ShuriZma
/
flycast
mirror of https://github.com/flyinghead/flycast.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
flycast/core/rec-ARM/rec_arm.cpp

2116 lines
45 KiB
C++
Raw Blame History

/*
This file is part of Flycast.
Flycast 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, either version 2 of the License, or
(at your option) any later version.
Flycast 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 for more details.
You should have received a copy of the GNU General Public License
along with Flycast. If not, see <https://www.gnu.org/licenses/>.
*/
#include "types.h"
#if FEAT_SHREC == DYNAREC_JIT && HOST_CPU == CPU_ARM
#include <unistd.h>
#include <array>
#include <map>
#include <aarch32/macro-assembler-aarch32.h>
using namespace vixl::aarch32;
#include "hw/sh4/sh4_opcode_list.h"
#include "hw/sh4/sh4_mmr.h"
#include "hw/sh4/sh4_rom.h"
#include "hw/sh4/sh4_interrupts.h"
#include "hw/sh4/sh4_core.h"
#include "hw/sh4/dyna/ngen.h"
#include "hw/sh4/dyna/ssa_regalloc.h"
#include "hw/sh4/sh4_mem.h"
#include "cfg/option.h"
//#define CANONICALTEST
/*
ARM ABI
r0~r1: scratch, params, return
r2~r3: scratch, params
8 regs, v6 is platform dependent
r4~r11
r12 is "The Intra-Procedure-call scratch register"
r13 stack
r14 link
r15 pc
Registers f0-s15 (d0-d7, q0-q3) do not need to be preserved (and can be used for passing arguments or returning results in standard procedure-call variants).
Registers s16-s31 (d8-d15, q4-q7) must be preserved across subroutine calls;
Registers d16-d31 (q8-q15), if present, do not need to be preserved.
Block linking
Reg alloc
r0~r4: scratch
r5,r6,r7,r10,r11: allocated
r8: sh4 cntx
r9: cycle counter
fpu reg alloc
d8:d15, single storage
*/
#ifdef __clang__
extern "C" char *stpcpy(char *dst, char const *src)
{
size_t src_len = strlen(src);
return (char *)memcpy(dst, src, src_len) + src_len;
}
#endif
#undef do_sqw_nommu
#define rcbOffset(x) (-sizeof(Sh4RCB) + offsetof(Sh4RCB, x))
struct DynaRBI: RuntimeBlockInfo
{
virtual u32 Relink();
virtual void Relocate(void* dst) { }
Register T_reg;
};
using FPBinOP = void (MacroAssembler::*)(DataType, SRegister, SRegister, SRegister);
using FPUnOP = void (MacroAssembler::*)(DataType, SRegister, SRegister);
using BinaryOP = void (MacroAssembler::*)(Register, Register, const Operand&);
class Arm32Assembler : public MacroAssembler
{
public:
Arm32Assembler() = default;
Arm32Assembler(u8 *buffer, size_t size) : MacroAssembler(buffer, size, A32) {}
void Finalize() {
FinalizeCode();
vmem_platform_flush_cache(GetBuffer()->GetStartAddress<void *>(), GetCursorAddress<u8 *>() - 1,
GetBuffer()->GetStartAddress<void *>(), GetCursorAddress<u8 *>() - 1);
}
};
static Arm32Assembler ass;
static u8 *sh4_dyna_context;
static void loadSh4Reg(Register Rt, u32 Sh4_Reg)
{
const int shRegOffs = (u8*)GetRegPtr(Sh4_Reg) - sh4_dyna_context;
ass.Ldr(Rt, MemOperand(r8, shRegOffs));
}
static void storeSh4Reg(Register Rt, u32 Sh4_Reg)
{
const int shRegOffs = (u8*)GetRegPtr(Sh4_Reg) - sh4_dyna_context;
ass.Str(Rt, MemOperand(r8, shRegOffs));
}
const int alloc_regs[] = { 5, 6, 7, 10, 11, -1 };
const int alloc_fpu[] = { 16, 17, 18, 19, 20, 21, 22, 23,
24, 25, 26, 27, 28, 29, 30, 31, -1 };
struct arm_reg_alloc: RegAlloc<int, int>
{
void Preload(u32 reg, int nreg) override
{
loadSh4Reg(Register(nreg), reg);
}
void Writeback(u32 reg, int nreg) override
{
if (reg == reg_pc_dyn)
// reg_pc_dyn has been stored in r4 by the jdyn op implementation
// No need to write it back since it won't be used past the end of the block
; //ass.Mov(r4, Register(nreg));
else
storeSh4Reg(Register(nreg), reg);
}
void Preload_FPU(u32 reg, int nreg) override
{
const s32 shRegOffs = (u8*)GetRegPtr(reg) - sh4_dyna_context;
ass.Vldr(SRegister(nreg), MemOperand(r8, shRegOffs));
}
void Writeback_FPU(u32 reg, int nreg) override
{
const s32 shRegOffs = (u8*)GetRegPtr(reg) - sh4_dyna_context;
ass.Vstr(SRegister(nreg), MemOperand(r8, shRegOffs));
}
SRegister mapFReg(const shil_param& prm)
{
return SRegister(mapf(prm));
}
Register mapReg(const shil_param& prm)
{
return Register(mapg(prm));
}
};
static arm_reg_alloc reg;
static const void *no_update;
static const void *intc_sched;
static const void *ngen_blockcheckfail;
static const void *ngen_LinkBlock_Generic_stub;
static const void *ngen_LinkBlock_cond_Branch_stub;
static const void *ngen_LinkBlock_cond_Next_stub;
static void (*ngen_FailedToFindBlock_)();
static void (*mainloop)(void *);
static std::map<shilop, ConditionType> ccmap;
static std::map<shilop, ConditionType> ccnmap;
void ngen_mainloop(void* context)
{
sh4_dyna_context = (u8 *)context;
mainloop(context);
}
static void jump(const void *code, ConditionType cond = al)
{
ptrdiff_t offset = reinterpret_cast<uintptr_t>(code) - ass.GetBuffer()->GetStartAddress<uintptr_t>();
verify((offset & 3) == 0);
if (offset < -32 * 1024 * 1024 || offset >= 32 * 1024 * 1024)
{
WARN_LOG(DYNAREC, "jump offset too large: %d", offset);
UseScratchRegisterScope scope(&ass);
Register reg = scope.Acquire();
ass.Mov(cond, reg, (u32)code);
ass.Bx(cond, reg);
}
else
{
Label code_label(offset);
ass.B(cond, &code_label);
}
}
static void call(const void *code, ConditionType cond = al)
{
ptrdiff_t offset = reinterpret_cast<uintptr_t>(code) - ass.GetBuffer()->GetStartAddress<uintptr_t>();
verify((offset & 3) == 0);
if (offset < -32 * 1024 * 1024 || offset >= 32 * 1024 * 1024)
{
WARN_LOG(DYNAREC, "call offset too large: %d", offset);
UseScratchRegisterScope scope(&ass);
Register reg = scope.Acquire();
ass.Mov(cond, reg, (u32)code);
ass.Blx(cond, reg);
}
else
{
Label code_label(offset);
ass.Bl(cond, &code_label);
}
}
static u32 relinkBlock(DynaRBI *block)
{
u32 start_offset = ass.GetCursorOffset();
switch(block->BlockType)
{
case BET_Cond_0:
case BET_Cond_1:
{
//quick opt here:
//peek into reg alloc, store actual sr_T register to relink_data
#ifndef CANONICALTEST
bool last_op_sets_flags = !block->has_jcond && !block->oplist.empty() &&
block->oplist.back().rd._reg == reg_sr_T && ccmap.count(block->oplist.back().op);
#else
bool last_op_sets_flags = false;
#endif
ConditionType CC = eq;
if (last_op_sets_flags)
{
shilop op = block->oplist.back().op;
if ((block->BlockType & 1) == 1)
CC = ccmap[op];
else
CC = ccnmap[op];
}
else
{
if (!block->has_jcond)
{
if (block->T_reg.IsRegister())
{
ass.Mov(r4, block->T_reg);
}
else
{
INFO_LOG(DYNAREC, "SLOW COND PATH %x", block->oplist.empty() ? -1 : block->oplist.back().op);
loadSh4Reg(r4, reg_sr_T);
}
}
ass.Cmp(r4, block->BlockType & 1);
}
if (block->pBranchBlock)
jump((void *)block->pBranchBlock->code, CC);
else
call(ngen_LinkBlock_cond_Branch_stub, CC);
if (block->pNextBlock)
jump((void *)block->pNextBlock->code);
else
call(ngen_LinkBlock_cond_Next_stub);
break;
}
case BET_DynamicRet:
case BET_DynamicCall:
case BET_DynamicJump:
ass.Sub(r2, r8, -rcbOffset(fpcb));
#if RAM_SIZE_MAX == 33554432
ass.Ubfx(r1, r4, 1, 24);
#else
ass.Ubfx(r1, r4, 1, 23);
#endif
ass.Ldr(pc, MemOperand(r2, r1, LSL, 2));
break;
case BET_StaticCall:
case BET_StaticJump:
if (block->pBranchBlock == nullptr)
call(ngen_LinkBlock_Generic_stub);
else
call((void *)block->pBranchBlock->code);
break;
case BET_StaticIntr:
case BET_DynamicIntr:
if (block->BlockType == BET_StaticIntr)
ass.Mov(r4, block->NextBlock);
//else -> already in r4 djump !
storeSh4Reg(r4, reg_nextpc);
call((void *)UpdateINTC);
loadSh4Reg(r4, reg_nextpc);
jump(no_update);
break;
default:
ERROR_LOG(DYNAREC, "Error, Relink() Block Type: %X", block->BlockType);
verify(false);
break;
}
return ass.GetCursorOffset() - start_offset;
}
u32 DynaRBI::Relink()
{
ass = Arm32Assembler((u8 *)code + relink_offset, host_code_size - relink_offset);
u32 size = relinkBlock(this);
ass.Finalize();
return size;
}
static Register GetParam(const shil_param& param, Register raddr = r0)
{
if (param.is_imm())
{
ass.Mov(raddr, param._imm);
return raddr;
}
if (param.is_r32i())
return reg.mapReg(param);
die("Invalid parameter");
return Register();
}
static void ngen_Binary(shil_opcode* op, BinaryOP dtop)
{
Register rs1 = GetParam(op->rs1);
Register rs2 = r1;
if (op->rs2.is_imm())
{
if (ImmediateA32::IsImmediateA32(op->rs2._imm))
{
(ass.*dtop)(reg.mapReg(op->rd), rs1, Operand(op->rs2._imm));
return;
}
ass.Mov(rs2, op->rs2._imm);
}
else if (op->rs2.is_r32i())
{
rs2 = reg.mapReg(op->rs2);
}
else
{
ERROR_LOG(DYNAREC, "ngen_Bin ??? %d", op->rs2.type);
verify(false);
}
(ass.*dtop)(reg.mapReg(op->rd), rs1, rs2);
}
static void ngen_fp_bin(shil_opcode* op, FPBinOP fpop)
{
SRegister rs1 = s0;
if (op->rs1.is_imm())
{
ass.Mov(r0, op->rs1._imm);
ass.Vmov(rs1, r0);
}
else
{
rs1 = reg.mapFReg(op->rs1);
}
SRegister rs2 = s1;
if (op->rs2.is_imm())
{
ass.Mov(r0, op->rs2._imm);
ass.Vmov(rs2, r0);
}
else
{
rs2 = reg.mapFReg(op->rs2);
}
(ass.*fpop)(DataType(F32), reg.mapFReg(op->rd), rs1, rs2);
}
static void ngen_fp_una(shil_opcode* op, FPUnOP fpop)
{
(ass.*fpop)(DataType(F32), reg.mapFReg(op->rd), reg.mapFReg(op->rs1));
}
struct CC_PS
{
CanonicalParamType type;
shil_param* par;
};
static std::vector<CC_PS> CC_pars;
void ngen_CC_Start(shil_opcode* op)
{
CC_pars.clear();
}
void ngen_CC_Param(shil_opcode* op,shil_param* par,CanonicalParamType tp)
{
switch(tp)
{
case CPT_f32rv:
#ifdef __ARM_PCS_VFP
// -mfloat-abi=hard
if (reg.IsAllocg(*par))
ass.Vmov(reg.mapReg(*par), s0);
else if (reg.IsAllocf(*par))
ass.Vmov(reg.mapFReg(*par), s0);
break;
#endif
case CPT_u32rv:
case CPT_u64rvL:
if (reg.IsAllocg(*par))
ass.Mov(reg.mapReg(*par), r0);
else if (reg.IsAllocf(*par))
ass.Vmov(reg.mapFReg(*par), r0);
else
die("unhandled param");
break;
case CPT_u64rvH:
verify(reg.IsAllocg(*par));
ass.Mov(reg.mapReg(*par), r1);
break;
case CPT_u32:
case CPT_ptr:
case CPT_f32:
{
CC_PS t = { tp, par };
CC_pars.push_back(t);
}
break;
default:
die("invalid tp");
break;
}
}
void ngen_CC_Call(shil_opcode* op, void* function)
{
Register rd = r0;
SRegister fd = s0;
for (int i = CC_pars.size(); i-- > 0; )
{
CC_PS& param = CC_pars[i];
if (param.type == CPT_ptr)
{
ass.Mov(rd, (u32)param.par->reg_ptr());
}
else
{
if (param.par->is_reg())
{
#ifdef __ARM_PCS_VFP
// -mfloat-abi=hard
if (param.type == CPT_f32)
{
if (reg.IsAllocg(*param.par))
ass.Vmov(fd, reg.mapReg(*param.par));
else if (reg.IsAllocf(*param.par))
ass.Vmov(fd, reg.mapFReg(*param.par));
else
die("Must not happen!");
continue;
}
#endif
if (reg.IsAllocg(*param.par))
ass.Mov(rd, reg.mapReg(*param.par));
else if (reg.IsAllocf(*param.par))
ass.Vmov(rd, reg.mapFReg(*param.par));
else
die("Must not happen!");
}
else
{
verify(param.par->is_imm());
ass.Mov(rd, param.par->_imm);
}
}
rd = Register(rd.GetCode() + 1);
fd = SRegister(fd.GetCode() + 1);
}
call(function);
}
void ngen_CC_Finish(shil_opcode* op)
{
CC_pars.clear();
}
enum mem_op_type
{
SZ_8,
SZ_16,
SZ_32I,
SZ_32F,
SZ_64F,
};
static mem_op_type memop_type(shil_opcode* op)
{
int sz = op->flags & 0x7f;
bool fp32 = op->rs2.is_r32f() || op->rd.is_r32f();
if (sz == 1)
return SZ_8;
else if (sz == 2)
return SZ_16;
else if (sz == 4)
return fp32 ? SZ_32F : SZ_32I;
else if (sz == 8)
return SZ_64F;
die("Unknown op");
return SZ_32I;
}
const u32 memop_bytes[] = { 1, 2, 4, 4, 8 };
static const void *_mem_hndl_SQ32[3][14];
static const void *_mem_hndl[2][3][14];
const void * const _mem_func[2][2] =
{
{ (void *)_vmem_WriteMem32, (void *)_vmem_WriteMem64 },
{ (void *)_vmem_ReadMem32, (void *)_vmem_ReadMem64 },
};
const struct
{
u32 mask;
u32 key;
bool read;
mem_op_type optp;
u32 offs;
}
op_table[] =
{
//LDRSB
{ 0x0E500FF0, 0x001000D0, true, SZ_8, 1 },
//LDRSH
{ 0x0E500FF0, 0x001000F0, true, SZ_16, 1 },
//LDR
{ 0x0E500010, 0x06100000, true, SZ_32I, 1 },
//VLDR.32
{ 0x0F300F00, 0x0D100A00, true, SZ_32F, 2 },
//VLDR.64
{ 0x0F300F00, 0x0D100B00, true, SZ_64F, 2 },
//
//STRB
{ 0x0FF00010, 0x07C00000, false, SZ_8, 1 },
//STRH
{ 0x0FF00FF0, 0x018000B0, false, SZ_16, 1 },
//STR
{ 0x0E500010, 0x06000000, false, SZ_32I, 1 },
//VSTR.32
{ 0x0F300F00, 0x0D000A00, false, SZ_32F, 2 },
//VSTR.64
{ 0x0F300F00, 0x0D000B00, false, SZ_64F, 2 },
{ 0, 0 },
};
union arm_mem_op
{
struct
{
u32 Ra:4;
u32 pad0:8;
u32 Rt:4;
u32 Rn:4;
u32 pad1:2;
u32 D:1;
u32 pad3:1;
u32 pad4:4;
u32 cond:4;
};
u32 full;
};
static void vmem_slowpath(Register raddr, Register rt, SRegister ft, DRegister fd, mem_op_type optp, bool read)
{
if (!raddr.Is(r0))
ass.Mov(r0, raddr);
if (!read)
{
if (optp <= SZ_32I)
ass.Mov(r1, rt);
else if (optp == SZ_32F)
ass.Vmov(r1, ft);
else if (optp == SZ_64F)
ass.Vmov(r2, r3, fd);
}
const void *funct = nullptr;
if (optp <= SZ_32I)
funct = _mem_hndl[read][optp][raddr.GetCode()];
else
funct = _mem_func[read][optp - SZ_32F];
verify(funct != nullptr);
call(funct);
if (read)
{
if (optp <= SZ_32I)
ass.Mov(rt, r0);
else if (optp == SZ_32F)
ass.Vmov(ft, r0);
else if (optp == SZ_64F)
ass.Vmov(fd, r0, r1);
}
}
bool ngen_Rewrite(host_context_t &context, void *faultAddress)
{
u32 *regs = context.reg;
arm_mem_op *ptr = (arm_mem_op *)context.pc;
mem_op_type optp;
u32 read;
s32 offs = -1;
u32 fop = ptr[0].full;
for (int i = 0; op_table[i].mask; i++)
{
if ((fop & op_table[i].mask) == op_table[i].key)
{
optp = op_table[i].optp;
read = op_table[i].read;
offs = op_table[i].offs;
}
}
if (offs == -1)
{
ERROR_LOG(DYNAREC, "%08X : invalid size", fop);
die("can't decode opcode\n");
}
ptr -= offs;
Register raddr, rt;
SRegister ft;
DRegister fd;
//Get used regs from opcodes ..
if ((ptr[0].full & 0x0FE00070) == 0x07E00050)
{
//from ubfx !
raddr = Register(ptr[0].Ra);
}
else if ((ptr[0].full & 0x0FE00000) == 0x03C00000)
{
raddr = Register(ptr[0].Rn);
}
else
{
ERROR_LOG(DYNAREC, "fail raddr %08X {@%08X}:(", ptr[0].full, regs[1]);
die("Invalid opcode: vmem fixup\n");
}
//from mem op
rt = Register(ptr[offs].Rt);
ft = SRegister(ptr[offs].Rt * 2 + ptr[offs].D);
fd = DRegister(ptr[offs].D * 16 + ptr[offs].Rt);
//get some other relevant data
u32 sh4_addr = regs[raddr.GetCode()];
u32 fault_offs = (uintptr_t)faultAddress - regs[8];
bool is_sq = (sh4_addr >> 26) == 0x38;
ass = Arm32Assembler((u8 *)ptr, 12);
//fault offset must always be the addr from ubfx (sanity check)
verify(fault_offs == 0 || fault_offs == (sh4_addr & 0x1FFFFFFF));
if (is_sq && !read && optp >= SZ_32I)
{
if (optp >= SZ_32F)
{
if (!raddr.Is(r0))
ass.Mov(r0, raddr);
else
ass.Nop();
raddr = r0;
}
switch (optp)
{
case SZ_32I:
ass.Mov(r1, rt);
break;
case SZ_32F:
ass.Vmov(r1, ft);
break;
case SZ_64F:
ass.Vmov(r2, r3, fd);
break;
default:
break;
}
call(_mem_hndl_SQ32[optp - SZ_32I][raddr.GetCode()]);
}
else
{
//Fallback to function !
if (optp >= SZ_32F)
{
if (!raddr.Is(r0))
ass.Mov(r0, raddr);
else
ass.Nop();
}
if (!read)
{
if (optp <= SZ_32I)
ass.Mov(r1, rt);
else if (optp == SZ_32F)
ass.Vmov(r1, ft);
else if (optp == SZ_64F)
ass.Vmov(r2, r3, fd);
}
const void *funct = nullptr;
if (offs == 1)
funct = _mem_hndl[read][optp][raddr.GetCode()];
else if (optp >= SZ_32F)
funct = _mem_func[read][optp - SZ_32F];
verify(funct != nullptr);
call(funct);
if (read)
{
if (optp <= SZ_32I)
ass.Mov(rt, r0);
else if (optp == SZ_32F)
ass.Vmov(ft, r0);
else if (optp == SZ_64F)
ass.Vmov(fd, r0, r1);
}
}
ass.Finalize();
context.pc = (size_t)ptr;
return true;
}
static Register GenMemAddr(shil_opcode* op, Register raddr = r0)
{
if (op->rs3.is_imm())
{
if (ImmediateA32::IsImmediateA32(op->rs3._imm))
{
ass.Add(raddr, reg.mapReg(op->rs1), op->rs3._imm);
}
else
{
ass.Mov(r1, op->rs3._imm);
ass.Add(raddr, reg.mapReg(op->rs1), r1);
}
}
else if (op->rs3.is_r32i())
{
ass.Add(raddr, reg.mapReg(op->rs1), reg.mapReg(op->rs3));
}
else if (!op->rs3.is_null())
{
ERROR_LOG(DYNAREC, "rs3: %08X", op->rs3.type);
die("invalid rs3");
}
else if (op->rs1.is_imm())
{
ass.Mov(raddr, op->rs1._imm);
}
else
{
raddr = reg.mapReg(op->rs1);
}
return raddr;
}
static bool ngen_readm_immediate(RuntimeBlockInfo* block, shil_opcode* op, bool optimise)
{
if (!op->rs1.is_imm())
return false;
mem_op_type optp = memop_type(op);
bool isram = false;
void* ptr = _vmem_read_const(op->rs1._imm, isram, std::min(4u, memop_bytes[optp]));
Register rd = (optp != SZ_32F && optp != SZ_64F) ? reg.mapReg(op->rd) : r0;
if (isram)
{
ass.Mov(r0, (u32)ptr);
switch(optp)
{
case SZ_8:
ass.Ldrsb(rd, MemOperand(r0));
break;
case SZ_16:
ass.Ldrsh(rd, MemOperand(r0));
break;
case SZ_32I:
ass.Ldr(rd, MemOperand(r0));
break;
case SZ_32F:
ass.Vldr(reg.mapFReg(op->rd), MemOperand(r0));
break;
case SZ_64F:
ass.Vldr(d0, MemOperand(r0));
ass.Vstr(d0, MemOperand(r8, op->rd.reg_nofs()));
break;
}
}
else
{
// Not RAM
if (optp == SZ_64F)
{
verify(!reg.IsAllocAny(op->rd));
// Need to call the handler twice
ass.Mov(r0, op->rs1._imm);
call(ptr);
ass.Str(r0, MemOperand(r8, op->rd.reg_nofs()));
ass.Mov(r0, op->rs1._imm + 4);
call(ptr);
ass.Str(r0, MemOperand(r8, op->rd.reg_nofs() + 4));
}
else
{
ass.Mov(r0, op->rs1._imm);
call(ptr);
switch(optp)
{
case SZ_8:
ass.Sxtb(r0, r0);
break;
case SZ_16:
ass.Sxth(r0, r0);
break;
case SZ_32I:
case SZ_32F:
break;
default:
die("Invalid size");
break;
}
if (reg.IsAllocg(op->rd))
ass.Mov(rd, r0);
else if (reg.IsAllocf(op->rd))
ass.Vmov(reg.mapFReg(op->rd), r0);
else
die("Unsupported");
}
}
return true;
}
static bool ngen_writemem_immediate(RuntimeBlockInfo* block, shil_opcode* op, bool optimise)
{
if (!op->rs1.is_imm())
return false;
mem_op_type optp = memop_type(op);
bool isram = false;
void* ptr = _vmem_write_const(op->rs1._imm, isram, std::min(4u, memop_bytes[optp]));
Register rs2 = r1;
SRegister rs2f = s0;
if (op->rs2.is_imm())
ass.Mov(rs2, op->rs2._imm);
else if (optp == SZ_32F)
rs2f = reg.mapFReg(op->rs2);
else if (optp != SZ_64F)
rs2 = reg.mapReg(op->rs2);
if (isram)
{
ass.Mov(r0, (u32)ptr);
switch(optp)
{
case SZ_8:
ass.Strb(rs2, MemOperand(r0));
break;
case SZ_16:
ass.Strh(rs2, MemOperand(r0));
break;
case SZ_32I:
ass.Str(rs2, MemOperand(r0));
break;
case SZ_32F:
ass.Vstr(rs2f, MemOperand(r0));
break;
case SZ_64F:
ass.Vldr(d0, MemOperand(r8, op->rs2.reg_nofs()));
ass.Vstr(d0, MemOperand(r0));
break;
default:
die("Invalid size");
break;
}
}
else
{
if (optp == SZ_64F)
die("SZ_64F not supported");
ass.Mov(r0, op->rs1._imm);
if (optp == SZ_32F)
ass.Vmov(r1, rs2f);
else if (!rs2.Is(r1))
ass.Mov(r1, rs2);
call(ptr);
}
return true;
}
static void ngen_compile_opcode(RuntimeBlockInfo* block, shil_opcode* op, bool optimise)
{
switch(op->op)
{
case shop_readm:
if (!ngen_readm_immediate(block, op, optimise))
{
mem_op_type optp = memop_type(op);
Register raddr = GenMemAddr(op);
if (_nvmem_enabled()) {
ass.Bic(r1, raddr, 0xE0000000);
switch(optp)
{
case SZ_8:
ass.Ldrsb(reg.mapReg(op->rd), MemOperand(r1, r8));
break;
case SZ_16:
ass.Ldrsh(reg.mapReg(op->rd), MemOperand(r1, r8));
break;
case SZ_32I:
ass.Ldr(reg.mapReg(op->rd), MemOperand(r1, r8));
break;
case SZ_32F:
ass.Add(r1, r1, r8); //3 opcodes, there's no [REG+REG] VLDR
ass.Vldr(reg.mapFReg(op->rd), MemOperand(r1));
break;
case SZ_64F:
ass.Add(r1, r1, r8); //3 opcodes, there's no [REG+REG] VLDR
ass.Vldr(d0, MemOperand(r1)); //TODO: use reg alloc
ass.Vstr(d0, MemOperand(r8, op->rd.reg_nofs()));
break;
}
} else {
switch(optp)
{
case SZ_8:
vmem_slowpath(raddr, reg.mapReg(op->rd), s0, d0, optp, true);
break;
case SZ_16:
vmem_slowpath(raddr, reg.mapReg(op->rd), s0, d0, optp, true);
break;
case SZ_32I:
vmem_slowpath(raddr, reg.mapReg(op->rd), s0, d0, optp, true);
break;
case SZ_32F:
vmem_slowpath(raddr, r0, reg.mapFReg(op->rd), d0, optp, true);
break;
case SZ_64F:
vmem_slowpath(raddr, r0, s0, d0, optp, true);
ass.Vstr(d0, MemOperand(r8, op->rd.reg_nofs()));
break;
}
}
}
break;
case shop_writem:
if (!ngen_writemem_immediate(block, op, optimise))
{
mem_op_type optp = memop_type(op);
Register raddr = GenMemAddr(op);
Register rs2 = r2;
SRegister rs2f = s2;
//TODO: use reg alloc
if (optp == SZ_64F)
ass.Vldr(d0, MemOperand(r8, op->rs2.reg_nofs()));
else if (op->rs2.is_imm())
{
ass.Mov(rs2, op->rs2._imm);
if (optp == SZ_32F)
ass.Vmov(rs2f, rs2);
}
else
{
if (optp == SZ_32F)
rs2f = reg.mapFReg(op->rs2);
else
rs2 = reg.mapReg(op->rs2);
}
if (_nvmem_enabled())
{
ass.Bic(r1, raddr, 0xE0000000);
switch(optp)
{
case SZ_8:
ass.Strb(rs2, MemOperand(r1, r8));
break;
case SZ_16:
ass.Strh(rs2, MemOperand(r1, r8));
break;
case SZ_32I:
ass.Str(rs2, MemOperand(r1, r8));
break;
case SZ_32F:
ass.Add(r1, r1, r8); //3 opcodes: there's no [REG+REG] VLDR, also required for SQ
ass.Vstr(rs2f, MemOperand(r1));
break;
case SZ_64F:
ass.Add(r1, r1, r8); //3 opcodes: there's no [REG+REG] VLDR, also required for SQ
ass.Vstr(d0, MemOperand(r1)); //TODO: use reg alloc
break;
}
} else {
switch(optp)
{
case SZ_8:
vmem_slowpath(raddr, rs2, s0, d0, optp, false);
break;
case SZ_16:
vmem_slowpath(raddr, rs2, s0, d0, optp, false);
break;
case SZ_32I:
vmem_slowpath(raddr, rs2, s0, d0, optp, false);
break;
case SZ_32F:
vmem_slowpath(raddr, r0, rs2f, d0, optp, false);
break;
case SZ_64F:
vmem_slowpath(raddr, r0, s0, d0, optp, false);
break;
}
}
}
break;
//dynamic jump, r+imm32.This will be at the end of the block, but doesn't -have- to be the last opcode
case shop_jdyn:
verify(op->rd.is_reg() && op->rd._reg == reg_pc_dyn);
if (op->rs2.is_imm())
{
ass.Mov(r2, op->rs2.imm_value());
ass.Add(r4, reg.mapReg(op->rs1), r2);
}
else
{
ass.Mov(r4, reg.mapReg(op->rs1));
}
break;
case shop_mov32:
verify(op->rd.is_r32());
if (op->rs1.is_imm())
{
if (op->rd.is_r32i())
{
ass.Mov(reg.mapReg(op->rd), op->rs1._imm);
}
else
{
if (op->rs1._imm==0)
{
//VEOR(reg.mapFReg(op->rd),reg.mapFReg(op->rd),reg.mapFReg(op->rd));
//hum, vmov can't do 0, but can do all kind of weird small consts ... really useful ...
//simd is slow on a9
#if 0
ass.Movw(r0, 0);
ass.Vmov(reg.mapFReg(op->rd), r0);
#else
//1-1=0 !
//should be slightly faster ...
//we could get rid of the imm mov, if not for infs & co ..
ass.Vmov(reg.mapFReg(op->rd), 1.f);;
ass.Vsub(reg.mapFReg(op->rd), reg.mapFReg(op->rd), reg.mapFReg(op->rd));
#endif
}
else if (op->rs1._imm == 0x3F800000)
ass.Vmov(reg.mapFReg(op->rd), 1.f);
else
{
ass.Mov(r0, op->rs1._imm);
ass.Vmov(reg.mapFReg(op->rd), r0);
}
}
}
else if (op->rs1.is_r32())
{
u32 type = 0;
if (reg.IsAllocf(op->rd))
type |= 1;
if (reg.IsAllocf(op->rs1))
type |= 2;
switch(type)
{
case 0: // reg = reg
ass.Mov(reg.mapReg(op->rd), reg.mapReg(op->rs1));
break;
case 1: // vfp = reg
ass.Vmov(reg.mapFReg(op->rd), reg.mapReg(op->rs1));
break;
case 2: // reg = vfp
ass.Vmov(reg.mapReg(op->rd), reg.mapFReg(op->rs1));
break;
case 3: // vfp = vfp
ass.Vmov(reg.mapFReg(op->rd), reg.mapFReg(op->rs1));
break;
}
}
else
{
die("Invalid mov32 size");
}
break;
case shop_mov64:
verify(op->rs1.is_r64() && op->rd.is_r64());
ass.Vldr(d0, MemOperand(r8, op->rs1.reg_nofs()));
ass.Vstr(d0, MemOperand(r8, op->rd.reg_nofs()));
break;
case shop_jcond:
verify(op->rd.is_reg() && op->rd._reg == reg_pc_dyn);
ass.Mov(r4, reg.mapReg(op->rs1));
break;
case shop_ifb:
if (op->rs1._imm)
{
ass.Mov(r1, op->rs2._imm);
storeSh4Reg(r1, reg_nextpc);
}
ass.Mov(r0, op->rs3._imm);
call((void *)OpPtr[op->rs3._imm]);
break;
#ifndef CANONICALTEST
case shop_neg:
ass.Rsb(reg.mapReg(op->rd), reg.mapReg(op->rs1), 0);
break;
case shop_not:
ass.Mvn(reg.mapReg(op->rd), reg.mapReg(op->rs1));
break;
case shop_shl:
ngen_Binary(op, &MacroAssembler::Lsl);
break;
case shop_shr:
ngen_Binary(op, &MacroAssembler::Lsr);
break;
case shop_sar:
ngen_Binary(op, &MacroAssembler::Asr);
break;
case shop_and:
ngen_Binary(op, &MacroAssembler::And);
break;
case shop_or:
ngen_Binary(op, &MacroAssembler::Orr);
break;
case shop_xor:
ngen_Binary(op, &MacroAssembler::Eor);
break;
case shop_add:
ngen_Binary(op, &MacroAssembler::Add);
break;
case shop_sub:
ngen_Binary(op, &MacroAssembler::Sub);
break;
case shop_ror:
ngen_Binary(op, &MacroAssembler::Ror);
break;
case shop_adc:
{
Register rs1 = GetParam(op->rs1, r1);
Register rs2 = GetParam(op->rs2, r2);
Register rs3 = GetParam(op->rs3, r3);
ass.Lsr(SetFlags, r0, rs3, 1); //C=rs3, r0=0
ass.Adc(SetFlags, reg.mapReg(op->rd), rs1, rs2); //(C,rd)=rs1+rs2+rs3(C)
ass.Adc(reg.mapReg(op->rd2), r0, 0); //rd2=C, (or MOVCS rd2, 1)
}
break;
case shop_rocr:
{
Register rd2 = reg.mapReg(op->rd2);
Register rs1 = GetParam(op->rs1, r1);
Register rs2 = GetParam(op->rs2, r2);
if (!rd2.Is(rs1)) {
ass.Lsr(SetFlags, rd2, rs2, 1); //C=rs2, rd2=0
ass.And(rd2, rs1, 1); //get new carry
} else {
ass.Lsr(SetFlags, r0, rs2, 1); //C=rs2, rd2=0
ass.Add(r0, rs1, 1);
}
ass.Rrx(reg.mapReg(op->rd), rs1); //RRX w/ carry :)
if (rd2.Is(rs1))
ass.Mov(rd2, r0);
}
break;
case shop_rocl:
{
Register rs1 = GetParam(op->rs1, r1);
Register rs2 = GetParam(op->rs2, r2);
ass.Orr(SetFlags, reg.mapReg(op->rd), rs2, Operand(rs1, LSL, 1)); //(C,rd)= rs1<<1 + (|) rs2
ass.Mov(reg.mapReg(op->rd2), 0); //clear rd2 (for ADC/MOVCS)
ass.Adc(reg.mapReg(op->rd2), reg.mapReg(op->rd2), 0); //rd2=C (or MOVCS rd2, 1)
}
break;
case shop_sbc:
{
Register rd2 = reg.mapReg(op->rd2);
Register rs1 = GetParam(op->rs1, r1);
if (rs1.Is(rd2))
{
ass.Mov(r1, rs1);
rs1 = r1;
}
Register rs2 = GetParam(op->rs2, r2);
if (rs2.Is(rd2))
{
ass.Mov(r2, rs2);
rs2 = r2;
}
Register rs3 = GetParam(op->rs3, r3);
ass.Eor(rd2, rs3, 1);
ass.Lsr(SetFlags, rd2, rd2, 1); //C=rs3, rd2=0
ass.Sbc(SetFlags, reg.mapReg(op->rd), rs1, rs2);
ass.Mov(cc, rd2, 1);
}
break;
case shop_negc:
{
Register rd2 = reg.mapReg(op->rd2);
Register rs1 = GetParam(op->rs1, r1);
if (rs1.Is(rd2))
{
ass.Mov(r1, rs1);
rs1 = r1;
}
Register rs2 = GetParam(op->rs2, r2);
ass.Eor(rd2, rs2, 1);
ass.Lsr(SetFlags, rd2, rd2, 1); //C=rs3, rd2=0
ass.Sbc(SetFlags, reg.mapReg(op->rd), rd2, rs1); // rd2 == 0
ass.Mov(cc, rd2, 1);
}
break;
case shop_shld:
{
verify(!op->rs2.is_imm());
ass.And(SetFlags, r0, reg.mapReg(op->rs2), 0x8000001F);
ass.Rsb(mi, r0, r0, 0x80000020);
Register rs1 = GetParam(op->rs1, r1);
ass.Lsr(mi, reg.mapReg(op->rd), rs1, r0);
ass.Lsl(pl, reg.mapReg(op->rd), rs1, r0);
}
break;
case shop_shad:
{
verify(!op->rs2.is_imm());
ass.And(SetFlags, r0, reg.mapReg(op->rs2), 0x8000001F);
ass.Rsb(mi, r0, r0, 0x80000020);
Register rs1 = GetParam(op->rs1, r1);
ass.Asr(mi, reg.mapReg(op->rd), rs1, r0);
ass.Lsl(pl, reg.mapReg(op->rd), rs1, r0);
}
break;
case shop_sync_sr:
//must flush: SRS, SRT, r0-r7, r0b-r7b
call((void *)UpdateSR);
break;
case shop_test:
case shop_seteq:
case shop_setge:
case shop_setgt:
case shop_setae:
case shop_setab:
{
Register rd = reg.mapReg(op->rd);
Register rs1 = GetParam(op->rs1, r0);
Register rs2 = r1;
bool is_imm = false;
if (op->rs2.is_imm())
{
if (!ImmediateA32::IsImmediateA32(op->rs2._imm))
ass.Mov(rs2, (u32)op->rs2._imm);
else
is_imm = true;
}
else if (op->rs2.is_r32i())
{
rs2 = reg.mapReg(op->rs2);
}
else
{
ERROR_LOG(DYNAREC, "ngen_Bin ??? %d", op->rs2.type);
verify(false);
}
if (op->op == shop_test)
{
if (is_imm)
ass.Tst(rs1, op->rs2._imm);
else
ass.Tst(rs1, rs2);
}
else
{
if (is_imm)
ass.Cmp(rs1, op->rs2._imm);
else
ass.Cmp(rs1, rs2);
}
static const ConditionType opcls2[] = { eq, eq, ge, gt, hs, hi };
ass.Mov(rd, 0);
ass.Mov(opcls2[op->op-shop_test], rd, 1);
}
break;
case shop_setpeq:
{
Register rs1 = GetParam(op->rs1, r1);
Register rs2 = GetParam(op->rs2, r2);
ass.Eor(r1, rs1, rs2);
ass.Mov(reg.mapReg(op->rd), 0);
ass.Tst(r1, 0xFF000000u);
ass.Tst(ne, r1, 0x00FF0000u);
ass.Tst(ne, r1, 0x0000FF00u);
ass.Tst(ne, r1, 0x000000FFu);
ass.Mov(eq, reg.mapReg(op->rd), 1);
}
break;
//UXTH for zero extention and/or more mul forms (for 16 and 64 bits)
case shop_mul_u16:
{
Register rs2 = GetParam(op->rs2, r2);
ass.Uxth(r1, reg.mapReg(op->rs1));
ass.Uxth(r2, rs2);
ass.Mul(reg.mapReg(op->rd), r1, r2);
}
break;
case shop_mul_s16:
{
Register rs2 = GetParam(op->rs2, r2);
ass.Sxth(r1, reg.mapReg(op->rs1));
ass.Sxth(r2, rs2);
ass.Mul(reg.mapReg(op->rd), r1, r2);
}
break;
case shop_mul_i32:
{
Register rs2 = GetParam(op->rs2, r2);
//x86_opcode_class opdt[]={op_movzx16to32,op_movsx16to32,op_mov32,op_mov32,op_mov32};
//x86_opcode_class opmt[]={op_mul32,op_mul32,op_mul32,op_mul32,op_imul32};
//only the top 32 bits are different on signed vs unsigned
ass.Mul(reg.mapReg(op->rd), reg.mapReg(op->rs1), rs2);
}
break;
case shop_mul_u64:
{
Register rs2 = GetParam(op->rs2, r2);
ass.Umull(reg.mapReg(op->rd), reg.mapReg(op->rd2), reg.mapReg(op->rs1), rs2);
}
break;
case shop_mul_s64:
{
Register rs2 = GetParam(op->rs2, r2);
ass.Smull(reg.mapReg(op->rd), reg.mapReg(op->rd2), reg.mapReg(op->rs1), rs2);
}
break;
case shop_pref:
{
ConditionType cc = eq;
if (!op->rs1.is_imm())
{
ass.Lsr(r1, reg.mapReg(op->rs1), 26);
ass.Mov(r0, reg.mapReg(op->rs1));
ass.Cmp(r1, 0x38);
}
else
{
// The SSA pass has already checked that the
// destination is a store queue so no need to check
ass.Mov(r0, op->rs1.imm_value());
cc = al;
}
if (CCN_MMUCR.AT)
{
call((void *)do_sqw_mmu, cc);
}
else
{
ass.Ldr(r2, MemOperand(r8, rcbOffset(do_sqw_nommu)));
ass.Sub(r1, r8, -rcbOffset(sq_buffer));
ass.Blx(cc, r2);
}
}
break;
case shop_ext_s8:
case shop_ext_s16:
if (op->op == shop_ext_s8)
ass.Sxtb(reg.mapReg(op->rd), reg.mapReg(op->rs1));
else
ass.Sxth(reg.mapReg(op->rd), reg.mapReg(op->rs1));
break;
case shop_xtrct:
{
Register rd = reg.mapReg(op->rd);
Register rs1 = reg.mapReg(op->rs1);
Register rs2 = reg.mapReg(op->rs2);
if (rd.Is(rs1))
{
verify(!rd.Is(rs2));
ass.Lsr(rd, rs1, 16);
ass.Lsl(r0, rs2, 16);
}
else
{
ass.Lsl(rd, rs2, 16);
ass.Lsr(r0, rs1, 16);
}
ass.Orr(rd, rd, r0);
}
break;
//
// FPU
//
case shop_fadd:
case shop_fsub:
case shop_fmul:
case shop_fdiv:
{
static const FPBinOP opcds[] = {
&MacroAssembler::Vadd, &MacroAssembler::Vsub, &MacroAssembler::Vmul, &MacroAssembler::Vdiv
};
ngen_fp_bin(op, opcds[op->op - shop_fadd]);
}
break;
case shop_fabs:
case shop_fneg:
{
static const FPUnOP opcds[] = { &MacroAssembler::Vabs, &MacroAssembler::Vneg };
ngen_fp_una(op, opcds[op->op - shop_fabs]);
}
break;
case shop_fsqrt:
ngen_fp_una(op, &MacroAssembler::Vsqrt);
break;
case shop_fmac:
{
SRegister rd = reg.mapFReg(op->rd);
SRegister rs1 = s1;
if (op->rs1.is_imm())
{
ass.Mov(r0, op->rs1.imm_value());
ass.Vmov(rs1, r0);
}
else
rs1 = reg.mapFReg(op->rs1);
SRegister rs2 = s2;
if (op->rs2.is_imm())
{
ass.Mov(r1, op->rs2.imm_value());
ass.Vmov(rs2, r1);
}
else
{
rs2 = reg.mapFReg(op->rs2);
if (rs2.Is(rd))
{
ass.Vmov(s2, rs2);
rs2 = s2;
}
}
SRegister rs3 = s3;
if (op->rs3.is_imm())
{
ass.Mov(r2, op->rs3.imm_value());
ass.Vmov(rs3, r2);
}
else
{
rs3 = reg.mapFReg(op->rs3);
if (rs3.Is(rd))
{
ass.Vmov(s3, rs3);
rs3 = s3;
}
}
if (!rd.Is(rs1))
ass.Vmov(rd, rs1);
ass.Vmla(rd, rs2, rs3);
}
break;
case shop_fsrra:
ass.Vmov(s1, 1.f);
ass.Vsqrt(s0, reg.mapFReg(op->rs1));
ass.Vdiv(reg.mapFReg(op->rd), s1, s0);
break;
case shop_fsetgt:
case shop_fseteq:
#if 1
//this is apparently much faster (tested on A9)
ass.Mov(reg.mapReg(op->rd), 0);
ass.Vcmp(reg.mapFReg(op->rs1), reg.mapFReg(op->rs2));
ass.Vmrs(RegisterOrAPSR_nzcv(APSR_nzcv), FPSCR);
if (op->op == shop_fsetgt)
ass.Mov(gt, reg.mapReg(op->rd), 1);
else
ass.Mov(eq, reg.mapReg(op->rd), 1);
#else
if (op->op == shop_fsetgt)
ass.Vcgt(d0, reg.mapFReg(op->rs1), reg.mapFReg(op->rs2));
else
ass.Vceq(d0, reg.mapFReg(op->rs1), reg.mapFReg(op->rs2));
ass.Vmov(r0, s0);
ass.And(reg.mapReg(op->rd), r0, 1);
#endif
break;
case shop_fsca:
//r1: base ptr
ass.Mov(r1, (u32)sin_table & 0xFFFF);
ass.Uxth(r0, reg.mapReg(op->rs1));
ass.Movt(r1, (u32)sin_table >> 16);
ass.Add(r0, r1, Operand(r0, LSL, 3));
ass.Vldr(d0, MemOperand(r0));
ass.Vstr(d0, MemOperand(r8, op->rd.reg_nofs()));
break;
case shop_fipr:
{
QRegister _r1 = q0;
QRegister _r2 = q0;
ass.Sub(r0, r8, op->rs1.reg_aofs());
if (op->rs2.reg_aofs() == op->rs1.reg_aofs())
{
ass.Vldm(r0, NO_WRITE_BACK, DRegisterList(d0, 2));
}
else
{
ass.Sub(r1, r8, op->rs2.reg_aofs());
ass.Vldm(r0, NO_WRITE_BACK, DRegisterList(d0, 2));
ass.Vldm(r1, NO_WRITE_BACK, DRegisterList(d2, 2));
_r2 = q1;
}
#if 1
//VFP
SRegister fs2 = _r2.Is(q0) ? s0 : s4;
ass.Vmul(reg.mapFReg(op->rd), s0, fs2);
ass.Vmla(reg.mapFReg(op->rd), s1, SRegister(fs2.GetCode() + 1));
ass.Vmla(reg.mapFReg(op->rd), s2, SRegister(fs2.GetCode() + 2));
ass.Vmla(reg.mapFReg(op->rd), s3, SRegister(fs2.GetCode() + 3));
#else
ass.Vmul(q0, _r1, _r2);
ass.Vpadd(d0, d0, d1);
ass.Vadd(reg.mapFReg(op->rd), f0, f1);
#endif
}
break;
case shop_ftrv:
{
Register rdp = r1;
ass.Sub(r2, r8, op->rs2.reg_aofs());
ass.Sub(r1, r8, op->rs1.reg_aofs());
if (op->rs1.reg_aofs() != op->rd.reg_aofs())
{
rdp = r0;
ass.Sub(r0, r8, op->rd.reg_aofs());
}
#if 1
//f0,f1,f2,f3 : vin
//f4,f5,f6,f7 : out
//f8,f9,f10,f11 : mtx temp
//f12,f13,f14,f15 : mtx temp
//(This is actually faster than using neon)
ass.Vldm(r2, WRITE_BACK, DRegisterList(d4, 2));
ass.Vldm(r1, NO_WRITE_BACK, DRegisterList(d0, 2));
ass.Vmul(s4, vixl::aarch32::s8, s0);
ass.Vmul(s5, s9, s0);
ass.Vmul(s6, s10, s0);
ass.Vmul(s7, s11, s0);
ass.Vldm(r2, WRITE_BACK, DRegisterList(d6, 2));
ass.Vmla(s4, s12, s1);
ass.Vmla(s5, s13, s1);
ass.Vmla(s6, s14, s1);
ass.Vmla(s7, s15, s1);
ass.Vldm(r2, WRITE_BACK, DRegisterList(d4, 2));
ass.Vmla(s4, vixl::aarch32::s8, s2);
ass.Vmla(s5, s9, s2);
ass.Vmla(s6, s10, s2);
ass.Vmla(s7, s11, s2);
ass.Vldm(r2, NO_WRITE_BACK, DRegisterList(d6, 2));
ass.Vmla(s4, s12, s3);
ass.Vmla(s5, s13, s3);
ass.Vmla(s6, s14, s3);
ass.Vmla(s7, s15, s3);
ass.Vstm(rdp, NO_WRITE_BACK, DRegisterList(d2, 2));
#else
//this fits really nicely to NEON !
// TODO
ass.Vldm(d16,r2,8);
ass.Vldm(d0,r1,2);
ass.Vmla(q2,q8,d0,0);
ass.Vmla(q2,q9,d0,1);
ass.Vmla(q2,q10,d1,0);
ass.Vmla(q2,q11,d1,1);
ass.Vstm(d4,rdp,2);
#endif
}
break;
case shop_frswap:
ass.Sub(r0, r8, op->rs1.reg_aofs());
ass.Sub(r1, r8, op->rd.reg_aofs());
//Assumes no FPU reg alloc here
//frswap touches all FPU regs, so all spans should be clear here ..
ass.Vldm(r1, NO_WRITE_BACK, DRegisterList(d0, 8));
ass.Vldm(r0, NO_WRITE_BACK, DRegisterList(d8, 8));
ass.Vstm(r0, NO_WRITE_BACK, DRegisterList(d0, 8));
ass.Vstm(r1, NO_WRITE_BACK, DRegisterList(d8, 8));
break;
case shop_cvt_f2i_t:
ass.Vcvt(S32, F32, s0, reg.mapFReg(op->rs1));
ass.Vmov(reg.mapReg(op->rd), s0);
break;
case shop_cvt_i2f_n: // may be some difference should be made ?
case shop_cvt_i2f_z:
ass.Vmov(s0, reg.mapReg(op->rs1));
ass.Vcvt(F32, S32, reg.mapFReg(op->rd), s0);
break;
#endif
default:
shil_chf[op->op](op);
break;
}
}
void ngen_Compile(RuntimeBlockInfo* block, bool force_checks, bool reset, bool staging, bool optimise)
{
ass = Arm32Assembler((u8 *)emit_GetCCPtr(), emit_FreeSpace());
block->code = (DynarecCodeEntryPtr)emit_GetCCPtr();
//reg alloc
reg.DoAlloc(block, alloc_regs, alloc_fpu);
u8* blk_start = ass.GetCursorAddress<u8 *>();
//pre-load the first reg alloc operations, for better efficiency ..
if (!block->oplist.empty())
reg.OpBegin(&block->oplist[0], 0);
//scheduler
if (force_checks)
{
s32 sz = block->sh4_code_size;
u32 addr = block->addr;
ass.Mov(r0, addr);
while (sz > 0)
{
if (sz > 2)
{
u32* ptr = (u32*)GetMemPtr(addr, 4);
if (ptr != nullptr)
{
ass.Mov(r2, (u32)ptr);
ass.Ldr(r2, MemOperand(r2));
ass.Mov(r1, *ptr);
ass.Cmp(r1, r2);
jump(ngen_blockcheckfail, ne);
}
addr += 4;
sz -= 4;
}
else
{
u16* ptr = (u16 *)GetMemPtr(addr, 2);
if (ptr != nullptr)
{
ass.Mov(r2, (u32)ptr);
ass.Ldrh(r2, MemOperand(r2));
ass.Mov(r1, *ptr);
ass.Cmp(r1, r2);
jump(ngen_blockcheckfail, ne);
}
addr += 2;
sz -= 2;
}
}
}
u32 cyc = block->guest_cycles;
if (!ImmediateA32::IsImmediateA32(cyc))
cyc &= ~3;
ass.Sub(SetFlags, r9, r9, cyc);
call(intc_sched, le);
//compile the block's opcodes
shil_opcode* op;
for (size_t i = 0; i < block->oplist.size(); i++)
{
op = &block->oplist[i];
op->host_offs = ass.GetCursorOffset();
if (i != 0)
reg.OpBegin(op, i);
ngen_compile_opcode(block, op, optimise);
reg.OpEnd(op);
}
if (block->BlockType == BET_Cond_0 || block->BlockType == BET_Cond_1)
{
// Store the arm reg containing sr.T in the block
// This will be used when the block in (re)linked
const shil_param param = shil_param(reg_sr_T);
if (reg.IsAllocg(param))
((DynaRBI *)block)->T_reg = reg.mapReg(param);
else
((DynaRBI *)block)->T_reg = Register();
}
reg.Cleanup();
//try to early-lookup the blocks -- to avoid rewrites in case they exist ...
//this isn't enabled for now, as the shared_ptr to this block doesn't exist yet
/*
if (block->BranchBlock != NullAddress)
{
block->pBranchBlock = bm_GetBlock(block->BranchBlock).get();
if (block->pNextBlock)
block->pNextBlock->AddRef(block);
}
if (block->NextBlock != NullAddress)
{
block->pNextBlock = bm_GetBlock(block->NextBlock).get();
if (block->pBranchBlock)
block->pBranchBlock->AddRef(block);
}
*/
//Relink written bytes must be added to the count !
block->relink_offset = ass.GetCursorOffset();
block->relink_data = 0;
emit_Skip(block->relink_offset);
emit_Skip(relinkBlock((DynaRBI *)block));
ass.Finalize();
u8* pEnd = ass.GetCursorAddress<u8 *>();
//blk_start might not be the same, due to profiling counters ..
block->host_opcodes = (pEnd - blk_start) / 4;
//host code size needs to cover the entire range of the block
block->host_code_size = pEnd - (u8*)block->code;
}
void ngen_ResetBlocks()
{
INFO_LOG(DYNAREC, "@@ ngen_ResetBlocks()");
}
void ngen_init()
{
INFO_LOG(DYNAREC, "Initializing the ARM32 dynarec");
ass = Arm32Assembler((u8 *)emit_GetCCPtr(), emit_FreeSpace());
// Stubs
Label ngen_LinkBlock_Shared_stub;
// ngen_LinkBlock_Generic_stub
ngen_LinkBlock_Generic_stub = ass.GetCursorAddress<const void *>();
ass.Mov(r1,r4); // djump/pc -> in case we need it ..
ass.B(&ngen_LinkBlock_Shared_stub);
// ngen_LinkBlock_cond_Branch_stub
ngen_LinkBlock_cond_Branch_stub = ass.GetCursorAddress<const void *>();
ass.Mov(r1, 1);
ass.B(&ngen_LinkBlock_Shared_stub);
// ngen_LinkBlock_cond_Next_stub
ngen_LinkBlock_cond_Next_stub = ass.GetCursorAddress<const void *>();
ass.Mov(r1, 0);
ass.B(&ngen_LinkBlock_Shared_stub);
// ngen_LinkBlock_Shared_stub
ass.Bind(&ngen_LinkBlock_Shared_stub);
ass.Mov(r0, lr);
ass.Sub(r0, r0, 4); // go before the call
call((void *)rdv_LinkBlock);
ass.Bx(r0);
// ngen_FailedToFindBlock_
ngen_FailedToFindBlock_ = ass.GetCursorAddress<void (*)()>();
ass.Mov(r0, r4);
call((void *)rdv_FailedToFindBlock);
ass.Bx(r0);
// ngen_blockcheckfail
ngen_blockcheckfail = ass.GetCursorAddress<const void *>();
call((void *)rdv_BlockCheckFail);
ass.Bx(r0);
// Main loop
Label no_updateLabel;
// ngen_mainloop:
mainloop = ass.GetCursorAddress<void (*)(void *)>();
RegisterList savedRegisters = RegisterList::Union(
RegisterList(r4, r5, r6, r7),
RegisterList(r8, r9, r10, r11),
RegisterList(r12, lr));
{
UseScratchRegisterScope scope(&ass);
scope.ExcludeAll();
ass.Push(savedRegisters);
}
ass.Mov(r9, SH4_TIMESLICE); // load cycle counter
ass.Mov(r8, r0); // load context
ass.Ldr(r4, MemOperand(r8, rcbOffset(cntx.pc))); // load pc
ass.B(&no_updateLabel); // Go to mainloop !
// this code is here for fall-through behavior of do_iter
Label do_iter;
Label cleanup;
// intc_sched:
intc_sched = ass.GetCursorAddress<const void *>();
ass.Add(r9, r9, SH4_TIMESLICE);
ass.Mov(r4, lr);
call((void *)UpdateSystem);
ass.Mov(lr, r4);
ass.Cmp(r0, 0);
ass.B(ne, &do_iter);
ass.Ldr(r0, MemOperand(r8, rcbOffset(cntx.CpuRunning)));
ass.Cmp(r0, 0);
ass.Bx(ne, lr);
// do_iter:
ass.Bind(&do_iter);
ass.Mov(r0, r4);
call((void *)rdv_DoInterrupts);
ass.Mov(r4, r0);
// no_update:
no_update = ass.GetCursorAddress<const void *>();
ass.Bind(&no_updateLabel);
// next_pc _MUST_ be on r4 *R4 NOT R0 anymore*
ass.Ldr(r0, MemOperand(r8, rcbOffset(cntx.CpuRunning)));
ass.Cmp(r0, 0);
ass.B(eq, &cleanup);
ass.Sub(r2, r8, -rcbOffset(fpcb));
#if RAM_SIZE_MAX == 33554432
ass.Ubfx(r1, r4, 1, 24); // 24+1 bits: 32 MB
// RAM wraps around so if actual RAM size is 16MB, we won't overflow
#elif RAM_SIZE_MAX == 16777216
ass.Ubfx(r1, r4, 1, 23); // 23+1 bits: 16 MB
#else
#error "Define RAM_SIZE_MAX"
#endif
ass.Ldr(pc, MemOperand(r2, r1, LSL, 2));
// cleanup:
ass.Bind(&cleanup);
{
UseScratchRegisterScope scope(&ass);
scope.ExcludeAll();
ass.Pop(savedRegisters);
}
ass.Bx(lr);
// Memory handlers
for (int s=0;s<6;s++)
{
const void* fn=s==0?(void*)_vmem_ReadMem8SX32:
s==1?(void*)_vmem_ReadMem16SX32:
s==2?(void*)_vmem_ReadMem32:
s==3?(void*)_vmem_WriteMem8:
s==4?(void*)_vmem_WriteMem16:
s==5?(void*)_vmem_WriteMem32:
0;
bool read=s<=2;
//r0 to r13
for (int i=0;i<=13;i++)
{
if (i==1 || i ==2 || i == 3 || i == 4 || i==12 || i==13)
continue;
const void *v;
if (i == 0)
v = fn;
else
{
v = ass.GetCursorAddress<const void *>();
ass.Mov(r0, Register(i));
jump(fn);
}
_mem_hndl[read][s % 3][i] = v;
}
}
for (int optp = SZ_32I; optp <= SZ_64F; optp++)
{
//r0 to r13
for (int reg = 0; reg <= 13; reg++)
{
if (reg == 1 || reg == 2 || reg == 3 || reg == 4 || reg == 12 || reg == 13)
continue;
if (optp != SZ_32I && reg != 0)
continue;
_mem_hndl_SQ32[optp - SZ_32I][reg] = ass.GetCursorAddress<const void *>();
if (optp == SZ_64F)
{
ass.Lsr(r1, r0, 26);
ass.Cmp(r1, 0x38);
ass.And(r1, r0, 0x3F);
ass.Add(r1, r1, r8);
jump((void *)&_vmem_WriteMem64, ne);
ass.Strd(r2, r3, MemOperand(r1, rcbOffset(sq_buffer)));
}
else
{
ass.And(r3, Register(reg), 0x3F);
ass.Lsr(r2, Register(reg), 26);
ass.Add(r3, r3, r8);
ass.Cmp(r2, 0x38);
if (reg != 0)
ass.Mov(ne, r0, Register(reg));
jump((void *)&_vmem_WriteMem32, ne);
ass.Str(r1, MemOperand(r3, rcbOffset(sq_buffer)));
}
ass.Bx(lr);
}
}
ass.Finalize();
emit_Skip(ass.GetBuffer()->GetSizeInBytes());
ngen_FailedToFindBlock = ngen_FailedToFindBlock_;
INFO_LOG(DYNAREC, "readm helpers: up to %p", ass.GetCursorAddress<void *>());
emit_SetBaseAddr();
ccmap[shop_test] = eq;
ccnmap[shop_test] = ne;
ccmap[shop_seteq] = eq;
ccnmap[shop_seteq] = ne;
ccmap[shop_setge] = ge;
ccnmap[shop_setge] = lt;
ccmap[shop_setgt] = gt;
ccnmap[shop_setgt] = le;
ccmap[shop_setae] = hs;
ccnmap[shop_setae] = lo;
ccmap[shop_setab] = hi;
ccnmap[shop_setab] = ls;
}
void ngen_GetFeatures(ngen_features* dst)
{
dst->InterpreterFallback=false;
dst->OnlyDynamicEnds=false;
}
RuntimeBlockInfo* ngen_AllocateBlock()
{
return new DynaRBI();
};
#endif
Reference in New Issue View Git Blame Copy Permalink