OpenPunk/citra
1
0
Fork 0
mirror of https://github.com/citra-emu/citra.git synced 2024-11-25 05:50:16 +00:00
Code Issues Projects Releases Wiki Activity

JitX64: Implement load and store multiple

Browse Source
This commit is contained in:
MerryMage 2016-04-02 12:55:41 +01:00
parent 67553cf56c
commit d4fc4e209a
8 changed files with 276 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/core/arm/decoder/arm.cpp
View File

@ -114,7 +114,7 @@ static std::unique_ptr<Matcher> MakeMatcher(const char format[32], Function fn)
return std::unique_ptr<Matcher>(std::move(ret));
}
static const std::array<Instruction, 230> arm_instruction_table = {{
static const std::array<Instruction, 220> arm_instruction_table = {{
// Branch instructions
{ "BLX (immediate)", MakeMatcher<2>("1111101hvvvvvvvvvvvvvvvvvvvvvvvv", &Visitor::BLX_imm) }, // ARMv5
{ "BLX (register)", MakeMatcher<2>("cccc000100101111111111110011mmmm", &Visitor::BLX_reg) }, // ARMv5
@ -266,21 +266,11 @@ static const std::array<Instruction, 230> arm_instruction_table = {{
{ "STRT (A2)", MakeMatcher<0>("----0110-010---------------0----", &Visitor::STRT) },
// Load/Store Multiple instructions
{ "LDMIA/LDMFD", MakeMatcher<0>("----100010-1--------------------", &Visitor::LDM) }, // all
{ "LDMDA/LDMFA", MakeMatcher<0>("----100000-1--------------------", &Visitor::LDM) }, // all
{ "LDMDB/LDMEA", MakeMatcher<0>("----100100-1--------------------", &Visitor::LDM) }, // all
{ "LDMIB/LDMED", MakeMatcher<0>("----100110-1--------------------", &Visitor::LDM) }, // all
{ "LDM (exc ret)", MakeMatcher<0>("----100--1-1----1---------------", &Visitor::LDM) }, // all
{ "LDM (usr reg)", MakeMatcher<0>("----100--1-1----0---------------", &Visitor::LDM) }, // all
{ "POP", MakeMatcher<0>("----100010111101----------------", &Visitor::LDM) }, // all
{ "POP", MakeMatcher<0>("----010010011101----000000000100", &Visitor::LDM) }, // all
{ "PUSH", MakeMatcher<0>("----100100101101----------------", &Visitor::STM) }, // all
{ "PUSH", MakeMatcher<0>("----010100101101----000000000100", &Visitor::STM) }, // all
{ "STMIA/STMEA", MakeMatcher<0>("----100010-0--------------------", &Visitor::STM) }, // all
{ "STMDA/STMED", MakeMatcher<0>("----100000-0--------------------", &Visitor::STM) }, // all
{ "STMDB/STMFD", MakeMatcher<0>("----100100-0--------------------", &Visitor::STM) }, // all
{ "STMIB/STMFA", MakeMatcher<0>("----100110-0--------------------", &Visitor::STM) }, // all
{ "STMIB (usr reg)", MakeMatcher<0>("----100--100--------------------", &Visitor::STM) }, // all
{ "LDM", MakeMatcher<6>("cccc100pu0w1nnnnxxxxxxxxxxxxxxxx", &Visitor::LDM) }, // all
{ "LDM (usr reg)", MakeMatcher<0>("----100--101--------------------", &Visitor::LDM_usr) }, // all
{ "LDM (exce ret)", MakeMatcher<0>("----100--1-1----1---------------", &Visitor::LDM_eret) }, // all
{ "STM", MakeMatcher<6>("cccc100pu0w0nnnnxxxxxxxxxxxxxxxx", &Visitor::STM) }, // all
{ "STM (usr reg)", MakeMatcher<0>("----100--100--------------------", &Visitor::STM_usr) }, // all
// Miscellaneous instructions
{ "CLZ", MakeMatcher<0>("----000101101111----11110001----", &Visitor::CLZ) }, // ARMv5

8
src/core/arm/decoder/decoder.h
View File

@ -58,6 +58,7 @@ using Imm11 = u32;
using Imm12 = u32;
using Imm24 = u32;
using Register = int;
using RegisterList = u16;
using ShiftType = int;
class Visitor {
@ -188,8 +189,11 @@ public:
virtual void STRT() = 0;
// Load/Store multiple instructions
virtual void LDM() = 0;
virtual void STM() = 0;
virtual void LDM(Cond cond, bool P, bool U, bool W, Register Rn, RegisterList list) = 0;
virtual void LDM_usr() = 0;
virtual void LDM_eret() = 0;
virtual void STM(Cond cond, bool P, bool U, bool W, Register Rn, RegisterList list) = 0;
virtual void STM_usr() = 0;
// Miscellaneous instructions
virtual void CLZ() = 0;

11
src/core/arm/decoder/thumb.cpp
View File

@ -352,10 +352,12 @@ static const std::array<Instruction, 27> thumb_instruction_table = { {
u32 reglist = bits<0, 7>(instruction);
if (!L) { // PUSH {reglist, <R>=LR}
reglist |= R << 14;
v->STM();
// Equivalent to STMDB SP!, {reglist}
v->STM(0xE, /*P=*/1, /*U=*/0, /*W=*/1, 13, reglist);
} else { // POP {reglist, <R>=PC}
reglist |= R << 15;
v->LDM();
// Equivalent to LDMIA SP!, {reglist}
v->LDM(0xE, /*P=*/0, /*U=*/1, /*W=*/1, 13, reglist);
}
})},
{ "SETEND", MakeMatcher("101101100101x000", [](Visitor* v, u32 instruction) {
@ -400,9 +402,10 @@ static const std::array<Instruction, 27> thumb_instruction_table = { {
Register Rn = bits<8, 10>(instruction);
u32 reglist = bits<0, 7>(instruction);
if (!L) { // STMIA Rn!, { reglist }
v->STM();
v->STM(0xE, /*P=*/0, /*U=*/1, /*W=*/1, Rn, reglist);
} else { // LDMIA Rn!, { reglist }
v->LDM();
bool w = reglist & (1 << Rn);
v->LDM(0xE, /*P=*/0, /*U=*/1, /*W=*/w, Rn, reglist);
}
})},
{ "B<cond>", MakeMatcher("1101xxxxxxxxxxxx", [](Visitor* v, u32 instruction) {

1
src/core/arm/jit_x64/common.h
View File

@ -12,6 +12,7 @@
namespace JitX64 {
using ArmReg = ArmDecoder::Register;
using ArmRegList = ArmDecoder::RegisterList;
using ArmImm4 = ArmDecoder::Imm4;
using ArmImm5 = ArmDecoder::Imm5;
using ArmImm8 = ArmDecoder::Imm8;

216
src/core/arm/jit_x64/instructions/load_store.cpp
View File

@ -2,6 +2,7 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/bit_set.h"
#include "common/swap.h"
#include "common/x64/abi.h"
@ -800,7 +801,218 @@ void JitX64::STRHT() { CompileInterpretInstruction(); }
void JitX64::STRT() { CompileInterpretInstruction(); }
// Load/Store multiple instructions
void JitX64::LDM() { CompileInterpretInstruction(); }
void JitX64::STM() { CompileInterpretInstruction(); }
static void LoadAndStoreMultiple_IncrementAfter(XEmitter* code, RegAlloc& reg_alloc, bool W, ArmReg Rn_index, ArmRegList list, std::function<void()> call) {
if (W) {
X64Reg Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->MOV(32, R(ABI_PARAM1), R(Rn));
reg_alloc.UnlockArm(Rn_index);
call();
Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->ADD(32, R(Rn), Imm8(4 * Common::CountSetBits(list)));
reg_alloc.UnlockArm(Rn_index);
} else {
OpArg Rn = reg_alloc.LockArmForReadWrite(Rn_index);
code->MOV(32, R(ABI_PARAM1), Rn);
reg_alloc.UnlockArm(Rn_index);
call();
}
}
static void LoadAndStoreMultiple_IncrementBefore(XEmitter* code, RegAlloc& reg_alloc, bool W, ArmReg Rn_index, ArmRegList list, std::function<void()> call) {
if (W) {
X64Reg Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->MOV(32, R(ABI_PARAM1), R(Rn));
code->ADD(32, R(ABI_PARAM1), Imm8(4));
reg_alloc.UnlockArm(Rn_index);
call();
Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->ADD(32, R(Rn), Imm8(4 * Common::CountSetBits(list)));
reg_alloc.UnlockArm(Rn_index);
} else {
OpArg Rn = reg_alloc.LockArmForReadWrite(Rn_index);
code->MOV(32, R(ABI_PARAM1), Rn);
code->ADD(32, R(ABI_PARAM1), Imm8(4));
reg_alloc.UnlockArm(Rn_index);
call();
}
}
static void LoadAndStoreMultiple_DecrementAfter(XEmitter* code, RegAlloc& reg_alloc, bool W, ArmReg Rn_index, ArmRegList list, std::function<void()> call) {
if (W) {
X64Reg Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->MOV(32, R(ABI_PARAM1), R(Rn));
code->SUB(32, R(ABI_PARAM1), Imm32(4 * Common::CountSetBits(list) - 4));
reg_alloc.UnlockArm(Rn_index);
call();
Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->SUB(32, R(Rn), Imm32(4 * Common::CountSetBits(list)));
reg_alloc.UnlockArm(Rn_index);
} else {
OpArg Rn = reg_alloc.LockArmForReadWrite(Rn_index);
code->MOV(32, R(ABI_PARAM1), Rn);
code->SUB(32, R(ABI_PARAM1), Imm32(4 * Common::CountSetBits(list) - 4));
reg_alloc.UnlockArm(Rn_index);
call();
}
}
static void LoadAndStoreMultiple_DecrementBefore(XEmitter* code, RegAlloc& reg_alloc, bool W, ArmReg Rn_index, ArmRegList list, std::function<void()> call) {
if (W && (list & (1 << Rn_index))) {
X64Reg Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->SUB(32, R(Rn), Imm32(4 * Common::CountSetBits(list)));
code->MOV(32, R(ABI_PARAM1), R(Rn));
reg_alloc.UnlockArm(Rn_index);
call();
} else if (W && (list & (1 << Rn_index))) {
X64Reg Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->MOV(32, R(ABI_PARAM1), R(Rn));
code->SUB(32, R(ABI_PARAM1), Imm32(4 * Common::CountSetBits(list)));
reg_alloc.UnlockArm(Rn_index);
call();
Rn = reg_alloc.BindArmForReadWrite(Rn_index);
code->SUB(32, R(Rn), Imm32(4 * Common::CountSetBits(list)));
reg_alloc.UnlockArm(Rn_index);
} else {
OpArg Rn = reg_alloc.LockArmForReadWrite(Rn_index);
code->MOV(32, R(ABI_PARAM1), Rn);
code->SUB(32, R(ABI_PARAM1), Imm32(4 * Common::CountSetBits(list)));
reg_alloc.UnlockArm(Rn_index);
call();
}
}
static void LoadAndStoreMultiple_Helper(XEmitter* code, RegAlloc& reg_alloc, bool P, bool U, bool W, ArmReg Rn_index, ArmRegList list, std::function<void()> call) {
reg_alloc.FlushX64(ABI_PARAM1);
reg_alloc.LockX64(ABI_PARAM1);
reg_alloc.FlushX64(ABI_PARAM2);
reg_alloc.LockX64(ABI_PARAM2);
reg_alloc.FlushX64(ABI_PARAM3);
reg_alloc.LockX64(ABI_PARAM3);
code->MOV(32, R(ABI_PARAM2), Imm32(list));
code->MOV(64, R(ABI_PARAM3), R(reg_alloc.JitStateReg()));
if (!P && !U) {
LoadAndStoreMultiple_DecrementAfter(code, reg_alloc, W, Rn_index, list, call);
} else if (!P && U) {
LoadAndStoreMultiple_IncrementAfter(code, reg_alloc, W, Rn_index, list, call);
} else if (P && !U) {
LoadAndStoreMultiple_DecrementBefore(code, reg_alloc, W, Rn_index, list, call);
} else if (P && U) {
LoadAndStoreMultiple_IncrementBefore(code, reg_alloc, W, Rn_index, list, call);
} else {
UNREACHABLE();
}
reg_alloc.UnlockX64(ABI_PARAM1);
reg_alloc.UnlockX64(ABI_PARAM2);
reg_alloc.UnlockX64(ABI_PARAM3);
}
static void ExecuteLDMLE(u32 start_address, u16 reg_list, JitState* jit_state) {
for (int i = 0; i < 16; i++) {
const u16 bit = 1 << i;
if (reg_list & bit) {
jit_state->cpu_state.Reg[i] = Memory::Read32(start_address);
start_address += 4;
}
}
}
static void ExecuteLDMBE(u32 start_address, u16 reg_list, JitState* jit_state) {
for (int i = 0; i < 16; i++) {
const u16 bit = 1 << i;
if (reg_list & bit) {
jit_state->cpu_state.Reg[i] = Common::swap32(Memory::Read32(start_address));
start_address += 4;
}
}
}
void JitX64::LDM(Cond cond, bool P, bool U, bool W, ArmReg Rn_index, ArmRegList list) {
cond_manager.CompileCond((ConditionCode)cond);
ASSERT_MSG(Rn_index != 15, "UNPREDICTABLE");
ASSERT_MSG(list != 0, "UNPREDICTABLE");
if (W && (list & (1 << Rn_index)))
ASSERT_MSG(false, "UNPREDICTABLE");
// TODO: Optimize
LoadAndStoreMultiple_Helper(code, reg_alloc, P, U, W, Rn_index, list,
[this](){ CompileCallHost(!current.EFlag ? &ExecuteLDMLE : &ExecuteLDMBE); });
current.arm_pc += GetInstSize();
if (list & (1 << 15)) {
code->BT(32, MJitStateArmPC(), Imm8(0));
code->SETcc(CC_C, MJitStateTFlag());
code->AND(32, MJitStateArmPC(), Imm32(0xFFFFFFFE));
CompileReturnToDispatch();
}
}
static void ExecuteSTMLE(u32 start_address, u16 reg_list, JitState* jit_state) {
for (int i = 0; i < 15; i++) {
const u16 bit = 1 << i;
if (reg_list & bit) {
Memory::Write32(start_address, jit_state->cpu_state.Reg[i]);
start_address += 4;
}
}
// Reading R15 here is IMPLEMENTATION DEFINED
if (reg_list & (1 << 15)) {
if (!jit_state->cpu_state.TFlag) {
Memory::Write32(start_address, jit_state->cpu_state.Reg[15] + 8);
} else {
Memory::Write32(start_address, jit_state->cpu_state.Reg[15] + 4);
}
}
}
static void ExecuteSTMBE(u32 start_address, u16 reg_list, JitState* jit_state) {
for (int i = 0; i < 16; i++) {
const u16 bit = 1 << i;
if (reg_list & bit) {
Memory::Write32(start_address, Common::swap32(jit_state->cpu_state.Reg[i]));
start_address += 4;
}
}
// Reading R15 here is IMPLEMENTATION DEFINED
if (reg_list & (1 << 15)) {
if (!jit_state->cpu_state.TFlag) {
Memory::Write32(start_address, Common::swap32(jit_state->cpu_state.Reg[15] + 8));
} else {
Memory::Write32(start_address, Common::swap32(jit_state->cpu_state.Reg[15] + 4));
}
}
}
void JitX64::STM(Cond cond, bool P, bool U, bool W, ArmReg Rn_index, ArmRegList list) {
cond_manager.CompileCond((ConditionCode)cond);
ASSERT(list != 0, "UNPREDICTABLE");
if (W && (list & (1 << Rn_index)))
ASSERT_MSG((list & ((1 << Rn_index) - 1)) == 0, "UNPREDICTABLE");
// TODO: Optimize
LoadAndStoreMultiple_Helper(code, reg_alloc, P, U, W, Rn_index, list,
[this](){ CompileCallHost(!current.EFlag ? &ExecuteSTMLE : &ExecuteSTMBE); });
current.arm_pc += GetInstSize();
}
void JitX64::LDM_usr() { CompileInterpretInstruction(); }
void JitX64::LDM_eret() { CompileInterpretInstruction(); }
void JitX64::STM_usr() { CompileInterpretInstruction(); }
}

7
src/core/arm/jit_x64/jit_x64.h
View File

@ -285,8 +285,11 @@ private:
void STRT() override;
// Load/Store multiple instructions
void LDM() override;
void STM() override;
void LDM(Cond cond, bool P, bool U, bool W, ArmReg Rn, ArmRegList list) override;
void LDM_usr() override;
void LDM_eret() override;
void STM(Cond cond, bool P, bool U, bool W, ArmReg Rn, ArmRegList list) override;
void STM_usr() override;
// Miscellaneous instructions
void CLZ() override;

2
src/tests/core/arm/jit_x64/fuzz_arm_common.cpp
View File

@ -160,7 +160,7 @@ void FuzzJit(const int instruction_count, const int instructions_to_execute_coun
if (interp_mem_recording != jit_mem_recording) {
printf("memory write recording mismatch *\n");
size_t i = 0;
while (i < interp_mem_recording.size() && i < jit_mem_recording.size()) {
while (i < interp_mem_recording.size() || i < jit_mem_recording.size()) {
if (i < interp_mem_recording.size())
printf("interp: %i %08x %08x\n", interp_mem_recording[i].size, interp_mem_recording[i].addr, interp_mem_recording[i].data);
if (i < jit_mem_recording.size())

36
src/tests/core/arm/jit_x64/fuzz_arm_load_store.cpp
View File

@ -93,4 +93,40 @@ TEST_CASE("Fuzz ARM load/store instructions (double-word)", "[JitX64]") {
SECTION("short blocks") {
FuzzJit(1, 2, 5000, instruction_select);
}
}
TEST_CASE("Fuzz ARM load/store multiple instructions", "[JitX64]") {
const std::array<std::pair<u32, u32>, 2> instructions = {{
FromBitString32("cccc100pu0w1nnnnxxxxxxxxxxxxxxxx"), // LDM
FromBitString32("cccc100pu0w0nnnnxxxxxxxxxxxxxxxx"), // STM
}};
auto instruction_select = [&]() -> u32 {
size_t inst_index = RandInt<size_t>(0, instructions.size() - 1);
u32 cond = 0xE;
// Have a one-in-twenty-five chance of actually having a cond.
if (RandInt(1, 25) == 1) {
cond = RandInt<u32>(0x0, 0xD);
}
u32 reg_list = RandInt<u32>(1, 0xFFFF);
u32 Rn = RandInt<u32>(0, 14);
u32 flags = RandInt<u32>(0, 0xF);
if (inst_index == 1 && (flags & 2)) {
if (reg_list & (1 << Rn))
reg_list &= ~((1 << Rn) - 1);
} else if (inst_index == 1 && (flags & 2)) {
reg_list &= ~(1 << Rn);
}
u32 assemble_randoms = (reg_list << 0) | (Rn << 16) | (flags << 24) | (cond << 28);
return instructions[inst_index].first | (assemble_randoms & (~instructions[inst_index].second));
};
SECTION("short blocks") {
FuzzJit(1, 1, 5000, instruction_select);
}
}