izzy/dolphin
0
0
Fork 0
mirror of https://github.com/izzy2lost/dolphin.git synced 2026-06-19 01:16:48 -07:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #1911 from Sonicadvance1/AArch64_paired_loadstores

Browse Source 
[AArch64] Implements paired loadstores
This commit is contained in:
Ryan Houdek
2015-01-20 16:33:08 -06:00
parent 714697faaf 4262d2199a
commit cb967a6ff2
11 changed files with 1138 additions and 96 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Source/Core/Common/Arm64Emitter.cpp
+502 -28
View File
File diff suppressed because it is too large Load Diff
Source/Core/Common/Arm64Emitter.h
+68 -17
View File
@@ -82,6 +82,8 @@ inline bool IsQuad(ARM64Reg reg) { return (reg & 0xC0) == 0xC0; }
inline bool IsVector(ARM64Reg reg) { return (reg & 0xC0) != 0; }
inline ARM64Reg DecodeReg(ARM64Reg reg) { return (ARM64Reg)(reg & 0x1F); }
inline ARM64Reg EncodeRegTo64(ARM64Reg reg) { return (ARM64Reg)(reg | 0x20); }
inline ARM64Reg EncodeRegToDouble(ARM64Reg reg) { return (ARM64Reg)((reg & ~0xC0) | 0x80); }
inline ARM64Reg EncodeRegToQuad(ARM64Reg reg) { return (ARM64Reg)(reg | 0xC0); }
enum OpType
{
@@ -217,10 +219,24 @@ private:
u32 m_shift;
public:
ArithOption(ARM64Reg Rd)
ArithOption(ARM64Reg Rd, bool index = false)
{
// Indexed registers are a certain feature of AARch64
// On Loadstore instructions that use a register offset
// We can have the register as an index
// If we are indexing then the offset register will
// be shifted to the left so we are indexing at intervals
// of the size of what we are loading
// 8-bit: Index does nothing
// 16-bit: Index LSL 1
// 32-bit: Index LSL 2
// 64-bit: Index LSL 3
if (index)
m_shift = 4;
else
m_shift = 0;
m_destReg = Rd;
m_shift = 0;
m_type = TYPE_EXTENDEDREG;
if (Is64Bit(Rd))
{
@@ -256,18 +272,20 @@ public:
{
return m_type;
}
ARM64Reg GetReg()
{
return m_destReg;
}
u32 GetData() const
{
switch (m_type)
{
case TYPE_EXTENDEDREG:
return (m_width == WIDTH_64BIT ? (1 << 31) : 0) |
(m_extend << 13) |
return (m_extend << 13) |
(m_shift << 10);
break;
case TYPE_SHIFTEDREG:
return (m_width == WIDTH_64BIT ? (1 << 31) : 0) |
(m_shifttype << 22) |
return (m_shifttype << 22) |
(m_shift << 10);
break;
default:
@@ -309,7 +327,7 @@ private:
void EncodeLoadStoreIndexedInst(u32 op, ARM64Reg Rt, ARM64Reg Rn, s32 imm, u8 size);
void EncodeMOVWideInst(u32 op, ARM64Reg Rd, u32 imm, ShiftAmount pos);
void EncodeBitfieldMOVInst(u32 op, ARM64Reg Rd, ARM64Reg Rn, u32 immr, u32 imms);
void EncodeLoadStoreRegisterOffset(u32 size, u32 opc, ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend);
void EncodeLoadStoreRegisterOffset(u32 size, u32 opc, ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void EncodeAddSubImmInst(u32 op, bool flags, u32 shift, u32 imm, ARM64Reg Rn, ARM64Reg Rd);
void EncodeLogicalImmInst(u32 op, ARM64Reg Rd, ARM64Reg Rn, u32 immr, u32 imms);
void EncodeLoadStorePair(u32 op, u32 load, IndexType type, ARM64Reg Rt, ARM64Reg Rt2, ARM64Reg Rn, s32 imm);
@@ -505,6 +523,8 @@ public:
void SXTB(ARM64Reg Rd, ARM64Reg Rn);
void SXTH(ARM64Reg Rd, ARM64Reg Rn);
void SXTW(ARM64Reg Rd, ARM64Reg Rn);
void UXTB(ARM64Reg Rd, ARM64Reg Rn);
void UXTH(ARM64Reg Rd, ARM64Reg Rn);
// Load Register (Literal)
void LDR(ARM64Reg Rt, u32 imm);
@@ -551,16 +571,16 @@ public:
void LDRSW(IndexType type, ARM64Reg Rt, ARM64Reg Rn, s32 imm);
// Load/Store register (register offset)
void STRB(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void LDRB(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void LDRSB(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void STRH(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void LDRH(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void LDRSH(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void STR(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void LDR(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void LDRSW(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void PRFM(ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm, ExtendType extend = EXTEND_LSL);
void STRB(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void LDRB(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void LDRSB(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void STRH(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void LDRH(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void LDRSH(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void STR(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void LDR(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void LDRSW(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
void PRFM(ARM64Reg Rt, ARM64Reg Rn, ArithOption Rm);
// Load/Store pair
void LDP(IndexType type, ARM64Reg Rt, ARM64Reg Rt2, ARM64Reg Rn, s32 imm);
@@ -588,11 +608,21 @@ public:
void STR(u8 size, IndexType type, ARM64Reg Rt, ARM64Reg Rn, s32 imm);
// Loadstore single structure
void LD1(u8 size, ARM64Reg Rt, u8 index, ARM64Reg Rn);
void LD1(u8 size, ARM64Reg Rt, u8 index, ARM64Reg Rn, ARM64Reg Rm);
void LD1R(u8 size, ARM64Reg Rt, ARM64Reg Rn);
void ST1(u8 size, ARM64Reg Rt, u8 index, ARM64Reg Rn);
void ST1(u8 size, ARM64Reg Rt, u8 index, ARM64Reg Rn, ARM64Reg Rm);
// Loadstore multiple structure
void LD1(u8 size, u8 count, ARM64Reg Rt, ARM64Reg Rn);
// Scalar - 2 Source
void FMUL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
// Scalar floating point immediate
void FMOV(ARM64Reg Rd, u32 imm);
// Vector
void AND(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
void BSL(ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
@@ -600,6 +630,9 @@ public:
void FABS(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void FADD(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
void FCVTL(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void FCVTN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn);
void FCVTZS(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void FCVTZU(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void FDIV(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
void FMUL(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
void FNEG(u8 size, ARM64Reg Rd, ARM64Reg Rn);
@@ -610,17 +643,24 @@ public:
void REV16(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void REV32(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void REV64(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void SCVTF(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void UCVTF(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void XTN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn);
// Move
void DUP(u8 size, ARM64Reg Rd, ARM64Reg Rn);
void INS(u8 size, ARM64Reg Rd, u8 index, ARM64Reg Rn);
void INS(u8 size, ARM64Reg Rd, u8 index1, ARM64Reg Rn, u8 index2);
void UMOV(u8 size, ARM64Reg Rd, ARM64Reg Rn, u8 index);
void SMOV(u8 size, ARM64Reg Rd, ARM64Reg Rn, u8 index);
// One source
void FCVT(u8 size_to, u8 size_from, ARM64Reg Rd, ARM64Reg Rn);
// Conversion between float and integer
void FMOV(u8 size, bool top, ARM64Reg Rd, ARM64Reg Rn);
void SCVTF(ARM64Reg Rd, ARM64Reg Rn);
void UCVTF(ARM64Reg Rd, ARM64Reg Rn);
// Float comparison
void FCMP(ARM64Reg Rn, ARM64Reg Rm);
@@ -647,6 +687,13 @@ public:
void TRN2(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
void ZIP2(u8 size, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
// Shift by immediate
void SSHLL(u8 src_size, ARM64Reg Rd, ARM64Reg Rn, u32 shift);
void USHLL(u8 src_size, ARM64Reg Rd, ARM64Reg Rn, u32 shift);
void SHRN(u8 dest_size, ARM64Reg Rd, ARM64Reg Rn, u32 shift);
void SXTL(u8 src_size, ARM64Reg Rd, ARM64Reg Rn);
void UXTL(u8 src_size, ARM64Reg Rd, ARM64Reg Rn);
// ABI related
void ABI_PushRegisters(BitSet32 registers);
void ABI_PopRegisters(BitSet32 registers, BitSet32 ignore_mask = BitSet32(0));
@@ -662,11 +709,15 @@ private:
void EmitCopy(bool Q, u32 op, u32 imm5, u32 imm4, ARM64Reg Rd, ARM64Reg Rn);
void Emit2RegMisc(bool U, u32 size, u32 opcode, ARM64Reg Rd, ARM64Reg Rn);
void EmitLoadStoreSingleStructure(bool L, bool R, u32 opcode, bool S, u32 size, ARM64Reg Rt, ARM64Reg Rn);
void EmitLoadStoreSingleStructure(bool L, bool R, u32 opcode, bool S, u32 size, ARM64Reg Rt, ARM64Reg Rn, ARM64Reg Rm);
void Emit1Source(bool M, bool S, u32 type, u32 opcode, ARM64Reg Rd, ARM64Reg Rn);
void EmitConversion(bool sf, bool S, u32 type, u32 rmode, u32 opcode, ARM64Reg Rd, ARM64Reg Rn);
void EmitCompare(bool M, bool S, u32 op, u32 opcode2, ARM64Reg Rn, ARM64Reg Rm);
void EmitCondSelect(bool M, bool S, CCFlags cond, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
void EmitPermute(u32 size, u32 op, ARM64Reg Rd, ARM64Reg Rn, ARM64Reg Rm);
void EmitScalarImm(bool M, bool S, u32 type, u32 imm5, ARM64Reg Rd, u32 imm);
void EmitShiftImm(bool U, u32 immh, u32 immb, u32 opcode, ARM64Reg Rd, ARM64Reg Rn);
void EmitLoadStoreMultipleStructure(u32 size, bool L, u32 opcode, ARM64Reg Rt, ARM64Reg Rn);
};
class ARM64CodeBlock : public CodeBlock<ARM64XEmitter>
Source/Core/Core/CMakeLists.txt
+2 -1
View File
@@ -165,6 +165,7 @@ set(SRCS ActionReplay.cpp
PowerPC/Interpreter/Interpreter_Paired.cpp
PowerPC/Interpreter/Interpreter_SystemRegisters.cpp
PowerPC/Interpreter/Interpreter_Tables.cpp
PowerPC/JitCommon/JitAsmCommon.cpp
PowerPC/JitCommon/JitBase.cpp
PowerPC/JitCommon/JitCache.cpp
PowerPC/JitILCommon/IR.cpp
@@ -195,7 +196,6 @@ if(_M_X86)
PowerPC/Jit64/Jit_Paired.cpp
PowerPC/Jit64/JitRegCache.cpp
PowerPC/Jit64/Jit_SystemRegisters.cpp
PowerPC/JitCommon/JitAsmCommon.cpp
PowerPC/JitCommon/JitBackpatch.cpp
PowerPC/JitCommon/Jit_Util.cpp
PowerPC/JitCommon/TrampolineCache.cpp)
@@ -230,6 +230,7 @@ elseif(_M_ARM_64)
PowerPC/JitArm64/JitArm64_LoadStore.cpp
PowerPC/JitArm64/JitArm64_LoadStoreFloating.cpp
PowerPC/JitArm64/JitArm64_Paired.cpp
PowerPC/JitArm64/JitArm64_LoadStorePaired.cpp
PowerPC/JitArm64/JitArm64_SystemRegisters.cpp
PowerPC/JitArm64/JitArm64_Tables.cpp)
endif()
Source/Core/Core/PowerPC/JitArm32/JitArm_LoadStorePaired.cpp
+4 -4
View File
@@ -33,7 +33,7 @@ void JitArm::psq_l(UGeckoInstruction inst)
UBFX(R12, R11, 16, 3); // Type
LSL(R12, R12, 2);
UBFX(R11, R11, 24, 6); // Scale
LSL(R11, R11, 2);
LSL(R11, R11, 3);
Operand2 off;
if (TryMakeOperand2(offset, off))
@@ -84,7 +84,7 @@ void JitArm::psq_lx(UGeckoInstruction inst)
UBFX(R12, R11, 16, 3); // Type
LSL(R12, R12, 2);
UBFX(R11, R11, 24, 6); // Scale
LSL(R11, R11, 2);
LSL(R11, R11, 3);
if (inst.RA || update) // Always uses the register on update
{
@@ -136,7 +136,7 @@ void JitArm::psq_st(UGeckoInstruction inst)
UBFX(R12, R11, 0, 3); // Type
LSL(R12, R12, 2);
UBFX(R11, R11, 8, 6); // Scale
LSL(R11, R11, 2);
LSL(R11, R11, 3);
Operand2 off;
if (TryMakeOperand2(offset, off))
@@ -187,7 +187,7 @@ void JitArm::psq_stx(UGeckoInstruction inst)
UBFX(R12, R11, 0, 3); // Type
LSL(R12, R12, 2);
UBFX(R11, R11, 8, 6); // Scale
LSL(R11, R11, 2);
LSL(R11, R11, 3);
if (inst.RA || update) // Always uses the register on update
{
Source/Core/Core/PowerPC/JitArm32/JitAsm.cpp
-40
View File
@@ -27,46 +27,6 @@ using namespace ArmGen;
JitArmAsmRoutineManager asm_routines;
static const float GC_ALIGNED16(m_quantizeTableS[]) =
{
(1 << 0), (1 << 1), (1 << 2), (1 << 3),
(1 << 4), (1 << 5), (1 << 6), (1 << 7),
(1 << 8), (1 << 9), (1 << 10), (1 << 11),
(1 << 12), (1 << 13), (1 << 14), (1 << 15),
(1 << 16), (1 << 17), (1 << 18), (1 << 19),
(1 << 20), (1 << 21), (1 << 22), (1 << 23),
(1 << 24), (1 << 25), (1 << 26), (1 << 27),
(1 << 28), (1 << 29), (1 << 30), (1 << 31),
1.0 / (1ULL << 32), 1.0 / (1 << 31), 1.0 / (1 << 30), 1.0 / (1 << 29),
1.0 / (1 << 28), 1.0 / (1 << 27), 1.0 / (1 << 26), 1.0 / (1 << 25),
1.0 / (1 << 24), 1.0 / (1 << 23), 1.0 / (1 << 22), 1.0 / (1 << 21),
1.0 / (1 << 20), 1.0 / (1 << 19), 1.0 / (1 << 18), 1.0 / (1 << 17),
1.0 / (1 << 16), 1.0 / (1 << 15), 1.0 / (1 << 14), 1.0 / (1 << 13),
1.0 / (1 << 12), 1.0 / (1 << 11), 1.0 / (1 << 10), 1.0 / (1 << 9),
1.0 / (1 << 8), 1.0 / (1 << 7), 1.0 / (1 << 6), 1.0 / (1 << 5),
1.0 / (1 << 4), 1.0 / (1 << 3), 1.0 / (1 << 2), 1.0 / (1 << 1),
};
static const float GC_ALIGNED16(m_dequantizeTableS[]) =
{
1.0 / (1 << 0), 1.0 / (1 << 1), 1.0 / (1 << 2), 1.0 / (1 << 3),
1.0 / (1 << 4), 1.0 / (1 << 5), 1.0 / (1 << 6), 1.0 / (1 << 7),
1.0 / (1 << 8), 1.0 / (1 << 9), 1.0 / (1 << 10), 1.0 / (1 << 11),
1.0 / (1 << 12), 1.0 / (1 << 13), 1.0 / (1 << 14), 1.0 / (1 << 15),
1.0 / (1 << 16), 1.0 / (1 << 17), 1.0 / (1 << 18), 1.0 / (1 << 19),
1.0 / (1 << 20), 1.0 / (1 << 21), 1.0 / (1 << 22), 1.0 / (1 << 23),
1.0 / (1 << 24), 1.0 / (1 << 25), 1.0 / (1 << 26), 1.0 / (1 << 27),
1.0 / (1 << 28), 1.0 / (1 << 29), 1.0 / (1 << 30), 1.0 / (1 << 31),
(1ULL << 32), (1 << 31), (1 << 30), (1 << 29),
(1 << 28), (1 << 27), (1 << 26), (1 << 25),
(1 << 24), (1 << 23), (1 << 22), (1 << 21),
(1 << 20), (1 << 19), (1 << 18), (1 << 17),
(1 << 16), (1 << 15), (1 << 14), (1 << 13),
(1 << 12), (1 << 11), (1 << 10), (1 << 9),
(1 << 8), (1 << 7), (1 << 6), (1 << 5),
(1 << 4), (1 << 3), (1 << 2), (1 << 1),
};
static void WriteDual32(u32 value1, u32 value2, u32 address)
{
Memory::Write_U32(value1, address);
Source/Core/Core/PowerPC/JitArm64/Jit.h
+4
View File
@@ -168,6 +168,10 @@ public:
void ps_sum0(UGeckoInstruction inst);
void ps_sum1(UGeckoInstruction inst);
// Loadstore paired
void psq_l(UGeckoInstruction inst);
void psq_st(UGeckoInstruction inst);
private:
Arm64GPRCache gpr;
Arm64FPRCache fpr;
Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStorePaired.cpp
+130
View File
@@ -0,0 +1,130 @@
// Copyright 2014 Dolphin Emulator Project
// Licensed under GPLv2
// Refer to the license.txt file included.
#include "Common/Arm64Emitter.h"
#include "Common/Common.h"
#include "Common/StringUtil.h"
#include "Core/Core.h"
#include "Core/CoreTiming.h"
#include "Core/PowerPC/PowerPC.h"
#include "Core/PowerPC/PPCTables.h"
#include "Core/PowerPC/JitArm64/Jit.h"
#include "Core/PowerPC/JitArm64/JitArm64_RegCache.h"
#include "Core/PowerPC/JitArm64/JitAsm.h"
using namespace Arm64Gen;
void JitArm64::psq_l(UGeckoInstruction inst)
{
INSTRUCTION_START
JITDISABLE(bJITLoadStorePairedOff);
FALLBACK_IF(js.memcheck || !SConfig::GetInstance().m_LocalCoreStartupParameter.bFastmem);
// X30 is LR
// X0 contains the scale
// X1 is the address
// X2 is a temporary
// Q0 is the return register
// Q1 is a temporary
bool update = inst.OPCD == 57;
s32 offset = inst.SIMM_12;
gpr.Lock(W0, W1, W2, W30);
fpr.Lock(Q0, Q1);
ARM64Reg arm_addr = gpr.R(inst.RA);
ARM64Reg scale_reg = W0;
ARM64Reg addr_reg = W1;
ARM64Reg type_reg = W2;
LDR(INDEX_UNSIGNED, scale_reg, X29, PPCSTATE_OFF(spr[SPR_GQR0 + inst.I]));
if (inst.RA || update) // Always uses the register on update
{
if (offset >= 0)
ADD(addr_reg, gpr.R(inst.RA), offset);
else
SUB(addr_reg, gpr.R(inst.RA), std::abs(offset));
}
else
{
MOVI2R(addr_reg, (u32)offset);
}
UBFM(type_reg, scale_reg, 16, 18); // Type
UBFM(scale_reg, scale_reg, 24, 29); // Scale
if (update)
MOV(arm_addr, addr_reg);
MOVI2R(X30, (u64)&asm_routines.pairedLoadQuantized[inst.W * 8]);
LDR(X30, X30, ArithOption(EncodeRegTo64(type_reg), true));
BLR(X30);
fpr.BindToRegister(inst.RS, false);
ARM64Reg VS = fpr.R(inst.RS);
m_float_emit.FCVTL(64, EncodeRegToDouble(VS), D0);
if (inst.W)
{
m_float_emit.FMOV(D0, 0x70); // 1.0 as a Double
m_float_emit.INS(64, VS, 1, Q0, 0);
}
gpr.Unlock(W0, W1, W2, W30);
fpr.Unlock(Q0, Q1);
}
void JitArm64::psq_st(UGeckoInstruction inst)
{
INSTRUCTION_START
JITDISABLE(bJITLoadStorePairedOff);
FALLBACK_IF(js.memcheck || !SConfig::GetInstance().m_LocalCoreStartupParameter.bFastmem);
// X30 is LR
// X0 contains the scale
// X1 is the address
// Q0 is the store register
bool update = inst.OPCD == 61;
s32 offset = inst.SIMM_12;
gpr.Lock(W0, W1, W2, W30);
fpr.Lock(Q0, Q1);
ARM64Reg arm_addr = gpr.R(inst.RA);
ARM64Reg scale_reg = W0;
ARM64Reg addr_reg = W1;
ARM64Reg type_reg = gpr.GetReg();
LDR(INDEX_UNSIGNED, scale_reg, X29, PPCSTATE_OFF(spr[SPR_GQR0 + inst.I]));
if (inst.RA || update) // Always uses the register on update
{
if (offset >= 0)
ADD(addr_reg, gpr.R(inst.RA), offset);
else
SUB(addr_reg, gpr.R(inst.RA), std::abs(offset));
}
else
{
MOVI2R(addr_reg, (u32)offset);
}
UBFM(type_reg, scale_reg, 0, 2); // Type
UBFM(scale_reg, scale_reg, 8, 13); // Scale
if (update)
MOV(arm_addr, addr_reg);
ARM64Reg VS = fpr.R(inst.RS);
m_float_emit.FCVTN(32, D0, VS);
MOVI2R(X30, (u64)&asm_routines.pairedStoreQuantized[inst.W * 8]);
LDR(X30, X30, ArithOption(EncodeRegTo64(type_reg), true));
BLR(X30);
gpr.Unlock(W0, W1, W2, W30, type_reg);
fpr.Unlock(Q0, Q1);
}
Source/Core/Core/PowerPC/JitArm64/JitArm64_Tables.cpp
+4 -4
View File
@@ -94,10 +94,10 @@ static GekkoOPTemplate primarytable[] =
{54, &JitArm64::stfXX}, //"stfd", OPTYPE_STOREFP, FL_IN_A}},
{55, &JitArm64::stfXX}, //"stfdu", OPTYPE_STOREFP, FL_OUT_A | FL_IN_A}},
{56, &JitArm64::FallBackToInterpreter}, //"psq_l", OPTYPE_PS, FL_IN_A}},
{57, &JitArm64::FallBackToInterpreter}, //"psq_lu", OPTYPE_PS, FL_OUT_A | FL_IN_A}},
{60, &JitArm64::FallBackToInterpreter}, //"psq_st", OPTYPE_PS, FL_IN_A}},
{61, &JitArm64::FallBackToInterpreter}, //"psq_stu", OPTYPE_PS, FL_OUT_A | FL_IN_A}},
{56, &JitArm64::psq_l}, //"psq_l", OPTYPE_PS, FL_IN_A}},
{57, &JitArm64::psq_l}, //"psq_lu", OPTYPE_PS, FL_OUT_A | FL_IN_A}},
{60, &JitArm64::psq_st}, //"psq_st", OPTYPE_PS, FL_IN_A}},
{61, &JitArm64::psq_st}, //"psq_stu", OPTYPE_PS, FL_OUT_A | FL_IN_A}},
//missing: 0, 5, 6, 9, 22, 30, 62, 58
{0, &JitArm64::FallBackToInterpreter}, //"unknown_instruction", OPTYPE_UNKNOWN, 0}},
Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp
+420
View File
@@ -7,6 +7,7 @@
#include "Core/PowerPC/PowerPC.h"
#include "Core/PowerPC/JitArm64/Jit.h"
#include "Core/PowerPC/JitArm64/JitAsm.h"
#include "Core/PowerPC/JitCommon/JitAsmCommon.h"
#include "Core/PowerPC/JitCommon/JitCache.h"
using namespace Arm64Gen;
@@ -89,9 +90,428 @@ void JitArm64AsmRoutineManager::Generate()
ABI_PopRegisters(regs_to_save);
RET(X30);
GenerateCommon();
FlushIcache();
}
void JitArm64AsmRoutineManager::GenerateCommon()
{
// X0 is the scale
// X1 is address
// X2 is a temporary on stores
// X30 is LR
// Q0 is the return for loads
// is the register for stores
// Q1 is a temporary
ARM64Reg addr_reg = X1;
ARM64Reg scale_reg = X0;
ARM64FloatEmitter float_emit(this);
const u32 GPR_CALLER_SAVE = 0x6007FFFF;
const u8* loadPairedIllegal = GetCodePtr();
BRK(100);
const u8* loadPairedFloatTwo = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LD1(32, 1, D0, addr_reg);
float_emit.REV32(8, D0, D0);
RET(X30);
}
const u8* loadPairedU8Two = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LDR(16, INDEX_UNSIGNED, D0, addr_reg, 0);
float_emit.UXTL(8, D0, D0);
float_emit.UXTL(16, D0, D0);
float_emit.UCVTF(32, D0, D0);
MOVI2R(addr_reg, (u64)&m_dequantizeTableS);
ADD(scale_reg, addr_reg, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
RET(X30);
}
const u8* loadPairedS8Two = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LDR(16, INDEX_UNSIGNED, D0, addr_reg, 0);
float_emit.SXTL(8, D0, D0);
float_emit.SXTL(16, D0, D0);
float_emit.SCVTF(32, D0, D0);
MOVI2R(addr_reg, (u64)&m_dequantizeTableS);
ADD(scale_reg, addr_reg, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
RET(X30);
}
const u8* loadPairedU16Two = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LD1(16, 1, D0, addr_reg);
float_emit.REV16(8, D0, D0);
float_emit.UXTL(16, D0, D0);
float_emit.UCVTF(32, D0, D0);
MOVI2R(addr_reg, (u64)&m_dequantizeTableS);
ADD(scale_reg, addr_reg, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
RET(X30);
}
const u8* loadPairedS16Two = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LD1(16, 1, D0, addr_reg);
float_emit.REV16(8, D0, D0);
float_emit.SXTL(16, D0, D0);
float_emit.SCVTF(32, D0, D0);
MOVI2R(addr_reg, (u64)&m_dequantizeTableS);
ADD(scale_reg, addr_reg, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
RET(X30);
}
const u8* loadPairedFloatOne = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LDR(32, INDEX_UNSIGNED, D0, addr_reg, 0);
float_emit.REV32(8, D0, D0);
RET(X30);
}
const u8* loadPairedU8One = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LDR(8, INDEX_UNSIGNED, D0, addr_reg, 0);
float_emit.UXTL(8, D0, D0);
float_emit.UXTL(16, D0, D0);
float_emit.UCVTF(32, D0, D0);
MOVI2R(addr_reg, (u64)&m_dequantizeTableS);
ADD(scale_reg, addr_reg, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
RET(X30);
}
const u8* loadPairedS8One = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LDR(8, INDEX_UNSIGNED, D0, addr_reg, 0);
float_emit.SXTL(8, D0, D0);
float_emit.SXTL(16, D0, D0);
float_emit.SCVTF(32, D0, D0);
MOVI2R(addr_reg, (u64)&m_dequantizeTableS);
ADD(scale_reg, addr_reg, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
RET(X30);
}
const u8* loadPairedU16One = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LDR(16, INDEX_UNSIGNED, D0, addr_reg, 0);
float_emit.REV16(8, D0, D0);
float_emit.UXTL(16, D0, D0);
float_emit.UCVTF(32, D0, D0);
MOVI2R(addr_reg, (u64)&m_dequantizeTableS);
ADD(scale_reg, addr_reg, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
RET(X30);
}
const u8* loadPairedS16One = GetCodePtr();
{
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.LDR(16, INDEX_UNSIGNED, D0, addr_reg, 0);
float_emit.REV16(8, D0, D0);
float_emit.SXTL(16, D0, D0);
float_emit.SCVTF(32, D0, D0);
MOVI2R(addr_reg, (u64)&m_dequantizeTableS);
ADD(scale_reg, addr_reg, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
RET(X30);
}
pairedLoadQuantized = reinterpret_cast<const u8**>(const_cast<u8*>(AlignCode16()));
ReserveCodeSpace(16 * sizeof(u8*));
pairedLoadQuantized[0] = loadPairedFloatTwo;
pairedLoadQuantized[1] = loadPairedIllegal;
pairedLoadQuantized[2] = loadPairedIllegal;
pairedLoadQuantized[3] = loadPairedIllegal;
pairedLoadQuantized[4] = loadPairedU8Two;
pairedLoadQuantized[5] = loadPairedU16Two;
pairedLoadQuantized[6] = loadPairedS8Two;
pairedLoadQuantized[7] = loadPairedS16Two;
pairedLoadQuantized[8] = loadPairedFloatOne;
pairedLoadQuantized[9] = loadPairedIllegal;
pairedLoadQuantized[10] = loadPairedIllegal;
pairedLoadQuantized[11] = loadPairedIllegal;
pairedLoadQuantized[12] = loadPairedU8One;
pairedLoadQuantized[13] = loadPairedU16One;
pairedLoadQuantized[14] = loadPairedS8One;
pairedLoadQuantized[15] = loadPairedS16One;
// Stores
const u8* storePairedIllegal = GetCodePtr();
BRK(0x101);
const u8* storePairedFloat = GetCodePtr();
{
BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
BitSet32 fprs(~3); // All except Q0/Q1
TST(DecodeReg(addr_reg), 6, 1);
FixupBranch argh = B(CC_NEQ);
float_emit.REV32(8, D0, D0);
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.ST1(32, Q0, 0, addr_reg, SP);
float_emit.ST1(32, Q0, 1, addr_reg, SP);
RET(X30);
SetJumpTarget(argh);
ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs);
float_emit.UMOV(64, X0, Q0, 0);
ORR(X0, SP, X0, ArithOption(X0, ST_ROR, 32));
MOVI2R(X30, (u64)Memory::Write_U64);
BLR(X30);
float_emit.ABI_PopRegisters(fprs);
ABI_PopRegisters(gprs);
RET(X30);
}
const u8* storePairedU8 = GetCodePtr();
const u8* storePairedS8 = GetCodePtr();
{
BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
BitSet32 fprs(~3); // All except Q0/Q1
MOVI2R(X2, (u64)&m_quantizeTableS);
ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
float_emit.FCVTZU(32, D0, D0);
float_emit.XTN(16, D0, D0);
float_emit.XTN(8, D0, D0);
TST(DecodeReg(addr_reg), 6, 1);
FixupBranch argh = B(CC_NEQ);
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.ST1(8, Q0, 0, addr_reg, SP);
float_emit.ST1(8, Q0, 1, addr_reg, SP);
RET(X30);
SetJumpTarget(argh);
ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs);
float_emit.UMOV(16, W0, Q0, 0);
REV16(W0, W0);
MOVI2R(X30, (u64)Memory::Write_U16);
BLR(X30);
float_emit.ABI_PopRegisters(fprs);
ABI_PopRegisters(gprs);
RET(X30);
}
const u8* storePairedU16 = GetCodePtr();
const u8* storePairedS16 = GetCodePtr();
{
BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
BitSet32 fprs(~3); // All except Q0/Q1
MOVI2R(X2, (u64)&m_quantizeTableS);
ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LD1R(32, D1, scale_reg);
float_emit.FMUL(32, D0, D0, D1);
float_emit.FCVTZU(32, D0, D0);
float_emit.XTN(16, D0, D0);
TST(DecodeReg(addr_reg), 6, 1);
FixupBranch argh = B(CC_NEQ);
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.ST1(16, Q0, 0, addr_reg, SP);
float_emit.ST1(16, Q0, 1, addr_reg, SP);
RET(X30);
SetJumpTarget(argh);
ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs);
float_emit.UMOV(32, W0, Q0, 0);
REV32(W0, W0);
MOVI2R(X30, (u64)Memory::Write_U32);
BLR(X30);
float_emit.ABI_PopRegisters(fprs);
ABI_PopRegisters(gprs);
RET(X30);
}
const u8* storeSingleFloat = GetCodePtr();
{
BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
BitSet32 fprs(~3); // All except Q0/Q1
TST(DecodeReg(addr_reg), 6, 1);
FixupBranch argh = B(CC_NEQ);
float_emit.REV32(8, D0, D0);
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.STR(32, INDEX_UNSIGNED, D0, addr_reg, 0);
RET(X30);
SetJumpTarget(argh);
ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs);
float_emit.UMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&Memory::Write_U32);
BLR(X30);
float_emit.ABI_PopRegisters(fprs);
ABI_PopRegisters(gprs);
RET(X30);
}
const u8* storeSingleU8 = GetCodePtr(); // Used by MKWii
{
BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
BitSet32 fprs(~3); // All except Q0/Q1
MOVI2R(X2, (u64)&m_quantizeTableS);
ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
float_emit.FMUL(32, D0, D0, D1);
float_emit.FCVTZU(32, D0, D0);
float_emit.XTN(16, D0, D0);
float_emit.XTN(8, D0, D0);
TST(DecodeReg(addr_reg), 6, 1);
FixupBranch argh = B(CC_NEQ);
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.ST1(8, Q0, 0, addr_reg);
RET(X30);
SetJumpTarget(argh);
ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs);
float_emit.UMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&Memory::Write_U8);
BLR(X30);
float_emit.ABI_PopRegisters(fprs);
ABI_PopRegisters(gprs);
RET(X30);
}
const u8* storeSingleS8 = GetCodePtr();
{
BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
BitSet32 fprs(~3); // All except Q0/Q1
MOVI2R(X2, (u64)&m_quantizeTableS);
ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
float_emit.FMUL(32, D0, D0, D1);
float_emit.FCVTZS(32, D0, D0);
float_emit.XTN(16, D0, D0);
float_emit.XTN(8, D0, D0);
TST(DecodeReg(addr_reg), 6, 1);
FixupBranch argh = B(CC_NEQ);
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.ST1(8, Q0, 0, addr_reg);
RET(X30);
SetJumpTarget(argh);
ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs);
float_emit.SMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&Memory::Write_U8);
BLR(X30);
float_emit.ABI_PopRegisters(fprs);
ABI_PopRegisters(gprs);
RET(X30);
}
const u8* storeSingleU16 = GetCodePtr(); // Used by MKWii
{
BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
BitSet32 fprs(~3); // All except Q0/Q1
MOVI2R(X2, (u64)&m_quantizeTableS);
ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
float_emit.FMUL(32, D0, D0, D1);
float_emit.FCVTZU(32, D0, D0);
float_emit.XTN(16, D0, D0);
TST(DecodeReg(addr_reg), 6, 1);
FixupBranch argh = B(CC_NEQ);
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.ST1(16, Q0, 0, addr_reg);
RET(X30);
SetJumpTarget(argh);
ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs);
float_emit.UMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&Memory::Write_U16);
BLR(X30);
float_emit.ABI_PopRegisters(fprs);
ABI_PopRegisters(gprs);
RET(X30);
}
const u8* storeSingleS16 = GetCodePtr();
{
BitSet32 gprs(GPR_CALLER_SAVE & ~7); // All except X0/X1/X2
BitSet32 fprs(~3); // All except Q0/Q1
MOVI2R(X2, (u64)&m_quantizeTableS);
ADD(scale_reg, X2, scale_reg, ArithOption(scale_reg, ST_LSL, 3));
float_emit.LDR(32, INDEX_UNSIGNED, D1, scale_reg, 0);
float_emit.FMUL(32, D0, D0, D1);
float_emit.FCVTZS(32, D0, D0);
float_emit.XTN(16, D0, D0);
TST(DecodeReg(addr_reg), 6, 1);
FixupBranch argh = B(CC_NEQ);
MOVK(addr_reg, ((u64)Memory::base >> 32) & 0xFFFF, SHIFT_32);
float_emit.ST1(16, Q0, 0, addr_reg);
RET(X30);
SetJumpTarget(argh);
ABI_PushRegisters(gprs);
float_emit.ABI_PushRegisters(fprs);
float_emit.SMOV(32, W0, Q0, 0);
MOVI2R(X30, (u64)&Memory::Write_U16);
BLR(X30);
float_emit.ABI_PopRegisters(fprs);
ABI_PopRegisters(gprs);
RET(X30);
}
pairedStoreQuantized = reinterpret_cast<const u8**>(const_cast<u8*>(AlignCode16()));
ReserveCodeSpace(16 * sizeof(u8*));
pairedStoreQuantized[0] = storePairedFloat;
pairedStoreQuantized[1] = storePairedIllegal;
pairedStoreQuantized[2] = storePairedIllegal;
pairedStoreQuantized[3] = storePairedIllegal;
pairedStoreQuantized[4] = storePairedU8;
pairedStoreQuantized[5] = storePairedU16;
pairedStoreQuantized[6] = storePairedS8;
pairedStoreQuantized[7] = storePairedS16;
pairedStoreQuantized[8] = storeSingleFloat;
pairedStoreQuantized[9] = storePairedIllegal;
pairedStoreQuantized[10] = storePairedIllegal;
pairedStoreQuantized[11] = storePairedIllegal;
pairedStoreQuantized[12] = storeSingleU8;
pairedStoreQuantized[13] = storeSingleU16;
pairedStoreQuantized[14] = storeSingleS8;
pairedStoreQuantized[15] = storeSingleS16;
}
Source/Core/Core/PowerPC/JitCommon/JitAsmCommon.cpp
+2 -2
View File
@@ -194,7 +194,7 @@ void CommonAsmRoutines::GenMfcr()
const u8 GC_ALIGNED16(pbswapShuffle1x4[16]) = { 3, 2, 1, 0, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 };
const u8 GC_ALIGNED16(pbswapShuffle2x4[16]) = { 3, 2, 1, 0, 7, 6, 5, 4, 8, 9, 10, 11, 12, 13, 14, 15 };
static const float GC_ALIGNED16(m_quantizeTableS[]) =
const float GC_ALIGNED16(m_quantizeTableS[]) =
{
(1ULL << 0), (1ULL << 0), (1ULL << 1), (1ULL << 1), (1ULL << 2), (1ULL << 2), (1ULL << 3), (1ULL << 3),
(1ULL << 4), (1ULL << 4), (1ULL << 5), (1ULL << 5), (1ULL << 6), (1ULL << 6), (1ULL << 7), (1ULL << 7),
@@ -222,7 +222,7 @@ static const float GC_ALIGNED16(m_quantizeTableS[]) =
1.0 / (1ULL << 2), 1.0 / (1ULL << 2), 1.0 / (1ULL << 1), 1.0 / (1ULL << 1),
};
static const float GC_ALIGNED16(m_dequantizeTableS[]) =
const float GC_ALIGNED16(m_dequantizeTableS[]) =
{
1.0 / (1ULL << 0), 1.0 / (1ULL << 0), 1.0 / (1ULL << 1), 1.0 / (1ULL << 1),
1.0 / (1ULL << 2), 1.0 / (1ULL << 2), 1.0 / (1ULL << 3), 1.0 / (1ULL << 3),
Source/Core/Core/PowerPC/JitCommon/JitAsmCommon.h
+2
View File
@@ -9,6 +9,8 @@
extern const u8 GC_ALIGNED16(pbswapShuffle1x4[16]);
extern const u8 GC_ALIGNED16(pbswapShuffle2x4[16]);
extern const float GC_ALIGNED16(m_one[]);
extern const float GC_ALIGNED16(m_quantizeTableS[]);
extern const float GC_ALIGNED16(m_dequantizeTableS[]);
class CommonAsmRoutinesBase
{