tacosukoneko/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

x64: Refactor to remove fake interfaces and general cleanups.

Browse source
This commit is contained in:
bunnei 2015-08-12 00:00:44 -04:00
parent cfb354f11f
commit bd7e691f78
16 changed files with 52 additions and 666 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
src/common/CMakeLists.txt
View file

@ -2,8 +2,8 @@
configure_file("${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp.in" "${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp" @ONLY) configure_file("${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp.in" "${CMAKE_CURRENT_SOURCE_DIR}/scm_rev.cpp" @ONLY)
set(SRCS set(SRCS
abi.cpp
break_points.cpp break_points.cpp
cpu_detect.cpp
emu_window.cpp emu_window.cpp
file_util.cpp file_util.cpp
hash.cpp hash.cpp
@ -22,7 +22,6 @@ set(SRCS
) )
set(HEADERS set(HEADERS
abi.h
assert.h assert.h
bit_field.h bit_field.h
break_points.h break_points.h
@ -61,19 +60,14 @@ set(HEADERS
vector_math.h vector_math.h
) )
if(_M_X86_64) if(ARCHITECTURE_X64)
set(SRCS ${SRCS} set(SRCS ${SRCS}
cpu_detect_x86.cpp x64/abi.cpp
x64_emitter.cpp) x64/emitter.cpp)
set(HEADERS ${HEADERS} set(HEADERS ${HEADERS}
x64_emitter.h) x64/abi.h
else() x64/emitter.h)
set(SRCS ${SRCS}
cpu_detect_generic.cpp)
set(HEADERS ${HEADERS}
fake_emitter.h)
endif() endif()
create_directory_groups(${SRCS} ${HEADERS}) create_directory_groups(${SRCS} ${HEADERS})

2
src/common/common_funcs.h
View file

@ -35,7 +35,7 @@
#ifndef _MSC_VER #ifndef _MSC_VER
#if defined(__x86_64__) || defined(_M_X86_64) #if defined(__x86_64__) || defined(ARCHITECTURE_X64)
#define Crash() __asm__ __volatile__("int $3") #define Crash() __asm__ __volatile__("int $3")
#elif defined(_M_ARM) #elif defined(_M_ARM)
#define Crash() __asm__ __volatile__("trap") #define Crash() __asm__ __volatile__("trap")

2
src/common/cpu_detect_x86.cpp → src/common/cpu_detect.cpp
View file

@ -62,7 +62,7 @@ CPUInfo::CPUInfo() {
// Detects the various CPU features // Detects the various CPU features
void CPUInfo::Detect() { void CPUInfo::Detect() {
memset(this, 0, sizeof(*this)); memset(this, 0, sizeof(*this));
#ifdef _M_X86_64 #ifdef ARCHITECTURE_X64
Mode64bit = true; Mode64bit = true;
OS64bit = true; OS64bit = true;
#endif #endif

19
src/common/cpu_detect_generic.cpp
View file

@ -1,19 +0,0 @@
// Copyright 2014 Dolphin Emulator Project
// Licensed under GPLv2+
// Refer to the license.txt file included.
#include "cpu_detect.h"
#include "hash.h"
namespace Common {
CPUInfo cpu_info;
CPUInfo::CPUInfo() {
}
std::string CPUInfo::Summarize() {
return "Generic";
}
} // namespace Common

465
src/common/fake_emitter.h
View file

@ -1,465 +0,0 @@
// Copyright (C) 2003 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/
// WARNING - THIS LIBRARY IS NOT THREAD SAFE!!!
#pragma once
#include <vector>
#include <stdint.h>
#include "assert.h"
#include "common_types.h"
// TODO: Check if Pandora still needs signal.h/kill here. Symbian doesn't.
// VCVT flags
#define TO_FLOAT 0
#define TO_INT 1 << 0
#define IS_SIGNED 1 << 1
#define ROUND_TO_ZERO 1 << 2
namespace FakeGen
{
enum FakeReg
{
// GPRs
R0 = 0, R1, R2, R3, R4, R5,
R6, R7, R8, R9, R10, R11,
// SPRs
// R13 - R15 are SP, LR, and PC.
// Almost always referred to by name instead of register number
R12 = 12, R13 = 13, R14 = 14, R15 = 15,
R_IP = 12, R_SP = 13, R_LR = 14, R_PC = 15,
// VFP single precision registers
S0, S1, S2, S3, S4, S5, S6,
S7, S8, S9, S10, S11, S12, S13,
S14, S15, S16, S17, S18, S19, S20,
S21, S22, S23, S24, S25, S26, S27,
S28, S29, S30, S31,
// VFP Double Precision registers
D0, D1, D2, D3, D4, D5, D6, D7,
D8, D9, D10, D11, D12, D13, D14, D15,
D16, D17, D18, D19, D20, D21, D22, D23,
D24, D25, D26, D27, D28, D29, D30, D31,
// ASIMD Quad-Word registers
Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7,
Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15,
// for NEON VLD/VST instructions
REG_UPDATE = R13,
INVALID_REG = 0xFFFFFFFF
};
enum CCFlags
{
CC_EQ = 0, // Equal
CC_NEQ, // Not equal
CC_CS, // Carry Set
CC_CC, // Carry Clear
CC_MI, // Minus (Negative)
CC_PL, // Plus
CC_VS, // Overflow
CC_VC, // No Overflow
CC_HI, // Unsigned higher
CC_LS, // Unsigned lower or same
CC_GE, // Signed greater than or equal
CC_LT, // Signed less than
CC_GT, // Signed greater than
CC_LE, // Signed less than or equal
CC_AL, // Always (unconditional) 14
CC_HS = CC_CS, // Alias of CC_CS Unsigned higher or same
CC_LO = CC_CC, // Alias of CC_CC Unsigned lower
};
const u32 NO_COND = 0xE0000000;
enum ShiftType
{
ST_LSL = 0,
ST_ASL = 0,
ST_LSR = 1,
ST_ASR = 2,
ST_ROR = 3,
ST_RRX = 4
};
enum IntegerSize
{
I_I8 = 0,
I_I16,
I_I32,
I_I64
};
enum
{
NUMGPRs = 13,
};
class FakeXEmitter;
enum OpType
{
TYPE_IMM = 0,
TYPE_REG,
TYPE_IMMSREG,
TYPE_RSR,
TYPE_MEM
};
// This is no longer a proper operand2 class. Need to split up.
class Operand2
{
friend class FakeXEmitter;
protected:
u32 Value;
private:
OpType Type;
// IMM types
u8 Rotation; // Only for u8 values
// Register types
u8 IndexOrShift;
ShiftType Shift;
public:
OpType GetType()
{
return Type;
}
Operand2() {}
Operand2(u32 imm, OpType type = TYPE_IMM)
{
Type = type;
Value = imm;
Rotation = 0;
}
Operand2(FakeReg Reg)
{
Type = TYPE_REG;
Value = Reg;
Rotation = 0;
}
Operand2(u8 imm, u8 rotation)
{
Type = TYPE_IMM;
Value = imm;
Rotation = rotation;
}
Operand2(FakeReg base, ShiftType type, FakeReg shift) // RSR
{
Type = TYPE_RSR;
ASSERT_MSG(type != ST_RRX, "Invalid Operand2: RRX does not take a register shift amount");
IndexOrShift = shift;
Shift = type;
Value = base;
}
Operand2(FakeReg base, ShiftType type, u8 shift)// For IMM shifted register
{
if(shift == 32) shift = 0;
switch (type)
{
case ST_LSL:
ASSERT_MSG(shift < 32, "Invalid Operand2: LSL %u", shift);
break;
case ST_LSR:
ASSERT_MSG(shift <= 32, "Invalid Operand2: LSR %u", shift);
if (!shift)
type = ST_LSL;
if (shift == 32)
shift = 0;
break;
case ST_ASR:
ASSERT_MSG(shift < 32, "Invalid Operand2: ASR %u", shift);
if (!shift)
type = ST_LSL;
if (shift == 32)
shift = 0;
break;
case ST_ROR:
ASSERT_MSG(shift < 32, "Invalid Operand2: ROR %u", shift);
if (!shift)
type = ST_LSL;
break;
case ST_RRX:
ASSERT_MSG(shift == 0, "Invalid Operand2: RRX does not take an immediate shift amount");
type = ST_ROR;
break;
}
IndexOrShift = shift;
Shift = type;
Value = base;
Type = TYPE_IMMSREG;
}
u32 GetData()
{
switch(Type)
{
case TYPE_IMM:
return Imm12Mod(); // This'll need to be changed later
case TYPE_REG:
return Rm();
case TYPE_IMMSREG:
return IMMSR();
case TYPE_RSR:
return RSR();
default:
ASSERT_MSG(false, "GetData with Invalid Type");
return 0;
}
}
u32 IMMSR() // IMM shifted register
{
ASSERT_MSG(Type == TYPE_IMMSREG, "IMMSR must be imm shifted register");
return ((IndexOrShift & 0x1f) << 7 | (Shift << 5) | Value);
}
u32 RSR() // Register shifted register
{
ASSERT_MSG(Type == TYPE_RSR, "RSR must be RSR Of Course");
return (IndexOrShift << 8) | (Shift << 5) | 0x10 | Value;
}
u32 Rm()
{
ASSERT_MSG(Type == TYPE_REG, "Rm must be with Reg");
return Value;
}
u32 Imm5()
{
ASSERT_MSG((Type == TYPE_IMM), "Imm5 not IMM value");
return ((Value & 0x0000001F) << 7);
}
u32 Imm8()
{
ASSERT_MSG((Type == TYPE_IMM), "Imm8Rot not IMM value");
return Value & 0xFF;
}
u32 Imm8Rot() // IMM8 with Rotation
{
ASSERT_MSG((Type == TYPE_IMM), "Imm8Rot not IMM value");
ASSERT_MSG((Rotation & 0xE1) != 0, "Invalid Operand2: immediate rotation %u", Rotation);
return (1 << 25) | (Rotation << 7) | (Value & 0x000000FF);
}
u32 Imm12()
{
ASSERT_MSG((Type == TYPE_IMM), "Imm12 not IMM");
return (Value & 0x00000FFF);
}
u32 Imm12Mod()
{
// This is an IMM12 with the top four bits being rotation and the
// bottom eight being an IMM. This is for instructions that need to
// expand a 8bit IMM to a 32bit value and gives you some rotation as
// well.
// Each rotation rotates to the right by 2 bits
ASSERT_MSG((Type == TYPE_IMM), "Imm12Mod not IMM");
return ((Rotation & 0xF) << 8) | (Value & 0xFF);
}
u32 Imm16()
{
ASSERT_MSG((Type == TYPE_IMM), "Imm16 not IMM");
return ( (Value & 0xF000) << 4) | (Value & 0x0FFF);
}
u32 Imm16Low()
{
return Imm16();
}
u32 Imm16High() // Returns high 16bits
{
ASSERT_MSG((Type == TYPE_IMM), "Imm16 not IMM");
return ( ((Value >> 16) & 0xF000) << 4) | ((Value >> 16) & 0x0FFF);
}
u32 Imm24()
{
ASSERT_MSG((Type == TYPE_IMM), "Imm16 not IMM");
return (Value & 0x0FFFFFFF);
}
};
// Use these when you don't know if an imm can be represented as an operand2.
// This lets you generate both an optimal and a fallback solution by checking
// the return value, which will be false if these fail to find a Operand2 that
// represents your 32-bit imm value.
bool TryMakeOperand2(u32 imm, Operand2 &op2);
bool TryMakeOperand2_AllowInverse(u32 imm, Operand2 &op2, bool *inverse);
bool TryMakeOperand2_AllowNegation(s32 imm, Operand2 &op2, bool *negated);
// Use this only when you know imm can be made into an Operand2.
Operand2 AssumeMakeOperand2(u32 imm);
inline Operand2 R(FakeReg Reg) { return Operand2(Reg, TYPE_REG); }
inline Operand2 IMM(u32 Imm) { return Operand2(Imm, TYPE_IMM); }
inline Operand2 Mem(void *ptr) { return Operand2((u32)(uintptr_t)ptr, TYPE_IMM); }
//usage: struct {int e;} s; STRUCT_OFFSET(s,e)
#define STRUCT_OFF(str,elem) ((u32)((u32)&(str).elem-(u32)&(str)))
struct FixupBranch
{
u8 *ptr;
u32 condition; // Remembers our codition at the time
int type; //0 = B 1 = BL
};
typedef const u8* JumpTarget;
// XXX: Stop polluting the global namespace
const u32 I_8 = (1 << 0);
const u32 I_16 = (1 << 1);
const u32 I_32 = (1 << 2);
const u32 I_64 = (1 << 3);
const u32 I_SIGNED = (1 << 4);
const u32 I_UNSIGNED = (1 << 5);
const u32 F_32 = (1 << 6);
const u32 I_POLYNOMIAL = (1 << 7); // Only used in VMUL/VMULL
u32 EncodeVd(FakeReg Vd);
u32 EncodeVn(FakeReg Vn);
u32 EncodeVm(FakeReg Vm);
u32 encodedSize(u32 value);
// Subtracts the base from the register to give us the real one
FakeReg SubBase(FakeReg Reg);
// See A.7.1 in the Fakev7-A
// VMUL F32 scalars can only be up to D15[0], D15[1] - higher scalars cannot be individually addressed
FakeReg DScalar(FakeReg dreg, int subScalar);
FakeReg QScalar(FakeReg qreg, int subScalar);
enum NEONAlignment {
ALIGN_NONE = 0,
ALIGN_64 = 1,
ALIGN_128 = 2,
ALIGN_256 = 3
};
class NEONXEmitter;
class FakeXEmitter
{
friend struct OpArg; // for Write8 etc
private:
u8 *code, *startcode;
u8 *lastCacheFlushEnd;
u32 condition;
protected:
inline void Write32(u32 value) {*(u32*)code = value; code+=4;}
public:
FakeXEmitter() : code(0), startcode(0), lastCacheFlushEnd(0) {
condition = CC_AL << 28;
}
FakeXEmitter(u8 *code_ptr) {
code = code_ptr;
lastCacheFlushEnd = code_ptr;
startcode = code_ptr;
condition = CC_AL << 28;
}
virtual ~FakeXEmitter() {}
void SetCodePtr(u8 *ptr) {}
void ReserveCodeSpace(u32 bytes) {}
const u8 *AlignCode16() { return nullptr; }
const u8 *AlignCodePage() { return nullptr; }
const u8 *GetCodePtr() const { return nullptr; }
void FlushIcache() {}
void FlushIcacheSection(u8 *start, u8 *end) {}
u8 *GetWritableCodePtr() { return nullptr; }
CCFlags GetCC() { return CCFlags(condition >> 28); }
void SetCC(CCFlags cond = CC_AL) {}
// Special purpose instructions
// Do nothing
void NOP(int count = 1) {} //nop padding - TODO: fast nop slides, for amd and intel (check their manuals)
#ifdef CALL
#undef CALL
#endif
void QuickCallFunction(FakeReg scratchreg, const void *func);
template <typename T> void QuickCallFunction(FakeReg scratchreg, T func) {
QuickCallFunction(scratchreg, (const void *)func);
}
}; // class FakeXEmitter
// Everything that needs to generate machine code should inherit from this.
// You get memory management for free, plus, you can use all the MOV etc functions without
// having to prefix them with gen-> or something similar.
class FakeXCodeBlock : public FakeXEmitter
{
protected:
u8 *region;
size_t region_size;
public:
FakeXCodeBlock() : region(NULL), region_size(0) {}
virtual ~FakeXCodeBlock() { if (region) FreeCodeSpace(); }
// Call this before you generate any code.
void AllocCodeSpace(int size) { }
// Always clear code space with breakpoints, so that if someone accidentally executes
// uninitialized, it just breaks into the debugger.
void ClearCodeSpace() { }
// Call this when shutting down. Don't rely on the destructor, even though it'll do the job.
void FreeCodeSpace() { }
bool IsInSpace(const u8 *ptr) const
{
return ptr >= region && ptr < region + region_size;
}
// Cannot currently be undone. Will write protect the entire code region.
// Start over if you need to change the code (call FreeCodeSpace(), AllocCodeSpace()).
void WriteProtect() { }
void UnWriteProtect() { }
void ResetCodePtr()
{
SetCodePtr(region);
}
size_t GetSpaceLeft() const
{
return region_size - (GetCodePtr() - region);
}
u8 *GetBasePtr() {
return region;
}
size_t GetOffset(const u8 *ptr) const {
return ptr - region;
}
};
} // namespace

2
src/common/platform.h
View file

@ -27,7 +27,7 @@
//////////////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////////////
// Platform detection // Platform detection
#if defined(__x86_64__) || defined(_M_X86_64) || defined(__aarch64__) #if defined(__x86_64__) || defined(ARCHITECTURE_X64) || defined(__aarch64__)
#define EMU_ARCH_BITS 64 #define EMU_ARCH_BITS 64
#elif defined(__i386) || defined(_M_IX86) || defined(__arm__) || defined(_M_ARM) #elif defined(__i386) || defined(_M_IX86) || defined(__arm__) || defined(_M_ARM)
#define EMU_ARCH_BITS 32 #define EMU_ARCH_BITS 32

6
src/common/abi.cpp → src/common/x64/abi.cpp
View file

@ -15,8 +15,8 @@
// Official SVN repository and contact information can be found at // Official SVN repository and contact information can be found at
// http://code.google.com/p/dolphin-emu/ // http://code.google.com/p/dolphin-emu/
#include "x64_emitter.h"
#include "abi.h" #include "abi.h"
#include "emitter.h"
using namespace Gen; using namespace Gen;
@ -27,7 +27,7 @@ void XEmitter::ABI_EmitPrologue(int maxCallParams)
{ {
#ifdef _M_IX86 #ifdef _M_IX86
// Don't really need to do anything // Don't really need to do anything
#elif defined(_M_X86_64) #elif defined(ARCHITECTURE_X64)
#if _WIN32 #if _WIN32
int stacksize = ((maxCallParams + 1) & ~1) * 8 + 8; int stacksize = ((maxCallParams + 1) & ~1) * 8 + 8;
// Set up a stack frame so that we can call functions // Set up a stack frame so that we can call functions
@ -43,7 +43,7 @@ void XEmitter::ABI_EmitEpilogue(int maxCallParams)
{ {
#ifdef _M_IX86 #ifdef _M_IX86
RET(); RET();
#elif defined(_M_X86_64) #elif defined(ARCHITECTURE_X64)
#ifdef _WIN32 #ifdef _WIN32
int stacksize = ((maxCallParams+1)&~1)*8 + 8; int stacksize = ((maxCallParams+1)&~1)*8 + 8;
ADD(64, R(RSP), Imm8(stacksize)); ADD(64, R(RSP), Imm8(stacksize));

4
src/common/abi.h → src/common/x64/abi.h
View file

@ -17,7 +17,7 @@
#pragma once #pragma once
#include "common_types.h" #include "common/common_types.h"
// x86/x64 ABI:s, and helpers to help follow them when JIT-ing code. // x86/x64 ABI:s, and helpers to help follow them when JIT-ing code.
// All convensions return values in EAX (+ possibly EDX). // All convensions return values in EAX (+ possibly EDX).
@ -55,7 +55,7 @@
// 32-bit bog standard cdecl, shared between linux and windows // 32-bit bog standard cdecl, shared between linux and windows
// MacOSX 32-bit is same as System V with a few exceptions that we probably don't care much about. // MacOSX 32-bit is same as System V with a few exceptions that we probably don't care much about.
#elif _M_X86_64 // 64 bit calling convention #elif ARCHITECTURE_X64 // 64 bit calling convention
#ifdef _WIN32 // 64-bit Windows - the really exotic calling convention #ifdef _WIN32 // 64-bit Windows - the really exotic calling convention

16
src/common/x64_emitter.cpp → src/common/x64/emitter.cpp
View file

@ -17,13 +17,13 @@
#include <cstring> #include <cstring>
#include "logging/log.h" #include "common/assert.h"
#include "common/cpu_detect.h"
#include "common/logging/log.h"
#include "common/memory_util.h"
#include "assert.h"
#include "x64_emitter.h"
#include "abi.h" #include "abi.h"
#include "cpu_detect.h" #include "emitter.h"
#include "memory_util.h"
#define PRIx64 "llx" #define PRIx64 "llx"
@ -164,7 +164,7 @@ void XEmitter::WriteSIB(int scale, int index, int base)
void OpArg::WriteRex(XEmitter *emit, int opBits, int bits, int customOp) const void OpArg::WriteRex(XEmitter *emit, int opBits, int bits, int customOp) const
{ {
if (customOp == -1) customOp = operandReg; if (customOp == -1) customOp = operandReg;
#ifdef _M_X86_64 #ifdef ARCHITECTURE_X64
u8 op = 0x40; u8 op = 0x40;
// REX.W (whether operation is a 64-bit operation) // REX.W (whether operation is a 64-bit operation)
if (opBits == 64) op |= 8; if (opBits == 64) op |= 8;
@ -236,7 +236,7 @@ void OpArg::WriteRest(XEmitter *emit, int extraBytes, X64Reg _operandReg,
_offsetOrBaseReg = 5; _offsetOrBaseReg = 5;
emit->WriteModRM(0, _operandReg, _offsetOrBaseReg); emit->WriteModRM(0, _operandReg, _offsetOrBaseReg);
//TODO : add some checks //TODO : add some checks
#ifdef _M_X86_64 #ifdef ARCHITECTURE_X64
u64 ripAddr = (u64)emit->GetCodePtr() + 4 + extraBytes; u64 ripAddr = (u64)emit->GetCodePtr() + 4 + extraBytes;
s64 distance = (s64)offset - (s64)ripAddr; s64 distance = (s64)offset - (s64)ripAddr;
ASSERT_MSG( ASSERT_MSG(
@ -1463,7 +1463,7 @@ void XEmitter::MOVD_xmm(const OpArg &arg, X64Reg src) {WriteSSEOp(0x66, 0x7E, sr
void XEmitter::MOVQ_xmm(X64Reg dest, OpArg arg) void XEmitter::MOVQ_xmm(X64Reg dest, OpArg arg)
{ {
#ifdef _M_X86_64 #ifdef ARCHITECTURE_X64
// Alternate encoding // Alternate encoding
// This does not display correctly in MSVC's debugger, it thinks it's a MOVD // This does not display correctly in MSVC's debugger, it thinks it's a MOVD
arg.operandReg = dest; arg.operandReg = dest;

8
src/common/x64_emitter.h → src/common/x64/emitter.h
View file

@ -17,11 +17,11 @@
#pragma once #pragma once
#include "assert.h" #include "common/assert.h"
#include "common_types.h" #include "common/common_types.h"
#include "code_block.h" #include "common/code_block.h"
#if defined(_M_X86_64) && !defined(_ARCH_64) #if defined(ARCHITECTURE_X64) && !defined(_ARCH_64)
#define _ARCH_64 #define _ARCH_64
#endif #endif

10
src/video_core/CMakeLists.txt
View file

@ -13,7 +13,6 @@ set(SRCS
rasterizer.cpp rasterizer.cpp
shader/shader.cpp shader/shader.cpp
shader/shader_interpreter.cpp shader/shader_interpreter.cpp
shader/shader_jit.cpp
utils.cpp utils.cpp
video_core.cpp video_core.cpp
) )
@ -39,17 +38,16 @@ set(HEADERS
renderer_base.h renderer_base.h
shader/shader.h shader/shader.h
shader/shader_interpreter.h shader/shader_interpreter.h
shader/shader_jit.h
utils.h utils.h
video_core.h video_core.h
) )
if(_M_X86_64) if(ARCHITECTURE_X64)
set(SRCS ${SRCS} set(SRCS ${SRCS}
shader/shader_jit_x64.cpp) shader/shader_jit_x64.cpp)
else()
set(SRCS ${SRCS} set(HEADERS ${HEADERS}
shader/shader_jit_fake.cpp) shader/shader_jit_x64.h)
endif() endif()
create_directory_groups(${SRCS} ${HEADERS}) create_directory_groups(${SRCS} ${HEADERS})

9
src/video_core/shader/shader.cpp
View file

@ -15,7 +15,10 @@
#include "shader.h" #include "shader.h"
#include "shader_interpreter.h" #include "shader_interpreter.h"
#include "shader_jit.h"
#ifdef ARCHITECTURE_X64
#include "shader_jit_x64.h"
#endif // ARCHITECTURE_X64
namespace Pica { namespace Pica {
@ -43,7 +46,7 @@ void Setup(UnitState& state) {
jit_shader = jit.Compile(); jit_shader = jit.Compile();
shader_map.emplace(cache_key, jit_shader); shader_map.emplace(cache_key, jit_shader);
} }
} #endif // ARCHITECTURE_X64
} }
void Shutdown() { void Shutdown() {
@ -92,7 +95,7 @@ OutputVertex Run(UnitState& state, const InputVertex& input, int num_attributes)
RunInterpreter(state); RunInterpreter(state);
#else #else
RunInterpreter(state); RunInterpreter(state);
#endif #endif // ARCHITECTURE_X64
#if PICA_DUMP_SHADERS #if PICA_DUMP_SHADERS
DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(), DebugUtils::DumpShader(setup.program_code.data(), state.debug.max_offset, setup.swizzle_data.data(),

36
src/video_core/shader/shader_jit.cpp
View file

@ -1,36 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "video_core/pica.h"
#include "shader.h"
#include "shader_jit.h"
namespace Pica {
namespace Shader {
JitShader::JitShader() : jitted(nullptr) {
}
void JitShader::DoJit(JitCompiler& jit) {
jitted = jit.Compile();
}
void JitShader::Run(UnitState& state) {
if (jitted)
jitted(&state);
}
JitCompiler::JitCompiler() {
AllocCodeSpace(1024 * 1024 * 4);
}
void JitCompiler::Clear() {
ClearCodeSpace();
}
} // namespace Shader
} // namespace Pica

91
src/video_core/shader/shader_jit_fake.cpp
View file

@ -1,91 +0,0 @@
// Copyright 2015 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/fake_emitter.h"
#include "video_core/shader/shader.h"
#include "video_core/shader/shader_jit.h"
namespace Pica {
namespace Shader {
using namespace FakeGen;
void Jit::Comp_ADD(Instruction instr) {
}
void Jit::Comp_DP3(Instruction instr) {
}
void Jit::Comp_DP4(Instruction instr) {
}
void Jit::Comp_MUL(Instruction instr) {
}
void Jit::Comp_FLR(Instruction instr) {
}
void Jit::Comp_MAX(Instruction instr) {
}
void Jit::Comp_MIN(Instruction instr) {
}
void Jit::Comp_MOVA(Instruction instr) {
}
void Jit::Comp_MOV(Instruction instr) {
}
void Jit::Comp_SLTI(Instruction instr) {
}
void Jit::Comp_RCP(Instruction instr) {
}
void Jit::Comp_RSQ(Instruction instr) {
}
void Jit::Comp_NOP(Instruction instr) {
}
void Jit::Comp_END(Instruction instr) {
}
void Jit::Comp_CALL(Instruction instr) {
}
void Jit::Comp_CALLC(Instruction instr) {
}
void Jit::Comp_CALLU(Instruction instr) {
}
void Jit::Comp_CMP(Instruction instr) {
}
void Jit::Comp_MAD(Instruction instr) {
}
void Jit::Comp_IF(Instruction instr) {
}
void Jit::Comp_LOOP(Instruction instr) {
}
void Jit::Comp_JMP(Instruction instr) {
}
void Jit::Comp_NextInstr(unsigned* offset) {
}
CompiledShader Jit::Compile() {
return nullptr;
}
} // namespace Shader
} // namespace Pica

20
src/video_core/shader/shader_jit_x64.cpp
View file

@ -4,12 +4,13 @@
#include <smmintrin.h> #include <smmintrin.h>
#include "common/abi.h"
#include "common/cpu_detect.h" #include "common/cpu_detect.h"
#include "common/x64_emitter.h"
#include "common/x64/abi.h"
#include "common/x64/emitter.h"
#include "shader.h" #include "shader.h"
#include "shader_jit.h" #include "shader_jit_x64.h"
namespace Pica { namespace Pica {
@ -134,7 +135,7 @@ static const u8 NO_DEST_REG_MASK = 0xf;
*/ */
void JitCompiler::Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg, X64Reg dest) { void JitCompiler::Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg, X64Reg dest) {
X64Reg src_ptr; X64Reg src_ptr;
std::size_t src_offset; int src_offset;
if (src_reg.GetRegisterType() == RegisterType::FloatUniform) { if (src_reg.GetRegisterType() == RegisterType::FloatUniform) {
src_ptr = UNIFORMS; src_ptr = UNIFORMS;
@ -451,7 +452,6 @@ void JitCompiler::Compile_NOP(Instruction instr) {
void JitCompiler::Compile_END(Instruction instr) { void JitCompiler::Compile_END(Instruction instr) {
ABI_PopAllCalleeSavedRegsAndAdjustStack(); ABI_PopAllCalleeSavedRegsAndAdjustStack();
RET(); RET();
done = true;
} }
void JitCompiler::Compile_CALL(Instruction instr) { void JitCompiler::Compile_CALL(Instruction instr) {
@ -655,7 +655,7 @@ CompiledShader* JitCompiler::Compile() {
MOVAPS(NEGBIT, MDisp(RAX, 0)); MOVAPS(NEGBIT, MDisp(RAX, 0));
looping = false; looping = false;
done = false;
while (offset < g_state.vs.program_code.size()) { while (offset < g_state.vs.program_code.size()) {
Compile_NextInstr(&offset); Compile_NextInstr(&offset);
} }
@ -663,6 +663,14 @@ CompiledShader* JitCompiler::Compile() {
return (CompiledShader*)start; return (CompiledShader*)start;
} }
JitCompiler::JitCompiler() {
AllocCodeSpace(1024 * 1024 * 4);
}
void JitCompiler::Clear() {
ClearCodeSpace();
}
} // namespace Shader } // namespace Shader
} // namespace Pica } // namespace Pica

10
src/video_core/shader/shader_jit.h → src/video_core/shader/shader_jit_x64.h
View file

@ -6,11 +6,7 @@
#include <nihstro/shader_bytecode.h> #include <nihstro/shader_bytecode.h>
#if defined(_M_X86_64) #include "common/x64/emitter.h"
#include "common/x64_emitter.h"
#else
#include "common/fake_emitter.h"
#endif
#include "video_core/pica.h" #include "video_core/pica.h"
@ -65,18 +61,16 @@ private:
void Compile_Block(unsigned stop); void Compile_Block(unsigned stop);
void Compile_NextInstr(unsigned* offset); void Compile_NextInstr(unsigned* offset);
#if defined(_M_X86_64)
void Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg, Gen::X64Reg dest); void Compile_SwizzleSrc(Instruction instr, unsigned src_num, SourceRegister src_reg, Gen::X64Reg dest);
void Compile_DestEnable(Instruction instr, Gen::X64Reg dest); void Compile_DestEnable(Instruction instr, Gen::X64Reg dest);
void Compile_EvaluateCondition(Instruction instr); void Compile_EvaluateCondition(Instruction instr);
void Compile_UniformCondition(Instruction instr); void Compile_UniformCondition(Instruction instr);
#endif
/// Pointer to the variable that stores the current Pica code offset. Used to handle nested code blocks. /// Pointer to the variable that stores the current Pica code offset. Used to handle nested code blocks.
unsigned* offset_ptr = nullptr; unsigned* offset_ptr = nullptr;
bool done = false; /// Set to true if currently in a loop, used to check for the existence of nested loops
bool looping = false; bool looping = false;
}; };