], rcx);
mov(qword[rsp + StackLayout::GUEST_RET_ADDR], rdx);
mov(qword[rsp + StackLayout::GUEST_CALL_RET_ADDR], 0);
mov(rdx, qword[rcx + 8]); // membase
}
uint64_t trace_base = runtime_->memory()->trace_base();
if (trace_base && trace_flags_ & TRACE_USER_CALLS) {
mov(rax, trace_base);
mov(r8d, static_cast<uint32_t>(sizeof(xdb::protocol::UserCallEvent)));
lock();
xadd(qword[rax], r8);
mov(rax, static_cast<uint64_t>(xdb::protocol::EventType::USER_CALL) |
(static_cast<uint64_t>(0) << 8) | (0ull << 32));
mov(qword[r8], rax);
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
}
// Body.
auto block = builder->first_block();
while (block) {
// Mark block labels.
auto label = block->label_head;
while (label) {
L(label->name);
label = label->next;
}
// Process instructions.
const Instr* instr = block->instr_head;
while (instr) {
const Instr* new_tail = instr;
// Special handling of TRACE_SOURCE.
if (instr->opcode == &OPCODE_TRACE_SOURCE_info) {
if (trace_flags_ & TRACE_SOURCE) {
EmitTraceSource(instr);
}
instr = instr->next;
continue;
}
if (!SelectSequence(*this, instr, &new_tail)) {
// No sequence found!
assert_always();
PLOGE("Unable to process HIR opcode %s", instr->opcode->name);
break;
}
instr = new_tail;
}
block = block->next;
}
// Function epilog.
L("epilog");
EmitTraceUserCallReturn();
if (emit_prolog) {
mov(rcx, qword[rsp + StackLayout::GUEST_RCX_HOME]);
add(rsp, (uint32_t)stack_size);
}
ret();
#if XE_DEBUG
nop();
nop();
nop();
nop();
nop();
#endif // XE_DEBUG
return 0;
}
void X64Emitter::MarkSourceOffset(const Instr* i) {
auto entry = source_map_arena_.Alloc<SourceMapEntry>();
entry->source_offset = i->src1.offset;
entry->hir_offset = uint32_t(i->block->ordinal << 16) | i->ordinal;
entry->code_offset = getSize();
source_map_count_++;
}
void X64Emitter::EmitTraceSource(const Instr* instr) {
uint64_t trace_base = runtime_->memory()->trace_base();
if (!trace_base || !(trace_flags_ & TRACE_SOURCE)) {
return;
}
// TODO(benvanik): make this a function call to some append fn.
uint8_t dest_reg_0 = instr->flags & 0xFF;
uint8_t dest_reg_1 = instr->flags >> 8;
xdb::protocol::EventType event_type;
size_t event_size = 0;
if (dest_reg_0 == 100) {
event_type = xdb::protocol::EventType::INSTR;
event_size = sizeof(xdb::protocol::InstrEvent);
} else if (dest_reg_1 == 100) {
if (dest_reg_0 & (1 << 7)) {
event_type = xdb::protocol::EventType::INSTR_R16;
event_size = sizeof(xdb::protocol::InstrEventR16);
} else {
event_type = xdb::protocol::EventType::INSTR_R8;
event_size = sizeof(xdb::protocol::InstrEventR8);
}
} else {
if (dest_reg_0 & (1 << 7) && dest_reg_1 & (1 << 7)) {
event_type = xdb::protocol::EventType::INSTR_R16_R16;
event_size = sizeof(xdb::protocol::InstrEventR16R16);
} else if (dest_reg_0 & (1 << 7) && !(dest_reg_1 & (1 << 7))) {
event_type = xdb::protocol::EventType::INSTR_R16_R8;
event_size = sizeof(xdb::protocol::InstrEventR16R8);
} else if (!(dest_reg_0 & (1 << 7)) && dest_reg_1 & (1 << 7)) {
event_type = xdb::protocol::EventType::INSTR_R8_R16;
event_size = sizeof(xdb::protocol::InstrEventR8R16);
} else if (!(dest_reg_0 & (1 << 7)) && !(dest_reg_1 & (1 << 7))) {
event_type = xdb::protocol::EventType::INSTR_R8_R8;
event_size = sizeof(xdb::protocol::InstrEventR8R8);
}
}
assert_not_zero(event_size);
mov(rax, trace_base);
mov(r8d, static_cast<uint32_t>(event_size));
lock();
xadd(qword[rax], r8);
// r8 is now the pointer where we can write our event.
// Write the header, which is the same for everything (pretty much).
// Some event types ignore the dest reg, and that's fine.
uint64_t qword_0 = static_cast<uint64_t>(event_type) |
(static_cast<uint64_t>(dest_reg_0) << 8) |
(instr->src1.offset << 32);
mov(rax, qword_0);
mov(qword[r8], rax);
// Write thread ID.
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
switch (event_type) {
default:
case xdb::protocol::EventType::INSTR:
break;
case xdb::protocol::EventType::INSTR_R8:
case xdb::protocol::EventType::INSTR_R16:
if (dest_reg_0 & (1 << 7)) {
EmitTraceSourceAppendValue(instr->src2.value, 8);
} else {
EmitTraceSourceAppendValue(instr->src2.value, 8);
}
break;
case xdb::protocol::EventType::INSTR_R8_R8:
case xdb::protocol::EventType::INSTR_R8_R16:
case xdb::protocol::EventType::INSTR_R16_R8:
case xdb::protocol::EventType::INSTR_R16_R16:
mov(word[r8 + 8], dest_reg_0 | static_cast<uint16_t>(dest_reg_1 << 8));
size_t offset = 8;
if (dest_reg_0 & (1 << 7)) {
EmitTraceSourceAppendValue(instr->src2.value, offset);
offset += 16;
} else {
EmitTraceSourceAppendValue(instr->src2.value, offset);
offset += 8;
}
if (dest_reg_1 & (1 << 7)) {
EmitTraceSourceAppendValue(instr->src3.value, offset);
offset += 16;
} else {
EmitTraceSourceAppendValue(instr->src3.value, offset);
offset += 8;
}
break;
}
}
void X64Emitter::EmitTraceSourceAppendValue(const Value* value,
size_t r8_offset) {
//
}
void X64Emitter::EmitGetCurrentThreadId() {
// rcx must point to context. We could fetch from the stack if needed.
mov(ax, word[rcx + runtime_->frontend()->context_info()->thread_id_offset()]);
}
void X64Emitter::EmitTraceUserCallReturn() {
auto trace_base = runtime_->memory()->trace_base();
if (!trace_base || !(trace_flags_ & TRACE_USER_CALLS)) {
return;
}
mov(rdx, rax);
mov(rax, trace_base);
mov(r8d, static_cast<uint32_t>(sizeof(xdb::protocol::UserCallReturnEvent)));
lock();
xadd(qword[rax], r8);
mov(rax, static_cast<uint64_t>(xdb::protocol::EventType::USER_CALL_RETURN) |
(static_cast<uint64_t>(0) << 8) | (0ull << 32));
mov(qword[r8], rax);
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
mov(rax, rdx);
ReloadEDX();
}
void X64Emitter::DebugBreak() {
// TODO(benvanik): notify debugger.
db(0xCC);
}
void X64Emitter::Trap(uint16_t trap_type) {
switch (trap_type) {
case 20:
// 0x0FE00014 is a 'debug print' where r3 = buffer r4 = length
// TODO(benvanik): debug print at runtime.
break;
case 0:
case 22:
// Always trap?
// TODO(benvanik): post software interrupt to debugger.
if (FLAGS_break_on_debugbreak) {
db(0xCC);
}
break;
default:
PLOGW("Unknown trap type %d", trap_type);
db(0xCC);
break;
}
}
void X64Emitter::UnimplementedInstr(const hir::Instr* i) {
// TODO(benvanik): notify debugger.
db(0xCC);
assert_always();
}
// Total size of ResolveFunctionSymbol call site in bytes.
// Used to overwrite it with nops as needed.
const size_t TOTAL_RESOLVE_SIZE = 27;
const size_t ASM_OFFSET = 2 + 2 + 8 + 2 + 8;
uint64_t ResolveFunctionSymbol(void* raw_context, uint64_t symbol_info_ptr) {
// TODO(benvanik): generate this thunk at runtime? or a shim?
auto thread_state = *reinterpret_cast<ThreadState**>(raw_context);
auto symbol_info = reinterpret_cast<FunctionInfo*>(symbol_info_ptr);
// Resolve function. This will demand compile as required.
Function* fn = NULL;
thread_state->runtime()->ResolveFunction(symbol_info->address(), &fn);
assert_not_null(fn);
auto x64_fn = static_cast<X64Function*>(fn);
uint64_t addr = reinterpret_cast<uint64_t>(x64_fn->machine_code());
// Overwrite the call site.
// The return address points to ReloadRCX work after the call.
#if XE_LIKE_WIN32
uint64_t return_address = reinterpret_cast<uint64_t>(_ReturnAddress());
#else
uint64_t return_address =
reinterpret_cast<uint64_t>(__builtin_return_address(0));
#endif // XE_LIKE_WIN32
#pragma pack(push, 1)
struct Asm {
uint16_t mov_rax;
uint64_t rax_constant;
uint16_t mov_rdx;
uint64_t rdx_constant;
uint16_t call_rax;
uint8_t mov_rcx[5];
};
#pragma pack(pop)
static_assert_size(Asm, TOTAL_RESOLVE_SIZE);
Asm* code = reinterpret_cast<Asm*>(return_address - ASM_OFFSET);
code->rax_constant = addr;
code->call_rax = 0x9066;
// We need to return the target in rax so that it gets called.
return addr;
}
void X64Emitter::Call(const hir::Instr* instr,
runtime::FunctionInfo* symbol_info) {
auto fn = reinterpret_cast<X64Function*>(symbol_info->function());
// Resolve address to the function to call and store in rax.
if (fn) {
mov(rax, reinterpret_cast<uint64_t>(fn->machine_code()));
} else {
size_t start = getSize();
// 2b + 8b constant
mov(rax, reinterpret_cast<uint64_t>(ResolveFunctionSymbol));
// 2b + 8b constant
mov(rdx, reinterpret_cast<uint64_t>(symbol_info));
// 2b
call(rax);
// 5b
ReloadECX();
size_t total_size = getSize() - start;
assert_true(total_size == TOTAL_RESOLVE_SIZE);
// EDX overwritten, don't bother reloading.
}
// Actually jump/call to rax.
if (instr->flags & CALL_TAIL) {
// Since we skip the prolog we need to mark the return here.
EmitTraceUserCallReturn();
// Pass the callers return address over.
mov(rdx, qword[rsp + StackLayout::GUEST_RET_ADDR]);
add(rsp, static_cast<uint32_t>(stack_size()));
jmp(rax);
} else {
// Return address is from the previous SET_RETURN_ADDRESS.
mov(rdx, qword[rsp + StackLayout::GUEST_CALL_RET_ADDR]);
call(rax);
}
}
// NOTE: slot count limited by short jump size.
const int kICSlotCount = 4;
const int kICSlotSize = 23;
const uint64_t kICSlotInvalidTargetAddress = 0x0F0F0F0F0F0F0F0F;
uint64_t ResolveFunctionAddress(void* raw_context, uint64_t target_address) {
// TODO(benvanik): generate this thunk at runtime? or a shim?
auto thread_state = *reinterpret_cast<ThreadState**>(raw_context);
// TODO(benvanik): required?
target_address &= 0xFFFFFFFF;
assert_not_zero(target_address);
Function* fn = NULL;
thread_state->runtime()->ResolveFunction(target_address, &fn);
assert_not_null(fn);
auto x64_fn = static_cast<X64Function*>(fn);
uint64_t addr = reinterpret_cast<uint64_t>(x64_fn->machine_code());
// Add an IC slot, if there is room.
#if XE_LIKE_WIN32
uint64_t return_address = reinterpret_cast<uint64_t>(_ReturnAddress());
#else
uint64_t return_address =
reinterpret_cast<uint64_t>(__builtin_return_address(0));
#endif // XE_LIKE_WIN32
#pragma pack(push, 1)
struct Asm {
uint16_t cmp_rdx;
uint32_t address_constant;
uint16_t jmp_next_slot;
uint16_t mov_rax;
uint64_t target_constant;
uint8_t jmp_skip_resolve[5];
};
#pragma pack(pop)
static_assert_size(Asm, kICSlotSize);
// TODO(benvanik): quick check table is full (so we don't have to enum slots)
// The return address points to ReloadRCX work after the call.
// To get the top of the table, look back a ways.
uint64_t table_start = return_address - 12 - kICSlotSize * kICSlotCount;
// NOTE: order matters here - we update the address BEFORE we switch the code
// over to passing the compare.
Asm* table_slot = reinterpret_cast<Asm*>(table_start);
bool wrote_ic = false;
for (int i = 0; i < kICSlotCount; ++i) {
if (poly::atomic_cas(kICSlotInvalidTargetAddress, addr,
&table_slot->target_constant)) {
// Got slot! Just write the compare and we're done.
table_slot->address_constant = static_cast<uint32_t>(target_address);
wrote_ic = true;
break;
}
++table_slot;
}
if (!wrote_ic) {
// TODO(benvanik): log that IC table is full.
}
// We need to return the target in rax so that it gets called.
return addr;
}
void X64Emitter::CallIndirect(const hir::Instr* instr, const Reg64& reg) {
// Check if return.
if (instr->flags & CALL_POSSIBLE_RETURN) {
cmp(reg.cvt32(), dword[rsp + StackLayout::GUEST_RET_ADDR]);
je("epilog", CodeGenerator::T_NEAR);
}
if (reg.getIdx() != rdx.getIdx()) {
mov(rdx, reg);
}
inLocalLabel();
Xbyak::Label skip_resolve;
// TODO(benvanik): make empty tables skippable (cmp, jump right to resolve).
// IC table, initially empty.
// This will get filled in as functions are resolved.
// Note that we only have a limited cache, and once it's full all calls
// will fall through.
// TODO(benvanik): check miss rate when full and add a 2nd-level table?
// 0000000264BD4DC3 81 FA 0F0F0F0F cmp edx,0F0F0F0Fh
// 0000000264BD4DC9 75 0C jne 0000000264BD4DD7
// 0000000264BD4DCB 48 B8 0F0F0F0F0F0F0F0F mov rax,0F0F0F0F0F0F0F0Fh
// 0000000264BD4DD5 EB XXXXXXXX jmp 0000000264BD4E00
size_t table_start = getSize();
for (int i = 0; i < kICSlotCount; ++i) {
// Compare target address with constant, if matches jump there.
// Otherwise, fall through.
// 6b
cmp(edx, 0x0F0F0F0F);
Xbyak::Label next_slot;
// 2b
jne(next_slot, T_SHORT);
// Match! Load up rax and skip down to the jmp code.
// 10b
mov(rax, kICSlotInvalidTargetAddress);
// 5b
jmp(skip_resolve, T_NEAR);
L(next_slot);
}
size_t table_size = getSize() - table_start;
assert_true(table_size == kICSlotSize * kICSlotCount);
// Resolve address to the function to call and store in rax.
// We fall through to this when there are no hits in the IC table.
CallNative(ResolveFunctionAddress);
// Actually jump/call to rax.
L(skip_resolve);
if (instr->flags & CALL_TAIL) {
// Since we skip the prolog we need to mark the return here.
EmitTraceUserCallReturn();
// Pass the callers return address over.
mov(rdx, qword[rsp + StackLayout::GUEST_RET_ADDR]);
add(rsp, static_cast<uint32_t>(stack_size()));
jmp(rax);
} else {
// Return address is from the previous SET_RETURN_ADDRESS.
mov(rdx, qword[rsp + StackLayout::GUEST_CALL_RET_ADDR]);
call(rax);
}
outLocalLabel();
}
uint64_t UndefinedCallExtern(void* raw_context, uint64_t symbol_info_ptr) {
auto symbol_info = reinterpret_cast<FunctionInfo*>(symbol_info_ptr);
PLOGW("undefined extern call to %.8llX %s", symbol_info->address(),
symbol_info->name().c_str());
return 0;
}
void X64Emitter::CallExtern(const hir::Instr* instr,
const FunctionInfo* symbol_info) {
assert_true(symbol_info->behavior() == FunctionInfo::BEHAVIOR_EXTERN);
uint64_t trace_base = runtime_->memory()->trace_base();
if (trace_base & trace_flags_ & TRACE_EXTERN_CALLS) {
mov(rax, trace_base);
mov(r8d, static_cast<uint32_t>(sizeof(xdb::protocol::KernelCallEvent)));
lock();
xadd(qword[rax], r8);
// TODO(benvanik): get module/ordinal.
uint32_t module_id = 0;
uint32_t ordinal = 0;
mov(rax, static_cast<uint64_t>(xdb::protocol::EventType::KERNEL_CALL) |
(static_cast<uint64_t>(0) << 8) | (module_id << 16) |
(ordinal));
mov(qword[r8], rax);
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
}
if (!symbol_info->extern_handler()) {
CallNative(UndefinedCallExtern, reinterpret_cast<uint64_t>(symbol_info));
} else {
// rcx = context
// rdx = target host function
// r8 = arg0
// r9 = arg1
mov(rdx, reinterpret_cast<uint64_t>(symbol_info->extern_handler()));
mov(r8, reinterpret_cast<uint64_t>(symbol_info->extern_arg0()));
mov(r9, reinterpret_cast<uint64_t>(symbol_info->extern_arg1()));
auto thunk = backend()->guest_to_host_thunk();
mov(rax, reinterpret_cast<uint64_t>(thunk));
call(rax);
ReloadECX();
ReloadEDX();
// rax = host return
}
if (trace_base && trace_flags_ & TRACE_EXTERN_CALLS) {
mov(rax, trace_base);
mov(r8d,
static_cast<uint32_t>(sizeof(xdb::protocol::KernelCallReturnEvent)));
lock();
xadd(qword[rax], r8);
mov(rax,
static_cast<uint64_t>(xdb::protocol::EventType::KERNEL_CALL_RETURN) |
(static_cast<uint64_t>(0) << 8) | (0));
mov(qword[r8], rax);
EmitGetCurrentThreadId();
mov(word[r8 + 2], ax);
}
}
void X64Emitter::CallNative(void* fn) {
mov(rax, reinterpret_cast<uint64_t>(fn));
call(rax);
ReloadECX();
ReloadEDX();
}
void X64Emitter::CallNative(uint64_t (*fn)(void* raw_context)) {
mov(rax, reinterpret_cast<uint64_t>(fn));
call(rax);
ReloadECX();
ReloadEDX();
}
void X64Emitter::CallNative(uint64_t (*fn)(void* raw_context, uint64_t arg0)) {
mov(rax, reinterpret_cast<uint64_t>(fn));
call(rax);
ReloadECX();
ReloadEDX();
}
void X64Emitter::CallNative(uint64_t (*fn)(void* raw_context, uint64_t arg0),
uint64_t arg0) {
mov(rdx, arg0);
mov(rax, reinterpret_cast<uint64_t>(fn));
call(rax);
ReloadECX();
ReloadEDX();
}
void X64Emitter::CallNativeSafe(void* fn) {
// rcx = context
// rdx = target host function
// r8 = arg0
// r9 = arg1
mov(rdx, reinterpret_cast<uint64_t>(fn));
auto thunk = backend()->guest_to_host_thunk();
mov(rax, reinterpret_cast<uint64_t>(thunk));
call(rax);
ReloadECX();
ReloadEDX();
// rax = host return
}
void X64Emitter::SetReturnAddress(uint64_t value) {
mov(qword[rsp + StackLayout::GUEST_CALL_RET_ADDR], value);
}
void X64Emitter::ReloadECX() {
mov(rcx, qword[rsp + StackLayout::GUEST_RCX_HOME]);
}
void X64Emitter::ReloadEDX() {
mov(rdx, qword[rcx + 8]); // membase
}
// Len Assembly Byte Sequence
// ============================================================================
// 2b 66 NOP 66 90H
// 3b NOP DWORD ptr [EAX] 0F 1F 00H
// 4b NOP DWORD ptr [EAX + 00H] 0F 1F 40 00H
// 5b NOP DWORD ptr [EAX + EAX*1 + 00H] 0F 1F 44 00 00H
// 6b 66 NOP DWORD ptr [EAX + EAX*1 + 00H] 66 0F 1F 44 00 00H
// 7b NOP DWORD ptr [EAX + 00000000H] 0F 1F 80 00 00 00 00H
// 8b NOP DWORD ptr [EAX + EAX*1 + 00000000H] 0F 1F 84 00 00 00 00 00H
// 9b 66 NOP DWORD ptr [EAX + EAX*1 + 00000000H] 66 0F 1F 84 00 00 00 00 00H
void X64Emitter::nop(size_t length) {
// TODO(benvanik): fat nop
for (size_t i = 0; i < length; ++i) {
db(0x90);
}
}
void X64Emitter::LoadEflags() {
#if STORE_EFLAGS
mov(eax, dword[rsp + STASH_OFFSET]);
btr(eax, 0);
#else
// EFLAGS already present.
#endif // STORE_EFLAGS
}
void X64Emitter::StoreEflags() {
#if STORE_EFLAGS
pushf();
pop(dword[rsp + STASH_OFFSET]);
#else
// EFLAGS should have CA set?
// (so long as we don't fuck with it)
#endif // STORE_EFLAGS
}
bool X64Emitter::ConstantFitsIn32Reg(uint64_t v) {
if ((v & ~0x7FFFFFFF) == 0) {
// Fits under 31 bits, so just load using normal mov.
return true;
} else if ((v & ~0x7FFFFFFF) == ~0x7FFFFFFF) {
// Negative number that fits in 32bits.
return true;
}
return false;
}
void X64Emitter::MovMem64(const RegExp& addr, uint64_t v) {
if ((v & ~0x7FFFFFFF) == 0) {
// Fits under 31 bits, so just load using normal mov.
mov(qword[addr], v);
} else if ((v & ~0x7FFFFFFF) == ~0x7FFFFFFF) {
// Negative number that fits in 32bits.
mov(qword[addr], v);
} else if (!(v >> 32)) {
// All high bits are zero. It'd be nice if we had a way to load a 32bit
// immediate without sign extending!
// TODO(benvanik): this is super common, find a better way.
mov(dword[addr], static_cast<uint32_t>(v));
mov(dword[addr + 4], 0);
} else {
// 64bit number that needs double movs.
mov(dword[addr], static_cast<uint32_t>(v));
mov(dword[addr + 4], static_cast<uint32_t>(v >> 32));
}
}
Address X64Emitter::GetXmmConstPtr(XmmConst id) {
static const vec128_t xmm_consts[] = {
/* XMMZero */ vec128f(0.0f),
/* XMMOne */ vec128f(1.0f),
/* XMMNegativeOne */ vec128f(-1.0f, -1.0f, -1.0f, -1.0f),
/* XMMFFFF */ vec128i(0xFFFFFFFFu, 0xFFFFFFFFu,
0xFFFFFFFFu, 0xFFFFFFFFu),
/* XMMMaskX16Y16 */ vec128i(0x0000FFFFu, 0xFFFF0000u,
0x00000000u, 0x00000000u),
/* XMMFlipX16Y16 */ vec128i(0x00008000u, 0x00000000u,
0x00000000u, 0x00000000u),
/* XMMFixX16Y16 */ vec128f(-32768.0f, 0.0f, 0.0f, 0.0f),
/* XMMNormalizeX16Y16 */ vec128f(
1.0f / 32767.0f, 1.0f / (32767.0f * 65536.0f), 0.0f, 0.0f),
/* XMM0001 */ vec128f(0.0f, 0.0f, 0.0f, 1.0f),
/* XMM3301 */ vec128f(3.0f, 3.0f, 0.0f, 1.0f),
/* XMM3333 */ vec128f(3.0f, 3.0f, 3.0f, 3.0f),
/* XMMSignMaskPS */ vec128i(0x80000000u, 0x80000000u,
0x80000000u, 0x80000000u),
/* XMMSignMaskPD */ vec128i(0x00000000u, 0x80000000u,
0x00000000u, 0x80000000u),
/* XMMAbsMaskPS */ vec128i(0x7FFFFFFFu, 0x7FFFFFFFu,
0x7FFFFFFFu, 0x7FFFFFFFu),
/* XMMAbsMaskPD */ vec128i(0xFFFFFFFFu, 0x7FFFFFFFu,
0xFFFFFFFFu, 0x7FFFFFFFu),
/* XMMByteSwapMask */ vec128i(0x00010203u, 0x04050607u,
0x08090A0Bu, 0x0C0D0E0Fu),
/* XMMByteOrderMask */ vec128i(0x01000302u, 0x05040706u,
0x09080B0Au, 0x0D0C0F0Eu),
/* XMMPermuteControl15 */ vec128b(15),
/* XMMPermuteByteMask */ vec128b(0x1F),
/* XMMPackD3DCOLORSat */ vec128i(0x404000FFu),
/* XMMPackD3DCOLOR */ vec128i(0xFFFFFFFFu, 0xFFFFFFFFu,
0xFFFFFFFFu, 0x0C000408u),
/* XMMUnpackD3DCOLOR */ vec128i(0xFFFFFF0Eu, 0xFFFFFF0Du,
0xFFFFFF0Cu, 0xFFFFFF0Fu),
/* XMMPackFLOAT16_2 */ vec128i(0xFFFFFFFFu, 0xFFFFFFFFu,
0xFFFFFFFFu, 0x01000302u),
/* XMMUnpackFLOAT16_2 */ vec128i(0x0D0C0F0Eu, 0xFFFFFFFFu,
0xFFFFFFFFu, 0xFFFFFFFFu),
/* XMMPackFLOAT16_4 */ vec128i(0xFFFFFFFFu, 0xFFFFFFFFu,
0x05040706u, 0x01000302u),
/* XMMUnpackFLOAT16_4 */ vec128i(0x09080B0Au, 0x0D0C0F0Eu,
0xFFFFFFFFu, 0xFFFFFFFFu),
/* XMMPackSHORT_2Min */ vec128i(0x403F8001u),
/* XMMPackSHORT_2Max */ vec128i(0x40407FFFu),
/* XMMPackSHORT_2 */ vec128i(0xFFFFFFFFu, 0xFFFFFFFFu,
0xFFFFFFFFu, 0x01000504u),
/* XMMUnpackSHORT_2 */ vec128i(0xFFFF0F0Eu, 0xFFFF0D0Cu,
0xFFFFFFFFu, 0xFFFFFFFFu),
/* XMMOneOver255 */ vec128f(1.0f / 255.0f),
/* XMMMaskEvenPI16 */ vec128i(0x0000FFFFu, 0x0000FFFFu,
0x0000FFFFu, 0x0000FFFFu),
/* XMMShiftMaskEvenPI16 */ vec128i(0x0000000Fu, 0x0000000Fu,
0x0000000Fu, 0x0000000Fu),
/* XMMShiftMaskPS */ vec128i(0x0000001Fu, 0x0000001Fu,
0x0000001Fu, 0x0000001Fu),
/* XMMShiftByteMask */ vec128i(0x000000FFu, 0x000000FFu,
0x000000FFu, 0x000000FFu),
/* XMMSwapWordMask */ vec128i(0x03030303u, 0x03030303u,
0x03030303u, 0x03030303u),
/* XMMUnsignedDwordMax */ vec128i(0xFFFFFFFFu, 0x00000000u,
0xFFFFFFFFu, 0x00000000u),
/* XMM255 */ vec128f(255.0f),
/* XMMPI32 */ vec128i(32),
/* XMMSignMaskI8 */ vec128i(0x80808080u, 0x80808080u,
0x80808080u, 0x80808080u),
/* XMMSignMaskI16 */ vec128i(0x80008000u, 0x80008000u,
0x80008000u, 0x80008000u),
/* XMMSignMaskI32 */ vec128i(0x80000000u, 0x80000000u,
0x80000000u, 0x80000000u),
/* XMMSignMaskF32 */ vec128i(0x80000000u, 0x80000000u,
0x80000000u, 0x80000000u),
/* XMMShortMinPS */ vec128f(SHRT_MIN),
/* XMMShortMaxPS */ vec128f(SHRT_MAX),
};
// TODO(benvanik): cache base pointer somewhere? stack? It'd be nice to
// prevent this move.
// TODO(benvanik): move to predictable location in PPCContext? could then
// just do rcx relative addression with no rax overwriting.
mov(rax, (uint64_t)&xmm_consts[id]);
return ptr[rax];
}
void X64Emitter::LoadConstantXmm(Xbyak::Xmm dest, const vec128_t& v) {
// http://www.agner.org/optimize/optimizing_assembly.pdf
// 13.4 Generating constants
if (!v.low && !v.high) {
// 0000...
vpxor(dest, dest);
} else if (v.low == ~0ull && v.high == ~0ull) {
// 1111...
vpcmpeqb(dest, dest);
} else {
// TODO(benvanik): see what other common values are.
// TODO(benvanik): build constant table - 99% are reused.
MovMem64(rsp + STASH_OFFSET, v.low);
MovMem64(rsp + STASH_OFFSET + 8, v.high);
vmovdqa(dest, ptr[rsp + STASH_OFFSET]);
}
}
void X64Emitter::LoadConstantXmm(Xbyak::Xmm dest, float v) {
union {
float f;
uint32_t i;
} x = {v};
if (!v) {
// 0
vpxor(dest, dest);
} else if (x.i == ~0U) {
// 1111...
vpcmpeqb(dest, dest);
} else {
// TODO(benvanik): see what other common values are.
// TODO(benvanik): build constant table - 99% are reused.
mov(eax, x.i);
vmovd(dest, eax);
}
}
void X64Emitter::LoadConstantXmm(Xbyak::Xmm dest, double v) {
union {
double d;
uint64_t i;
} x = {v};
if (!v) {
// 0
vpxor(dest, dest);
} else if (x.i == ~0ULL) {
// 1111...
vpcmpeqb(dest, dest);
} else {
// TODO(benvanik): see what other common values are.
// TODO(benvanik): build constant table - 99% are reused.
mov(rax, x.i);
vmovq(dest, rax);
}
}
Address X64Emitter::StashXmm(int index, const Xmm& r) {
auto addr = ptr[rsp + STASH_OFFSET + (index * 16)];
vmovups(addr, r);
return addr;
}
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe

217
src/xenia/cpu/backend/x64/x64_emitter.h Normal file
View File

@@ -0,0 +1,217 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_X64_X64_EMITTER_H_
#define XENIA_BACKEND_X64_X64_EMITTER_H_
#include "xenia/cpu/hir/value.h"
#include "poly/arena.h"
#include "third_party/xbyak/xbyak/xbyak.h"
namespace xe {
namespace cpu {
namespace hir {
class HIRBuilder;
class Instr;
} // namespace hir
namespace runtime {
class DebugInfo;
class FunctionInfo;
class Runtime;
class SymbolInfo;
} // namespace runtime
} // namespace cpu
} // namespace xe
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
using vec128_t = poly::vec128_t;
class X64Backend;
class X64CodeCache;
enum RegisterFlags {
REG_DEST = (1 << 0),
REG_ABCD = (1 << 1),
};
enum XmmConst {
XMMZero = 0,
XMMOne,
XMMNegativeOne,
XMMFFFF,
XMMMaskX16Y16,
XMMFlipX16Y16,
XMMFixX16Y16,
XMMNormalizeX16Y16,
XMM0001,
XMM3301,
XMM3333,
XMMSignMaskPS,
XMMSignMaskPD,
XMMAbsMaskPS,
XMMAbsMaskPD,
XMMByteSwapMask,
XMMByteOrderMask,
XMMPermuteControl15,
XMMPermuteByteMask,
XMMPackD3DCOLORSat,
XMMPackD3DCOLOR,
XMMUnpackD3DCOLOR,
XMMPackFLOAT16_2,
XMMUnpackFLOAT16_2,
XMMPackFLOAT16_4,
XMMUnpackFLOAT16_4,
XMMPackSHORT_2Min,
XMMPackSHORT_2Max,
XMMPackSHORT_2,
XMMUnpackSHORT_2,
XMMOneOver255,
XMMMaskEvenPI16,
XMMShiftMaskEvenPI16,
XMMShiftMaskPS,
XMMShiftByteMask,
XMMSwapWordMask,
XMMUnsignedDwordMax,
XMM255,
XMMPI32,
XMMSignMaskI8,
XMMSignMaskI16,
XMMSignMaskI32,
XMMSignMaskF32,
XMMShortMinPS,
XMMShortMaxPS,
};
// Unfortunately due to the design of xbyak we have to pass this to the ctor.
class XbyakAllocator : public Xbyak::Allocator {
public:
virtual bool useProtect() const { return false; }
};
class X64Emitter : public Xbyak::CodeGenerator {
public:
X64Emitter(X64Backend* backend, XbyakAllocator* allocator);
virtual ~X64Emitter();
runtime::Runtime* runtime() const { return runtime_; }
X64Backend* backend() const { return backend_; }
int Initialize();
int Emit(hir::HIRBuilder* builder, uint32_t debug_info_flags,
runtime::DebugInfo* debug_info, uint32_t trace_flags,
void*& out_code_address, size_t& out_code_size);
public:
// Reserved: rsp
// Scratch: rax/rcx/rdx
// xmm0-2 (could be only xmm0 with some trickery)
// Available: rbx, r12-r15 (save to get r8-r11, rbp, rsi, rdi?)
// xmm6-xmm15 (save to get xmm3-xmm5)
static const int GPR_COUNT = 5;
static const int XMM_COUNT = 10;
static void SetupReg(const hir::Value* v, Xbyak::Reg8& r) {
auto idx = gpr_reg_map_[v->reg.index];
r = Xbyak::Reg8(idx);
}
static void SetupReg(const hir::Value* v, Xbyak::Reg16& r) {
auto idx = gpr_reg_map_[v->reg.index];
r = Xbyak::Reg16(idx);
}
static void SetupReg(const hir::Value* v, Xbyak::Reg32& r) {
auto idx = gpr_reg_map_[v->reg.index];
r = Xbyak::Reg32(idx);
}
static void SetupReg(const hir::Value* v, Xbyak::Reg64& r) {
auto idx = gpr_reg_map_[v->reg.index];
r = Xbyak::Reg64(idx);
}
static void SetupReg(const hir::Value* v, Xbyak::Xmm& r) {
auto idx = xmm_reg_map_[v->reg.index];
r = Xbyak::Xmm(idx);
}
void MarkSourceOffset(const hir::Instr* i);
void DebugBreak();
void Trap(uint16_t trap_type = 0);
void UnimplementedInstr(const hir::Instr* i);
void UnimplementedExtern(const hir::Instr* i);
void Call(const hir::Instr* instr, runtime::FunctionInfo* symbol_info);
void CallIndirect(const hir::Instr* instr, const Xbyak::Reg64& reg);
void CallExtern(const hir::Instr* instr,
const runtime::FunctionInfo* symbol_info);
void CallNative(void* fn);
void CallNative(uint64_t (*fn)(void* raw_context));
void CallNative(uint64_t (*fn)(void* raw_context, uint64_t arg0));
void CallNative(uint64_t (*fn)(void* raw_context, uint64_t arg0),
uint64_t arg0);
void CallNativeSafe(void* fn);
void SetReturnAddress(uint64_t value);
void ReloadECX();
void ReloadEDX();
void nop(size_t length = 1);
// TODO(benvanik): Label for epilog (don't use strings).
void LoadEflags();
void StoreEflags();
// Moves a 64bit immediate into memory.
bool ConstantFitsIn32Reg(uint64_t v);
void MovMem64(const Xbyak::RegExp& addr, uint64_t v);
Xbyak::Address GetXmmConstPtr(XmmConst id);
void LoadConstantXmm(Xbyak::Xmm dest, float v);
void LoadConstantXmm(Xbyak::Xmm dest, double v);
void LoadConstantXmm(Xbyak::Xmm dest, const vec128_t& v);
Xbyak::Address StashXmm(int index, const Xbyak::Xmm& r);
size_t stack_size() const { return stack_size_; }
protected:
void* Emplace(size_t stack_size);
int Emit(hir::HIRBuilder* builder, size_t& out_stack_size);
void EmitTraceSource(const hir::Instr* instr);
void EmitTraceSourceAppendValue(const hir::Value* value, size_t r8_offset);
void EmitGetCurrentThreadId();
void EmitTraceUserCallReturn();
protected:
runtime::Runtime* runtime_;
X64Backend* backend_;
X64CodeCache* code_cache_;
XbyakAllocator* allocator_;
hir::Instr* current_instr_;
size_t source_map_count_;
poly::Arena source_map_arena_;
size_t stack_size_;
uint32_t trace_flags_;
static const uint32_t gpr_reg_map_[GPR_COUNT];
static const uint32_t xmm_reg_map_[XMM_COUNT];
};
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_X64_X64_EMITTER_H_

52
src/xenia/cpu/backend/x64/x64_function.cc Normal file
View File

@@ -0,0 +1,52 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/x64/x64_function.h"
#include "xenia/cpu/backend/x64/x64_backend.h"
#include "xenia/cpu/runtime/runtime.h"
#include "xenia/cpu/runtime/thread_state.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
using xe::cpu::runtime::Breakpoint;
using xe::cpu::runtime::Function;
using xe::cpu::runtime::FunctionInfo;
using xe::cpu::runtime::ThreadState;
X64Function::X64Function(FunctionInfo* symbol_info)
: Function(symbol_info), machine_code_(nullptr), code_size_(0) {}
X64Function::~X64Function() {
// machine_code_ is freed by code cache.
}
void X64Function::Setup(void* machine_code, size_t code_size) {
machine_code_ = machine_code;
code_size_ = code_size;
}
int X64Function::AddBreakpointImpl(Breakpoint* breakpoint) { return 0; }
int X64Function::RemoveBreakpointImpl(Breakpoint* breakpoint) { return 0; }
int X64Function::CallImpl(ThreadState* thread_state, uint64_t return_address) {
auto backend = (X64Backend*)thread_state->runtime()->backend();
auto thunk = backend->host_to_guest_thunk();
thunk(machine_code_, thread_state->raw_context(), (void*)return_address);
return 0;
}
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe

47
src/xenia/cpu/backend/x64/x64_function.h Normal file
View File

@@ -0,0 +1,47 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2013 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_X64_X64_FUNCTION_H_
#define XENIA_BACKEND_X64_X64_FUNCTION_H_
#include "xenia/cpu/runtime/function.h"
#include "xenia/cpu/runtime/symbol_info.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
class X64Function : public runtime::Function {
public:
X64Function(runtime::FunctionInfo* symbol_info);
virtual ~X64Function();
void* machine_code() const { return machine_code_; }
size_t code_size() const { return code_size_; }
void Setup(void* machine_code, size_t code_size);
protected:
virtual int AddBreakpointImpl(runtime::Breakpoint* breakpoint);
virtual int RemoveBreakpointImpl(runtime::Breakpoint* breakpoint);
virtual int CallImpl(runtime::ThreadState* thread_state,
uint64_t return_address);
private:
void* machine_code_;
size_t code_size_;
};
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_X64_X64_FUNCTION_H_

758
src/xenia/cpu/backend/x64/x64_sequence.inl Normal file
View File

@@ -0,0 +1,758 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2014 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
namespace {
enum KeyType {
KEY_TYPE_X = OPCODE_SIG_TYPE_X,
KEY_TYPE_L = OPCODE_SIG_TYPE_L,
KEY_TYPE_O = OPCODE_SIG_TYPE_O,
KEY_TYPE_S = OPCODE_SIG_TYPE_S,
KEY_TYPE_V_I8 = OPCODE_SIG_TYPE_V + INT8_TYPE,
KEY_TYPE_V_I16 = OPCODE_SIG_TYPE_V + INT16_TYPE,
KEY_TYPE_V_I32 = OPCODE_SIG_TYPE_V + INT32_TYPE,
KEY_TYPE_V_I64 = OPCODE_SIG_TYPE_V + INT64_TYPE,
KEY_TYPE_V_F32 = OPCODE_SIG_TYPE_V + FLOAT32_TYPE,
KEY_TYPE_V_F64 = OPCODE_SIG_TYPE_V + FLOAT64_TYPE,
KEY_TYPE_V_V128 = OPCODE_SIG_TYPE_V + VEC128_TYPE,
};
#pragma pack(push, 1)
union InstrKey {
struct {
uint32_t opcode : 8;
uint32_t dest : 5;
uint32_t src1 : 5;
uint32_t src2 : 5;
uint32_t src3 : 5;
uint32_t reserved : 4;
};
uint32_t value;
operator uint32_t() const {
return value;
}
InstrKey() : value(0) {}
InstrKey(uint32_t v) : value(v) {}
InstrKey(const Instr* i) : value(0) {
opcode = i->opcode->num;
uint32_t sig = i->opcode->signature;
dest = GET_OPCODE_SIG_TYPE_DEST(sig) ? OPCODE_SIG_TYPE_V + i->dest->type : 0;
src1 = GET_OPCODE_SIG_TYPE_SRC1(sig);
if (src1 == OPCODE_SIG_TYPE_V) {
src1 += i->src1.value->type;
}
src2 = GET_OPCODE_SIG_TYPE_SRC2(sig);
if (src2 == OPCODE_SIG_TYPE_V) {
src2 += i->src2.value->type;
}
src3 = GET_OPCODE_SIG_TYPE_SRC3(sig);
if (src3 == OPCODE_SIG_TYPE_V) {
src3 += i->src3.value->type;
}
}
template <Opcode OPCODE,
KeyType DEST = KEY_TYPE_X,
KeyType SRC1 = KEY_TYPE_X,
KeyType SRC2 = KEY_TYPE_X,
KeyType SRC3 = KEY_TYPE_X>
struct Construct {
static const uint32_t value =
(OPCODE) | (DEST << 8) | (SRC1 << 13) | (SRC2 << 18) | (SRC3 << 23);
};
};
#pragma pack(pop)
static_assert(sizeof(InstrKey) <= 4, "Key must be 4 bytes");
template <typename... Ts>
struct CombinedStruct;
template <>
struct CombinedStruct<> {};
template <typename T, typename... Ts>
struct CombinedStruct<T, Ts...> : T, CombinedStruct<Ts...> {};
struct OpBase {};
template <typename T, KeyType KEY_TYPE>
struct Op : OpBase {
static const KeyType key_type = KEY_TYPE;
};
struct VoidOp : Op<VoidOp, KEY_TYPE_X> {
protected:
template <typename T, KeyType KEY_TYPE> friend struct Op;
template <hir::Opcode OPCODE, typename... Ts> friend struct I;
void Load(const Instr::Op& op) {}
};
struct OffsetOp : Op<OffsetOp, KEY_TYPE_O> {
uint64_t value;
protected:
template <typename T, KeyType KEY_TYPE> friend struct Op;
template <hir::Opcode OPCODE, typename... Ts> friend struct I;
void Load(const Instr::Op& op) {
this->value = op.offset;
}
};
struct SymbolOp : Op<SymbolOp, KEY_TYPE_S> {
FunctionInfo* value;
protected:
template <typename T, KeyType KEY_TYPE> friend struct Op;
template <hir::Opcode OPCODE, typename... Ts> friend struct I;
bool Load(const Instr::Op& op) {
this->value = op.symbol_info;
return true;
}
};
struct LabelOp : Op<LabelOp, KEY_TYPE_L> {
hir::Label* value;
protected:
template <typename T, KeyType KEY_TYPE> friend struct Op;
template <hir::Opcode OPCODE, typename... Ts> friend struct I;
void Load(const Instr::Op& op) {
this->value = op.label;
}
};
template <typename T, KeyType KEY_TYPE, typename REG_TYPE, typename CONST_TYPE, int TAG = -1>
struct ValueOp : Op<ValueOp<T, KEY_TYPE, REG_TYPE, CONST_TYPE, TAG>, KEY_TYPE> {
typedef REG_TYPE reg_type;
static const int tag = TAG;
const Value* value;
bool is_constant;
virtual bool ConstantFitsIn32Reg() const { return true; }
const REG_TYPE& reg() const {
assert_true(!is_constant);
return reg_;
}
operator const REG_TYPE&() const {
return reg();
}
bool IsEqual(const T& b) const {
if (is_constant && b.is_constant) {
return reinterpret_cast<const T*>(this)->constant() == b.constant();
} else if (!is_constant && !b.is_constant) {
return reg_.getIdx() == b.reg_.getIdx();
} else {
return false;
}
}
bool IsEqual(const Xbyak::Reg& b) const {
if (is_constant) {
return false;
} else if (!is_constant) {
return reg_.getIdx() == b.getIdx();
} else {
return false;
}
}
bool operator== (const T& b) const {
return IsEqual(b);
}
bool operator!= (const T& b) const {
return !IsEqual(b);
}
bool operator== (const Xbyak::Reg& b) const {
return IsEqual(b);
}
bool operator!= (const Xbyak::Reg& b) const {
return !IsEqual(b);
}
void Load(const Instr::Op& op) {
const Value* value = op.value;
this->value = value;
is_constant = value->IsConstant();
if (!is_constant) {
X64Emitter::SetupReg(value, reg_);
}
}
protected:
REG_TYPE reg_;
};
template <int TAG = -1>
struct I8 : ValueOp<I8<TAG>, KEY_TYPE_V_I8, Reg8, int8_t, TAG> {
typedef ValueOp<I8<TAG>, KEY_TYPE_V_I8, Reg8, int8_t, TAG> BASE;
const int8_t constant() const {
assert_true(BASE::is_constant);
return BASE::value->constant.i8;
}
};
template <int TAG = -1>
struct I16 : ValueOp<I16<TAG>, KEY_TYPE_V_I16, Reg16, int16_t, TAG> {
typedef ValueOp<I16<TAG>, KEY_TYPE_V_I16, Reg16, int16_t, TAG> BASE;
const int16_t constant() const {
assert_true(BASE::is_constant);
return BASE::value->constant.i16;
}
};
template <int TAG = -1>
struct I32 : ValueOp<I32<TAG>, KEY_TYPE_V_I32, Reg32, int32_t, TAG> {
typedef ValueOp<I32<TAG>, KEY_TYPE_V_I32, Reg32, int32_t, TAG> BASE;
const int32_t constant() const {
assert_true(BASE::is_constant);
return BASE::value->constant.i32;
}
};
template <int TAG = -1>
struct I64 : ValueOp<I64<TAG>, KEY_TYPE_V_I64, Reg64, int64_t, TAG> {
typedef ValueOp<I64<TAG>, KEY_TYPE_V_I64, Reg64, int64_t, TAG> BASE;
const int64_t constant() const {
assert_true(BASE::is_constant);
return BASE::value->constant.i64;
}
bool ConstantFitsIn32Reg() const override {
int64_t v = BASE::value->constant.i64;
if ((v & ~0x7FFFFFFF) == 0) {
// Fits under 31 bits, so just load using normal mov.
return true;
} else if ((v & ~0x7FFFFFFF) == ~0x7FFFFFFF) {
// Negative number that fits in 32bits.
return true;
}
return false;
}
};
template <int TAG = -1>
struct F32 : ValueOp<F32<TAG>, KEY_TYPE_V_F32, Xmm, float, TAG> {
typedef ValueOp<F32<TAG>, KEY_TYPE_V_F32, Xmm, float, TAG> BASE;
const float constant() const {
assert_true(BASE::is_constant);
return BASE::value->constant.f32;
}
};
template <int TAG = -1>
struct F64 : ValueOp<F64<TAG>, KEY_TYPE_V_F64, Xmm, double, TAG> {
typedef ValueOp<F64<TAG>, KEY_TYPE_V_F64, Xmm, double, TAG> BASE;
const double constant() const {
assert_true(BASE::is_constant);
return BASE::value->constant.f64;
}
};
template <int TAG = -1>
struct V128 : ValueOp<V128<TAG>, KEY_TYPE_V_V128, Xmm, vec128_t, TAG> {
typedef ValueOp<V128<TAG>, KEY_TYPE_V_V128, Xmm, vec128_t, TAG> BASE;
const vec128_t& constant() const {
assert_true(BASE::is_constant);
return BASE::value->constant.v128;
}
};
struct TagTable {
struct {
bool valid;
Instr::Op op;
} table[16];
template <typename T, typename std::enable_if<T::key_type == KEY_TYPE_X>::type* = nullptr>
bool CheckTag(const Instr::Op& op) {
return true;
}
template <typename T, typename std::enable_if<T::key_type == KEY_TYPE_L>::type* = nullptr>
bool CheckTag(const Instr::Op& op) {
return true;
}
template <typename T, typename std::enable_if<T::key_type == KEY_TYPE_O>::type* = nullptr>
bool CheckTag(const Instr::Op& op) {
return true;
}
template <typename T, typename std::enable_if<T::key_type == KEY_TYPE_S>::type* = nullptr>
bool CheckTag(const Instr::Op& op) {
return true;
}
template <typename T, typename std::enable_if<T::key_type >= KEY_TYPE_V_I8>::type* = nullptr>
bool CheckTag(const Instr::Op& op) {
const Value* value = op.value;
if (T::tag == -1) {
return true;
}
if (table[T::tag].valid &&
table[T::tag].op.value != value) {
return false;
}
table[T::tag].valid = true;
table[T::tag].op.value = (Value*)value;
return true;
}
};
template <typename DEST, typename... Tf>
struct DestField;
template <typename DEST>
struct DestField<DEST> {
DEST dest;
protected:
bool LoadDest(const Instr* i, TagTable& tag_table) {
Instr::Op op;
op.value = i->dest;
if (tag_table.CheckTag<DEST>(op)) {
dest.Load(op);
return true;
}
return false;
}
};
template <>
struct DestField<VoidOp> {
protected:
bool LoadDest(const Instr* i, TagTable& tag_table) {
return true;
}
};
template <hir::Opcode OPCODE, typename... Ts>
struct I;
template <hir::Opcode OPCODE, typename DEST>
struct I<OPCODE, DEST> : DestField<DEST> {
typedef DestField<DEST> BASE;
static const hir::Opcode opcode = OPCODE;
static const uint32_t key = InstrKey::Construct<OPCODE, DEST::key_type>::value;
static const KeyType dest_type = DEST::key_type;
const Instr* instr;
protected:
template <typename... Ti> friend struct SequenceFields;
bool Load(const Instr* i, TagTable& tag_table) {
if (InstrKey(i).value == key &&
BASE::LoadDest(i, tag_table)) {
instr = i;
return true;
}
return false;
}
};
template <hir::Opcode OPCODE, typename DEST, typename SRC1>
struct I<OPCODE, DEST, SRC1> : DestField<DEST> {
typedef DestField<DEST> BASE;
static const hir::Opcode opcode = OPCODE;
static const uint32_t key = InstrKey::Construct<OPCODE, DEST::key_type, SRC1::key_type>::value;
static const KeyType dest_type = DEST::key_type;
static const KeyType src1_type = SRC1::key_type;
const Instr* instr;
SRC1 src1;
protected:
template <typename... Ti> friend struct SequenceFields;
bool Load(const Instr* i, TagTable& tag_table) {
if (InstrKey(i).value == key &&
BASE::LoadDest(i, tag_table) &&
tag_table.CheckTag<SRC1>(i->src1)) {
instr = i;
src1.Load(i->src1);
return true;
}
return false;
}
};
template <hir::Opcode OPCODE, typename DEST, typename SRC1, typename SRC2>
struct I<OPCODE, DEST, SRC1, SRC2> : DestField<DEST> {
typedef DestField<DEST> BASE;
static const hir::Opcode opcode = OPCODE;
static const uint32_t key = InstrKey::Construct<OPCODE, DEST::key_type, SRC1::key_type, SRC2::key_type>::value;
static const KeyType dest_type = DEST::key_type;
static const KeyType src1_type = SRC1::key_type;
static const KeyType src2_type = SRC2::key_type;
const Instr* instr;
SRC1 src1;
SRC2 src2;
protected:
template <typename... Ti> friend struct SequenceFields;
bool Load(const Instr* i, TagTable& tag_table) {
if (InstrKey(i).value == key &&
BASE::LoadDest(i, tag_table) &&
tag_table.CheckTag<SRC1>(i->src1) &&
tag_table.CheckTag<SRC2>(i->src2)) {
instr = i;
src1.Load(i->src1);
src2.Load(i->src2);
return true;
}
return false;
}
};
template <hir::Opcode OPCODE, typename DEST, typename SRC1, typename SRC2, typename SRC3>
struct I<OPCODE, DEST, SRC1, SRC2, SRC3> : DestField<DEST> {
typedef DestField<DEST> BASE;
static const hir::Opcode opcode = OPCODE;
static const uint32_t key = InstrKey::Construct<OPCODE, DEST::key_type, SRC1::key_type, SRC2::key_type, SRC3::key_type>::value;
static const KeyType dest_type = DEST::key_type;
static const KeyType src1_type = SRC1::key_type;
static const KeyType src2_type = SRC2::key_type;
static const KeyType src3_type = SRC3::key_type;
const Instr* instr;
SRC1 src1;
SRC2 src2;
SRC3 src3;
protected:
template <typename... Ti> friend struct SequenceFields;
bool Load(const Instr* i, TagTable& tag_table) {
if (InstrKey(i).value == key &&
BASE::LoadDest(i, tag_table) &&
tag_table.CheckTag<SRC1>(i->src1) &&
tag_table.CheckTag<SRC2>(i->src2) &&
tag_table.CheckTag<SRC3>(i->src3)) {
instr = i;
src1.Load(i->src1);
src2.Load(i->src2);
src3.Load(i->src3);
return true;
}
return false;
}
};
template <typename... Ti>
struct SequenceFields;
template <typename I1>
struct SequenceFields<I1> {
I1 i1;
protected:
template <typename SEQ, typename... Ti> friend struct Sequence;
bool Check(const Instr* i, TagTable& tag_table, const Instr** new_tail) {
if (i1.Load(i, tag_table)) {
*new_tail = i->next;
return true;
}
return false;
}
};
template <typename I1, typename I2>
struct SequenceFields<I1, I2> : SequenceFields<I1> {
I2 i2;
protected:
template <typename SEQ, typename... Ti> friend struct Sequence;
bool Check(const Instr* i, TagTable& tag_table, const Instr** new_tail) {
if (SequenceFields<I1>::Check(i, tag_table, new_tail)) {
auto ni = i->next;
if (ni && i2.Load(ni, tag_table)) {
*new_tail = ni;
return i;
}
}
return false;
}
};
template <typename I1, typename I2, typename I3>
struct SequenceFields<I1, I2, I3> : SequenceFields<I1, I2> {
I3 i3;
protected:
template <typename SEQ, typename... Ti> friend struct Sequence;
bool Check(const Instr* i, TagTable& tag_table, const Instr** new_tail) {
if (SequenceFields<I1, I2>::Check(i, tag_table, new_tail)) {
auto ni = i->next;
if (ni && i3.Load(ni, tag_table)) {
*new_tail = ni;
return i;
}
}
return false;
}
};
template <typename I1, typename I2, typename I3, typename I4>
struct SequenceFields<I1, I2, I3, I4> : SequenceFields<I1, I2, I3> {
I4 i4;
protected:
template <typename SEQ, typename... Ti> friend struct Sequence;
bool Check(const Instr* i, TagTable& tag_table, const Instr** new_tail) {
if (SequenceFields<I1, I2, I3>::Check(i, tag_table, new_tail)) {
auto ni = i->next;
if (ni && i4.Load(ni, tag_table)) {
*new_tail = ni;
return i;
}
}
return false;
}
};
template <typename I1, typename I2, typename I3, typename I4, typename I5>
struct SequenceFields<I1, I2, I3, I4, I5> : SequenceFields<I1, I2, I3, I4> {
I5 i5;
protected:
template <typename SEQ, typename... Ti> friend struct Sequence;
bool Check(const Instr* i, TagTable& tag_table, const Instr** new_tail) {
if (SequenceFields<I1, I2, I3, I4>::Check(i, tag_table, new_tail)) {
auto ni = i->next;
if (ni && i5.Load(ni, tag_table)) {
*new_tail = ni;
return i;
}
}
return false;
}
};
template <typename SEQ, typename... Ti>
struct Sequence {
struct EmitArgs : SequenceFields<Ti...> {};
static bool Select(X64Emitter& e, const Instr* i, const Instr** new_tail) {
EmitArgs args;
TagTable tag_table;
if (!args.Check(i, tag_table, new_tail)) {
return false;
}
SEQ::Emit(e, args);
return true;
}
};
template <typename T>
const T GetTempReg(X64Emitter& e);
template <>
const Reg8 GetTempReg<Reg8>(X64Emitter& e) {
return e.al;
}
template <>
const Reg16 GetTempReg<Reg16>(X64Emitter& e) {
return e.ax;
}
template <>
const Reg32 GetTempReg<Reg32>(X64Emitter& e) {
return e.eax;
}
template <>
const Reg64 GetTempReg<Reg64>(X64Emitter& e) {
return e.rax;
}
template <typename SEQ, typename T>
struct SingleSequence : public Sequence<SingleSequence<SEQ, T>, T> {
typedef Sequence<SingleSequence<SEQ, T>, T> BASE;
typedef T EmitArgType;
// TODO(benvanik): find a way to do this cross-compiler.
#if XE_COMPILER_MSVC
static uint32_t head_key() { return T::key; }
#else
static constexpr uint32_t head_key() { return T::key; }
#endif // XE_COMPILER_MSVC
static void Emit(X64Emitter& e, const typename BASE::EmitArgs& _) {
SEQ::Emit(e, _.i1);
}
template <typename REG_FN>
static void EmitUnaryOp(
X64Emitter& e, const EmitArgType& i,
const REG_FN& reg_fn) {
if (i.src1.is_constant) {
e.mov(i.dest, i.src1.constant());
reg_fn(e, i.dest);
} else {
if (i.dest != i.src1) {
e.mov(i.dest, i.src1);
}
reg_fn(e, i.dest);
}
}
template <typename REG_REG_FN, typename REG_CONST_FN>
static void EmitCommutativeBinaryOp(
X64Emitter& e, const EmitArgType& i,
const REG_REG_FN& reg_reg_fn, const REG_CONST_FN& reg_const_fn) {
if (i.src1.is_constant) {
assert_true(!i.src2.is_constant);
if (i.dest == i.src2) {
if (i.src1.ConstantFitsIn32Reg()) {
reg_const_fn(e, i.dest, static_cast<int32_t>(i.src1.constant()));
} else {
auto temp = GetTempReg<typename decltype(i.src1)::reg_type>(e);
e.mov(temp, i.src1.constant());
reg_reg_fn(e, i.dest, temp);
}
} else {
e.mov(i.dest, i.src1.constant());
reg_reg_fn(e, i.dest, i.src2);
}
} else if (i.src2.is_constant) {
if (i.dest == i.src1) {
if (i.src2.ConstantFitsIn32Reg()) {
reg_const_fn(e, i.dest, static_cast<int32_t>(i.src2.constant()));
} else {
auto temp = GetTempReg<typename decltype(i.src2)::reg_type>(e);
e.mov(temp, i.src2.constant());
reg_reg_fn(e, i.dest, temp);
}
} else {
e.mov(i.dest, i.src2.constant());
reg_reg_fn(e, i.dest, i.src1);
}
} else {
if (i.dest == i.src1) {
reg_reg_fn(e, i.dest, i.src2);
} else if (i.dest == i.src2) {
reg_reg_fn(e, i.dest, i.src1);
} else {
e.mov(i.dest, i.src1);
reg_reg_fn(e, i.dest, i.src2);
}
}
}
template <typename REG_REG_FN, typename REG_CONST_FN>
static void EmitAssociativeBinaryOp(
X64Emitter& e, const EmitArgType& i,
const REG_REG_FN& reg_reg_fn, const REG_CONST_FN& reg_const_fn) {
if (i.src1.is_constant) {
assert_true(!i.src2.is_constant);
if (i.dest == i.src2) {
auto temp = GetTempReg<typename decltype(i.src2)::reg_type>(e);
e.mov(temp, i.src2);
e.mov(i.dest, i.src1.constant());
reg_reg_fn(e, i.dest, temp);
} else {
e.mov(i.dest, i.src1.constant());
reg_reg_fn(e, i.dest, i.src2);
}
} else if (i.src2.is_constant) {
if (i.dest == i.src1) {
if (i.src2.ConstantFitsIn32Reg()) {
reg_const_fn(e, i.dest, static_cast<int32_t>(i.src2.constant()));
} else {
auto temp = GetTempReg<typename decltype(i.src2)::reg_type>(e);
e.mov(temp, i.src2.constant());
reg_reg_fn(e, i.dest, temp);
}
} else {
e.mov(i.dest, i.src1);
if (i.src2.ConstantFitsIn32Reg()) {
reg_const_fn(e, i.dest, static_cast<int32_t>(i.src2.constant()));
} else {
auto temp = GetTempReg<typename decltype(i.src2)::reg_type>(e);
e.mov(temp, i.src2.constant());
reg_reg_fn(e, i.dest, temp);
}
}
} else {
if (i.dest == i.src1) {
reg_reg_fn(e, i.dest, i.src2);
} else if (i.dest == i.src2) {
auto temp = GetTempReg<typename decltype(i.src2)::reg_type>(e);
e.mov(temp, i.src2);
e.mov(i.dest, i.src1);
reg_reg_fn(e, i.dest, temp);
} else {
e.mov(i.dest, i.src1);
reg_reg_fn(e, i.dest, i.src2);
}
}
}
template <typename FN>
static void EmitCommutativeBinaryXmmOp(
X64Emitter& e, const EmitArgType& i, const FN& fn) {
if (i.src1.is_constant) {
assert_true(!i.src2.is_constant);
e.LoadConstantXmm(e.xmm0, i.src1.constant());
fn(e, i.dest, e.xmm0, i.src2);
} else if (i.src2.is_constant) {
e.LoadConstantXmm(e.xmm0, i.src2.constant());
fn(e, i.dest, i.src1, e.xmm0);
} else {
fn(e, i.dest, i.src1, i.src2);
}
}
template <typename FN>
static void EmitAssociativeBinaryXmmOp(
X64Emitter& e, const EmitArgType& i, const FN& fn) {
if (i.src1.is_constant) {
assert_true(!i.src2.is_constant);
e.LoadConstantXmm(e.xmm0, i.src1.constant());
fn(e, i.dest, e.xmm0, i.src2);
} else if (i.src2.is_constant) {
e.LoadConstantXmm(e.xmm0, i.src2.constant());
fn(e, i.dest, i.src1, e.xmm0);
} else {
fn(e, i.dest, i.src1, i.src2);
}
}
template <typename REG_REG_FN, typename REG_CONST_FN>
static void EmitCommutativeCompareOp(
X64Emitter& e, const EmitArgType& i,
const REG_REG_FN& reg_reg_fn, const REG_CONST_FN& reg_const_fn) {
if (i.src1.is_constant) {
assert_true(!i.src2.is_constant);
if (i.src1.ConstantFitsIn32Reg()) {
reg_const_fn(e, i.src2, static_cast<int32_t>(i.src1.constant()));
} else {
auto temp = GetTempReg<typename decltype(i.src1)::reg_type>(e);
e.mov(temp, i.src1.constant());
reg_reg_fn(e, i.src2, temp);
}
} else if (i.src2.is_constant) {
if (i.src2.ConstantFitsIn32Reg()) {
reg_const_fn(e, i.src1, static_cast<int32_t>(i.src2.constant()));
} else {
auto temp = GetTempReg<typename decltype(i.src2)::reg_type>(e);
e.mov(temp, i.src2.constant());
reg_reg_fn(e, i.src1, temp);
}
} else {
reg_reg_fn(e, i.src1, i.src2);
}
}
template <typename REG_REG_FN, typename REG_CONST_FN>
static void EmitAssociativeCompareOp(
X64Emitter& e, const EmitArgType& i,
const REG_REG_FN& reg_reg_fn, const REG_CONST_FN& reg_const_fn) {
if (i.src1.is_constant) {
assert_true(!i.src2.is_constant);
if (i.src1.ConstantFitsIn32Reg()) {
reg_const_fn(e, i.dest, i.src2, static_cast<int32_t>(i.src1.constant()), true);
} else {
auto temp = GetTempReg<typename decltype(i.src1)::reg_type>(e);
e.mov(temp, i.src1.constant());
reg_reg_fn(e, i.dest, i.src2, temp, true);
}
} else if (i.src2.is_constant) {
if (i.src2.ConstantFitsIn32Reg()) {
reg_const_fn(e, i.dest, i.src1, static_cast<int32_t>(i.src2.constant()), false);
} else {
auto temp = GetTempReg<typename decltype(i.src2)::reg_type>(e);
e.mov(temp, i.src2.constant());
reg_reg_fn(e, i.dest, i.src1, temp, false);
}
} else {
reg_reg_fn(e, i.dest, i.src1, i.src2, false);
}
}
};
static const int ANY = -1;
typedef int tag_t;
static const tag_t TAG0 = 0;
static const tag_t TAG1 = 1;
static const tag_t TAG2 = 2;
static const tag_t TAG3 = 3;
static const tag_t TAG4 = 4;
static const tag_t TAG5 = 5;
static const tag_t TAG6 = 6;
static const tag_t TAG7 = 7;
template <typename T>
void Register() {
sequence_table.insert({ T::head_key(), T::Select });
}
template <typename T, typename Tn, typename... Ts>
void Register() {
Register<T>();
Register<Tn, Ts...>();
};
#define EMITTER_OPCODE_TABLE(name, ...) \
void Register_##name() { \
Register<__VA_ARGS__>(); \
}
#define MATCH(...) __VA_ARGS__
#define EMITTER(name, match) struct name : SingleSequence<name, match>
#define SEQUENCE(name, match) struct name : Sequence<name, match>
} // namespace

5887
src/xenia/cpu/backend/x64/x64_sequences.cc Normal file
View File

File diff suppressed because it is too large Load Diff

37
src/xenia/cpu/backend/x64/x64_sequences.h Normal file
View File

@@ -0,0 +1,37 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2014 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_X64_X64_SEQUENCES_H_
#define XENIA_BACKEND_X64_X64_SEQUENCES_H_
namespace xe {
namespace cpu {
namespace hir {
class Instr;
} // namespace hir
} // namespace cpu
} // namespace xe
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
class X64Emitter;
void RegisterSequences();
bool SelectSequence(X64Emitter& e, const hir::Instr* i,
const hir::Instr** new_tail);
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_X64_X64_SEQUENCES_H_

147
src/xenia/cpu/backend/x64/x64_thunk_emitter.cc Normal file
View File

@@ -0,0 +1,147 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2014 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/x64/x64_thunk_emitter.h"
#include "third_party/xbyak/xbyak/xbyak.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
using namespace Xbyak;
X64ThunkEmitter::X64ThunkEmitter(X64Backend* backend, XbyakAllocator* allocator)
: X64Emitter(backend, allocator) {}
X64ThunkEmitter::~X64ThunkEmitter() {}
HostToGuestThunk X64ThunkEmitter::EmitHostToGuestThunk() {
// rcx = target
// rdx = arg0
// r8 = arg1
const size_t stack_size = StackLayout::THUNK_STACK_SIZE;
// rsp + 0 = return address
mov(qword[rsp + 8 * 3], r8);
mov(qword[rsp + 8 * 2], rdx);
mov(qword[rsp + 8 * 1], rcx);
sub(rsp, stack_size);
mov(qword[rsp + 48], rbx);
mov(qword[rsp + 56], rcx);
mov(qword[rsp + 64], rbp);
mov(qword[rsp + 72], rsi);
mov(qword[rsp + 80], rdi);
mov(qword[rsp + 88], r12);
mov(qword[rsp + 96], r13);
mov(qword[rsp + 104], r14);
mov(qword[rsp + 112], r15);
/*movaps(ptr[rsp + 128], xmm6);
movaps(ptr[rsp + 144], xmm7);
movaps(ptr[rsp + 160], xmm8);
movaps(ptr[rsp + 176], xmm9);
movaps(ptr[rsp + 192], xmm10);
movaps(ptr[rsp + 208], xmm11);
movaps(ptr[rsp + 224], xmm12);
movaps(ptr[rsp + 240], xmm13);
movaps(ptr[rsp + 256], xmm14);
movaps(ptr[rsp + 272], xmm15);*/
mov(rax, rcx);
mov(rcx, rdx);
mov(rdx, r8);
call(rax);
/*movaps(xmm6, ptr[rsp + 128]);
movaps(xmm7, ptr[rsp + 144]);
movaps(xmm8, ptr[rsp + 160]);
movaps(xmm9, ptr[rsp + 176]);
movaps(xmm10, ptr[rsp + 192]);
movaps(xmm11, ptr[rsp + 208]);
movaps(xmm12, ptr[rsp + 224]);
movaps(xmm13, ptr[rsp + 240]);
movaps(xmm14, ptr[rsp + 256]);
movaps(xmm15, ptr[rsp + 272]);*/
mov(rbx, qword[rsp + 48]);
mov(rcx, qword[rsp + 56]);
mov(rbp, qword[rsp + 64]);
mov(rsi, qword[rsp + 72]);
mov(rdi, qword[rsp + 80]);
mov(r12, qword[rsp + 88]);
mov(r13, qword[rsp + 96]);
mov(r14, qword[rsp + 104]);
mov(r15, qword[rsp + 112]);
add(rsp, stack_size);
mov(rcx, qword[rsp + 8 * 1]);
mov(rdx, qword[rsp + 8 * 2]);
mov(r8, qword[rsp + 8 * 3]);
ret();
void* fn = Emplace(stack_size);
return (HostToGuestThunk)fn;
}
GuestToHostThunk X64ThunkEmitter::EmitGuestToHostThunk() {
// rcx = context
// rdx = target function
// r8 = arg0
// r9 = arg1
const size_t stack_size = StackLayout::THUNK_STACK_SIZE;
// rsp + 0 = return address
mov(qword[rsp + 8 * 2], rdx);
mov(qword[rsp + 8 * 1], rcx);
sub(rsp, stack_size);
mov(qword[rsp + 48], rbx);
mov(qword[rsp + 56], rcx);
mov(qword[rsp + 64], rbp);
mov(qword[rsp + 72], rsi);
mov(qword[rsp + 80], rdi);
mov(qword[rsp + 88], r12);
mov(qword[rsp + 96], r13);
mov(qword[rsp + 104], r14);
mov(qword[rsp + 112], r15);
// TODO(benvanik): save things? XMM0-5?
mov(rax, rdx);
mov(rdx, r8);
mov(r8, r9);
mov(r9, r10);
call(rax);
mov(rbx, qword[rsp + 48]);
mov(rcx, qword[rsp + 56]);
mov(rbp, qword[rsp + 64]);
mov(rsi, qword[rsp + 72]);
mov(rdi, qword[rsp + 80]);
mov(r12, qword[rsp + 88]);
mov(r13, qword[rsp + 96]);
mov(r14, qword[rsp + 104]);
mov(r15, qword[rsp + 112]);
add(rsp, stack_size);
mov(rcx, qword[rsp + 8 * 1]);
mov(rdx, qword[rsp + 8 * 2]);
ret();
void* fn = Emplace(stack_size);
return (HostToGuestThunk)fn;
}
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe

143
src/xenia/cpu/backend/x64/x64_thunk_emitter.h Normal file
View File

@@ -0,0 +1,143 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2014 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_X64_X64_THUNK_EMITTER_H_
#define XENIA_BACKEND_X64_X64_THUNK_EMITTER_H_
#include "xenia/cpu/backend/x64/x64_backend.h"
#include "xenia/cpu/backend/x64/x64_emitter.h"
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
/**
* Stack Layout
* ----------------------------
* NOTE: stack must always be 16b aligned.
*
* Thunk stack:
* +------------------+
* | arg temp, 3 * 8 | rsp + 0
* | |
* | |
* +------------------+
* | scratch, 16b | rsp + 32
* | |
* +------------------+
* | rbx | rsp + 48
* +------------------+
* | rcx / context | rsp + 56
* +------------------+
* | rbp | rsp + 64
* +------------------+
* | rsi | rsp + 72
* +------------------+
* | rdi | rsp + 80
* +------------------+
* | r12 | rsp + 88
* +------------------+
* | r13 | rsp + 96
* +------------------+
* | r14 | rsp + 104
* +------------------+
* | r15 | rsp + 112
* +------------------+
* | (return address) | rsp + 120
* +------------------+
* | (rcx home) | rsp + 128
* +------------------+
* | (rdx home) | rsp + 136
* +------------------+
*
*
* TODO:
* +------------------+
* | xmm6 | rsp + 128
* | |
* +------------------+
* | xmm7 | rsp + 144
* | |
* +------------------+
* | xmm8 | rsp + 160
* | |
* +------------------+
* | xmm9 | rsp + 176
* | |
* +------------------+
* | xmm10 | rsp + 192
* | |
* +------------------+
* | xmm11 | rsp + 208
* | |
* +------------------+
* | xmm12 | rsp + 224
* | |
* +------------------+
* | xmm13 | rsp + 240
* | |
* +------------------+
* | xmm14 | rsp + 256
* | |
* +------------------+
* | xmm15 | rsp + 272
* | |
* +------------------+
*
* Guest stack:
* +------------------+
* | arg temp, 3 * 8 | rsp + 0
* | |
* | |
* +------------------+
* | scratch, 48b | rsp + 32
* | |
* +------------------+
* | rcx / context | rsp + 80
* +------------------+
* | guest ret addr | rsp + 88
* +------------------+
* | call ret addr | rsp + 96
* +------------------+
* ... locals ...
* +------------------+
* | (return address) |
* +------------------+
*
*/
class StackLayout {
public:
const static size_t THUNK_STACK_SIZE = 120;
const static size_t GUEST_STACK_SIZE = 104;
const static size_t GUEST_RCX_HOME = 80;
const static size_t GUEST_RET_ADDR = 88;
const static size_t GUEST_CALL_RET_ADDR = 96;
};
class X64ThunkEmitter : public X64Emitter {
public:
X64ThunkEmitter(X64Backend* backend, XbyakAllocator* allocator);
virtual ~X64ThunkEmitter();
// Call a generated function, saving all stack parameters.
HostToGuestThunk EmitHostToGuestThunk();
// Function that guest code can call to transition into host code.
GuestToHostThunk EmitGuestToHostThunk();
};
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_X64_X64_THUNK_EMITTER_H_

202
src/xenia/cpu/backend/x64/x64_tracers.cc Normal file
View File

@@ -0,0 +1,202 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2014 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/cpu/backend/x64/x64_tracers.h"
#include "xenia/cpu/backend/x64/x64_emitter.h"
#include "xenia/cpu/runtime/runtime.h"
#include "xenia/cpu/runtime/thread_state.h"
using namespace xe;
using namespace xe::cpu::backend::x64;
using namespace xe::cpu::runtime;
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
#define ITRACE 0
#define DTRACE 0
#define TARGET_THREAD 1
#define IFLUSH() fflush(stdout)
#define IPRINT \
if (thread_state->thread_id() == TARGET_THREAD) printf
#define DFLUSH() fflush(stdout)
#define DPRINT \
DFLUSH(); \
if (thread_state->thread_id() == TARGET_THREAD) printf
uint32_t GetTracingMode() {
uint32_t mode = 0;
#if ITRACE
mode |= TRACING_INSTR;
#endif // ITRACE
#if DTRACE
mode |= TRACING_DATA;
#endif // DTRACE
return mode;
}
void TraceString(void* raw_context, const char* str) {
auto thread_state = *((ThreadState**)raw_context);
IPRINT("XE[t] :%d: %s\n", thread_state->thread_id(), str);
IFLUSH();
}
void TraceContextLoadI8(void* raw_context, uint64_t offset, uint8_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%d (%X) = ctx i8 +%llu\n", (int8_t)value, value, offset);
}
void TraceContextLoadI16(void* raw_context, uint64_t offset, uint16_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%d (%X) = ctx i16 +%llu\n", (int16_t)value, value, offset);
}
void TraceContextLoadI32(void* raw_context, uint64_t offset, uint32_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%d (%X) = ctx i32 +%llu\n", (int32_t)value, value, offset);
}
void TraceContextLoadI64(void* raw_context, uint64_t offset, uint64_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%lld (%llX) = ctx i64 +%llu\n", (int64_t)value, value, offset);
}
void TraceContextLoadF32(void* raw_context, uint64_t offset, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%e (%X) = ctx f32 +%llu\n", poly::m128_f32<0>(value),
poly::m128_i32<0>(value), offset);
}
void TraceContextLoadF64(void* raw_context, uint64_t offset,
const double* value) {
auto thread_state = *((ThreadState**)raw_context);
auto v = _mm_loadu_pd(value);
DPRINT("%le (%llX) = ctx f64 +%llu\n", poly::m128_f64<0>(v),
poly::m128_i64<0>(v), offset);
}
void TraceContextLoadV128(void* raw_context, uint64_t offset, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("[%e, %e, %e, %e] [%.8X, %.8X, %.8X, %.8X] = ctx v128 +%llu\n",
poly::m128_f32<0>(value), poly::m128_f32<1>(value),
poly::m128_f32<2>(value), poly::m128_f32<3>(value),
poly::m128_i32<0>(value), poly::m128_i32<1>(value),
poly::m128_i32<2>(value), poly::m128_i32<3>(value), offset);
}
void TraceContextStoreI8(void* raw_context, uint64_t offset, uint8_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("ctx i8 +%llu = %d (%X)\n", offset, (int8_t)value, value);
}
void TraceContextStoreI16(void* raw_context, uint64_t offset, uint16_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("ctx i16 +%llu = %d (%X)\n", offset, (int16_t)value, value);
}
void TraceContextStoreI32(void* raw_context, uint64_t offset, uint32_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("ctx i32 +%llu = %d (%X)\n", offset, (int32_t)value, value);
}
void TraceContextStoreI64(void* raw_context, uint64_t offset, uint64_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("ctx i64 +%llu = %lld (%llX)\n", offset, (int64_t)value, value);
}
void TraceContextStoreF32(void* raw_context, uint64_t offset, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("ctx f32 +%llu = %e (%X)\n", offset, poly::m128_f32<0>(value),
poly::m128_i32<0>(value));
}
void TraceContextStoreF64(void* raw_context, uint64_t offset,
const double* value) {
auto thread_state = *((ThreadState**)raw_context);
auto v = _mm_loadu_pd(value);
DPRINT("ctx f64 +%llu = %le (%llX)\n", offset, poly::m128_f64<0>(v),
poly::m128_i64<0>(v));
}
void TraceContextStoreV128(void* raw_context, uint64_t offset, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("ctx v128 +%llu = [%e, %e, %e, %e] [%.8X, %.8X, %.8X, %.8X]\n", offset,
poly::m128_f32<0>(value), poly::m128_f32<1>(value),
poly::m128_f32<2>(value), poly::m128_f32<3>(value),
poly::m128_i32<0>(value), poly::m128_i32<1>(value),
poly::m128_i32<2>(value), poly::m128_i32<3>(value));
}
void TraceMemoryLoadI8(void* raw_context, uint32_t address, uint8_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%d (%X) = load.i8 %.8X\n", (int8_t)value, value, address);
}
void TraceMemoryLoadI16(void* raw_context, uint32_t address, uint16_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%d (%X) = load.i16 %.8X\n", (int16_t)value, value, address);
}
void TraceMemoryLoadI32(void* raw_context, uint32_t address, uint32_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%d (%X) = load.i32 %.8X\n", (int32_t)value, value, address);
}
void TraceMemoryLoadI64(void* raw_context, uint32_t address, uint64_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%lld (%llX) = load.i64 %.8X\n", (int64_t)value, value, address);
}
void TraceMemoryLoadF32(void* raw_context, uint32_t address, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%e (%X) = load.f32 %.8X\n", poly::m128_f32<0>(value),
poly::m128_i32<0>(value), address);
}
void TraceMemoryLoadF64(void* raw_context, uint32_t address, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("%le (%llX) = load.f64 %.8X\n", poly::m128_f64<0>(value),
poly::m128_i64<0>(value), address);
}
void TraceMemoryLoadV128(void* raw_context, uint32_t address, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("[%e, %e, %e, %e] [%.8X, %.8X, %.8X, %.8X] = load.v128 %.8X\n",
poly::m128_f32<0>(value), poly::m128_f32<1>(value),
poly::m128_f32<2>(value), poly::m128_f32<3>(value),
poly::m128_i32<0>(value), poly::m128_i32<1>(value),
poly::m128_i32<2>(value), poly::m128_i32<3>(value), address);
}
void TraceMemoryStoreI8(void* raw_context, uint32_t address, uint8_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("store.i8 %.8X = %d (%X)\n", address, (int8_t)value, value);
}
void TraceMemoryStoreI16(void* raw_context, uint32_t address, uint16_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("store.i16 %.8X = %d (%X)\n", address, (int16_t)value, value);
}
void TraceMemoryStoreI32(void* raw_context, uint32_t address, uint32_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("store.i32 %.8X = %d (%X)\n", address, (int32_t)value, value);
}
void TraceMemoryStoreI64(void* raw_context, uint32_t address, uint64_t value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("store.i64 %.8X = %lld (%llX)\n", address, (int64_t)value, value);
}
void TraceMemoryStoreF32(void* raw_context, uint32_t address, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("store.f32 %.8X = %e (%X)\n", address, poly::m128_f32<0>(value),
poly::m128_i32<0>(value));
}
void TraceMemoryStoreF64(void* raw_context, uint32_t address, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("store.f64 %.8X = %le (%llX)\n", address, poly::m128_f64<0>(value),
poly::m128_i64<0>(value));
}
void TraceMemoryStoreV128(void* raw_context, uint32_t address, __m128 value) {
auto thread_state = *((ThreadState**)raw_context);
DPRINT("store.v128 %.8X = [%e, %e, %e, %e] [%.8X, %.8X, %.8X, %.8X]\n",
address, poly::m128_f32<0>(value), poly::m128_f32<1>(value),
poly::m128_f32<2>(value), poly::m128_f32<3>(value),
poly::m128_i32<0>(value), poly::m128_i32<1>(value),
poly::m128_i32<2>(value), poly::m128_i32<3>(value));
}
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe

72
src/xenia/cpu/backend/x64/x64_tracers.h Normal file
View File

@@ -0,0 +1,72 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2014 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_BACKEND_X64_X64_TRACERS_H_
#define XENIA_BACKEND_X64_X64_TRACERS_H_
#include <xmmintrin.h>
#include <cstdint>
namespace xe {
namespace cpu {
namespace backend {
namespace x64 {
class X64Emitter;
enum TracingMode {
TRACING_INSTR = (1 << 1),
TRACING_DATA = (1 << 2),
};
uint32_t GetTracingMode();
inline bool IsTracingInstr() { return (GetTracingMode() & TRACING_INSTR) != 0; }
inline bool IsTracingData() { return (GetTracingMode() & TRACING_DATA) != 0; }
void TraceString(void* raw_context, const char* str);
void TraceContextLoadI8(void* raw_context, uint64_t offset, uint8_t value);
void TraceContextLoadI16(void* raw_context, uint64_t offset, uint16_t value);
void TraceContextLoadI32(void* raw_context, uint64_t offset, uint32_t value);
void TraceContextLoadI64(void* raw_context, uint64_t offset, uint64_t value);
void TraceContextLoadF32(void* raw_context, uint64_t offset, __m128 value);
void TraceContextLoadF64(void* raw_context, uint64_t offset,
const double* value);
void TraceContextLoadV128(void* raw_context, uint64_t offset, __m128 value);
void TraceContextStoreI8(void* raw_context, uint64_t offset, uint8_t value);
void TraceContextStoreI16(void* raw_context, uint64_t offset, uint16_t value);
void TraceContextStoreI32(void* raw_context, uint64_t offset, uint32_t value);
void TraceContextStoreI64(void* raw_context, uint64_t offset, uint64_t value);
void TraceContextStoreF32(void* raw_context, uint64_t offset, __m128 value);
void TraceContextStoreF64(void* raw_context, uint64_t offset,
const double* value);
void TraceContextStoreV128(void* raw_context, uint64_t offset, __m128 value);
void TraceMemoryLoadI8(void* raw_context, uint32_t address, uint8_t value);
void TraceMemoryLoadI16(void* raw_context, uint32_t address, uint16_t value);
void TraceMemoryLoadI32(void* raw_context, uint32_t address, uint32_t value);
void TraceMemoryLoadI64(void* raw_context, uint32_t address, uint64_t value);
void TraceMemoryLoadF32(void* raw_context, uint32_t address, __m128 value);
void TraceMemoryLoadF64(void* raw_context, uint32_t address, __m128 value);
void TraceMemoryLoadV128(void* raw_context, uint32_t address, __m128 value);
void TraceMemoryStoreI8(void* raw_context, uint32_t address, uint8_t value);
void TraceMemoryStoreI16(void* raw_context, uint32_t address, uint16_t value);
void TraceMemoryStoreI32(void* raw_context, uint32_t address, uint32_t value);
void TraceMemoryStoreI64(void* raw_context, uint32_t address, uint64_t value);
void TraceMemoryStoreF32(void* raw_context, uint32_t address, __m128 value);
void TraceMemoryStoreF64(void* raw_context, uint32_t address, __m128 value);
void TraceMemoryStoreV128(void* raw_context, uint32_t address, __m128 value);
} // namespace x64
} // namespace backend
} // namespace cpu
} // namespace xe
#endif // XENIA_BACKEND_X64_X64_TRACERS_H_