mirror of
https://github.com/dolphin-emu/dolphin.git
synced 2024-09-22 04:11:46 +02:00
84711122ee
align the stack on the code we generate. Fix some existing code to do what the comments claim it already does, and then actually use that code. :) git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@1650 8ced0084-cf51-0410-be5f-012b33b47a6e
313 lines
6.8 KiB
C++
313 lines
6.8 KiB
C++
// Copyright (C) 2003-2008 Dolphin Project.
|
|
|
|
// This program is free software: you can redistribute it and/or modify
|
|
// it under the terms of the GNU General Public License as published by
|
|
// the Free Software Foundation, version 2.0.
|
|
|
|
// This program is distributed in the hope that it will be useful,
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
// GNU General Public License 2.0 for more details.
|
|
|
|
// A copy of the GPL 2.0 should have been included with the program.
|
|
// If not, see http://www.gnu.org/licenses/
|
|
|
|
// Official SVN repository and contact information can be found at
|
|
// http://code.google.com/p/dolphin-emu/
|
|
|
|
#include "Common.h"
|
|
#include "x64Emitter.h"
|
|
#include "ABI.h"
|
|
|
|
using namespace Gen;
|
|
|
|
// Shared code between Win64 and Unix64
|
|
// ====================================
|
|
|
|
// Sets up a __cdecl function.
|
|
void XEmitter::ABI_EmitPrologue(int maxCallParams)
|
|
{
|
|
#ifdef _M_IX86
|
|
// Don't really need to do anything
|
|
#elif defined(_M_X64)
|
|
#if _WIN32
|
|
int stacksize = ((maxCallParams + 1) & ~1) * 8 + 8;
|
|
// Set up a stack frame so that we can call functions
|
|
// TODO: use maxCallParams
|
|
SUB(64, R(RSP), Imm8(stacksize));
|
|
#endif
|
|
#else
|
|
#error Arch not supported
|
|
#endif
|
|
}
|
|
|
|
void XEmitter::ABI_EmitEpilogue(int maxCallParams)
|
|
{
|
|
#ifdef _M_IX86
|
|
RET();
|
|
#elif defined(_M_X64)
|
|
#ifdef _WIN32
|
|
int stacksize = ((maxCallParams+1)&~1)*8 + 8;
|
|
ADD(64, R(RSP), Imm8(stacksize));
|
|
#endif
|
|
RET();
|
|
#else
|
|
#error Arch not supported
|
|
#endif
|
|
}
|
|
|
|
#ifdef _M_IX86 // All32
|
|
|
|
// Shared code between Win32 and Unix32
|
|
// ====================================
|
|
|
|
void XEmitter::ABI_CallFunctionC(void *func, u32 param1) {
|
|
ABI_AlignStack(1 * 4);
|
|
PUSH(32, Imm32(param1));
|
|
CALL(func);
|
|
ABI_RestoreStack(1 * 4);
|
|
}
|
|
|
|
void XEmitter::ABI_CallFunctionCC(void *func, u32 param1, u32 param2) {
|
|
ABI_AlignStack(2 * 4);
|
|
PUSH(32, Imm32(param2));
|
|
PUSH(32, Imm32(param1));
|
|
CALL(func);
|
|
ABI_RestoreStack(2 * 4);
|
|
}
|
|
|
|
// Pass a register as a paremeter.
|
|
void XEmitter::ABI_CallFunctionR(void *func, X64Reg reg1) {
|
|
ABI_AlignStack(1 * 4);
|
|
PUSH(32, R(reg1));
|
|
CALL(func);
|
|
ABI_RestoreStack(1 * 4);
|
|
}
|
|
|
|
void XEmitter::ABI_CallFunctionRR(void *func, Gen::X64Reg reg1, Gen::X64Reg reg2)
|
|
{
|
|
ABI_AlignStack(2 * 4);
|
|
PUSH(32, R(reg2));
|
|
PUSH(32, R(reg1));
|
|
CALL(func);
|
|
ABI_RestoreStack(2 * 4);
|
|
}
|
|
|
|
void XEmitter::ABI_CallFunctionAC(void *func, const Gen::OpArg &arg1, u32 param2)
|
|
{
|
|
ABI_AlignStack(2 * 4);
|
|
PUSH(32, arg1);
|
|
PUSH(32, Imm32(param2));
|
|
CALL(func);
|
|
ABI_RestoreStack(2 * 4);
|
|
}
|
|
|
|
void XEmitter::ABI_PushAllCalleeSavedRegsAndAdjustStack() {
|
|
ABI_AlignStack(0);
|
|
// Note: 4 * 4 = 16 bytes, so alignment is preserved.
|
|
PUSH(EBP);
|
|
PUSH(EBX);
|
|
PUSH(ESI);
|
|
PUSH(EDI);
|
|
}
|
|
|
|
void XEmitter::ABI_PopAllCalleeSavedRegsAndAdjustStack() {
|
|
POP(EDI);
|
|
POP(ESI);
|
|
POP(EBX);
|
|
POP(EBP);
|
|
ABI_RestoreStack(0);
|
|
}
|
|
|
|
unsigned int XEmitter::ABI_GetAlignedFrameSize(unsigned int frameSize) {
|
|
frameSize += 4; // reserve space for return address
|
|
unsigned int alignedSize =
|
|
#ifdef __GNUC__
|
|
(frameSize + 15) & -16;
|
|
#else
|
|
frameSize;
|
|
#endif
|
|
return alignedSize;
|
|
}
|
|
|
|
|
|
void XEmitter::ABI_AlignStack(unsigned int frameSize) {
|
|
// Mac OS X requires the stack to be 16-byte aligned before every call.
|
|
// Linux requires the stack to be 16-byte aligned before calls that put SSE
|
|
// vectors on the stack, but since we do not keep track of which calls do that,
|
|
// it is effectively every call as well.
|
|
// Windows binaries compiled with MSVC do not have such a restriction, but I
|
|
// expect that GCC on Windows acts the same as GCC on Linux in this respect.
|
|
// It would be nice if someone could verify this.
|
|
#ifdef __GNUC__
|
|
unsigned int fillSize =
|
|
ABI_GetAlignedFrameSize(frameSize) - (frameSize + 4);
|
|
if (fillSize != 0) {
|
|
SUB(32, R(ESP), Imm8(fillSize));
|
|
}
|
|
#endif
|
|
}
|
|
|
|
void XEmitter::ABI_RestoreStack(unsigned int frameSize) {
|
|
unsigned int alignedSize = ABI_GetAlignedFrameSize(frameSize);
|
|
alignedSize -= 4; // return address is POPped at end of call
|
|
if (alignedSize != 0) {
|
|
ADD(32, R(ESP), Imm8(alignedSize));
|
|
}
|
|
}
|
|
|
|
#else
|
|
|
|
void XEmitter::ABI_CallFunctionC(void *func, u32 param1) {
|
|
MOV(32, R(ABI_PARAM1), Imm32(param1));
|
|
CALL(func);
|
|
}
|
|
|
|
void XEmitter::ABI_CallFunctionCC(void *func, u32 param1, u32 param2) {
|
|
MOV(32, R(ABI_PARAM1), Imm32(param1));
|
|
MOV(32, R(ABI_PARAM2), Imm32(param2));
|
|
CALL(func);
|
|
}
|
|
|
|
// Pass a register as a paremeter.
|
|
void XEmitter::ABI_CallFunctionR(void *func, X64Reg reg1) {
|
|
if (reg1 != ABI_PARAM1)
|
|
MOV(32, R(ABI_PARAM1), R(reg1));
|
|
CALL(func);
|
|
}
|
|
|
|
// Pass a register as a paremeter.
|
|
void XEmitter::ABI_CallFunctionRR(void *func, X64Reg reg1, X64Reg reg2) {
|
|
if (reg1 != ABI_PARAM1)
|
|
MOV(32, R(ABI_PARAM1), R(reg1));
|
|
if (reg2 != ABI_PARAM2)
|
|
MOV(32, R(ABI_PARAM2), R(reg2));
|
|
CALL(func);
|
|
}
|
|
|
|
void XEmitter::ABI_CallFunctionAC(void *func, const Gen::OpArg &arg1, u32 param2)
|
|
{
|
|
if (!arg1.IsSimpleReg(ABI_PARAM1))
|
|
MOV(32, R(ABI_PARAM1), arg1);
|
|
MOV(32, R(ABI_PARAM2), Imm32(param2));
|
|
CALL(func);
|
|
}
|
|
|
|
unsigned int XEmitter::ABI_GetAlignedFrameSize(unsigned int frameSize) {
|
|
return frameSize;
|
|
}
|
|
|
|
void XEmitter::ABI_AlignStack(unsigned int /*frameSize*/) {
|
|
}
|
|
|
|
void XEmitter::ABI_RestoreStack(unsigned int /*frameSize*/) {
|
|
}
|
|
|
|
#ifdef _WIN32
|
|
|
|
// Win64 Specific Code
|
|
// ====================================
|
|
void XEmitter::ABI_PushAllCalleeSavedRegsAndAdjustStack() {
|
|
//we only want to do this once
|
|
PUSH(RBX);
|
|
PUSH(RSI);
|
|
PUSH(RDI);
|
|
PUSH(RBP);
|
|
PUSH(R12);
|
|
PUSH(R13);
|
|
PUSH(R14);
|
|
PUSH(R15);
|
|
//TODO: Also preserve XMM0-3?
|
|
SUB(64, R(RSP), Imm8(0x28));
|
|
}
|
|
|
|
void XEmitter::ABI_PopAllCalleeSavedRegsAndAdjustStack() {
|
|
ADD(64, R(RSP), Imm8(0x28));
|
|
POP(R15);
|
|
POP(R14);
|
|
POP(R13);
|
|
POP(R12);
|
|
POP(RBP);
|
|
POP(RDI);
|
|
POP(RSI);
|
|
POP(RBX);
|
|
}
|
|
|
|
// Win64 Specific Code
|
|
// ====================================
|
|
void XEmitter::ABI_PushAllCallerSavedRegsAndAdjustStack() {
|
|
PUSH(RCX);
|
|
PUSH(RDX);
|
|
PUSH(RSI);
|
|
PUSH(RDI);
|
|
PUSH(R8);
|
|
PUSH(R9);
|
|
PUSH(R10);
|
|
PUSH(R11);
|
|
//TODO: Also preserve XMM0-15?
|
|
SUB(64, R(RSP), Imm8(0x28));
|
|
}
|
|
|
|
void XEmitter::ABI_PopAllCallerSavedRegsAndAdjustStack() {
|
|
ADD(64, R(RSP), Imm8(0x28));
|
|
POP(R11);
|
|
POP(R10);
|
|
POP(R9);
|
|
POP(R8);
|
|
POP(RDI);
|
|
POP(RSI);
|
|
POP(RDX);
|
|
POP(RCX);
|
|
}
|
|
|
|
#else
|
|
// Unix64 Specific Code
|
|
// ====================================
|
|
void XEmitter::ABI_PushAllCalleeSavedRegsAndAdjustStack() {
|
|
PUSH(RBX);
|
|
PUSH(RBP);
|
|
PUSH(R12);
|
|
PUSH(R13);
|
|
PUSH(R14);
|
|
PUSH(R15);
|
|
PUSH(R15); //just to align stack. duped push/pop doesn't hurt.
|
|
}
|
|
|
|
void XEmitter::ABI_PopAllCalleeSavedRegsAndAdjustStack() {
|
|
POP(R15);
|
|
POP(R15);
|
|
POP(R14);
|
|
POP(R13);
|
|
POP(R12);
|
|
POP(RBP);
|
|
POP(RBX);
|
|
}
|
|
|
|
void XEmitter::ABI_PushAllCallerSavedRegsAndAdjustStack() {
|
|
PUSH(RCX);
|
|
PUSH(RDX);
|
|
PUSH(RSI);
|
|
PUSH(RDI);
|
|
PUSH(R8);
|
|
PUSH(R9);
|
|
PUSH(R10);
|
|
PUSH(R11);
|
|
PUSH(R11);
|
|
}
|
|
|
|
void XEmitter::ABI_PopAllCallerSavedRegsAndAdjustStack() {
|
|
POP(R11);
|
|
POP(R11);
|
|
POP(R10);
|
|
POP(R9);
|
|
POP(R8);
|
|
POP(RDI);
|
|
POP(RSI);
|
|
POP(RDX);
|
|
POP(RCX);
|
|
}
|
|
|
|
#endif
|
|
|
|
#endif
|
|
|